Thesis Hap Mimo 2011

ΠΑΝΕΠΙΣΤΗΜΙΟ ΠΕΙΡΑΙΩΣ
Τμήμα Ψηφιακών Συστημάτων

ΜΟΝΤΕΛΟΠΟΙΗΣΗ ΡΑΔΙΟΔΙΑΥΛΟΥ
ΠΟΛΛΑΠΛΩΝ–ΕΙΣΟΔΩΝ ΠΟΛΛΑΠΛΩΝ–ΕΞΟΔΩΝ
ΓΙΑ ΑΣΥΡΜΑΤΑ ΕΥΡΥΖΩΝΙΚΑ ΣΥΣΤΗΜΑΤΑ
ΣΤΡΑΤΟΣΦΑΙΡΙΚΩΝ ΕΠΙΚΟΙΝΩΝΙΩΝ

ΔΙΔΑΚΤΟΡΙΚΗ ΔΙΑΤΡΙΒΗ
του
ΕΜΜΑΝΟΥΗΛ Θ. ΜΙΧΑΗΛΙΔΗ

Πειραιάς, Μάιος 2011

ΠΑΝΕΠΙΣΤΗΜΙΟ ΠΕΙΡΑΙΩΣ
Τμήμα Ψηφιακών Συστημάτων

ΜΟΝΤΕΛΟΠΟΙΗΣΗ ΡΑΔΙΟΔΙΑΥΛΟΥ
ΠΟΛΛΑΠΛΩΝ–ΕΙΣΟΔΩΝ ΠΟΛΛΑΠΛΩΝ–ΕΞΟΔΩΝ
ΓΙΑ ΑΣΥΡΜΑΤΑ ΕΥΡΥΖΩΝΙΚΑ ΣΥΣΤΗΜΑΤΑ
ΣΤΡΑΤΟΣΦΑΙΡΙΚΩΝ ΕΠΙΚΟΙΝΩΝΙΩΝ

ΔΙΔΑΚΤΟΡΙΚΗ ΔΙΑΤΡΙΒΗ
του
ΕΜΜΑΝΟΥΗΛ Θ. ΜΙΧΑΗΛΙΔΗ

Η διατριβή υποβάλλεται για την μερική κάλυψη των απαιτήσεων
με στόχο την απόκτηση του Διδακτορικού Διπλώματος


UNIVERSITY OF PIRAEUS
Department of Digital Systems

MULTIPLE–INPUT MULTIPLE–OUTPUT
RADIO CHANNEL MODELING FOR BROADBAND WIRELESS
STRATOSPHERIC COMMUNICATIONS SYSTEMS

A Thesis
Presented to
The Academic Faculty
by
EMMANOUEL T. MICHAILIDIS

In Partial Fulfillment
of the Requirements for the Degree
Doctor of Philosophy

Piraeus, Greece, May 2011

-
nANEntITHMIO nEIPAIOr
TI..ul~a lJ.If"\q.HaKWV rUm:f"\~Ct"(WV
MONTEAonOIH}:H PA~IO~IAYAOY
nOAAAnAON-EI}:O~ON nOAAAnAON-E=:O~ON
riA A}:YPMATA EYPYZONIKA }:Y}:THMATA
}:TPATO}:(])AIPIKON EnlKOINONION
~I~AKTOPIKH ~IATPIBH
"(OU
EMMANOYHA e. MIXAHAIl\H
rU~~OUAEUUKr\ EnL"(porrr\: ASavCtCJlo<; r. KavCt"(a<;

rEWPYlO<; EUSU~OyAOU
Cl>i.Amno<; KWVm:aVllVOU
EYKpiSf"\KE arro "(f"\V Err"(a~EAr\ E~E"(am:lKr\ EnL"(pOrrr\ "(f"\V gr'f Ma.Tou 2011.
A. r. Kavchac; r. Eu8uJ.l.oyAou <D. KwvatavTlvou

Ava:rrA. Ka8I)Yl)t~<; ETtlK. Ka8I)yl)t~<; Ka8I)yl)t~<;
nav/J.l.l u nElpmw<; nav/J.l.lou nElpmw<; E.M.
I. . KaTolKac; <D.-N. nauAL60u

Ka8I)Yl)t~<;
nav/J.l.iou nElpalw<;
-1--
Copyright© 2011- E~~avou~A 0. MlXaI")AlOI")C;
ME ETIl¢UAa~1") navtOC; olKau.il~aTOC;.
AnayopEuETat I") avtLypa¢~, an09~KEUOI") Kat OLaVO~~ TI")<; napouoa<; Epyaoia<;, E~

OAOKA~POU ~ T~~~aTO<; aUT~<;, Yla E~nOpLKO OKono. ETIlTpEnETat I") avaTunWOI"),
ano9~KEUOI") Kat Olavo~~ YLa OKono ~I") KEPOOOKOTIlKO, EKnaLOEunK~<; t)
EpEuVl")nK~<; ¢UOI")<;, uno TI")V npoOno9£01") va ava¢EpETat I") nl")y~ npOEAWOI")<; Kat
va o LaTI") PE iTat TO napov ~~vu~a. EpWT~J..laTa nou a¢opouv TI") XP~OI") TI")<; £pyaoLa<;
Yla K£POOOKOTIlKO oKono npEn£L va anw9uvovtat npo<; TOV ouyypa¢ta. OL
anoljJ£l<; KaL Ta ou~£pa.o~aTa nou n£plEXOvtat O£ aUTo TO Eyypa¢o £K¢Pa.~ouv
TOV ouyypa¢Ea KaL liEv nptn£l va £p~l")v£u9£i on avtmpoown£uouv n<; £niol")~£<;
9EO£l<; TOU nav£mOTI")~iou n£Lpau.il<;.
--------~---.-- .... -- ~-~-------

UNIVERSITY OF PIRAEUS
Department of Digital Systems
MULTIPLE-INPUT MULTIPLE-OUTPUT
RADIO CHANNEL MODELING FOR BROADBAND WIRELESS
STRATOSPHERIC COMMUNICATIONS SYSTEMS
Doctor of Philosophy Thesis
EMMANOUEL T. MICHAILIDIS
Approved by:.
A. G. Kanatas, Supervisor G. Efthymoglou, Advisor

/~-
P. Constantinou, Advisor
Assoc. Professor Assist. Professor Professor
University of Piraeus University of Piraeus N.T.U.A.
F.-N. Pavlidou
Professor
University of Piraeus A.U.TH.
Lecturer
N.T.U.A.
Date Approved: May 9th 2011

-_··It ----.--
Dr. Emmanouel T. Michailidis
Copyright© 2011- Emmanouel T. Michailidis
All rights reserved.
Copying, storing or distributing of this thesis, as part or as a whole, for commercial

purposes is prohibited. Copying, storing or distributing this thesis for non-profit,
educational or research purposes is allowed, provided the source is explicitly
mentioned and the nonce-text is preserved. Questions concerning the non-profit
use of this thesis will have to be addressed to the author. The concepts and
conclusions included in this work express the author's personal opinion and should
not be interpreted that they represent University of Piraeus official concepts.

στους γονείς μου
Θεόδωρο και Πέρσα,
και στον αδερφό μου Σπύρο…

to my parents
Theodoros and Persa,
and my brother Spiros…

ΕΥΧΑΡΙΣΤΙΕΣ
Η πορεία μας στη ζωή βασίζεται στις ευκαιρίες που θα βρεθούν στο δρόμο μας
και στις αποφάσεις μας αν θα τις εκμεταλλευτούμε ή όχι. Η παρούσα Διδακτορική
Διατριβή ξεκίνησε ως μια ευκαιρία για την κατάκτηση της γνώσης και αποτελεί το
επιστέγασμα μιας δύσκολης, επίπονης και αδιάλειπτης προσωπικής προσπάθειας,
η οποία απαίτησε όχι μόνο κόπο και χρόνο, αλλά και ψυχική δύναμη. Εντούτοις, η
ολοκλήρωση της Διατριβής αυτής θα ήταν αδύνατη χωρίς την αμέριστη
συμπαράσταση και ουσιαστική συμβολή κάποιων ατόμων, τα οποία αισθάνομαι
την ανάγκη να ευχαριστήσω.
Καταρχάς, θα ήθελα να ευχαριστήσω θερμά τον Αναπληρωτή Καθηγητή του
Πανεπιστημίου Πειραιά και επιβλέποντα της Διδακτορικής μου Διατριβής Δρ.
Αθανάσιο Κανάτα για την εμπιστοσύνη που έδειξε στις δυνατότητες μου, για την
άριστη συνεργασία μας, για την υποδειγματική ερευνητική καθοδήγηση και για
την αδιάκοπη στήριξη, βοήθεια, επιμονή και υπομονή του. Με την εμπειρία του
και τις υψηλές επιστημονικές του αξίες αποτέλεσε σύμμαχο σε κάθε ερευνητική
μου δραστηριότητα. Του εύχομαι καλή συνέχεια στο εκπαιδευτικό και ερευνητικό
του έργο, καθώς και προσωπική και οικογενειακή ευτυχία.
Τις ειλικρινείς μου ευχαριστίες θα ήθελα επίσης να εκφράσω προς τον
Επίκουρο Καθηγητή του Πανεπιστημίου Πειραιά και μέλος της τριμελούς
συμβουλευτικής μου επιτροπής Δρ. Γεώργιο Ευθύμογλου για τις πολύτιμες
συμβουλές του στις μεταξύ μας συνεργασίες και τη συνεχή ενθάρρυνση που μου
παρείχε.
Επίσης, θα ήθελα να ευχαριστήσω μερικούς συναδέλφους, με τους οποίους
μοιραζόμαστε τα ίδια όνειρα και προσπάθειες, για την άψογη συνεργασία σε
διάφορες εκπαιδευτικές και ερευνητικές δραστηριότητες, για τις εποικοδομητικές
συζητήσεις και για την αλληλοσυμπαράσταση στην κοινή μας πορεία.
Συγκεκριμένα, θα ήθελα να ευχαριστήσω τους Βλάση Μπαρούση, Δρ. Παναγιώτη
Θεοφιλάκο, Δρ. Χρήστο Σπηλιωτόπουλο, Δρ. Γεώργιο Μπράβο, Κώστα Κυρίτση και
i

Δρ. Άγγελο Σαατσάκη. Τους εύχομαι από καρδιάς καλή σταδιοδρομία και πολλές
επαγγελματικές και προσωπικές επιτυχίες.
Επιπλέον, θα ήθελα να ευχαριστήσω το νονό μου και Αναπληρωτή Καθηγητή
της Οικονομικής Επιστήμης του Πανεπιστημίου Mid‐Sweden University της
Σουηδίας Δρ. Νικία Σαράφογλου για τις χρήσιμες υποδείξεις, παρατηρήσεις και
συμβουλές του.
Δεν παραλείπω να ευχαριστήσω ιδιαίτερα τους φίλους μου για τη συνεχή και
υπομονετική στήριξή τους και τις ευχάριστες και ξένοιαστες στιγμές που περνάμε
μαζί εδώ και χρόνια. Θα ήθελα αρχικά να ευχαριστήσω τoν αδερφικό μου φίλο και
Μηχανολόγο Μηχανικό ΕΜΠ Σπύρο Καπαγερίδη για τη φιλία, το ήθος και τη
συμπαράσταση του στα εύκολα και τα δύσκολα από τις πρώτες τάξεις του Λυκείου
μέχρι σήμερα. Επίσης, ευχαριστώ από καρδιάς για τη φιλία τους τον Κώστα
Κουφόγλου, τη Νεκταρία Χουλιαρά, το Γιώργο Παπαδόπουλο, τη Δώρα Μπέλλου,
τον Τάκη Αλεφραγκή, το Γιώργο Αγιοργιωτάκη, τον Παναγιώτη Ιασωνίδη, το
Νεκτάριο Φαρασσόπουλο, το Θανάση Τσαγκάρη, το Σωτήρη Τραχιώτη, το Γιώργο
Ολυμπίου, το Γιάννη Θωμαΐδη, το Στέφανο Κόκλα, τη Μαριάνθη Καραγιάννη και
την αγαπημένη μου Έλενα Τριανταφυλλίδη.
Τέλος, θα ήθελα να ευχαριστήσω ολόψυχα την οικογένειά μου, τον πατέρα μου
Θεόδωρο, τη μητέρα μου Πέρσα και τον αδερφό μου Σπύρο για την παντοτινή
τους αγάπη, την κατανόηση και την αδιάκοπη ηθική και υλική στήριξη που μου
παρέχουν απλόχερα από τη στιγμή που γεννήθηκα. Ως μικρή ένδειξη
ευγνωμοσύνης και αγάπης, τους αφιερώνω το πιο ξεχωριστό μου επίτευγμα, τη
Διδακτορική μου Διατριβή…

Εμμανουήλ Θ. Μιχαηλίδης
ii

ACKNOWLEDGEMENTS
We are defined by opportunities, even the ones we miss. This doctoral thesis
started as an opportunity for the conquest of knowledge and stands for the
outcome of a difficult, painful, and uninterruptible struggle, which not only
required exertion and time, but also inward power. Nevertheless, this thesis would
not have been completed without the unreserved solidarity and substantial
contribution of some persons, for which I truly feel gratitude.
To begin with, I would like to warmly thank Dr. Athanasios Kanatas, supervisor
of my doctoral thesis and Associate Professor at the University of Piraeus, for his
trust and faith to my capabilities and potential, for our excellent cooperation, his
exemplary guidance to my research, and his continuous support, help, persistence,
and forbearance, which all strongly and successfully raised my research activity. I
wish him to effectually continue his educational and research tasks and have
personal happiness and family prosperity.
In addition, I would like to truthfully thank Dr. George Efthymoglou, Assistant
Professor at the University of Piraeus and advisor of my thesis, for his continuous
provision with valuable advices and encouragement.
Furthermore, I would also like to thank my colleagues for our excellent
cooperation, constructive discussions, mutual understanding and support in our
common course, with whom I perfectly collaborated in various educational and
research activities and we share the same dreams and endeavors. In particular, I
would like to thank Vlasis Barousis, Dr. Panagiotis Theofilakos, Dr. Christos
Spiliotopoulos, Dr. George Bravos, Kostas Kyritsis, and Dr. Aggelos Saatsakis. I
heartily wish them to have a brilliant career and many successful achievements.
I would also like to thank my godfather Dr. Nikias Sarafoglou, Associate Professor
of Economics at the Mid‐Sweden University for his useful advices and comments.
I would not omit to especially thank my friends for their constant and patient
reliance and the joyful and carefree moments we have shared together for many
iii

years. I would initially like to thank my brotherly friend Spiros Kapageridis,
Mechanical Engineer, NTUA, for his friendship, his morality, and his solidarity in
both good and tough moments, since the first grades of high school. I would also
like to sincerely thank Kostas Koufoglou, Nektaria Houliara, George Papadopoulos,
Dora Bellou, Takis Alefragis, George Agiorgiotakis, Panagiotis Iasonidis, Nektarios
Farassopoulos, Thanasis Tsagaris, Sotiris Trahiotis, George Olymbiou, John
Thomaidis, Steve Koklas, Marianthi Karagianni, and my beloved Elena Triantafillidi
for their friendship.
Finally, I would like to express my deepest thankfulness to my family, my father
Theodoros, my mother Persa, and my brother Spiros for their everlasting love, their
encouragement, and their incessant moral and material support, since the day I
was born. As the least token of gratitude and love, I dedicate to them my most
especial achievement, this doctoral thesis...

Emmanouel T. Michailidis
Piraeus, Greece, May 2011

iv

ΠΕΡΙΛΗΨΗ
Οι στρατοσφαιρικές πλατφόρμες (high altitude platforms – HAPs) αποτελούν
μία από τις πολλά υποσχόμενες εναλλακτικές τεχνολογίες για τη δημιουργία
ασύρματων δικτύων υψηλών ταχυτήτων νέας γενιάς. Οι επιτακτικές ανάγκες για
αποδοτικότερη εκμετάλλευση του φάσματος συχνοτήτων, για αύξηση του ρυθμού
μετάδοσης πληροφορίας και χωρητικότητας του διαύλου, καθώς και για βελτίωση
της αξιοπιστίας της ασύρματης ζεύξης έχουν οδηγήσει στην ραγδαία ανάπτυξη της
αρχιτεκτονικής πολλαπλών‐εισόδων πολλαπλών‐εξόδων (multiple‐input multiple‐
output – MIMO). Η αρχιτεκτονική αυτή έχει χαρακτηριστεί ως η βασική τεχνολογία
φυσικού στρώματος για τα μελλοντικά ευρυζωνικά συστήματα επικοινωνιών.
Εντούτοις, η επιτυχής εφαρμογή πολλαπλών κεραιών και στα δύο άκρα της
ασύρματης ζεύξης προϋποθέτει κατάλληλο χαρακτηρισμό και μοντελοποίηση του
διαύλου επικοινωνίας. Η παρούσα διδακτορική διατριβή πραγματεύεται την
ανάπτυξη και αξιολόγηση μοντέλων διαύλου τεχνολογίας ΜΙΜΟ για ευρυζωνικά
ασύρματα συστήματα επικοινωνιών βασισμένα σε HAPs.
Συγκεκριμένα, αναπτύσσονται καινοτόμα θεωρητικά γεωμετρικά μοντέλα

τριών‐διαστάσεων για διαύλους Ricean HAP‐MIMO στενής και ευρείας ζώνης στις
εκχωρημένες για κινητές επικοινωνίες μέσω HAPs ζώνες συχνοτήτων L και S. Με
χρήση των προτεινόμενων μοντέλων χαρακτηρίζεται λεπτομερώς η
πολυδιαδρομική διάδοση και μελετώνται διεξοδικά οι στατιστικές ιδιότητες και η
απόδοση της χωρητικότητας του διαύλου, λαμβάνοντας υπόψη διάφορες
παραμέτρους, όπως η γωνία ανύψωσης της πλατφόρμας, η διάταξη των
στοιχειοκεραιών, η διασπορά Doppler, η διασπορά καθυστέρησης και η τριών‐
διαστάσεων μη‐ισοτροπική κατανομή των τοπικών σκεδαστών που περιβάλλουν
τον επίγειο σταθμό. Επιπρόσθετα, τα προτεινόμενα μοντέλα επιτρέπουν την
εκτίμηση της απαιτούμενης απόστασης των στοιχείων της κεραίας των HAPs, ώστε
να εξασφαλίζονται ασυσχέτιστες αποκρίσεις στον πίνακα ΜΙΜΟ του διαύλου.
Επιπλέον, αναπτύσσονται καινοτόμα ντετερμινιστικά και στατιστικά μοντέλα

προσομοίωσης τριών διαστάσεων για διαύλους HAP‐MIMO, τα οποία βασίζονται
v

στα προαναφερθέντα θεωρητικά μοντέλα στενής και ευρείας ζώνης. Για την
ανάπτυξη των μοντέλων προσομοίωσης χρησιμοποιείται η μέθοδος αθροίσματος
ημιτόνων (sum‐of‐sinusoids – SoS), η οποία έχει γίνει αποδεκτή ευρέως στον
ακαδημαϊκό χώρο και στη βιομηχανία. Η αποδοτικότητα, η ακρίβεια και η
πολυπλοκότητα των προτεινόμενων μοντέλων προσομοίωσης διερευνάται με
γνώμονα τις στατιστικές ιδιότητες. Τα αποτελέσματα επιβεβαιώνουν την υψηλή
επίδοση των μοντέλων αυτών.
Τέλος, προτείνεται μια καινοτόμος γεωμετρική σχεδιαστική μέθοδος

μεγιστοποίησης της χωρητικότητας του διαύλου HAP‐MIMO σε συστήματα
επικοινωνιών σταθερής ασύρματης πρόσβασης στις εκχωρημένες ζώνες
συχνοτήτων Ka και V για σταθερές επικοινωνίες μέσω HAPs. Σε αυτές τις
χιλιοστομετρικές συχνότητες η παρουσία ισχυρής απευθείας συνιστώσας είναι
απαραίτητη και η βροχή επηρεάζει σημαντικά την ποιότητα της ζεύξης. Το
ενδεχόμενο αυξημένο κέρδος χωρητικότητας του διαύλου της προτεινόμενης
βελτιστοποιημένης αρχιτεκτονικής διερευνάται, υπό συνθήκες καθαρού ουρανού
και βροχής.
Τα προτεινόμενα μοντέλα διαύλου HAP‐MIMO παρέχουν σε άλλους ερευνητές

τις κατάλληλες κατευθυντήριες γραμμές για τον χαρακτηρισμό, την ανάλυση, την
αξιολόγηση και την σχεδίαση μελλοντικών συστημάτων κινητών και σταθερών
επικοινωνιών τεχνολογίας HAP‐MIMO.

Λέξεις Κλειδιά: Διαλείψεις Ricean, δίαυλος πολλαπλών‐εισόδων πολλαπλών‐

εξόδων (ΜΙΜΟ), επίδραση βροχής, μοντέλο προσομοίωσης, στρατοσφαιρικές
πλατφόρμες (HAPs), συσχέτιση, τριών διαστάσεων (3‐D) σκέδαση, χωρητικότητα.

vi

ABSTRACT
High altitude platforms (HAPs) are one of the most promising alternative
infrastructures for realizing next generation high data‐rate wireless
communications networks. Considering the continuous demand for enhanced
spectral efficiency, increased channel capacity, and improved link reliability,
multiple‐input multiple‐output (MIMO) technology is the leading candidate for
future communications systems. The development of MIMO‐based
communications networks depends on a proper characterization and modeling of
the propagation channel. To enable the successful design and performance
evaluation of broadband wireless HAP‐MIMO systems, this doctoral thesis focuses
on modeling of HAP‐MIMO channels.
Novel three‐dimensional (3‐D) geometry‐based reference models for Ricean

fading channels are proposed that encompass narrowband and wideband HAP‐
MIMO channel scenarios with line‐of‐sight (LoS) and non‐line‐of‐sight (NLoS)
connections at L and S frequency bands licensed for mobile communications
through HAPs. From these models, the statistical properties and the capacity
performance are analytically and thoroughly studied in terms of various
parameters, such as the elevation angle of the platform, the antenna array
configuration, the Doppler and delay spread, and the 3‐D non‐isotropic distribution
of the local scatterers. Using the theoretical expressions one can easily evaluate
numerically the HAP antenna inter‐element spacing required to achieve
uncorrelated responses in the HAP‐MIMO channel matrix.
Moreover, novel 3‐D deterministic and statistical simulation models for HAP‐
MIMO channels based on the reference narrowband and wideband models are
developed by using the sum‐of‐sinusoids (SoS) method, which has been widely
accepted by academia and industry. The efficiency, the accuracy, and the
complexity of these models are thoroughly investigated with respect to the
statistical properties. The results indicate that the proposed simulation models
vii

yield high‐performance and satisfactorily approximate the statistical properties of
the reference models.
Finally, a novel geometrical design approach to construct high‐capacity HAP‐

MIMO communications systems in LoS propagation environments is proposed,
considering the Ka and V frequency bands licensed for fixed broadband
communications through HAPs. In these mm‐wave frequencies, the rain has a
significant effect on the quality of the link. The potential channel capacity gain of
the optimized LoS‐HAP‐MIMO architecture is investigated, under clear sky and rain
conditions.
The proposed models for HAP‐MIMO channels would provide other researchers
a convenient framework and guidelines for the characterization, analysis, test, and
design of future mobile and fixed HAP‐MIMO communications systems.

Keywords: Capacity, correlation, high‐altitude platforms (HAPs), multiple‐input

multiple‐output (MIMO) channel, rain effects, Ricean fading, simulation model,
three‐dimensional (3‐D) scattering.
viii

TABLE OF CONTENTS
ΕΥΧΑΡΙΣΤΙΕΣ………………………………………………………………..………………………………….. i
ACKNOWLEDGEMENTS ………….…………………………..…………………………………………… iii
ΠΕΡΙΛΗΨΗ.………………………………………………………………..……………………………………. v
ABSTRACT…….………………………………………………………..………………………………………… vii
TABLE OF CONTENTS………………………….…………………………………………………………..… ix
LIST OF TABLES………………………………………………………………………………………………… xiii
LIST OF FIGURES………………………………………………………………………………………………. xv
ABBREVIATIONS…………………………………………………………………………………………..…. xxi
1 INTRODUCTION..………………..…………………………………………………………….……….. 1
1.1 High‐Altitude Platforms for Wireless Communications…….………………..…. 1
1.1.1 Structures………………………………………………………………………..……….... 3
1.1.2 Advantages……………………………………………………………………………..…... 5
1.1.3 Spectrum Allocation and Licensed Frequency Bands………..…………. 5
1.1.4 Applications and Services…………………….…………..……………………..….. 9
1.1.4.1 Broadband Fixed Wireless Access…………….….....……………...… 10
1.1.4.2 High‐Speed Mobile Services…………………………………………….. 10
1.2 Future Trends in Wireless Communications……………………….…………...…... 11
1.3 Thesis Motivation and Objectives…………………………………………………...….... 14
1.4 Research Contributions and Author’s Publications………..………………….….. 16
1.5 Thesis Outline………………………………………………………………..………………….…. 18
2 OVERVIEW ON PROPAGATION MECHANISMS, MIMO PRINCIPLE, AND
CHANNEL MODELING……………………………………………………………………………….… 19
2.1 Introduction…………….…………………………………………………………………………… 19
2.2 The Wireless Propagation Channel…………………..………….………………………. 19
2.2.1 Path Loss…………………………….…………………………………………………..….. 21
2.2.2 Rain Effects……………………………………………………………………………..….. 22
2.2.3 Shadow Fading………………………………………………………………………..….. 25
2.2.4 Multipath Fading…………………..………………………………………………..…... 26
ix

2.2.5 Time, Frequency, and Space Selectivity……………....….…………...……… 27
2.2.6 The Diversity Concept……………….…………………………......................... 29
2.3 The MIMO Technology…………….……………………………………….…………………… 31
2.3.1 Benefits…………………………….…………………………………………………..….... 32
2.3.2 Channel Statistics…………………………………………………………………..….... 33
2.4 The Earth‐Stratosphere Channel: From SISO to MIMO Modeling……..…… 36
2.4.1 Survey of HAP‐SISO Channels………………………………………………………… 36
2.4.1.1 Switched‐Channel Models…………………………………………….…… 37
2.4.1.2 Empirical Channel Models…………………………………………….…… 38
2.4.1.3 Geometry‐based Channel Models…………………………..…….…… 39
2.4.2 Classification of MIMO Channel Models………………………………………… 41
2.4.3 Recent Advances in Satellite/HAP MIMO Channel Modeling……….… 42
2.4.3.1 Physical‐Statistical Modeling……………………………………….…… 43
2.4.3.2 Analytical Modeling……………………………………………..……….…… 44
2.4.3.3 Polarized Satellite/HAP MIMO Channels……………………….…… 45
2.4.4 Potential Application of MIMO Technology over Single HAPs…..…… 45
3 THREE‐DIMENSIONAL MODELING OF SPACE‐TIME CORRELATED
HAP‐MIMO CHANNELS……….....………………………………………………………………….. 47
3.1 Introduction…….……..……………………………………………………………………………. 47
3.2 The HAP‐MIMO System Model………….………………….………………………………. 49
3.3 A Reference Model for HAP‐MIMO Channels.……..………………………………… 50
3.3.1 Channel Impulse Response…….....…………………………….…….…..………. 56
3.4 Space‐Time Correlation Function..……..…………...…………………………………..… 59
3.4.1 Investigation of Space‐Time Correlation of HAP‐MIMO Channels... 66
3.4.1.1 Effect of HAP Antenna Displacement………….……………………. 77
3.5 Analysis of HAP‐MIMO Channel Capacity…………….……………………………….. 79
3.5.1 Definition of HAP‐MIMO Channel Capacity……….....……….…..……….. 79
3.5.2 Numerical Results…….....……….…………………………………………...……….. 82
3.6 Summary…………………………………………………………..………………...………………… 87

x

4 THREE‐DIMENSIONAL SIMULATION OF HAP‐MIMO CHANNELS………………… 89
4.1 Introduction………………………………………………………………………………………….. 89
4.2 The Stochastic Simulation Model…………………………………………….……………. 91
4.3 The Deterministic Simulation Model……………………………………….……………. 94
4.3.1 The Deterministic Simulation Model Based on Set Partitioning…..… 98
4.4 The Statistical Simulation Model…………………………………………………………… 101
4.5 Simulation Results………………………………………………………………………………… 103
4.6 Summary……………………………………………………………………………..……………….. 111
5 THREE‐DIMENSIONAL MODELING AND SIMULATION OF WIDEBAND
HAP‐MIMO CHANNELS…....…………………………………….………………………………….. 113
5.1 Introduction………………………..……………………………………………………………..... 113
5.2 The Wideband HAP‐MIMO System Model……………………………………..……… 115
5.3 A Reference Model for Wideband HAP‐MIMO Channels…….…………….….. 116
5.3.1 Input Delay‐Spread and Time‐Variant Transfer Functions…………..… 120
5.4 Statistical Properties of Wideband HAP‐MIMO Channels……..…..………….. 123
5.4.1 Space‐Time‐Frequency Correlation Function……….….…..……….....…. 123
5.4.2 Space‐Doppler Power Spectrum……….….…..……….....……………………. 126
5.4.3 Power Space‐Delay Spectrum…………………....…………….…………..……… 130
5.4.4 Numerical Results………………..…………………....…………….…………..……… 133
5.5 Simulation Models for Wideband HAP‐MIMO Channels……..…....…………. 137
5.5.1 Simulation Results…………………………………………….….…..………….....…. 140
5.6 Summary……………………………………….………………………..………….………..……… 141
6 MODELING OF LINE‐OF‐SIGHT HAP‐MIMO CHANNELS….…………………………… 143
6.1 Introduction……….….….…………………………………………………………………………. 143
6.2 The LoS‐HAP‐MIMO System Model…………………………………………………………145
6.2.1 A Geometrical Model for LoS‐HAP‐MIMO Channels………..………….… 147
6.3 Capacity Optimization Techniques……………………………….…………………………148
6.3.1 Practical Considerations…………………………………………………………………150
6.4 Numerical Results……….………………………………………………….……………………… 151
6.4.1 Sensitivity Study………………………………………………………………………….. 155
xi

6.5 Summary……………………………………………………………………………………….……… 158
7 CONCLUSIONS AND FUTURE RESEARCH DIRECTIONS……….………..……………… 159
7.1 Conclusions………………………………………………………….…..…………………………… 159
7.2 Future Research Directions……………………………………................................. 161
APPENDIX A…………………………….………….………………………..…………………………………. 165
APPENDIX B…………………….………………….………………………..…………………………………. 169
REFERENCES………………………………………………………………..………………………….………. 173
ΣΥΝΤΟΜΟ ΒΙΟΓΡΑΦΙΚΟ ΣΗΜΕΙΩΜΑ………………….…………………………………………… 187
VITA………….……………………………..………………………………..……………………………………. 189

xii

LIST OF TABLES
1 Radio frequency bands...………………….……………………………………………………. 7
2 Microwave frequency bands………….……………………………………………………... 7
3 Frequency spectrum available for HAP‐based applications…………………….. 8
4 Regression coefficients for estimating specific rain attenuation……..……... 24
5 Definition of the parameters used in the reference model……….……..…….. 54
6 The complexity of the simulation models…….………….……….……………………. 102
7 Performance valuation of the deterministic simulation model (A)…………. 104
8 Performance valuation of the deterministic simulation model (B)…………. 105
9 Accuracy evaluation of the statistical simulation model……...……..…………. 110
10 Efficiency evaluation of the statistical simulation model………....……………. 110
11 Definition of the parameters used in the reference wideband model……. 119
12 Link budget parameters…………………………………………………………...……………. 154

xiii

xiv

LIST OF FIGURES
1 The troposphere, stratosphere, and mesosphere layers of the
atmosphere....………….....….……………………………………………………................. 2
2 The average wind velocity for different altitude above the Earth’s
surface……......………….....….…………………………………………………….................. 3
3 The HAP and UAV structures: (a) solar‐powered unmanned aircraft, (b)
manned aircraft, (c) unmanned aerial vehicle, and (d) solar‐powered
unmanned airship.....…………………….………………………………………................. 4
4 The ITU region map....………….....………..………………………….......................... 6
5 Visualization of an integrated satellite‐HAP‐terrestrial 4G wireless
communications system providing fixed and mobile broadband
services.....……………….....….……………………………………………………................. 12
6 High level view of the future 4G wireless reconfigurable radio systems... 13
7 Architecture of a generic wireless communications system.....……........... 21
8 The specific rain attenuation as a function of the rain rate for different
carrier frequency.....…….….……………………………………………………................. 24
9 Rain attenuation path geometry.....………………………………………................. 25
10 Visualization of a multipath propagation environment.....…………………..... 27
11 The relationships of the time, frequency, and space domains…............... 29
12 Basic representation of SISO, SIMO, MISO, MIMO, and MU‐MIMO
systems…….....………….....….……………………………………………………................. 31
13 Visualization of the MIMO radio channel..............…………………................. 34
14 The typical geometry of a HAP‐based system..............………………............. 39
15 The ellipsoid geometrical model for HAP‐SISO channels......................…… 40
16 The circular cone geometrical model for HAP‐SISO channels………..………… 40
17 Visualization of isotropic scattering conditions in an open area............... 48
18 Visualization of non‐isotropic scattering conditions in a narrow street.... 48
19 Basic concept of a HAP‐based communication system and a cylindrical
scattering region inside the coverage area.....…………................................ 51
20 The LoS paths of the 3‐D geometrical model for a 2  2 HAP‐MIMO
channel.....……………….....….…………………………………………………….................. 52
21 The LoS paths of the projection of the 3‐D geometrical model for a
2  2 HAP‐MIMO channel.....………………………………………………….................. 52
22 The NLoS paths of the 3‐D geometrical model for a 2  2 HAP‐MIMO
channel.....………….....….………………………………………………………….................. 53
23 The NLoS paths of the projection of the 3‐D geometrical model for a
2  2 HAP‐MIMO channel..…......……………………………………………................. 53

xv

24 The probability density for the distribution of the angle of arrival of the
scattered waves for different degree of scattering....…………………….……... 62
25 The probability density for the distribution of the distance between
TMS and the scatterers for different values of the parameter a ............. 63
26 The mean distance between the TMS and the scatterers for different
values of the parameter a ....…………………………………………………................ 64
27 The probability density for the distribution of the building heights in
London (U.K.) and Guildford (U.K.).....………………..………………….................. 65
28 3‐D building distribution in a typical urban propagation environment..... 67
29 The absolute correlation of HAP‐MIMO channels for various values of
the Ricean factor............….……………………………………………………................. 67
30 The transmit SCF of a 2  2 HAP‐MIMO Rayleigh channel for various
elevation angles of the SBS, in isotropic ( k  0 ) and non‐isotropic
( k  3 ) scattering environments…….……...………………................................. 68
31 The transmit (a) and receive (b) SCF of a 2  2 HAP‐MIMO Rayleigh
channel for different orientation of the SBS and TMS antenna array,
respectively..………………………………………….................................................. 70
channel for different degree of local scattering at the TMS………………..... 70
channel for different spread of the scatterers around the TMS……………… 70
channel for different maximum distance between the projections to
the x ‐y plane of the TMS and an effective scatterer................................ 71
35 The receive SCF of a 2  2 HAP‐MIMO Rayleigh channel with
horizontally (ψR  0 ) or vertically (ψR  90 ) placed TMS antennas, for
o o
various mean values of the scatterer height.......……………………….……....... 71
36 The SCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean channel……….. 72
37 The TCF of a HAP‐SISO Rayleigh channel for different different degree
of local scattering at the TMS..................…………………………………………....... 73
38 The TCF of a HAP‐SISO Rayleigh channel for different spread of the
scatterers around the TMS and different maximum distance between
the projections to the x‐y plane of TMS and an effective scatterer…........ 74
39 The TCF of a HAP‐SISO Rayleigh channel for various mean values of the
scatterer height.…...................................................................................... 74
40 The TCF of a HAP‐SISO Rayleigh channel for different maximum Doppler
frequency and different moving direction at the TMS…………………………… 75
41 The TCF of a HAP‐SISO Ricean channel for different HAP structures……... 75

xvi

42 The transmit STCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean
channel…………………………………………………………………………………………………. 76
43 The receive STCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean
channel……………………………………………………………..………………….………………… 76
44 Geometrical representation of roll, pitch, and yaw effects......…………...... 77
45 The transmit SCF of a 2  2 HAP‐MIMO Rayleigh channel for different
directions of HAP displacement due to stratospheric winds…………………… 78
46 The transmit SCF of a 2  2 HAP‐MIMO Rayleigh channel for different
elevation angle of the pth SBS antenna element (pitch effect)……............. 79
47 Comparison of the ergodic channel capacities obtained using a 2  2
HAP‐MIMO architecture and a HAP‐SISO architecture as a function of
the received SNR for different Ricean factor.........……………………………...... 83
48 Comparison of the ergodic channel capacities obtained using a 2  2
HAP‐MIMO architecture and a HAP‐SISO architecture as a function of
the Ricean factor for different received SNR……………………….…………........ 83
49 The ergodic capacity of a 2  2 HAP‐MIMO channel as a function of the
spacing between the SBS antenna elements for different SBS elevation
angle…………………………………………………………………......…………………….…….... 84
50 The complementary cumulative distribution function of the capacity of
2  2 HAP‐MIMO channel for different antenna array orientation at the
SBS………………………………………………………………............................................ 85
2  2 HAP‐MIMO channel for different degree of scattering, spread of
the scatterers around TMS, and maximum distance between the TMS
and the scatterers....…………….………….…..................................................... 85
2  2 HAP‐MIMO channel for horizontally or vertically placed TMS
antennas and different mean of the scatterer height............................... 86
2  2 HAP‐MIMO channel for different normalized Doppler frequency
and different moving direction at the TMS…………………............................ 87
54 The cumulative distribution of the angle of arrival of the scattered
waves for different degree of scattering.........……………………….................. 92
55 The cumulative distribution of the distance between TMS and the
scatterers for different values of the parameter a ………………….………….... 93
56 The cumulative distribution of the building heights in London (U.K.)
and Guildford (U.K.)……...…………………………………....................................... 94
57 Visualization of the three scatterer sub‐regions, each consisting of N1,
N2, and N3 coordinates (“  ”) of scattering points, respectively................ 95
xvii

58 The set partitioning of three scatterer constellations into V1, V2, and V3
sub‐constellations, each consisting of 3 coordinates (“  ”) of relevant
scattering points……………………………………………………………………………………. 99
59 Comparison between the TCFs of the reference model and the
deterministic simulation model…………………………………………………….......... 105
60 The absolute error between the SCFs of the reference model and the
deterministic simulation model...............................................……............ 106
deterministic SP simulation model.......…………………………………………..…..... 107
deterministic SP simulation model..................................................……..... 107
statistical simulation model......………………………………………………………....... 109
statistical simulation model…………………….………………………………………….... 109
65 Comparison of the ergodic capacities obtained using the reference
model and the simulation models.……………...…..............…….….................. 111
66 The LoS paths of the 3‐D geometrical wideband model for a 2  2
HAP‐MIMO channel………………………………..............…………………................. 117
67 The LoS paths of the projection of the 3‐D geometrical wideband
model for a 2  2 HAP‐MIMO channel…………..……..…………………............... 117
68 The NLoS paths of the 3‐D geometrical wideband model for a 2  2
HAP‐MIMO channel………………………………..............…………………................. 118
69 The NLoS paths of the projection of the 3‐D geometrical wideband
model for a 2  2 HAP‐MIMO channel………….…………………………............... 118
70 The relative maximum propagation delay for different maximum
scatterer height and different maximum distance in the azimuth
domain between the TMS and the scatterers……………………….…………………122
71 The FCF of a HAP‐SISO Rayleigh channel for different elevation angles
of the SBS……………………………………………………………………………………………… 134
72 The transmit SFCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean
channel…………………………………………………..............…………………................. 134
73 The receive SFCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean
channel…………………………………………………..............…………………................. 135
74 The TFCF of a HAP‐SISO (a) Rayleigh and (b) Ricean channel……………………135
75 The SDPS of a HAP‐SISO and a 2  2 HAP‐MIMO Rayleigh channel………... 136
76 The relative PSDS of a HAP‐SISO and a 2  2 HAP‐MIMO Rayleigh
channel……….............................................................................................. 137
77 Comparison between the STFCFs of the reference and the simulation
models………............................................................................................... 141
xviii

78 A geometrical model for a 2  2 LoS‐HAP‐MIMO channel………………..……. 148
79 Comparison of the channel ergodic capacities obtained using the
full‐rank and rank‐one 2  2 LoS‐HAP‐MIMO, LoS‐HAP‐SISO, and the
Rayleigh i.i.d. 2  2 MIMO architectures, under clear sky conditions......... 153
80 The complementary cumulative distribution function of the channel
capacity achieved with the full‐rank and rank‐one 2  2 LoS‐HAP‐MIMO
architectures for different rain rates…………………………………...…………….…… 154
81 The ergodic capacity obtained using the full‐rank 2  2 LoS‐HAP‐MIMO
architecture as a function of the rain rate for different carrier frequency…155
82 The normalized capacity obtained using the full‐rank 2  2
LoS‐HAP‐MIMO architecture as a function of the deviation factor……..…… 156
LoS‐HAP‐MIMO architecture for different orientation of the SBS
antenna array………………………………………………………………………………………… 157
LoS‐HAP‐MIMO architecture as a function of the antenna array shifting
from the optimal point…………………………………………………………………………… 158
xix

xx

ABBREVIATIONS
2‐D Two‐Dimensional
3‐D Three‐Dimensional
3G Third Generation
3GPP Third Generation Partnership Project
4G Fourth Generation
ΕΜΠ Εθνικό Μετσόβιο Πολυτεχνείο
ΕΤΣ Εργαστήριο Τηλεπικοινωνιακών Συστημάτων
ΜΔΕ Μεταπτυχιακό Δίπλωμα Ειδίκευσης
NTUA National Technical University of Athens
ΤΕ Τεχνολογική Εκπαίδευση
ΤΕΙ Τεχνολογικό Εκπαιδευτικό Ίδρυμα
ACE Antenna Centre of Excellence
AESP Antenna Elements Spacing Product
AFD Average Fade Duration
AoA Angle of Arrival
AoD Angle of Departure
AWGN Additive White Gaussian Noise
BAN Body Area Network
BER Bit Error Rate
BFWA Broadband FWA
CCDF Complementary Cumulative Distribution Function
CDF Cumulative Density Function
CNES Centre National d’Etudes Spatiales
COST European COoperation in Science and Technology
CSI Channel State Information
DAB Digital Audio Broadcasting
dB Decibel
DoF Degrees of Freedom
xxi

DSP Digital Signal Processor
DVB Digital Video Broadcasting
EHF Extra High Frequency
EM Electromagnetic
ESA European Space Agency
FCC Federal Communications Commission
FCF Frequency Correlation Function
FSL Free‐Space Loss
FT Fourier Transform
FWA Fixed Wireless Access
Gbps Gigabits per second
GBSB Geometry‐Based Single‐Bounce
HAP High‐Altitude Platform
HDTV High‐Definition Television
HSDPA High‐speed Downlink Packet Access
HSPA High‐Speed Packet Access
IEEE Institute of Electrical and Electronics Engineers
IFT Inverse Fourier Tranform
i.i.d. Independent and identically distributed
IMT International Mobile Telecommunications
IP Internet Protocol
IST Information Society Technologies
ITU International Telecommunication Union
Kbps Kilobits per second
LAN Local Area Network
LCR Level Crossing Rate
LMDS Local Multipoint Distribution Services
LMS Land Mobile Satellite
LoS Line‐of‐Sight
LPNM Lp ‐norm method
xxii

LTE Long Term Evolution
MAN Metropolitan Area Network
MBMS Multimedia Broadcast and Multicast Services
Mbps Megabits per second
MCM Management Committing Meeting
MD Maximal Deviation
MIMO Multiple‐Input Multiple‐Output
MISO Multiple‐Input Single‐Output
MU‐MIMO Multi‐user MIMO
NLoS Non‐Line‐of‐Sight
NTUA National Technical University of Athens
OECD Organization for Economic Cooperation and Development
OFDMA Orthogonal Frequency Division Multiple Access
PAN Personal Area Network
PC Personal Computer
PDA Personal Digital Assistant
PDF Probability density function
PSD Power Spectral Density
PSDS Power Space‐Delay Spectrum
QoS Quality of Service
RMSE Root Mean Square Error
SAA Spherical Antenna Array
SBS Stratospheric Base Station
SCF Spatial Correlation Function
SDMA Space‐Division Multiple Access
SDPS Space‐Doppler Power Spectrum
SFCF Space‐Frequency Correlation Function
SHF Super High Frequency
SIMO Single‐Input Multiple‐Output
SISO Single‐Input Single‐Output
xxiii

SNR Signal‐to‐Noise Ratio
SoS Sum‐of‐Sinusoids
SP Set Partitioning
STCF Space‐Time Correlation Function
STFCF Space‐Time‐Frequency Correlation Function
SU Subscriber Unit
SU‐MIMO Single‐user MIMO
TCF Temporal Correlation Function
TDL Tapped Delay Line
TFCF Time‐Frequency Correlation Function
TFS Terrestrial Fixed Station
TMS Terrestrial Mobile Station
TSL Telecommunications Systems Laboratory
UAV Unmanned Aerial Vehicle
UCA Uniform Circular Array
UHF Ultra High Frequency
ULA Uniform Linear Array
UMB Ultra Mobile Broadband
UMTS Universal Mobile Telephone System
UPA Uniform Planar Array
V‐MIMO Virtual MIMO
WAN Wide Area Network
WiMAX Worldwide Interoperability for Microwave Access
WRC World Radiocommunications Conference
WSSUS Wide‐Sense Stationary Uncorrelated Scattering
xxiv

Chapter 1
Chapter 1
Introduction

1.1 High‐Altitude Platforms for Wireless Communications
For several years, the wireless communications sector [1], [2] has definitely been
the fastest growing market segment in communications around the world with
dazzling new applications and products emerging on an almost daily basis. Wireless
communications have changed the way people communicate and have directly and
significantly influenced our everyday life by offering nomadicity, i.e., the ability to
connect to a network from different locations via different base stations,
portability, i.e., the ability to readily transport from one location to another, and
mobility, i.e., the ability to keep ongoing connections active while moving at
vehicular speeds. Strongly supported by the advances in semiconductor and signal
processing technologies, new communications architectures are being envisioned
to improve as well as complement current networks to meet future needs.
Wireless communications networks enable the provision of both mobile and
fixed services to serve a variety of commercial and business applications and
services. Nowadays, third generation (3G) mobile cellular systems [3]‐[5] have
enabled access to advanced services, such as video communications, multimedia,
and high‐speed internet access. Typical example of 3G standards is the Universal
Mobile Telephone System (UMTS) [3]‐[5], especially supplemented with High‐speed
Downlink Packet Access (HSDPA) [6], [7]. In addition, Fixed Wireless Access (FWA)
[8]‐[10] communications systems attempt to provide a set of reliable, cost‐effective
and secure communications services in case where existing wired infrastructure is
impractical, expensive or temporary. In general, FWA can serve two main
categories of services, Internet access for the public and businesses and private
high‐speed data communications for small, medium, and large businesses.
1

Chapter 1
Wireless communications services are traditionally provided by terrestrial and

satellite systems. Terrestrial links are widely used to provide services in areas with
complex propagation conditions, while satellite links are usually used to provide
high‐speed connections, where terrestrial infrastructure is not available. These
systems represent two well established infrastructures that have been dominant in
the telecommunications arena for years. In recent years, a new alternative wireless
communications technology has emerged known as high‐altitude platforms (HAPs)
[11]‐[17] and has attracted attention worldwide [18]‐[23]. The term HAPs defines
aerial platforms flying at an altitude between 17 and 22 km above the Earth’s
surface, in the stratosphere [11]‐[17].
Mesosphere
Stratosphere
HAP  50 km
 20 km
 35 km
Ozone Troposphere
Layer Airplane
Mt. Everest  17 km
 12 km
 9 km

Figure 1: The troposphere, stratosphere, and mesosphere layers of the atmosphere.
As shown in Fig. 1, the stratosphere is the second major layer of Earth's
atmosphere, just above the troposphere, and below the mesosphere. This
operating altitude was chosen because it represents a layer of relatively mild wind
and turbulence in most regions of the world. At this altitude, HAPs can succeed in
maintaining station‐keeping and flying against the wind without excessive power
demands. Fig. 2 illustrates the average wind velocities for different altitudes
obtained from worldwide measurements [12], [13]. Although this wind profile may
considerably vary with latitude, season and location, it generally follows this rough
distribution. The proposed range in altitude is above commercial air‐traffic heights
2

Chapter 1
(the maximum altitude of airplane lanes is about 12 km), which would otherwise
prove a potentially prohibitive constraint. However, as the altitude increases the air
density is reduced, rendering the placement of the HAPs very difficult. For example,
considering 12 km altitude, the density is about 25% compared to that at the sea
level, while at 24 km the density is only about 3.6% [12].
70
60
50
Wind velocity (m/s)
40
30
20
10
0
5 10 15 20 25 30
Altitude (km)

Figure 2: The average wind velocity for different altitude above the Earth’s surface.
1.1.1 Structures
Two major HAP structures are considered in the literature, the circling aircrafts
(approximately 30 m in length or with wingspan of about 35‐70 m) and the lighter‐
than‐air quasi‐stationary airships (of about 150‐200 m in length) [11]‐[17]. The
aircrafts fly in a roughly circular tight path (about 2 km radius or more) and can be
either solar‐powered and unmanned with continuous flight duration in the order of
months or manned with average flight duration of some hours due to the fuel
constraints and human factors. The unmanned aerial vehicles (UAVs) are another
type of small fuelled unmanned aircraft employed only for military short time
surveillance (up to 40 hours) at modest altitudes. The airships use very large semi‐
3

Chapter 1
rigid or non‐rigid helium‐filled containers and electric motors and propellers for
station‐keeping. Prime power required for propulsion and station‐keeping as well
as for the payload and applications is provided from lightweight solar cells in the
form of large flexible sheets. The achievable mission duration for airships is hoped
to be 5 years at least. HAP stability problems and displacements due to the winds
or pressure variations of the stratosphere are a problem to be faced for both
aircrafts and airships [24]‐[26]. Although, it is easier for airships than for aircrafts to
remain quasi‐stationary, it is rather difficult to remotely control the airship’s
position. Recently, advances in composite materials, computers, navigation
systems, aerodynamics, and propulsions systems, have made station keeping more
realizable and hence HAP‐based systems more achievable.

(a) (b)

(c) (d)
Figure 3: The HAP and UAV structures: (a) solar‐powered unmanned aircraft, (b) manned
aircraft, (c) unmanned aerial vehicle, and (d) solar‐powered unmanned airship.
4

Chapter 1
1.1.2 Advantages
HAPs preserve some of the best characteristics of terrestrial and satellite
communication systems, while avoiding many of their drawbacks [11]‐[17]. In
comparison to terrestrial wireless technologies, HAPs require considerably less
communications infrastructure and they can serve potentially large coverage areas.
When compared to satellite communication systems, HAPs provide lower
propagation delays (a serious issue for voice communications over satellite links)
and less free space loss, as well as easy maintenance and upgrading of the payload
during the lifetime of the platform. In addition, the cost for the development of
satellite systems is much greater (the eventual cost of HAPs is expected to be about
10% of that of a satellite) and it may be economically more efficient to cover a
large area with many HAPs rather than with many terrestrial base stations or with a
satellite system. Hence, HAPs represent an economically attractive technology.
HAPs are also well‐suited for temporary provision of basic or additional
communications services due to their rapid deployment and movable on demand,
providing network flexibility and adaptability. Finally, HAPs can use most of
conventional base station technology and terminal equipment, while solar‐
powered HAPs are non‐pollutant and environmentally friendly. Although HAPs
provide substantial advantages over terrestrial and satellite systems, their
successful deployment requires integration of available and emerging technologies
to make long‐term operation feasible and profitable. Specifically, some issues on
communications systems need to be addressed and are of fundamental
importance, such as the design and implementation of the onboard antennas, the
channel characterization and modeling, the resource management, and the
coordination and inter‐operability between different systems.
1.1.3 Spectrum Allocation and Licensed Frequency Bands
Radio transmission involves the use of part of the electromagnetic spectrum and
different parts of the radio spectrum are used for different radio transmission
5

Chapter 1
technologies, services and applications. The tasks of internationally coordinating

the use of radio spectrum and setting global standards are undertaken by the
International Telecommunication Union (ITU) [27]. The ITU divided the world into
three regions for the purposes of managing the global radio spectrum and each
region has its own set of frequency allocations. As shown in Fig. 4, Region 1
comprises Europe, Africa, Russia and Middle East, Region 2 comprises North and
South America, and Region 3 comprises Asia, Oceania and Pacific Countries.
Figure 4: The ITU region map.
According to Table 1, the radio frequency spectrum involves electromagnetic
(EM) radiation with frequencies between 3 Hz and 300 GHz [28]. As the demand for
new and different communication services increases, more radio spectrum space at
higher frequencies is required. Hence, the utilization of the Ultra High Frequency
(UHF), Super High Frequency (SHF), and Extra High Frequency (EHF) bands has
emerged. The common microwave band designations are summarized in Table 2.
The properties of the EM energy depend on the frequency. Hence, different
frequencies have different properties, and propagation at high frequencies is
dominated by different mechanisms as compared to propagation at low
frequencies (these mechanisms will be analytically studied in the next chapter). In
particular, high rain attenuation and atmospheric losses at millimeter‐wave
6

Chapter 1
frequencies make communications difficult and unreliable [29]. Therefore,

frequency bands less susceptible to path loss, absorption and noise should be
chosen to minimize the cost of the transmission and maximize its information
carrying rate.
TABLE 1
RADIO FREQUENCY BANDS
BAND NOMENCLATURE FREQUENCY WAVELENGTH

ELF Extra low frequency 3 – 30 Hz 100,000 – 10,000 km

SLF Super low frequency 30 – 300 Hz 10,000 – 1,000 km
ULF Ultra low frequency 300 – 3,000 Hz 1,000 – 100 km
VLF Very low frequency 3 – 30 KHz 100 – 10 km
LF Low frequency 30 – 300 KHz 10 – 1 km
MF Medium frequency 300 – 3,000 KHz 1000 – 100 m
HF High frequency 3 – 30 MHz 100 – 10 m
VHF Very high frequency 30 – 300 MHz 10 – 1 m
UHF Ultra high frequency 300 – 3,000 MHz 1 – 0.1 m
SHF Super high frequency 3 – 30 GHz 10 – 1 cm
EHF Extra high frequency 30 – 300 GHz 10 – 1 mm

TABLE 2
MICROWAVE FREQUENCY BANDS
BAND NAME FREQUENCY WAVELENGTH
L‐band 1 – 2 GHz 30 – 15 cm
S‐band 2 – 4 GHz 15 – 7.5 cm
C‐band 4 – 8 GHz 7.5 – 3.75 cm
X‐band 8 – 12 GHz 3.75 – 2.5 cm
Ku‐band 12 – 18 GHz 2.5 – 1.67 cm
K‐band 18 – 27 GHz 1.67 – 1.11 cm
Ka‐band 27 – 40 GHz 11.1 – 7.5 mm
V‐band 40 – 75 GHz 7.5 – 4 mm
W‐band 75 – 110 GHz 4 – 2.7 mm
7

Chapter 1
TABLE 3
FREQUENCY SPECTRUM AVAILABLE FOR HAP‐BASED APPLICATIONS
FREQUENCY MICROWAVE AREA/ TYPE OF

BAND FREQUENCY BAND COUNTRY SERVICE
47.9 – 48.2 GHz V Global Fixed (uplink)
47.2 – 47.5 GHz V Global Fixed (downlink)
31.0 – 31.3 GHz Ka 40 countries Fixed (uplink)
27.5 – 28.35 GHz Ka 40 countries Fixed (downlink)
2.160 – 2.170 GHz S Regions 1 and 3 Mobile
2.110 – 2.160 GHz S Global Mobile
2.010 – 2.025 GHz S Regions 1 and 3 Mobile
1.885 – 1.985 GHz L Global Mobile

Several frequency bands (see Table 3) have been licensed for communications
through HAPs on a global, regional or national basis and for each frequency band
appropriate regulatory provisions have been established based on technical and
operational studies [30]. The choice of the frequency bands at which HAPs can
operate is mainly determined by the frequency sharing and compatibility with
other existing services provided by terrestrial/satellite systems, which would cause
interference problems if spectrum allocation is not carefully controlled [31], [32].
At the ITU World Radiocommunications Conference in 1997 (WRC‐97) the
concept of HAPs was formally introduced into the radio regulations as a separate
category of the communications infrastructures and was decided that HAPs can
provide fixed wireless services. Specifically, the bands 47.2-47.5 GHz (downlink)
and 47.9-48.2 (uplink) were regulated for use by HAPs on a global basis. Use of
these bands is subject to the provisions of Resolution 122, which was revised at
WRC‐07 [33]. At WRC‐00, a need for a lower frequency band for fixed wireless
services was expressed due to the severe rain attenuation occurring at 47/48 GHz
bands. Hence, the bands 27.5-28.35 GHz (downlink) and the band 31.0-31.3 GHz
(uplink) were regulated for use by HAPs in several countries located mainly at
Region 3. The WRC‐03 increased the number of countries with access to the HAPs
allocations including Region 2. Use of these bands is subject to the provisions of
Resolution 145, which was revised at WRC‐07 [34]. Hence, HAPs can serve as base
8

Chapter 1
stations of FWA networks at 47/48 and 28/31 GHz bands providing data, voice, and
video services with a total bandwidth of up to 300 MHz in each direction shared on
a non‐harmful interference, non‐protection basis with fixed satellite and terrestrial
services.
The WRC‐00 also decided that HAPs may be used as base stations within the
terrestrial component of the International Mobile Telecommunications 2000 (IMT‐
2000) system. This alternative way of delivering the IMT-2000 services by HAPs on
a co‐primary basis was regulated in the frequency ranges 1885-1980, 2010-2025
and 2110-2170 MHz in Region 1 and Region 3, and 1885‐1980 and 2110‐2160 MHz
in Region 2. Such use of the bands above is subject to the provisions of Resolution
221, which was revised at WRC‐07 [35]. Hence, HAPs can serve as base stations of
3G cellular networks providing mobile data/voice/video services with a total
bandwidth of up to 50/60 MHz (amount of bandwidth varies slightly with region)
worldwide.
Finally, spectrum identification in the range between 5,850-7,075 MHz to
facilitate gateway links to and from HAPs has been also envisioned, since WRC‐07
agenda, Resolution 734 [36].
1.1.4 Applications and Services
An application can be considered as the end product offered to public and
businesses over which services are accessed. The major wireless communications
applications intend to be facilitated by HAPs can be divided technologically into
two types, the broadband FWA (BFWA) [12], [15], [19], [37] and the provision of
mobile services [38]‐[40]. These types of applications are briefly described in the
following subsections. Note that the flexibility of HAP‐based communications
systems allows not only for carrying wireless communications payloads but also for
serving other applications, such as navigation and positioning [41], monitoring,
remote sensing, and surveillance [42], disaster management and relief [43], [44],
telemedicine [45], and military applications [46].
9

Chapter 1
1.1.4.1 Broadband Fixed Wireless Access
As mentioned above, HAPs have already been accepted by the ITU as alternative
base stations of serving BFWA at the 47/48 GHz (V band) and 28/31 GHz (Ka band).
BFWA communications provide in general high rate of data transmission. The ITU
has defined broadband as a transmission capacity that is faster than 1.5‐2 megabits
per second (Mbps) [47], while the Federal Communications Commission (FCC)
definition of broadband is 768 kilobits per second (Kbps). The Organization for
Economic Cooperation and Development (OECD) has defined broadband as 256
Kbps in at least one direction and this bit rate is the most common baseline that is
marketed as "broadband" around the world. High‐speed broadband services
include but are not limited to conversational services (voice/video telephony, high
resolution video conferencing), streaming and broadcast services (real‐time radio
and television, video on demand), local multipoint distribution services (LMDS),
interactive data access (broadband internet access, web browsing), and large file
transfers. In addition, HAPs could potentially be used as an alternative solution for
the provision of Digital Video Broadcasting (DVB) and Digital Audio Broadcasting
(DAB). A feasibility study of HAPs‐DVB/DAB has been conducted by European Space
Agency (ESA) under STRATOS project [22]. Although these services are offered
mainly to fixed terminals, their application could be extended to mobile scenarios
in order to benefit from existing air interfaces and accelerate the development of
new applications, such as the provision of high‐speed Internet access, audio and
video on demand and file transfer to cars, trains, airplanes, boats, etc. [48].
1.1.4.2 High‐Speed Mobile Services
In addition to BFWA, ITU has also endorsed the use of HAPs in the IMT‐2000
spectrum for the provision of 3G/UMTS mobile services, as well as enhanced
HSDPA services offering data transfer speeds up to 14 Mbps on the downlink.
These services are expected to have the same functionality and meeting the same
service and operational requirements as traditional terrestrial tower‐based
10

Chapter 1
systems. HAPs can be designed to serve as the sole station in a stand‐alone
infrastructure (essentially, replacing the tower base station network with a “base
station network in the sky”) [38]‐[40]. In particular, a single HAP can replace a large
number of terrestrial base stations and their backhaul infrastructure (microwave or
optical links) and can serve a large city, a suburban area, a rural region [49], or even
a whole state. Moreover, HAPs could be able to deliver Worldwide Interoperability
for Microwave Access (WiMAX) [50], [51] mobile services based on the IEEE
802.16e standard (the most popular implementation of the IEEE 802.16 standard)
at frequencies between 2‐6 GHz [52]‐[53]. WiMAX adopts Orthogonal Frequency
Division Multiple Access (OFDMA) and enables wireless transmission of data using
a variety of transmission modes at speeds up to 40 Mbps [52]‐[53].
1.2 Future Trends in Wireless Communications
The growing demand for greater bandwidth, higher data rates, improved quality
of service (QoS) and ubiquitous access have prompted the development of next‐
generation wireless communications systems, which can by far surpass the
capabilities of existing systems and attempt to bring broadband applications to
new experience scenarios. Fourth generation (4G) communications systems will
employ a comprehensive and secure all‐Internet‐Protocol (IP) network and will
support fixed and mobile ultra‐broadband (gigabit speed) access and multi‐carrier
transmission [54]‐[57]. New broadband applications are expected to thrive with the
advent of 4G, such as multimedia broadcast and multicast services (MBMS), IP
telephony, high‐quality voice, high‐definition television (HDTV), mobile television,
ultra‐broadband Internet access, gaming services, and high‐quality streamed
multimedia. The available data rates of 3G/HSDPA cellular networks will further be
increased with the Evolved High‐Speed Packet Access (HSPA+) technology [7], [8],
which intends to provide theoretical data rates up to 84 Mbps on the downlink. The
high performance Third Generation Partnership Project (3GPP) Long Term
Evolution (LTE) technology [58], [59] will bring 3G cellular networks a step toward
11

Chapter 1
to the 4G providing data rates of at least 320 Mbps on the downlink, while LTE
Advanced technology will fully support the 4G requirements offering data rates up
to 1 gigabit per second (Gbps) [60]. Moreover, next generation Mobile WiMAX
based on the IEEE 802.16m [61] standard will support at least 100 Mbps at mobile
and 1 Gbps at fixed‐nomadic services, and hence will be compliant with 4G next
generation mobile networks.

Figure 5: Visualization of an integrated satellite‐HAP‐terrestrial 4G wireless
communications system providing fixed and mobile broadband services.
Indeed, the term 4G is widely used to include several types of broadband fixed
and mobile wireless communications systems. As shown in Figs. 5 and 6, 4G
systems will seamlessly integrate the entire communications infrastructure
(terrestrial base stations, satellites, and HAPs), the modern communications
terminals (mobile phones, personal digital assistants (PDAs), tablet computers,
laptops, etc.), the transportation vehicles (cars, trains, airplanes, boats, etc.), the
12

Chapter 1
available networks (body area networks‐BANs, personal area networks‐PANs, local
area networks‐LANs, metropolitan area networks‐MANs, and wide area networks‐
WANs), and the next generation applications (as mentioned above) to satisfy
increasing user demands [54]‐[57]. Hence, complex network topologies of
heterogeneous nature will be deployed. Each network has different capabilities, in
terms of capacity, coverage, mobility support, cost, and is best suited for handling
certain situations. However, these different networks will have to co‐exist, and be
complementarily and efficiently exploited. This can be achieved by introducing the
reconfigurability concept [62], [63], which offers service differentiation,
customization, and personalization according to the environment requirements. In
particular, reconfigurability dynamically tackles the changeable behavior of
wireless networks and associated equipment, specifically in the fields of radio
spectrum, radio access technologies, protocol stacks, and application services.
4G as the integrator
Access
Mobile Points
Terminals
and HSPA+
Vehicles WiMAX
Mobile
LTE
Accesses
Mobile phones
PDAs
Laptops
Vehicles Terrestrial
Trains Base Stations
Airplanes IP Backbone
Ships Applications
and
Air Services
Interfaces
HAPs Network
Users LMDS
Homes HDTV
Public INTERNET
Businesses DVB/DAB
Fixed
Backbone
Fixed Satellites
Terminals
Reconfigurable Radio
Figure 6: High level view of the future 4G wireless reconfigurable radio systems.
13

Chapter 1
1.3 Thesis Motivation and Objectives
As new requirements for access to wireless networks are emerged within the
communications society, HAPs can play an important role in the evolution of current
and future communications systems [64]‐[67]. In particular, HAPs are expected to
alternatively or complementary fulfill the vision of optimal connectivity anywhere
providing high data throughputs in virtually every possible scenario at low cost.
However, wireless systems designers are facing a number of challenges. These
include the limited availability of the radio frequency spectrum and a complex
space‐time varying wireless environment [68]. In addition, next generation wireless
communications services will require increased network coverage, higher data
rates, enhanced capacity, enriched QoS, and improved efficiency. Thus, new
technologies have to continuously be under development in order to meet these
growing demands.
In recent years, the multiple‐input multiple‐output (MIMO) technology has

emerged as the most promising technology in these measures due to its powerful
performance‐enhancing capabilities [69]. The MIMO communications systems can
be defined intuitively by considering that multiple antennas are used at both the
transmitting and the receiving end. The core idea behind MIMO is the use of space‐
time processing in which time (the natural dimension of digital communication
signals) is complemented with the spatial dimension inherent in the use of multiple
spatially distributed antennas. Then, the signals sampled in the spatial domain at
both ends are combined is such a way that they either create effective multiple
parallel spatial data pipes (therefore increasing the data rate), and/or add diversity
to improve the quality, i.e., the bit‐error‐rate (BER) of the communication link. This
can in turn significantly improve spectral efficiency and enduringly preserve the
QoS for the range of services offered. Different aspects of MIMO technology are
being planned or have already been incorporated in wireless standards, such as the
IEEE 802.11n, 802.16e, 802.16m, 802.20, 802.22, 3GPP Releases 7, 8, and 9, 3GPP2
ultra mobile broadband (UMB), DVB‐T2 among others. Thus, MIMO technology is
14

Chapter 1
seen as an ideal and critical technology of physical layer for next generation
wireless communications networks.
This thesis investigates the application of MIMO techniques to HAP‐based
communications systems. It is well known that before the introduction of a new
technology into a communications system, a large number of theoretical and
experimental investigations have to be made. In particular, the successful design
and deployment of HAP‐MIMO communications systems requires a detailed
knowledge of the wireless radio propagation channel involved in communicating
between two ends of the link. Channel knowledge is required to understand
various impairments, such as the attenuation, the multipath fading [70], and the
space‐time variations of the channel [68]. Using this knowledge, communications
systems can be designed to obtain optimal or near optimal performance. This
philosophy has been the driving force behind the research activity on all types of
wireless communications systems. Traditional research topics include the following
steps:
 Developing reference (theoretical) non‐realizable channel models.

 Conducting real‐world channel measurements and deriving empirical

channel models from these measurements.

 Developing accurate, efficient and flexible simulation models for

computer/software simulation of communications systems.

 Developing hardware‐based channel emulators for real‐time system

analysis and performance testing.
This piece of research was driven to understand, characterize and model the
HAP‐MIMO channel, verify the applicability of MIMO techniques to HAP‐based
systems, and quantify the available MIMO gain. The approach used in this research
followed a serial path of objectives. To enable successful design of HAP‐MIMO
15

Chapter 1
systems, this research initially focuses on the study and characterization of HAP‐
MIMO multipath fading channels at L and S bands. The first and second objectives
of this thesis are to develop three‐dimensional (3‐D) theoretical models for HAP‐
MIMO narrowband and wideband mobile fading channels, respectively. Moreover,
the third objective of this thesis is to develop 3‐D simulation models for HAP‐MIMO
narrowband and wideband mobile fading channels. Except for delivering future
mobile services, HAPs can also provide broadband fixed and mobile services at Ka‐
and V‐ bands, where the rain has a significant effect on the quality of the link.
Hence, the fourth objective of this thesis is to design HAP‐MIMO channel models
for BFWA services, under clear sky and rain conditions.
1.4 Research Contributions and Author’s Publications
The major achievements and novelty in this thesis are:
1. Design and development of a novel 3‐D parametric reference model for
narrowband HAP‐MIMO channels and analysis of space‐time correlation and
channel capacity.

2. Design and development of a novel 3‐D parametric reference model for
wideband HAP‐MIMO channels and analysis of statistical properties.

3. Design and development of novel 3‐D deterministic simulation models for
narrowband and wideband HAP‐MIMO mobile fading channels.

4. Design and development of novel 3‐D statistical simulation models for
narrowband and wideband HAP‐MIMO mobile fading channels.

5. Design and development of a novel 3‐D channel model for fixed HAP‐MIMO
communications in mm‐wave frequencies.

Part of this thesis resulted in publications in several international journals and
conference proceedings (including a Best Paper Award), and in contribution to the
European COoperation in Science and Technology (COST) Action IC0802 entitled
16

Chapter 1
“Propagation Tools and Data for Integrated Telecommunication, Navigation, and

Earth Observation Systems”.
 Journal Publications
[1] E. T. Michailidis and A. G. Kanatas, “Three‐Dimensional HAP‐MIMO Channels:
Modeling and Analysis of Space‐Time Correlation,” IEEE Transactions on Vehicular
Technology, vol. 59, no. 5, pp. 2232‐2242, Jun. 2010.

[2] P.‐D. Arapoglou, E. T. Michailidis, A. D. Panagopoulos, A. G. Kanatas, and R. Prieto‐
Cerdeira, “The Land Mobile Earth‐Space Channel: SISO to MIMO Modeling from L‐ to
Ka‐Bands,” IEEE Vehicular Technology Magazine, Special Issue on Trends in Mobile
Radio Channels: Modeling, Analysis, and Simulation, vol. 6, no. 2, pp. 44‐53, Jun. 2011.

[3] E. T. Michailidis and A. G. Kanatas, “Statistical Simulation Modeling of 3‐D HAP‐MIMO
Channels,” accepted for publication in Wireless Personal Communications.

[4] E. T. Michailidis and A. G. Kanatas, “Capacity Analysis and Simulation of 3‐D Space‐
Time Correlated HAP‐MIMO Channels,” accepted for publication in International
Journal on Advances in Telecommunications. Invited paper

[5] E. T. Michailidis and A. G. Kanatas, “Wideband HAP‐MIMO Channels: A 3‐D Modeling
Approach,” submitted to IEEE Transactions on Wireless Communications.

 Conference Publications
[1] E. T. Michailidis and A. G. Kanatas, “A Three Dimensional Model for Land Mobile‐HAP‐
MIMO Fading Channels,” in Proc. 10th International Workshop on Signal Processing
for Space Communications (SPSC) 2008, pp. 1‐6, Rhodes, Greece, 6‐8 Oct. 2008.

[2] E. T. Michailidis, G. Efthymoglou, and A. G. Kanatas, “Spatially Correlated 3‐D HAP‐
MIMO Fading Channels,” in Proc. International Workshop on Aerial & Space Platforms:
Research, Applications, Vision of IEEE Global Communications Conference (Globecom)
2008, pp. 1‐7, New Orleans, LA, USA, 4 Dec. 2008.

[3] E. T. Michailidis and A. G. Kanatas, “Capacity Optimized Line‐of‐Sight HAP‐MIMO
Channels for Fixed Wireless Access,” in Proc. International Workshop on Satellite and
Space Communications (IWSSC) 2009, pp. 73‐77, Siena‐Tuscany, Italy, 10‐11 Sep. 2009.

[4] E. T. Michailidis and A. G. Kanatas, “On the Capacity of 3‐D Space‐Time Correlated
HAP‐MIMO Channels,” in Proc. 2nd International Conference on Advances in Satellite
and Space Communications (SPACOMM) 2010, Athens‐Glyfada, Greece, 13‐19 Jun.
2010. Best Paper Award

[5] Ε. T. Michailidis and A. G. Kanatas, “Modeling and Simulation of 3‐D Wideband HAP‐
MIMO Channels,” accepted for publication in IEEE 74th Vehicular Technology
Conference (VTC Fall) 2011, San Francisco, USA, 5‐8 Sep. 2011.
17

Chapter 1
 Contribution to COST Action IC0802
[1] E. T. Michailidis and A. G. Kanatas, “Three‐dimensional HAP MIMO Channels:
Modeling and Analysis of Spatial Correlation,” COST Action IC0802, 1st MCM,
ESA/ESTEC, Noordwijk, Netherlands, 9‐11 Mar. 2009.

[2] E. T. Michailidis and A. G. Kanatas, “Effect of 3‐D Spatial Correlation on the Capacity of
HAP‐MIMO Channels,” COST Action IC0802, 3rd MCM, NTUA, Athens, Greece, 26‐28
Apr. 2010.

[3] E. T. Michailidis and A. G. Kanatas, “A Three‐Dimensional Statistical Simulation Model
for HAP‐MIMO Channels,” COST Action IC0802, 3rd MCM, NTUA, Athens, Greece, 26‐
28 Apr. 2010.
1.5 Thesis Outline
The remainder of this thesis is organized as follows. Chapter 2 thoroughly
presents the basic concepts of radio propagation, describes the characteristics of
MIMO technology, and surveys existing models for earth‐space radio channels.
Chapter 3 proposes a 3‐D reference model for narrowband HAP‐MIMO

channels, studies the channel statistics, defines and investigates the HAP‐MIMO
channel capacity, and provides numerical results.
Chapter 4 proposes 3‐D deterministic and statistical simulation models for HAP‐
MIMO mobile fading channels and provides simulation results to validate the
theoretical results of the reference model.
Chapter 5 proposes a 3‐D reference model for wideband HAP‐MIMO channels,
studies its statistical properties, and provides numerical results. This chapter also
proposes simulation models for wideband HAP‐MIMO channels and provides
simulation results to validate the reference model.
Chapter 6 proposes a HAP‐MIMO channel model for BFWA services at mm‐wave
frequencies, suggests capacity optimization techniques, and provides numerical
results to demonstrate the theoretical derivations.
Finally, Chapter 7 concludes this thesis and addresses open research areas in
MIMO channel modeling for wireless communications systems based on HAPs.
18

Chapter 2
Chapter 2
Overview on Propagation Mechanisms,
MIMO Principle, and Channel Modeling

2.1 Introduction
The analysis, the design, the operation, and the performance of a wireless
communications network are heavily dependent on the characteristics of the
wireless propagation channel. Indeed, communicating through a wireless channel is
a challenging task because the medium may introduce severe impairments [2],
[71]‐[74]. Hence, an accurate and thorough characterization of the propagation
effects influencing radio waves propagating in a wireless channel is essential, in
order to design and implement reliable and efficient HAP‐MIMO channel models.
These propagation effects vary depending on the operating frequency. Thus, it is
important to emphasize considerable differences between the communications
systems operating at low frequency bands and the ones operating in upper bands.
With these in mind, this chapter aims to describe in detail the fundamental
propagation mechanisms that influence the links between HAPs and terrestrial
fixed/mobile stations at the licensed L‐, S‐, Ka‐, and V‐ frequency bands. Moreover,
this chapter describes the MIMO principle, reviews existing models for earth‐space
radio channels, and provides an introduction to HAP‐MIMO channel modeling.
2.2 The Wireless Propagation Channel
The propagation characteristics crucially determine the performance of any

potentially adopted MIMO technique. This section reviews the most important
channel impairments. Particular emphasis is given to issues related to the rain
effects and the multipath propagation, which dramatically affect MIMO
19

Chapter 2
performance and control MIMO applicability to fixed and mobile HAP‐based
systems at frequencies well above 10 GHz and well below 10 GHz, respectively.
The architecture of a generic wireless communications system is illustrated in
Fig. 7. Based on this architecture, the transmitter sends a signal through a wireless
communications channel, the channel itself distorts the signal in such ways that
may be unpredictable to the receiver, and the receiver must be designed to
overcome these modifications to deliver the signal to its final destination with as
few distortions as possible. Specifically, in the wireless channel, there are
multiplicative and additive effects [2], [71], [72].
The multiplicative effects arise from the various processes encountered by the
transmitted waves on their way from the transmit end to the receive end. It is
conventional to further subdivide the multiplicative processes in the channel into
two types of fading, the large‐scale fading and the small‐scale fading. The large‐
scale fading is due to the shadowing of the transmitted signal by objects and
scatterers in the channel and is referred to as shadow fading, while the small‐scale
fading is caused by the multipath propagation and is referred to as multipath fading
[2], [71], [72]. Shadow fading and multipath fading will be discussed later in this
section.
On the contrary, the additive effects arise from the noise generated within the
receiver itself (although external noise contributions may also be significant) and
their impact on the performance of a communications system is important, since
they determine the received signal‐to‐noise ratio (SNR). The additive white
Gaussian noise (AWGN) is a good example of additive impairments and
corresponds to a random process with each realization modeled as a random
variable following a Gaussian distribution. The average received SNR is expressed as
SNR  PR  dB   PN  dB  . (1)

The received signal power PR is given by

PR  dBW   PT  dBW   GT  dBi  GR  dBi   L  dB  , (2)

20

Chapter 2
where PT is the transmitted power, GT and GR are the transmit and receive
antenna gains, respectively, and L is the path loss. The noise power PN is given by
PN  dBW   10log  N0B   10log  kTeB  , (3)

where N0 is the noise power spectral density (PSD), B is the channel bandwidth, k
is the Boltzmann’s constant ( 1.38  10 23 Joules/Kelvin ), and Te is the system noise
temperature.
The Wireless Channel
Source Transmitter x x + Receiver Destination
Large Small
Additive
Scale Scale
Noise
Fading Fading

Figure 7: Architecture of a generic wireless communications system.
2.2.1 Path Loss
The path loss (or path attenuation) in any wireless communications radio link is
an important effect that contributes to signal degradation by reducing its power.
The value of the path loss is usually expressed in decibels (dB) and is highly
dependent on many factors, such as the terrain contours, the propagation
environment (urban or rural, vegetation and foliage), the propagation medium (dry
or moist air), the distance between the transmitter and the receiver, and the height
and location of antennas. In the study of wireless communications, path loss is
usually represented by the path loss exponent [2]. Path loss models for predicting
path loss in urban environments are described in [75], [76].
For a link between HAPs and terrestrial stations, the minimum (reference) path
loss is given by the free‐space loss (FSL) LF , which assumes a line‐of‐sight (LoS) link
21

Chapter 2
between the transmitter and the receiver and propagation in free‐space and is
given by [11], [37]
LF  dB   32.4  20log dkm  20log fMHz , (4)

where fMHz is the carrier frequency in MHz, and dkm is the distance between the
transmitter and the receiver in km. Nevertheless, the signals do not experience
free‐space propagation due to other sources of loss, such as obstacles in the first
Fresnel zone, rain attenuation [11], [37], [77]‐[79], gaseous absorption (oxygen
absorption and water vapour) [78], cloud attenuation, and attenuation due to
vegetation [80], [81]. The summation of any losses caused by the aforementioned
propagation effects is considered as the excess loss Lex . Hence, the total loss is
given by
L  dB   LF  dB   Lex  dB  . (5)

Note that rain attenuation is the dominant mechanism that significantly affects the
quality of the link at frequencies above 10 GHz [11].
2.2.2 Rain Effects
The rain is confined to the first 2.5‐5 km of the atmosphere depending on the
latitude [11]. Hence, an EM wave propagating in the troposphere is directly
affected by rain effects (see Fig. 8). Rain effects primarily refer to the attenuation
of a signal. The troposphere consists of a mixture of particles having a wide range
of sizes and characteristics. The attenuation is the result of the conversion of EM
energy to thermal energy within an attenuating particle. This attenuation increases
with the number of raindrops along the path, the size of the drops, the length of
the path through the rain, and the carrier frequency. The main particles of interest
are hydrometeors, including raindrops, fog and clouds. Note that attenuation is
negligible for snow or ice crystals, in which the particles are tightly bound and do
not interact with the waves.
22

Chapter 2
The rain attenuation Lr can be empirically obtained using the specific rain
attenuation  r (dB/km) [82], which is defined as

 r  ar Rrbr , (6)
where Rr is the rain rate measured on the ground in millimeters per hour (mm/h)
and is strongly dependent on the geographical location. Worldwide rain rate
contour maps can be found in [83]. Typical rain rate values for Europe are around
30 mm/h, while for some Mediterranean regions the rain rate exceeds 50 mm/h
and for equatorial regions the rain rate may reach 150 mm/h. The rain rate
corresponds to the measure of the average size of the raindrops. However, the
period of time for which the rain rate exceeds a certain value is more important
than the total amount of rain falling during a year. The values of the empirical
regression coefficients ar and br can be obtained from [82] and depend on the
climatic zone, the transmission frequency, and the polarization.
Table 4 shows typical values for ar and br at various frequencies for horizontal
and vertical polarization extracted from [82], while Fig. 8 utilizes (6) and
demonstrates the specific rain attenuation as a function of the rain rate for
different carrier frequency. One observes that the rain attenuation is only
significant to communications systems operating above 10 GHz. At these
frequencies, the wavelength and raindrop size (about 1.5 millimeters) are
comparable and the attenuation is quite large. Thus, HAP‐based communications
systems operating at Ka‐ and V‐ frequency bands are susceptible to rain, while the
ones operating at L‐ and S‐ frequency bands are not significantly affected by rain.
The total rain attenuation Lr can be obtained as follows
Lr   r dr , (7)
where dr is the total rainy path length and can be geometrically obtained as
follows (see Fig. 9)
dr   Hr  HR  / sin T , (8)

23

Chapter 2
where Hr is the effective rain height, HR is the height of the terrestrial station, and
T is the elevation angle of the platform. Representative values for Hr vary
according to the latitude of the terrestrial station [11], [37]. In practice, high rain
attenuations are sometimes avoided by using site diversity [84], [85], in which two
widely separated terrestrial stations are used. Then, the probability that both
terrestrial stations are within the same area of rain concentration is small.
Alternatively, a portion of spectrum in a lower frequency band may be used.
TABLE 4
REGRESSION COEFFICIENTS FOR ESTIMATING SPECIFIC RAIN ATTENUATION

HORIZONTAL VERTICAL
FREQUENCY POLARIZATION POLARIZATION
ar br ar br
2 GHz 0.000154 0.963 0.000138 0.923
10 GHz 0.0101 1.276 0.00887 1.264
20 GHz 0.0751 1.099 0.0691 1.065
30 GHz 0.187 1.021 0.167 1.000

40 GHz 0.350 0.939 0.310 0.929
50 GHz 0.536 0.873 0.479 0.868

50
2 GHz
10 GHz
40 20 GHz
30 GHz
Attenuation (dB/km)
40 GHz
30
50 GHz
20
10
0
0 50 100 150
Rain Rate (mm/h)

Figure 8: The specific rain attenuation as a function of the rain rate for different carrier
frequency.
24

Chapter 2
HAP dr
T Hr
HR Terrestrial
Station

Figure 9: Rain attenuation path geometry.
The aforementioned empirical modeling of the long‐term behavior of the rain
attenuation effects does not convey any information on the time dynamics of these
effects. An alternative and more accurate way of modeling the rain attenuation is
through the use of time‐series [86]. Since rain is characterized by a significant
spatial inhomogeneity within the distances of interest, rain attenuation time‐series
can be obtained using actual meteorological data accumulated over many years of
carefully performed propagation measurements. The possibility of making a
computer synthesis of these time‐series allows for the implementation of statistical
channel models, which are characterized by short‐term statistics.
Apart from introducing signal attenuation, rain also introduces scattering

processes [87], [88]. This scattering results from redirection of the radio waves in
various directions, so that only a fraction of the incident energy is transmitted
onwards in the direction of the receiver. The scattering process is strongly
frequency‐dependent, since wavelengths which are long compared to the particle
size will be only weakly scattered.
2.2.3 Shadow Fading
The predicted path loss for a system operated in a particular environment is
constant for a given distance. However, the emitted EM waves often do not reach
the receive antenna directly due to obstacles that block the LoS link resulting in
25

Chapter 2
random variations of the received power with respect to the nominal value given
by the path loss models. This effect is called shadowing or slow (long term) fading
and controls the reliability of coverage and the link budget [2], [71]. The density of
the obstacles significantly depends on the physical environment. The randomness
in this environment is captured by modeling the density of obstacles and their
absorption behavior as random numbers. The amplitude variations caused by
shadowing is often modeled using a log‐normal distribution with a standard
deviation according to the log‐distance path loss model. The interested reader is
referred to [89], [90] for a detailed discussion of the shadow fading characteristics
for HAP‐based systems. This thesis focuses only on the multipath fading, since this
is the dominant fading mechanism that affects the performance of MIMO systems
[68], [69].
2.2.4 Multipath Fading
Although LoS conditions are required in the Ka‐ and V‐ frequency bands [11],
[37] due to the severe attenuation of possible obstructed signal components, non‐
line‐of‐sight (NLoS) connections can be used and should also be evaluated for HAP‐
based communication systems operating in L‐ and S‐ frequency bands. Then, it is
considered that a signal propagating through a wireless channel usually arrives at
its destination along a number of different paths, referred to as multipath
components, which arise from single‐bounce and/or multiple‐bounce scattering,
reflection or diffraction of the radiated energy by fixed and/or mobile intervening
objects in the environment [2], [71], [91]. These mechanisms affect the amplitude,
the direction, and the phase of the propagating radio waves and this effect is
known as multipath fading or fast (short term) fading owing to the random rapid
fluctuations of the transmitted wave over small areas. Indeed, these fluctuations
result in the spreading of the transmitted wave in Doppler and/or time delay
and/or angular domain. The received wave is a superposition of the impinging
multipath components. Depending on the phase of each partial wave, this
superposition can be constructive or destructive. The phase relationships depend
26

Chapter 2
on the speed of motion, frequency of transmission and relative path lengths. Fig.
10 presents a typical scenario affected by multipath fading. One observes that the
LoS component reaches the receiver directly. However, single‐bounce and double
bounce NLoS rays arrive also to the receiver causing multipath.
HAP
Shadowing
LoS
Single‐bounce Double‐bounce
NLoS NLoS

Figure 10: Visualization of a multipath propagation environment.
2.2.5 Time, Frequency, and Space Selectivity
Time selectivity refers to a time‐varying multipath channel originated by the
motion of the transmitter and/or the receiver and/or the scattering elements [68]‐
[70]. When viewed in the frequency domain by applying Fourier transform (FT) to
the time correlation function, time selectivity appears as different Doppler
(frequency) shifts of the individual multipath components over a finite spectral
bandwidth and corresponds to a frequency dispersion of the Doppler power
spectrum of the transmitted signal. The value of this dispersion and the
corresponding maximum Doppler frequency mainly depend on the angle between
the direction of motion of the terrestrial mobile station and direction of arrival of
the scattered waves, and the amplitudes of the received partial waves. Based on
the rate with which the channel impulse response changes relative to the signal
transmission rate, channels may be classified as fast fading, i.e., the channel
27

Chapter 2
changes within the transmitted symbol duration, or slow fading, i.e., the channel is
approximately constant within symbol duration. A good measure of channel time
selectivity is given by the channel coherence time (in the time domain), i.e., the
time duration for which the channel can be considered as approximately time‐
invariant. The larger the coherence time, the slower the channel fluctuation, or
equivalently the Doppler spread (in the frequency domain).
Frequency selectivity refers to a frequency‐varying multipath channel. When

viewed in the time domain by applying inverse Fourier transform (IFT) to the
frequency correlation function, frequency selectivity appears as different time
delays of the individual multipath components over a finite spectral bandwidth,
which corresponds to a time dispersion of the power delay spectrum of the
transmitted signal. The span of the delays is called the delay spread. Delay spread
causes frequency selective fading as the channel acts like a tapped delay line filter.
Based on their degree of frequency selectivity, channels may be classified as
frequency‐flat or frequency‐selective channels. In particular, if all the transmitted
frequencies undergo approximately identical amplitude and phase changes, the
channel is called frequency‐flat. Otherwise, the channel is called frequency‐
selective and the transmission is wideband. Frequency selectivity is measured in
terms of the coherence bandwidth, i.e., the bandwidth over which the channel's
frequency response remains constant. The channel can be considered frequency‐
flat only if the transmission is narrowband compared to the channel's coherence
bandwidth. Otherwise, the channel is frequency‐selective.
Finally, space selectivity refers to a space‐varying multipath channel. When

viewed in the angle domain by applying FT to the spatial correlation function,
spatial selectivity appears as different azimuth and/or elevation angles of arrival (or
departure) of the multipath components at the receive (or transmit) antenna array
over a finite spectral bandwidth, which corresponds to an angular dispersion of the
power azimuth spectrum and/or the power elevation spectrum of the transmitted
signal. Hence, the signal amplitude depends on the spatial location of the antennas.
Space selectivity is measured in terms of the coherence distance, i.e., the spatial
28

Chapter 2
separation for which the channel's spatial response remains constant, which is
inversely proportional to the angle spread.
Fig. 11 illustrates the relationship through FT between the variances of time t ,
frequency f , and space (represented by the receive x R and transmit xT
antenna position vectors) and the spreads of Doppler frequency shift v ,
propagation delay  , and direction of arrival or departure (represented by the
receive ΩR and transmit ΩT direction vectors, respectively).
VARIANCE SPREAD
Fourier Tranform
t v
Fourier Tranform
f 
Fourier Tranform
x R ΩR
Fourier Tranform
xT ΩT

Figure 11: The relationships of the time, frequency, and space domains.
2.2.6 The Diversity Concept
The randomness of the communications channel can be exploited to enhance
the performance of a communications system through diversity [2], [68]‐[70].
Diversity refers to a method that improves the reliability of the transmission of a
signal by using two or more individual communication channels with different
characteristics, i.e., different level of fading and interference. In particular, the
probability that all channels are simultaneously in a fade condition reduces
dramatically. Hence, diversity helps stabilizing the link through channel hardening,
which leads to improved performance in terms of the BER. Diversity techniques
exploit the multipath fading, resulting in diversity gain. Since multipath fading may
take place in time, frequency, and space, diversity techniques may similarly be
exploited in each of these domains.
29

Chapter 2
Time diversity can be obtained if multiple versions of the same signal are
transmitted at different time instants. Adding a redundant forward error correction
code, the message is spread in time by means of bit‐interleaving before it is
transmitted. Thus, error bursts are avoided, which simplifies the error correction
procedure.
Frequency diversity exploits the temporal spreading of the channel through
equalization techniques or multi‐carrier modulations. Thus, the signal is
transmitted using several frequency channels or spread over a wide spectrum that
is affected by frequency‐selective fading. Naturally, both time and frequency
diversity techniques incur a loss in time or bandwidth to allow for the introduction
of redundancy.
Although conventional single‐input single‐output (SISO) systems, i.e., systems

utilizing one antenna only at both the transmitter and the receiver, can exploit time
and frequency diversity, spatial diversity can be obtained using multiple transmit
antennas (transmit diversity) and/or multiple receive antennas (receive diversity).
Indeed, single‐input multiple‐output (SIMO) systems, i.e., systems utilizing one
antenna at the transmitter and multiple antennas at the receiver, multiple‐input
single‐output (MISO) systems, i.e., systems utilizing multiple antennas at the
transmitter and one antenna at the receiver, and MIMO systems, i.e., systems
utilizing multiple antennas at both the transmitter and the receiver, are the
potential candidates for providing spatial diversity schemes.
The multi‐user MIMO (MU‐MIMO) systems also intend to provide diversity by
exploiting the availability of multiple independent terminals [92]. Hence,
enhancement of the communications capabilities of each individual terminal can be
obtained. Although single‐user MIMO (SU‐MIMO) techniques only consider access
to multiple antennas that are physically connected to each individual terminal, MU‐
MIMO can be seen as the extended concept of space‐division multiple access
(SDMA) that allows a terminal to transmit (or receive) signal to (or from) multiple
users in the same band simultaneously. Fig. 12 demonstrates the aforementioned
architectures.
30

Chapter 2
SISO Transmitter Receiver
SIMO Transmitter
 Receiver
MISO Transmitter
 Receiver
MIMO Transmitter
  Receiver
Receiver/
Transmitter
Transmitter/
MU‐MIMO
Receiver  
Receiver/
Transmitter
Figure 12: Basic representation of SISO, SIMO, MISO, MIMO, and MU‐MIMO systems.
Finally, diversity techniques may also exploit the polarization of antennas.
Specifically, multiple versions of a signal are transmitted and received via antennas
with different polarization [93], [94]. Polarization diversity can overcome possible
space limitation and still achieve the advantages predicted by MIMO theory.
2.3 The MIMO Technology
As modern communications services demand increased channel capacity and

the radio spectrum is a limited resource, the research efforts have been focused on
the design and development of highly reliable and high‐speed wireless
communication systems. It is well known that multipath propagation severely
affects the quality, the reliability, and the performance of SISO communications
systems. Although diversity through SIMO and MISO systems was the first step to
mitigate multipath propagation, the emergence of MIMO systems has effectively
31

Chapter 2
converted the multipath effect into a benefit for the communications systems [95],
[96]. In a very general way, MIMO technology improves the quality of the received
signal and increases the data rate by using digital signal processing techniques to
shape and combine the transmitted signals from multiple wireless paths created by
the use of multiple receive and transmit antennas. The advent of MIMO technology
represents one of the most important advancements in wireless communications
and has attracted attention worldwide, since it offers a number of benefits that
help achieve significant performance gains and meet the challenges posed by the
impairments in the wireless channel as well as resource constraints.
2.3.1 Benefits
The major benefits of MIMO techniques are array gain, spatial diversity gain,
and spatial multiplexing gain. Although SIMO and MISO systems offer array gain
and spatial diversity gain, only MIMO systems can exploit environments with rich
scattering and take advantage of spatial multiplexing gain, which offers a potential
increase in channel capacity beyond the Shannon capacity for SISO systems at no
cost of extra radio spectrum and/or transmitted power [68]‐[70].
Array gain refers to a coherent combining of the transmitted signals at the
receiver. In particular, the same signal is emitted from each of the transmit
antennas with appropriate phase and amplitude weightings such that the signal
power is maximized at the receiver input. This coherent combining realized through
spatial processing at the receive antenna array and/or spatial pre‐processing at the
transmit antenna array improves the resistance to noise and increases the signal
gain from constructive interference. Hence, array gain increases the average
receive SNR and improves the coverage and range of a wireless communications
network. However, this coherent combining requires knowledge of the channel
state information (CSI).
Spatial diversity is realized by providing the receiver with multiple (ideally
independent) copies of the transmitted signal separated in space. Then, as the
32

Chapter 2
number of these copies increases, the probability that at least one of these copies
do not experience deep fading also increases. A MIMO channel with nT transmit
antennas and nR receive antennas potentially offers nT  nR independently fading
links, and hence a spatial diversity order of nT  nR . Spatial diversity techniques can
be applied using space‐time coding, when there is no CSI knowledge. Then, the
array gain cannot be exploited.
Spatial multiplexing requires MIMO antenna configuration and takes full

advantage of the spatial dimension of the propagation radio channel offering a
linear increase in data rate by transmitting multiple, independent data streams
within the bandwidth of operation [95]. In particular, a high rate signal is divided
into multiple lower rate streams and each stream is transmitted from a different
transmit antenna in the same frequency channel. Under suitable channel
conditions, such as rich scattering in the propagation environment, these data
streams arrive at the receiver antenna array with sufficiently different spatial
signatures and can be separated at the receiver end. Each data stream experiences
at least the same channel quality that would be experienced by a SISO scenario.
Thus, the capacity is effectively enhanced by a multiplicative factor equal to the
number of streams. In general, the number of data streams that can be reliably
supported by a MIMO channel equals the minimum of the number of transmit and
receive antennas, i.e., minnT , nR . Spatial multiplexing can be used with or without

transmit channel knowledge, but the practical gain is limited by correlation, which
means that some of the parallel streams may have very weak channel gains [97]‐
[99].
2.3.2 Channel Statistics
Considering a MIMO communications system with nT transmit antennas and nR
receive antennas and assuming frequency non‐selective (flat fading) over the
bandwidth of interest, the MIMO radio channel at a given time instant may be
represented as a nR  nT matrix H as follows [68], [69]
33

Chapter 2
h11
h21
h12
h22
hnR 2
nT Transmitter
 h1nT H hnR 1  Receiver nR
h2nT
hnR nT
MIMO
Radio Channel
Figure 13: Visualization of the MIMO radio channel.
 h11 h12  h1nT 

 
 h21 h22  h2nT 
H  , (9)
    
 
hnR 1 hnR 2  hnR nT 

where hij is the (SISO) channel gain between the i th receive and j th transmit
antenna pair. Note that channel capacity linearly increases according to rank1 of
the channel matrix H. Fig. 13 visualized the MIMO radio channel.
In the presence of a sufficiently large (theoretically infinite) number of non‐
coherent diffuse components in dense scattering propagation environment, the
fading is described by a Rayleigh distribution [70], [100], the channel matrix is full‐
rank, and the elements of this matrix are uncorrelated, i.e., independent and
identically distributed (i.i.d.). However, when a strong LoS signal also arrives at the
receiver, the fading is described by a Rician distribution. Considering coherent
diffuse components or sparse scattering and increased correlation in space and/or
time and/or frequency, the rank of the channel matrix is deficient and the spectral
efficiency is low. Indeed, the degree of correlation is a complicated function of the

1
The rank of a matrix is equal to the number of linearly independent rows (or columns) of the matrix or equivalently the
number of non‐zero eigenvalues.
34

Chapter 2
degree of scattering in a specific propagation environment and the antenna inter‐
element spacing at both the transmitter and the receiver. Hence, an increase in
this spacing is not sufficient to ensure decorrelation between the responses in the
MIMO channel matrix. On the other hand, dense scattering in the propagation
environment in combination with adequate2 antenna spacing ensures
decorrelation. These two factors control the performance measures, i.e., the
diversity gain and the spatial multiplexing gain, and therefore the applicability of
MIMO techniques to any communications system.
A HAP‐based communications channel is expected to be Ricean in its general
form, i.e., Rice distribution can be employed to describe the statistics of the
channel [11]. Considering this general case of Ricean fading, the HAP‐MIMO
channel matrix can be modeled as [68]
K 1
H HLoS  HNLoS , (10)
K 1 K 1
where HLoS is the nR  nT matrix containing the free‐space LoS responses among
the nT and nR antenna elements, HNLoS is the nR  nT matrix containing the
responses due to the scattered waves (NLoS), and K is the Ricean factor, which
expresses the relative power of the direct and scattered components of the
received signal and provides an indication of the link quality. When K  0 the
channel is described by a Rayleigh distribution, whereas a very large value of K ,
i.e., K  , implies the presence of a Gaussian channel. Recent studies have
shown that the performance of MIMO systems depends on K [101]. In particular,
as K increases, the correlation between MIMO subchannels increases [102].
Hence, efficient and accurate methods for estimating K are of considerable
interest [103]. Several values of K have been reported in the literature from
measurement campaigns and studies performed in the L and S frequency bands,
for satellite [104] and HAP‐based [105] communications systems. According to
these measurements, the value of K depends on the elevation angle of the

2
Antenna spacing equal to half wavelength is usually required for terrestrial mobile terminals [96].
35

Chapter 2
satellite/platform and the operating frequency. In particular, the results of
propagation measurements conducted in a semi‐urban area [105] showed that K
varies between 0.9 and 18.6 dB at frequency 1.2 GHz, and between 1.4 and 16.8 dB
at frequency 2.4 GHz, for the elevation angle ranging from 10 to 90.
Nevertheless, the value of K also depends on the propagation area, and the
degree of urbanization. Thus, K is expected to be lower in highly urbanized areas,
where the scatterers are usually dense and tall. Note that HAP and/or user
movement and HAP displacement cause a continuous change in the sub‐platform
point and in the elevation angle of platform. Thus, the Ricean factor K also varies
with the elevation angle variation.
2.4 The Earth‐Stratosphere Channel: From SISO to MIMO

Modeling
Since many factors influence the channel propagation characteristics, it is of
great interest to properly characterize and model the HAP‐MIMO channel for
different conditions. Then, efficient and reliable HAP‐MIMO systems can be
designed and accurately evaluated, before their implementation. This section
provides an overview of the available HAP‐SISO models and discusses some open
issues related to channel features not sufficiently reproduced by these models in
order to support MIMO extension. In addition, this section outlines the various
MIMO channel modeling approaches and categorizes them into main groups. This
section also carries out a review of MIMO techniques that have been recently
proposed for satellite/HAP communications [106]. Finally, special attention is given
to the particular characteristics of the stratospheric channel, which ultimately
determine the viability of MIMO over HAP.
2.4.1 Survey of HAP‐SISO Channel Models

The propagation effects suffered by a signal transmitted over a HAP‐based

system are similar to those present over a terrestrial or a satellite system, in terms
36

Chapter 2
mainly of shadowing and multipath, while tropospheric effects substantially control
the quality of the link of both satellite and HAP‐based systems. Therefore, channel
models for HAPs could be based on models applied to terrestrial and satellite
systems [11]. However, classical terrestrial modeling artificially separates slow and
fast variations due to shadowing and multipath, respectively, and models them
independently. Conversely, in satellite systems, these two processes are usually
treated statistically in a combined manner. The reason is that terrestrial
propagation rarely exhibits LoS conditions and the propagation environment
contributes to both slow and fast variations. Nevertheless, direct signal is usually
present, when the transmitter is located above the earth due to the higher
elevations and impairments of the signal are mainly caused only by the local
environment. Indeed, although terrestrial, satellite and stratospheric channels
exhibit similar multipath fading, the intensity of this small‐scale effect is not the
same, since the effective scatterers are present only at specific positions of the link.
Nevertheless, HAP‐based systems exhibit distinct characteristics compared to
satellite and terrestrial systems, with regard to the size of the coverage area, the
length of the radio path between transmitter and receiver, the link geometry, and
the propagation time delay. Several models for HAP‐SISO channels are reported in
the literature, and can be categorized as switched‐channel models, empirical
channel models, and geometry‐based channel models [91], [105]‐[112].

2.4.1.1 Switched‐Channel Models
The changes in shadowing and multipath are typically modeled by different
propagation states, e.g., “good” and “bad” states corresponding to LoS/open/light
shadowing areas and NLoS/blocked/heavy shadowing areas, respectively [111]‐
[116]. Then, the goal of a switched channel model is to properly characterize a
time‐variant channel, which switches from one state to another at any time. These
propagation states can be described by a first‐order Markov chain with specific
state and transition probabilities. The differences between the states are related to
the type of fading that is affecting the channel. The classic narrowband model of
37

Chapter 2
switched‐channel is the one presented in [113] for the L‐ band. This model
characterizes the process of fading through a switch between two states, one
defined as good (Rice distribution) and the other defined as bad (Rayleigh‐
lognormal distribution). There are also three‐state channel models proposed in
[114], [115]. The characteristics that define the channel, such as the finite number
of states present in a discrete time, make the use of Markov chains possible.
A possible weakness of the Markov model is that utilizes the same distributions
for all the channel states. Alternatively, a semi‐Markov model for HAP‐SISO
broadband channels is proposed in [112], which considers that the time duration
between states transition is random and can be characterized by some type of
probability distribution. One of the greatest advantages of the semi‐Markovian
process is the possibility to make a clear distinction between the probability
distribution of the duration of the fading present in a state and the type of fading
present in that state, which helps to get a better approximation of the link channel
conditions [116]. A general feature of the Markov and semi‐Markov approaches is
the memoryless property of the channel, where one state is uncorrelated to other
instances of the same state at different times.
2.4.1.2 Empirical Channel Models
Empirical channel models can be obtained based on experimental

measurements. In [105], an empirical model for the link between HAPs and
terrestrial users was defined and analyzed, based on experiments carried out in a
“semi‐urban” environment. From this model, narrowband channel characteristics
were presented in terms of the Ricean factor and the local mean received power
over a wide range of elevation angles. However, this model treated only the total
received power, without differentiating the LoS power and the multipath scattered
power during measurements. In [117], [118], empirical models to predict
penetration loss into buildings in an urban area as a function of the elevation angle
were developed, while results from a measurement campaign emulating the HAP‐
to‐indoor channel at S‐band were presented in [119]. These empirical models were
38

Chapter 2
established based on experiments performed in a specific area. Hence, it is
questionable whether they can be applied to other propagation environments.
2.4.1.3 Geometry‐based Channel Models
The channel characterization strongly depends on the location of the
transmitter and the receiver. As shown in Fig. 14, the fundamental parameters,
which describe the geometry of a basic HAP‐based system, are the elevation angle
of the platform T , the height of the platform HT , and the distance D between
the mobile station and the sub‐platform point.
HAP
HT
Mobile
Station
T
D
Figure 14: The typical geometry of a HAP‐based system.
To the author’s best knowledge, two different geometries for HAP‐SISO
channels were previously proposed. In [91] an ellipsoid was considered as the
volume containing all the scatterers in the terrain with transmitter and receiver as
foci (see Fig. 15). This model is based on the theoretical model proposed in [120]
for a terrestrial station and is extended to the case of a stratospheric station. In this
extension, the receiver and the transmitter are no longer on the horizontal plane
and the height of the transmitter must be taken into account This model provides a
really convenient method for estimating the small‐scale fading of the
communication links between HAPs and terrestrial stations and characterizes the
mobile channel through the power delay profile and the Doppler spectrum.
39

Chapter 2
However, this model overestimates the effects of large delay components. This
deficiency is due to the assumption of uniformly distributed single bounce echoes.
Therefore, a blockage‐based channel model for HAP‐SISO channels was proposed in
[109], quantified the existing probability of the single bounce echoes from different
scatterers, and obtained an improved distribution function of the excess delay
through numerical integration. Although the blockage‐based model seems more
realistic than the ellipsoid model, it greatly complicates the derived process.
HAP
NLoS
LoS
y
Scatterer
Terrestrial
Station
x
Figure 15: The ellipsoid geometrical model for HAP‐SISO channels.
z
HAP
LoS Scatterer
NLoS
y
Terrestrial
Station
x
Figure 16: The circular cone geometrical model for HAP‐SISO channels.
40

Chapter 2
A multipath model based on circular straight cone geometry (see Fig. 16) was
also proposed in [110]. This geometry could be a better approximation to simulate
the multipath propagation, since it represents the coverage area of a HAP‐based
system more accurately. The energy from the scatterers is not considered
significant close to the platform, as [91] suggests, but more concentrated close to
the base of the cone. This new geometric definition can resemble more accurately
the scenario of a directive antenna onboard the HAP illuminating a specific
coverage area on the ground.
2.4.2 Classification of MIMO Channel Models
The MIMO models can be classified into categories depending on specific
criteria. A potential way of distinguishing the individual models is with regard to
the bandwidth. In particular, the MIMO models can be divided into narrowband
models, i.e., the multipath fading is frequency flat, and wideband models, i.e., the
multipath fading is frequency selective. MIMO models can also be separated into
outdoor, indoor, urban, suburban or rural depending on the propagation
environment, as well as they can be characterized as mobile or fixed depending on
the existence or absence of mobility, respectively. Nevertheless, MIMO models can
be generally classified into physical or non‐physical (analytical) based on the
modeling philosophy [121], [122].
Physical channel models use important physical parameters to provide

reasonable description of the MIMO channel characteristics and the surrounding
scattering environment. Depending on the chosen complexity, these models allow
for an accurate reproduction of the real channel. Indeed, choosing a small number
of physical parameters makes it difficult, if not impossible, to identify and validate
the models. Physical models can further be classified into deterministic models,
geometry‐based stochastic models, and non‐geometric stochastic models.
Deterministic models characterize the physical propagation parameters in a
completely deterministic manner, i.e., using ray tracing techniques [123]. With
geometry‐based stochastic models, the impulse response is characterized by the
41

Chapter 2
entire system and the geometry of the scattering environement. Moreover, non‐
geometric stochastic models describe physical parameters in a mathematical
(analytical) way without assuming an underlying geometry.
On the contrary, non‐physical models synthesize the MIMO channel matrices in
the context of system and algorithm development and verification. Although the
non‐physical models are easy to simulate and provide accurate channel
characterization for the situations under which they are identified, they give limited
insight to the propagation characteristics of the MIMO channels and depend on the
measurement equipment, e.g., the bandwidth, the configuration and aperture of
the arrays, the heights and response of transmit and receive antennas in the
measurements. Non‐physical models can be further subdivided into propagation‐
motivated models and correlation‐based models. The interest reader is referred to
[121], [122] for a detailed comparison between physical and non‐physical models.
Finally, another way to model the MIMO channel is through field measurements
of the MIMO channel responses. Empirical models based on MIMO channel
measurement campaigns are reported in [124] for terrestrial systems and in [125]
for satellite systems. However, the parameters used to generate these models are
based on specific measurement campaigns for particular environments. Thus, these
models can be used to generate fading effects for similar types of environments,
which limit their merit.
2.4.3 Recent Advances in Satellite/HAP MIMO Channel Modeling
The wireless world has seen substantial increase in the demand of high quality
broadband wireless services during recent years, which has resulted in the
evolution of MIMO communications systems. In an effort to remain competitive
with terrestrial systems, satellite and HAP‐based systems are trying to follow the
progress in MIMO technology and profit from the significant enhancements [106],
[126]. In this section some of the recently developed models for satellite/HAP
MIMO channels are presented.
42

Chapter 2
2.4.3.1 Physical‐Statistical Modeling
As mentioned before, different approaches can be used to build models for
MIMO channels. For instance, physical or deterministic channel models based on
ray‐tracing algorithms can provide accurate results for a particular scenario.
However, this approach is not often used due to increased computational
complexity. On the other hand, statistical models are built around measurement
data and provide reasonable reproduction of the real channel. However, they
provide little insight into the propagation mechanisms and depend on the accuracy
of the measurements. An intermediate approach between these models is the
physical‐statistical model. This type of modeling is the most appropriate in
predicting the ‘ON/OFF’ nature and investigating the small‐scale fading effects over
large coverage areas applicable to satellite/HAP‐based communications systems
[127].
In [128], a physical‐statistical model for land mobile satellite (LMS) and HAP
MIMO channels was proposed. This model generates high‐resolution time‐series
data and power delay profile for communication links between satellite (or HAP)
and terrestrial terminal antennas and also predicts the correlation between these
links. In this model, the obstacles (e.g., buildings, trees) are grouped into clusters of
spherical shapes and the clusters centers are randomly positioned. Multiple
scatterers are placed randomly around the cluster center and their position follows
the Laplacian distribution, while the building heights follow the lognormal
distribution. Three paths between a satellite and a mobile terminal are considered:
a LoS path, a blocked LoS path, and an attenuated path by trees. To validate this
model, the parameters obtained from experimental data collected in Munich,
Germany at L‐band (1.54 GHz) for urban and high‐way environments. The small‐
scale fading and the wideband parameters can be approximated using the output
time‐series and spatial power delay profile data of the model.
As the maximum MIMO gain can be achieved with low correlation between
antenna elements at both ends of a MIMO communications system, a fundamental
43

Chapter 2
way of achieving low antenna correlation is to use antenna elements with
adequate separation. However, owing to size restriction, the physical‐statistical
model suggests that multiple antennas with large separation cannot be deployed at
a single satellite. Hence, two satellites are required to achieve diversity.
Nevertheless, employing two satellites gives rise to new challenges, such as waste
of the limited satellite bandwidth for the transmission of the same signal, lack of
synchronization in reception, and high implementation cost. The synchronization
issues can be dealt with by employing cooperative satellite diversity concept [129],
[130] or by using compact antennas [94], [131], in which the problem of
synchronization does not exist. From this physical‐statistical model, one also
obtains that the application of multiple antennas at a single HAP may be viable.
However, further and extensive research is required to verify the results.
2.4.3.2 Analytical Modeling
The physical‐statistical channel model described above is designed for the L‐ and
S‐ frequency bands. The application of MIMO technology to satellite systems
operating at the Ku‐ frequency band and above was discussed in [132]. In this
model, two features of MIMO technology are presented: (i) a 2  2 MIMO spatial
multiplexing system is used to achieve capacity improvements and a closed form
expression for the outage capacity is derived (ii) a MIMO spatial diversity scheme
with receive antenna selection is applied in order to reduce interference in LMS
communication links. In addition, an analytical closed form expression for
interference mitigation on forward link of a satellite 2  2 MIMO diversity system
with antenna selection is also obtained. In order to discuss the features of MIMO
techniques, the model assumes high antenna directivity and propagation
phenomena, such as clear LoS, rain fading, and rainfall spatial homogeneity. The
propagation delay offset (synchronization problem) in LMS communications is also
considered and a practical solution to this problem is suggested by applying
matched filters, first to the received signals for the detection of propagation delay
offset and then the resulting signals are fed to the timing aligner.
44

Chapter 2
2.4.3.3 Polarized Satellite/HAP MIMO Channels
The overwhelming majority of analytical and experimental work carried out on
LMS and HAP MIMO channels at L‐ and S‐ frequency bands focuses on exploiting
polarization diversity at both the transmitter and receiver to form a MIMO channel
matrix [93], [94], [131], [133]‐[135]. This approach seems beneficial, since a single
satellite cannot provide the necessary antenna spacings required by MIMO theory
to provide a high degree of channel decorrelation. Spatial, temporal and
polarization SIMO and MISO measurements in S‐ and C‐ bands for mobile satellite
systems have been recently carried out by ESA employing existing satellites [136]
and by Centre National d’Etudes Spatiales (CNES) employing a helicopter [137]. A
relevant SIMO channel modeling approach is presented in [138]. With regard to
MIMO measurement campaigns, the relevant attempts are extremely scarce and
have been conducted mainly in the frame of [125] in Guildford, UK, at 2.45 GHz. In
order to get benefit from polarization dimension, the cross‐polar transmissions
(e.g., transmission from vertically polarized antenna to horizontally polarized
antenna) should be zero. However, in real scenarios there is always some
polarization mismatch since linearly polarized antenna arrays have non‐zero
patterns for cross‐polar fields. In addition, multipath effects (e.g., diffraction,
scattering, reflection, etc.) may change the plane of polarization of incident
electromagnetic waves at the receiver. A physical‐statistical model for dual‐
polarized satellite links is presented in [134], while the capacity enhancement
resulting from the use of a multiple HAP system employing compact MIMO antenna
arrays and multiple spatial‐polarization states was investigated in [93]‐[94].
2.4.4 Potential Application of MIMO Technology over Single HAPs
The prerequisite so that single HAP configurations fully exploit the spatial
diversity and spatial multiplexing advantages predicted by information theory is the
existence of sufficient antenna spacing at the transmitter and receiver, as well as a
rich scattering environment, which renders the fading paths between the antenna
45

Chapter 2
elements of the transmitter and the receiver independent. Insufficient spacing
between antenna elements and limited scattering correspond to lower MIMO
channel capacity compared to the ideal i.i.d. case. Therefore, future HAP‐MIMO
channel models should take the distribution of the scatterers and the correlations
among signal carriers into account. The application of multiple antennas to single
HAPs may be plausible to be pursued in the frame of HAP‐MIMO systems, since
spatial limitations on board are not as stringent as in satellites. Moreover, the
length of the radio path between HAPs and terrestrial stations is significantly
smaller than the corresponding one associated with satellites.
Although several MIMO models have been proposed for terrestrial and satellite
communications channels, newer models are required to accurately characterize
specific issues concerned with HAP‐MIMO channels. For these channels, the
physical‐geometrical characteristics, which define the HAP‐MIMO system itself, are
highly critical. Most of the physical models of terrestrial channels postulate a
scattering environment and attempt to capture the channel characteristics by
involving scattering parameters. Such models can often illustrate the essential
characteristics of the MIMO channel, as long as the constructed scattering
environment is reasonable. The conventional geometry‐based models for HAP‐SISO
channels provide a convenient framework for characterizing the time and
frequency domain of the fading channels. Notwithstanding, extending these
models to the MIMO case is not straightforward or even not applicable mainly due
to the need of utilization of the space domain, which is the essence of MIMO
technology. Specifically, it is necessary to incorporate new parameters, such as the
angles of arrival and departure, the angle spread, and the utilization of multiple
antennas at both the ends of the link. Consequently, the establishment of a
particular geometry allows for an accurate characterization of the multipath effects
and supports the investigation of MIMO applicability.
Chapter 3 introduces and justifies a novel cylindrical geometry. Based on this
geometry, a 3‐D reference model for narrowband HAP‐MIMO channels is
proposed.
46

Chapter 3
Chapter 3
Three‐Dimensional Modeling of
Space‐Time Correlated HAP‐MIMO Channels

3.1 Introduction
Chapter 2 reviewed the MIMO principle and underlined its potential advantages.
However, the invariable non‐zero correlation of MIMO subchannels in space and/or
time can significantly degrade and limit the prospective MIMO gain [97], [98].
Spatial correlation is measured among the elements of the channel matrix for each
realization, while temporal correlation is measured between channel matrix
realizations at different time instants. In practice, due to the length of the radio
path between HAPs and terrestrial stations, the transmit and/or receive antennas
should be placed at significant distances from each other to ensure that the paths
are really diverse. In [93], [94] virtual MIMO (V‐MIMO) techniques were used
exploiting the diversity provided by multiple HAPs, while in [128] a general
estimation of the required antenna element separation at a HAP to achieve
uncorrelated responses in the MIMO channel matrix was provided. Nevertheless,
whether the size of a single HAP can support spatial diversity is still questionable.
Hence, further and extensive research is necessary.
Analysis and design of HAP‐MIMO communications systems require the

development of space‐time models, which enable us to properly characterize the
stochastic fading channel and thoroughly study the channel statistics. Since
impairments of the signal are mainly caused by the environment near to the user, a
realistic positioning of the scatterers is essential for an accurate channel model.
Thus, when terrain and scattering distributions are available, physical‐geometrical
models are preferred. In [91], a geometrical model applied to HAP‐SISO channels
47

Chapter 3
was proposed. According to this model, the scatterers are uniformly distributed in
space, i.e., the user receives signals from all directions with equal probabilities (see
Fig. 17). However, this assumption deviates from practical situations. In addition,
the influence of the HAP elevation angle was not considered for the performance
evaluation. In [128], a physical‐statistical MIMO model applied to satellites/HAPs
was presented. According to this model, the scatterers are non‐uniformly
distributed, i.e., the user receives the signal only from particular directions (see Fig.
18). Nevertheless, analytical expressions for the space‐time correlation function
(STCF) were not derived and the effect of the array configuration was not studied.
Thus, newer models are essential on mathematically analyzing the impact of any
system parameter on the performance and providing in‐depth understanding and
description of the statistical properties of HAP‐MIMO channels.

Figure 17: Visualization of isotropic scattering conditions in an open area.
Figure 18: Visualization of non‐isotropic scattering conditions in a narrow street.
This chapter proposes a novel 3‐D geometry‐based single‐bounce (GBSB)

reference model [139] for HAP‐MIMO channels and constructs the received
48

Chapter 3
complex faded envelope as a superposition of the LoS and the NLoS rays. This
model utilizes L‐ (1/2 GHz) and S‐ (2/4 GHz) frequency bands and assumes that the
local scatterers in the vicinity of the mobile user are non‐uniformly distributed
within a cylinder. An alternative version of the proposed model was presented in
[140] and [141], but only the first tier of scatterers lying on the surface of a cylinder
was taken into account. These studies were based on existing terrestrial 3‐D MIMO
channel models that deal with fixed‐to‐mobile [142] or mobile‐to‐mobile [143]
cases. The proposed model derives the STCF considering several parameters, such
as the elevation angle of the platform, the array configuration, the Doppler spread,
and the distribution of the scatterers. The numerical results demonstrate the
impact of these parameters on the space‐time correlation and the channel
capacity. These results also estimate the required HAP antenna spacing to achieve
uncorrelated HAP‐MIMO channels and show that if HAP length is adequate, MIMO
techniques are applicable on a single HAP.
The remainder of this chapter is organized as follows. In Section 3.2 the HAP‐
MIMO communication system is described. Section 3.3 introduces a 3‐D
geometrical model for HAP‐MIMO channels, proposes a reference model, and
derives its channel impulse response. Section 3.4 derives the STCF, investigates the
influence of model parameters on the STCF, and studies the effect of the
stratospheric winds on the correlation. Section 3.5 defines and investigates the
HAP‐MIMO channel capacity for different values of the model parameters. Finally,
Section 3.6 summarizes this chapter.
3.2 The HAP‐MIMO System Model
The system model employed throughout this chapter considers a downlink

narrowband MIMO Stratospheric‐Terrestrial communication channel with nT
transmit and nR receive antenna elements. All antennas are fixed, omni‐
directional, and are numbered as 1  p  q  nT and 1  l  m  nR , respectively. The
49

Chapter 3
nT antenna elements of the stratospheric base station (SBS) are situated
approximately 20 km above the ground and it is assumed that the nR antenna
elements of the terrestrial mobile station (TMS) are in motion. Considering slowly
varying and frequency‐flat‐fading channels, the link between the SBS and TMS
antenna arrays is represented using the following complex baseband vector
equation
r  t   H  t  s  t   n  t  , (11)

where s  t   nT 1 is the transmitted signal vector, r  t   nR 1 is the received
signal vector, and n  t   nR 1 is the noise vector, which denotes the AWGN at the
receiver branches. The entries of the noise vector are i.i.d. complex Gaussian
random variables with zero‐mean and variance N0 , where N0 is the noise PSD.
Finally, H  t    hij  t     nR nT is the matrix of complex faded channel gains.

nR nT
3.3 A Reference Model for HAP‐MIMO Channels
In this section, a 3‐D GBSB reference model for HAP‐MIMO channels is
introduced and its geometrical characteristics are described. Based on this model,
the channel impulse response is derived.
Let us initially consider a stratospheric platform free of local scattering, and a
terrestrial single user, which is located in the center of a disc‐shaped region of
scattering elements. As shown in Fig. 19, the distance between the user and the
scatterers is not constant in reality. Moreover, the radius of the disc corresponds to
the maximum distance between the user and an effective scatterer. This approach
is realistic for HAP‐based scenarios, due to the high elevation angles and the use of
omni‐directional antennas at the user terminals. In addition, the platform is placed
in the stratosphere and the user is often located lower than the surrounding
scatterers. Therefore, the waves may travel in both horizontal and vertical planes,
50

Chapter 3
and the propagation environment has to be characterized by 3‐D scattering
conditions, i.e., the heights of the scatterers need to be taken into account. Since
the scattered waves are mostly owing to buildings, poles, trees, and hills that can
be found only up to a predefined maximum height, a cylinder is finally considered
as the volume containing all the scatterers, and the maximum scatterer height
corresponds to the height of the cylinder.
The geometrical characteristics of the proposed model and the definition of the
Cartesian coordinate system are discussed in Figs. 20‐23. Fig. 20 shows the LoS
paths of the 3‐D geometrical model for a 2  2 HAP‐MIMO channel, while Fig. 21
presents the projection of this model to the x ‐y plane. Figs. 22 and 23 show the
NLoS paths for the same channel. Based on this simple configuration, uniform
linear arrays (ULAs) with arbitrary number of antennas can be constructed. Note
that shadowing within the cylinder is neglected in this model. In addition,
attenuation due to rain, oxygen absorption, water vapor, clouds, precipitation, and
vegetation is bypassed, since these atmospheric‐environmental phenomena do not
significantly affect communication links at L‐ and S‐ frequency bands [11]. For ease
of reference, the parameters of the proposed model are summarized in Table 5.
HAP
NLoS LoS
COVERAGE AREA
USER
SCATTERING REGION

Figure 19: Basic concept of a HAP‐based communication system and a cylindrical scattering
region inside the coverage area.
51

Chapter 3
p
z
q OT
HS ,max
R l
y T
p
OR l vR
T vT Rl
aLoS R
T m R
O 
m O RS ,max x
q
Figure 20: The LoS paths of the 3‐D geometrical model for a 2  2 HAP‐MIMO channel.
y
p vT
l vR
Rl 
T T aLoS R
R
O O x

m
q
D RS ,max

Figure 21: The LoS paths of the projection of the 3‐D geometrical model for a 2  2 HAP‐
MIMO channel.
52

Chapter 3
p
z
q OT
S 
n
HS ,max
  R l
n
HS 
n
S
y
p
OR
T vT  n  n l vR
S  
n RS aR R
m R
O  T a n  O x
T 
m RS ,max
q
Figure 22: The NLoS paths of the 3‐D geometrical model for a 2  2 HAP‐MIMO channel.
y
p vT
S  
n
l vR
T n
aR R
T  n
aT RS n  R
O O x

m
q
D RS ,max

Figure 23: The NLoS paths of the projection of the 3‐D geometrical model for a 2  2 HAP‐
MIMO channel.
53

Chapter 3
TABLE 5
DEFINITION OF THE PARAMETERS USED IN THE REFERENCE MODEL
D The distance between the centre O of the projections of the SBS antenna
elements to the x ‐y plane and the lower centre O of the cylinder.
RS ,max , HS ,max The radius and the height of the cylinder with the scatterers, respectively.
T ,  R The spacing between two adjacent antenna elements at the SBS and TMS,
respectively.
T , R The orientation of the SBS and TMS antenna array in the x ‐y plane
(relative to the x ‐axis), respectively.
 T ,  R The elevation angle of the p th SBS and l th TMS antenna element, relative

to the x ‐y plane, respectively.
vT , vR The velocities of the SBS and TMS, respectively.
 T ,  R The moving directions of the SBS and TMS, respectively.
fT ,max , fR ,max The maximum Doppler frequency shift of the SBS and TMS, respectively.
T The elevation angle of the SBS relative to OR .
HT , HR , HS  The height of the SBS, the TMS, and the n th scatterer, respectively.

n
Rl
aLoS The azimuth angle of arrival of the LoS paths.
aT  , aR  The azimuth angle of arrival and the azimuth angle of departure at/from
n n
the n th scatterer, respectively
RS  The distance between O and the projection of the n th scatterer to the
n
x ‐y plane.
 S  The elevation angle of the n th scatterer relative to OR .
n
  n  , g  n  The random phase and the amplitude introduced by the n th scatterer,
respectively.
 The mean azimuth angle at which the scatterers are distributed in the x ‐y
plane (von Mises pdf).
k The spread of the scatterers around the mean azimuth angle (von Mises
pdf).
a The spread of the scatterers around the TMS (hyperbolic pdf).
HS ,mean The mean of scatterer’s height (log‐normal pdf).
 The standard deviation of scatterer’s height (log‐normal pdf).
54

Chapter 3
According to Figs. 20‐23, the x ‐axis is the line that connects coordinate origin O

(the centre of the projections p and q of the SBS antenna elements p and q to
the x ‐y plane), and O (the lower center of the cylinder). In addition, OT and OR
are the array centers at the SBS and TMS, respectively. To aid the analysis, the
distance between two points a and b is denoted as d  a, b  . Then, the distance
between O and O is d  O ,O   D , and the heights of the SBS and TMS antennas
arrays are d  O ,OT   HT and d  O, OR   HR , respectively. Note that HT  HR and
that the elevation angle of SBS relative to OR is given by
T  arctan HT / D  . (12)

The spacing between two adjacent antenna elements at the SBS and TMS is
denoted by T and  R , respectively. Angles T and R represent the orientation of
the SBS and TMS antenna arrays respectively, relative to the x ‐axis. In addition,
the angle  R describes the elevation angle of the l th TMS antenna element,
Rl
relative to the x ‐y plane, and the angle aLoS denotes the azimuth angle of arrival
(AoA) of the LoS paths. Finally, it is considered that SBS (aircraft case) and TMS are
moving with speeds vT and vR in the directions determined by the angles  T and
 R , respectively. Since aircrafts fly on a circle, vT is related to the angular velocity
of SBS. Then, vT  RT  2 RT / T , where  is the angular frequency, T is the
period, and RT is the radius of the rotation. Therefore,  T is always perpendicular
to RT .
The proposed model assumes N   fixed, omni‐directional scatterers and
single scattering conditions. Then, the nth scatterer is denoted by S   , the
n
distance between its projection to the x ‐y plane and O is denoted by
 
d S   ,O  RS    0, RS ,max  , and its height is designated by d S   , S   
n n n n
 
HS    0, HS ,max  , where RS ,max and HS ,max represent the radius and the height of
n
55

Chapter 3
the cylinder, respectively. Note that HS ,max  HT and that the elevation angle of
the nth scatterer relative to OR is given by

n

 S   arctan HS  / RS  .
n n
 (13)
The angles aT  and aR  denote the azimuth angle of departure (AoD) of the waves
n n
that impinge on the scatterer S   and the azimuth AoA of the waves scattered
n
from S   , respectively. It is assumed that aT  , aR  , RS  , HS  , and  S  are
n n n n n n
random variables. According to Fig. 22, aT  is dependent on aR  and RS  , while

n n n
 S n  is dependent on RS n  and H S n  . However, aR n  , RS n , and HS n  are
independent.
It is assumed that RS ,max is much smaller than D, i.e., RS ,max  D.3 Then, one
 
concludes that aT   arcsin RS  / D  arcsin  RS ,max / D  is a small angle for every
n n
RS    0, RS ,max  . Furthermore, it is assumed that R is much smaller than RS  ,
n n
 
i.e.,  R  min RS  . Finally, it is assumed that T is much smaller than D, i.e.,
n
 T  D.3 Then, aLoS
Rl
.
3.3.1 Channel Impulse Response
Based on the 3‐D reference model, the impulse response hpl  t  of the
subchannel p‐l is a superposition of the LoS and NLoS rays and can be written as
hpl  t   hpl ,LoS  t   hpl ,NLoS  t  . (14)

3
Note that as the SBS elevation angle T increases, the distance D decreases. Without loss of generality, it is assumed that
T  85. From (12), for T  85 (maximum value of the aforementioned inequality) and HT  20 km, one can obtain that
D  1750 m. As long as RS ,max and T are equal to a few hundreds of wavelengths, the assumptions RS ,max  D and
T  D are valid.
56

Chapter 3
Since the number of local scatterers is infinite, central limit theorem [144] implies
that hpl  t  is a low‐pass nonzero‐mean complex Gaussian process. Hence, the
envelope hpl  t  is a Rice process. In addition, hpl ,NLoS  t  is a low‐pass zero‐mean
complex Gaussian process, and the envelope hpl ,NLoS  t  is Rayleigh distributed,
while hpl ,LoS  t  is a deterministic process.
The LoS and NLoS components of the impulse response can be written as
hpl ,LoS  t  
K pl  pl
e
j
2
 e 

d  p ,l  j 2 t  fT ,max cos  aLoS
Rl
 Rl
 T  fR ,max cos aLoS R 
 ,  (15)
K pl  1

 pl 1 N  n  j 
2 

d p ,S   
 n  d S  n  ,l 
hpl ,NLoS  t   g e

lim
K pl  1 N  N n1

 n n  n 
j 2 t  fT ,max cos aT  T   fR ,max cos aR  R  cos  S 
j  
n
     
e e , (16)

where j 2  1, fT ,max  vT /  and fR ,max  vR /  are the maximum Doppler shifts
associated with SBS and TMS, respectively, and  is the carrier wavelength. It is
assumed that the phase   n  introduced by the nth scatterer is an i.i.d. random
variable uniformly distributed in the interval   ,  and independent from aT  ,

n
aR  , RS  , HS  , and  S n  . Considering a unit transmit power,

n n n
 pl  E  hpl  t    1 denotes the transmitted power of the subchannel p‐l , where

2
 
E is the statistical expectation operator. In addition,

K pl  hpl ,LoS  t  / E  hpl ,NLoS  t   denotes the Ricean factor of the subchannel p‐l

2 2
 
and g  n  represents the amplitude of the nth scattered wave such that

 
N 2 
N 1
E  g n   1 as N  . The set g

n
n 1
consists of independent positive
n1
 

random variables with finite variances, independent of   n  .
n1
57

Chapter 3
As shown in the Appendix A, the distances d  p, l  , d p, S   , and d S   , l are

n n
   
approximated as


d  p, l   D  DT1  DR1 / cos T , 
(17)

 n
 
d p, S    D  DT1  DT2 RS  sin aR  / cos T ,
n n

(18)

 n

d S   , l  RS  / cos  S   DR1 cos aR  cos  S 
n n n n
DR2 sin  S   DR3 sin aR  cos  S  ,

n n n
(19)

where
DT1  0.5  nT  1  2p   T cosT , (20)

DT2  0.5  nT  1  2p   T sinT / D , (21)

DR1  0.5  nR  1  2l   R cos R cos R , (22)

DR2  0.5  nR  1  2l   R sin R , (23)

DR3  0.5  nR  1  2l   R sin R cos R . (24)

Finally, using (13) and (17)‐(24), and the approximation aT   sin aR RS  / D (see
n n n
Appendix A), the LoS and NLoS components of the impulse response become,
respectively

2 D DT1  DR1 
K pl  pl j
j 2 t  FT ,LoS FR ,LoS 
hpl ,LoS  t   e  cos T
e , (25)
K pl  1

 pl 1 N n  n  j 2 t  FT ,NLoS FR ,NLoS 
hpl ,NLoS  t  
K pl  1 N 
lim 
N n1
g A  n A  n  e j e
p ,S S ,l
, (26)

where

FT ,LoS  fT ,max cos   aLoS
Rl

 T ,  (27)

FR ,LoS  fR ,max cos aLoS
Rl
R ,   (28)
58

Chapter 3
2   n n 
j  D DT1 DT2 RS sin aR 
 cos T  
A e , (29)
p ,S  
n

 n
2 RS
j  n   n  n 
2 DR1 cos R cos arctan HS /RS  
 cos arctan HS  /RS    j
 n n    
   
A e e
S   ,l
n

  n  n 
2 DR2 sinarctan HS /RS  
 n   n  n 
2 DR3 sin R cos arctan HS / RS  
     
j j
e  e  , (30)

n

FT ,NLoS  fT ,max RS  sin T sin aR  / D  cos  T ,
n
 (31)

FR ,NLoS  fR ,max cos aR    R
n
  cos arctanH  / R   . (32)
S
n
S
n
3.4 Space‐Time Correlation Function
Considering a 3‐D non‐isotropic scattering environment, the STCF between two
arbitrary subchannels hpl  t  and hqm  t  is defined as
Rpl ,qm T , R , ,t   E hpl  t  hqm

*
 t     / pl qm , (33)
where   denotes complex conjugate operation and E is the statistical
*
expectation operator. Since both hpl ,NLoS  t  and hqm,NLoS  t  are zero‐mean
processes, i.e., E hpl ,NLoS  t    E  hqm ,NLoS  t    0, (33) can be written as
Rpl ,qm T , R , , t 
 E  hpl ,LoS  t   hpl ,NLoS  t   hqm

*

,LoS  t     hqm ,NLoS  t     /  pl qm
*  
 hpl ,LoS  t  hqm
*
,LoS  t    /  pl qm
 hpl ,LoS  t  E hqm

*
,NLoS  t     /  pl qm

 hqm ,LoS  t    E 

 hpl ,NLoS  t   /  pl qm
*

 E hpl ,NLoS  t  hqm

*
,NLoS  t     /  pl qm

 RplLoS,qm  T , R , , t   RplNLoS
,qm   T , R , , t  , (34)
59

Chapter 3
where RplLoS,qm  T , R , , t  and RplNLoS

,qm   T ,  R , , t  denote the STCFs of the LoS and
the NLoS component, respectively. Using (15), (27), (28), and (33), the STCF of the
LoS component can be written as
RplLoS,qm T , R , , t 

K pl Kqm
e
j
2
  e  
d  p ,l d  q ,m    j 2  fT ,max cos  aLoS
Rm
 Rm
 T  fR ,max cos aLoS R 
 
 K pl  1 Kqm  1
 e  
j 2 tfT ,max cos  aLoS
Rl
 
Rm
 T cos  aLoS

e

 T  j 2 tfR ,max cos aLoS

Rl
 Rm
 R cos aLoS  R   
 .  (35)

Rl
Using (17), (20), (22), and the approximation aLoS  aLoS
Rm
  , (35) becomes

RplLoS,qmT , R , ,t 
 RplLoS,qm T , R , 
2  q  p T cos T  m l  R cos R cos R 
K pl Kqm j
 cos T
 e
 K pl  1 Kqm  1
 j 2  fT ,max cos  T  fR ,max cos  R 
 e . (36)
Moreover, using (26) and (33), the STCF of the NLoS component can be written as
,qm   T , R , , t 
RplNLoS
,qm   T , R , 
 RplNLoS
1

 K pl  1 Kqm  1
1 N   n 2  j 2  FT ,NLoS FR ,NLoS  
 lim  E  g A  n A  n A*  n A* n e . (37)

n1 
N  N p , S S ,l q , S S ,m
The number of scatterers in the reference model is infinite. Thus, the discrete
variables aR  , RS  , and HS  can be replaced with continuous random variables
n n n
aR , RS , and HS with joint probability density function (pdf) f  aR , RS , HS  . Since
aR  , RS  , and HS  are independent, the joint pdf f  aR , RS , HS  can be
n n n
60

Chapter 3
decomposed to f  aR  f  RS  f  HS  . Hence, using (29)‐(32), the STCF of the NLoS
component becomes [145, p.23]
,qm   T , R , 
RplNLoS
HS ,max RS ,max  2  q  p T cos T 2  q  p T sinT RS sin aR

j j
1  cos T D cos T
   e e
 K pl  1 Kqm  1 0 0 

2  m l  R sin R sinarctan HS /RS   2  m l  R sinR cos R sin aR cos arctan HS / RS  
j j
 e  e 
2  m l  R cos R cos R cos aR cos arctan HS / RS    R sin  T sin aR 

j  j 2 fT ,max  S cos  T 
 e  e  D 
 j 2 fR ,max cos aR  R  cos arctan HS / RS  

 e f  aR  f  RS  f  HS  daR dRS dHS . (38)
Several different distributions, such as uniform [97], von Mises [102] and
Laplacian [128], [146], were used in prior work to characterize the azimuth angle
aR . In this work, the von Mises pdf (also known as the circular normal distribution)
is used, because it approximates many of the aforementioned distributions, and is
empirically justified in urban and suburban areas in [147]. The von Mises pdf is
defined as
e R 
k cos a  
f  aR   , ‐  aR   , (39)
2 I0  k 

where I0   is the zeroth‐order modified Bessel function of the first kind,
   ,  is the mean angle at which the scatterers are distributed in the x ‐y
plane, and k  0 controls the spread around the mean. This pdf was initially
introduced by R. von Mises in 1918 to study the deviations of measured atomic
weights from integral values [149] and has played a prominent role in statistical
modeling and analysis of angular variables [150, pp. 57‐68]. It is interesting to
note that the von Mises pdf has appeared in a number of other communication
contexts. For example, this pdf is referred to as the Tikhonov pdf in partially
coherent communication [151, p. 406], has been used in phase‐locked loop related
61

Chapter 3
problems [152], and has been shown to represent the pdf of the phase of a sine
wave in Gaussian noise for large signal to noise ratios [153].
Fig. 24 demonstrates the von Mises pdf for different values of the parameter k
and   0. One observes that setting k  0, i.e., f  aR   1 / 2 , incurs isotropic
scattering. As k increases, the scattering becomes increasingly non‐isotropic. For
small k , the von Mises pdf approximates the cardioid pdf [149, p.60], which is
rather similar to the cosine pdf [153], while for large k it resembles a Gaussian pdf
[149, p.60] with mean  and standard deviation 1/ k . For k   (extremely non‐

isotropic scattering), the von Mises pdf becomes a Dirac delta function
concentrated at aR   , i.e., f  aR   δ  aR    , where δ   denotes the Dirac

delta function.
2
k=0
k=0.5
1.5 k=1
Probability density
k=2
k=5
1 k=10
k=20
0.5
0
‐180 ‐120 ‐60 0 60 120 180
α (degrees)
R
Figure 24: The probability density for the distribution of the angle of arrival of the
scattered waves for different degree of scattering.
In order to characterize the distance RS , the hyperbolic pdf [154] is used, which
is empirically justified in [155], [156]. This distribution is more realistic and flexible
than other commonly used distributions, such as the uniform distribution [91],
62

Chapter 3
[120], because it allows scatterers to assume higher concentration in a flexible area
in the vicinity of the TMS. The hyperbolic pdf is defined as follows [154]‐[156]
a
f  RS   , 0<RS  RS ,max . (40)
tanh  aRS ,max  cosh2  aRS 

The parameter a controls the spread (standard deviation) of the scatterers around
the TMS and its applicable value is in the interval  0,1 .
Figs. 25 and 26 show the hyperbolic pdf and the mean distance between TMS
and the scatterers, respectively, for several values of the parameter a and
RS ,max  200 m. One observes that decreasing a increases the spread of the pdf of

RS and increases the mean distance between the scatterers and the terrestrial
user. Considering a HAP‐based communications system, as the elevation angle of
the platform decreases, the scatterers are expected to be more widely distributed,
which corresponds to a possible decrease of a. Nevertheless, a can be accurately
obtained through measurements in different propagation environments, i.e.,
urban, suburban or rural environments.

0.08
α=0.08
α=0.06
α=0.02
0.06
α=0.01
Probability density
α=0.005
0.04
0.02
0
0 50 100 150 200
R (m)
S
Figure 25: The probability density for the distribution of the distance between TMS and the
scatterers for different values of the parameter a.
63

Chapter 3
100
Mean distance between TMS‐scatterers (m)
80
60
40
20
0
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Parameter α
Figure 26: The mean distance between the TMS and the scatterers for different values of
the parameter a.
The building height was found to follow normal [157], log‐normal [157]‐[159] or
Rayleigh distribution [160], depending on the selected urban or suburban
environment. Specifically, the corresponding distributions matched geographical
data values obtained from measurements of building heights conducted in high and
medium built‐up density areas in different European countries. In order to
characterize the height HS , the log‐normal pdf is adopted , which is defined as
ln HS ln HS ,mean  

2
 
2
e 2
f  HS   , 0<HS  HS ,max , (41)
HS 2
where the parameters HS ,mean and  are the mean and standard deviation of HS ,
respectively. Fig. 27 demonstrates the experimentally verified log‐normal

distribution of building heights in London, U.K. (densely built‐up district) and
Guildford, U.K. (medium‐size town) with parameters HS ,mean  17.6 m and
  0.31, and HS ,mean  7.2 m and   0.26, respectively [157].
64

Chapter 3
0.25
London, H =17.6 m, σ=0.31
S,mean
Guildford, H =7.2 m, σ=0.26
0.2 S,mean
Probability density
0.15
0.1
0.05
0
0 10 20 30 40 50
H (m)
S
Figure 27: The probability density for the distribution of the building heights in London
(U.K.) and Guildford (U.K.).
Finally, using (39)‐(41), and the equality


 e
a sin c   b cos c 
dc  2 I0  a2  b2 
[161, eq.3.338‐4], (38) becomes
 
HS ,max RS ,max
RplNLoS
,qm T , R ,   r1   r2I0 r32  r42 dRS dHS , (42)
0 0
where
  q  p T cos T 
j 2   fT ,max cos  T 
 cos 
ae  T 
r1  , (43)

 2 tanh aRS ,max  I0  k   Kpl  1 Kqm  1
ln HS ln HS ,mean  

2
2  ml R sin R sinarctan HS /RS  
 
j
2
e 2 e 
r2  , (44)
HS cosh2  aRS 

2  q  p  T sinT RS
r3  j  j2 fT ,max sin T RS / D
D cos T
 j2  m  l   R sinR cos R cos arctan  HS / RS   / 

 j2 fR ,max sin R cos arctan  HS / RS    k sin  , (45)
65

Chapter 3
r4  j2  m  l   R cosR cos R cos arctan  HS / RS   / 

 j2 fR ,max cos  R cos arctan  HS / RS    k cos  . (46)
To obtain the STCF of the NLoS component, the integral in (42) has to be
numerically evaluated, since there is no closed‐form solution.
Finally, the STCF between two arbitrary subchannels hpl  t  and hqm  t 
becomes a summation of the STCFs of the LoS and NLoS components defined in
(36) and (42)‐(46), respectively.
3.4.1 Investigation of Space‐Time Correlation of HAP‐MIMO Channels
This subsection investigates the absolute spatial and temporal correlation, and
demonstrates the impact of the antenna elements spacing and arrangement.
Unless indicated otherwise, the values of the model parameters used to obtain the
curves are nT  nR  2, T  60 , T  R  90 ,  R  0 , HT  20 km, k  3,   0
and RS ,max  200 m. Moreover, a typical densely built‐up district (London, U.K.
[157]) is considered to be the scattering region, i.e., the surrounding buildings act
as scatterers, and the values of the corresponding model parameters used are
HS ,max  100 m, HS ,mean  17.6 m, and   0.31. In addition, it is assumed that
a  0.01, which corresponds to a reasonable average distance between the TMS
and an effective scatterer of approximately 60 m (see Fig. 26). This value was
observed by applying ray‐tracing techniques to 3‐D digital maps of European typical
urban environments (Fig. 28). Specifically, the AWE Communications Winprop
software package was used, which is designed to accurately predict the
propagation characteristics between the link of a transmitter and a receiver
including all important parameters of the mobile radio channel. To properly
simulate the desired HAP‐based scenario, single bounce scattering conditions, i.e.,
either reflection or diffraction, were considered for an elevated transmitter at 20
km and an elevation angle of 60 . Finally, in order to clearly present the influence
66

Chapter 3
of each model parameter on the correlation, and indicate the minimum achieved
correlation, it is considered that K pl  Kqm  K  0 (Rayleigh channel). However, in
many cases a LoS component exists, i.e., K  0. Fig. 29 demonstrates the
correlation as a function of the Ricean factor, in the general case. One observes
that the absolute correlation easily exceeds 0.15, 0.5 and 0.7, as soon as the Ricean
factor exceeds ‐7 dB, 0 dB and 3 dB, respectively.
Figure 28: 3‐D building distribution in a typical urban propagation environment.
0.8
Correlation
0.6
0.4
0.2
0
‐20 ‐15 ‐10 ‐5 0 5 10 15 20
Ricean K‐factor (dB)
Figure 29: The absolute correlation of HAP‐MIMO channels for various values of the Ricean
factor.
67

Chapter 3
,ql   T  as a
Fig. 30 depicts the transmit spatial correlation function (SCF) RplNLoS
function of the antenna separation at the SBS for different elevation angle T .
Observe that as the elevation angle  T increases, the SCF drastically decreases,
under both isotropic and non‐isotropic scattering environments. From Fig. 30 the
proper HAP antenna separation to achieve an uncorrelated MIMO channel matrix
can be also indicated. For instance, assuming that an absolute correlation of 0.15 or
less amounts to nearly uncorrelated links, and considering 2.1 GHz carrier
frequency [35], i.e., the wavelength is equal to approximately 0.14 m, and
T  60 , HAP antennas require a separation of around 10 and 18 m in isotropic
and non‐isotropic scattering environments, respectively. This suggests that utilizing
MIMO techniques on a single aircraft or airship is a viable solution, as long as the
Ricean factor is small. Moreover, if HAP length is adequate, more than two
antennas may be used to further increase the prospective MIMO
multiplexing/capacity gain.4
1
β =50ο, k=0
Τ
0.8 β =60ο, k=0
Τ
β =70ο, k=0
Spatial Correlation
Τ
0.6 β =50ο, k=3
Τ
β =60ο, k=3
Τ
0.4
β =70ο, k=3
Τ
0.2
0
0 20 40 60 80 100 120 140 160 180
δ /λ
Τ
Figure 30: The transmit SCF of a 2  2 HAP‐MIMO Rayleigh channel for various elevation
angles of the SBS, in isotropic  k  0  and non‐isotropic  k  3  scattering environments.

4
The capacity of MIMO channels linearly increases with the number of antenna elements [95].
68

Chapter 3
,ql   T  and receive Rpl ,pm   R  SCF as a function

Fig. 31 shows the transmit RplNLoS NLoS
of the antenna separation at the SBS/TMS for different orientation of the SBS/TMS
antenna array. Note that SBS broadside antenna arrays, i.e., T  90 , provide
minimum spatial correlation, while SBS inline antennas, i.e., T  0 , are fully
correlated. In addition, the receive correlation depends on the relative angle
between the TMS antenna array, and the local scatterers around TMS, i.e., R ‐ ,
and is minimized, when the difference between R and  is 90.

Fig. 32 depicts the transmit RplNLoS NLoS
of the antenna separation at the SBS/TMS, for different degrees of local scattering
(in the azimuth domain) at the TMS. Note that as k increases, the scattering
becomes increasingly non‐isotropic and the correlation significantly increases. For
highly concentrated scatterers around the mean angle, i.e., for k  10, the required
separation at the SBS for uncorrelated subchannels is  T  200 , while the
required separation at the TMS is  R  1.5 .

of the antenna separation at the SBS/TMS for different spread of the scatterers
around the TMS, which is controlled by the parameter a of the hyperbolic
distribution. In particular, a values of 0.1, 0.05, and 0.005 are used, which
correspond to an average distance between the TMS and an effective scatterer of
approximately 7 m, 13.5 m, and 88 m, respectively (see Fig. 26). Observe that as a
increases, the spread of scatterers decreases and the correlation drastically
increases.

of the antenna separation at the SBS/TMS for different distance RS ,max between
the projections to the x ‐y plane of TMS and an effective scatterer, when a  0.01.
One observes that increasing RS ,max beyond 150 m has a negligible effect on the
correlation.
69

Chapter 3
1 0.8
o
θ =0
Spatial Correlation
Spatial Correlation
0.8 T θ =0o
R
0.6
0.6 θ =90o θ =30o

T R
θ =60o 0.4
T
0.4 θ =30o
T θ =60o
R
0.2
0.2
θ =90o
R
0 0
0 50 100 150 200 0 0.5 1 1.5
δ /λ δ /λ
Τ R
(a) (b)
Figure 31: The transmit (a) and receive (b) SCF of a 2  2 HAP‐MIMO Rayleigh channel for
different orientation of the SBS and TMS antenna array, respectively.
1 1
k=0 k=0
k=1 k=1
0.8 k=2 0.8 k=2
k=5 k=5
Spatial Correlation
Spatial Correlation
0.6 k=10 0.6 k=10
0.4 0.4
0.2 0.2
0 0
0 50 100 150 200 0 0.5 1 1.5
δ /λ δ /λ
T R
(a) (b)
different degree of local scattering at the TMS.
1 1
α=0.1
0.8 0.8
α=0.05
Spatial Correlation
Spatial Correlation
α=0.05 α=0.1
0.6 0.6
α=0.005
0.4 0.4
α=0.01
α=0.005
0.2 0.2
α=0.01
0 0
0 50 100 150 200 0 0.5 1 1.5
δ /λ
Τ δ /λ
R
(a) (b)
different spread of the scatterers around the TMS.
70

Chapter 3
1 1
R =100m R =100m
S,max S,max
R =150m R =150m
0.8 S,max 0.8 S,max
R =200m R =200m
S,max S,max
Spatial Correlation
Spatial Correlation
R =250m R =250m
0.6 S,max
0.6 S,max
R =300m R =300m
S,max
S,max
0.4 0.4
0.2 0.2
0 0
0 20 40 60 80 100 120 140 0 0.5 1 1.5
δ /λ δ /λ
T R
(a) (b)
different maximum distance between the projections to the x ‐y plane of the TMS and an
effective scatterer.
1
H =23.1m
S,mean
o
ψ =90 H =17.6m
R S,mean
0.8 H =7.2m
S,mean
Spatial Correlation
0.6
0.4
ψ =0o
R
0.2
0
0 0.5 1 1.5 2 2.5 3
δ /λ
R
Figure 35: The receive SCF of a 2  2 HAP‐MIMO Rayleigh channel with horizontally
 R   
 0 or vertically  R  90 placed TMS antennas, for various mean values of the
scatterer height.
, pm   R  of horizontally, i.e.,  R  0 , and


Fig. 35 shows the receive SCF RplNLoS
vertically, i.e.,  R  90 , placed TMS antenna arrays as a function of the antenna
separation at the TMS for different mean of the scatterer height HS ,mean . Note that

71

Chapter 3
for horizontally placed arrays, changing HS ,mean has an insignificant influence on
the spatial correlation, whereas for vertically placed arrays the spatial correlation

drastically decreases, as HS ,mean increases. Fig. 35 suggests that two vertically
placed antennas can have small correlations and provide considerable diversity
gain in highly urbanized areas. It also justifies the inclusion of the third dimension
of the model.
,qm   T , R  is illustrated in Fig. 36. Both Rayleigh and

The shape of the SCF RplNLoS
Ricean fading conditions are considered, and the Ricean factor is set to 0 and 3 dB,
respectively. It is clear that the SCF is significantly affected not only by the spacing
but also by the arrangement of the antenna elements. Providing that the Ricean
factor is small, and the SBS and TMS antenna elements are carefully arranged, low
spatial correlation can be obtained.
1 1
Spatial Correlation
Spatial Correlation
0.9
0.5 0.8
0.7
0
0 0
200 1 200
1 150 150
100 2 100
δ /λ 2 50
δ /λ
R 50
R δ /λ δ /λ
3 0 T 3 0 T
(a) (b)
Figure 36: The SCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean channel.
Figs. 37‐40 investigate the effect of model parameters on the temporal
correlation function (TCF), i.e.,  T   R  0, RplNLoS   of a Rayleigh HAP‐SISO

channel. The quasi‐stationary airship is considered and the corresponding
,max  0 Hz. In addition, the maximum

maximum Doppler frequency shift used is fTAirship
Doppler frequency shift and the moving direction of TMS are fR ,max  100 Hz and
 R  60 , respectively.
72

Chapter 3
1
k=0
k=3
0.8
Temporal Correlation k=8
0.6
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.1
τ (s)
Figure 37: The TCF of a HAP‐SISO Rayleigh channel for different degree of local scattering
at the TMS.
Fig. 37 depicts the TCF RplNLoS   for different degrees of local scattering (in the

azimuth domain) at the TMS. Note that as k increases, the temporal correlation
increases.
Fig. 38 shows the TCF RplNLoS   for different spread of the scatterers around the

TMS, i.e., different a, and different distances RS ,max between the projections to
the x ‐y plane of TMS and an effective scatterer. Observe that as a increases, the
temporal correlation drastically increases. However, changing RS ,max has a
negligible effect on the temporal correlation.
Fig. 39 depicts the TCF RplNLoS   for different mean of the scatterer height. Note

that as HS ,mean increases, the temporal correlation fairly decreases.
Fig. 40 demonstrates the TCF RplNLoS   for different maximum Doppler

frequency fR ,max and moving direction  R at the TMS. One observes that an

increase in fR ,max results in a decrease in temporal correlation. Furthermore, as  R
73

Chapter 3
increases, the TMS moves toward the SBS and the temporal correlation drastically
decreases.
1
R =150m, α=0.01
S,max
R =200m, α=0.01
S,max
0.8
R =300m, α=0.01
Temporal Correlation
S,max
R =200m, α=0.05
S,max
0.6
R =200m, α=0.1
S,max
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.1
τ (s)

Figure 38: The TCF of a HAP‐SISO Rayleigh channel for different spread of the scatterers
around the TMS and different maximum distance between the projections to the x ‐y
plane of TMS and an effective scatterer.
1.4
H =7.2m
S,mean
1.2
H =17.6m
S,mean
1 H =23.1m
S,mean
0.8
0.6
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.1
τ (s)

Figure 39: The TCF of a HAP‐SISO Rayleigh channel for various mean values of the scatterer
height.
74

Chapter 3
1
f =100 Hz, γ =60ο
R,max R
0.8 f =200 Hz, γ =60ο

R,max R
Temporal Correlation f =100 Hz, γ =0ο
R,max R
0.6
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.1
τ (s)
Figure 40: The TCF of a HAP‐SISO Rayleigh channel for different maximum Doppler
frequency and different moving direction at the TMS.
1
Airship
Aircraft
0.9
0.8
0.7
0.6
0.5
0 0.02 0.04 0.06 0.08 0.1
τ (s)
Figure 41: The TCF of a HAP‐SISO Ricean channel for different HAP structures.
Fig. 41 depicts the TCF of a Ricean HAP‐SISO channel, when K pl  K  3 dB. Both

the quasi‐stationary airship and the moving aircraft structures are considered, and
,max  0 Hz and

the corresponding maximum Doppler frequency shifts used are fTAirship
75

Chapter 3
,max  300 Hz, respectively, while the moving direction of SBS (aircraft case) is
fTAircraft
set to  T  30. One observes that the two structures have almost identical
temporal correlation properties. Fig. 41 also implies that velocity variations at the
SBS due to the stratospheric winds do not significantly affect the temporal
correlation.
,ql   T ,  and the receive Rpl ,pm   R , 

Finally, the shape of the transmit RplNLoS NLoS
STCF of HAP‐MIMO Rayleigh  K  0  and Ricean  K  3 dB  channels is illustrated
in Figs. 42 and 43, respectively.
1 1
Space‐Time Correlation
0.8
0.5
0.6
0 0.4
0 0
0.1 0.1
100 100
0.05 0.05
δ /λ δ /λ
Τ τ (sec) T τ (sec)
200 0 200 0
(a) (b)
Figure 42: The transmit STCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean channel.
1 1
0.8
0.5 0.6
0.4
0 0.2
0 0
1 0.1 1 0.1
δ /λ 2 0.05 δR/λ 2 0.05

R τ (sec) τ (sec)
3 0 3 0
(a) (b)
Figure 43: The receive STCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean channel.
76

Chapter 3
3.4.1.1 Effect of HAP Antenna Displacement
This subsection investigates the effect of HAP antenna displacement due to the
stratospheric winds on the transmit spatial correlation of HAP‐MIMO channels. In
practice, there are six degrees of freedom (DoF) to which HAPs are subjected.
Specifically, HAPs may be displaced in any direction at a varying speed, and the
displacements can be shifting along the x ‐, y‐, and z ‐axes, as well as roll, pitch,
and yaw [24]‐[26]. According to the geometry of the reference model and
considering that SBS antenna elements are installed along the SBS, rotation with
respect to the x ‐axis corresponds to pitch, rotation on the y ‐axis corresponds to
roll and rotation along z‐axis corresponds to yaw (see Fig. 44). Roll and pitch can
be considered identical to each other, since both are based on similar principles.
Thus, the effect of only pitch and yaw on the correlation is studied. As shown in Fig.
44, pitch corresponds to an elevation of the pth SBS antenna element, which is
described by the angle  T .
p
T
Roll
Yaw
q
y
T
x
Pitch

Figure 44: Geometrical representation of roll, pitch, and yaw effects.
77

Chapter 3
The influence of each DoF is examined independently (keeping all other
parameters fixed) in the following figures. The values of the parameters used to
obtain the curves are the same as in the previous subsection. Fig. 45 demonstrates
the influence of 5 km shifting along either x ‐, y ‐ or z‐axis on the transmit SCF.
One observes that the correlation is relatively insensitive to displacements along
y ‐axis, while fairly increases when the displacements take place along z ‐axis.
However, shifting along x ‐axis significantly increases correlation.
,ql   T  as a
Fig. 46 studies the pitch effect and shows the transmit SCF RplNLoS

function of the angle  T . One observes that increasing angle  T from 0 to 30
has a negligible effect on the transmit spatial correlation. However, further
increase in  T drastically increases the correlation, and when  T  90 , the
antennas are fully correlated.
Finally, yaw corresponds to a change in the orientation of the SBS antenna array,
which is described by the angle T . Therefore, Fig. 31(a) represents also the yaw
effect.
1
Initial point
x‐direction
0.8
y‐direction
Spatial Correlation
z‐direction
0.6
0.4
0.2
0
0 50 100 150 200
δ /λ
T
Figure 45: The transmit SCF of a 2  2 HAP‐MIMO Rayleigh channel for different directions
of HAP displacement due to stratospheric winds.
78

Chapter 3
Spatial Correlation ψ =90o
0.8 T
ψ =0o
0.6 T
ψ =30o
T
ψ =60o
0.4 T
0.2
0
0 50 100 150 200
δ /λ
T
Figure 46: The transmit SCF of a 2  2 HAP‐MIMO Rayleigh channel for different elevation
angle of the p th SBS antenna element (pitch effect).
3.5 Analysis of HAP‐MIMO Channel Capacity
This section utilizes the reference model described above and defines the HAP‐
SISO and HAP‐MIMO channel capacity.5 It also compares the capacity of a 2  2 HAP‐
MIMO channel with that of a HAP‐SISO channel and investigates the impact of
model parameters on the capacity of HAP‐MIMO channels [162], [163].
3.5.1 Definition of HAP‐MIMO Channel Capacity
For a HAP‐SISO channel, the Shannon upper bound limit on capacity is given by

C SISO  log2 1  SNR h11
2
 bps/Hz, (47)

where h11 is a normalized complex faded channel gain and SNR corresponds to the
average signal‐to‐noise ratio at the input of the receiver. Assuming that the channel
is known to the TMS and unknown to the SBS, the available capacity of a HAP‐
MIMO channel can be obtained as follows [95]

5
Channel capacity is defined as the maximum rate at which data can be transmitted at an arbitrarily small error probability.
79

Chapter 3
  SNR  H 
CMIMO  log2 det  InR    HH  bps/Hz, (48)
 n
 T  

where H is the nR  nT matrix of complex faded channel gains, InR is an identity
matrix of size nR , nT and nR are the number of SBS and TMS antennas respectively,
H denotes the complex conjugate (Hermitian) transpose operator, and det  
denotes the matrix determinant. Throughout this section, the notion of ergodic
capacity [95] is also employed, which corresponds to the expectation of the capacity
   SNR  H  
C erg ,MIMO  E log2 det  InR    HH   bps/Hz. (49)

   nT   
Note that when the channel is known to the TMS, the temporal correlations do not
directly affect the ergodic capacity. However, the performance of practical coding
schemes is affected by the temporal correlation. In particular, strong temporal
correlations signify a slowly varying channel that would require longer codewords to
realize the ergodic capacity. On the contrary, many low complexity coding schemes
rely on the channel remaining constant over several symbols, and therefore may
perform better for slowly varying channels.
Since the HAP‐MIMO channel is Ricean in general, the matrix H can be obtained
using (10). For a unit transmit power and assuming that the value of the Ricean
factor of each subchannel is approximately equal to K , the normalized matrix HLoS
can be generated using (25), (27) and (28) as follows
 h11,LoS  h1nT ,LoS 

K 1  
HLoS      . (50)
K  
hnR 1,LoS  hnR nT ,LoS 
Considering a 2  2 HAP‐MIMO channel, HLoS is simplified as follows

K  1  h11,LoS h12,LoS 
HLoS   h22,LoS 
. (51)
K h21,LoS
80

Chapter 3
The matrix HNLoS can be evaluated from a large number of channel realizations
using the following equation [98]
vec  HNLoS   R1/2

NLoS vec  Hiid  , (52)

where vec   denotes matrix vectorization6, RNLoS is the nR nT  nR nT correlation
matrix associated with the NLoS component, R1/2
NLoS is the square root of RNLoS that
NLoS RNLoS  RNLoS , and Hiid is a nR  nT stochastic matrix with i.i.d. zero‐
satisfies R1/2 H /2
mean complex Gaussian entries and unit variance. The  nR nT  elements of RNLoS

2
can be obtained using (42)‐(46). In particular, RNLoS can be generated as
 R11,11
NLoS NLoS
R11,21 NLoS
 R11, nR nT

 
 R21,11
NLoS
 NLoS
R21, 
RNLoS   K  1 K  1  nR nT
 . (53)
    
 NLoS 
RnRnT ,11  RnNLoS 
R nT ,nR nT 

Considering a 2  2 HAP‐MIMO channel, RNLoS is simplified as follows
R11,11
NLoS NLoS
R11,21 NLoS
R11,12 NLoS 
R11,22
 NLoS NLoS NLoS NLoS

R21,11 R21,21 R21,12 R21,22 
RNLoS   K  1  K  1  NLoS NLoS NLoS NLoS 
. (54)
R12,11 R12,21 R12,12 R12,22 
R NLoS NLoS
R22,21 NLoS
R22,12 NLoS 
R22,22
 22,11 
Finally, in order to remove the effects of the average SNR on the channel
capacity, the channel response is normalized so that the constraint
E  H F   nT nR
2
(55)
 
is fulfilled. The term  F corresponds to the Frobenius norm of a matrix.

6
The vectorization of a matrix is a linear transformation which converts the matrix into a column vector. Specifically, the
vectorization of a m  n matrix A denoted by vec  A  , is the mn  1 column vector obtained by stacking the columns of
the matrix A on top of one another.
81

Chapter 3
3.5.2 Numerical Results
In this subsection, the effect of the model parameters on the capacity of a 2  2
HAP‐MIMO channel is investigated. Unless indicated otherwise, the values of the
parameters used to obtain the following curves are nT  nR  2, SNR  18 dB,
K  3 dB, HT  20 km, T  60 ,  T  50 ,  R  0.5 , T  R  90 ,  R  0 ,
k  3,   0 , RS ,max  200 m, a  0.01, HS ,max  70 m, HS ,mean  17.6 m,   0.31,
fT ,max  0 Hz, fR ,max  100 Hz,  R  60. Finally, a normalized Doppler frequency
fn  fR ,maxTs  0.01 is used, where Ts is the sampling period.
Figs. 47 and 48 compare the ergodic capacity obtained using a 2  2 HAP‐MIMO
architecture, i.e., nT  nR  2, and a HAP‐SISO architecture, i.e., nT  nR  1, by
varying the Ricean factor and the received SNR. One observes that maximum
capacity is obtained in an ideal HAP‐MIMO Rayleigh channel scenario. Then, the
capacity increases up to 10.6 bps/Hz compared to the capacity of a HAP‐SISO
channel. However, the potential MIMO gain drastically degrades with the strength
of the LoS component, i.e., increasing the Ricean factor increases the correlation
and decreases the multiplexing/capacity gain. On the contrary, HAP‐SISO channels
take advantage of the existence of a strong LoS signal, i.e., increasing the Ricean
factor drastically increases the capacity. Note that the differences between HAP‐
MIMO and HAP‐SISO architectures become more obvious as the SNR increases.
Then, the rank of the channel matrix plays an important role. In addition, the
channel capacity achieved using both architectures is invariable for K  10 dB.
Figs. 49‐53 investigate the effect of spatial and temporal correlation on the
capacity. Fig. 49 shows the ergodic capacity as a function of the spacing T between
SBS antenna elements for different SBS elevation angle T and K pl  Kqm  K  0.
One observes that increasing T beyond approximately 70 , 50 and 30 , when
T  50 , T  60 and T  70 , respectively, has a negligible effect on the
capacity.
82

Chapter 3
20
K=0, Rayleigh channel
K=3 dB, Ricean Channel
K=10 dB, Ricean Channel
Channel Capacity (bps/Hz)
15

2x2
HAP‐MIMO
10
HAP‐SISO
0
0 5 10 15 20 25 30
SNR (dB)

Figure 47: Comparison of the ergodic channel capacities obtained using a 2  2 HAP‐
MIMO architecture and a HAP‐SISO architecture as a function of the received SNR for
different Ricean factor.
12
SNR=10 dB
SNR=15 dB
10
2x2
HAP‐MIMO SNR=20 dB
8
4
HAP‐SISO
2
0
‐20 ‐15 ‐10 ‐5 0 5 10 15 20
Ricean K‐factor (dB)

Figure 48: Comparison of the ergodic channel capacities obtained using a 2  2 HAP‐
MIMO architecture and a HAP‐SISO architecture as a function of the Ricean factor for
different received SNR.
83

Chapter 3
11
10
Channel capacity (bps/Hz)
8
β =50o
T
7 β =60o
T
β =70o
T
6
0 20 40 60 80 100 120 140
δ /λ
T
Figure 49: The ergodic capacity of a 2  2 HAP‐MIMO channel as a function of the spacing
between the SBS antenna elements for different SBS elevation angle.
Fig. 50 depicts the complementary cumulative distribution function (CCDF) of the
capacity and examines the effect of SBS antenna array orientation on the capacity.
One observes that SBS broadside antenna arrays, i.e., T  90 , provide maximum
capacity. Decreasing T from 90 to 60 has a negligible effect on the capacity.
However, further decrease in T significantly decreases the capacity. Note that
according to [143], capacity also depends on the relative angle between the TMS
antenna array and the local scatterers around the TMS, i.e., R ‐ . In particular, the
capacity is maximized, when the difference between R and  is equal to 90.
Fig. 51 depicts the influence of the parameters k , RS ,max , and a on the capacity,

i.e., the influence of 2‐D scattering. Considering that the influence of each
parameter is examined independently (keeping all other parameters fixed), one
observes that increasing k decreases the capacity. Moreover, increasing a
decreases the spread of scatterers and decreases the capacity. However, changing
RS ,max has a negligible effect on the capacity.
84

Chapter 3
Probability (capacity > abscissa)
0.8
0.6
0.4
o θ =90o
θ =60 T
T
0.2 θ =45o
T
θ =30o
T
0
6.5 7 7.5 8 8.5 9 9.5 10 10.5
Capacity (bps/Hz)

Figure 50: The complementary cumulative distribution function of the capacity of a 2  2
HAP‐MIMO channel for different antenna array orientation at the SBS.
1
k=3, α=0.1,
R =200m
S,max
0.8 k=3, α=0.03,
R =200m
S,max
0.6 k=0, α=0.01,
R =200m
S,max
k=5, α=0.01,
0.4 R =200m
S,max
k=3, α=0.01,
R =300m
0.2 S,max
k=3, α=0.01, k=3, α=0.01,
R =200m R =100m
S,max
S,max
0
6 7 8 9 10 11
Capacity (bps/Hz)

HAP‐MIMO channel for different degree of scattering, spread of the scatterers around TMS,
and maximum distance between the TMS and the scatterers.
Fig. 52 shows the CCDF of the capacity, when the TMS antenna array is
horizontally, i.e.,  R  0 , or vertically placed, i.e.,  R  90 , for different HS ,mean .
85

Chapter 3
Note that changing HS ,mean has an insignificant influence on the channel capacity,
when the TMS antenna array is horizontally placed. However, the capacity depends
on the degree of urbanization, i.e., the heights of the scatterers, when the TMS
antenna array is vertically placed. Then, increasing HS ,mean increases the capacity. In
addition, the difference between capacities of systems with horizontally and
vertically placed TMS antenna array is nullified in highly urbanized areas. Fig. 52
suggests that if the available area in the x ‐y plane is limited for antenna array
realization, the TMS antenna array can be tilted without significant loss of channel
capacity.
Finally, the effect of the temporal correlation on the capacity is investigated in
Fig. 53 by separately varying the normalized Doppler frequency fn and the moving
direction  R of the TMS. Observe that an increase in fn results in an increase in
channel capacity due to the decrease in temporal correlation. Furthermore, as  R
increases (up to 180 ), the TMS moves toward the SBS, the temporal correlation
decreases, and the channel capacity increases.
1
H =7.2m
S,mean
H =17.6m
S,mean
0.8 H =23.1m
S,mean
0.6
0.4
ψ =90o ψ =0o
R R
0.2
0
6 7 8 9 10 11
Capacity (bps/Hz)
HAP‐MIMO channel for horizontally or vertically placed TMS antennas and different mean
of the scatterer height.
86

Chapter 3
0.8
f =0.01, γ =60o
n R
0.6
0.4
f =0.1, γ =60o
n R
0.2
f =0.01, γ =0o
n R
0
6 7 8 9 10 11 12
Capacity (bps/Hz)

HAP‐MIMO channel for different normalized Doppler frequency and different moving
direction at the TMS.
3.6 Summary
In this chapter, a reference model for HAP‐MIMO Ricean channels has been
proposed, where both the HAP and the terrestrial mobile user are equipped with
uniform linear multi‐element antenna arrays. In particular, a 3‐D cylinder model
has been designed to characterize the scattering environment that encloses the
user. Several parameters related to the physical properties of the stratospheric
communication system have been considered, e.g., the elevation angle of the
platform, the array orientation and elevation, the degree and spread of scattering
for the terrestrial user, the height of the scatterers, and the HAP displacement due
to the stratospheric winds. In this model, the effects of time and space selectivity
have been taken into account, as a result of the user/HAP mobility, and the 3‐D
non‐isotropic scattering conditions, respectively.
Based on this model the STCF has been derived, under a 3‐D non‐isotropic
fading environment. The distributions for the azimuth AoA of the scattered waves,
the distance between the effective scatterers and the user, and the height of the
87

Chapter 3
effective scatterers have been modeled by the von Mises, the hyperbolic and the
log‐normal distributions, respectively, which all have previously shown to be
successful in describing measured data. Extensive numerical calculations have been
carried out to demonstrate theoretical derivations. The results have pointed out
the effect of the model parameters on the correlation. The results have also
indicated the required HAP antenna separation to attain uncorrelated HAP‐MIMO
channels in different propagation environments. In particular, as long as the HAP
length is adequate and the Ricean factor is small, i.e., the propagation conditions
are characterized by rich multipath and weak LoS component (considering dense
urban areas, where the scatterers are usually dense and tall), a single airship or
aircraft can be employed to successfully exploit MIMO benefits.
This chapter has also utilized the proposed model to study the HAP‐MIMO
channel capacity. The results have revealed that the capacity depends on the
strength of the LoS signal, and the received SNR. In particular, increasing the Ricean
factor decreases the capacity, while increasing the SNR increases the capacity.
These results have also demonstrated the influence of the spatial and temporal
correlation on the capacity. Specifically, it has been shown that increasing the
elevation angle of the platform increases the capacity, while increasing the density
of the scatterers in the vicinity of the user and the spacing between the antennas
increases the capacity. Moreover, broadside HAP antennas maximize the capacity,
while vertically placed antennas at the mobile terminal provide considerable
capacity gain in highly urbanized areas. Finally, the results have underlined that
changing the velocity and the moving direction of the user significantly affects the
capacity.
The proposed model provides theoretical guidelines and a useful framework for
designing and testing future HAP‐MIMO communication systems at L‐ and S‐
frequency bands, under LoS and NLoS conditions. Notwithstanding, future real‐
time field measurements or at least simulation trials are needed to verify the
theoretical results. Hence, next chapter proposes deterministic and statistical
simulation models for HAP‐MIMO channels and examines their performance.
88

Chapter 4
Chapter 4
Three‐Dimensional Simulation of
HAP‐MIMO Channels

4.1 Introduction
Although theoretical channel models can be ideally verified through

experimental real‐time field trials, simulation of the radio propagation environment
is commonly used as an alternative, cost‐effective, and time‐saving approach to the
test, optimization, and performance evaluation of wireless communications
systems. Hence, the development and design of accurate and efficient simulation
models for HAP‐MIMO channels is a major research issue. The prime requirement
of a simulation set‐up is to capture the fading effects created by the radio channel
and the goal of any simulation model is to properly reproduce the channel
properties. Indeed, many different methods have been adopted for the simulation
of fading channels. The most widely accepted methods are the filtered noise
models [164]‐[166] and the sum‐of‐sinusoids (SoS) models [143], [167]‐[177].
Before the contributions of this chapter are discussed, it is necessary to understand
the major characteristics of these two different simulation philosophies commonly
employed to simulate fading channels.
The filtered noise models intend to simulate the channel properties by means of
signal processing techniques, without considering the underlying propagation
mechanisms. These models filter Gaussian noise through appropriately designed
linear time‐invariant filters to generate the channel waveform with the desired
channel PSD and capture the important first‐ and second‐order channel statistics.
However, the efficiency of this approach is limited by the utilized filter. In
particular, a high‐order filter is usually required to accurately approximate the
89

Chapter 4
desired statistical properties. Nevertheless, high‐order filters have a long impulse

response, which significantly increases the simulation time. Furthermore, since
filters with sharp stop‐bands are difficult to implement in practice, the power
spectrum obtained using this method is not band‐limited.
The SoS principle introduced by Rice [168] has been widely accepted by
academia and industry as an adequate basis for the design of simulation models
due to its reasonably low computational costs. According to this principle, the
overall channel waveform is the sum of several complex sinusoids having
frequencies, amplitudes and phases that are appropriately selected to accurately
approximate the desired statistical properties. Two main categories of SoS‐based
simulation models are reported in the literature, the deterministic simulation
models [170] and the statistical simulation models [172]. The deterministic (ergodic
statistical) models are easy to implement and have short simulation times.
Specifically, they have fixed parameters for all simulation trials and converge to the
desired properties in a single simulation trial leading to deterministic statistical
properties. On the contrary, the statistical (Monte Carlo) models have at least one
of the parameters as random variables that vary with each simulation trial. Hence,
their statistical properties also vary for each simulation trial and converge to the
desired ones in the statistical sense, i.e., when averaged over a sufficiently large
number of simulation trials. In contrast to filtered noise models, SoS‐based models
produce channel waveforms that have high accuracy and a perfectly band‐limited
spectrum. In addition, their complexity is typically reduced by cleverly choosing the
model parameters to reduce the computation load. Furthermore, SoS‐based
models can be easily extended to develop simulation channel models for MIMO
communications systems due to the explicit inclusion of spatial information, such
as the multipath angles of arrival and departure. Owing to these advantages, our
research focuses on SoS‐based models and proposes deterministic and statistical
simulation models for HAP‐MIMO channels, under the framework of the reference
model described in Chapter 3. Theoretical and simulation results are employed to
verify the reliability and the usefulness of the proposed models.
90

Chapter 4
The remainder of this chapter is organized as follows. Section 4.2 derives a 3‐D
generic non‐efficient stochastic simulation model, while Sections 4.3 and 4.4 derive
3‐D efficient deterministic and statistical simulation models, respectively. Section
4.5 present simulations results and examines the performance and the complexity
of the proposed simulation models. Finally, Section 4.6 summarizes this chapter.
4.2 The Stochastic Simulation Model
The reference model for HAP‐MIMO channels proposed in Chapter 3 assumes an
infinite number of scatterers, which prevents practical realization, i.e.,
software/hardware implementation. However, this model can be used as the
starting point for the design of simulation models with a finite number of
scatterers, which have similar statistical properties. Indeed, a stochastic simulation
model can be directly obtained by using a finite number N of scatterers. Then, the
impulse response of the sub‐channel p‐l is given by
hˆpl  t   hpl ,LoS  t   hˆpl ,NLoS  t  , (56)

where
 pl 1 N j 
2 
 n
  n

d p ,S   d S   ,l 
 e j e j 2 t  FT ,NLoS FR ,NLoS  (57)
 n
hˆpl ,NLoS  t   e
K pl  1 N n1

is the NLoS component of the impulse response, hpl ,LoS  t  is defined in (25), (27),
   
(28), d p, S   , d S   , l , FT ,NLoS , and FR ,NLoS are defined in (18)‐(24), (31), (32),
n

n
and “ ̂ ” describes a stochastic process . The random variables aR  , RS  , and HS 

n n n
can be generated utilizing the distributions defined in the previous chapter, while
  n  is generated as a random variable uniformly distributed in the interval
  ,  . All other parameters of the simulation model are identical with those of
the reference model. Specifically, aR  is modeled using the von Mises pdf and can

n
be generated as
91

Chapter 4
aR   FaR1   ,
n
(58)

where  is a random variable uniformly distributed in the interval  0,1 . The
function FaR1   denotes the inverse function of the von Mises cumulative
distribution function (cdf) and can be evaluated using the numerical method
presented in [149] (no analytical method is available). Fig. 54 depicts the von Mises
cdf for different values of k and   0.
0.8
Cumulative distribution
0.6
k=0
0.4
k=0.5
k=1
0.2 k=2
k=5
k=10
0
‐180 ‐120 ‐60 0 60 120 180
α (degrees)
R
Figure 54: The cumulative distribution of the angle of arrival of the scattered waves for
different degree of scattering.
In addition, RS  is modeled using the hyperbolic pdf and can be generated as

n
RS   FRS1   ,
n
(59)

where  is a random variable uniformly distributed in the interval  0,1 . The
function FRS   denotes the hyperbolic cdf, which is given by
FRS  RS   tanh  aRS  / tanh  aRS ,max  . (60)

92

Chapter 4
Specifically, using (59), (60), RS  is generated as follows

n
n 1
a

RS   tanh1  tanh  aRS ,max  .  (61)

Fig. 55 demonstrates the hyperbolic cdf for several values of the parameter a and
RS ,max  200 m.

0.8
0.6
0.4
α=0.08
α=0.06
0.2 α=0.02
α=0.01
α=0.005
0
0 50 100 150 200
R (m)
S
Figure 55: The cumulative distribution of the distance between TMS and the scatterers for
different values of the parameter a.
Finally, HS  is modeled using the log‐normal pdf and can be generated as

n
HS   FHS1   ,
n
(62)

where  is a random variable uniformly distributed in the interval (0,1). The
function FHS   denotes the log‐normal cdf, which is given by
1  ln  HS   ln  HS ,mean  
FHS  HS   erfc   , (63)
2   2 

93

Chapter 4
 
 u2
where erfc  x   2 /  e du is the complementary error function. Fig. 56
x
shows the log‐normal cdf for the corresponding values of HS ,mean and 
associated with London (U.K.) and Guildford (U.K.) [157].
To simulate a HAP‐MIMO channel, this generic stochastic simulation model can
be directly applied by generating the variables aR  , RS  , HS  , and   n  . However,

n n n
due to the high degree of randomness during the computation of these variables,
an infinite number of simulation trials is required and the stochastic model is non‐
realizable and inefficient. Hence, one must otherwise determine the model
parameters, in order to fully exploit the inherent advantages of the SoS principle.
This procedure is called the parameter computation method and the type of a
simulation model is directly related to this method.
0.8
0.6
0.4
0.2 London, H =17.6 m, σ=0.31

S,mean
Guildford, H =7.2 m, σ=0.26
S,mean
0
0 10 20 30 40 50
H (m)
S
Figure 56: The cumulative distribution of the building heights in London (U.K.) and
Guildford (U.K.).
4.3 The Deterministic Simulation Model
This section proposes a SoS‐based deterministic simulation model for HAP‐

MIMO channels. As mentioned in Section 4.1, the deterministic models have all
94

Chapter 4
parameters fixed for all simulation trials, thereby, leading to deterministic
properties, i.e., only one simulation trial is needed to obtain the desired properties
and the statistical behavior does not vary from one simulation trial to another.
Since the location of each scatterer inside the cylinder can be fully described by
aR  , RS  , and HS  , the entire scattering region is considered as a 3D‐point lattice

n n n
partitioned into three scattering sub‐regions, which are individually associated with
one of these variables [179]. In particular, it is considered that N1 coordinates of
scattering points are non‐uniformly distributed around a ring, N2 coordinates of
scattering points are non‐uniformly distributed along a radial line, and N3
coordinates of scattering points are non‐uniformly distributed along a vertical line.
Then, the product N  N1N2N3 is the total finite number of discrete scatterers. This
design method is visualized in Fig. 57, for ease of comprehension. Note that a
random scatterer is now designated by S 

n1 ,n2 ,n3 
and its location is controlled by
aR 1  , RS 2  , and HS 3  , where n1 1,..., N1  , n2 1,..., N2  , and n3 1,..., N3 .

n n n

RS ,max
S
n1 ,n2 ,n3  HS ,max
n 
HS 3
RS 2 
n
aR 1 
n
TMS
3rd scattering
1st scattering
sub‐region 2nd scattering sub‐region
sub‐region
Figure 57: Visualization of the three scatterer sub‐regions, each consisting of N1 , N2 , and

N3 coordinates (“  ”) of scattering points, respectively.
95

Chapter 4
The impulse response of the sub‐channel p‐l can be written as
hpl  t   hpl ,LoS  t   hpl ,NLoS  t  , (64)

where
 pl 1 N1 N2 N3  n1 ,n2 ,n3 

hpl ,NLoS  t  
K pl  1 N1N2N3
   e j
n1 1 n2 1 n3 1
2    n1 ,n2 ,n3     n1 ,n2 ,n3   
j d  p ,S d  S
e      
e
 T ,NLoS FR ,NLoS 
,l   j 2 t F 
(65)

is the NLoS component of the impulse response, hpl ,LoS  t  is defined in (25), (27),
(28), “  ” describes a deterministic time averaged quantity, n1 1,..., N1  ,
n2 1,..., N2  , n3 1,..., N3  , and d p, S 

 n1 ,n2 ,n3 
 , d  S n1 ,n2 ,n3 

, l , FT,NLoS , and
FR,NLoS are defined, respectively, as follows

d p, S 
n1 ,n2 ,n3 
  D  D T1 
 DT2 RS 2  sin aR 1  / cos T , (66)
n n

d  S
n1 ,n2 ,n3 
 n

n n

, l  RS 2  / cos arctan HS 3  / RS 2  
 
 DR1 cos aR 1  cos arctan HS 3  / RS 2  
n

n n
  

 
 DR2 sin arctan HS 3  / RS 2  
n n
 

n


n n

DR3 sin aR 1  cos arctan HS 3  / RS 2   ,
 
(67)


FT,NLoS  fT ,max RS 2  sin T sin aR 1  / D  cos  T ,
n n
 (68)

n

FR,NLoS  fR ,max cos aR 1    R  cos arctanH  / R   . (69)
S
n3
S
n2
Providing that a sufficient large number of scatterers is used, i.e.,
N  N1N2N3  20, central limit theorem implies that hpl ,NLoS  t  is close to a low‐
pass zero‐mean complex Gaussian process. Then, using a similar derivation to that
described in Chapter 3, the STCF between hpl  t  and hqm  t  can be simplified as
R pl ,qm  T , R ,   RplLoS,qm  T , R ,   RplNLoS

,qm   T , R ,  , (70)

96

Chapter 4
where
,qm   T , R , 
RplNLoS
:  hpl ,NLoS  t  hqm
*
,NLoS  t     /  pl qm

2  q  p T cos T
n  n 
j 2  q  p T sinT RS 2 sin aR 1
N1 N2 N3 j
1 1
  e  cos T
e  D cos T

 K pl  1 Kqm  1 N1N2N3 n1 1 n2 1 n3 1
 n  n  
j 2  m  l  R sin R sinarctan HS 3 / RS 2  
  
 e 
n   n  n  
j 2  m  l  R cos R cos R cos aR 1 cos arctan HS 3 / RS 2  
  
 e 
n   n  n  
j 2  m l  R sinR cos R sin aR 1 cos arctan HS 3 / RS 2  
  
  j 2  FT ,NLoS FR ,NLoS 
e e (71)
is the STCF of the NLoS component and RplLoS,qm  T , R ,  is defined in (36). From
     
N1 N2 N3
(71), it is obvious that proper finite sets aR 1  , RS 2  , and HS 3 
n n n
for
n1 1 n2 1 n3 1

the parameters aR 1  , RS 2  , and HS 3  must be determined, which lead to a good
n n n
approximation to the statistical properties of the reference model. Note that no
parameter computation method will be applied to the phases  
n1 ,n2 ,n3 
, since the
STCF does not depend on them. Notwithstanding, the interest reader is referred to
[170], where an efficient method to compute the phases is given.
Although the Lp ‐norm method (LPNM) used in [171], [180] provides maximum
accuracy and versatility, it requires the minimization of two error functions in order
to obtain proper temporal and spatial correlation properties. Since this
minimization must be simultaneously proceeded for aR 1  , RS 2  , and HS 3  , it is
n n n
obvious that simpler and less complex parameter computation methods are
desirable. Hence, following a similar approach to that proposed in [143], the
aforementioned parameters are generated as follows [163]
aR 1   FaR1    , (72)

n
97

Chapter 4
RS 2   FRS1    , (73)

n
HS 3   FHS1    ,
n
(74)

where
    n1  0.5  / N1 , (75)
    n2  0.5 / N2 , (76)
    n3  0.5 / N3 (77)
for n1  1,..., N1 , n2  1,..., N2 , and n3  1,..., N3 , respectively.
4.3.1 The Deterministic Simulation Model Based on Set Partitioning
The performance of the proposed deterministic simulation model can only be
improved by increasing the number of the discrete scatterers, i.e., the number of
N1 , N2 , and N3 , and thus the complexity, but not by averaging over the simulation
results achieved by multiple simulation runs. Moreover, the performance has to be
investigated for practical values of these numbers, which may not lead to proper
and accurate statistical properties. In [181] and [182] an alternative deterministic
parameter computation method was proposed for isotropic and non‐isotropic
scattering environments, respectively. This method utilizes the principle of set
partitioning (SP), which was initially used to design trellis‐coded modulation
schemes [183]. Based on this principle, it is assumed that a constellation (scatter
diagram) with N relevant scatterers is separated into V sub‐constellations, each of
which consists of N scatterers, such that N  V  N, where V is an integer number.
This subsection applies the principle of SP to the deterministic simulation model
for HAP‐MIMO channels described above. Hence, it is similarly assumed that
N1  V1  N1 coordinates of scattering points are located around a ring constellation.
This ring constellation is separated into V1 sub‐constellations (partitions), each of
which consists of N1 coordinates of relevant scattering points. In addition, it is
98

Chapter 4
assumed that N2  V2  N2 coordinates of scattering points are located along a radial
line constellation. This radial line constellation is separated into V2 sub‐
constellations, each of which consists of N2 coordinates of relevant scattering
points. Finally, it is assumed that N3  V3  N3 coordinates of scattering points are
located along a vertical line constellation. This vertical line constellation is
separated into V3 sub‐constellations, each of which consists of N3 coordinates of
relevant scattering points. Fig. 58 explains the proposed SP method and illustrates
the v1th , v2th , and v3th sub‐constellations of relevant scattering points, when
N1  N2  N3  3.
Under these considerations and for n1  1,..., N1 , n2  1,..., N2 , n3  1,..., N3 ,
v1  1,...,V1 , v2  1,...,V2 , and v3  1,...,V3 the model parameters are generated
using a similar approach to that proposed in [181], [182]

aR 1 1   FaR1    ,
n ,v
(78)

RS ,max
v3th sub‐constellation
HS ,max
Figure 58: The set partitioning of three scatterer constellations into V1 , V2 and V3 sub‐
constellations, each consisting of 3 coordinates (“  ”) of relevant scattering points.
99

Chapter 4
RS 2 2   FRS1    ,
n ,v
(79)

HS
n3 ,v3 
 FHS1    , (80)

where
n1  0.25
     v1 , (81)
N1
n2  0.25
     v2 , (82)
N2
n3  0.25
     v3 , (83)
N3
2v1  V1  1
 v1  , (84)
2V1N1
2v2  V2  1
 v2  , (85)
2V2N2
2v3  V3  1
 v3  . (86)
2V3N3
In practice, it is recommended to simulate the deterministic processes hpl 1 2 3   t 
v ,v ,v
successively by computing from time to time a new set of parameters
a      
N1 N N3
, and HS
n3 ,v3 
, RS
n1 ,v1 n2 ,v2  2
R using (78)‐(80).
n1 1 n2 1 n3 1
The STCF associated with the proposed deterministic SP simulation model is
given by
1 V1 V2 V3  v1 ,v2 ,v3 

Rpl ,qm T , R ,  :    Rpl ,qm T , R , ,
V1V2V3 v1 1 v2 1 v3 1
(87)

where “  ” describes an average process and Rpl 1,qm2 3  T , R ,  is calculated using

v ,v ,v
(70) and (78)‐(86) for n1  1,..., N1 , n2  1,..., N2 , and n3  1,..., N3 .
Note that Rpl ,qm T , R ,  is equal to Rpl ,qm T , R ,  defined in (70) for
N  N  V scatterers, where N  N1 N2 N3 and V  V1V2V3 . Thus, instead of simulating
a single deterministic process with N scatterers, V mutually uncorrelated
100

Chapter 4
deterministic processes with reduced complexity determined by N  N / V can be
alternatively simulated. Although the statistical properties of this deterministic
model vary for each simulation trial, they will converge to the statistical properties
of the reference model after a sufficient number of simulation trials controlled by
V1 , V2 , and V3 and for an arbitrary number of scatterers, i.e., for any

N1 , N2 , N3   1. They will also converge to the desired statistical properties after an
arbitrary number of simulation trials, i.e., for any V1 ,V2 ,V3   1, but with an
sufficient number of scatterers, i.e., for sufficient N1 , N2 , and N3 .
4.4 The Statistical Simulation Model
This section proposes a SoS‐based statistical simulation model for HAP‐MIMO

channels. As mentioned in Section 4.1, the statistical models are not ergodic and
have at least one of the parameters as a random variable for each simulation trial.
Thus, their properties have to be analyzed by statistical averages after a number of
simulation trials. Based on [143], the parameters aR 1  , RS 2  , and HS 3  are
n n n
generated as random variables for each simulation trial as follows [179]
aR 1   FaR1    ,

n
(88)

RS 2   FRS1    ,
n
(89)

HS 3   FHS1    ,
n
(90)

where
    n1  1  1 / N1 , (91)

    n2  2  1 / N2 , (92)

    n3   3  1 / N3 , (93)

where 1 ,  2 , and  3 are independent random variables uniformly distributed in
the interval 0,1 . Since the proposed statistical model has three of its parameters
101

Chapter 4
as random variables for each simulation trial, its statistical properties have to be
analyzed by statistical averages. Hence, although the statistical properties vary for
each simulation trial, they will converge to the statistical properties of the
reference model after a sufficient number of simulation trials for an arbitrary
number of scatterers, i.e., for any N1 , N2 , N3   1.
The STCF associated with the statistical model is given by
Ntrials
1
 ,qm T , R ,  
Rpl 
Ntrials ntrials 1
 ,qm T , R , 
Rˆpl (94)

for ntrials  1,..., Ntrials , where Ntrials denotes the number of simulation trials and
 ,qm T , R ,  can be obtained using (70) and (88)‐(93), providing that a sufficient

Rˆpl
number of scatterers is used, i.e., N  N1N2N3  20. Since the proposed statistical

simulation model statistically converges to the desired properties, it is important to
determine the number of simulation trials needed to achieve a desired

convergence level. Indeed, the complexity of the statistical simulation model is

controlled by N1 , N2 , N3 , and Ntrials . Thus, N1 , N2 , and N3 should be carefully
determined to reduce the number of trials required for convergence.
Table 6 reviews and compares the relative complexity of the proposed
simulation models, which generally depends on the number of the utilized
scatterers, the number of the performed simulation trials, and the number of the
required random variables.
TABLE 6
THE COMPLEXITY OF THE SIMULATION MODELS
Simulation Number Number of Relative Number of
Model of simulation number of random
scatterers trials calculations variables
Stochastic NStoc   3
Deterministic NDet 1 NDet 0
Deterministic SP NDet ,SP V  V1V2V3 NDet ,SPV 0
Statistical NStat Ntrials NStat Ntrials 3
102

Chapter 4
4.5 Simulation Results
In this section, the performance of the proposed simulation models is evaluated
for different complexity and a finite number of scatterers. The values of the model
parameters used to obtain the following curves are nT  nR  2, K pl  Kqm  3 dB,
HT  20 km, T  60 , D  H / tanT  11.5 km, T  R  90 ,  R  30 , k  3,

  30 , a  0.01, RS ,max  200 m, HS ,max  70 m, HS ,mean  17.6 m,   0.31,

 T  60 ,  R  30 , fT ,max  300 Hz, and fR ,max  100 Hz.

Tables 7 and 8 investigate the performance of the deterministic model for
different complexity, i.e., different N1 , N2 , and N3 , in terms of the TCF for the
time delay range 0 s    0.1 s, which is typically of interest for many

communications systems, and the SCF for the ranges 0   T  200 and

0   R  3 , respectively. As soon as the value of the Ricean factor is sufficiently
small, the maximum values of the range of the inter‐element distances chosen, i.e.,
200 and 3 , ensure low spatial correlation in many propagation scenarios, as
shown in Chapter 3. The effect of the variation of each number of coordinates of
scattering points is examined independently, provided that the other numbers are
sufficiently large and do not affect the results. The performance evaluation is
realized in terms of the root mean square error (RMSE) between the absolute TCFs
and SCFs of the simulation model and the reference model, as well as the
maximum deviation (MD) of the absolute TCF and SCF of the simulation model
from the corresponding theoretical values. According to Tables 7 and 8, the

performance can be improved by increasing the number of the discrete scatterers
and thus the complexity. Tables 7 and 8 also suggest that reasonable values for N1
(associated with aR ), N2 (associated with RS ), and N3 (associated with HS ) are in
the order of 30, 20, and 5, respectively.
Fig. 59 illustrates the absolute value of the TCFs of the reference and the

deterministic simulation model, respectively, for N1  30, N2  20, N3  5 and
103

Chapter 4
N1  40, N2  30, N3  8, respectively. One observes that the TCF of deterministic
model approximates the theoretical one for the range 0    s   0.1, as the

number of the scatterers increases. From the simulation results, the RMSE is
approximately 20.5  103 and 19.3  10 3 , respectively, while the MD is
approximately 33.4  103 and 31.9  103 , respectively.
Fig. 60 depicts the absolute error between the absolute SCFs of the reference
model and the simulation model for the ranges 0   T  200 and 0   R  3.

From the simulation results obtained for N1  30, N2  20, and N3  5, the RMSE is
approximately 2.1  10 3 , while the MD is approximately 7.3  103. Overall, the
deterministic simulation model satisfactorily matches the temporal and spatial
correlation properties of the reference model.
TABLE 7
PERFORMANCE EVALUATION OF THE DETERMINISTIC SIMULATION MODEL (A)
TEMPORAL CORRELATION
Number of N1 N2 N3
scatterers
RMSE MD RMSE MD RMSE MD
1 88.7  103 135  103 62.1  103 101  103 24.5  103 48.1  103
2 71.4  103 124  10 3 34.5  103 62.6  103 19.6  103 38.9  103
5 51.8  103 109  103 24.2  103 36.3  103 15.1  103 32.4  103
6 48.1  103 83.2  103 23.2  103 35.2  103 15.1  103 31.2  103
7 44.9  103 78.3  103 20.5  103 34.5  103 15.1  103 26.7  103
8 41.1  103 71.4  103 19.4  103 29.8  103 14.9  103 28.5  103
9 38.7  103 67.9  103 18.9  103 30.9  103 14.9  103 26.3  103
10 33.3  103 55.8  103 17.7  103 28.9  103 14.8  103 25.2  103
15 29.4  103 45.5  103 17.3  103 28.2  103 14.8  103 28.3  103
20 25.5  103 39.6  103 16.4  103 27.4  103 14.8  103 23.1  103
30 19.5  103 33.1  103 16.1  103 28.1  103 14.8  103 23.6  103
40 18.5  103 31.3  103 15.8  103 27.2  103 14.7  103 22.4  103
60 16.3  103 28.5  103 15.2  103 24.5  103 14.7  103 22.3  103
104

Chapter 4
TABLE 8
PERFORMANCE EVALUATION OF THE DETERMINISTIC SIMULATION MODEL (B)
SPATIAL CORRELATION
Number of N1 N2 N3
scatterers
RMSE MD RMSE MD RMSE MD
1 163  103 329  103 49.4  103 141  103 2.4  103 7.3  103
2 117  103 202  103 36.3  103 93.0  103 2.3  103 5.9  103
5 45.2  103 62.2  103 16.7  103 35.2  103 1.5  103 4.2  10 3
6 31.3  103 57.2  103 14.9  103 31.8  103 1.5  103 4.2  10 3
7 22.7  103 44.2  103 13.7  103 28.5  103 1.5  103 4.1  10 3
8 18.1  103 37.3  103 10.2  103 23.7  103 1.4  10 3 4.0  103
9 13.7  103 33.6  103 8.3  103 20.8  103 1.4  10 3 4.0  103
10 10.6  103 25.8  103 7.2  103 16.2  103 1.4  10 3 4.0  103
15 7.4  103 19.5  103 4.9  103 12.2  103 1.4  10 3 4.0  103
20 4.8  103 13.2  103 2.4  103 9.3  103 1.3  103 3.9  10 3
30 1.9  103 7.1  103 1.7  103 7.1  103 1.3  103 3.9  10 3
40 1.7  103 6.9  103 1.6  103 5.9  103 1.3  103 3.9  10 3
60 1.6  103 5.3  103 1.4  103 4.9  103 1.2  10 3 3.8  10 3

1
Reference model
0.9 Deterministic model, N =30, N =20, N =5
1 2 3
Deterministic model, N =40, N =30, N =8
1 2 3
Temporal correlation
0.8
0.7
0.6
0.5
0.4
0 0.02 0.04 0.06 0.08 0.1

τ (s)
Figure 59: Comparison between the TCFs of the reference model and the deterministic
simulation model.
105

Chapter 4
0.01
Absolute error
0.005
0
0
1 200
150
δ 2
R 100
50 δ
3 0 T

Figure 60: The absolute error between the SCFs of the reference model and the
deterministic simulation model.
Fig. 61 depicts the absolute value of the TCF of the reference model and the TCF
of the deterministic SP simulation model averaged over V  1 simulation trial,
when V1  1, V2  1, and V3  1, over V  10 simulation trials, when V1  5, V2  2,
and V3  1, and over V  50 simulation trials, when V1  5, V2  5, and V3  2.
Moreover, N1  10, N2  8, and N3  3, i.e., N  240. One observes that increasing
the number of trials significantly improves the performance of this simulation
model for the range 0 s    0.1 s. From the simulation results, the RMSE between

the absolute TCFs of the deterministic SP model and the reference model is

approximately 35.1  10 3 , 18.2  10 3 , and 11.2  10 3 , respectively, while the MD of
the absolute TCFs of the deterministic SP model from the corresponding theoretical
values is approximately 59.2  103 , 34.6  103 , and 18.9  103 , respectively.

Fig. 62 depicts the absolute error between the absolute SCFs of the reference
model and the deterministic SP model after V  50 simulation trials, when V1  5,
V2  5, and V3  2 and for N  240, when N1  10, N2  8, and N3  3. From the
simulation results, the RMSE between the absolute SCFs of the deterministic SP
model and the reference model is approximately 1.8  10 3 , while the MD of the

106

Chapter 4
absolute SCFs of the deterministic SP model from the corresponding theoretical
values is approximately 4.8  103. Consequently, the deterministic SP simulation

model closely matches the TCF of the reference model, while approximates the SCF
of the reference model with high precision.
1
Reference model
0.9 Deterministic SP, V=1
Deterministic SP, V=10
0.8
Deterministic SP, V=50
0.7
0.6
0.5
0.4
0 0.02 0.04 0.06 0.08 0.1

τ (s)
Figure 61: Comparison between the TCFs of the reference model and the deterministic SP
simulation model.
‐3
x 10
4
Absolute error
0
0
1
200
δ /λ 150
R 2 100
50 δ /λ
T
3 0
deterministic SP simulation model.
107

Chapter 4
Fig. 63 illustrates the absolute value of the TCFs of the reference and the

statistical model averaged over 50 simulation trials for N  240, when N1  10,
N2  8, and N3  3. One observes that the TCF of the statistical model almost
perfectly matches the theoretical one for the range 0 s    0.1 s. From the
simulation results, the RMSE between the absolute TCFs of the statistical model
and the reference model is approximately 2.4  10 3 , while the MD of the absolute

TCFs of the statistical model from the corresponding theoretical values is

approximately 7.6  103.
Fig. 64 demonstrates the absolute error between the absolute SCFs of the
reference model and the statistical model after 50 simulation trials. From the
simulation results the RMSE between the absolute SCFs of the statistical model and
the reference model is approximately 1.6  10 3 , while the MD of the absolute SCFs

of the statistical model from the corresponding theoretical values is approximately

is approximately 4.7  103. Hence, the statistical simulation model approximates
the space‐time correlation properties of the reference model in all cases.
Tables 9 and 10 thoroughly investigate the performance of the statistical
simulation model for 0 s    0.1 s, 0  T  200 , and 0   R  3. Table 9

examines the accuracy of the simulation model for different values of the model
parameters that control the location of the scatterers inside the cylinder. The
performance evaluation is realized in terms of the RMSE between the absolute
TCFs and SCFs of the statistical model and the reference model, as well as the MD
of the absolute TCF and SCF of the statistical model from the corresponding
theoretical values. One observes that all quantities are very small in the scenarios
simulated. The close agreement between the channel statistics confirms the utility
of the simulation model and the method used to estimate the model parameters.
Table 10 studies the efficiency of the simulation model and tests its performance
for different complexity, i.e., different number of simulation trials, in terms of the
RMSE and the MD. From the results, adequate statistics can be achieved with only
10 simulation trials. It is also evident that by averaging over more simulation trials,
108

Chapter 4
the performance of the simulation model improves. From the entire simulation
results, the statistical simulation model outperforms the other simulation models
and provides the best solution for the design of simulation models for HAP‐MIMO
mobile fading channels.
1
Reference model
Statistical model
0.8
0.6
0.4
0.2
0 0.02 0.04 0.06 0.08 0.1
τ (s)
Figure 63: Comparison between the TCFs of the reference model and the statistical
simulation model.
‐3
x 10
8
Absolute error
0
0
1 200
150
δ 2 100
R 50 δ
3 0 Τ
statistical simulation model.
109

Chapter 4
TABLE 9
ACCURACY EVALUATION OF THE STATISTICAL SIMULATION MODEL
Values of the TCF SCF
model
parameters RMSE MD RMSE MD
0 2.5  103 8.2  10 3 1.8  103 6.2  103

k
3 2.3  103 7.4  103 1.5  103 5.1  103

8 2.1  103 6.8  10 3 1.3  103 4.9  103

30 2.3  103 7.4  103 1.5  103 5.1  103

60 2.3  103 8.2  10 3 1.5  103 5.6  103

90 2.3  103 7.5  103 1.4  10 3 4.9  103

0.01 2.3  103 7.4  103 1.5  10 3 5.1  103


0.05 2.1  103 6.8  10 3 1.2  10 3 4.5  103

0.1 1.8  10 3 5.4  103 0.8  103 3.7  103

8 m 2.2  10 3 6.9  10 3 1.4  10 3 4.8  103

HS ,mean 17 m 2.3  103 7.4  103 1.5  103 5.1  103

23 m 2.4  10 3 7.7  10 3 1.6  103 5.5  103

TABLE 10
EFFICIENCY EVALUATION OF THE STATISTICAL SIMULATION MODEL
Simulation TCF SCF
trials
RMSE MD RMSE MD
10 5.3  10 3 14.6  10 3 3.4  103 12.4  10 3
20 3.8  10 3 11.3  10 3 2.5  103 7.2  103
30 3.2  10 3 9.3  103 2.2  10 3 6.5  10 3
40 2.8  10 3 8.7  10 3 1.9  10 3 5.8  103
50 2.3  10 3 7.4  103 1.5  103 5.1  103
100 1.6  103 5.2  10 3 1.2  103 3.3  10 3
110

Chapter 4
Finally, Fig. 65 compares the ergodic capacity defined in (48), (49) against SNR
obtained using the reference model and the simulation models for T  60 ,
 R  0.5 , and a normalized Doppler frequency fn  fR ,maxTs  0.01. Due to the
different way of generation of HNLoS in (49), the reference model is referred to as
the non‐physical model (see Chapter 3, Eq. (52)), while the simulation models are
referred to as the physical models, since the elements of H can be directly
obtained using the procedures described in this chapter. The results show very
good agreement between the non‐physical and the physical models [163].
20
Reference model
Deterministic model, N1=30, N2=20, N3=5
Deterministic SP model, N1=10, N2=8, N3=3, V1=5, V2=5, V3=2
15
Statistical model, N1=10, N2=8, N3=3, Ntrials=50
10
0
0 5 10 15 20 25 30
SNR (dB)
Figure 65: Comparison between the ergodic capacities obtained using the reference model
and the simulation models.
4.6 Summary
In this chapter, methodologies for simulating HAP‐MIMO Ricean faded channels
have been presented, under 3‐D non‐isotropic scattering conditions. Specifically,
parametric deterministic and statistical simulation models based on the SoS
principle have been developed and the reference non‐realizable HAP‐MIMO
channel model has served as the initial point. The proposed deterministic
simulation model enables the generation of multiple deterministic waveforms and
111

Chapter 4
provides an adequate approach for the simulation of HAP‐MIMO channels in a
single simulation trial. Nevertheless, its performance can only be improved by
increasing the number of sinusoids, i.e., the number of the discrete scatterers. An
alternative advantageous deterministic design method using the principle of SP has
also been introduced. This method keeps the computational costs low, while
significantly improves the performance simply by averaging across several
simulation runs. Finally, a method that applies Monte Carlo techniques has been
proposed to construct a statistical simulation model with at least one parameter as
random variable. Since, this method statistically converges to the desired statistical
properties, it is important to determine the number of simulation trials needed to
achieve a specific convergence level.
By assuming that the AoA of the scattered waves, the distance between the user
and the scatterers in azimuth domain and the height of the scatterers follow the
von Mises, the hyperbolic and the log‐normal distribution, respectively, the
performance and the complexity of the proposed simulation models have been
demonstrated by extensive simulation results. Based on these results, one
concludes that the simulation models accurately and efficiently reproduce the SCFs
and TCFs of the reference model and provide similar channel capacity estimations.
Although the proposed deterministic simulation model has the potential of
becoming a standard procedure due to its simplicity, efficiency, and reproducibility,
the slightly complex statistical simulation model provides the highest performance
with a relative small number of simulation trials, while the deterministic SP
simulation model provides a good trade‐off between complexity and performance.
Heretofore, 3‐D narrowband reference and simulation models for HAP‐MIMO

channels have been developed. Notwithstanding, wideband channel modeling has
received significant attention worldwide due to the increasing demand for high‐
speed broadband communications. To completely characterize HAP‐MIMO
multipath fading channels, it is necessary to develop realistic wideband HAP‐MIMO
channel models. Hence, Chapter 5 deals with this objective and proposes 3‐D
reference and simulation models for wideband HAP‐MIMO channels.
112

Chapter 5
Chapter 5
Three‐Dimensional Modeling and Simulation of
Wideband HAP‐MIMO Channels

5.1 Introduction
Prospective 4G advanced multimedia and broadcasting communications services
demand increased bandwidth and enhanced data rates. These services are mainly
based on wideband transmissions and can be potentially upgraded if MIMO
techniques are used. Thus, realistic wideband HAP‐MIMO channel models are
essential for the design and concise evaluation of future HAP‐based systems.
Wideband effects usually impose a time‐variant impulse response, which can be
modeled by a tapped delay line (TDL) with each tap defined using different weights
and distributions [184]. In agreement with the TDL concept, a statistical switched
(state‐oriented) broadband channel model for HAP‐based communications systems
was presented in [111], while a model based on semi‐Markovian processes was
presented in [112]. These models allow for convenient and time‐efficient system
analysis of the links between HAPs and terrestrial stations, when the channel
characteristics are dynamic and vary in time. Nevertheless, the propagation
environment necessitates fundamental limitations on the performance. Hence,
when terrain and scattering distributions are available, physical‐geometrical
channel modeling is preferred to ensure model accuracy and versatility.
To the author’s best knowledge, only terrestrial geometry‐based wideband

MIMO channel models are available. Two‐dimensional (2‐D) models for non‐
isotropic scattering environments were presented in [185]‐[187] and assumed that
the radio waves travel only horizontally, which is an acceptable approximation for
certain scenarios. As described in Chapter 3, HAP‐based channel models should
113

Chapter 5
consider 3‐D scattering conditions. In [188] and [189], a 3‐D wideband channel
model was presented and introduced an elevation angle for the incidence of the
radio waves in addition to the azimuth angle. Nevertheless, this model is not
applicable to HAP‐based systems. In particular, this model assumes that the
distance between the user and the scatterers in the azimuth plane is constant.
Moreover, the derivation of this model and its statistical properties was ground on
the simplified assumption that the maximum elevation angle of arrival of the
scattered waves is smaller than 20°, which typically represents terrestrial mobile‐
to‐mobile systems (e.g., two cars driving through streets) [190]. If “above rooftop”
propagation is present as in HAP‐based systems, the maximum elevation angle
approximates 90° [191]. Hence, other channel modeling approaches are necessary.
In Chapters 3 and 4, 3‐D reference and simulation models for HAP‐MIMO
multipath fading channels have been proposed. The realization of these models is
limited to narrowband, i.e., frequency‐nonselective, communications. In particular,
it was assumed that the propagation delays of all incoming scattered waves are
approximately equal and small in comparison to the data symbol duration. In
wideband communications, the data symbol duration is small and multipath delay
spread may be introduced. Therefore, the propagation delay differences cannot be
neglected. In this chapter, the 3‐D reference model for narrowband HAP‐MIMO
channels is extended with respect to frequency‐selectivity and a 3‐D reference
model for wideband HAP‐MIMO channels is derived. The proposed model utilizes
carrier frequencies well below 10 GHz. Hence, both LoS and NLoS links should be
considered, while rain effects are insignificant [1]. To describe the reference model,
a 3‐D geometrical model for wideband HAP‐MIMO channels is introduced, referred
to as the “two concentric‐cylinders" model. This model is a modified version of the
cylinder model for narrowband HAP‐MIMO channels in Chapter 3. Several critical
parameters are considered, e.g., the elevation angle of the platform, the array
configuration, the Doppler and delay spread, and the distribution of the scatterers.
Based on the reference model, the space‐time‐frequency correlation function
(STFCF), the space‐Doppler power spectrum (SDPS), and the power space‐delay
114

Chapter 5
spectrum (PSDS) are derived for a 3‐D non‐isotropic scattering environment.

Numerical results demonstrate the theoretical derivations. This reference
wideband model assumes an infinite number of scatterers, which prevents
practical implementation. Thus, 3‐D deterministic and statistical SoS simulation
models for wideband HAP‐MIMO channels are proposed and their statistical
properties are verified by simulations.
The remainder of this chapter is organized as follows. Section 5.2 presents the
wideband HAP‐MIMO system model, while Section 5.3 proposes the 3‐D reference
model for wideband HAP‐MIMO channels and constructs the corresponding input
delay‐spread and time‐variant transfer functions. Section 5.4 studies the statistical
properties of this reference model and presents numerical results. Section 5.5
details the simulation models and provides simulation results to verify their
accuracy and efficiency. Finally, Section 5.6 summarizes this chapter.
5.2 The Wideband HAP‐MIMO System Model
This chapter considers a downlink wideband HAP‐MIMO communication

channel with nT transmit at the SBS and nR receive antenna elements at the TMS.
All antennas are fixed, omni‐directional, and are numbered as 1  p  q  nT and
1  l  m  nR , respectively. The complex low‐pass equivalent time‐variant channel
response between SBS and TMS is given by hpl  t ,  , and the complex low‐pass
equivalent input‐output relation for the frequency selective HAP‐MIMO fading

channel is given by
r  t   H  t ,   s  t   n  t  , (95)
where H  t ,    hp ,l  t ,     nR nT is the matrix of the input delay‐spread

nR nT
functions [160], s  t  nT 1 is the transmitted signal vector, r  t   nR 1 is the
received signal vector, n  t   nR 1 is the noise vector, which denotes the AWGN
115

Chapter 5
at the receiver branches, and   denotes convolution. The entries of the noise
vector are i.i.d. complex Gaussian random variables with zero‐mean and variance
N0 , where N0 is the noise PSD.
5.3 A Reference Model for Wideband HAP‐MIMO Channels
In this section, a 3‐D GBSB reference model for wideband HAP‐MIMO channels
is introduced and its geometrical characteristics are described. Then, the input
delay‐spread and time‐variant transfer functions are derived as a superposition of
the LoS and NLoS rays.
The geometrical model proposed in Chapter 3 considers that the waves emitted
from the SBS antenna elements travel over paths with different lengths and
impinge the TMS array elements from different directions due to the scattering
conditions within a cylinder. This section proposes a modified “two concentric‐
cylinders” version of this geometry [192]. The geometrical characteristics of the
proposed model and the definition of the Cartesian coordinate system are
discussed in Figs. 66‐69 for a 2  2 HAP‐MIMO channel, which is the basis to study
ULAs with arbitrary number of antennas. Fig. 66 shows the LoS paths of the 3‐D
geometrical model, while Fig. 67 presents the projection of this model to the x ‐y
plane. Figs. 68 and 69 show the NLoS paths for the channel in Figs. 66 and 67. For
ease of reference, the parameters of the proposed reference model for wideband
HAP‐MIMO channels are summarized in Table 11.
The proposed reference model for wideband HAP‐MIMO channels assumes that
N   fixed effective local scatterers surround the TMS. Then, the nth scatterer is
designated by S   and its projection to the x ‐y plane is denoted by S   . The
n n
n
 n

height of S   is d S   , S    HS   HS ,min , HS ,max  , and the distance between
n n

S   and O is d  S   ,O   R    R

n n
S
n
S ,min,R S ,max 
,  
where HS ,min  min HS  ,
n
 max H    , R  minR    , and R  

 max RS  .
n n n
H S ,max S S ,min S S ,max

116

Chapter 5
p
z
q OT
RS ,max
RS ,min
Hc
HS ,max
R l
T
y vT HS ,min
p
OR l vR
T R
T m R
O   aLoS
Rl O x

m
q
Figure 66: The LoS paths of the 3‐D geometrical wideband model for a 2  2 HAP‐MIMO
channel.
y vT
p vR
l
R
Rl
T  T aLoS
R
O O x

m
q
RS ,min
D RS ,max

Figure 67: The LoS paths of the projection of the 3‐D geometrical wideband model for a
2  2 HAP‐MIMO channel.
117

Chapter 5
p
z
q OT
RS ,max
RS ,min
Hc
S 
n
HS ,max
HS   S 
n n
R l
y vT HS ,min
p S  
n
OR
T l vR
RS  aR n  R
n
m R
O  T n O x
aT

m
q
Figure 68: The NLoS paths of the 3‐D geometrical wideband model for a 2  2 HAP‐MIMO
channel.
y
p vT S  
n
vR
T RS n  aR n  l
T n R R
aT
O O x

m
q
RS ,min
D RS ,max

Figure 69: The NLoS paths of the projection of the 3‐D geometrical wideband model for a
2  2 HAP‐MIMO channel.
118

Chapter 5
TABLE 11
DEFINITION OF THE PARAMETERS USED IN THE REFERENCE WIDEBAND MODEL
D The distance between the centre O of the projections of the SBS antenna
elements to the x ‐y plane and the lower centre O of the cylinder.
RS ,max , HS ,max The maximum distance between O and S   , and the maximum height of
n
the scatterers, respectively.
RS ,min , HS ,min The minimum distance between O and S   , and the minimum height of the
n
scatterers, respectively.
T ,  R The spacing between two adjacent antenna elements at the SBS and TMS,
respectively.
T , R The orientation of the SBS and TMS antenna array in the x ‐y plane (relative
to the x ‐axis), respectively.
 T ,  R The elevation angle of the p th SBS and l th TMS antenna element, relative to

the x ‐y plane, respectively.
vT , vR The velocities of the SBS and TMS, respectively.
 T ,  R The moving directions of the SBS and TMS, respectively.
fT ,max , fR ,max The maximum Doppler frequency shift of the SBS and TMS, respectively.
T The elevation angle of the SBS relative to OR .
HT , HR , HS  The height of the SBS, the TMS, and the n th scatterer, respectively.

n
Rl
aLoS The azimuth angle of arrival of the LoS paths.
aT  , aR  The azimuth angle of arrival and the azimuth angle of departure at/from the

n n
n th scatterer, respectively
RS  The distance between O and S   .

n n
 S  The elevation angle of the n th scatterer relative to OR .
n
 S ,min ,  S ,max The minimum and maximum elevation angle of the n th scatterer relative to
OR .
  n  , g  n  The random phase and the amplitude introduced by the n th scatterer,
respectively.
 , k The mean azimuth angle at which the scatterers are distributed in the x ‐y
plane and the spread of the scatterers around this angle, respectively.
a The spread of the scatterers around the TMS.
HS ,mean ,  The mean and standard deviation of scatterer’s height, respectively.
119

Chapter 5
Based on the modified model geometry, the scatterers occupy the volume
between two concentric‐cylinders with height Hc  HS ,max  HS ,min and radius RS ,min
and RS ,max , respectively. This new geometry degrades to the corresponding one of

the narrowband channel model in Chapter 3, when HS ,min  RS ,min  0. Thus, the

geometry of the narrowband model is a special case of the geometry of the
wideband one. Using (13), this new geometry enables the determination of the

minimum  S ,min and the maximum  S ,max values of  S  as follows
n

 S ,min  arctan  HS ,min / RS ,max  , (96)

 S ,max  arctan  HS ,max / RS ,min . (97)

Note that shadowing within the scattering region is neglected in this model.

5.3.1 Input Delay‐Spread and Time‐Variant Transfer Functions
Considering Ricean fading, the input delay‐spread function of the link p‐l is a
superposition of the LoS and NLoS rays and can be written as follows [188], [189]
hpl  t ,   hpl ,LoS  t ,   hpl ,NLoS  t ,  , (98)
where
j 2
 pl K pl  d  p ,l  j 2 t  F
T ,LoS FR ,LoS 
hpl ,LoS  t ,   e  e     LoS  , (99)
K pl  1
 pl
1 N n 
g   e j e 
j 2 

d p , S   
 n  d S  n  ,l 
hpl ,NLoS  t ,    n 
lim
K pl  1 N  N n1
e
j 2 t  FT ,NLoS FR ,NLoS 
 t     NLoS
 n
 , (100)
where  pl denotes the transmitted power of the subchannel p‐l , K pl denotes the
Ricean factor of the subchannel p‐l , g  n  represents the amplitude of the nth
120

Chapter 5
scattered wave,   n     ,  is the phase introduced by the nth scatterer,  LoS
 
and  NLoS
n
are the propagation delays (travel times) of the waves associated with
the LoS and NLoS links, respectively, and  denotes the Dirac delta function. In

n n
   
addition, the parameters d  p, l  , d p, S   , d S   , l , FT ,LoS , FR ,LoS , FT ,NLoS , and

FR ,NLoS are defined in (17)‐(24), (27), (28), (31), and (32), respectively.
From Fig. 66, since HT  HR the propagation delay from the SBS to the TMS, i.e.,
the minimum propagation delay, is given by
D
 LoS   min  c cos  , (101)
0 T
where c0 is the speed of light. From Fig. 68, considering that HS ,max  HT and
RS   D , and using the cosine law, the travel time of a wave originated from SBS,
n

reflected from S   and received by TMS is given by
n
   2DR  cos   a   
2
D2  RS 
n n n
RS 
S R n
 n
 NLoS  . (102)
c0 cos T c0 cos  S 
n
Using the approximation 1  x  1  x / 2 for small x , (102) becomes
RS  cos aR  RS 

n n n
 
 NLoS   LoS 
n
 . (103)
c0 cos T c0 cos  S 
n
Moreover, using (13), (103) and considering that cos  arctan x   1 / 1  x 2 ,
cos aR   1, RS   RS ,max , and HS   HS ,max , the maximum possible propagation
n n n
delay can be derived as follows
RS ,max RS2,max  HS2,max

 max   LoS   . (104)
c0 cos T c0
121

Chapter 5
Note that the minimum propagation delay  LoS is common to all paths. Thus, it can
be neglected, without loss of generality. Then, the relative minimum and maximum
propagation delays, as well as the relative propagation delay associated with the
NLoS link are given, respectively, by
 r ,LoS   r ,min  0, (105)
 r ,max   max   LoS , (106)
 r ,NLoS
  
  NLoS   LoS . (107)
n n
From (106), RS ,max can be easily determined, considering that HS ,max and  r ,max are

known (after measurements campaigns in different propagation environments and
carrier frequencies). Estimated values of  r ,max achieved with conventional aircrafts
are in the order of hundreds of nanoseconds, which are much smaller than the
corresponding values in terrestrial systems [193]. This is possibly due to the fact
that the effective scatterers in a terrestrial system might be expected to be more
widely distributed than those for a flying platform system operating at high
elevation angles. Using (106), Fig. 70 demonstrates  r ,max for T  60 and
different RS ,max and HS ,max .

1.5
(μsec)
1
r,max
0.5
τ
0
10
20 100
30 80
60
H (m) 40 40
S,max R (m)
50 20 S,max
Figure 70: The relative maximum propagation delay for different maximum scatterer
height and maximum distance in the azimuth plane between the TMS and the scatterers.
122

Chapter 5
Finally, the time‐variant transfer function is the FT of the input delay‐spread
function [184] and can be written as
Tpl  t , f    hpl  t ,   Tpl ,LoS  t , f   Tpl ,NLoS  t , f  , (108)
where the LoS and NLoS components of the time‐variant transfer function are,
respectively,
j 2
 pl K pl  d  p ,l  j 2 t  F
T ,LoS  FR ,LoS   j 2 f LoS
Tpl ,LoS  t , f   e  e e , (109)
K pl  1
 pl 1 N  n  j  n  
j 2 
  
d p ,S   d S   ,l 
n n

Tpl ,NLoS  t , f   lim g e e
K pl  1 N  N n1
  
 n
j 2 t  FT ,NLoS FR ,NLoS   j 2 f NLoS
e e . (110)
5.4 Statistical Properties of Wideband HAP‐MIMO Channels
Considering a 3‐D non‐isotropic scattering environment, the STFCF of the

complex faded envelope is derived. Then, using the STFCF, the SDPS and the PSDS
are also derived.
5.4.1 Space‐Time‐Frequency Correlation Function
The normalized STFCF between two time‐variant transfer functions Tpl  t , f  , and
Tqm  t , f  is defined as [184], [186], [194]
 Tpl  t , f  Tqm  t  t , f  f  
Rpl ,qm  T , R , t , f   . (111)
 pl qm
Since the number of local scatterers is infinite, central limit theorem implies that
both Tpl ,NLoS  t , f  and Tqm ,NLoS  t , f  are zero‐mean complex Gaussian random

processes, i.e., E Tpl ,NLoS  t , f    E Tqm ,NLoS  t , f    0. Then, (111) can be written as

follows (see Eq. (34) in Section 3 for a similar derivation)
123

Chapter 5
Rpl ,qm T , R , t , f   RplLoS,qm T , R , t , f   RplNLoS

,qm   T , R , t , f  , (112)
where RplLoS,qm  T , R , t , f  and RplNLoS

,qm   T , R , t , f  denote the STFCF of the LoS
and the NLoS components, respectively. Using (17), (20), (22), (27), (28), (101),
Rl
(109), and (111), and the approximate relation aLoS  aLoS
Rm
  for  T  D , the

STFCF of the LoS component is derived as follows
RplLoS,qm  T , R , t , f 
j 2  q  p T cos T  m l  R cos R cos R 
K pl Kqm
 cos T
 e
 K pl  1 Kqm  1
D
 j 2f
j 2t  fT ,max cos  T  fR ,max cos  R  c0 cos T
e e . (113)
Moreover, using (13), (18)‐(24), (31), (32), (110), and (111), the STFCF of the NLoS
component can be written as
,qm   T , R , t , f 
RplNLoS
j 2  q  p T cos T j 2  q  p T sinT RS
 n  sin a n 
 N R
1 1  n 2
 lim   g  e  cos T
e D cos T
 K pl  1 Kqm  1 N  N
n1


  n n 
j 2  m  l  R sin R sinarctan HS / RS  
n   n n 
j 2  m l  R cos R cos R cos aR cos arctan HS / RS  
     
 e  e 
 n   n  n   R n  sin  sin a n  
j 2  m  l  R sinR cos R sin aR cos arctan HS / RS  
   j 2tfT ,max  S T R cos  
T
 D
 e  e  
 n   n n 
j 2tfR ,max cos aR  R  cos arctan HS / RS  
    
 e
 
  n
RS cos aR
 n  n
RS 
D
 j 2f    
 c0 cos T c0 cos T   n  n 
c0 cos arctan HS / RS   
     
e . (114)
The infinite number of local scatterers in the reference model also enables the
replacement of the discrete variables aR  , RS  , and HS  with the continuous
n n n
random variables aR , RS , and HS with joint pdf f  aR , RS , HS  . According to Fig. 68,
124

Chapter 5
aR  , RS  , and HS  are independent. Thus, the joint pdf f  aR , RS ,  S  can be
n n n

decomposed to f  aR  f  RS  f  HS  .
In agreement with the narrowband model in Chapter 3, the von Mises pdf in
(39) is used to characterize aR . In addition, the following truncated7 hyperbolic pdf
is used to characterize RS for the range RS ,min  RS  RS ,max
f  RS 
fT  RS   , (115)
FRS  RS ,max   FRS  RS ,min 
where f  RS  and FRS  RS  are defined in (40) and (60), respectively. Furthermore,
the following truncated log‐normal pdf is adopted to characterize HS for the range
HS ,min  HS  HS ,max
f  HS 
fT  HS   , (116)
FHS  HS ,max   FHS  HS ,min 
where f  HS  and FHS  HS  are defined in (41) and (63), respectively.
Finally, using the equality


 e
a sin c  b cos c 
dc  2 I0  
a2  b2 [161, eq.3.338‐
4], and (39), (115), and (116), the STFCF of the NLoS component becomes
 
HS ,max RS ,max
RplNLoS
,qm T , R , t , f   w1   w2I0 w32  w42 dRS dHS , (117)
HS ,min RS ,min
where w1 , w2 , w3 , and w 4 are, respectively
  q  p T cos T fD 

1/2 j 2  tfT ,max cos  T  
a  K pl  1 Kqm  1 
 cos  c cos T 
e  T 0
w1  , (118)
 2 I0  k   tanh  aRS ,max   tanh  aRS ,min   FHS  HS ,max   FHS  HS ,min  

7
A truncated distribution is a conditional distribution that results from restricting the domain of some other probability
distribution.
125

Chapter 5
  m l  R sin R sinarctan HS / RS   fRS   1 2  HS  

j 2     ln  

  c0 cos arctan HS / RS     2 2  HS ,mean  
e e
w2  , (119)
HS cosh2  aRS 
j2  q  p   T sinT RS j2tfT ,max sin T RS

w3  
D cos T D

j2  m  l   R sinR cos R cos arctan  HS / RS  


 j2tfR ,max sin R cos arctan  HS / RS    k sin  , (120)
j2  m  l   R cosR cos R cos arctan  HS / RS  

w4 

 j2tfR ,max cos  R cos arctan  HS / RS  
RS
 j2f  k cos . (121)
c0 cos T
The double integral in (117) has to be evaluated numerically, since there is no
closed‐form solution.
Finally, the STFCF between two time‐variant transfer functions Tpl  t , f  , and
Tqm  t , f  becomes a summation of the STFCFs of the LoS and the NLoS
components defined in (113) and (117)‐(121), respectively.
5.4.2 Space‐Doppler Power Spectrum
Considering a wide‐sense stationary uncorrelated scattering (WSSUS) time‐

varying HAP‐MIMO channel, the SDPS is derived by applying FT to the STCF, i.e.,
Rpl ,qm T , R , t , f  0  , in (112). Other approaches for determining the Doppler

spectrum of terrestrial and satellites systems by assuming 3‐D scattering were
previously proposed in [188], [189], [195]‐[198].
The SDPS of the LoS component is obtained by calculating the FT of the STCF of
the LoS component in (113) as follows
126

Chapter 5


SplLoS,qm T , R ,v   t RplLoS,qm T , R , t , f  0  
2  q  p T cos T  m l  R cos R cos R 
K pl K qm j
 cos T
 e

 K pl  1 Kqm  1
  v  fT ,max cos  T  fR ,max cos  R  , (122)
where v is the Doppler frequency shift relative to the carrier frequency.
The SDPS of the NLoS component is obtained by calculating the FT of the STCF of
the NLoS component in (117)‐(121)

SplNLoS NLoS
 
,qm   T , R , v   t Rpl ,qm   T , R , t , f  0  . (123)
For f  0, the Bessel function I0  
w32  w42 of (117) can be written as follows
I0  w32  w42   J s

f  0
0 1  t  s2 2  s32  , (124)

where
s1  2 fR ,max cos arctan  HS / RS   , (125)
q1  q2  q3
s2  , (126)
fR ,max cos arctan HS / RS  
q4  q5  q6
s3  , (127)
fR ,max cos arctan HS / RS  
q1 
 q  p T sinT RS sin R , (128)
D cos T
 m  l  R cos R cos R   R  cos arctan HS / RS 

q2  , (129)

jk cos     R 
q3  , (130)
2
q4 
 q  p T sinT RS cos  R , (131)
D cos T
127

Chapter 5
 m  l  R cos R sinR   R  cos arctan HS / RS 

q5  , (132)

jk sin     R 
q6  . (133)
2
Using (124)‐(133), the SDPS of the NLoS component becomes
HS ,max RS ,max 
 j 2t  v  fT ,max cos  T 
SplNLoS
,qm   T , R , v     A e
HS ,min RS ,min 

 J0  s1  t  s2 2  s32  d tdRS dHS , (134)
 

where
  q  p T cos T  m l  R sin R sinarctan HS / RS     1 2  HS  

j 2      2 ln 
  cos T    2  
    HS ,mean  
e e
A
HS cosh2  aRS  2 I0  k   K pl  1 Kqm  1
a
 . (135)
FH  HS ,max   FH  HS ,min    tanh  aRS ,max   tanh  aRS ,min  
 S S  
It is well known that e  jx  cos x  j sin x (Euler’s formula), the integral of the
product of an odd function, i.e., sin x , and an even function, i.e., cos x and J0  x  ,
from  to  is equal to zero, the product of two even functions is an even
function, and the integral of an even function from  to  is twice the integral
from 0 to . Under these considerations and using the equality
0

   
J0 a x 2  z 2 cos  bx  dx  cos z a2  b2 / a2  b2 [161, eq. 6.677‐3], (134)

becomes
HS ,max RS ,max j 2 s2  v  fT ,max cos  T 

Ae
SplNLoS
,qm   T , R , v    
 HS / RS   v  fT ,max cos  T 
2
HS ,min RS ,min  fR2,max cos2 arctan
 2
 cos 2 s3 fR2,max cos2 arctan  HS / RS     v  fT ,max cos  T   dRS dHS . (136)
 

128

Chapter 5
closed‐form solution. One observes that (136) is applicable for the range
0  v  fT ,max cos  T  fR ,max cos arctan  HS / RS  

 0  v  fT ,max cos  T  vmin , (137)

where

vmin  min fR ,max cos arctan  HS / RS  
HS ,RS

 fR ,max cos arctan  HS ,max / RS ,min   . (138)

However, the SDPS should be sketched for the range
0  v  fT ,max cos  T  vmax ,

(139)
where


vmax  max fR ,max cos arctan  HS / RS  
HS ,RS

 fR ,max cos arctan  HS ,min / RS ,max   . (140)

Hence, the following function is finally proposed as the SDPS of the NLoS
component
x1 R,max j 2 s2  v  fT ,max cos  T 

Ae
SplNLoS
,qm   T , R , v    
arctan  HS / RS     v  fT ,max cos  T 
2
HS ,min x2  fR2,max cos2
 2
 cos 2 s3 fR2,max cos2 arctan  HS / RS     v  fT ,max cos  T   dRS dHS , (141)
 

where
  q  p T cos T  m l  R sin R sinarctan HS / RS     1 2  HS  

j 2      2 ln   
  cos T    2
 HS ,mean  
e  e 
A 
a
 , (142)
FH  x1   FH  HS ,min    tanh  aRS ,max   tanh  ax2  
 S S  
129

Chapter 5
HS ,max , 0  v  fT ,max cos  T  vmin

x1   , (143)
   
 S ,min
R B , vmin v fT ,max cos T vmax
RS ,min , 0  v  fT ,max cos  T  vmin

x2   , (144)
 S ,max
H / B , vmin  v  fT ,max cos  T  vmax

  v  fT ,max cos  T 
B  tan  arccos   . (145)
  fR ,max 
   
5.4.3 Power Space‐Delay Spectrum
Considering a WSSUS frequency‐varying HAP‐MIMO channel, the PSDS is derived
by applying IFT to the space‐frequency correlation function (SFCF) in (34) i.e.,
Rpl ,qm T , R , t  0, f  . Another approach for determining the delay spectrum of
terrestrial systems by assuming 3‐D scattering was previously proposed in [188],
[189].
The relative PSDS of the LoS component can be obtained by calculating the IFT
of the SFCF of the LoS component in (113) as follows
PplLoS 1 LoS

,qm   T , R , r   f Rpl ,qm  T , R , t  0, f  
2  q  p T cos T  m  l  R cos R cos R 
K pl Kqm j
 e  cos T
  r  , (146)
 K pl  1 Kqm  1
where  r     LoS  0, r ,max  is the relative propagation delay.
The relative PSDS of the NLoS component can be obtained by calculating the IFT
of the SFCF of the NLoS component in (117)‐(121)

PplNLoS 1 NLoS
 
,qm  T , R , r   f Rpl ,qm  T , R , t  0, f  . (147)
For t  0, the Bessel function I0  
w32  w42 of (117) can be written as follows
130

Chapter 5
I0  w32  w42   J 2 p

t  0
0 1  f  p2 2  p32  , (148)

where
RS
p1  , (149)
c0 cos T
 m  l  R cosR cos R c0 cos T cos arctan  HS / RS 

p2 
 RS
jkc0 cos  cos T
 , (150)
2 RS
 m  l  R sinR cos R c0 cos T cos arctan HS / RS 

p3  .
 RS

 q  p T sinT c0  jkc0 sin  cos T . (151)
D 2 RS
Using (148)‐(151), (147) becomes

HS ,max RS ,max 
J0  2 p1  f  p2 2  p32  d fdRS dHS ,
j 2f  r  p4 
,qm   T , R , r  
PplNLoS   A e
HS ,min RS ,min 
 

(152)
where
RS
p4  . (153)
c0 cos arctan  HS / RS  

Using the equality [161, eq. 6.677‐3] and after extensive calculations, the relative
PSDS of the NLoS component is derived as follows

cos  2 p3 p12   r  p4  
j 2 p2  r  p4  2
HS ,max RS ,max Ae
  dR dH . (154)
,qm   T , R , r  
PplNLoS   S S
 p12   r  p4 
2
HS ,min RS ,min
closed‐form solution.
131

Chapter 5
One observes that (154) is applicable for the range

 r  p4  p1

 p4  p1   r  p4  p1
 1   r  2 , (155)

where
 1  max p4  p1
HS ,RS
RS2,min  HS2,max RS ,min

  , (156)
c0 c0 cos T
 2  min p4  p1 
HS ,RS
RS2,min  HS2,min RS ,min

  . (157)
c0 c0 cos T
However, the relative PSDS should be sketched for the range
0   r   r ,max ,

(158)
where  r ,max is defined in (106). Hence, the following function is finally proposed as
the relative PSDS of the NLoS component

Ae 2  r 4  cos  2 p3 p12   r  p4  
j 2 p   p 2
y2 RS ,max
  dR dH ,
PplNLoS
,qm T , R , r     S S (159)
 p1   r  p4 
2 2
y1 y3

where
  q  p T cos T  m l  R sin R sinarctan HS / RS     1 2  HS  
j 2     ln  
  cos T    2 2  HS ,mean  
    
e e
A 
a
 , (160)
FH  y2   FH  y1    tanh  aRS ,max   tanh  ay3  
 S S  
132

Chapter 5
 HS ,min , 0   r   2

y1   , (161)
 tan 2
 R 2
T S ,min  2c 
0 r sec  R
T S ,min  c 
2 2
0 r ,  2   r   r ,max
 tan T RS ,min  2c0 r sec T RS ,min  c0 r , 0   r   1

2 2 2 2
y2   , (162)

 H S ,max ,  1   r   r ,max
 c  sec   c2 2  H 2
S ,max tan T
2
 0 r T 0 r
, 0   r   1
 tan2 T

y3   RS ,min ,  1   r   2 . (163)

 c0 r sec T  c0 r  HS ,min tan T
2 2 2 2
 ,  2   r   r ,max
 tan 2
 T
5.4.4 Numerical Results
This section demonstrates the theoretical derivations described above. Figs. 71‐
74 depict the absolute STFCF for a 3‐D non‐isotropic scattering environment.
Unless indicated otherwise, the values of the model parameters used to obtain the
curves are nT  nR  2, K pl  Kqm  0 for Rayleigh fading, K pl  Kqm  3 dB for

Ricean fading, HT  20 km, T  60 , T  60 , R  30 ,  R  20 , a  0.01,
RS ,min  7 m, RS ,max  140 m, HS ,min  5 m, HS ,max  50 m, HS ,mean  17.6 m,

  0.31, k  4,   90 , fT ,max  300 Hz, fR ,max  100 Hz,  T  30 , and  R  60.
Fig. 71 shows the absolute frequency correlation function (FCF)
Rpl ,qm T  0, R  0, t  0, f  of a HAP‐SISO Rayleigh channel for different T .
One observes that the correlation decreases as T increases. From Fig. 71, the
coherence bandwidth Bc can also be estimated. Considering that the coherence
bandwidth is the bandwidth over which the absolute FCF is above 0.5 [2], Bc is
approximately 0.9 MHZ, 0.7 MHz, and 0.4 MHz, when T  60 , T  70 , and
133

Chapter 5
T  80 , respectively. Figs. 72 and 73 illustrate the shape of the absolute SFCFs
Rpl ,qm T , R  0, t  0, f  and Rpl ,qm T  0, R , t  0, f  of a HAP‐MIMO

channel, while Fig. 74 depicts the shape of the absolute time‐frequency correlation
function (TFCF) Rpl ,qm T  0, R  0, t , f  of a HAP‐SISO channel. Both Rayleigh
and Ricean fading conditions are considered.
1
β =60ο
Τ
0.8 β =70ο
Τ
Frequency Correlation
β =80ο
Τ
0.6
0.4
0.2
0
0 1 2 3 4 5
Frequency Separation Δf (MHz)
Figure 71: The FCF of a HAP‐SISO Rayleigh channel for different elevation angles of the SBS.
Space‐Frequency Correlation
1 1
0.9
0.5 0.8
0.7
0
0 0
5 5
100 4 100 4
3 3
δ /λ 2 δ /λ 2
Τ 1 Δf (MHz) Τ 1 Δf (MHz)
200 0 200 0
(a) (b)
Figure 72: The transmit SFCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean channel.
134

Chapter 5
1
1
0.8
0.5 0.6
0.4
0 0.2
0 0
5 1 5
1 4 4
3 δ /λ 3
δ /λ 2 2 R
2 2
R 1 Δf (MHz) 1 Δf (MHz)
3 0 3 0
(a) (b)
Figure 73: The receive SFCF of a 2  2 HAP‐MIMO (a) Rayleigh and (b) Ricean channel.
Time‐Frequency Correlation
1
Time‐Frequency Correlation
0.8
0.5
0.6
0 0.4
0 0
5 5
4 0.05 4
0.05 3 3
Δt (sec) 2 Δt (sec) 2
1 Δf (MHz) 1 Δf (MHz)
0.1 0 0.1 0
(a) (b)
Figure 74: The TFCF of a HAP‐SISO (a) Rayleigh and (b) Ricean channel.
Fig. 75 shows the SDPSs of a HAP‐SISO Rayleigh channel and a 2  2 HAP‐MIMO
Rayleigh channel for a 3‐D non‐isotropic scattering environment. The values of the
model parameters used are K pl  Kqm  0, HT  20 km, T  60 ,  T  80 ,

 R   , T  R  45 ,  R  30 , a  0.01, RS ,min  15 m, RS ,max  150 m,
HS ,min  5 m, HS ,max  40 m, HS ,mean  17.6 m,  S ,min  arctan  HS ,min / RS ,max   2 ,

 S ,max  arctan  HS ,max / RS ,min   70 ,   0.31, k  3,   0 , fT ,max  0 Hz,

fR ,max  150 Hz, and  R  30. A normalized Doppler frequency vn  v / vmax is

used, such that 1  vn  1. The results show that the SDPSs are similar to the
135

Chapter 5
typical U‐shaped spectrum of fixed‐to‐mobile cellular channels. These results also
show that the SDPSs do not have any undesirable discontinuities8 and exhibit
realistic finite9 values for the maximum Doppler frequencies.
Finally, Fig. 76 shows the relative PSDSs of a HAP‐SISO Rayleigh channel and a
2  2 HAP‐MIMO Rayleigh channel for a 3‐D non‐isotropic scattering environment.
The values of the model parameters used are K pl  Kqm  0, HT  20 km, T  60 ,

T  80 ,  R   , T  R  45 ,  R  10 , a  0.01, RS ,min  12 m, RS ,max  140 m,
HS ,min  6 m, HS ,max  50 m, HS ,mean  17.6 m,   0.31, k  8, and   90. A

normalized relative propagation delay  r ,n   r /  r ,max is used, such that

0   r ,n  1. To the author’s best knowledge, there are no experimental data
available in the literature to verify the theoretical results. Thus, the need for
experimental verification data is pointed out to confirm the validity of this
reference wideband HAP‐MIMO channel model.
20
Space‐Doppler Power Spectrum (dB)
‐20
‐40
‐60
HAP‐MIMO
HAP‐SISO
‐80
‐1 ‐0.5 0 0.5 1
Normalized Doppler Frequency
Figure 75: The SDPS of a HAP‐SISO and a 2  2 HAP‐MIMO Rayleigh channel.

8
The Doppler spectrum obtained from the Aulin’s 3‐D model is constant for fmax cos  max  v  fmax , where fmax and  max
are the maximum Doppler frequency and the maximum elevation angle of the scattered waves, respectively [195].
9
The Doppler spectrum obtained from the Clarke’s 2‐D model becomes infinite at the maximum Doppler frequency [167].
136

Chapter 5
10
Relative Power Space‐Delay Spectrum (dB)
HAP‐MIMO
5 HAP‐SISO
‐5
‐10
‐15
‐20
‐25
0 0.2 0.4 0.6 0.8 1
Normalized Propagation Delay

Figure 76: The relative PSDS of a HAP‐SISO and a 2  2 HAP‐MIMO Rayleigh channel.
5.5 Simulation Models for Wideband HAP‐MIMO Channels
The reference wideband model described above assumes an infinite number of
scatterers. Although this model is practically non‐realizable, it can be used to derive
simulation models, which have similar statistical properties as the reference model
and utilize a finite (preferably small) number of scatterers. This section proposes
deterministic and statistical simulation models for wideband HAP‐MIMO channels.
Assuming a 3‐D non‐isotropic scattering environment and using the design

method of HAP‐MIMO simulation models described in Chapter 4, the following
time‐variant transfer function of the sub‐channel p‐l is proposed [199]
Tpl  t , f   Tpl ,LoS  t , f   Tpl ,NLoS  t , f  , (164)

where
 pl 1 N1 N2 N3  n1 ,n2 ,n3 
Tpl ,NLoS  t , f  
K pl  1 N1N2N3
   e j
n1 1 n2 1 n3 1
 j 2    n1 ,n2 ,n3     n1 ,n2 ,n3     n1 ,n2 ,n3 

d  p , S d  S  T ,NLoS FR ,NLoS  e  j 2 f NLoS
,l   j 2 t F 
e      
e (165)
137

Chapter 5
is the NLoS component of the time‐variant transfer function, Tpl ,LoS  t , f  is defined

in (109),  
n1 ,n2 ,n3 
   ,  is the random phase, d p, S   n1 ,n2 ,n3 
 , d  S n1 ,n2 ,n3 

,l ,
FT,NLoS , and FR,NLoS are defined in (66), 67), (68), and (69), respectively,
n1 1,..., N1  , n2 1,..., N2  , n3 1,..., N3  , the product N1N2N3  N is the total

 n1 ,n2 ,n3 
finite number of scatterers and  NLoS is given by
RS 2  cos aR 1  RS 2 

n n n
 n1 ,n2 ,n3 
 NLoS   LoS   . (166)
c0 cos T

n n

c0 cos arctan HS 3  / RS 2  
 
Note that the input delay‐spread function of the simulation models is obtained as
  t ,   f 1 Tpl  t , f .
the IFT of the time‐variant transfer function in (164), i.e., hpl

Providing that a sufficient number of scatterers is used, i.e., N1N2N3  20,
Tpl ,NLoS  t , f  and Tqm

 ,NLoS  t , f  are close to a low‐pass zero‐mean complex Gaussian
process and the STFCF between Tpl  t , f  , and Tqm

  t , f  is given by (see Chapter 3)
 ,qm T , R , t , f   RplLoS,qm T , R , t , f   RplNLoS

Rpl ,qm  T , R , t , f  , (167)

where
2  q  p T cos T
N1 N2 N3 j
1
NLoS
Rpl T , R , t , f     e  cos T
 K pl  1 Kqm  1 n1 1 n2 1 n3 1
,qm
N1N2N3
n  n   n  n  
j 2  m l  R sin R sinarctan HS 3 / RS 2  
j 2  q  p T sinT RS 2 sin aR 1  

 D cos T 
e e
n   n  n  
j 2  m l  R cos R cos R cos aR 1 cos arctan HS 3 / RS 2  
  
e 
n   n  n  
j 2  m l  R sinR cos R sin aR 1 cos arctan HS 3 / RS 2  
  
e 
 
 n  n 
RS 2 cos aR 1
n 
RS 2 
D
 j 2f    
 c0 cos T c0 cos T    n3   n2    
j 2t  FT ,NLoS FR ,NLoS   c0 cos arctan HS /RS   
    
e e . (168)
138

Chapter 5
is the STFCF of the NLoS component and RplLoS,qm  T , R , t , f  is defined in (113).

To successfully approximate the statistical properties of the reference model, aR 1  ,
n
RS 2  , and HS 3  must be properly determined.

n n
First, a deterministic model is proposed. This model computes constant values

for the model parameters, needs a single run to obtain the desired statistical
properties and hence, preserves short simulation times. Based on the deterministic
approach, aR 1  can be generated using (72) and (75). Moreover, RS 2  can be
n n
generated as
RS 2   FT,R1S    ,
n
(169)

where
FRS  RS 
FT ,RS  RS   (170)
FRS  RS ,max   FRS  RS ,min 
is a truncated hyperbolic cdf applicable for the range RS ,min  RS  RS ,max , FRS  RS  is
defined in (60) and   is defined in (76). Finally, HS 3  can be generated as

n
HS 3   FT,H1 S    ,
n
(171)

where
FHS  HS 
FT ,HS  HS   (172)
FHS  HS ,max   FHS  HS ,min 
is a truncated log‐normal cdf applicable for the range HS ,min  HS  HS ,max , FHS  H S 
is defined in (63) and   is defined in (77).
A statistical model is also proposed. The statistical properties of this model are
stochastic for each simulation trial and converge to the theoretical statistical
properties after averaging over a sufficient number of simulation trials for an
arbitrary finite number of scatterers, i.e., for any N1 , N2 , N3   1. Considering that

1 ,  2 , and  3 are independent random variables uniformly distributed in the
139

Chapter 5
interval 0,1 , aR 1  is generated using (88) and (91), while RS 2  and HS 3  are
n n n
generated as
RS 2   FT,R1S    , (173)

n
HS 3   FT,H1 S    , (174)

n
where   and   are defined in (92) and (93), respectively. The STFCF of the
 ,qm T , R , t , f  over Ntrials

statistical model is obtained after averaging Rpl
simulation trials

Ntrials
1
 ,qm T , R , t , f  
Rpl 
Ntrials ntrials 1
 ,qm T , R , t , f .
Rpl (175)

It is obvious that the statistical model is more complex than the deterministic
model due to the average procedure and the generation of three random variables.
Hence, the number of the scatterers and the number of simulation trials should be
carefully chosen to keep the complexity to a minimum.
5.5.1 Simulation Results
This subsection examines the performance of simulation models. The values of
the model parameters used are K pl  Kqm  0 dB, nT  nR  2, Δf = 200 Hz,

HT  20 km, T  60 ,  T  60 ,  R   / 2, T  60 , R  30 ,  R  20 , k  4,
  90 , a  0.01, RS ,max  140 m, HS ,max  60 m, HS ,mean  17.6 m,   0.31,

fT ,max  0 Hz, fR ,max  100 Hz, and  R  60. Moreover, a normalized sampling

period fR ,maxTS  0.02 is used, where TS is the sampling period.
Fig. 77 compares the absolute STFCFs of the reference and simulation models.
One observes that the STFCF of the deterministic model closely matches the
reference one for N1  30, N2  20, and N3  8, while the statistical model almost
perfectly reproduces the theoretical STFCF for N1  10, N2  8, N3  3, and
140

Chapter 5
Ntrials  40. From the results, the RMSE between the STFCFs of the deterministic
and the reference model is 0.031 for N1  30, N2  20, and N3  8, and 0.027 for
N1  40, N2  30, and N3  10. Hence, increasing the number of scatterers,
enhances the performance of the deterministic model. Finally, the RMSE between
the STFCFs of the statistical and the reference model is 0.019 for Ntrials  40 and
0.014 for Ntrials  60. Thus, increasing the number of trials, improves the
performance of the statistical model.
0.7
Space‐Time‐Frequency Correlation
0.6
0.5
0.4
Reference model
0.3 Deterministic model, N =30, N =20, N =8
1 2 3
Statistical model, N =10, N =8, N =3, N =40
1 2 3 trials
0.2
0 0.02 0.04 0.06 0.08 0.1
Δt (s)

Figure 77: Comparison between the STFCFs of the reference and the simulation models.
5.6 Summary
In this chapter, a theoretical parametric model for wideband Ricean HAP‐MIMO
channels has been proposed to completely characterize HAP‐MIMO multipath
fading channels. In particular, the narrowband 3‐D HAP‐MIMO channel model has
been extended to the wideband channel scenario with respect to frequency
selectivity. Several parameters related to the physical properties of wideband
stratospheric communications have been considered, in order to properly and
thoroughly characterize the wideband HAP‐MIMO channels. Based on the
141

Chapter 5
proposed model the channel statistics have been analytically studied. Specifically,
the STFCF, the SDPS, and the PSDS have been derived, under a 3‐D non‐isotropic
scattering environment. Useful numerical results have been provided and the need
for further experimental verification data has been pointed out. Future empirical
results could be easily be compared with theoretical results, since the proposed
model is flexible and applicable to a wide range of propagation environments, i.e.,
one may choose proper values for the model parameters to fit a particular
environment.
The reference model assumes an infinite number of scatterers and proves to be
too complex. Therefore, realizable simulation models are essential to faithfully
reproduce its statistical properties. Based on the SoS principle, deterministic and
statistical simulation models for wideband HAP‐MIMO channels have been
developed. The statistical properties of these simulation models have been verified
by comparison with the corresponding statistical properties of the reference
model. The results have shown that the proposed simulation model satisfactorily
approximate the STFCF of the reference model and have nearly the same statistics
as the reference model, while keeping the complexity to a minimum.
Overall, the proposed models are applicable to a variety of propagation
environments and scenarios, and provide helpful insights and a useful framework
for analyzing, designing, optimizing, and testing emerging wideband HAP‐MIMO
mobile communications systems, under LoS and NLoS conditions, in frequencies
well below 10 GHz.
Nowadays, as the telecommunication market is driven by the increasing demand
for broadcast services and high‐speed ubiquitous Internet access, BFWA is gaining
increased popularity. Nevertheless, frequency bandwidths wide enough to carry
such services are only available at mm‐wave frequency bands. MIMO technology
may enhance BFWA networks, although signal propagation is severely affected by
rainfall in mm‐wave frequencies. Hence, Chapter 6 investigates the application of
MIMO techniques to fixed HAP‐based systems at Ka‐ and V‐ bands.
142

Chapter 6
Chapter 6
Modeling of Line‐of‐Sight HAP‐MIMO Channels

6.1 Introduction
Apart from providing mobile communications services, HAPs can also serve as
base stations for the provision of BFWA services at the licensed Ka‐ and V‐
frequency bands. The growing exigencies for spectral efficiency and improved
channel capacity have prompted the development of advanced physical layer
theory and techniques. Hence, such theory and techniques should be applied,
evolved and/or originated in order to construct high‐performance HAP‐based fixed
communications systems. Previous studies investigated V‐MIMO techniques based
on constellations of multiple HAPs that intend to provide BFWA services at mm‐
wave frequency bands [200]‐[202]. As MIMO technology is nowadays deemed a
necessity due to its potentially high‐bandwidth efficiency, whether the size of a
single HAP can effectively exploit MIMO advantages at the licensed Ka‐ and V‐
frequency bands is an open research issue. Specifically, at these frequencies, the
propagation characteristics are drastically different compared to those observed in
frequencies well below 10 GHz, i.e., L‐ and S‐ bands. Therefore, new wireless
system design approaches at the physical layer and realistic HAP‐MIMO channel
models are required.
As shown in Chapter 3, the MIMO gain strongly depends on the channel
characteristics, which are mainly determined by the antenna configuration and the
richness of scattering at frequencies well below 10 GHz. Indeed, the MIMO
research has been initiated by the mathematical derivation of the MIMO channel
capacity assuming Rayleigh i.i.d. fading [68], [69]. However, propagation at mm‐
wave bands requires a strong, dominant LoS signal for sufficient coverage due to
the severe attenuation of the NLoS links [11], [37]. Conceptually, this corresponds
143

Chapter 6
to a rank‐deficient10 MIMO channel matrix and low spectral efficiency due to the
increased spatial correlation introduced by the linear relationship of the phases of
the received signals [203], [204]. Contrary to these observations, a number of
studies have suggested that the LoS response is not inherently correlated and have
shown the possibility of getting high MIMO gain for LoS channels by preserving the
orthogonality of the received signals [205]‐[208]. In detail, using specifically
designed antenna arrays, i.e., placing the antenna elements in positions, where the
LoS rays are orthogonal, a full‐rank MIMO channel may be achieved.
In this chapter, geometrical design recommendations are introduced and the
applicability of MIMO techniques to LoS‐HAP channels with propagation models for
the Ka‐ and V‐ frequency bands is investigated [88]. The parameters of interest are
the height and the elevation angle of the platform, the carrier frequency, and the
array configuration. Since rain has an important impact on the quality of the link at
mm‐wave frequencies [11], both clear sky and rain conditions should be
considered. The results show that the proposed design method overcomes the
problem of reduced capacity in LoS conditions and outperforms the conventional
rank‐one LoS architecture in terms of channel capacity. These results also show
that the optimized scheme actually performs better than MIMO schemes based on
Rayleigh i.i.d. fading. The sensitivity to possible deviation of the values of the
parameters chosen to satisfy the optimal design constraints is also investigated.
Thus, a deviation factor is introduced and an analysis of the sensitivity to non‐
optimal design with regard to channel capacity is performed.
The remainder of this chapter is organized as follows. The LoS‐HAP‐MIMO system
model and the corresponding geometrical model are described in Section 6.2, while
the maximum capacity criteria are derived in Section 6.3. Section 6.4 demonstrates
the theoretical derivations by comparing the performance of different architectures,
under clear sky and rain conditions, and investigating the sensitivity of the capacity
performance. Finally, Section 6.5 summarizes this chapter.

10
Rank‐deficient channel matrix brings correlations between MIMO subchannels and, hence, is incapable of supporting many
multiple parallel data streams.
144

Chapter 6
6.2 The LoS‐HAP‐MIMO System Model
The system model employed throughout this chapter considers a SBS and a
terrestrial fixed station (TFS) utilizing ULAs with nT transmit and nR receive antenna
elements, respectively. All antennas are fixed, highly directional11 and are numbered
as 1  p  q  nT and 1  l  m  nR , respectively. Considering slowly varying and
frequency‐flat fading channels over the band of interest, the link between the SBS
and TFS antenna arrays is represented using the complex baseband vector notation
r  GHs  n, (176)
where s  nT 1 is the transmitted signal vector, r   nR 1 is the received signal
vector, H   hij   nR nT is the channel matrix of complex channel gains, and

nR nT
n   nR 1 is the noise vector, which denotes the AWGN at the receiver branches.
Moreover, G is the common power gain over the channel and is given by
G  GT  dBi  GR  dBi  LF  dB   Lr  dB  , (177)
where GT and GR are the transmit and receive antenna gains, respectively, LF is
the FSL, and Lr is the rain attenuation. Finally, the entries of the noise vector are
i.i.d. complex Gaussian random variables with zero mean and variance N0 , where
N0 is the noise PSD. It is assumed that all the subchannels experience the same path
loss due to the significant distance between SBS and TFS. This common path loss is
incorporated in the G factor.
As the propagation environment is related to BFWA in mm‐wave frequencies, the
multipath propagation due to local scatterers in the vicivity of TFS in neglected and
H corresponds to a pure LoS channel, under clear sky conditions. However, rain
introduces not only severe attenuation but also a short‐term variation to the

11
Since it is more difficult to generate power at the susceptible to propagation impairments Ka/V bands than at the lower L/S
bands, omni‐directional antennas are essentially not applicable. Thus, it is easier to realize highly directive antennas, which
can increase the fade margin by adding more gain, and improve the link availability.
145

Chapter 6
received signal. In [209], a wideband 200 MHz measurement campaign was
conducted in a campus environment for short terrestrial 38‐GHz BFWA links, where
LoS conditions, fixed reception, high antenna directivity, and vertical polarization
were maintained. An important conclusion that is drawn from this experiment was
that multipath was observed even in unobstructed LoS links during rain but not
observed during clear sky conditions. Hence, the MIMO channel matrix H is given by
Kr 1
H HLoS  r
HNLoS , (178)
Kr  1 Kr  1
where HLoS is a deterministic nR  nT matrix containing the non‐fading LoS

r
responses between all array elements, HNLoS is a stochastic nR  nT matrix

containing the responses of the scattered waves due to rain, and Kr is the Ricean
factor, which directly refers to the rainy conditions, i.e., it incorporates the channel
dynamics as a result of radiowave propagation through the rainfall medium. Note
that the channel matrix H is normalized, which means that its elements have an
average power of one. This normalization corresponds to a system with perfect
power control [210] and makes the SNR independent of H. Considering free‐space
propagation, the normalized matrix HLoS is given by
 h11,LoS h12,LoS  h1nT ,LoS 

 
h   
HLoS   21,LoS
    
 
 hnR 1,LoS  hnR nT ,LoS 
 e  j 2 d 1,1 /  e
 j 2 d 1,2  / 
 e
 j 2 d 1,nT  / 

 
 e  j 2 d 2,1 /   
  , (179)
    
  j 2 d  nR ,1/   j 2 d  nR ,nT  /  
e  e 
 j 2 d  i , j  / 
where hij ,LoS  e is the complex response between the i th receive and the
j th transmit antenna element, d  i , j  is the distance between these two elements,
146

Chapter 6
and  is the carrier wavelength. Hence, the elements of HLoS depend only on the
distances between the SBS and TFS antenna elements. These distances can be
estimated by applying ray‐tracing12 techniques to the geometrical model in Fig. 75.
The scattering introduced by rain can be modeled as a stochastic process [88], [208].
r
Therefore, the entries of HNLoS are modeled as i.i.d. random variables (Rayleigh
fading). Finally, the Ricean factor K r is given by [209]
16.88  0.04Rr dB, Rr  0

K r  dB    . (180)
  dB, Rr  0
This empirical relationship between the Ricean factor K r (in dB) and the rainfall
rate Rr (in mm/h) is elevation‐, frequency‐, polarization‐, climate‐, and site‐specific.
Although an analytical prediction model for the CDF of the Rician factor can
incorporate the mathematically intractable physical input of rain fading into the
Rician channel model (see [211]) on a global scale, this chapter adopts the simplified
expression in (180) to make the analysis traceable and the design less complex.
6.2.1 A Geometrical Model for LoS‐HAP‐MIMO Channels
The geometrical characteristics of the proposed model for a 2  2 LoS‐HAP‐MIMO
channel with nT  nR  2 antenna elements and the definition of the Cartesian
coordinate system are discussed in Fig. 78. The x ‐axis is the line that connects the

sub‐platform point p and the TFS antenna element l , while the z ‐axis is the line
that connects the sub‐platform point p and the SBS antenna element p. The
spacing between SBS and TFS antenna elements is denoted T and  R , respectively.
In addition, angles T and R represent the orientation of the SBS and TFS antenna
arrays, respectively, while T , HT , HR , and Hr denote the elevation angle of SBS,
the height of SBS, the height of TFS, and the effective rain height, respectively.

12
Ray‐tracing is based on finding the path length from each of the SBS antennas to each of the TFS antennas, and employing
these path lengths to find the corresponding received phases.
147

Chapter 6
z q
p
HT Hr m
T l
q HR
y 
m
p T l R x

Figure 78: A geometrical model for a 2  2 LoS‐HAP‐MIMO channel.
6.3 Capacity Optimization Techniques
Using (179), the correlation matrix of the LoS component is equal to
 nt
 j 2 d 1, j d  nR , j   /  
 nT  e 
 j 1 
H
HLoSHLoS     . (181)
 
 nt  j2 d  nR , j d 1, j  / 
 e    nT 
 j 1 
Moreover, using the definition of HAP‐MIMO channel capacity in (47) and
considering clear sky conditions, the minimum capacity is obtained for
H
HLoS HLoS  nT 1nR , (182)

where 1nR is an nR  nR all‐ones matrix. Then, the channel matrix HLoS is rank‐one
(entirely correlated) and the capacity is equal to that of a SIMO channel [212], i.e.,
no multiplexing gain is obtained
148

Chapter 6
Cmin  log2 1  nR SNR  . (183)

On the contrary, the maximum capacity is obtained for
H
HLoS HLoS  nT InR , (184)

where InR is an nR  nR identity matrix. Then, HLoS is full‐rank and the MIMO
subchannels are perfectly orthogonal. In addition, the capacity is equivalent to that
of nR independent SISO channels as follows
Cmax  nR log2 1  SNR  . (185)

Considering a 2  2 LoS‐HAP‐MIMO channel, the normalized channel matrix HLoS

can be written as
hpl ,LoS hpm ,LoS 
HLoS  
hqm ,LoS 
. (186)
 hql ,LoS
Then, the maximum capacity criterion is simplified as follows
H
HLoS HLoS  2I2

 2 ALoS 
 *   2I2 , (187)
 ALoS 2 
where
 j 2 d  l ,p  d  m ,p   /   j 2 d  l ,q d  m ,q   / 
ALoS  e e . (188)
The solution to (188) can be written as [182]
2 d  l , p   d  m, p   d  l , q   d  m , q   / =  2v  1  

 d  l , p   d  m, p   d  l , q   d  m, q  =  2v  1  / 2, (189)

where v  and  represents a set of integers. As shown in Appendix B, from
(189) one can easily derive the simplified 2  2 LoS‐HAP‐MIMO maximum capacity
criterion in pure LoS as a function of T ,  R , T , R , T , HT , and  as follows
149

Chapter 6
HT
T  R   2v  1 , T ,T ,R   0. (190)
2sin T sinT sinR
Physically, the approximate maximum capacity criterion corresponds to systems

where the sum of the path differences is an odd integer multiple of a half
wavelength. The product between T and  R is the key design quantity on the part
of maximizing the channel capacity and leads to a versatile system design, i.e.,
maximum capacity can be achieved using small array sizes at one end of the link and
large array sizes at the other end. This product is referred to as the antenna
elements spacing product (AESP). The criterion in (190) for a 2  2 LoS‐HAP‐MIMO
channel can be extended to the case of a nR  nT LoS‐HAP‐MIMO channel with an
arbitrary number of antenna elements. As shown in Appendix B, the nR  nT LoS‐
HAP‐MIMO design criterion to achieve maximum channel capacity can be expressed
as follows
 1  HT
T  R   v , T ,T ,R   0. (191)
n
 T  sin  T sinT sin R
From (191), one observes that the required AESP reduces with increasing nT .
Hence, high space efficiency at the SBS is obtained, i.e., smaller spacing per element
at the SBS antenna array can be adjusted and a large number of elements can be
accommodated. This enables the utilization of multiple antenna elements at the SBS
without the problems of mutual coupling [213] and correlation [97], [98], while the
performance does not depend on the existence of a rich scattering environment as
in the case of the traditional MIMO. In addition, although a strong LoS signal is
considered, the capacity linearly increases with increasing nT .
6.3.1 Practical Considerations
From (190) and (191), one observes that considerable antenna spacing is needed
to introduce phase differences among antennas when operating MIMO system in
LoS environment. Hence, the solution corresponding to the smallest AESP is
150

Chapter 6
probably the most interesting from a practical and realistic design point of view,
because it requires the shortest antenna arrays. Equations (190) and (191) imply
that the minimum AESP is obtained for broadside antenna array orientation, i.e.,
T  R  90 , and for location of the SBS just above TFS, i.e., T  90. Note that

HAPs are either airships (of about 150‐200 m in length) or circling aircrafts
(approximately 30 m in length) [11], [12]. Hence, based on the HAP structure used
and the available HAP length, the spacing between TFS antenna elements can be
ideally adjusted to satisfy (190). For instance, assuming that v  0, HT  20 km
(typical HAP height), T  60 (typical value in urban environments), T  R  90 ,
and considering signal transmission at 28 GHz and 48 GHz, the required AESP is
equal to approximately 124 and 72, respectively. This suggests that applying LoS‐
HAP‐MIMO techniques to a single HAP (aircraft or airship) and a typical TFS, i.e.,
installing multiple antenna elements at the roof of a building, is viable. In addition,
the significantly small distance between the antenna elements ensures identical
atmospheric impairments for all the signals propagating to the same TFS array, i.e.,
the propagation is realized through the same rainfall medium for all the antenna
elements at the TFS. However, as T and R decrease, the required AESP increases,
e.g., the AESP redoubles, when T  R  45. In the worst case, in which
T ,R   0 , very large non‐practical antenna spacing at the TFS and/or
utilization of multiple HAPs may be required. Note that using very large antenna
spacing at the SBS/TFS introduces propagation delay offsets, which may lead to
synchronization problems. These problems do not occur in conventional terrestrial
MIMO systems, where the multiple transmitted signals are simultaneously received
[214] and the inter‐element distances are usually equal to a half wavelength.
6.4 Numerical Results
This section investigates the performance of the proposed design

recommendations for a 2  2 LoS‐HAP‐MIMO system. As mentioned in the previous
151

Chapter 6
section, capacity optimized LoS‐HAP‐MIMO techniques are feasible in a single HAP,
if the TFS antenna inter‐element spacing is carefully arranged. Using (190) and
considering that v  0, HT  20 km, T  80 , and T  R  90 , the required AESP
is approximately 63.4 and 108.8 for the 48 and the 28 GHz frequency band,
respectively. Thus, assuming that the TFS could be a subscriber unit (SU) in an
outdoor position and considering that SBS antenna inter‐element spacing is fixed
and equal to T  30 m, the TFS antenna inter‐element spacing is set  R  2.11 m,
and  R  3.63 m, respectively.
Fig. 79 examines the effect of the array design on the ergodic capacity, under
clear sky conditions, i.e., Rr  0. In order to remove the effects of the average SNR
on the channel capacity, the channel response is normalized so that the constraint
in (55) is fulfilled. Specifically, the full‐rank and rank‐one 2  2 LoS‐HAP‐MIMO
capacity, as well as the LoS‐HAP‐SISO capacity are plotted against SNR, in order to
compare the performance of different architectures. Moreover, the 2  2 MIMO
Rayleigh i.i.d. capacity (evaluated after a large number of channel realizations)
against SNR is plotted, for reference. Note that there is a substantial gain of the full‐
rank 2  2 LoS‐MIMO architecture over the rank‐one 2  2 LoS‐HAP‐MIMO and LoS‐
HAP‐SISO architectures and the capacity dramatically increases up to 8.5 and 9.5
bps/Hz, respectively. Hence, the rank‐one scenario validates the common belief that
a strong LoS signal results in poor capacity gain. One observes that the optimized
2  2 LoS‐HAP‐MIMO design can offer better performance than the Rayleigh i.i.d.
2  2 MIMO system and the capacity increases up to 1.4 bps/Hz. Hence, using
multiple elements at both SBS and TFS, under LoS conditions and design constraints,
can effectively boost channel capacity.
Fig. 80 shows the CCDF of the channel capacity for full‐rank and rank‐one LoS‐
HAP‐MIMO architectures under the effect of rain, i.e., Rr  0, using the
propagation model described in Chapter 2 and the 28 GHz frequency band.
According to (178) and (180) the channel becomes Ricean for rain rates greater
than zero. In addition, the SNR depend on the rain rate. A set of realistic parameter
152

Chapter 6
values are chosen to produce the results. Most of these values are found in [215]
and are reproduced for convenience in Table 12 together with all the parameters
employed. Considering horizontally polarized waves, the 30 GHz frequency band
corresponds to ar  0.187 and br  1.021 [82] (see Chapter 2). Then, using (7) and
(180), and assuming that Rr takes values 12, 28 and 42 mm/h and the rain height is
3.5 km, the corresponding Lr and K r approximated values are 8.3, 19.6, and 29.7
dB, and 16.4, 15.8, and 15.2 dB, respectively. Using (1)‐(3), (177) the corresponding
SNR values are 25.9, 14.6, and 4.5 dB, respectively. Note that increasing rain rate
r r
decreases the SNR and increases the influence of HNLoS . Since HNLoS is full‐rank,
increasing rain rate, degrades the capacity gain of full‐rank 2  2 LoS‐HAP‐MIMO
architecture over rank‐one 2  2 LoS‐HAP‐MIMO architecture. Specifically, one
observes that for over 50% of the channels, the capacity is greater than 16.7 and
12.7 bits/Hz for the full‐rank and rank‐one LoS‐HAP‐MIMO systems, respectively,
for 12 mm/h rain rate, greater than 9.3 and 6.5 bits/Hz, respectively, for 28 mm/h
rain rate, and greater than 3.6 and 2.8 bits/Hz, respectively, for 42 mm/h rain rate.
20
Full‐rank 2x2 LoS‐HAP‐MIMO
Rank‐one 2x2 LoS‐HAP‐MIMO
Ergodic Capacity (bps/Hz)
15 LoS‐HAP‐SISO
Rayleigh i.i.d. 2x2 MIMO
10
0
0 5 10 15 20 25 30
SNR (dB)
Figure 79: Comparison of the channel ergodic capacities obtained using the full‐rank and
rank‐one 2  2 LoS‐HAP‐MIMO, the LoS‐HAP‐SISO, and the Rayleigh i.i.d. 2  2 MIMO
architectures, under clear sky conditions.
153

Chapter 6
0.8
R =42 mm/h R =12 mm/h
r r
0.6
0.4
R =28 mm/h
r
0.2
Full‐rank
Rank‐one
0
2 4 6 8 10 12 14 16 18
Capacity (bits/Hz)
Figure 80: The complementary cumulative distribution function of the channel capacity
achieved with the full‐rank and rank‐one 2  2 LoS‐HAP‐MIMO architectures for different
rain rates.
TABLE 12
LINK BUDGET PARAMETERS
Frequency  f  28 GHz 48 GHz
Free space loss  LF  147.4 dB 152 dB
Bandwidth  B  20 MHz
HAP transmit power  PT  ‐15 dBW
HAP antenna gain  GT  34 dBi
SU antenna gain  GR  34 dBi
Noise Temperature  Te  500 K
Noise PSD  N0  ‐201.6 dBW/Hz
Noise power  PN  ‐128.6 dBW

Fig. 81 shows the ergodic capacity against rain rate obtained using the optimized
full‐rank 2  2 LoS‐HAP‐MIMO architecture and compares the 28 GHz and the 48
154

Chapter 6
GHz frequency bands. According to [82] and considering horizontally polarized
waves, the 50 GHz frequency band corresponds to ar  0.537 and br  0.873. One
observes that the 48 GHz frequency band is more susceptible to rain attenuation,
and as the rain rate increases from 0 mm/h (clear sky) to 80 mm/h, the ergodic
capacity sharply degrades compared to the ergodic capacity achieved at the 28 GHz
frequency band.
25
28 GHz
48 GHz
20
Ergodic capacity (bps/Hz)
15
10
0
0 10 20 30 40 50 60 70 80
R (mm/h)
r
Figure 81: The ergodic capacity obtained using the full‐rank 2  2 LoS‐HAP‐MIMO
architecture as a function of the rain rate for different carrier frequency.
6.4.1 Sensitivity Study
The maximum capacity criterion corresponds to very specific array geometries
fixed at optimal locations. In this section, the sensitivity of the optimized LoS‐HAP‐
MIMO systems to deviations from the optimal AESP value in (191) is studied. For
this purpose, the deviation factor  is introduced, which is defined as the ratio
between the actual AESP and the optimal AESP
AESPactual Tactual Ractual sin T sinT sinR

  . (192)
AESPoptimal 1 / nT  v  HT
155

Chapter 6
From (192), one observes that if the deviation factor is smaller than one, the actual
AESP is too small compared to the optimal value, while a deviation factor larger
than unity indicates that the actual AESP is too large. Indeed, the optimal AESP may
be too large for practical systems. To examine this sensitivity to non‐optimal
design, the capacity is now evaluated in terms of the variation of the deviation
factor defined in (192) for SNR  20 dB. From Fig. 82, a large sensitivity (in terms of
capacity) to the deviation factor is observed. In detail, the channel capacity is seen
to vary between the minimum (7.65 bps/Hz) and the maximum (13.32 bps/Hz)
values. This sensitivity needs to be taken into account in the design and fixing of
LoS‐HAP‐MIMO communications systems.
14
13
Normalized Capacity (bps/Hz)
12
11
10
7
0 0.2 0.4 0.6 0.8 1
Deviation factor
Figure 82: The normalized capacity obtained using the full‐rank 2  2 LoS‐HAP‐MIMO
architecture as a function of the deviation factor.
In practical situations, the parameters T , T , and R in (190) and (191) may be
difficult to be determined with sufficient accuracy. Moreover, there is usually a
need for high capacity over an area, rather than to a fixed point, while the
stratospheric winds may vary the position of the platform. Hence, a compromise
may be required, under more realistic deployment and propagation conditions or
156

Chapter 6
imperfectly positioned arrays. The sensitivity of the performance to the orientation
and positioning of the arrays is discussed and evaluated in the following figures.
Fig. 83 demonstrates the channel capacity for different orientation of the SBS
antenna array. The capacity is evaluated for SNR  20 dB, T  R  90 and
0  T  90. According to Fig. 83, decreasing T from 90 to 45 has an
insignificant effect on the capacity, while a further decrease in T drastically
decreases the capacity. Hence, the results show a large sensitivity (in terms of
capacity) to the orientation of the arrays. From (190) and (191), it is clear that the
same results apply for 0  R  90 and 0  T  90.
Finally, Fig. 84 examines the sensitivity of the performance to SBS (or TFS) array
shifting from the optimal point and demonstrates the influence of up to 20 km
shifting along either x‐ or y‐axis on the channel capacity for SNR  20 dB. One
observes that the capacity is relatively insensitive to shifting along x‐ or y ‐axis. In
particular, for a maximum shifting of 20 km, the capacity is only 3.3% and 2.2%
lower than the maximum capacity, respectively.
14
13
12
11
10
7
0 10 20 30 40 50 60 70 80 90
θ (degrees)
T
architecture for different orientation of the SBS antenna array.
157

Chapter 6
13.4
x‐direction
y‐direction
13.3
13.2
13.1
13
12.9
12.8
0 0.5 1 1.5 2
Antenna array shifting (m) 4
x 10
architecture as a function of the antenna array shifting from the optimal point.
6.5 Summary
In this chapter, a method to achieve orthogonality between spatially

multiplexed signals in LoS channels of HAP‐MIMO systems employed with ULAs has
been theoretically derived. In particular, antenna array design recommendations
have been introduced, in order to construct a full‐rank HAP‐MIMO channel matrix
at Ka and V frequency bands. Providing that the system parameters are carefully
selected, the results have demonstrated the substantial benefits of the proposed
architecture and have also illustrated the impact of real world effects, such as rain
attenuation and scattering and array length constraints, on the performance. The
optimized LoS‐HAP‐MIMO architecture outperforms the conventional rank‐one
LoS‐HAP‐MIMO and the Rayleigh i.i.d. MIMO architectures, in terms of channel
capacity, under clear sky and rain conditions. However, the capacity gain is
confined, as rain rate increases. The sensitivity of the performance to the
positioning and orientation of the arrays has been also discussed and evaluated.
Nevertheless, future experimental data are required to verify these results.
158

Chapter 7
Chapter 7
Conclusions and Future Research Directions

7.1 Conclusions
The aim of this research was to quantify the benefit (if any) of applying MIMO
technology to HAP‐based communications systems. Since it is the channel which
principally dictates any benefit, the primary focus of this thesis was to discover the
nature of the HAP‐MIMO radio propagation channel and develop accurate and
reliable channel models that can be widely used by the research community. At the
beginning of this research, a literature survey suggested that no channel model or
measurement data were in existence for the HAP‐MIMO channel. Furthermore,
whether the size of a single HAP can effectively support multiple antennas was
uncertain. Motivated by these facts, this section concludes this thesis.
In Chapter 3, the problem of modeling the narrowband HAP‐MIMO channels has
been addressed. A 3‐D cylinder geometrical propagation model for narrowband
HAP‐MIMO communications has been proposed. Based on this geometrical model,
a 3‐D parametric reference model for narrowband HAP‐MIMO Ricean fading
channels has been developed. From this reference model, the STCF for a 3‐D non‐
isotropic scattering environment has been mathematically derived. Then, the effect
of the system parameters on the spatial and temporal correlation and the capacity
of ULAs has been thoroughly studied. The numerical results have suggested that
MIMO techniques are applicable on a single HAP, as long as the Ricean factor is
sufficiently small.
To validate the accuracy of the aforementioned theoretical model, 3‐D SoS‐
based deterministic and statistical simulation models have been developed in
Chapter 4. The statistical properties of the simulation models have been derived
159

Chapter 7
and verified by simulation. The simulation results have shown that the proposed
simulation models are a good approximation of the reference model. Specifically,
the close agreement between the theoretical‐based and simulation‐based channel
statistics have confirmed the utility of the proposed simulation models and the
method used to estimate the model parameters. These results have also evaluated
the performance vs. complexity trade‐offs between the statistical and deterministic
simulation models. In most applications, software‐based simulation is performed
on a workstation or a personal computer (PC). Nevertheless, the feasibility of
porting these simulation models into hardware by using digital signal processors
(DSPs) is also of interest. In particular, hardware channel simulations can
significantly increase the usefulness of the simulation models by enabling
simulations in real‐time.
Currently, a number of standardization bodies supported by industries and

research institutes are trying to establish new system standards for future high‐
speed 4G wireless systems employing MIMO techniques. One important feature of
future communications systems is that they demand considerably larger
bandwidths than today’s systems. In Chapter 5, the reference and simulation
narrowband HAP‐MIMO channel models have been extended to the corresponding
frequency‐selective models for high‐bandwidth transmissions. In particular, a 3‐D
two concentric‐cylinder model for wideband HAP‐MIMO channels has been
proposed. Based on this geometrical model, a 3‐D parametric reference model for
wideband HAP‐MIMO Ricean fading channels has been developed. From the
reference model, the STFCF has been mathematically derived for a 3‐D non‐
isotropic scattering environment. From the STFCF, the SDPS and the PSDS have
been also derived. The numerical results have demonstrated the theoretical
derivations. To validate the reference wideband HAP‐MIMO channel model, SoS‐
based deterministic and statistical simulation models for wideband HAP‐MIMO
Ricean fading channels have been developed and the statistics of these models
have been verified by simulation. The simulation results have underlined the
outstanding performance of the proposed simulation models.
160

Chapter 7
Finally, the possibility of applying the MIMO technology to fixed broadband
HAP‐based communications systems operating at mm‐wave frequencies has been
explored. Although most research on MIMO technology has focused on the
potential gain arising due to rich scattering environments, Chapter 6 has suggested
that specifically designed antenna arrays provide a superior solution to the
problem of reduced MIMO performance in LoS. In particular, maximum capacity
can be reached when the relation between the antenna elements separation at
both ends, the transmission distance, the wavelength, the array orientation, and
the elevation angle of the platform fulfill a certain criterion. The results have
demonstrated the potential channel capacity gain of the proposed optimized
architecture. These results have also studied the influence of the rainfall and the
deviation from the optimal relation (e.g. due to array length constraints) on the
channel capacity performance.
7.2 Future Research Directions
With their unique characteristics, HAPs seem to represent an efficacious
alternative infrastructure, which can revolutionize the telecommunication industry.
Since MIMO architecture is the key technology, which will make 4G visions a
reality, it is envisaged that HAP‐MIMO systems will be potentially capable of
providing and delivering a compelling range of current and next generation
mobile/fixed services. Hence, HAP‐MIMO research will gain much interest from
academia, research centers, and industry worldwide. Although this work
constitutes a robust basis for HAP‐MIMO channel modeling, it could be further
improved or extended into different areas. In this section, some topics that could
be interesting to investigate in the future are briefly reviewed.
Measurement campaigns: Due to the lack of channel‐sounding measurement

campaigns, the contribution of this thesis has been limited to developing reference
and simulation models for HAP‐MIMO channels. However, it is important to verify
these models in real‐world propagation conditions. Specifically, HAPs could be
161

Chapter 7
emulated by using a helicopter, a small plane or a balloon containing two or more
antennas sufficiently separated. However, access to a real HAP would be even
more ideal. Therefore, future research efforts may be devoted to collecting
measured channel data and developing empirical models.
2nd order statistics: The first‐order statistics of the channel impulse response are
not sufficient to assess system characteristics such as handover, velocities of the
transmitter and receiver, and fading rate. To assess the aforementioned system
characteristics, accurate characterization of the second‐order statistics, such as the
envelope level crossing rate (LCR) and the average fade duration (AFD), is
necessary. Specifically, the envelope LCR is an important statistical quantity
revealing information about how fast the received signal changes with time. It is
basically a measure to describe the average number of times the signal envelope
crosses a certain threshold level from up to down (or from down to up) per second.
Moreover, the envelope AFD is defined as the expected value of the time intervals
over which the fading signal envelope remains below a certain threshold level.
Based on [188], [216] and using the proposed 3‐D reference model for narrowband
HAP‐MIMO channels, the corresponding analytical expressions for the envelope
LCR and AFD can be derived for a 3‐D non‐isotropic scattering environment.
Other types of antenna arrays: The analysis in this thesis is restricted to HAP‐
MIMO systems employing ULAs at both sides of the communication link. However,
the proposed channel models can be modified to employ other antenna array
geometries, such as uniform planar arrays (UPAs), uniform circular arrays (UCAs),
and spherical antenna arrays (SAAs), or a combination of them.
Multiple‐bounce scattering: Apart from considering only single‐bounce rays, this

work can be extended to additionally support double‐bounce or multiple‐bounce
rays (due to multiple scattering, reflection or diffraction of the radiated energy),
that individually contribute in the total transmitted power of the link. Then, more
sophisticated and realistic channel models can be obtained.
162

Chapter 7
Mobile scatterers: The proposed HAP‐MIMO channel models assume that all
scattering objects are stationary. However, this assumption does not hold in some
cases. Therefore, these models can be extended to include non‐stationary
scattering objects, such as vehicles and people.
Mobile applications at mm‐wave bands: Although signal transmissions at mm‐
wave frequencies are usually associated with fixed broadband services, currently
there is a clear trend to extend the applicability of these services to mobile
scenarios in order to benefit from existing air interfaces and accelerate the
development of new applications, such as the provision of high‐speed Internet
access, audio and video on demand and file transfer to vehicles, airplanes, trains
and ships [217]. However, modeling the entire possible propagation effects
associated with the licensed Ka and V frequency bands is an essential precondition.
The propagation effects are usually classified into two main categories, the local
environment propagation effects, e.g., multipath, shadowing, and blockage, and
the tropospheric effects, e.g., rainfall, oxygen absorption, water vapor, clouds, and
precipitation. The local environment propagation effects, which are also
encountered at lower frequency bands, i.e., L and S, exist due to the local
environment in the vicinity of the TMS. The atmospheric effects introduce severe
attenuation to the transmitted signal in frequencies well above 10 GHz [11].
This thesis has independently treated the local environment propagation effects
and the tropospheric effects depending on the operation frequency. Hence, their
possible correlation has not been yet investigated. However, an accurate and
realistic HAP‐MIMO channel modeling of broadband mobile communications
systems operating at mm‐wave frequencies should treat these effects together.
Under such conditions, it could be interesting to theoretically and experimentally
investigate the relation between the multipath introduced by the local scatterers
and the rainfall effects. The interested reader is referred to [218] for a first
theoretical approach to this open research problem, where a novel statistical
analysis for SISO mobile satellite systems is presented.
163

164

Appendix A
Appendix A
Derivations of Equations (17)‐(19)
1. Derivation of Equation (17)
   and pOO
From Fig. 18, applying the cosine law to the triangles plO  , we obtain
 
d 2 p , l
 d 2  p , O    0.5  nR  1  2l   R   d  p ,O  nR  1  2l   R cos aLoS

2 Rl

 R ,  (193)
d 2  p , O 

 D2   0.5  nT  1  2p   T   D  nT  1  2p   T cosT .
2
(194)

Rl
From (193) and (194), considering that aLoS   and using the approximate relation

1  x  1  x / 2 for small x , we obtain

 
d p , l  d  p ,O    nR  1  2l   R cosR ,

(195)

d  p , O   D   nT  1  2p  T cosT . (196)

Using (196), (195) becomes

 
d p , l  D   nT  1  2p  T cosT   nR  1  2l   R cosR .

(197)

Finally, from Fig. 17, since HT  HR , we finally obtain that

d  p, l   D   nT  1  2p  T cosT   nR  1  2l   R cosR cos R  / cos T . (198)

From Fig. 20, applying the sine law to the triangle OS  n O, we obtain
165

Appendix A
RS 
n

 
d O , S  
n
D
. (199)
sin aT n  sin   a   sin   a     a   
n n n
R
 T R
Since RS ,max  D , aT  is a small angle (see Chapter 3). Then, the approximate
n
relations sin x  x and cos x  1 for small x can be used and (199) becomes

RS 
n


d O , S  
n
 D
. (200)
aT n  sin aR 
n
sin aR 
n
From (200), we obtain
 
d O , S    D , (201)
n
aT   sin aR RS  / D. (202)

n n n
   n  , we also obtain
From Fig. 20, applying the cosine law to the triangle OpS

d 2 p , S  
n


 
 d 2 O , S    0.5  nT  1  2p  T 
n 2
 d  O, S     n  1  2p   cos   (203)

 aT  .
n n
T T T

Using (201) and the approximate relation 1  x  1  x / 2 for small x , (203)

becomes

 
d p , S    D  0.5  nT  1  2p  T  cos T  aT  .
n n
  (204)
Using (202), (204) can be written as follows

d p , S  
n

 D  0.5  nT  1  2p  T  cosT

 0.5  nT  1  2p  T  sinT sin aR RS  / D.
n n
(205)

166

Appendix A
Finally, from Fig. 19, since D  RS ,max , i.e., D  RS ,max  D  RS ,max  D , and

HT  HS ,max , we finally obtain that

   
d p, S    d p , S   / cos T . (206)
n n
From Fig. 19, the distance d S   , l can be written as
n
 

d2 S  , l
n

   
2
 d 2 S   , l  d OR , S   sin  S   d  OR , l  sin R 
n n n
 
 d  S  , l 
2 n

 d  S   , l   d  O , S    sin     d  O , l  sin 

2 n 2 n 2 n 2 2
R S R R

 2d  O , S    sin   d  O , l  sin .

n n
R S R R (207)

By applying the cosine law to the triangle OS  n l , we obtain

d 2 S   , l
n

 
 d 2 OR , S   cos2  S   d 2  OR , l  cos2  R
n n

 
2d OR , S   cos  S d  OR , l  cos R cos aR   R .
n n n
  (208)
Using (208), (207) becomes


d2 S  ,l
n

  
 d 2 OR , S    d 2  OR , l   2d OR , S   sin  S d  OR , l  sin R
n n n

 
 2d OR , S   cos  S d  OR , l  cos R cos aR   R .
n n n
  (209)
167

Appendix A
Using the approximate relation 1  x  1  x / 2 for small x , (209) finally becomes

d S  , l
n

 n
 
n n n

 d OR , S    d  OR , l  sin R sin  S   cos R cos  S  cos aR   R  .
 

 d S  ,l
n

  
 d OR , S    d  OR , l  sin R sin  S   d  OR , l  cos  S  cos R cos aR   R
n n n n


 d S  , l
n

 RS  / cos  S   0.5  nR  1  2l   R sin R sin  S 
n n n

 n  n
 0.5  nR  1  2l   R cosR cos R cos aR cos  S
 0.5  nR  1  2l   R sinR cos R sin aR  cos  S  . (210)

n n
168

Appendix B
Appendix B
Derivation of Equations (190) and (191)
From Fig. 75, considering that HT  HR , the Euclidean distances (path lengths)
between each pair of elements are given as follows
d  l , p   r  HT / sin T , (211)

1/2
d  m , p    r   R cosR    R sinR  
2 2
, (212)
 
1/2
d  l , q    r   T cosT    T sinT  
2 2
, (213)
 
1/2
d  m , q    r   T cosT   R cosR    T sinT   R sinR  
2 2
. (214)
 
Equations (212)‐(214) can be simplified using the following first‐order Taylor
series approximation
1/2
1/2  C2  C2
 A  B   C
2 2
  A  B  1     A  B  . (215)
    A  B  
2
2 A  B

This approximation is valid for  A  B   C 2 , which is a reasonable assumption for
2
the targeted application, providing that r  T , R . Using (215), (212)‐(214) are
approximated as follows
d  m, p   r   R cosR 
  R sinR 
2
, (216)
2  r   R cosR 

d  l , q   r   T cosT 
T sinT 2 , (217)
2  r   T cosT 

d  m, q   r  T cosT   R cosR 
T sinT   R sinR 2 . (218)
2  r  T cosT   R cosR 

169

Appendix B
Since r  T , R  , the denominators in equations (216)‐(218) are approximately
equal to 2r , without any loss of accuracy, i.e., the error introduced by these
approximations is minor for all practical systems. For instance, in a system with
r  23 km, T  30 m, and  R  5 m, the maximum error, i.e., the maximum
difference between (218) and (221) for all values of T and R is less than 103. The
error is even smaller for larger r and smaller T and  R . Then, (216)‐(218) become,
respectively
d  m, p   r   R cosR 
  R sinR 
2
, (219)
2r

d  l , q   r  T cosT 
T sinT 2 ,
(220)
2r

 sinT   R sinR  .
d  m, q   r  T cosT   R cosR  T
2
(221)
2r

Finally, using (211) and (219)‐(221), (189) becomes
T  R sinT sinR
  2v  1  / 2
r
HT
 T  R   2v  1 , T ,T ,R   0 , (222)
2sin T sinT sin R
where v  and  represents a set of integers.
The simplified maximum capacity criterion can be extended to the case of a
nR  nT LoS‐HAP‐MIMO channel. It is trivial to show that (184) can be written as
nt
 j 2 d  a , j  d  b , j   / 
 e , a, b 1,2,..., nR . (223)
j 1
Since ULAs are used at both ends of the communication link, the phase difference
2 d  a , j   d  a , j  1  /  is virtually constant for all j 1,2,..., nT  1 and for all
170

Appendix B
the practical values of distance between transmitter‐receiver and inter‐element
spacing. Therefore, 2 d  a , j   d  b, j   /   2 d  a , j  1  d  b, j  1  /  is also
constant for all j 1,2,..., nT  1. Hence, the solution to (223) for j  p and
j  1  q can be written as [206]
2
2 d  a , p   d  b, p   /   2 d  a , q   d  b, q   /    2v . (224)
nT
Since the array elements at both ends are uniformly distributed and r  T , R 

the following approximation d  a , j   d  b, j    a  b  d  l , j   d  m, j   can be
used. Thus, (224) can be simplified to
2
2 d  l , p   d  m, p   d  l , q   d  m, q   /   2v (225)
nT
and the maximum capacity criterion can be expressed as
 1 
d  l , p   d  m , p   d  l , q   d  m , q  =   v  . (226)
 nT 
Using the derivation for the simplified 2  2 LoS‐HAP‐MIMO maximum capacity

criterion described above, the nR  nT LoS‐HAP‐MIMO maximum capacity criterion
can be written as
 1  HT
T  R   v , T ,T ,R   0 , (227)
 nT  sin T sinT sinR
where v  and  represents a set of integers.
171

172

References
REFERENCES
[1] A. Goldsmith, Wireless Communications, New York, USA: Cambridge University
Press, 2005.
[2] T. S. Rappaport, Wireless Communications: Principles and practice, 2nd ed. Upper
Saddle River, New Jersey, USA: Prentice Hall PTR, 2002.
[3] T. Ojanpera and R. Prasad, “An Overview of Air Interface Multiple Access for IMT
2000/UMTS,” IEEE Communications Magazine, vol. 36, no. 9, pp. 82‐89, 91‐5, Sep.
1998.
[4] D. Collins and C. Smith, 3G Wireless Networks, New York, USA: McGraw‐Hill, 2001.
[5] J. P. Castro, The UMTS network and radio access technology: Air interface
techniques for future mobile systems, Chichester, West Sussex, England: John
Wiley & Sons, 2001.
[6] http://www.3gpp.org/HSPA
[7] H. Holma and A. Toskala, HSDPA/HSUPA for UMTS: High Speed Radio Access for
Mobile Communications, New York, USA: John Wiley & Sons, 2006.
[8] W. Webb, “Broadband fixed wireless access as a key component of the future
integrated communications environment,” IEEE Communications Magazine, vol.
39, no. 9, pp. 115‐121, Sep. 2001.
[9] H. Bolcskel, A. J. Paulraj, K. V. S. Hari, R. U. Nabar, W. W. Lu, “Fixed broadband
wireless access: state of the art, challenges, and future directions,”
IEEE Communications Magazine, vol. 39, no. 1, pp. 100‐108, Jan. 2001.
[10] H. R. Anderson, Fixed Broadband Wireless System Design, Chichester, West Sussex,
England: John Wiley & Sons, Jan. 2003.
[11] A. Aragón‐Zavala, J. L. Cuevas‐Ruíz, and J. A. Delgado‐Penín, High‐Altitude
Platforms for Wireless Communications, New York, USA: John Wiley & Sons, Dec.
2008.
[12] S. Karapantazis and F.‐N. Pavlidou, “Broadband communications via high‐altitude
platforms: A survey,” IEEE Communications Surveys and Tutorials, vol. 7, no. 1, pp.
2‐31, First Quarter 2005.
[13] T. C. Tozer and D. Grace, “High‐altitude platforms for wireless communications,”
IEE Electronics & Communication Engineering Journal, vol. 13, no. 3, pp. 127‐137,
Jun. 2001.
[14] F.‐N. Pavlidou, M. Ruggieri, M. Gerla, and R. Miura, “Communications via High
Altitude Platforms: Technologies and Trials,” International Journal of Wireless
Information Networks, vol. 13, no. 1, pp. 1‐4, Jan. 2006.
[15] A. Mohammed and Z. Yang, “Broadband Communications and Applications from
High Altitude Platforms,” International Journal of Recent Trends in Engineering, vol.
1, no. 3, pp. 239‐243, May 2009.
[16] G. M. Djunnic and J. Freidenfelds, “Establishing wireless communications services
via high‐altitude aeronautical platforms: a concept whose time has come?,” IEEE
Communications Magazine, vol. 35, no. 9, pp. 128‐135, Sep. 1997.
[17] A. K. Widiawan and R. Tafazolli, “High Altitude Platform Station (HAPS): A Review
of New Infrastructure Development for Future Wireless Communications,”
Wireless Personal Communications, vol. 42, no. 3, pp. 387‐404, Aug. 2007.
[18] N. Colella, J. Martin, and I. Akyildiz, “The HALO network,” IEEE Communications
Magazine, vol. 38, no. 6, pp. 142‐148, 2000.
173

References
[19] J. Thornton et.al., “Broadband communications from high altitude platforms – the
European Helinet programme,” IEE Electronics & Communication Engineering
Journal, vol. 13, no. 3, pp. 138‐144, 2001.
[20] Y. Lee and H. Ye, “Sky station stratospheric telecommunications system, a high
speed low latency switched wireless network,” in Proc. 17th AIAA International
Communications Satellite Systems Conference, pp. 25‐32, 1998.
[21] R. Tafazolli, R. Wansch, C. Plenge, J. Horwath, and U. Apel, “Stratospheric
Services,” Technical Note of Stratospheric Platforms ‐ A definition study for an ESA
platform (TN1), 2004.
[22] ESA‐ESTEC contract 162372/02/NL/US, STRATOS: Stratospheric Platforms, a
Definition Study for an ESA Platform, Final Report, 1‐34, Sep. 2005.
[23] Lindstrand, “ESA‐HALE airship research and development program,” in Proc. 2nd
stratospheric platform systems workshop (SPSW) 2000, pp. 99‐107, Tokyo, Japan,
Sep. 2000.
[24] J. Thornton and D. Grace, “Effect of Lateral Displacement of a High Altitude
Platform on Cellular Interference and Handover,” IEEE Transaction on Wireless
Communications, vol. 4. no. 4, pp. 1483‐1490, Jul. 2005.
[25] D. I. Axiotis, M. E. Theologou, and E. D. Sykas, “The effect of platform instability on
the system level performance of HAPS UMTS,” IEEE Communications Letters, vol. 8,
no. 2, pp. 111‐113, Feb. 2004.
[26] K. Katzis, D. A. J. Pearce, and D. Grace, “Impact of high altitude platform
movements on cellular handover,” in Proc. International Workshop on High
Altitude Platform Systems, Athens, Greece, Sep. 2005.
[27] ITU‐R Recommendation V.431: Nomenclature of the frequency and wavelength
bands used in telecommunications, International Telecommunication Union,
Geneva.
[28] IEEE Standard 521‐2002: Standard Letter Designations for Radar‐Frequency Bands
[29] A. D. Panagopoulos, P.‐D. M. Arapoglou, and P. G. Cottis, “Satellite communications
at Ku, Ka, and V bands: Propagation impairments and mitigation techniques,” IEEE
Communications Surveys and Tutorials, vol. 6, no. 3, 3rd Quarter 2004.
[30] J.‐M. Park, B.‐J. Ku, and D.‐S. Oh, “Technical and Regulatory Studies on HAPS,” in
Proc. International Workshop on Aerial & Space Platforms: Research, Applications,
Vision of IEEE Global Communications Conference (Globecom) 2008, pp. 1‐5, New
Orleans, LA, USA, Dec. 4, 2008.
[31] B.‐J. Ku, D.‐S. Ahn, and N. Kim, “An Evaluation of Interference Mitigation Schemes
for HAPS Systems,” EURASIP Journal on Wireless Communications and Networking,
vol. 2008, 2008.
[32] V. F. Milas and P. Constantinou, “Interference Environment between High Altitude
Platform Networks (HAPN) Geostationary (GEO) Satellite and Wireless Terrestrial
Systems,” Wireless Personal Communications, vol. 32, pp. 257‐274, Feb. 2005.
[33] RESOLUTION 122 (Rev.WRC‐07), Use of the bands 47.2‐47.5 GHz and 47.9‐48.2
GHz by high‐altitude platform stations in the fixed service and by other services,
International Telecommunication Union, Geneva, Switzerland, 2007.
[34] RESOLUTION 145 (Rev.WRC‐07), Use of the bands 27.9‐28.2 GHz and 31‐31.3 GHz
by high‐altitude platform stations in the fixed service, International
Telecommunication Union, Geneva, Switzerland, 2007.
[35] RESOLUTION 221 (Rev.WRC‐07), Use of high‐altitude platform stations providing
IMT‐2000 in the bands 1885‐1980 MHz, 2010‐2025 MHz and 2110‐2170 MHz in
Regions 1 and 3 and 1885‐1980 MHz and 2110‐2160 MHz in Region 2, International
Telecommunication Union, Geneva, Switzerland, 2007.
174

References
[36] RESOLUTION 734 (Rev.WRC‐07), Studies for spectrum identification for gateway
links for high‐altitude platform stations in the range from 5850 to 7075 MHz,
International Telecommunication Union, Geneva, Switzerland, 2007.
[37] D. Grace, N. E. Daly, T. C. Tozer, A. G. Burr, and D. A. J. Pearce, “Providing
multimedia communications services from high altitude platforms,” International
Journal of Satellite Communications, vol. 19, no. 6, pp. 559‐580, Nov. 2001.
[38] M. Mondin, F. Dovis, and P. Mulassano, “On the use of HALE platforms as GSM
base stations,” IEEE Personal Communications, vol. 8, no. 2, pp. 37‐44, Apr. 2001.
[39] B. T. Ahmed, M. C. Ramon, and L. H. Ariet, “On the UMTS‐HSDPA in high altitude
platforms (HAPs) communications,” in Proc. 3rd International Symposium on
Wireless Communication Systems (ISWCS) 2006, pp. 704‐708, Valencia, Spain, Sep.
2006.
[40] B. Taha‐Ahmed, M. Calvo‐Ramon, and L. de Haro‐Ariet, “High altitude platforms
(HAPs) W‐CDMA system over cities,” in Proc. IEEE 61st Vehicular Technology
Conference (VTC Spring) 2005, vol. 4, pp. 2673‐2677, Stockholm, Sweden, 30 May‐
1 Jun. 2005.
[41] T. Toshiaki and H. Masatoshi, “Navigation and Positioning System Using High
Altitude Platforms Systems (HAPS),” Journal of the Japan Society for Aeronautical
and Space Sciences, vol. 52, no. 603, pp. 175‐185, 2004.
[42] K. Akalestos, T. C. Tozer, and D. Grace, “Emergency Communications from High
Altitude Platforms,” in Proc. Int. Workshop on High Altitude Platform Systems,
Athens, Greece, Sep. 2005.
[43] J. Holis and P. Pechac, “Coexistence of Terrestrial and HAP 3G Networks during
Disaster Scenarios,” Radioengineering, vol. 17, no. 4, pp. 1‐7, 2008.
[44] J. D. Deaton, “High Altitude Platforms for Disaster Recovery: Capabilities,
Strategies, and Techniques for Emergency Telecommunications,” EURASIP Journal
on Wireless Communications and Networking, vol. 2008, May 2008.
[45] M. J. H. Lum, J. Rosen, H. King, D. C.W. Friedman, G. Donlin, G. Sankaranarayanan,
B. Harnett, L. Huffman, C. Doarn, T. Broderick, and B. Hannaford, “Telesurgery Via
Unmanned Aerial Vehicle (UAV) with a Field Deployable Surgical Robot,” Studies in
health technology and informatics, vol. 125, pp. 313‐315, 2006.
[46] L. Jamison, G.S. Sommer, and I.R. Porche, “High‐Altitude Airships for the Future
Force Army,” Technical Report 234, RAND Arroyo Center, Jan. 2005.
[47] ITU‐T, Vocabulary of terms for broadband aspects of ISDN, I.113 Recommendation,
1992.
[48] C. Morlet, A. Bolea‐Alamañac, G. Gallinaro, L. Erup, P. Takats, and A. Ginesi,
“Introduction of Mobility Aspects for DVB‐S2/RCS Broadband Systems,”
International Journal on Space Communications, vol. 21, no. 1‐2, pp. 5‐17, 2007.
[49] J. Holis and P. Pechac, “Provision of 3G Mobile Services in Sparsely Populated Areas
Using High Altitude Platforms,” Radioengineering, vol. 17, no.1, pp. 43‐49, Apr.
2008.
[50] I. R. Palma‐Lázgare, J. A. Delgado‐Penín, and F. Pérez‐Fontán, “An Advance in
Wireless Broadband Communications based on a WiMAX‐HAPS Architecture,” in
Proc. 26th International Communications Satellite Systems Conference (ICSSC), San
Diego, CA, 10‐12 Jun. 2008.
[51] J. Thornton, A. D. White, and T. C. Tozer, “A WiMAX Payload for High Altitude
Platform Experimental Trials,” EURASIP Journal on Wireless Communications and
Networking, vol. 2008, May 2008.
175

References
[52] Air Interface for Fixed and Mobile Broadband Wireless Access Systems‐
Amendment for Physical and Medium Access Control Layers for Combined Fixed
and Mobile Operation in Licensed Band, IEEE Std. 802.16e‐2005, 2005.
[53] J. G. Andrews, A. Ghosh, and R. Muhamed, Fundamentals of Wimax:
Understanding Broadband Wireless Networking, Upper Saddle River, NJ: PTR
Prentice Hall, Feb. 2007.
[54] A. Bria, M. Flament, F. Gessler, O. Queseth, R. Stridh, M. Unbehaun, J. Wu, and J.
Zander, “4th Generation wireless infrastructures ‐ scenarios & research
challenges,” IEEE Personal Communications‐ Special Edition, Dec. 2001.
[55] B. G. Evans and K. Baughan, “Visions of 4G,” IEE Electronics & Communication
Engineering Journal, vol. 12, no. 6, pp. 293‐303, Dec. 2000.
[56] H.‐H. Chen and M. Guizani, Next Generation Wireless Systems and Networks,
Chichester, Hoboken, NJ: Wiley, Apr. 2006.
[57] S. Y. Hui and K. H. Yeung, ”Challenges in the Migration to 4G Mobile Systems,” IEEE
Communications Magazine, vol. 41, no. 12, pp. 54‐59, Dec. 2003.
[58] H. Ekström, A. Furuskar, J. Karlsson, M. Meyer, S. Parkvall, and J. Torsner,
“Technical Solutions for the 3G Long‐Term Evolution,” IEEE Communications
Magazine, vol. 44, no. 3, pp. 38‐45, 2006.
[59] S. Sesia, M. Baker, and I. Toufik, LTE‐The UMTS Long Term Evolution: From Theory
to Practice, 2nd ed. New York, USA: Wiley, Oct. 2010.
[60] http://www.3gpp.org/LTE‐Advanced
[61] http://www.wimaxforum.org/
[62] P. Demestichas, G. Vivier, K. El‐Khazen, and M. Theologou, “Evolution in wireless
systems management concepts: from composite radio to reconfigurability,” IEEE
Communications Magazine, vol. 42, no. 5, pp. 90‐98, May 2004.
[63] V. Stavroulaki, S. Buljore, P. Roux, and E. Melin, “Equipment management issues in
B3G, end‐to‐end reconfigurable systems,” IEEE Wireless Communications, vol. 13,
no. 3, pp. 24‐32, Jun. 2006.
[64] S. Karapantazis and F.‐N. Pavlidou, “The role of high altitude platforms in beyond
3G networks,” IEEE Wireless Communications, vol. 12, no. 6, pp. 33‐41, Dec. 2005.
[65] F. Dovis, R. Fantini, M. Mondin, and P. Savi, “4G Communications Based on High
Altitude Stratospheric Platforms: Channel Modeling and Performance Evaluation,”
in Proc. IEEE Global Communications Conference (Globecom) 2001, vol. 1, pp. 557‐
561, San Antonio, TX, USA, Nov. 2001.
[66] D. Grace, M. Mohorcic, M. H. Capstick, M. B. Pallavicini, and M. Fitch, “Integrating
users into the wider broadband network via high altitude platforms,” IEEE
Transactions on Wireless Communications, vol. 12, no. 5, pp. 98‐105, Oct. 2005.
[67] E. Falletti, M. Laddomada, M. Mondin, and F. Sellone, “Integrated services from
high‐altitude platforms: a flexible communication system,” IEEE Communications
Magazine, vol. 44, no. 2, pp. 85‐94, Feb. 2006.
[68] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space‐Time Wireless
Communications, Cambridge, UK: Cambridge University Press, 2003.
[69] E. Biglieri, R. Calderbank, A. Constantinides, A. Goldsmith, A. Paulraj, and H.
Vincent Poor, MIMO Wireless Communications, Cambridge, UK: Cambridge
University Press, 2007.
[70] M. Pätzold, Mobile Fading Channels, Chichester, West Sussex, UK: John Wiley &
Sons, Feb. 2002.
[71] S. R. Saunders and A. Aragón‐Zavala, Antennas and Propagation for Wireless
Communication Systems, 2nd ed. Chichester, West Sussex, UK: John Wiley & Sons,
2007.
176

References
[72] W. Pawlowski, “Radio wave propagation effects in high‐altitude platform systems,”
in Proc. International Conference on Microwaves, Radar and Wireless
Communications (MIKON) 2000, vol. 1, pp. 185‐188, Wroclaw, Poland, May 2000.
[73] G. Kandus, M. Mohorcic, E. Leitgeb, and T. Javornik, “Modelling of atmospheric
impairments in stratospheric communications,” in Proc. 2nd WSEAS international
Conference on Circuits, Systems, Signal and Telecommunications, pp. 86‐91,
Acapulco, Mexico, Jan. 25 ‐ Feb. 27 2008.
[74] M. Smolnikar, M. Mohorcic, T. Javornik, and D. Grace, “Propagation Impairment
Countermeasures in Mobile Stratospheric Operating Environment,” in Proc. IEEE
69th Vehicular Technology Conference (VTC Spring) 2009, pp. 1‐5, Barcelona, Spain,
26‐29 Apr. 2009.
[75] Q. Feng, J. McGeehan, E. K. Tameh, and A. R. Nix, “Path Loss Models for Air‐to‐
Ground Radio Channels in Urban Environments,” in Proc. IEEE 63rd Vehicular
Technology Conference (VTC Spring) 2006, vol. 6, pp. 2901‐2905, Melbourne,
Australia, 7‐10 May 2006.
[76] Iskandar and S. Shimamoto, “Prediction of Propagation Path Loss for Stratospheric
Platforms Mobile Communications in Urban Site LOS/NLOS Environment,” in Proc.
IEEE International Conference on Communications (ICC) 2006, vol. 12, pp. 5643‐
5648, Istanbul, Turkey, Jun. 2006.
[77] C. L. Spillard, T. C. Tozer, B. Gremont, and D. Grace, “The Performance of High‐
Altitude Platform Networks in Rainy Conditions,” in Proc. 22nd AIAA International
Communications Satellite Systems Conference, Monterey, USA, 9‐12 May 2004.
[78] S. Zvanovec, P. Piksa, M. Mazanek, and P. Pechac, “A study of gas and rain
propagation effects at 48 GHz for HAP scenarios,” EURASIP Journal on Wireless
Communications and Networking, vol. 2008, pp. 1‐7, Jan. 2008.
[79] A. Mohammed and T. Hult, “Evaluation of Depolarization Effects on the
Performance of High Altitude Platforms (HAPs),” in Proc. IEEE 67th Vehicular
Technology Conference (VTC Spring) 2008, pp. 2942‐2946, Singapore, 11‐14 May
2008.
[80] S. K. Agrawal and P. Garg, “Calculation of Channel Capacity and Rician Factor in the
Presence of Vegetation in Higher Altitude Platforms Communication Systems,” in
Proc. 15th International Conference on Advanced Computing and Communications
(ADCOM) 2007, pp. 243‐248, India, Dec. 2007.
[81] S. K. Agrawal and P. Garg, “Effect of urban‐site and vegetation on channel capacity
in higher altitude platform communication system,” IET Microwaves, Antennas and
Propagation, vol. 3, no. 4, pp. 703‐713, Jun. 2009.
[82] ‘‘Specific attenuation model for rain for use in prediction methods,’’ International
Telecommunication Union, Geneva, Switzerland, ITU‐R P.838‐1, 1997.
[83] ‘‘Characteristics of precipitation for propagation modeling,” International
Telecommunication Union, Geneva, Switzerland, ITU‐R P.837‐4, 2003.
[84] A. D. Panagopoulos, E. M. Georgiadou, and J. D. Kanellopoulos, “Selection
Combining Site Diversity Performance in High Altitude Platform Networks,” IEEE
Communications Letters, vol. 11, no. 10, pp.787‐789, Oct. 2007.
[85] T. Konefal, C. L. Spillard, and D. Grace, “Site diversity for high‐altitude platforms: a
method for the prediction of joint site attenuation statistics,” IEE Proceedings of
Microwaves, Antennas and Propagation, vol. 149, no. 2, pp. 124‐128, Apr. 2002.
[86] J. L. Cuevas‐Ruíz and J. A. Delgado‐Penín, “HAPS Systems Performance Using a Ka‐
Band Channel Model Based on a Time Series Generator,” in Proc. 15th
International Conference on Electronics, Communications and Computers
(CONIELECOMP) 2005, pp. 10‐15, Puebla, Mexico, Feb. 28 ‐ 2 Mar. 2005.
177

References
[87] C. L. Spillard, D. Grace, J. Thornton, and T. C. Tozer, “Effect of ground station
antenna beamwidth on rain scatter interference in high‐altitude platform links,”
Electronics Letters, vol. 38, no. 20, pp. 1211‐1213, Sep. 2002.
[88] E. T. Michailidis and A. G. Kanatas, “Capacity Optimized Line‐of‐Sight HAP‐MIMO
Channels for Fixed Wireless Access,” in Proc. International Workshop on Satellite
and Space Communications (IWSSC) 2009, pp. 73‐77, Siena‐Tuscany, Italy, 10‐11
Sep. 2009.
[89] J. Holis and P. Pechac, “Elevation Dependent Shadowing Model for Mobile
Communications via High Altitude Platforms in Built‐Up Areas,” IEEE Transactions
on Antennas and Propagation, vol. 56, no. 4, pp. 1078‐1084, Apr. 2008.
[90] M. Kong, O. Yorkinov, and S. Shimamoto, “Evaluations of Urban Shadowing
Characteristics for HAPS Communications,” in Proc. IEEE 5th Consumer
Communications and Networking Conference (CCNC) 2008, pp. 555‐559, Las Vegas,
USA, Jan. 2008.
[91] F. Dovis, R. Fantini, M. Mondin, and P. Savi, “Small‐scale fading for high‐altitude
platform (HAP) propagation channels,” IEEE Journal on Selected Areas in
Communications, vol. 20, no. 3, pp. 641‐647, Apr. 2002.
[92] D. Gesbert, M. Kountouris, R. W. Heath, Jr., C.‐B. Chae, and T. Salzer, “Shifting the
MIMO Paradigm: From Single User to Multiuser Communications,” IEEE Signal
Processing Magazine, vol. 24, no. 5, pp. 36‐46, Oct. 2007.
[93] T. Hult, A. Mohammed, Z. Yang, and D. Grace, “Performance of a multiple HAP
system employing multiple polarization,” Wireless Personal Communications, vol.
52, no. 1, pp. 105‐117, Jan. 2010.
[94] A. Mohammed and T. Hult, “Capacity evaluation of a high altitude platform
diversity system equipped with compact MIMO antennas,” International Journal of
Recent Trends in Engineering, vol. 1, no. 3, pp. 244‐247, May 2009.
[95] E. Telatar, “Capacity of multi‐antenna Gaussian channels,” European Transactions
on Telecommunications, vol. 10, no. 6, pp. 585‐595, 1999.
[96] G. J. Foschini and M. J. Gans, “On limits of wireless communications in a fading
environment when using multiple antennas,” Wireless Personal Communications,
vol. 6, no. 3, pp. 311‐335, 1998.
[97] J. Salz and J. H. Winters, “Effect of fading correlation on adaptive arrays in digital
mobile radio,” IEEE Transactions on Vehicular Technology, vol. 43, no. 4, pp. 1049‐
1057, Nov. 1994.
[98] D.‐S. Shiu, G. J. Foschini,M. J. Gans, and J.M. Kahn, “Fading correlation and its
effect on the capacity of multielement antenna systems,” IEEE Transactions on
Communications, vol. 48, no. 3, pp. 502‐513, Mar. 2000.
[99] C.‐N. Chuah, D. N. C. Tse, J. M. Kahn, and R. A. Valenzuela, “Capacity scaling in
MIMO wireless systems under correlated fading,” IEEE Transactions on Information
Theory, vol. 48, no. 3, pp. 637‐650, Mar. 2002.
[100] J. G. Proakis and M. Salehi, Digital Communications, 5th ed. New York, USA:
McGraw‐Hill, 2008.
[101] V. Tarokh, N. Seshadri, and A. R. Calderbank, “Space‐time codes for high data rate
wireless communication: Performance criterion and code construction,” IEEE
Transactions on Information Theory, vol. 44, no. 2, pp. 744‐765, Mar. 1998.
[102] A. Abdi and M. Kaveh, “A space‐time correlation model for multielement antenna
systems in mobile fading channels,” IEEE Journal on Selected Areas in
178

References
[103] A. Abdi, C. Tepedelenlioglu, M. Kaveh, and G. Giannakis, “On the estimation of the
K parameter for the Rice fading distribution,” IEEE Communications Letters, vol. 5,
no. 3, pp. 92‐94, Mar. 2001.
[104] A. Jahn, “Propagation considerations and fading countermeasures for mobile
multimedia services,” International Journal on Satellite Communications, vol. 19,
no. 3, pp. 223‐250, 2001.
[105] Iskandar and S. Shimamoto, “Channel characterization and performance
evaluation of mobile communication employing stratospheric platforms,” IEICE
Transactions on Communications, vol. E89‐B, no. 3, pp. 937‐944, Mar. 2006.
[106] P.‐D. Arapoglou, E. T. Michailidis, A. D. Panagopoulos, A. G. Kanatas, and R. Prieto‐
Cerdeira, “The Land Mobile Earth‐Space Channel: SISO to MIMO Modeling from L‐
to Ka‐Bands,” IEEE Vehicular Technology Magazine, Special Issue on Trends in
Mobile Radio Channels: Modeling, Analysis, and Simulation, vol. 6, no. 2, pp. 44‐53,
Jun. 2011.
[107] M. Vázquez‐Castro, F. Pérez‐Fontán, B. Arbesser‐Rastburg, “Channel modeling for
satellite and HAPS system design,” Wireless Communications and Mobile
Computing, vol. 2, no. 3, pp. 285‐300, 2002.
[108] Y. Yang, R. Zong, X. Gao, and J. Cao, “Channel modeling for High‐Altitude Platform:
A review,” in Proc. International Symposium on Intelligent Signal Processing and
Communication Systems (lSPACS) 2010, pp. 1‐4, Cheng Du, China, 6‐8 Dec. 2010.
[109] S. Liu, Z. Niu, and Y. Wu, “A blockage based channel model for high altitude
platform communications,” in Proc. IEEE 57th IEEE Vehicular Technology
Conference (VTC Spring) 2003, vol. 2, pp. 1051‐1055, Seoul, Korea, 22‐25 Apr.
2003.
[110] J. L. Cuevas‐Ruíz and A. Aragón‐Zavala, G. A. Medina‐Acosta, and J. A. Delgado‐
Penín, “Multipath propagation model for High Altitude Platform (HAP) based on
circular straight cone geometry,” in Proc. International Workshop on Satellite and
Space Communications (IWSSC) 2009, pp. 235‐239, Siena‐Tuscany, Italy, 9‐11 Sep.
2009.
[111] J. L. Cuevas‐Ruíz and J. A. Delgado‐Penín, “A statistical switched broadband
channel model for HAPS links,” in Proc. IEEE Wireless Communications and
Networking Conference (WCNC) 2004, Atlanta, USA, Mar. 2004.
[112] J. L. Cuevas‐Ruíz and J. A. Delgado‐Penín, “Channel model based on semi‐
Markovian processes, an approach for HAPS systems,” in Proc. XIV International
Conference on Electronics, Communications, and Computers (CONIELECOMP) 2004,
Veracruz, Mexico, pp. 52‐56, Feb. 2004.
[113] E. Lutz, D. Cygan, M. Dippold, F. Dolainsky, and W. Papke, “The land mobile
satellite communications channel ‐ Recording, statistics, and channel model,” IEEE
Transactions on Vehicular Technology, vol. 40, no. 2, pp. 375‐386, May 1991.
[114] Y. Karasawa, K. Kimura, and K. Minamisono, “Analysis of availability improvements
in LMSS by means of satellite diversity based on three‐state propagation channel
model,” IEEE Transactions on Vehicular Technology, vol. 46, no. 4, pp. 1047‐1056,
Nov. 1997.
[115] F. Perez‐Fontan, J. P. González, M. J. S. Ferreiro, M. A. V. Castro, S. Buonomo, and J.
P. Baptista, “Complex envelope three‐state Markov model based simulator for the
narrow‐band LMS channel,” International Journal of Satellite Communications, vol.
15, no. 1, pp. 1‐15, Jan. 1997.
[116] L.E. Bråten and T. Tjelta, “Semi‐Markov multistate modeling of the land mobile
propagation channel for geostationary satellites,” IEEE Transactions on Antennas
and Propagation, vol. 50, no. 12, pp. 1795‐1802, Dec. 2002.
179

References
[117] J. Holis and P. Pechac, “Penetration loss measurement and modeling for HAP
mobile systems in urban environment,” EURASIP Journal on Wireless
Communications and Networking, 2008, vol. 2008, Jan. 2008.
[118] F. Perez‐Fontan, V. Hovinen, M. Schönhuber, R. Prieto‐Cerdeira, J. A. Delgado‐
Penín, F. Teschl, J. Kyrölänen, and P. Valtr,” Building entry loss and delay spread
measurements on a simulated HAP‐to‐indoor link at S‐band,” EURASIP Journal on
Wireless Communications and Networking, vol. 2008, Jan. 2008.
[119] D. I. Axiotis and M. E. Theologou, “An empirical model for predicting building
penetration loss at 2 GHz for high elevation angles,” IEEE Antennas and Wireless
Propagation Letters, vol.2, no.1, pp.234‐237, 2003.
[120] T. S. Rappaport, and J. C. Liberti, “A geometrical‐based model for line‐of sight
multipath radio channel,” in Proc. IEEE 46th Vehicular Technology Conference (VTC)
1996, Atlanta, GA, USA, 844‐848, April/May 1996.
[121] P. Almers, E. Bonek, A. Burr, et al., “Survey of Channel and Radio Propagation
Models for Wireless MIMO Systems,” EURASIP Journal on Wireless
Communications and Networking, vol. 2007, 2007.
[122] K. Yu and B. Ottersten, “Models for MIMO propagation channels: a review,”
Wireless Communications and Mobile Computing, vol. 2, no. 7, pp. 653‐666, Nov.
2002.
[123] Iskandar, S. Shimamoto, “Ray Tracing for Urban Site Propagation in Stratospheric
Platform Mobile Communications,” in Proc. Asia‐Pacific Conference on
Communications, pp. 212‐216, Perth, Australia, 3‐5 Oct. 2005.
[124] A. F. Molisch, M. Steinbauer, M. Toeltsch, E. Bonek, and R.S. Thoma, “Capacity of
MIMO systems based on measured wireless channels,” IEEE Journal on Selected
Areas in Communications, vol. 20, no. 3, pp. 539‐549, Apr. 2002.
[125] P. R. King, Modelling and Measurement of the Land Mobile Satellite MIMO Radio
Propagation Channel, PhD Thesis, University of Surrey, 2007.
[126] P.‐D. Arapoglou, K. Liolis, M. Bertinelli, A. Panagopoulos, P. Cottis, and R. De
Gaudenzi, “MIMO over Satellite: A Review,” IEEE Communications Surveys &
Tutorials, vol. 13, no. 1, pp.27‐51, First Quarter 2011.
[127] C. Tzaras, B.G. Evans and S.R. Saunders, “Physical‐statistical analysis of land
mobile‐satellite channel,” Electronics Letters, vol. 34, no. 13, pp. 1355‐1357, Jun.
1998.
[128] P. R. King, B. G. Evans, and S. Stavrou, “Physical‐statistical model for the land
mobile‐satellite channel applied to satellite/HAP MIMO,” in Proc. 11th European
Wireless Conference 2005, vol. 1, pp. 198‐204, Nicosia, Cyprus, Apr. 2005.
[129] G. Zang, B. Huang, and J. Mu, “One scheme of cooperative diversity with two
satellites based on the Alamouti code,” in Proc. IET 3rd International Conference on
Wireless, Mobile and Multimedia Networks (ICWMNN) 2010, pp. 151‐154, 26‐29
Sept. 2010.
[130] D. S. Ahn, S. Kim, H. W. Kim, and D.‐C. Park, “A cooperative transmit diversity
scheme for mobile satellite broadcasting systems,” International Journal of
Satellite Communications and Networking, vol. 28, no. 5‐6, pp. 352‐368, Dec. 2010.
[131] P. Horvath and I. Frigyes, “Application of the 3D Polarization Concept in Satellite
MIMO Systems,” in Proc. IEEE Global Communications Conference (Globecom)
2006, pp. 1‐5, Nov. 27 2006 ‐ Dec. 1 2006.
[132] K. P. Liolis, A. D. Panagopoulos, and P. G. Cottis, “Multi‐satellite MIMO
communications at Ku band and above: Investigations on spatial multiplexing for
capacity improvement and selection diversity for interference mitigation,”
EURASIP Journal on Wireless Communications and Networking, vol. 2007, 2007.
180

References
[133] P.‐D. Arapoglou, M. Zamkotsian, and P. Cottis, “Dual polarization MIMO in LMS
broadcasting systems: Possible benefits and challenges,” International Journal of
Satellite Communications and Networking, 2010,
[134] P. Horvath, G. Karagiannidis, P.R. King, S. Stavrou, I. Frigyes, “Investigations in
satellite MIMO channel modeling: Accent on polarization,” EURASIP Journal on
Wireless Communications and Networking, vol. 2007, 2007.
[135] M. Sellathurai, P. Guinand, and J. Lodge, “Space‐time coding in mobile satellite
communications using dual‐polarized channels,” IEEE Transactions on Vehicular
Technology, vol. 55, no. 1, pp. 188‐199, Jan. 2006.
[136] See for details of the MiLADY (Mobile satellite channeL with Angle DiversitY)
project: [Online]. Available:
http://telecom.esa.int/telecom/www/object/index.cfm?fobjectid=29020
[137] F. Lacoste, F. Carvalho, F. Perez Fontan, A. Nunez Fernandez, V. Fabro, G. Scot,
“Polarization and spatial diversity measurements of the Land Mobile Satellite
propagation channel at S‐band,” COST Action IC0802, 2nd MCM, ESA/ESTEC,
Toulouse, France, 4‐6 Nov. 2009.
[138] K. Liolis, I. Andrikopoulos, and P. Cottis, “On statistical modeling and performance
evaluation of SIMO land mobile satellite channels,” in Proc. 4th Advanced Satellite
Mobile Systems (ASMS) 2008, pp. 76‐81, Bologna, 26‐28 Aug. 2008.
[139] E. T. Michailidis and A. G. Kanatas, “Three‐Dimensional HAP‐MIMO Channels:
Modeling and Analysis of Space‐Time Correlation,” IEEE Transactions on Vehicular
Technology, vol. 59, no. 5, pp. 2232‐2242, Jun. 2010.
[140] E. T. Michailidis, and A. G. Kanatas, “A Three Dimensional Model for Land Mobile‐
HAP‐MIMO Fading Channels,” in Proc. 10th International Workshop on Signal
Processing for Space Communications (SPSC) 2008, Rhodes Island, Greece, 6‐8 Oct.
2008.
[141] E. T. Michailidis, G. Efthymoglou, and A. G. Kanatas, “Spatially Correlated 3‐D HAP‐
MIMO Fading Channels,” in Proc. International Workshop on Aerial & Space
Platforms: Research, Applications, Vision of IEEE Global Communications
Conference (Globecom) 2008, pp. 1‐7, New Orleans, LA, USA, 4 Dec. 2008.
[142] S.‐Y. Leong, Y. R. Zheng, and C. Xiao, “Space‐Time Fading Correlation Functions of a
3‐D MIMO Channel Model,” in Proc. IEEE Wireless Communications Networking
Conference (WCNC) 2004, vol. 2, pp. 1127‐1132, Atlanta, 21‐25 Mar. 2004.
[143] A. G. Zajić, and G. L. Stüber, “Three‐dimensional modeling, simulation, and capacity
analysis of space‐time correlated mobile‐to‐mobile channels,” IEEE Transactions on
Vehicular Technology, vol. 57, no. 4, pp. 2042‐2054, Jul. 2008.
[144] G. L. Stüber, Principles of Mobile Communications, 2nd edition. Norwell, MA:
Kluwer, 2001.
[145] W. C. Jakes, Jr., “Multipath interference,” in Microwave Mobile communications,
New York, Wiley, 1974, pp. 11‐78.
[146] K. I. Pedersen, P. E. Mogensen, and B. H. Fleury, “Power azimuth spectrum in
outdoor environments,” Electronics Letters, vol. 33, no. 18, pp. 1583‐1584, Aug.
1997.
[147] A. Abdi, J. A. Barger, and M. Kaveh, “A parametric model for the distribution of the
angle of arrival and the associated correlation function and power spectrum at the
mobile station,” IEEE Transactions on Vehicular Technology, vol. 51, no. 3, pp. 425‐
434, 2002.
[148] R. Von Mises, “Über die “Ganzzahligkeit” der Atomgewicht und verwandte
Fragen,” Physikalische Zeitschrift, vol. 19, pp. 490‐500, 1918.
181

References
[149] K. V. Mardia and P. E. Jupp, Directional Statistics, New York, USA: John Wiley &
Sons, 2000.
[150] M. K. Simon, S. M. Hinedi, and W. C. Lindsey, Digital Communication Techniques‐
Signal Design and Detection, Upper Saddle River, NJ: PTR Prentice Hall, 1995.
[151] A. J. Viterbi, “Phase‐locked loop dynamics in the presence of noise by Fokker‐
Planck techniques,” in Proc. of the IEEE, vol. 51, no. 12, pp. 1737‐1753, Jun. 1963.
[152] H. Leib and S. Pasupathy, “The phase of a vector perturbed by Gaussian noise and
differentially coherent receivers”, IEEE Transactions on Information Theory, vol. 34,
no. 6, pp. 1491‐1501, 1988.
[153] M. D. Austin and G. L. Stuber, “Velocity adaptive handoff algorithms for
microcellular systems,” IEEE Transactions on Vehicular Technology, vol. 43, no. 3,
pp. 549‐561, 1994.
[154] S. S. Mahmoud, Z. M. Hussain, and P. O'Shea, “Space‐time model for mobile radio
channel with hyperbolically distributed scatterers,” IEEE Antennas and Wireless
Propagation Letters, vol. 1, pp. 211‐214.
[155] S. S. Mahmoud, Z. M. Hussain, and P. O'Shea, P., “A geometrical‐based microcell
mobile radio channel model,” Wireless Networks, vol. 12, no. 5, pp. 653‐664, Sep.
2006.
[156] S. S. Mahmoud, F. S. Al‐Qahtani, Z. M. Hussain, and A. Gopalakrishnan, “Spatial and
temporal statistics for the geometrical‐based hyperbolic macrocell channel
model,” Digital Signal Processing, vol. 18, no. 2, pp. 151‐167, Mar. 2008.
[157] M.A. Vázquez‐Castro, F. Perez‐Fontan and S. R. Saunders, “Shadowing correlation
assessment and modeling for satellite diversity in urban environments,”
International Journal of Satellite Communications, vol. 20, no. 2, March/April 2002,
pp. 151‐166.
[158] M. A. Vázquez‐Castro, D. Belay‐Zeleke, A. Curieses‐Guerrero, “Availability of

systems based on satellites with spatial diversity and HAPS,” Electronics Letters,
vol. 38, no. 6, pp. 286‐288, Mar. 2002.
[159] C. Tzaras, B.G. Evans and S.R. Saunders, “Physical‐statistical analysis of land
mobile‐satellite channel,” Electronics Letters, vol. 34, no. 13, pp. 1355‐1357, Jun.
1998.
[160] S. R. Saunders and B. G. Evans, “Physical model of shadowing probability for land
mobile satellite propagation,” Electronics Letters, vol. 32, no. 17, pp. 1548‐1549,
Aug. 1996.
[161] I. S. Gradshteyn, and I. M. Ryzhik, Table of Integrals, Series and Products, 5th
edition, Edited by A. Jeffrey, New York, USA: Academic Press, 1994.
[162] E. T. Michailidis and A. G. Kanatas, “On the Capacity of 3‐D Space‐Time Correlated
HAP‐MIMO Channels,” in Proc. 2nd International Conference on
Advances in Satellite and Space Communications (SPACOMM) 2010, Athens‐
Glyfada, Greece, 13‐19 Jun. 2010.
[163] E. T. Michailidis and A. G. Kanatas, “Capacity Analysis and Simulation of 3‐D HAP‐
MIMO Channels,” accepted for publication in International Journal on Advances in
Telecommunications.
[164] D. J. Young and N. C. Beaulieu, “The generation of correlated Rayleigh random
variates by inverse discrete Fourier transform,” IEEE Transactions on
Communications, vol. 48, no. 7, pp. 1114‐1127, Jul. 2000.
[165] D. Verdin and T. C. Tozer, “Generating a fading process for the simulation of land‐
mobile radio communications,” Electronics Letters, vol. 29, no.23, pp. 2011‐2012,
Nov. 1993.
182

References
[166] S. A. Fechtel, “A novel approach to modeling and efficient simulation of frequency‐
selective fading radio channels,” IEEE Journal on Selected Areas in
[167] R. H. Clarke, “A statistical theory of mobile‐radio reception,” Bell Systems Technical
Journal, vol. 47, pp. 957‐1000, Jul. 1968.
[168] M. F. Pop and N. C. Beaulieu, “Limitations of sum‐of‐sinusoids fading channel
simulators,” IEEE Transactions on Communications, vol. 49, no. 4, pp. 699‐708, Apr.
2001.
[169] C. Xiao, Y. R. Zheng, and N. C. Beaulieu, “Novel sum‐of‐sinusoids simulation models
for Rayleigh and Rician fading channels,” IEEE Transactions on Wireless
Communications, vol. 5, no. 12, pp. 3667‐3679, Dec. 2006.
[170] M. Pätzold, U. Killat, F. Laue, and Y. Li, “On the statistical properties of
deterministic simulation models for mobile fading channels,” IEEE Transactions on
Vehicular Technology, vol. 47, no. 1, pp. 254‐269, Feb. 1998.
[171] C.‐X. Wang, M. Patzold, and D. Yuan, “Accurate and efficient simulation of multiple
uncorrelated Rayleigh fading waveforms,” IEEE Transactions on Wireless
Communications, vol.6, no.3, pp.833‐839, Mar. 2007.
[172] C. S. Patel, G. L. Stüber, and T. G. Pratt, “Comparative analysis of statistical models
for the simulation of Rayleigh faded cellular channels,” IEEE Transactions on
Communications, vol. 53, no. 6, pp. 1017‐1026, Jun. 2005.
[173] Y. R. Zheng and C. Xiao, “Simulation models with correct statistical properties for
Rayleigh fading channels,” IEEE Transactions on Communications, vol. 51, no. 6, pp.
920‐928, Jun. 2003.
[174] Y. R. Zheng and C. Xiao, “Improved models for the generation of multiple
uncorrelated Rayleigh fading waveforms,” IEEE Communications Letters, vol. 6, no.
6, pp. 256‐258, Jun. 2002.
[175] A. S. Akki and F. Haber, “A statistical model for mobile‐to‐mobile land
communication channel,” IEEE Transactions on Vehicular Technology, vol. 35, no.
1, pp. 2‐10, Feb. 1986.
[176] A. S. Akki, “Statistical properties of mobile‐to‐mobile land communication
channels,” IEEE Transactions on Vehicular Technology, vol. 43, no. 4, pp. 826‐831,
Nov. 1994.
[177] C. S. Patel, G. L. Stüber, and T. G. Pratt, “Simulation of Rayleigh‐faded mobile‐to‐
mobile communication channels,” IEEE Transactions on Communications, vol. 53,
no. 11,pp. 1876‐1884, Nov. 2005.
[178] S. O. Rice, “Mathematical analysis of random noise,” Bell Systems Technical
Journal, vol. 23, pp. 282‐332, Jul. 1944.
[179] E. T. Michailidis and A. G. Kanatas, “Statistical Simulation Modeling of 3‐D HAP‐
MIMO Channels,” accepted for publication in Wireless Personal Communications.
[180] M. Pätzold, B. O. Hogstad, “A space‐time channel simulator for MIMO channels
based on the geometrical one‐ring scattering model,” Wireless Communications
and Mobile Computing, Special Issue on Multiple‐Input Multiple‐Output (MIMO)
Communications, vol. 4, no. 7, pp. 727‐737, Nov. 2004.
[181] M. Pätzold, B. O. Hogstad, and D. Kim, “A New Design Concept for High‐
Performance Fading Channel Simulators Using Set Partitioning,” Wireless Personal
Communications, vol. 40, no. 2, pp. 267‐279, Feb. 2007.
[182] C. A. Gutierrez‐Diaz‐de‐Leon, M. Patzold, “Sum‐of‐sinusoids‐based simulation of
flat fading wireless propagation channels under non‐isotropic scattering
conditions,” in Proc. IEEE Global Communications Conference (Globecom) 2007, pp.
3842‐3846, 26‐30 Nov. 2007.
183

References
[183] G. Ungerboeck, “Channel coding with multilevel/phase signals,” IEEE Transactions
Information Theory, vol. 28, no. 1, pp. 55‐67, Jan. 1982.
[184] P. A. Bello, “Characterization of randomly time‐variant linear channels,” IEEE
Transactions on Communications Systems, vol. 11, pp. 360‐393, 1963.
[185] Z. Latinovic, A. Abdi, and Y. Bar‐Ness, “A wideband space‐time model for MIMO
mobile fading channels,” in Proc. IEEE Wireless Communications and Networking
Conference (WCNC) 2003, pp. 338‐342, New Orleans, USA, 16‐20 Mar. 2003.
[186] Z. Latinovic, A. Abdi, and Y. Bar‐Ness, “On the utility of the circular ring model for
wideband MIMO channels,” in Proc. IEEE 60th Vehicular Technology Conference
(VTC Fall) 2004, vol. 1, pp. 96‐100, Los Angeles, USA, 26‐29 Sep. 2004.
[187] M. Yuanyuan and M. Pätzold, “A Wideband One‐Ring MIMO Channel Model Under
Non‐Isotropic Scattering Conditions,” in Proc. IEEE Vehicular Technology
Conference (VTC Spring) 2008, pp. 424‐429, Singapore, 11‐14 May 2008.
[188] A. G. Zajić, G. L. Stüber, T. G. Pratt, and S. Nguyen, “Wideband MIMO mobile‐to‐
mobile channels: Geometry‐based statistical modeling with experimental
verification,” IEEE Transactions on Vehicular Technology, vol. 58, no. , pp. 517‐534,
Feb. 2009.
[189] A. G. Zajić and G. L. Stüber, “Three‐dimensional modeling and simulation of
wideband MIMO mobile‐to‐mobile channels,” IEEE Transactions on Wireless
Communications, vol. 8, no. , pp. 1260‐1275, Mar. 2009.
[190] Y. Yamada, Y. Ebine, and N. Nakajima, “Base station/vehicular antenna design
techniques employed in high capacity land mobile communications system,”
Review of the Electrical Communication Laboratory, vol. 35, no. 2, pp. 115‐121,
1987.
[191] A. Kuchar, J. P. Rossi, and E. Bonek, “Directional macro‐cell channel
characterization from urban measurements,” IEEE Transactions on Antennas and
Propagation, vol. 48, no. 2, pp. 137‐146, Feb. 2000.
[192] E. T. Michailidis and A. G. Kanatas, “Wideband HAP‐MIMO Channels: A 3‐D
Modeling Approach,” submitted to IEEE Transactions on Wireless Communications.
[193] M. Döttling, A. Jahn, J. Kunisch, and S. Buonomo, “A versatile propagation channel
simulator for land mobile satellite applications,” in Proc. 48th IEEE Vehicular
Technology Conference (VTC) 1998, Ottawa, Canada, pp. 213‐217, 18‐21 May 1998.
[194] B. H. Fleury, “First‐ and Second‐order characterization of direction dispertion and
space selectivity in the radio channel,” IEEE Transactions on Information Theory,
vol. 46, pp. 2027‐2044, 2000.
[195] T. Aulin, “A modified model for the fading signal at a mobile radio channel,” IEEE
Transactions on Vehicular Technology, vol. 28, no. 3, pp. 182‐203, Aug. 1979.
[196] F. Vatalaro and A. Forcella, “Doppler spectrum in mobile‐to‐mobile
communications in the presence of three‐dimensional multipath scattering,” IEEE
Transactions on Vehicular Technology, vol. 46, no. 1, pp. 213‐219, Feb. 1997.
[197] S. Qu and T. Yeap, “A three‐dimensional scattering model for fading channels in
land mobile environment,” IEEE Transactions on Vehicular Technology, vol. 48, no.
3, pp. 765‐781, May 1999.
[198] C. Kasparis, P. R. King, and B. G. Evans, “Doppler spectrum of the multipath fading
channel in mobile satellite systems with directional terminal antennas,” IET
Communications, vol. 1, no. 6, pp. 1089‐1094, Dec. 2007.
[199] Ε. T. Michailidis and A. G. Kanatas, “Modeling and Simulation of 3‐D Wideband
HAP‐MIMO Channels,” accepted for publication in IEEE 74th Vehicular Technology
Conference (VTC Fall) 2011, San Francisco, USA, 5‐8 Sep. 2011.
184

References
[200] D. Grace, J. Thornton, C. Guanhua, G. P. White, and T. C. Tozer, “Improving the
system capacity of broadband services using multiple high‐altitude platforms,” IEEE
Transactions on Wireless Communications, vol. 4, no. 2, pp. 700‐709, Mar. 2005.
[201] G. Chen, D. Grace, and T. C. Tozer, “Performance of Multiple High Altitude
Platforms using Directive HAP and User Antennas,” Wireless Personal
Communications, vol. 32, no. 3‐4, pp. 275‐299, Feb. 2005.
[202] T. Celcer, G. Kandus, T. Javornik, M. Mohorcic, and S. Plevel, “Evaluation of
diversity gain and system capacity increase in a multiple HAP system,” in Proc.
International Workshop on Satellite and Space Communications (IWSSC) 2006, pp.
114‐118, Leganes‐Madrid, Spain, 14‐15 Sep. 2006.
[203] L. Cottatellucci and M. Debbah, “On the capacity of MIMO rice channels,” in Proc.
42nd Allerton Conference on Communication, Control, and Computing 2004, pp.
1506‐1516, Monticello‐Illinois, USA, 29 Sep. ‐ 1 Oct. 2004.
[204] K. Sakaguchi, H. Y. E. Chua, and K. Araki, “MIMO channel capacity in an indoor line‐
of‐sight (LoS) environment,” IEICE Transactions on Communications, vol. E88‐B, no.
7, pp. 3010‐3019, Jul. 2005.
[205] P. F. Driessen and G. J. Foschini, “On the capacity formula for multiple input‐
multiple output wireless channels: A geometric interpretation,” IEEE Transactions
on Communications, vol. 47, no. 2, pp. 173‐176, Feb. 1999.
[206] I. Sarris and A. R. Nix, “Design and performance assessment of high‐capacity MIMO
architectures in the presence of a line‐of‐sight component,” IEEE Transactions on
Vehicular Technology, vol. 56, no. 4, pp. 2194‐2202, Jul. 2007.
[207] F. Bohagen, P. Orten, and G. E. Oien, “Design of Optimal High‐Rank Line‐of‐Sight
MIMO Channels,” IEEE Transactions on Wireless Communications, vol. 6, no. 4, pp.
1420‐1425, Apr. 2007.
[208] F. Bohagen, P. Orten, and G. E. Oien, “Modeling and analysis of a 40 GHz MIMO
system for fixed wireless access,” in Proc. IEEE 61st Vehicular Technology
Conference (VTC Spring) 2005, vol. 3, pp. 1691‐1695, Stockholm, Sweden, 30 May ‐
1 Jun. 2005.
[209] H. Xu, T. S. Rappaport, R. J. Boyle, and J. H. Schaffner, “Measurements and models
for 38‐GHz point‐to‐multipoint radiowave propagation,” IEEE Journal on Selected
Areas in Communications, vol. 18, pp. 310‐321, Apr. 2000.
[210] R. U. Nabar, Ö. Oyman, H. Bolcskei, and A. J. Paulraj, “Capacity scaling laws in
MIMO wireless networks,” in Proc. 41st Allerton Conference on Communication,
Control, and Computing 2003, pp. 378‐389, Monticello‐Illinois, USA, Oct. 2003.
[211] D. Panagopoulos, K. P. Liolis, and P. G. Cottis, “Rician K‐Factor Distribution in
Broadband Fixed Wireless Access Channels under Rain Fades,” IEEE
Communications Letters, vol. 11, no. 4, pp. 301‐303, Apr. 2007.
[212] M. A. Khalighi, J.‐M. Brossier, G. Jourdain, and K. Raoof, “On capacity of Rician
MIMO channels,” in Proc. IEEE 12th International Symposium on Personal, Indoor
and Mobile Radio Communications (PIMRC) 2001, vol. 1, pp. A150‐A154, San
Diego, USA, September 30 ‐ October 3, 2001.
[213] S. Lu, H. T. Hui, and M. Bialkowski, “Optimizing MIMO Channel Capacities Under
the Influence of Antenna Mutual Coupling,” IEEE Antennas and Wireless
Propagation Letters, vol. 7, pp. 287‐290, 2008.
[214] D. Gesbert, M. Shafi, D.‐S. Shiu, P. J. Smith, and A. Naguib, “From theory to
practice: an overview of MIMO space‐time coded wireless systems,” IEEE Journal
on Selected Areas in Communications, vol. 21, no. 3, pp. 281‐302, Apr. 2003.

185

References
[215] J. N. Martin and N. J. Colella, “Broadband Wireless Services from High Altitude
Long Operation (HALO) Aircraft,” in Proc. SPIE International Symposium on Voice,
Video, and Data Communications: Broadband Engineering for Multimedia Markets,
1997.
[216] A. G. Zajić, G. L. Stüber, T. G. Pratt, and S. Nguyen, “Envelope level crossing rate
and average fade duration in mobile‐to‐mobile Ricean fading channels,” in Proc.
IEEE International Conference on Communications (ICC) 2008, pp. 4446‐4450,
Beijing, China, 19‐23 May 2008.
[217] C. Morlet, A. B. Alamañac, G. Gallinaro, L. Erup, P. Takats, and A. Ginesi,
“Introduction of Mobility Aspects for DVB‐S2/RCS Broadband Systems,”
International Journal on Space Communications, vol. 21, no. 1‐2, pp. 5‐17, 2007.
[218] K. P. Liolis, A. D. Panagopoulos, and S. Scalise, “On the Combination of
Tropospheric and Local Environment Propagation Effects for Mobile Satellite
Systems Above 10 GHz,” IEEE Transactions on Vehicular Technology, vol. 59, no. 3,
pp. 1109‐1120, Mar. 2010.
186

Σύντομο Βιογραφικό Σημείωμα
Σύντομο Βιογραφικό Σημείωμα
Ο Εμμανουήλ Θ. Μιχαηλίδης γεννήθηκε στην Αθήνα τον Ιούλιο του 1980. Είναι
πτυχιούχος Ηλεκτρονικός Μηχανικός Τεχνολογικής Εκπαίδευσης (TE) από το τμήμα
Ηλεκτρονικής του Τεχνολογικού Εκπαιδευτικού Ιδρύματος (ΤΕΙ) Πειραιώς από τον
Ιούνιο του 2004. Έλαβε το Μεταπτυχιακό Δίπλωμα Ειδίκευσης (ΜΔΕ) στο
αντικείμενο των Ψηφιακών Επικοινωνιών και Δικτύων από το τμήμα Ψηφιακών
Συστημάτων του Πανεπιστημίου Πειραιώς το Σεπτέμβριο του 2006.
Από το Σεπτέμβρη του 2006 είναι μέλος του Εργαστηρίου Τηλεπικοινωνιακών
Συστημάτων (ΕΤΣ) του τμήματος Ψηφιακών Συστημάτων του Πανεπιστημίου
Πειραιώς. Από τον Ιούλιο του 2006 έως το Δεκέμβριο του 2007 συμμετείχε ενεργά
στο χρηματοδοτούμενο από την Ευρωπαϊκή Ένωση ερευνητικό πρόγραμμα
αριστείας “European Information Society Technologies (IST) Antenna Centre of
Excellence (ACE)”. Από το 2009 συμμετέχει στην ευρωπαϊκή ερευνητική δράση
COST IC0802 με τίτλο “Propagation Tools and Data for Integrated
Telecommunication, Navigation and Earth Observation Systems”. Παράλληλα με
την ερευνητική του δραστηριότητα είναι εργαστηριακός συνεργάτης στο Τμήμα
Ηλεκτρονικής του ΤΕΙ Πειραιώς από τον Οκτώβριο του 2007 και έχει συμμετάσχει
ως εισηγητής πληροφορικής σε διάφορα προγράμματα επαγγελματικής
κατάρτισης από το Μάρτιο του 2007.
Τα τρέχοντα ερευνητικά του ενδιαφέροντα εστιάζονται στην ανάπτυξη νέων

ψηφιακών τεχνικών για ασύρματα, δορυφορικά και στρατοσφαιρικά συστήματα
επικοινωνιών, καθώς και στο χαρακτηρισμό, τη μοντελοποίηση και την
προσομοίωση του ραδιοδιαύλου διάδοσης μελλοντικών ασύρματων συστημάτων
επικοινωνιών.
Ο κ. Μιχαηλίδης βραβεύτηκε με τη διάκριση καλύτερης εργασίας (Best Paper
Award) στο διεθνές συνέδριο “International Conference on Advances in Satellite
and Space Communications (SPACOMM) 2010” και είναι μέλος του Ινστιτούτου
Ηλεκτρολόγων και Ηλεκτρονικών Μηχανικών (IEEE).
187

188

Vita
Vita
Emmanouel T. Michailidis was born in Athens, Greece, in July 1980. He received
the B.Sc. degree in Electronics Engineering in June 2004 from the Technological
Educational Institute of Piraeus, Piraeus, Greece, and the M.Sc. degree in Digital
Communications and Networks in September 2006 from the University of Piraeus.
Since September 2006, he has been a Graduate Research Engineer with the
Telecommunications Systems Laboratory (TSL), Department of Digital Systems,
University of Piraeus. From July 2006 to December 2007, he was with the European
Information Society Technologies (IST)‐funded research project Antenna Centre of
Excellence (ACE). Since 2009, he has participated in the COST IC0802 action entitled
“Propagation Tools and Data for Integrated Telecommunication, Navigation, and
Earth Observation Systems”. Since 2007, he has been a Laboratory Assistant with
the Department of Electronics, Technological Educational Institute of Piraeus and
he has also been an informatics trainer in vocational training institutes.
His current research interests include the development of new digital
techniques for wireless, satellite, and stratospheric communications systems, and
the radio channel characterization, modeling, and simulation for future wireless
communications systems.
Mr. Michailidis received the Best Paper Award at the “International Conference
on Advances in Satellite and Space Communications (SPACOMM) 2010” and he is a
member of the Institute of Electrical and Electronics Engineers (IEEE).

189

190

Thesis Hap Mimo 2011

Uploaded by

Copyright:

Available Formats

Thesis Hap Mimo 2011

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Thesis Hap Mimo 2011

Uploaded by

Copyright:

Available Formats

TI..ul~a lJ.If"\q.HaKWV rUm:f"\~Ct"(WV

riA A}:YPMATA EYPYZONIKA }:Y}:THMATA

rU~~OUAEUUKr\ EnL"(porrr\: ASavCtCJlo<; r. KavCt"(a<;

A. r. Kavchac; r. Eu8uJ.l.oyAou <D. KwvatavTlvou

I. . KaTolKac; <D.-N. nauAL60u

Copyright© 2011- E~~avou~A 0. MlXaI")AlOI")C;

ME ETIl¢UAa~1") navtOC; olKau.il~aTOC;.

AnayopEuETat I") avtLypa¢~, an09~KEUOI") Kat OLaVO~~ TI")<; napouoa<; Epyaoia<;, E~

--------~---.-- .... -- ~-~-------

Department of Digital Systems

RADIO CHANNEL MODELING FOR BROADBAND WIRELESS

STRATOSPHERIC COMMUNICATIONS SYSTEMS

Doctor of Philosophy Thesis

A. G. Kanatas, Supervisor G. Efthymoglou, Advisor

University of Piraeus University of Piraeus N.T.U.A.

University of Piraeus A.U.TH.

Date Approved: May 9th 2011

Dr. Emmanouel T. Michailidis

Copyright© 2011- Emmanouel T. Michailidis

All rights reserved.

Copying, storing or distributing of this thesis, as part or as a whole, for commercial

Συγκεκριμένα, αναπτύσσονται καινοτόμα θεωρητικά γεωμετρικά μοντέλα

Επιπλέον, αναπτύσσονται καινοτόμα ντετερμινιστικά και στατιστικά μοντέλα

Τέλος, προτείνεται μια καινοτόμος γεωμετρική σχεδιαστική μέθοδος

Τα προτεινόμενα μοντέλα διαύλου HAP‐MIMO παρέχουν σε άλλους ερευνητές

Λέξεις Κλειδιά: Διαλείψεις Ricean, δίαυλος πολλαπλών‐εισόδων πολλαπλών‐

Novel three‐dimensional (3‐D) geometry‐based reference models for Ricean

Finally, a novel geometrical design approach to construct high‐capacity HAP‐

Keywords: Capacity, correlation, high‐altitude platforms (HAPs), multiple‐input

Wireless communications services are traditionally provided by terrestrial and

technologies, services and applications. The tasks of internationally coordinating

frequencies make communications difficult and unreliable [29]. Therefore,

ELF Extra low frequency 3 – 30 Hz 100,000 – 10,000 km

FREQUENCY MICROWAVE AREA/ TYPE OF

In recent years, the multiple‐input multiple‐output (MIMO) technology has

 Conducting real‐world channel measurements and deriving empirical

 Developing accurate, efficient and flexible simulation models for

 Developing hardware‐based channel emulators for real‐time system

“Propagation Tools and Data for Integrated Telecommunication, Navigation, and

Chapter 3 proposes a 3‐D reference model for narrowband HAP‐MIMO

The propagation characteristics crucially determine the performance of any

PR  dBW   PT  dBW   GT  dBi  GR  dBi   L  dB  , (2)

antenna gains, respectively, and L is the path loss. The noise power PN is given by

PN  dBW   10log  N0B   10log  kTeB  , (3)

where N0 is the noise power spectral density (PSD), B is the channel bandwidth, k

is the Boltzmann’s constant ( 1.38  10 23 Joules/Kelvin ), and Te is the system noise

Source Transmitter x x + Receiver Destination

LF  dB   32.4  20log dkm  20log fMHz , (4)

where fMHz is the carrier frequency in MHz, and dkm is the distance between the

Table 4 shows typical values for ar and br at various frequencies for horizontal

where Hr is the effective rain height, HR is the height of the terrestrial station, and

Apart from introducing signal attenuation, rain also introduces scattering

Frequency selectivity refers to a frequency‐varying multipath channel. When

Finally, space selectivity refers to a space‐varying multipath channel. When

Although conventional single‐input single‐output (SISO) systems, i.e., systems

SISO Transmitter Receiver

As modern communications services demand increased channel capacity and

antennas and nR receive antennas potentially offers nT  nR independently fading

Spatial multiplexing requires MIMO antenna configuration and takes full

 h11 h12  h1nT 