This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
IEEE JSAC SatCom'17 1570375434
Journal on Selected Areas in Communications

1  
2  
3  
4  
5  
6  
7  
Beam Tracking for UAV Mounted SatCom
8  
9  
on-the-Move with Massive Antenna Array
10   Jianwei Zhao, Feifei Gao, Qihui Wu, Shi Jin, Yi Wu, and Weimin Jia
11  
12  
13  
1DYLJDWLRQ
14   Abstract—Unmanned aerial vehicle (UAV)-satellite communi- 6DWHOOLWH
6SDFH
15   cation has drawn dramatic attention for its potential to build
the integrated space-air-ground network and the seamless wide-
16   area coverage. A key challenge to UAV-satellite communication
0RQLWRULQJ
6DWHOOLWH
17   is its unstable beam pointing due to the UAV navigation, which &RPPXQLFDWLRQ
18   is a typical SatCom on-the-move scenario. In this paper, we
6DWHOOLWH

19   propose a blind beam tracking approach for Ka-band UAV-

satellite communication system, where UAV is equipped with a 6N\
20  
hybrid large-scale antenna array. The effects of UAV navigation
21   are firstly released through the mechanical adjustment, which
+RWDLU
EDOORRQ
22   could approximately point the beam towards the target satellite 8$9
23   through beam stabilization and dynamic isolation. Specially, the
24   attitude information for mechanical adjustment can be realtimely
25   derived from data fusion of low-cost sensors. Then, the precision (DUWK
of beam pointing is blindly refined through electrically adjusting
26   the weight of the massive antennas, where an array structure 6HD
27   based simultaneous perturbation algorithm is designed. Simula-
28   tion results are provided to demonstrate the superiority of the 'LVDVWHU &URZGHG
29   proposed method over the existing ones. 8UJHQWFRPPXQLFDWLRQ
DUHD
+RFFRPPXQLFDWLRQ
DUHD

30   Index Terms—UAV, SatCom on-the-move, beam tracking,

31   mechanical adjustment, electrical adjustment, massive antenna Fig. 1. The integrated space-air-ground network.
32   array, hybrid.
33  
34   destroyed, or could assist the existing communications, e.g.,
I. I NTRODUCTION
35   the rapid service recovery and the base station (BS) offloading
36   Satellite communication has become a promising solution of the extremely crowded areas. In these circumstances, UAV
37   to realize high data rates anywhere, anytime, and towards the acts as a relay, connecting to terrestrial networks via a satellite
38   seamless wide-area coverage, in which unmanned aerial vehi- link, and serves user terminals via a ground link.
39   cle (UAV)-satellite communication1 is a key part for building On the other side, spectrum crowding and increased data
40   the integrated space-air-ground network [1]–[5]. As is shown rates encourage the migration to millimeter wave (mmWave)
41   in Fig. 1, UAV communication could be adopted to serve users band, typically Ka-band [6]–[9]. Though the heavy path loss
42   without infrastructure coverage such as disaster areas after of Ka-band seemingly blocks it from practical usage, the
43   earthquake, where the ground communication infrastructure is micro wavelength of Ka-band permits the employment of large
44   amount of antennas onto a small area to achieve adequate
Manuscript received June 28, 2017; revised November 15, 2017; accepted spatial gains and combat such propagation losses [10]–[12].
45   December 11, 2017. This work was supported in part by the National Natural
46   Science Foundation of China under Grant {61422109, 61771274}. This work Meanwhile, multiple-antennas with small size would be specif-
47  
of S. Jin was supported in part by the National Science Foundation (NSFC) for ically suitable for UAV mounted transmission, which also
Distinguished Young Scholars of China with Grant 61625106 and the National removes the requirement of the expensive high-gain directional
48   Natural Science Foundation of China under Grant 61531011. This work of
49   Yi Wu was supported in part by the National Natural Science Foundation of antenna in satellite communication, and thus greatly reduces
50  
China (No. 61571128). (Corresponding author: Feifei Gao.) the overall system cost.
J. Zhao and F. Gao are with Department of Automation, Tsinghua Universi- In general, the performances of the communication systems
51   ty, State Key Lab of Intelligent Technologies and Systems, Tsinghua National
52   Laboratory for Information Science and Technology (TNList) Beijing, P. R. critically rely on the channel state information (CSI), which
53  
China (e-mail: [email protected], [email protected]). J. Zhao brings a large number of CSI estimation works [13]–[20].
is also with High-Tech Institute of Xian, Xian, Shaanxi 710025, China. Different from the conventional wireless communications sys-
54   Q. Wu is with the Department of Electronics and Information Engineering,
tems, UAV based one would encounter new challenges. Due
55   Nanjing University of Aeronautics and Astronautics, Nanjing 210007, China
(email: [email protected]). to the continuous navigation of UAV, the channel would vary
56   S. Jin is with the National Communications Research Laboratory, Southeast constantly for both the UAV-satellite link and the UAV-ground
57   University, Nanjing 210096, P. R. China (email: [email protected]).
link. Furthermore, because of the constraints of the overall
60   Yi Wu is with College of Photonic and Electronic Engineering, Fujian
Normal University, Fuzhou, 350117, China (e-mail: [email protected]). system cost and the power consumption, the available radio
61   W. Jia is with High-Tech Institute of Xian, Xian, Shaanxi 710025, China frequency (RF) chains are limited for UAV-satellite communi-
62   (e-mail: [email protected]).
1 Geosynchronous satellite is usually used for UAV-satellite communication cation systems. A beam sweeping method was adopted in [18]
63  
since its location relative to earth keeps invariant. for beam tracking, which spends significant resources to train
64  
65  

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
1
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

the beam in all directions. A priori knowledge aided channel

tracking method was proposed in [19] for Teraherz massive
MIMO system, where a linear motion is adopted to derive
angles of the incident signals and then the complex gains are 1

estimated by pilots. However, the linear movement model re- 0.8

stricts its application for practical nonlinear movement scenes.
0.6

Amplitude
An angle division multiple access (ADMA) based channel
tracking scheme was proposed in [20] for massive MIMO 0.4
systems, where channel tracking is transmitted to estimating
0.2
the direction of arrival (DOA) information of the incoming
signals and the time-varying complex gain information, which 0
1.5
greatly reduces the complexity of channel tracking. 1.5
1
Albeit most methods [13]–[20] can be extended to the UAV- 1
0.5
ground link, the channel tracking for UAV-satellite link still 0.5
needs to be investigated due to its distinct properties. For UAV- β(rad) 0 0
α(rad)
satellite link, the premise of establishing a successful link is
the alignment of the spatial beam from UAV to satellite, i.e.,
Fig. 2. The channel characteristics of Ka-band massive array antenna.
the spatial beam should keep “locking” the direction of the
target satellite during UAV navigation [21]–[24], which is a
typical SatCom on-the-move. However, the attitude variation denoted by AT , A∗ , AH , A−1 and A† , respectively.
of UAV (including yaw angle, pitch angle and roll angle)
would constantly affect the beam pointing. Therefore, UAV II. P ROBLEM F ORMULATION
needs to overcome the effects of its navigation, and persis-
tently adjusts the spatial beam towards the target satellite. An A. Channel Model
electrical tracking method for phased-array antenna system Let us consider the UAV-satellite communication system,
was developed in [24], which adopts the electrical feedback to where UAV is equipped with UPA of M × N antennas with
track the beam pointing variation. However, for UAV-satellite M  1, N  1, while a single antenna is installed on the
link, it is always preferred that the steered spatial beam should geosynchronous satellite [25], [26]. Due to the constraints
be pointed towards the normal of the array antenna plane of system cost and power consumption, the available radio
since the array antenna gain is proportional to the projected frequency (RF) chains are limited for UAV-satellite commu-
area of the antenna aperture in the direction of the target nication system. In this paper, it is assumed that UAV has
satellite [21], which demands for the mechanical adjustment only one dedicated RF chain, i.e., single channel receiver
as did in conventional SatCom on-the-move [23]. mode [27], [28].
In this paper, we propose a blind hybrid beam tracking Define A(α, β) ∈ CM ×N as the array response matrix,
method for the UAV-satellite communication with a massive whose (m, n)-th element can be expressed as
uniform plane array (UPA). The mechanical adjustment is 2πd
λc [(m−1) sin α cos β+(n−1) sin α sin β]
A(α, β) = ej , (1)
firstly adopted to alleviate the effects of UAV navigations,
where the spatial beam can be pointed towards the target where α and β are the azimuth angle and elevation angle from
satellite through the beam stabilization algorithm and dynamic the target satellite.
isolation algorithm. However, due to the limited precision of For UAV-satellite link, there are limited scatters in the
low cost attitude sensors, the system errors, and measurement propagation surroundings and thus the number of the incident
errors, the mechanical adjustment could only roughly point the paths would be very small. Propagation measurement says that
spatial beam of UAV to the target satellite. We then design an the overwhelming majority of the total energy arrives at the
electrical adjustment to further calibrate beam pointing, where receiver from the direct path in satellite communication [21].
the spatial beam is steered to DOA of the incident signals in Meanwhile, fixed satellite communication systems above 10
a more accurate manner with an array structure based simul- GHz have been verified to operate under line of sight (LOS)
taneous perturbation algorithm. Finally, simulation results are only [22]. Therefore, LOS ray is the dominant one while the
provided to verify the effectiveness of the proposed method other Non-LOS (NLOS) components are usually ignored for
and demonstrate its superiority over the existing candidates. the Ka-band UAV-satellite communication link. Therefore, the
The rest of the paper is organized as follows. The system M ×N channel matrix H between the target satellite and UAV
model is described in section II. Beam tracking with me- can be expressed as [17], [22]
chanical adjustment is presented in section III, while beam
ae−j2πd/λc
tracking with electrical adjustment is presented in section √ H=
A(α, β), (2)
IV. The simulations are then displayed in section V. Finally, MN
conclusions are drawn in section VI. where d is the distance between antenna 1 of UAV and
Notations: Vectors and matrices are denoted by boldface the satellite along the LOS path, and λc is the carrier
small and capital letters; the transpose, complex conjugate, wavelength. Moreover, a is the complex channel attenuation
Hermitian, inverse, and pseudo-inverse of the matrix A are due to gaseous, cloud, rain, and scintillation attenuation. A

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
2
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

7DUJHWVDWHOOLWH Z
G
ɔ

= 83$QRUPDO T

3KDVH6KLIWHU
Ă
%HDPSRLQWLQJ
&RPELQLQJ
HUURU
%HDPIRUPLQJ

$&8 6HQVRUV
<
8$9 &RQWURO0RWRU
;

Fig. 3. The error analysis of UAV beam pointing, where the beam direction Fig. 4. The structure of the UAV communication system, where a mechanical
should be stabilized at the vicinity of the plane normal for high quality of adjustment subsystem is installed at BS and each antenna is equipped with a
communication. analog shifters instead of a dedicated RF chain.

combination method considering the above effects can be seen to overcome the effects of UAV navigation for UAV-satellite
in [22]. communication systems.
Define the spatial domain channel matrix as the 2-dimension To remedy the effects of UAV navigation, we propose a
discrete Fourier transform (2D-DFT) of H, i.e., new beam tracking method based on the joint mechanical
adjustment and electrical adjustment. The structure of the
H̃ = FM HFN , (3) designed beam tracking system is depicted in Fig. 4, which
consists of three segments: (1) the perception part contains
where FM and FN are the normalized DFT matrices with
the measurement sensors and is responsible to sense the UAV
dimension M × M and N × N . According to the analysis
attitude variations as well as to measure the corresponding
in [17], the significant values of the spatial channel H̃ are
angular rate; (2) the decision part, i.e., the arithmetic and
centered around DOA α and β of the impinging signal, as
control unit (ACU), provides control commands to stabilize
shown in Fig. 2. Therefore, the massive number of antennas
beam pointing according to the outputs of the perception
in mmWave band can be packed into a small area and form
part; (3) the operative part contains the control motors (the
narrow beam with high spatial gain, which makes it very
azimuth angle control monitor, the polarization angle control
suitable for UAV-satellite communication.
monitor, and the elevation angle control monitor) and phase
shifters, etc., which implements the directives of the decision
B. Problem Formulation part.
Traditionally, the high precision inertial navigation sys-
For UAV-satellite link, the beam pointing error would tem [23] is used to obtain the attitude information in UAV
severely degrade the communication quality, which should be military applications. However, the cost is not affordable
kept as small as possible. In addition, it has been verified for the civil applications. Recently, with the development of
that the optimal working range of UPA is in the vicinity of its the micro electro mechanical system (MEMS), there appear
normal, and the array gain is proportional to the projected area many low-cost attitude sensors, such as MEMS gyros, MEMS
of the antenna aperture in the direction of the target satellite, accelerator, global position system (GPS). Therefore, we pro-
which declines by sine of the elevation angle [21]. For example pose to utilize these low-cost sensors to derive the attitude
in Fig. 3, the spatial beam would be distorted if it deviates far information and reduce the total system cost. The attitude
from the direction of the plane normal. determination principles of these low-cost sensors can be
In general, there are two main factors affecting beam found in Appendix A.
pointing: satellite perturbation and UAV navigation. In order
to overcome the influence of the satellite perturbation, the
position-keeping technique is usually adopted so that the III. C OARSE B EAM A LIGNMENT WITH M ECHANICAL
latitude and longitude errors of the satellite position are kept A DJUSTMENT
within the allowable range. A good example of position- Due to small wavelength of Ka-band, the size of massive
keeping technique is [29], where satellite perturbation has array antenna is not large for small UAVs with limited load,
little effect on beam pointing within one hour and can be which makes the mechanical adjustment based beam tracking
neglected. On the other side, the UAV navigation, including applicable [23]. In general, the mechanical adjustment for
the carrier’s heading, pitching and rolling motion, are very beam tracking consists of two steps: the beam stabilization
detrimental. Unfortunately, there is no corresponding method and the dynamic isolation, which is shown in Fig. 5.

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
3
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

8$9
QDYLJDWLRQ
&RRUGLQDWH $WWLWXGH WIUDPH
n
WUDQVIRUPDWLRQ
]3

3
g3

3
$WWLWXGH
'\QDPLF VHQVRUV
0HFKDQLFDO LVRODWLRQ
DGMXVWPHQW 3 3 

QIUDPH EIUDPH
%HDP  %HDP
VWDELOL]DWLRQ SRLQWLQJ
Fig. 7. The relationships between different frames.

Fig. 5. Mechanical adjustment with beam stabilization and dynamic isolation.

[ Besides, its axis yf is parallel to ya , while its xf axis

\ points to the satellite. The axis zf is perpendicular to
7DUJHW ] the plane spanned by the axis xf and yf . On the basis
VDWHOOLWH of the relationship of f -frame and a-frame, the a-frame
WIUDPH would be consistent with the f -frame when we rotate the
elevation angle β around the axis ya .
5) The spatial beam coordinate frame (t-frame): Its origin
is also AACG, and axis yt is identical with the direction
EIUDPH of electric field, while xt is parallel to xf . The axis zt
R is perpendicular to the plane spanned by the axis xt and
QIUDPH yt . Based on the relationship between the f -frame and t-
8$9 frame, the f -frame would be consistent with the t-frame
when we rotate the polarization angle γ around the axis
xt .
Fig. 6. The definition of different reference frames, including n-frame, b-
frame, a-frame, f -frame and t-frame. The UAV attitude relative to the n-frame is determined by
the yaw angle Ψ, the pitch angle θ, and the roll angle φ, while
the beam attitude of UAV array antenna relative to the b-frame
A. Beam Stabilization is dependent on the azimuth angle α, the elevation angle β,
and the polarization angle γ. Moreover, the beam attitude of
The proposed beam stabilization jointly utilizes the UAV
UAV array antenna relative to the n-frame can be expressed
attitude information and satellite location information to track
by the Euler angles o, e, and v. The relationships between
the spatial beam. To represent UAV attitude information, we
different frames are shown in Fig. 7, and the transformation
need to explain the reference frames first, which describe
between different frames can be realized by the coordinate
the relationships between UAV attitude and beam pointing.
transformation matrix [23].
Different reference frames are also shown in Fig. 6.
1) The geodetic coordinate frame (n-frame): Its origin is Denote Ctb as the coordinates transformation matrix be-
chosen as the UAV center of gravity (UAVCG), and its tween the b-frame and the t-frame, which can be realized
three axes xn , yn , zn are separately aligned with the by three continuous spatial rotation: ⎡rotating α around the ⎤
direction of north, east, and geocentric. cos α sin α 0
2) The UAV-fixed coordinate frame (b-frame): Its origin is axis zb that is represented by T1 (α)=⎣ − sin α cos α 0 ⎦,
UAVCG, and its three axes xb , yb , zb , i.e., roll axis, pitch 0 0 1
axis, and yaw axis, are aligned with its longitudinal (for- rotating β around the
⎡ ⎤ axis ya that is represented by T2 (β)=
cos β 0 − sin β
ward), lateral (right), and vertical (downward) direction ⎣ 0 1 0 ⎦, and rotating γ around the axis xt that
respectively.
sin β 0 cos β ⎡ ⎤
3) The azimuth-rotary coordinate frame (a-frame): Its origin
1 0 0
is the array antenna center of gravity (AACG). Its axis
is represented by T3 (γ) = ⎣ 0 cos γ sin γ ⎦. Each of
za is parallel to zb , while axis ya is identical with
0 − sin γ cos γ
the direction of UPA elevation axis. The axis xa is
the three rotations are illustrated in Fig. 8. Thus, Ctb can be
perpendicular to the plane spanned by the axis ya and za .
mathematically expressed as (4), shown on the top of the next
According to the relationship of b-frame and a-frame, the
page.
b-frame would be consistent with the a-frame when we
rotate the azimuth angle α around the axis zb . Then, according to the position relationships of UAV and
4) The UPA coordinate frame (f -frame): Its origin is AACG. satellite, we can obtain the Euler angles o, e, v of the spatial

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
4
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

⎡ ⎤
cos α cos β sin α cos β − sin β
Ctb = T3 (γ) T2 (β) T1 (α) = ⎣ cos α sin β sin γ − sin α cos γ sin α sin β sin γ + cos α cos γ cos β sin γ ⎦ . (4)
sin α sin γ + cos α sin β cos γ sin α sin β cos γ − cos α sin γ cos β cos γ

ωut =Ctb ωub = T3 (γ) T2 (β) T1 (α) ωub

⎡ ⎤⎡ ⎤⎡ ⎤⎡ ⎤
1 0 0 cos β 0 − sin β cos α sin α 0 ωubx
= ⎣ 0 cos γ sin γ ⎦ ⎣ 0 1 0 ⎦ ⎣ − sin α cos α 0 ⎦ ⎣ ωuby ⎦ . (10)
0 − sin γ cos γ sin β 0 cos β 0 0 1 ωubz

]E ]D ]D ]I
]I E ]W J

\I
[E 27
2E 27 \D 27 E [I
D D \D J
[D [D [ I [W \ I \W
\E

 Zx^jYe] Yx^jYe]  Yx^jYe] ^x^jYe]  ^x^jYe] lx^jYe]

Fig. 8. The transformation from the b-frame to t-frame.

beam relative to the n-frame as where Tm,n is the (m, n)-th element in Ctb .
⎧
⎪
⎪ o = 180◦ + tan−1 tan(λ−λ s)
,
⎪
⎪ sin ϑ
⎨ RE
cos ϕ cos(λ−λs )− R +h
e = tan−1 √ E E
, (5) B. Dynamic Isolation
⎪
⎪ 1−[cos ϕ cos(λ−λs )]2
⎪
⎪ When UAV navigates, its attitude will change and would
⎩ v = tan−1 sin(λ−λs ) ,
tan ϑ affect beam pointing. The corresponding angular rate in the
where ϑ is the UAV latitude, λ is UAV longitude, λs is b-frame, denoted as ωub , is coupled with the direction of
longitude of target satellite, RE and hE are the earth radius the spatial beam through geometric constraints and friction
and the UAV altitude. constraints, and could be measured by gyros. Let us further
Similarly, we can derive the coordinates transformation denote the coupled angular rate of the spatial beam due to
matrix between the n-frame and the t-frame Ctn as UAV navigation in the t-frame as ωut . Then, it can be derived
as (10), shown on the top of the page. Moreover, in (10), ωubx ,
Ctn = T3 (v) T2 (e) T1 (o) . (6)
ωuby and ωubz are the elements of ωub .
Furthermore, when the UAV attitude information (ψ, θ, φ) Next, we propose the dynamic isolation algorithm to com-
is obtained (the derivation is presented in Appendix B), the pensate ωut as well as release the effects of UAV navigation,
coordinates transformation matrix between the n-frame and and the detailed scheme is shown in Fig. 9. Specially, the
the b-frame Cbn can be expressed as control monitors of the azimuth angle, the polarization angle,
Cbn = T3 (φ) T2 (θ) T1 (ψ) . (7) and the elevation angle are jointly utilized to make the spatial
beam point to the target satellite.
As shown in Fig. 7, the coordinates transformation from the Denote ωαa , ωβf and ωγt as the control monitor outputs
n-frame to the t-frame can be decomposed into the transfor- of the azimuth angle, the polarization angle, and the elevation
mation from n-frame to the b-frame and the transformation angle respectively. According to the inter-relations of the a-
from the b-frame to the t-frame, namely, frame, the f -frame, and the t-frame, the azimuth angular rate
Ctn = Cbn Ctb , or Ctb = (Cbn )−1 Ctn . (8) in the t-frame, denoted as ωαt , can be obtained from azimuth
angle control monitor as
From (8), the final angle in the b-frame for aligning the
spatial beam of UAV to the target satellite can be derived as ωαt = T3 (γ) T2 (β) ωαa
⎧ ⎡ ⎤⎡ ⎤⎡ ⎤
⎨ α = arctan T12
T
11
, 1 0 0 cos β 0 − sin β 0
β = − arcsin T13 , (9) =⎣ 0 cos γ sin γ ⎦⎣ 0 1 0 ⎦⎣ 0 ⎦ , (11)
⎩
γ = arctan TT23
33
, 0 − sin γ cos γ sin β 0 cos β ωα

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
5
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

Zx^jYe] Yx^jYe] ^x^jYe] lx^jYe]

mZ qY ml 5 L+ !L* !L) !mZ
rZ L) !mZ L* !L) !mZ
pl

†
lx^jYe]
Yx^jYe] ^x^jYe] lx^jYe] ml # el 5 (
 Y qY L* ! Y
pl L+ !L* ! Y
  

^x^jYe]
 ^ pl
lx^jYe]
L+ ! ^
† lx^jYe]
el 5  l #  l #  l

lx^jYe]
 l

Fig. 9. The procedure of dynamic isolation.

while the elevation angular rate in the t-frame, denoted as ωβt , A. Electrical Adjustment with Simultaneous Perturbation
can be obtained from the elevation control monitor as As is mentioned earlier, we consider the hybrid scheme to
⎡ ⎤⎡ ⎤
1 0 0 0 decrease the system cost and power consumption. Therefore,
ωβt = T3 (γ) ωβf = ⎣ 0 cos γ sin γ ⎦ ⎣ωβ ⎦ . (12) the received signal at UAV can be expressed as
0 − sin γ cos γ 0
y = wH hs + wH n, (16)
Moreover, the polarization angular rate in the t-frame can be
T
directly obtained by ωγt = [ωγ 0 0] . where h = vec(H), w is the M N × 1 receive beamforming
Then, the total beam pointing angular rates from the control vector with |wi | = 1, s is the transmitted signal of the satellite,
monitors in the t-frame, denoted as ωmt , are the sum of ωαt , and n is the M N × 1 noise vector.
ωjtβ , and ωγt , i.e., Then, the instant power of the received signal at UAV can
be expressed as
ωmt = ωγt + T3 (γ) ωβf + T3 (γ) T2 (β) ωαa . (13)
P = |y|2 = wH Rs w + wH Rn w, (17)
Take the t-frame as the reference, the eventual physical
angular rate of beam pointing, denoted as ω, is the result of where Rs = hsshH and Rn = nnH are the temporal
a combination of the UAV’s own navigation and the control covariance matrices of the signal and the noise respectively.
monitor’s outputs, namely, Let [α, β] be the relative location of the target satellite after
mechanical adjustment. Since mechanical adjustment could
ω = ωmt + ωut . (14)
approximately stabilize the beam direction near the normal
To overcome the effect of UAV attitude variation, it is re- of UPA, the value of α and β would be small. To achieve
quired that ω = 0 such that the spatial beam is always pointing a superior quality of service in UAV-satellite communication,
to the target satellite. After some tedious computations, the the power of the received signal at UAV should be made as
angular rates of the control monitors can be obtained from (14) high as possible to guarantee the beam pointing in a precise
as direction [24].
⎧
⎨ ωα=− cos α · tan β · ωubx −sin α · tan β · ωuby −ωubz , It has been demonstrated in [17] that P can be maximized
ωβ = sin α · ωubx − cos α · ωuby , by aligning the beam direction with the desired signal source
⎩ direction, i.e., w = vec (A(α, β)). In this case, the phase of
ωγ = (− cos α · ωubx − sin α · ωuby ) sec β.
(15) the (m, n)-th element of A(α, β), denoted as p(m,n) can be
expressed as
Remark 1: The mechanical adjustment is performed ac-
cording to the attitude information rather than the training 2πd sin α
p(m,n) = [(m − 1) cos β + (n − 1) sin β]
symbols, which is a blind way for beam tracking. While the λc
beam stabilization is utilized to derive the prior directions of 2πd sin α 
= (m − 1)2 + (n − 1)2
the spatial beam, the dynamic isolation is adopted to release λ
 c 
the effects of UAV navigation. m−1
. sin β + arcsin  . (18)
IV. F INE B EAM A LIGNMENT WITH E LECTRICAL (m − 1)2 + (n − 1)2
A DJUSTMENT From (18), the DOA information is required for receive
Due to the system error and the measurement error, purely beamforming. However, the single RF chain at UAV restricts
counting on mechanical adjustment is not sufficient to yield the application of many advanced DOA estimation algorithms,
an accurate beam pointing. To further improve the precision such as MUSIC, ESPRIT, and Capon [32]–[35]. For single RF
of beam pointing, we propose an electrical adjustment method chain receiver, a zero-knowledge beamforming method was
for UAV communications with large antenna array. suggested in [36], but this method is time-consuming due to

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
6
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

sequential perturbation of the phase shifters. Besides, the con- Algorithm 1 : Electrical adjustment
vergence speed would become much slower with the increase • Step 1: Initialization: initialize the spatial beam direction
of the massive array antennas. A simultaneous perturbation of the array antenna as w = 1M N ×1 , and obtain the UAV
method was proposed in [37] to accelerate the convergence latitude ϑ, the UAV longitude λ and the satellite longitude
speed. However, the method is fully isotropic and still requires λs ;
many iterations for convergence. Here, we propose to utilize • Step 2: Beam coarse alignment: adopt beam stabilization
the array structure to further accelerate the converge speed for (9) and dynamic isolation (12) to realize the rough
UAV with massive array antennas. alignment of the spatial beam;
Note from (18) that the values of the phase shifters for beam • Step 3: Beam fine alignment: perform the array structure
alignment are related to the antenna locations (m, n). Hence, aided DOA tracking. Set k = 0, and pick parameters
we could utilize the array antenna structure to accelerate the a, b, c, ζ, Ω, ξ according to the guide lines [38];
convergence speed. Specially, the instant power of the received • Step 4: Simultaneous perturbation: generate the pertur-
signals are utilized for the electrical adjustment instead of the bation scalar ξk and Δk , and derive two instant power
training symbols, which is also a blind way. measurements of the UAV received signals P (w+ ) and
Denote ĝk (wk ) as the estimation of the gradien- P (w− ), where w+ = wk + b/(k + 1)Ω Dξk + c/(k +
t ∂P (wk )/∂wk , which can be derived from two noisy mea- 1)Ω Δk and w− = wk −b/(k +1)Ω Dξk −c/(k +1)Ω Δk ;
surements of the received signal power as (19), shown on the • Step 5: The beam pointing updating: estimate the gra-
top of the next page. dient ĝt (wk ) using (19) and update the beam direction
Moreover, in (19), ξk and the elements of Δk are ran- wk+1 by (21);
dom perturbation parameter at the k-th iteration following a • Step 6: Decision: terminate algorithm when the power
Bernoulli ±1 distribution with probability 1/2. Meanwhile, b, of the received signal increases little, or else return to
c as well as Ω are all the small perturbation parameters, and Step 3.
D can be expressed as
⎡ ⎤
d11 · · · d1N
⎢ .. ⎥ , minimum size of the training sequence is 1 for both uplink
D = ⎣ ... ..
. . ⎦ (20) and downlink channel.
dM 1 · · · dM N Take the channel from the satellite to the UAV as exam-
 ple, the received signal at UAV side, defined as Y, can be
where dmn = (m − 1)2 + (n − 1)2 . expressed as
Then, the received beamforming vector can be adjusted 
according to the following formula P
Y= Hsp + B, (22)
σp2
wk+1 = wk − ηk ĝk (wk ), (21)
where P is the training power constraint, sp is the training
where ηk = a/(ζ + k)ξ is the step size for the adjustment at
symbol with |sp |2 = σp2 , and the elements of B ∈ M × N are
the k-th iteration, and ξ, ζ are step perturbation parameters. 2
distributed as i.i.d. CN (0, σB ).
The perturbation parameters in (19) and (21) can be obtained −j2πd/λc

by the guide lines in [38]. Then, complex gain can be derived as ae √M N =


1 −1
P Y(A(αv , βv ))
The detailed steps of the proposed method are shown in . Note that the gain information can be
Algorithm 1. Since the proposed electrical adjustment utilizes utilized for transmit beamforming, i.e., when the channel gain
the array structure, it can be named as array structure based is small, UAV should allocate much more power for transmis-
simultaneous perturbation (ASSP). sion. Therefore, the performance of the UAV communication
Remark 2: Since the value of α and β are small, the time system would be greatly enhanced.
for convergence would not be large. Besides, by utilizing Besides, it is worth to mention that the proposed method
the array structure, the convergence speed of the proposed would have a much lower training overhead than the traditional
algorithm is not relevant to the number of the phase shifter or beam cycling method [18], since the spatial beam can be
equivalently the number of the array antennas, and would be realtimely pointed to the target satellite without extra training.
much faster than the traditional techniques in [36], [37].There-
fore, the proposed method has a much lower complexity than
V. S IMULATIONS AND R ESULTS
the existing methods.
In this section, numerical results are provided to demon-
strate the effectiveness of the proposed method. We set M =
B. Complex Gains Estimation 128, N = 64 and d = λ2 , and the channel vector between
Through the joint mechanical adjustment and electrical UAV and target satellite can be produced according to (2).
adjustment, the spatial beam can be pointed to the target The experimental platform, which consists of the outdoor
satellite without training, i.e., a blind way. With the tracked part and the indoor part, is installed at the three-axis flight
beam, we can further derive the gain information by sending simulation turntable to imitate the navigation of UAV. The
training symbols towards the target satellite as did in [17]. outdoor part are composed of XW-ADU3601, XW-IMU5220
Since there is only one dominant beam for transmission, the and XW-ADU7612 [39]–[41]. Specifically, XW-ADU3601 is

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
7
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

bDξk bDξk
P wk + (k+1)Ω
+ cΔk
(k+1)Ω
− P wk − (k+1)Ω
− cΔk
(k+1)Ω
ĝt (wk ) =   . (19)
bDξk
2 (k+1) Ω + (k+1)Ω
cΔk

181
1.5
180 α
β
ψ/°

179 1
Reference Data fusion
178
0 10 20 30 40 50
0 0.5

Angle error/°
−1
θ/°

0
−2

−3
0 10 20 30 40 50 −0.5
1.5

1
−1
φ/°

0.5

−1.5
0 0 10 20 30 40 50
0 10 20 30 40 50
t/s t/s

Fig. 10. The performance of attitude determination. Fig. 12. The errors of the beam pointing from mechanical adjustment.

0.5
Fig. 10 shows the attitude from Kalman filter based data
fusion, where the attitude from the XW-ADU7612 is also dis-
δψ/°

0
played as comparison, while Fig. 11 shows the corresponding
−0.5 attitude error. It can be seen from Fig. 10 that the regulation
0 10 20 30 40 50
0.5 of attitude variation from data fusion coincides with the
practical attitude value, which verifies that the data fusion can
δθ/°

0 effectively integrate the measurements from different sensors

and provide relatively precise attitude information. However,
−0.5 we can see clearly from Fig. 11 that there are still attitude
0 10 20 30 40 50
0.5
errors from data fusion, and the maximum error of the attitude
reaches 0.5◦ , which is consistent with the result in (43)-(45).
δφ/°

0
With the tracked attitude, we then derive the beam pointing
−0.5 through mechanical adjustment, and the beam pointing errors
0 10 20 30 40 50
t/s of the azimuth angle α and the elevation angle β are shown in
Fig. 12. It can be seen that the pointing errors of both α and β
Fig. 11. The error of the estimated attitude. are within 0.5◦ in most time, and mechanical adjustment can
only approximately make the beam point to the target satellite.
The reason is that the attitude error from data fusion cannot
a two-antenna GPS outputting the measurement of ψm ; XW- be neglected, which would be coupled into beam pointing and
IMU5220 is an MEMS inertial measurement unit that can make the beam deviate from the target satellite.
output the attitude angular rates ωub and the measurements Next, we show the performance of electrical adjustment
of θm and φm ; XW-ADU7612 is an attitude and heading in Fig. 13 and Fig. 14 with SNR=10 dB and SNR=20
reference system, which can provide the attitude to an accuracy dB respectively, and the method in [36] and [37] are also
of 0.1◦ as comparison. The DSP processor runs in the indoor displayed as comparison, where SNR is the receive signal
parts with the frequency 400 MHz, which is ACU of the noise radio after beamforming. According to the guide lines
platform. Note that the platform can provide measurements in [38], the step parameters are set as a = 0.7, b = 0.02, c =
of the attitude information, and it can be used to verify the 0.01, ζ = 0.1, τ = 0.1, ξ = 0.602. The performance metric
validity of the mechanical adjustment. The target satellite is is the normalized received signal power (NRSP) which is
set as AsiaSat-3S [42] with longitude 105.5◦ , while the testing defined as the ratio between the instantaneous received power
location is Xi’an shaanxi with latitude 34.27◦ and longitude and the maximum received power. The prior NRSP is 0.952,
108.95◦ . According to (5), the Euler angles of the beam which further verifies that the mechanical adjustment could
alignment is a = 6.11◦ , e = 50.1◦ , and v = −5.05◦ . approximately point the spatial beam to the target satellite. It

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
8
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

1 0.5
α
0.995 0.45 β

0.99 0.4

0.985 0.35

Angle error/deg
0.98 0.3
NRSP

0.975 0.25

0.97 0.2

0.965 0.15
The proposed method
0.96 0.1
Method in [36]
0.955 Method in [37] 0.05

0.95 0
1 2 3 4 5 6 7 8 9 10 11 12 2 4 6 8 10 12
Number of the iterations Number of the iterations

Fig. 13. The performance of the electrical adjustment, where SNR= 20dB. Fig. 15. The angle error of the electrical adjustment, where SNR=10dB.

1 precisely pointed to the target satellite.

0.995
VI. C ONCLUSIONS
0.99
In this paper, we proposed a blind beam tracking method
0.985
for Ka-band UAV-satellite communication system with hybrid
0.98 massive antennas. The proposed method mainly consists of
NRSP

0.975 two procedures: mechanical adjustment and electrical adjust-

ment. The former one, including the beam stabilization and
0.97
dynamic isolation, is used to overcome the effect of UAV
0.965
navigation and realize the coarse alignment, while the latter
0.96
The proposed method
one is used to further improve the tracking precision and
0.955
Method in [36] realize the fine alignment. Moreover, tracking the spatial beam
Method in [37]
can be simplified to tracking the DOA information of the
0.95
1 2 3 4 5 6 7 8
Number of the iterations
9 10 11 12 satellite, which greatly decreases the training overhead and
improves the system efficiency. Various examples are provided
to verify the efficiency of the proposed method.
Fig. 14. The performance of the electrical adjustment, where SNR=10dB.
A PPENDIX A
ATTITUDE D ETERMINATION P RINCIPLES OF T HE
can be seen that all the perturbation methods can converge L OW-C OST S ENSORS
to the maximum received power, i.e., the spatial beam of
(1) MEMS Gyros: The three MEMS gyros (roll gyro, pitch
UAV is pointed to the practical DOA of the incident signals.
gyro and yaw gyro) are installed near the UAVCG in the
The proposed method can converge with much faster speed
three directions of the b-frame, which can measure the attitude
than other two methods, since the array structure is utilized
variation of UAV. Denote the angular rate outputs of the
for beam tracking. Besides, we can find that the number of T
gyros as ωub = [ωubx (τ ) ωuby (τ )ωubz (τ )] , and denote the
iteration is increasing with the decrease of SNR for all the
displayed methods. The number of iteration for convergence variation of the UAV attitudes as φ̇(τ ) θ̇(τ ) ψ̇(τ ) , where τ
is 4 for SNR=20dB, while it increases to 7 at SNR=10 dB is the time index. According to the definitions of the reference
for the proposed method. The convergence of the methods frames, the relationship between ωub and φ̇(τ ) θ̇(τ ) ψ̇(τ )
in [36] and [37] is much worse compared to their higher can be expressed as (23), shown on the top of the next page.
SNR case. This is because for low SNR case, the noise Then, we can obtain a rough measurement of the three-
is dominant and would affect the convergence speed of the dimension attitude angles, denoted as φm (τ ), θm (τ ), and
electrical adjustment. ψm (τ ), from the gyro integration as
In the last example, we show the final angle error of ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
φm (τ ) φm (τ − 1) φ̇(τ )
the proposed simultaneous perturbation with SNR=10 dB in ⎣ θm (τ ) ⎦ = ⎣ θm (τ − 1) ⎦ + ⎣ θ̇(τ ) ⎦ . (24)
Fig. 15. We see that with the increase of the iteration number,
ψm (τ ) ψm (τ − 1) ψ̇(τ )
the angle errors of both α and β become smaller, and the
eventual angle errors stay in the magnitude of 10−2 , which is (2) Accelerator: The accelerator is also installed at UAVCG,
consistent with the result in Fig. 13. Hence, through the joint which can derive the attitude of the UAV by measur-
mechanical and electrical adjustment, the spatial beam can be ing the local gravity acceleration directly. Denote f (τ ) =

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
9
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

10

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
ωubx (τ ) φ̇(τ ) 0 0
⎣ ωuby (τ ) ⎦ = ⎣ 0 ⎦ + T3 [φ(τ − 1)] ⎣ θ̇(τ ) ⎦ + T3 [φ(τ − 1)] T2 [θ(τ − 1)] ⎣ 0 ⎦
ωubz (τ ) 0 0 ψ̇(τ )
⎡ ⎤⎡ ⎤
1 sin φ(τ − 1) tan θ(τ − 1) cos φ(τ − 1) tan θ(τ − 1) ωubx (τ )
=⎣ 0 cos φ(τ − 1) − sin φ(τ − 1) ⎦ ⎣ ωuby (τ ) ⎦ . (23)
sin φ(τ −1) cos φ(τ −1)
0 cos θ(τ −1) cos θ(τ −1)
ωubz (τ )

T
[fx (τ ), fy (τ ), fz (τ )] as the measurement of the accelerator, The quaternion uses four parameters, denoted as q1 , q2 , q3 ,
and the relationship between the output of accelerator and the and q4 , to represent the three dimension attitude of the UAV,
attitude of UAV can be expressed as which is defined as
⎡ ⎤ ⎡ ⎤  T
fx (τ ) − sin θ(τ ) q = q1 ρ , (30)
f (τ ) = ⎣ fy (τ ) ⎦ ≈ −g ⎣ sin φ(τ ) cos θ(τ ) ⎦ , (25)  T
fz (τ ) cos φ(τ ) cos θ(τ ) where ρ = q2 q3 q4 .
With the quaternion, the attitude updating equation can be
where g is the gravitational acceleration. represented by
Then, the measured pitch angle and the roll angle can be Ts T
obtained by q(τ ) = q(τ − 1) + Ξ [q(τ − 1)]ωub (τ ), (31)
2
 
fx (τ ) where Ts is the sampling interval. Moreover, Ξ[q(τ − 1)] can
θm (τ ) = arcsin , (26)
g be derived as (32), shown on the top of the next page.
  After some tedious transformations of (31), we can further
fy (τ )
φm (τ ) = arctan . (27) derive that
fz (τ )
q(τ ) = Γq(τ − 1) = q(τ − 1), (33)
(3) GPS: When two GPS antennas are installed at the front
and back of UAV, they can provide measured yaw angle where ⎛ ⎡ ⎤⎞
by using carrier-phase differential technology. If the baseline −ωubx −ωuby −ωubz
0
⎜ ⎢
Ts ⎢ ωubx 0 ωubz −ωuby ⎥ ⎟
length between the two GPS antennas is d, the coordinate of Γ=⎜ ⎝I4×4 + 2 ⎣ ωuby −ωubz
⎥ ⎟.
⎦ ⎠
baseline can be expressed as [d 0 0] in the b-frame. Meanwhile, 0 ωubx
we can derive the coordinate of baseline in the n-frame as ωubz ωuby −ωubx 0
[x(τ ) y(τ ) z(τ )]. Then, the coordinates of the same baseline Therefore, the system equation can be expressed as
in the b-frame and the n-frame can be transformed by Cbn as q(τ ) = Γq(τ − 1) + χ, (34)
⎡ ⎤ ⎡ ⎤
d x(τ ) where χ is the system noise vector, and each element belongs
⎣ 0 ⎦ = Cbn ⎣ y(τ ) ⎦ . (28) to CN (0, Qχ ).
0 z(τ ) When the quaternion is used for attitude updating, it should
meet the normalized condition:
Then, we can acquire the measured yaw angle from (28) as
  q1 2 + q2 2 + q3 2 + q4 2 = 1. (35)
z(τ )
ψm (τ ) = − arctan . (29)
x(τ ) Then, the normalized quaternion can be expressed as
q
q =  . (36)
A PPENDIX B qT q
L OW C OST ATTITUDE D ETERMINATION WITH DATA Since we can directly obtain coarse values of the attitude
F USION from the low-cost sensors, the measurement of the quaternion
The low cost attitude sensors can measure the attitude can be obtained from
information. However, there would exist large error due to ψm θm φm ψm θm φm
q0m = cos cos cos + sin sin sin , (37)
the imprecision of the low-cost gyros, e.g., the drift error and 2 2 2 2 2 2
sideslip angle. Therefore, we try to utilize the Kalman filter ψm θm φm ψm θm φm
q1m = cos cos sin − sin sin cos , (38)
based data fusion to derive an accurate UAV attitude. 2 2 2 2 2 2
Direction cosine matrix, quaternion and Euler angle are ψm θm φm ψm θm φm
q2m = cos sin cos + sin cos sin , (39)
the most commonly used methods for attitude description, 2 2 2 2 2 2
among which the quaternion method not only has low com- ψm θm φm ψm θm φm
q3m = sin cos cos + cos sin sin . (40)
putation complexity, but can also be realtimely updated with 2 2 2 2 2 2
high accuracy. Hence, we adopt the quaternion for attitude Note that in the above equations, we omit the time index τ
representation. for simplicity.

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
10
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

11

⎡ ⎡ ⎤ ⎤
0 −q3 (τ − 1) q2 (τ − 1)
⎢ q4 (τ − 1)I3×3 + ⎣ q3 (τ − 1) 0 −q1 (τ − 1) ⎦ ⎥
Ξ[q(τ − 1)] = ⎢
⎣
⎥.
⎦ (32)
−q2 (τ − 1) q1 (τ − 1) 0
−ρT

Algorithm 2 : Attitude determination by Kalman filter [3] Z. Xiao, P. Xia, and X. Xia, “Enabling UAV cellular with millimeter-
• Step 1: Initialization: q(0), Qχ , Qu ; wave communication: potentials and approaches,” IEEE Commun. Mag.,
vol. 54, no. 5, pp. 66–73, May 2016.
• Step 2: Prediction: q(τ |τ − 1) = Γq(τ − 1); [4] L. Fan, R. Zhao, F. Gong, N. Yang, and G. K. Karagiannidis, “Secure
• Step 3: Minimum prediction MSE: κ(τ |τ − 1) = Γ(τ − multiple amplify-and-forward relaying over correlated fading channels,”
1)κ(τ − 1)Γ(τ − 1)H + Qχ ; IEEE Trans. Commun., vol. 65, no. 7, pp. 2811-2820, July 2017.
[5] L. Fan, X. Lei, N. Yang, T. Q. Duong, and G. K. Karagiannidis, “Secrecy
• Step 4: Kalman gain matricx: Υ(τ ) = κ(τ |τ − cooperative networks with outdated relay selection over correlated fading
1)[κ(τ |τ − 1) + Qu ]−1 ; channels,” IEEE Trans. Veh. Technol., vol. 66, no. 8, pp. 7599–7603,
• Step 5: Updation and normalization: q̂(τ ) = Aug. 2017.
q(τ |τ−1)+Υ(τ )[z(τ )−q(τ |τ−1)] [6] S. Gong, D. Wei, X. Xue, and M. Y. Chen,“Study on the channel
{q(τ |τ−1)+Υ(τ )[z(τ )−q(τ |τ−1)]}T{q(τ |τ−1)+Υ(τ )[z(τ )−q(τ |τ−1)]}
; model and BER performance of single-polarization satellite-earth MIMO
• Step 6: MMSE: κ(τ ) = [I − k (τ )]κ(τ |τ − 1); communication systems at Ka band,” IEEE Trans. Antennas Propag.,
vol. 62, no. 10, pp. 5282–5297, Oct. 2014.
• Step 7: Go to next time instant τ + 1. [7] Z. Pi and F. Khan, “An introduction to millimeter-Wave mobile broad-
band systems,” IEEE Commun. Mag., vol. 49, no. 6, pp. 101–107, June
2011.
[8] M. L. Bhavsar, R. Sharma, and A. Bhattacharya, “Monolithic Ka to Ku-
Therefore, the measurement equation can be expressed as band all balanced sub-harmonic resistive MHEMT mixer for satellite
transponder,” IEEE Microw. Compon. Lett., vol. 25, no. 5, pp. 316–318,
z(τ ) = [q0m q1m q2m q3m ]T + u, (41) Apr. 2015.
[9] D. Fan, F. Gao, G. Wang, Z. Zhong, and A. Nallanathan, “Angle domain
where u is the measurement noise vector, and each element signal processing aided channel estimation for indoor 60GHz TDD/FDD
meets CN (0, Qωk ). massive MIMO systems,” IEEE J. Select. Areas Commun., vol. 35, no. 9,
According to the system equation (34) and the measurement pp. 1948–1961, Sep. 2017.
[10] H. Xie, F. Gao, and S. Jin, “An overview of low-rank channel estimation
equation (41), the UAV attitude can be tracked by Kalman filter for massive MIMO systems,” IEEE Access, vol. 4, pp. 7313–7321, Nov.
based data fusion [30], [31], which integrates the outputs of 2016.
the low-cost sensors to provide a precise attitude information. [11] F. Rusek, “Scaling up MIMO: opportunities and challenges with very
large arrays,” IEEE Signal Process. Mag., vol. 30, no. 1, pp. 40–60, Jan.
The detailed steps are shown in Algorithm 2. 2013.
For attitude adjustment, we need to further transform the [12] H. Xie, B. Wang, F. Gao, and S. Jin, “A new view of multi-user hybrid
quaternion to the attitude angle. Since Cnb can be derived from massive MIMO: non-orthogonal angle division multiple access,” IEEE
J. Select. Areas Commun., vol. 35, no. 10, pp. 2268–2280, Oct. 2016.
the tracked quaternion as
⎡ ⎡ ⎤ ⎤ [13] H. Xie, B. Wang, F. Gao, and S. Jin, “A full-space spectrum-sharing
0 −q̂3 q̂2 strategy for massive MIMO cognitive radio,” IEEE J. Select. Areas
⎢ q̂4 I3×3 − ⎣ q̂3 0 −q̂1 ⎦ ⎥
Commun., vol. 34, no. 10, pp. 2537–2549, Oct. 2016.
Cnb (q̂) = ΞT (q̂) ⎢
⎣
⎥
⎦
[14] W. Tan, M. Matthaiou, S. Jin, and X. Li, “Spectral efficiency of
−q̂2 q̂1 0 DFT-based processing hybrid architectures in massive MIMO,” IEEE
T
−ρ̂ Commun. Lett., vol. 6, no. 5, pp. 586–589, Oct. 2017.
⎡ ⎤ [15] F. Gao, R. Zhang, and Y.-C. Liang, “Optimal channel estimation and
C11 C12 C13 training design for two-way relay networks,” IEEE Trans. Commun.,
= ⎣ C21 C22 C23 ⎦ , (42) vol. 57, no. 10, pp. 3024–3033, Oct. 2009.
[16] F. Gao, R. Zhang, and Y.-C. Liang, “Channel estimation for OFDM
C31 C32 C33 modulated two-way relay networks,” IEEE Trans. Signal Process., vol.
57, no. 11, pp. 4443–4455, Nov. 2009.
the attitude can be obtained from the definition of Cnb (q̂) as [17] H. Xie, F. Gao, S. Zhang, and S. Jin, “An unified transmission strategy
C12 for TDD/FDD massive MIMO systems with spatial basis expansion
ψ̂(τ ) = arctan = ψ(τ ) + δψ(τ ), (43) model,” IEEE Trans. Veh. Technol., vol. 66, no. 4, pp. 3170–3184,
C11 April 2016.
[18] R. W. Heath, N. G. Prelcic, S. Rangan, W. Roh, and A. Sayeed, “An
θ̂(τ ) = arcsin (−C13 ) = θ(τ ) + δθ(τ ), (44) overview of signal processing techniques for millimeter wave MIMO
C23 systems,” IEEE J. Sel. Topic Signal Process., vol. 10, no. 3, pp. 436–
φ̂(τ ) = arctan = φ(τ ) + δφ(τ ), (45) 453, Apr. 2016.
C33 [19] X. Gao, L. Dai, Y. Zhang, T. Xie, and X. Dai, “Fast channel tracking
where δψ(τ ), δθ(τ ), δφ(τ ) are the attitude errors due to the for Terahertz beamspace massive MIMO systems,” to appear in IEEE
Trans. Veh. Technol., 2016, earlier access available online.
system error and measurement error, while ψ(τ ), θ(τ ), φ(τ ) [20] J. Zhao, F. Gao, W. Jia, S. Zhang, S. Jin, and H. Lin, “Angle domain
are the true values of the UAV attitudes. hybrid precoding and channel tracking for mmWave massive MIMO
systems,” IEEE Trans. Wireless Commun., vol. 16, no. 10, Oct. 2017,
pp. 6868–6880.
R EFERENCES [21] Y. Jiang, S. Foti, A. Sambell, and D. Smith, “A low profile radiating
[1] Y. Zeng, R. Zhang, and T. J. Lim, “Wireless communications with element with nearly hemispheric coverage for satellite communications
unmanned aerial vehicles: opportunities and challenges,” IEEE Commun. on-the-move hybrid array antenna,” in Proc of CSNDSP’10, pp. 123–
Mag., vol. 54, no. 5, pp. 36–42, May 2016. 127, Newcastle, UK., July 2010.
[2] L. Fan, X. Lei, N. Yang, T. Q. Duong, and G. K. Karagiannidis, “Secure
multiple amplify-and-forward relaying with cochannel interference,”
IEEE J. Sel. Topics Signal Process., vol. 10, no. 8, pp. 1494-1505, Dec.
2016.

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
11
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JSAC.2018.2804239, IEEE
Journal on Selected Areas in Communications

12

[22] ITU-R Recommendation P.618-9, “Propagation data and prediction [32] A. Hu, T. Lv, and H. Gao, “An ESPRIT-based approach for 2-D local-
methods required for the design of Earth-space telecommunication ization of incoherently distributed sources in massive MIMO systems,”
systems,” Geneva, Switzerland, 2007. IEEE J. Sel. Topics Signal Process., vol. 8, no. 5, pp. 996–1011,
[23] Z. Wu, M. Yao, H. Ma, W. Jia, and F. Tian, “Low-cost antenna attitude Oct. 2014.
estimation by fusing inertial sensing and two-antenna GPS for vehicle- [33] Y. Xin, “Estimating the DOA and the polarization of a polynomial-
mounted satcom-on-the-move,” IEEE Trans. Veh. Technol., vol. 62, no. phase signal using a single polarized vector-sensor,” IEEE Trans. Signal
3, pp. 1084–1096, Mar. 2013. Process., vol. 60, no. 3, pp. 1270–1282, Mar. 2012.
[24] H. Bolandhemmat, M. Fakharzadeh, P. Mousavi, S. H. Jamali, G. Z. Rafi, [34] Y. Zhou, P. C. Yip, and H. Leung, “Tracking the direction-of-arrival
and S. Safavi-Naeini, “Active stabilization of vehicle-mounted phased- of multiple moving targets by passive arrays: algorithm,” IEEE Trans.
array antennas,” IEEE Trans. Veh. Technol., vol. 58, no. 6, pp. 2638– Signal Process., vol. 47, no. 10, pp. 2655–2666, Oct. 1999.
2650, July 2009. [35] F. Gao and A. B. Gershman, “A generalized ESPRIT approach to
[25] P. Arapoglou, “MIMO over satellite: a review,” IEEE Commun. Surveys direction-of-arrival estimation,” IEEE Signal Process. Lett., vol. 12, no.
& Tutorials,, vol. 13, no. 1, pp. 27–51, First Quarter 2011. 3, pp. 254–257, Mar. 2005.
[26] B. Aaron, “MIMO for mobile satellite digital broadcasting: from theory [36] M. Fakharzadeh, H. Jamali, P. Mousary, and S. Safavinaeini, “Fast
to practice,” IEEE Trans. Veh. Technol., vol. 65, no. 7, pp. 4839–4853, beamforming for mobile satellite receiver phased arrays: theory and
July 2016. experiment,” IEEE Trans. Antennas Propag., vol. 57, no. 6, pp. 1645–
[27] A. Alkhateeb, O. Ayach, G. Leus, and R. Heath, “Channel estimation 1654, June 2009.
and hybrid precoding for millimeter wave cellular systems,” IEEE J. Sel. [37] P. Mousavi, M. Fakharzadeh, S. H. Jamali, K. Narimani, M. Hossu, H.
Areas Signal Process., vol. 8, no. 5, pp. 831–846, Oct. 2014. Bolandhemmat, G. Rafi, and S. Safavi-Naeini, “A low-cost ultra low
[28] F. Khalid and J. Speidel, “Robust hybrid precoding for multiuser MIMO profile phased array system for mobile satellite reception using zero-
wireless communication systems,” IEEE Trans. Wireless Commun., knowledge beamforming algorithm,” IEEE Trans. Antennas Propag.,
vol. 13, no. 6, pp. 3353–3363, Jun. 2014. vol. 56, no. 12, pp. 3667–3679, Dec. 2008.
[29] Y. Liu, Y. Li, and J. Xi, “Design and simulation of satellite formation fly- [38] J. C. Spall, “Implementation of the simultaneous perturbation algorithm
ing position-keeping control method,” in Third International Workshop for stochastic optimization,” IEEE Trans. Aerosp. Electron. Syst., vol.
on Advanced Computational Intelligence, pp. 502–505, Aug. 2010. 34, no. 3, pp. 817–823, July 1998.
[30] C. Luo, S. I. McClean, G. Parr, L. Teacy, and R. De Nardi, “UAV [39] http://www.gkzhan.com/st2851/product 319828.html
position estimation and collision avoidance using the extended Kalman [40] http://www.gkzhan.com/st2851/product 319908.html
filter,” Proc. IEEE, vol. 62, no. 6, pp. 2749–2762, July 2013. [41] http://www.hi1718.com/company/28658/products/2011516152856984.html
[31] S. J. Julier and J. K. Uhlmann, “Unscented filtering and nonlinear [42] http://space.skyrocket.de/doc sdat/asiasat-3.html
estimation,” Proc. IEEE, vol. 92, no. 3, pp. 401–422, Mar. 2004.

0733-8716 (c) 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
12