Preamble Structure and Synchronization for IEEE

802.15.6 Impulse–Radio Ultra–Wideband Physical

Layer
Igor Dotlić∗ and Ryuji Kohno∗
∗ Medical ICT Institute, National Institute for Information and Communications Technology
3-4, Hikarino-oka, Yokosuka, 239-0847, Japan
Email: {dotlic, kohno}@nict.go.jp

Abstract—The paper develops a method of synchronization for arbitrary, while SFD structure is not defined in the standard
synchronization header (SHR) structure described in the draft draft at all. Specific of this IR-UWB standard, compared to
of IEEE 802.15.6 standard for body area network (BAN) and its older IEEE 802.15.4a IR-UWB [5] and most IR-UWB system
Impulse–Radio Ultra–Wideband (IR-UWB) physical layer (PHY)
with differential phase modulation. Uniqueness of this PHY is discussions is that transmitted waveform shape is not known
that transmitted waveform shape is not known to the receiver at to the receiver. Thus, classic approach of synchronization by
the time of synchronization and therefore classic synchronization using correlation with transmitted waveform is not possible
methods based on correlation cannot be employed. Based on and novel synchronization method needs to be developed.
the developed synchronization method, synchronization perfor- The purpose of this paper is to investigate possible structures
mances with the different SHR structures are evaluated and
solutions are suggested for the SHR structure described in the of the preamble and the SFD and recommend solutions for
standard. the standard. This will be done, first, by investigating the
Index Terms—IEEE 802.15.6, Body Area Network (BAN), SHR structure currently in the draft in Sec. II. Then, receiver
Ultra–Wideband (UWB), synchronization. architecture suitable for the PHY specs given in the standard
draft will be described in Sec. III. After that, in Sec. IV we
I. I NTRODUCTION will develop synchronization method that will be employed
Body area network (BAN) is emerging technology of wire- in the rest of the paper. In Sec. V, numerical results of
less communication between sensor nodes located on the synchronization performance with different SHR structures
human body and on-body BAN coordinator and between on- will be shown. We end our discussion with conclusions in
body BAN coordinator and of-body data hub. Possible appli- Sec. VI.
cations of the BAN are numerous and can be divided between
medical and non-medical. For medical applications, that are II. SHR STRUCTURE
of foremost interest in this work, main requirements are high Symbol structure of the IEEE 802.15.6 IR-UWB PHY [2]
reliability of communication, especially considering severe with differential modulation is using a single long waveform
BAN channel characteristics [1], and low power consumption, per information symbol and synchronization chip. This symbol
i.e. long battery life. structure was also employed in our previous work [3], [4],
BAN is currently under standardization effort in IEEE where we discussed different receiver architectures suitable for
802.15.6 [2], group for BAN. Draft of the standard includes such symbol structure. The difference between IEEE 802.15.6
different wireless Physical Layers or PHYs. One of the most IR-UWB PHY information symbols and IEEE 802.15.6 IR-
promising PHYs for highly reliable medical communication is UWB PHY SHR is that IEEE 802.15.6 IR-UWB PHY SHR
Impulse–Radio Ultra–Wideband (IR-UWB) PHY with differ- chip waveform is considerably shorter than informational
ential modulation. According to the standard draft, this PHY symbol waveform. Furthermore, in our previous work [3],
targets high quality of service, i.e. foremostly medical, appli- [4] it was assumed that the receiver has a prior knowledge
cations. The BAN communication devices designed according of the type of the waveform that transmitter will use in a
to specifications of this PHY exhibit good performance in package. Nevertheless, draft of the standard [2] includes 4
multipath channels [3] and good interference resistivity [4]. types of transmitted waveforms: chirp pulse, chaotic pulse,
However, at the time of writing this paper, specifications short–pulse burst with static scrambling and short–pulse burst
of this PHY are not fully complete. The main open issue with dynamic scrambling. What is common both for SHR
is the structure of the synchronization header (SHR). SHR and different payload data rates is that duration of the chip
consists of two parts: preamble and start-of-frame delimiter transmitted waveform, denoted in a baseband complex repre-
(SFD). The purpose of preamble is for a receiver to execute sentation as w(t), is not a function of type of a waveform used.
timing acquisition and other synchronization tasks, while by Therefore, when transmitting a package, transmitter can decide
detecting the SFD, receiver is informed about the end of SHR. to use any of aforementioned types of a transmitted waveforms
Preamble parameters currently in the draft were decided rather in the preamble, excluding short–pulse burst with dynamic
 vkQ (t) sQ
k (t) Mathematical model of the sampling receiver for the infor-
× hLP (t) ADC
mation symbol detection is given in [3], [4]. Here, we will
give a model suitable for synchronization with SHR structure
90◦ sQ
k described in the previous section.
Let us consider a single chip waveform – w(t); when it
Carrier generator Detection
passes through multipath channel with baseband–equivalent
rk (t) impulse response h(t) and low–pass anti–aliasing filter with
sIk impulse response hLP (t) shown in Fig. 1, it will have a form
× hLP (t) ADC u(t) = w(t) ⊗ h(t) ⊗ hLP (t), (4)
vkI (t) sIk (t)
where “⊗” denotes convolution. Thus, complex received base-
Fig. 1. Sampling receiver block diagram. band SHR signal, at the input of the ADC converters in Fig.
1 will have a form
N
X
scrambling. Thus, when starting synchronization, receiver does s(t) = an u (t − nLTw ). (5)
not have knowledge of a type of SHR waveform transmitted, n=0
but only knows its duration – Tw and time between consecutive Since RMS delay spread of the channel models [2], [5] is
waveforms – (L−1)Tw . The transmitted SHR structure – g(t) considerably smaller than time between chips (L−1)Tw , there
is of a form will be negligible inter–chip–interference (ICI).
N
X IV. S YNCHRONIZATION METHOD
g(t) = an w (t − nLTw ), (1)
n=0
A. Preamble synchronization
where an are differentially–encoded chips Let τR be reference time of the receiver and the timing
ambiguity between transmitter and receiver be |τR | < ∆τ . It
an+1 = an bn , for n > 0, (2) is assumed that synchronization does not begin before there is
a preamble signal s(t) present, thus receiver begins sampling
with a0 being of arbitrary phase: a0 = ejφ , φ ∈ [0, 2π}. at instance τ0 = τR + ∆τ . Complex signal sampled in the
N −1
b = [bn ]n=0 is a column–vector of bipolar SHR sequences received will then have a form:
b = [CP A ; CP A ; . . . ; CP A ; CSF D ]. (3) M KS +KC −1
S
| {z } sP R = [s (τ0 + kTs )]k=0 . (6)
MP A Here, Ts is the sampling period of the receiver. Ts is chosen
The structure of b (3) is MP A times repetition of a preamble so that the number of samples per chip KC = LTw /Ts is an
sequence – CP A after which there is one instance of a SFD integer; MS < MP A is the number of sequences that receiver
sequence – CSF D . Due to limited space, we will limit our uses in the synchronization and KS is the number of samples
discussion to the case of lengths of CP A and CSF D being per each synchronization sequence: KS = KC Θ.
the same, i.e. dim {CP A } = dim {CSF D } = Θ, although It is assumed that sampling of the preamble finishes before
generally this does not have to be a case. start of the SFD. Hence ∆τ is bounded by
In the current standard draft Tw is set to be 8 ns; L is 32 in ∆τ < (MP A − MS ) KS Ts = (MP A − MS ) LTw Θ. (7)
order to keep a constant duty–cycle of 3.1% [4] for SFD, PHY
header (PHR) and payload, similarly to the mandatory mode In other words, the maximum ambiguity of time between
of IEEE 802.15.4a standard [5]. CP A is decided to be one of 8 transmitter and receiver is bounded by the difference between
small Kasami set sequences of length 63, divided in 2 sets of 4 duration of the sequences that are transmitted in the preamble
sequences for channels with odd and even indexes respectively and duration of the preamble sequences receiver uses in the
with MP A = 4 repetitions, while CSF D is undefined. synchronization.
Since receiver does not have any knowledge of transmitted
III. R ECEIVER ARCHITECTURES chip waveform type, it cannot use classic matched–filtering or
For a several reasons, quadrature analog correlation (QAC) correlation based methods. Instead of that, received samples
receiver is not suitable for a payload symbol structure that are sample–by–sample differentiated
M KS −1
is described in the standard draft [3]. Furthermore, both QAC S
qP R = [ℜ {s∗P R (k) sP R (k + KC )}]k=0 , (8)
and chirp receivers [3], in order to work need for a transmitted
where superscript “∗” denotes complex conjugation. Since ICI
waveform to be known, however, this is not a case within
is negligible, from (2), (3), (5), (6) and (8) it can be concluded
the current standard specs. The only receiver architecture MS KS −1
that qP R = [qP R (kTs + τ0 )]k=0 , where qP R (t) is given
known to the authors that can work with differential encoding,
by:
independently of the type of the transmitted waveform is the N −1
X
sampling receiver, shown in Fig. 1, with Sample–wise DPSK qP R (t) = b (n) |u(t − nLTw )|2 . (9)
detection (S-DPSK) [4]. n=0
In other words, vector qP R represents samples of MS repeti- Consider detection of SFD at time instance τ̂SHR . Since in
tions of CP A modulated by |u(t)|2 . SFD detection timing of the rectangular window corresponding
To improve SNR and reduce size of vector it in reality to each chip is known, receiver needs only to sample at time
operates with, the receiver is able to sum–up samples of qP R instances located within this window:
which are KS samples apart, i.e. correspond to the same CP A ν−1
siSF D (τ̂SHR ) = [s (τ̂SHR + (iKC + ks ) Ts )]ks =0 , (16)
element and the same sample of |u(t)|2 :
MX
for i = 0, . . . , Θ. Sample vectors of consecutive chips are
S −1
correlated:
zP R = qP R (mKs , . . . , (m + 1) Ks − 1). (10) n
H o
m=0 qSF D (i) = ℜ siSF D (τ̂SHR ) si+1 SF D (τ̂ SHR ) , (17)
In order to estimate timing of the SFD, receiver circularly Θ−1
correlates the vector zP R with the kernel vector y: with qSF D = [qSF D (i)]i=0 , where superscript “H” denotes
Hermitian transpose. In order to determine if at τ̂SHR is
x P R = y ⊙ zP R , (11) indeed located CSF D (hypothesis H1 ) or it is one of possible
MP A − MS − 1 CP A sequences remaining after preamble
where “⊙” denotes circular correlation. Kernel vector y is synchronization (hypothesis H0 ), receiver performs likelihood
constructed as the sequence CP A modulated by a rectangular test
KS −1
window of ν < KC samples width; y = [y(k)]k=0 , with H 1
>
( CTSF D qSF D < CTP A qSF D . (18)
CP A (k ÷ KC ) , (k mod KC ) < ν,
y(k) = (12) H0

0, (k mod KC ) ≥ ν. If hypothesis H1 is chosen, receiver concludes that SFD is

If maximum of vector xP R is above threshold value – T H detected and continues with PHR and payload detection. If
calculated with regard to predefined false–alarm rate, hypothesis H0 is chosen and kτ < MP A − MS − 1, receiver
increments kτ and repeats (15), (16), (17) and (18). Otherwise,
max {xP R } > T H, (13) receiver assumes there is no SFD present and stops.
receiver concludes that there is indeed packet preamble present C. Timing estimation loss
and continues with SFD detection. Otherwise, receiver con- Optimum value of τSHR is one that, besides being at the
cludes that there is no packet preamble present and stops. beginning of CSF D , maximizes the energy of siSF D (τSHR )
If (13) is satisfied, timing of the first start of the CP A within vector:
qP A is estimated as n 2 o
opt
τSHR = arg max siSF D (τSHR ) . (19)
τ̂P A = τ0 + Ts arg max {xP R (k)} . (14) opt
Any value of τSHR other than τSHR will produce a perfor-
More precisely, τ̂P A represents estimate of a first time instance mance loss:
within samples of qP A at which preamble sequence can be i opt
2
SF D τSHR
s
S-DPSK detected with maximum energy using rectangular ητ (τSHR ) = 2 . (20)
si
window of ν samples per chip. Possible MP A − MS locations SF D (τSHR )

of the SFD are estimated as Other performance losses of the sampling receiver not directly
τ̂SHR (kτ ) = τ̂P A + (MS + kτ ) LTw Θ, (15) related to the symbol timing, i.e. synchronization method, were
described in [3], [4].
with kτ = 0, . . . , MP A − MS − 1.
V. N UMERICAL ANALYSIS
B. S-DPSK SFD detection A. Numerical simulation parameters
As τ̂P A is estimated time instance at which CP A should In all cases considered, Nyquist sampling period of Ts =
be detected with maximum energy with window of ν samples 2 ns is used. Employed window size is ν = 16 samples,
per chip (12), τ̂SHR (kτ ) are timing instances in which CSF D which for Tw = 8 ns gives a good windowing loss [3], [4]
should be S-DPSK detected with window of of ν samples per of less than 1.5 dB for all [2], [5] channel models. Packet
chip. payload is BCH encoded and transmitted with parameters of
In [4] we have shown that, on one hand, S-DPSK detection IEEE 802.15.6 IR-UWB mandatory mode of 490 kbps DBPSK
performance deteriorates with the increase of the number of [2], with Tw = 64 ns and detected with ν = 32 samples S-
samples per detected symbol if energy in the symbol samples DPSK detection [4]. Size of the packet payload is random
is fixed. On the other hand, energy of the symbol samples and between 0 and 256 octets. Although both performance of
used for detection is increased with an increase of the number the proposed synchronization method and S-DPSK detection
of samples per symbol. Thus, in S-DPSK detection, there proved to be relatively independent of a type of transmitted
should be a careful consideration of the number of samples waveform employed, it is worth mentioning that short pulse
per symbol, or in the case discussed here number of samples burst was used as the transmitted waveform. All simulations
per chip ν. are done on IEEE 802.15.6 CM3 [2] channel model.
 TABLE I
P ROPERTIES OF S YNCHRONIZATION SEQUENCE SETS CONSIDERED . of the preamble sequence for the SFD i.e. CSF D = −CP A ;
second is to use different sequence from a same set for CSF D
Set name Length Size No./Channel Max x-corr. (dB) as for CP A , which will be denoted as |CSF D | 6= |CP A |.
Small Kasami 63 8 4 -9.54 First approach maximizes the distance between hypothesis
Large Kasami 63 72 36 -6.2 H0 and H1 in (18) and thus is optimal under assumption
Gold 31 33 16 -6.46
dim {CP A } = dim {CSF D }. This can be also concluded
from Fig. 3, since in both cases of large Kasami set and Gold
set SFD detection performs significantly better in the case of
B. Synchronization sequence sets considered CSF D = −CP A compared to the case |CSF D | 6= |CP A |.
Furthermore, because of the length of the Kasami sequences
As described in Sec. II, standard draft currently uses small
of 63, compared to 31 of the Gold sequences, it can be noticed
Kasami synchronization sequence set of length 63. Although
that signal to noise ratio for the SFD detection is significantly
having very good circular autocorrelation and circular cross-
better in the case of large Kasami set compared to the Gold
correlation properties, this set is very small and consists of
set. However, even Gold sequence has large enough SNR for
only 8 sequences, i.e. 4 sequences per channel; this results
SFD detection failure probability to be lower than preamble
in high probability of collision in synchronization between
detection failure probability in the case CSF D = −CP A .
BANs in multi–BAN environments. Furthermore, because of
The reason of choosing MS = 3 for both Kasami sets
relatively large length of 63 duration of each sequence is
and MS = 6 for Gold set can also be understood from Fig.
16.128µs which with Nyquist sampling period of Ts = 2ns
3; by choosing these numbers of sequences repetitions the
results in a rather large dim {zP R } = 8064, i.e. rather large
probabilities of packet error due to preamble detection failure
circular correlation complexity (11).
In this work we will also consider two other synchronization and SFD detection failure are less than about ten times lower
sequence sets; first is Large Kasami set of length 63 and that packet error probabilities due to payload detection failure.
second one is Gold set of length 31. General properties of In other words, roughly less than 10% of the total packets lost
all 3 sequence sets considered are summarized in Table I. are lost due to the synchronization failure which is common
system design criteria in wireless communications.
C. Preamble structure
As described in Sec. II, preamble structure is determined VI. C ONCLUSIONS
by choosing a sequence CP A and a number of sequence
Uniqueness of the current preamble structure of the IEEE
repetitions MP A . As it will be shown in the next subsec-
802.15.6 IR-UWB PHY with differential modulation is that at
tion, it has been determined by the comparison of preamble
the time of synchronization receiver does not have information
detection failure probability performance with payload error
of a type of the transmitted waveform. For such conditions
performance that for length of preamble sequence of 63 (large
we developed synchronization method based on the sampling
and small Kasami sets), MS = 3 is enough, which gives
receiver with S-DPSK detection.
MP A = 4, already in the standard draft, as a minimal value.
Similarly for length of 31 of Gold sequences, MS = 6 Preamble structure currently in the draft of the standard
gives about the same SNR in preamble detection. In order uses the small Kasami set of length 63 of only 4 sequences
for preamble to have similar duration in all cases considered, per channel. Since each BAN choses synchronization sequence
MP A = 8 is used for the Gold sequences. from the set randomly, this causes high probability of collision
Probability of preamble detection failure performances for in synchronization among BANs in multi–BAN scenarios. For
3 cases considered and false alarm rate of PF A = 10−7 are this reason, we tested performance of two other larger syn-
shown in Fig. 2(a), while timing estimation loss performance chronization sequences sets: the large Kasami set of length 63
is shown in Fig. 2(b). It can be seen that all sequence sets and the Gold set of length 31. In the preamble detection all 3
perform similarly, although Gold set and large Kasami set sets considered perform very close. Furthermore, Gold set with
have poorer circular autocorrelation properties compared to the length of 31 results in two times complexity reduction of
the small Kasami set. Furthermore, from Fig. 2(b) it can be circular correlation employed in the synchronization method.
noticed that synchronization method has intrinsic timing loss Thus, we recommend usage of the Gold set of length 31 for
of about ητ = 0.3 dB for the payload. The reason for this is the preamble.
the different symbol structures of SHR and payload. Hence, SFD is currently undefined in the standard; according to
optimal timings of the SHR and the payload are different. developed synchronization method, we concluded that opti-
Thus, by estimating optimal timing for the SFD receiver mum approach is to use inversion of the preamble sequence
cannot estimate exact timing for the payload. for the SFD. SNR level in detection with a single repetition
of the Gold sequence of length 31 as a SFD proved to be
D. SFD structure good enough. Reusing the same sequence for the preamble
As explained in Sec. II, only cases when dim {CP A } = and the SFD also minimizes the number of sequences used
dim {CSF D } will be considered here. There are two ap- in a BAN and thus minimizes the probability of collision in
proaches that we will examine; first is to use the inversion synchronization among BANs in multi–BAN scenarios.
 100 0.8
Small Kasami set Small Kasami set, Payload
Large Kasami set Small Kasami set, SFD
Gold set Large Kasami set, Payload
Preamble detection failure probability

Large Kasami set, SFD

10−1 0.6
Gold set, Payload
Gold set, SFD

ητ (dB)
10−2 0.4

10−3 0.2

10−4 0
10 12 14 16 18 10 12 14 16 18 20
Eb /N0 (dB) in payload Eb /N0 (dB) in payload

(a) (b)

Fig. 2. Preamble detection performance for 3 different sequence sets considered. (a) Probability of preamble detection failure. (b) Timing estimation loss –
ητ .

100 100
Preamble, CSF D = −CP A Preamble, CSF D = −CP A
SFD, CSF D = −CP A SFD, CSF D = −CP A
Payload, CSF D = −CP A Payload, CSF D = −CP A
Preamble, |CSF D | =6 |CP A | Preamble, |CSF D | =6 |CP A |
10−1 10−1
SFD, |CSF D | =
6 |CP A | SFD, |CSF D | =
6 |CP A |
Payload, |CSF D | =
6 |CP A | Payload, |CSF D | =
6 |CP A |
Packet error rate

Packet error rate

10−2 10−2

10−3 10−3

10−4 10−4
10 12 14 16 18 10 12 14 16 18
Eb /N0 (dB) in payload Eb /N0 (dB) in payload

(a) (b)

Fig. 3. Different causes of packet error rate for two cases: CSF D = −CP A , |CSF D | =
6 |CP A |. (a) Large Kasami set. (b) Gold set.

Therefore, our overall recommendation is to use the Gold [3] I. Dotlić and R. Kohno, “Low Complexity Chirp Pulsed Ultra–Wideband
set of length 31 for the preamble with the SFD being a single System with Near–Optimum Multipath Performance,” IEEE Transactions
on Wireless Communications, 2011, accepted for publication.
inverted instance of the preamble sequence. [4] I. Dotlic and R. Kohno, “Performance analysis of Impulse Radio Ultra–
Wideband differential detection schemes for Body Area Networks,” in
R EFERENCES IEEE 21st International Symposium on Personal, Indoor and Mobile
[1] D. Miniutti, D. Smith, L. Hanlen, A. Zhang, A. Boulis, D. Rodda, and Radio Communications Workshops (PIMRC Workshops), 2010, pp. 72
B. Gilbert, “Sleeping channel measurements for body area networks,” –77.
IEEE 802.15 WPAN, no. IEEE 802.15-09-0778-01-0006, January 2010. [5] “IEEE 802.15 WPAN low rate alternative PHY task group 4a (TG4a).”
[2] “IEEE 802.15 WPAN task group 6 body area networks (BAN).” [Online]. Available: http://www.ieee802.org/15/pub/TG4a.html
[Online]. Available: http://www.ieee802.org/15/pub/TG6.html