Documente Academic
Documente Profesional
Documente Cultură
Multipath Channels
899
is substantially long, thus, this case cannot be ignored and where (t) is the Dirac delta function, L is the number
will be included in our analysis in a strict way. of resolvable paths, and k,l and k,l are the channel gain
Consequently, we can accurately evaluate the mean, vari- and timing delay of the kth users l-path. We assume k,l
ance, and cdf of the acquisition time. Simulation and nu- remains constant for a block of symbols, but will vary from
merical results verify our analysis and show the large per- one block to another.
formance improvement of the proposed acquisition schemes After undergoing a multipath channel, the received signal
over the serial search. Our result can be readily used for from the K users may be represented by,
practical UWB communication design and analysis for mul-
tiuser systems over multipath channels.
K
L
(j)
r(t) = Pk k,l (j)xk (t k,l ) + n(t) (4)
j=0 k=1 l=1
2. SYSTEM MODEL (j)
where Pk is the average signal power for user k, and xk (t)
2.1 Signalling Schemes was dened in (2) and (1) for UWB-BPSK and UWB-PPM
signals, respectively.
We consider pulse position modulation (PPM) and bi-
Without loss of generality, we assume that 1,1 = 0 and
nary pulse amplitude modulation (PAM), and the latter is
0 k,l < Ts +Tm for all k and l, where Tm is the maximum
equivalent to binary phase shift keying (BPSK).
multipath delay spread. n(t) is a zero-mean additive white
2.1.1 UWB-PPM Gaussian noise (AWGN) with power spectrum density N0 .
Consider a system with K active users. The received 2.3 Acquisition Methods
UWB-PPM pure signal the kth user over the j-th symbol
duration (k = 1, . . . , K, j = 0, 1, . . .) can be represented by 2.3.1 Conventional Serial Search
(j+1)Ns 1 In this scheme, we assume that the correlation receiver
(j)
(j)
xk (t) = w(t nTf ck,n Tc dk,PPM ) (1) output is tested consecutively, with the searching step size
n=jNs
tSer . For the case of uniformly-spaced multipath we assume
the multipath resolvable interval is ts = 1/B = Tw /M ,
where Tf is a frame interval, and there are Ns frames in one where B is the RF bandwidth of w(t), and M (assumed to
symbol period Ts , so that Ts = Ns Tf . w(t) is the received be an integer) is the number of paths in a pulse duration.
mono-cycle waveform, with major energy concentrated in The number of tested cells in the uncertainty region of
the duration t (0, Tw ) and Tw Tf . The autocorrelation one symbol interval Ts is
T /2
function of w(t) is dened as Rw ( ) = Tww /2 w(t)w(t Ns Tf
)dt, with Rw ( ) = 0 for (Tw , Tw ), and Rw ( ) = Nu = +1 (5)
tSer
0 otherwise. w(t) may be dierent from the transmitted
monopulse due to channel distortion. In our scheme, we where x is the largest integer not larger than x.
assume the receiver has the knowledge of w(t).
In (1), Tc is a chip interval, and ck,n , for n = 0, 1, . . . , Np
2.3.2 NCS/PD
1, is a time hopping (TH) sequence (or called hopping code) To utilize the eect of long-spread multipath and cy-
for the kth user, with period Np . When Np = Ns , this clostationarity of UWB signals, we may employ the path-
sequence is normally referred to as a short TH sequence; diversity and nonconsecutive search [11, 7]. For the PD,
otherwise if Np > Ns , it is called a long TH sequence. we assume that the correlator outputs for N dierent cells
In the acquisition mode for the desired user, short TH are combined, with a time delay dierence of Tw /2 between
code with period Ts is used. Accordingly, the correspond- the nth and (n + 1)th correlators, for n = 1, . . . , N . For
ing uncertainty region to be searched for acquisition pur- example, when the rst correlator is testing the ith cell, the
(j) second correlator is aligned to the (i + M/2)th cell. The
pose is Ts . We assume dk,PPM = 0 in the acquisition mode.
N cells under test is called a cell group (CG). For NCS,
the next CG to be tested has a time delay of N tSer with
2.1.2 UWB-BPSK
respect to the current CG, with N 2N .
For BPSK modulation, the signal may be represented by With noncoherent acquisition, the correlator outputs are
(j+1)Ns 1 summed by using the square-law combining to form a new
(j)
(j)
xk (t) = dk,BP w(t nTf ck,n Tc ) (2) decision variable (DV), which is compared to a detection
n=jNs
threshold. Faster acquisition than noncoherent acquisition
may be achieved by using the dierentially coherent ac-
(j) quisition [17], in which the correlation outputs at adja-
where dk,BP = 1 is the data for jth symbol of the kth
user. cent symbol durations are multiplied to form the DV. For
NCS/PD the acquisition is achieved when a cluster of cells
2.2 Channel Model within the H1 region is positively declared as signal cells.
IEEE 802.15.3a Model [10, 1] has been proposed for the Thus, our schemes achieve the pulse-level acquisition, in-
UWB channel modelling for wireless personal area net- stead of the more rough symbol-level and frame-level ac-
works (WPANs). In this model, the path delays are non- quisitions.
uniformly spaced according to clustered structures. The
channel impulse response for the ith user is given by 3. CORRELATION OUTPUT STATISTICS
AND ACQUISITION PROBABILITIES
L
hk (t) = k,l (t k,l ) (3) We derive the correlation receiver output statistics and
l=1
the related detection and false alarm probabilities, which
900
are then used to analyze the statistics of the acquisition the unconditional variance for ysi as
time in the next section. For tractability of the analysis, we
assume that the multipath channel has a uniformly-spaced s2i = s2i ,SIG + s2i ,MAI + Ns Rw (0)N0 (10)
paths. Our analytical result may be extended to the case
where
of non-uniformly spaced paths (e.g. the IEEE 802.15.3a
Model [10, 1]). Without loss of generality, we assume that
M
the rst user (k = 1) is the desired user. s2i ,SIG = Ns2 P1 2
1,i+l 2 (lt )
Rw s (11)
l=M
3.1 Statistics for Single Cells
Tm K
The correlator output for testing the ith cell at the jth s2i ,MAI = Ns 2
Pk Rw (12)
symbol duration is given by L Tf
k=2
1,i +(j+1)Ts
(j) which are are the variances of the desired signal and MAI
yi (j) = r(t)x1 (t 1,i )dt (6) 2 (lt ) = 1
1,i +jTs parts, respectively. In (11) and (12), Rw s
tSer /2 2 1
Tw 2
tSer
where 1,i is the delay position being tested. t /2 Rw (lts + )d and Rw = 2Tw Tw Rw
2 ()d.
Ser
When the i-th H1 cell is under test, ideally, the code For the i -th H0 cell (i = L + 1, . . . , Nu ), the variance of
delay of the local correlator is set to 1,i = 1,1 + (i 1)ts . the correlator output ysi is derived as
The actual test position (delay) closest to 1,i is denoted as
1,i , which has an alignment error denoted as 1 = 1,i Tm K
2
1,i . 1 can be modelled as a uniformly distributed random R = s2i = Ns 2 + N R (0)N
Pk Rw s w 0 (13)
L Tf
variable with 1 (0.5tSer , 0.5tSer ). k=1
The correlator output for testing i-th H1 signal cell is
given by 3.2 Statistics and Acquisition Probabilities For
1,i +(j+1)Ts NCS/PD
(j)
ysi (j) = r(t)x1 (t 1,i )dt For NCS/PD with N th-order diversity, the correlation
1,i +jTs outputs for N -cells (a cell group) are combined. Using the
l=M noncoherent squared-law combining the DV for the ith CG
= P1 Ns 1,i Rw (1 ) + P1 Ns 1,i+l Rw (1 + lts ) (with i = 1, . . . , Nu /N + 1, and i = i/N + 1 holds)
l=M is given by
l=0
Ns 1
K
M
N1
2
+ Pk k,l Rw (k + lts ) + nw (7) xnoncoh
i,N = yi+M k/2 (j) (14)
n=0 k=2 l =M k=0
where the rst term in the right hand-side of (7) is the where yi (j) is given by (6) (which may be an H1 or H0
desired signal output, and the second and the third terms cell).
are due to the interpath interference (IPI) and MAI re- For dierential acquisition with NCS/PD, the DV is given
spectively. The output noise nw is a zero-mean Gaussian by
variable with variance Ns Rw (0)N0 .
For the i -th H0 cell (for i = L + 1, . . . , Nu ), the output
N1
+(j+1)Ts
1,i
(j) In the H1 region , when xdiff > 1 or xnoncoh > 1 , where
ysi (j) = r(t)x1 (t 1,i )dt i,N i,N
1 is the detection threshold, the detection event is de-
1,i +jTs
clared. For the H0 region, when xdiff
i,N
> 1 or xnoncoh
i,N
> 1 ,
s 1
N
K M
the false alarm event occurs. For constant false alarm rate
= Pk k,l Rw (lts + k ) + nw (8) 2
(CFAR) design we may set 1 = R , where is a scalar
n=0 k=1 l =M
to be adjusted.
Since we let Tc Tw in our design, the collisions of The closed-form cdfs of the DVs for NCS/PD with dif-
dierent users pulses occur in a more random pattern than ferential and noncoherent acquisitions are in general dif-
the case of Tc Tw . Consequently, the resulting MAI to cult to derive. Here, we observe that the detection and
the desired user approaches a Gaussian distribution, which false-alarm probabilities is related to the complementary
is derived below (also c.f. the numerical results in [7]). cdf (ccdf) of the DV. Thus, we design a technique by de-
Conditioned on a set of time delays {k,1 }K k=1 and step riving the MGF of the DV and then employ the inverse
quantization errors {k }K
k=1 , the variance (or power) of ysi Laplace transform to obtain the acquisition probabilities,
is given by valid for arbitrary multipath intensity prole (MIP) with
M uniformly-spaced paths.
s2i (, ) = Ns2 P1 2
1,i+l 2
Rw (lts + 1 ) The DV xnoncoh
i,N
is a Gaussian quadratic form. When it is
l=M related to the ith H1 cell group (CG), its MGF is derived
as
K
M
2 2
+Ns Pk k,l Rw (lts + k ) + Ns Rw (0)N0 (9)
xnoncoh,H1 (s) = E[exp(sxnoncoh
i,N )]
k=2 l =M i,N
901
where E[] denotes expectation. For any of the H0 CGs, Equations (23) and (24) can be evaluated by using an
the MGF is obtained as ecient Gauss-Chebyshev quadrature (GCQ) formula. For
1 example, letting s = c + j and = c tan(/2), (23) is
xnoncoh,H0 (s) = N1 2 1/2
(17) reduced to
i,N
q=0 (1 sR )
1
M
For dierentially coherent acquisition, xdiff may be ex- (2n 1)
i,N Pd (i, N ) = 1 Re xi,N + RM
pressed in a quadratic form 2M n=1 2M
T
xdiff
i,N = v Q2N v (18) where RM is a residual term which vanishes quickly to zero
where v is the transpose of vector v, v = [yi (j 1), . . . ,
T as M increases, and xi,N () = xi,N (c + jc tan(/2))(1
yi+M N/2(j 1), yi (j), . . . ,yi+M N/2 (j)]T , and e[c+jc tan(/2)]T )[1 j tan(/2)].
0NN 0.5IN We note that in [11] for a CDMA multipath channel the
Q2N = , where 0NN and IN represent acquisition probabilities for the case of N = 3 and non-
0.5IN 0NN
the N N all-zero and identity matrices, respectively. The coherent acquisition were derived based on a closed-form
MGFs of xdiff
i,N
for H1 and H0 CGs are derived as cdf. In comparison, the acquisition probabilities for UWB
signals are more complex to evaluate and our technique is
1 more general in that it is valid for arbitrary N and also for
xdiff,H1 (s) = (19)
i,N det[I2N Pv,H1 Q2N ]m dierential acquisition.
1
xdiff,H0 (s) = (20)
i,N det[I2N Pv,H0 Q2N ]m 4. GENERATING FUNCTION OF ACQUI-
where SITION TIME
2
R IN 0NN In the literature for analyzing the acquisition probabil-
Pv,H0 = 2 . (21) ity, it is usually assumed that the cell-search starts from
0NN R IN
an H0 cell [8], or from the rst cell of the H1 region [6,
and 16], and the case that the search starts from the middle of
the H1 region has not been strictly analyzed. For UWB
Pv,H1 = channels, however, due to long multipath spread, this case
diag({s2i+2n }n ) b1 (j) diag({s2i+2n ,SIG }n ) must be accurately evaluated. Below, we derive a new and
b1 (j) diag({s2i+2n ,SIG }n ) diag({s2i+2n }n ) general GF result applicable to multipath with arbitrary
(22) statistics (e.g., delay spread, path powers), and then ob-
tain the mean, variance and cdf of the acquisition time TA .
(j1) (j)
where b1 (j) = d1,BP d1,BP for dierential UWB-BPSK and
b1 (j) = 1 for UWB-PPM. In (22), diag({s2i+2n }n ) = 4.1 Conventional Scheme
diag(s2i , . . . , s2i+2(N 1) ), and diag({s2i+2n ,SIG }n ) = There are L cells in the H1 region, and Nu L cells in
the H0 region. Without loss of generality, we dene the
diag(s2i ,SIG , . . . , s2i+2(N 1) ,SIG ), respectively. Here, rst L cells to be H1 cells. The initial position of the cell
diag(x1 , . . . , xL ) is a diagonal matrix that is formed by set- search (say iST ) is a random variable uniformly distributed
ting x1 , . . . , xL on its main diagonal. s2i+2n and s2i+2n ,SIG in [1, Nu ]. Without loss of generality, we let iST = 1 corre-
can be evaluated using (10) and (11), respectively. sponds to the rst signal path in the H1 region.
The probability of detection given the threshold 1 can In order to obtain the GF of TA , we derive some useful
be evaluated by expressions below. The GF for the time of miss detection
in the nth H1 cell is given by
Pd (i, N ) = Pr(xi (N ) > 1 ) = fxi (x)dx
1 Hm (s|n) = [1 Pd (n)]estd
s1
1 c+j
1e
=1 xi,N (s) ds (23) where td = Ts + tSer accounts for the time for the signal
2j cj s correlation and threshold comparison. The GF for the miss
detection time in the whole H1 region is
where j = 1, c is a positive real constant in the con-
vergence region, and xi,N (s) is given by (16) and (19) for
L
noncoherent and dierential acquisition schemes, respec- HML (s) = Hm (s|n) (25)
tively. The miss detection probability is obtained as n=1
Pm (i, N ) = 1 Pd (i, N ). The GF for the signal detection time in the nth H1 cell
is obtained as
The false alarm probability is derived as
Hd (s|n) = Pd (n1 )estd .
1 c+j
1 es1
Pf (i, N ) = 1 xi,N (s) ds (24)
2j cj s Assuming a serial search and L paths, the GF for the
signal detection time in the whole H1 region (consisting of
where xi,N (s) is given by (17) and (20) for noncoherent L signal paths) is derived as
and dierential acquisition schemes, respectively. It follows
n1 1
that the correct rejection probability is
L
902
The GF for the time used for the rejection of an false Mean TA, noncoherent acquisition
903
700 45
25
300
A
20
200
15
100
10
0 5
0 10 20 30 40 50 60 70 0 2 4 6 8 10 12 14 16
Step size N Diversity order N
thors would also like to thank Dr. Kenneth Peterson for algorithms for code acquisition in UWB impulse
some useful discussions. radio communications. IEEE J. Sel. Areas.
Commun., 23(5):898908, May 2005.
8. REFERENCES [10] A. Saleh and R. Valenzuela. A statistical model for
indoor multipath propagation. IEEE J. Sel. Areas
[1] I. P. 02/368r5 SG3a. Channel modeling Commun., 5(2):128137, Feb. 1987.
sub-committee report.
[11] O.-S. Shin and K. B. Lee. Utilization of multipaths
URL:grouper.ieee.org/groups/802/15/pub/, Dec.
for spread-spectrum code acquisition in
2002.
frequency-selective Rayleigh fading channels. IEEE
[2] G. E. Corazza. On the MAX/TC criterion for code Trans. Commun., 49(4):734743, April 2001.
acquisition and its application to DS-SSMA systems.
[12] W. Suwansantisuk, M. Z. Win, and L. A. Shepp. On
IEEE Trans. Commun., 44(9):11731182, Sep. 1996.
the performance of wide-bandwidth signal acquisition
[3] FCC. First report and order, revision of part 15 of in dense multipath channels. EEE Trans. Veh.
the commissions rules regarding ultra-wideband Technol., 54(5):15841594, Sept. 2005.
transmission systems. ET Docket 98-153, Feb. 14
[13] Z. Tian and G. B. Giannakis. Training sequence
2002.
design for data-aided timing acquisition in UWB
[4] E. A. Homier and R. A. Scholtz. Rapid acquisition of radios. In Proc. IEEE Int. Conf. Commun. (ICC),
ultra-wideband signals in the dense multi-path pages 33993403, June 2004.
channels. In Proc. IEEE Conference on Ultra
[14] M. Z. Win and R. A. Scholtz. Ultra-wide bandwidth
Wideband Systems and Technologies (UWBST), May
time-hopping spread-spectrum impulse radio for
2002.
wireless multiple-access communications. IEEE
[5] E. A. Homier and R. A. Scholtz. A generalized signal Trans. Commun., 48(4):679691, April 2000.
ow graph approach for hybrid acquisition of
[15] L. Yang and G. B. Giannakis. Blind UWB timing
ultra-wideband signals. International Journal of
with a dirty template. In Proc. IEEE Int. Conf.
Wireless Information Networks, 10(4):179191, Oct.
Acoustics, Speech, and Signal Processing (ICASSP),
2003.
pages 509512, May 2004.
[6] B. B. Ibrahim and A. H. Aghvami. Direct sequence
[16] L. L. Yang and L. Hanzo. Serial acquisition of
spread spectrum matched lter acquisition in
DS-CDMA signals in multipath fading mobile
frequency-selective Rayleigh fading channels. IEEE
channels. IEEE Trans. Veh. Technol., 50(2):617628,
J. Sel. Areas Commun., 12(5):885890, June 1994.
March 2001.
[7] Y. Ma, F. Chin, B. Kannan, and S. Pasupathy.
[17] M. H. Zarrabizadeh and E. S. Sousa. A dierentially
Acquisition performance of an ultra wide-band
coherent PN code acquisition receiver for CDMA
communications system over a multiple-access fading
systems. IEEE Trans. Commun., 45(11):14561465,
channel. In IEEE Conference on Ultra Wideband
Nov. 1997.
Systems and Technologies, pages 99103, 2002.
[18] Z. Zhang, L. Ge, B. Chen, and C. Li. Dierential
[8] A. Polydoros and C. L. Weber. A unied approach to
detection acquisition in multiple-access time-hopping
serial search spread-spectrum code acquisition part
ultra-wideband communications. In Proc. IEEE Joint
I: General theory. IEEE Trans. Commun.,
UWBST & IWUWBS, pages 283287, May 2004.
32(5):542549, May 1984.
[9] L. Reggiani and G. M. Maggio. Rapid search
904