Progress In Electromagnetics Research B, Vol.

57, 139–155, 2014

PN Code Acquisition Using Smart Antenna and Adaptive

Thresholding CFAR Based on Ordered Data Variability
for CDMA Communications

Kamel Berbra1 , Mourad Barkat2, * , and Abderrahmane Anou1

Abstract—Recently, a novel approach for PN code acquisition of direct sequence code division multiple-
access (DS-CDMA) systems in Rayleigh fading multipath channel was proposed in [1]. The authors
considered a combination of adaptive thresholding constant false alarm rate (CFAR) and smart antennas
to increase the system capacity and consequently enhance the detection performance. This paper
considers still the problem of PN code acquisition for DS-CDMA communication systems over Rayleigh
fading channels under the presence of multipath and multiple-access interference (MAI) signals. We
propose and analyze an adaptive array acquisition system, which integrates an adaptive thresholding
technique based on ordered data variability (ODV) index constant false alarm rate (ODV-CFAR) and
digital beamforming where a low complexity least mean square (LMS) algorithm is used to calculate
the optimal weighting coefficients. This approach is expected to mitigate interferences caused by the
presence of multiple access interference and multipath. Unlike other approaches based on a fixed
censoring point when the number of interferences is assumed known, ODV-CFAR processing does not
require prior knowledge about the number of interferences. The simulation results show a robust
performance of the proposed system in varying mobile communication channels.

1. INTRODUCTION

In direct sequence code division multiple access (DS-CDMA) communication systems, spectral expansion
is achieved by modulating each unit of information to be transmitted with a pseudo-noise (PN) random-
like code sequence, where each pulse is denominated as a chip. At the receiver the demodulation is
performed by multiplying the received signal by an aligned replica of the code used in the transmitter.
The alignment condition is essential for a successful demodulation since only when the codes are perfectly
aligned the received signal is reverted to its original format. The alignment operation known as code
synchronization is usually carried out in two steps; an initial coarse synchronization process known as a
code acquisition followed by a fine synchronization process known as code tracking. Acquisition is the
process of searching through an uncertainty region of code phases until the correct phase is found. The
uncertainty region is divided into a number of cells where a cell may be as small as a fraction of a PN
chip interval. In serial acquisition, the received signal is correlated with a locally generated replica of
the transmitted code and the correlation value is compared to a threshold in order to make a decision
about synchronization [2].
The setting of the threshold plays an important role on the performance of the acquisition process
and is the most challenging task. A lot of research on PN code acquisition has been carried out,
but focused on fixed–threshold acquisition where the threshold value is selected for a specific radio
environment. This results in a high false alarm rate and/or low probability of detection [3–6]. Since
the power of the received signals in mobile communication decreases with distance and are also subject
Received 24 September 2013, Accepted 4 December 2013, Scheduled 4 December 2013
* Corresponding author: Mourad Barkat (mbarkat58@gmail.com).
1 Department of Electronics, University of Saad-Dahlab-Blida, Algeria. 2 Department of Electrical and Computer Engineering

Technology, Valencia College, Orlando, Florida 32811, USA.

140 Berbra, Barkat, and Anou

to fast fading and slowly varying mean signal strength, fixed threshold detection schemes are unable
to adapt to these radio varying environments. One of the most promising approaches to meet these
challenges is the use of adaptive thresholding techniques. The goal of these approaches is to improve the
acquisition performance by keeping a constant probability of false alarm (Pfa ) [7]. Adaptive thresholding
CFAR processing is well developed in automatic radar signal detection [8].
In conventional wireless communications, a single antenna is used at the receiver. Such systems are
vulnerable to multipath and multiple access interference (MAI) caused by other users. Since in practice
the codes are not perfectly orthogonal, while the MAI caused by any one user is generally small, as the
number of interferences or their power increases, MAI becomes substantial. In CDMA systems, MAI is
the major factor limiting the performance and hence, limits the number of supported users to share the
same channel known also as the capacity of the system. The effect of the multiple access interference,
distinctive of CDMA systems, also has to be incorporated in the analysis of code acquisition. Therefore,
analysis of the effect of MAI on the system performance as well as ways to suppress MAI has been the
major focus of recent research on CDMA. A large amount of the research has focused on developing
these theories to address the difficulties of wireless communication. A number of different techniques
have been developed in this field. Many techniques are used to mitigate this effect as shown in some
very recent papers [9, 10]; the work in [9] compares maximal length sequences (m-sequences) codes and
Gold sequences codes in order to select a spreading sequence possessing desirable characteristics which
would minimize the effect of multiple access interference for implementation in CDMA systems. A blind
paralind multiuser detection over multipath fading channels to further suppress MAIs in CDMA systems
has been proposed and analyzed in [10]. A possible solution to overcome these channel impairments is to
use a smart antenna by steering beams toward the desired user while decreasing the total power level of
multiuser access interference (MAI). The smart antennas can increase the capacity of CDMA systems
through interference suppression by exploiting the spatial diversity [11]. In [13, 14] smart antennas
have been considered for acquisition where a correlator was used in each antenna element and the
outputs of the correlators were combined using a beamformer (BF). Smart antenna systems attract a
lot of attentions now and believably more in the future, as they can increase the performance of mobile
communication systems dramatically.
In [1], Sofwan and Barkat considered a novel approach using adaptive thresholding CFAR processing
combined to smart antennas as a powerful method to improve the PN code acquisition performance.
They used the trimmed-mean (TM) CFAR processor for adaptive thresholding [14]. The same
acquisition system was adopted in [15], but using the censored mean level detector CMLD-CFAR [16] in a
Rayleigh fading multipath channel. The above mentioned systems used adaptive order statistic detection
based on a fixed censoring point and they perform robustly as long as the interferences are properly
discarded. However, due to the lack of a priori knowledge about the radio environment, the censored
cells may be poorly selected causing significant performance degradation [17]. Hence, to overcome
the above mentioned problem, we propose in this paper the use of an automatic censoring acquisition
processor based on ordered data variability index ODV-CFAR [18]. The censored cells are automatically
determined according to the actual number of interferences and thus the adaptive threshold is set
accordingly. We recall that automatic censoring was extensively used in the literature [19–24]. The
proposed system which uses a combination of smart antennas and adaptive ODV-CFAR processing
demonstrates a robust performance for PN code acquisition regardless of the surroundings as will be
shown in the next coming sections.
The remainder of this paper is organized as follows. In Section 2, a description of the proposed PN
code acquisition system and the problem formulation are presented. The automatic censoring processor
considered with the performance analysis as well as the expression for the probability of false alarm
are provided in Section 3. Monte Carlo simulation results and discussions are given in Section 4. The
findings and conclusions are then discussed in Section 5. The mathematical derivations of statistics of
the correlator output and probability of false alarm Pfa are given in the Appendices A and B respectively.

2. SYSTEM DESCRIPTION AND PROBLEM FORMULATION

The block diagram of the proposed serial search acquisition DS-CDMA system with smart antennas
is shown in Figure 1. A linear array with M elements and spacing d equal to one half of the carrier
Progress In Electromagnetics Research B, Vol. 57, 2014 141

Optimum
weights
1 Y1 Tracking
W
Correlator 1
fc
H1
D users 2 Y2 z ODV-based automatic
Correlator 2 LMS | .| 2 censoring
fc CFAR processor H0

M
Correlator M Update phase
YM
fc By ∆Tc

Figure 1. Block diagram of the proposed acquisition scheme with smart antennas.

V Im
H* (w) Correlator Y 1m

rm(t) 2 cos(wc t)
C 1 (t - jTc /2) Local PN code
generator
2 sin(w c t)
V mQ
H* (w) Correlator Y mQ

Figure 2. Correlator block diagram.

wavelength is assumed (d = 0.5λ), where λ is the wavelength of the carrier transmitted signal. The
array antenna elements are also assumed to have an identical response to any direction.
In this reverse link, we assume that there are D actively transmitting users in the system; the
first user is considered as the initial synchronization user whose performance is to be evaluated. Each
user is assigned a unique CDMA code sequence Ck (t) which spreads the data sequence while all the
users transmit with orthogonal PN codes and their signals fade independently. In each branch Ym of
Figure 1, z represents a complex signal with the in-phase I and quadrature-phase Q components while
W = [w1 w2 . . . wM ]T are the complex beam former weights and T denotes transpose.
The received signal is first down-converted into in-phase I and quadrature-phase Q components by
using a matched filter to the rectangular shaping pulse of duration Tc to discard the carrier frequency
at 2wc as shown in Figure 2. Two I-Q correlators perform a correlation over an observation interval or
dwell time interval τD = RTc between the locally generated PN code C1 (t − ∆Tc ) of the desired user and
the received signals; in our case ∆ = 0.5. The LMS block is used to process the information obtained
by the antenna array in order to optimize the complex weights, multiplying the incoming signals with
the complex weights obtained and summing them together (spatial correlation) to obtain the desired
radiation pattern. The optimization is based on minimizing the contribution from the interferences
while producing maximum beam gain at the desired user direction. Then, the output of the LMS
processor goes through an automatic censoring processor unit which is a CFAR based on ordered data
variability index (ODV-CFAR) [18]. The proposed detector does not require any a priori information
about the radio channel and uses the variability index statistic to reject the corrupted cells so that a
threshold can then be set in order to decide as to whether there is acquisition or not.

2.1. Transmitted Signals and Channel Model

The complex transmitted signal for the kth user can be expressed as
p
Sk (t) = 2Pk bk (t) C k (t) ej(wc t+ϕk ) (1)
where
142 Berbra, Barkat, and Anou

Pk is the transmitted power of the kth signal,

bk (t) is the data waveform,
Ck (t) is the spreading sequence of the kth user,
wc the common angular carrier frequency, and
ϕk the phase introduced by the kth modulator.
Assuming D users communicating over the same channel, the first one is the initial synchronization user
whose PN code is being acquired (acquisition phase), while the others are the data transmission users
who have finished acquisition. During the preamble-assisted acquisition phase, the transmitter aids the
initial synchronization by transmitting an unmodulated PN sequence b1 (t) = 1.
Each signal Sk (t), k = 1, 2, . . . , D is transmitted through a communication channel assumed to
be a Rayleigh fading multipath channel. The mobile radio channel is modeled by using an accepted
model [25] consisting of an L tapped-delay-line structure with a tap spacing of one chip and where
L corresponds to the number of resolvable multipath. The tap weights αkl , k = 1, 2, . . . , D and
l = 0, 1, . . . , L − 1 are assumed to be independent and identically distributed (I.I.D) Rayleigh random
variables with a probability density function (PDF) given by [25]
x2
2x − σ2
fαkl (x) = 2 e f ; x≥0 (2)
σf
2 ) = σ 2 is the fading channel power and the fading phases ψ are (I.I.D) random variables
where E(αkl f kl
uniformly distributed over the interval [0, 2π). We also assume that the fading amplitude is constant
during an observation interval but changes from one observation time to another.
The complex received signal at the mth branch of the array antenna can be expressed as [26]
(L−1 )
p X
j(wc t+∅1l ) −jπ(m−1) sin(θs )
rm (t) = 2PR α1l C1 (t − τ1 − lTc ) e e
l=0
( D L−1
)
p X X
+ 2PI αkl bk (t−τk − lTc )Ck (t−τk −lTc ) ej(wc t+∅kl ) e−jπ(m−1) sin(θk ) +nBP m (t) ;
k=2 l=0
m = 1, 2, . . . , M (3)
where PR is the average received power of the initial synchronization user, PI is the average
received power of each interfering signal, τk the relative time delay associated with the asynchronous
communication channel, ∅kl = ϕk − ψkl − wc (τk + lTc ) represents I.I.D random variables uniformly
distributed in [0, 2π), Tc is the chip time duration, θs is the direction of arrival (DOA) of the desired
user and θk is the DOA of the interfering user. Assuming that (D − 1) interfering users are engaged
in the data transmission, then their corresponding signals are ideally power-controlled and the average
received power of each interfering user is PI . However, for the initial synchronization user, it is not
possible to consider ideal power control before successful code acquisition. Hence, the average received
power of the initial synchronization user is usually different from that of the data transmission users.
This power is denoted by PR . In Equation (3) the received signal can be divided into three parts,
the received signal from the desired user, the multiple access interferences (MAIs) from the (D − 1)
interfering users and an additive white Gaussian bandpass noise (AWGN) with a double-sided power
spectral density of N0 /2.

2.2. Statistics of the Correlators Outputs

In this section, we derive the probability density function (PDF) of the in-phase I and quadrature
phase Q components at the output of the active correlator for each of the m branches. The following
assumptions are used [27, 28]
(1) The search step size is Tc /2,
(2) R À 1 such that the correlation between the received and the locally generated PN sequence is
zero when they are not aligned,
Progress In Electromagnetics Research B, Vol. 57, 2014 143

(3) The self-interference of the desired user caused by the (L − 1) paths, and the multiple access
interference MAI caused by the other (D − 1) users can be modeled as an additive white Gaussian
noise.
The complex equivalent baseband signal Vm (t) at the input of the correlator for the mth element can
be written as [26]
(L−1 )
√ X
Vm (t) = P R α1l C1 (t − τ1 − lTc ) ( cos (∅1l ) + j sin (∅1l )) e−jπ(m−1) sin(θs )
l=0
( D L−1
)
√ X X −jπ(m−1) sin(θk )
+ PI αkl bk (t−τk − lTc ) C k (t − τk − lTc ) ( cos (∅kl )+j sin (∅kl )) e +nm (t) ;
k=2 l=0

m = 1, 2, . . . , M (4)
where nm (t) = nIm (t) + jnQ
m (t)
is a complex AWGN with mean zero and variance N0 /2. The correlator
output can then be expressed as
(L−1 ) ( D L−1 )
X³ Q
´ XX³ Q
´
I −jπ(m−1) sin(θs ) I −jπ(m−1) sin(θk )
Ym = YD,l + jYD,l e + YMAI ,k,l + jYMAI ,k,l e
l=0 k=2 l=0
| {z } | {z }
1 2
© I Q
ª
+ Nm + jNm ; m = 1, 2, . . . , M. (5)
| {z }
3
The in-phase and quadrature phase components of the self-interference (term 1), the multiple access
interference (term 2) and thermal noise (term 3) can be modeled as Gaussian random variables with
corresponding means and variances given by (A7) to (A13) as shown in Appendix A. These results will
be used in the next section to determine the PDF at the output of the LMS processor under the aligned
and non-aligned hypotheses.

2.3. Smart Antenna and Adaptive Beamforming

Smart antenna in the proposed system performs beamforming. This operation is achieved by adjusting
weights of each of the antenna elements used in the array so that the signals received from all elements
are in phase in a particular direction. This will direct the main array pattern towards the desired
user and introduces nulls at interfering directions. We use the least mean square (LM S) algorithm
which incorporates an iterative procedure that makes successive corrections to the weight vector
W = [w1 w2 . . . wM ]T in order to achieve a minimum mean square error (MMSE ). The LMS algorithm
is important because of its simplicity and ease of implementation [12].
The updated weights can be written as
wm (n + 1) = wm (n) + µe∗ (n)Ym (n); m = 1, 2, . . . , M (6)
where
e (n) = M − WH (n) Y (n) (6a)
and Y(n) =[Y1 (n) Y2 (n) . . . YM (n)]T , we choose µ = 1/M to insure the convergence of the LMS
algorithm [12].
When the optimum weight vector W is obtained and the MMSE is achieved, this vector is used to
generate a spatial correlation output z given by [13]
z = WH Y (7)
T
where Y =[Y1 Y2 . . . YM ] are the correlators outputs from the M branches. Equation (5) can be
expressed in a vector form as
³ ´ XD ³ ´
Y = YDI + jYDQ a(θs ) + I
YMAI ,k + jY Q I
MAI ,k a(θk ) + N + jN
Q
(8)
k=2
144 Berbra, Barkat, and Anou

where
X³
L−1 ´ L−1
X h iT
YDI,Q = I,Q
YD,l ; I,Q
YMAI ,k = I,Q
YMAI ,k,l and NI,Q = N1I,Q N2I,Q . . . NM
I,Q
(8a)
l=0 l=0
and
£ ¤T
a (θ) = 1 e−jπ sin(θ) . . . e−jπ(M −1) sin(θ) (8b)
is the array steering vector.
Perfect weights are that combine different signals in order to compensate the different phase shifts
created by the steering vector at each branch, if we assume that the smart antenna can track perfectly
the desired user then
W = a (θs ) (9)
H
Substituting for W in (7) and (8) and using a(θs ) a(θs ) = M , then z can be found to be
³ ´ D ³
X ´
H¡ I ¢
z=M YDI + jYDQ + I
YMAI ,k + jY Q H
MAI ,k a (θs ) a(θk ) + a (θs ) N + jNQ (10)
k=2
where
(M −1) sin( M
2 ψ)
a (θs )H a (θk ) = e−j 2
ψ
, M × AFn (θk ) (10a)
sin( ψ2 )
ψ = π[sin(θk ) − sin (θs )] and AFn (ψ) is by definition the normalized array factor which represents a
pattern function associated with the array geometry. We observe from (10a) that the normalized array
factor |AFn (ψ)| has a maximum value of 1 in the direction of the desired user, while the other directions
of the interfering users are affected by different values smaller than 1. In conclusion by steering the
main beam toward the desired user, the smart antennas provide an alternative mean to cope with MAIs
by decreasing the total power level of interfering users. Consequently z can be written as
³ ´ D ³
X ´
I Q I Q H¡ I ¢
z = M YD + jYD + M YMAI ,k + jY MAI ,k AFn (θk ) + a (θs ) N + jNQ (11)
k=2

In addition, the noise components N1I,Q , . . . , NM

I,Q
are assumed to be independent from one branch to
another and thus by exploiting the circular property of complex Gaussian random variables (RVs) the
variance of the in-phase thermal noise term in (11) becomes (see Appendix A)
³ H
´ 1
E a (θs )H NI NI a (θs ) =M (12)
2RSchip
In the case of aligned hypothesis H1 , the in-phase and quadrature-phase components of z are
Gaussian random variables with mean zero while the variance can be obtained using (A7), (A8) and (A9)
such that
à !
h 2
i h 2
i 9σ 2
f
E (z I ) = E (z Q ) = M 2 + M σ 20 = M σ 20 (1 + M ν) (13)
32
where
9σf2
ν = (13a)
32σ02
¡ 2¢
σf2 = E αkl (13b)
and
 PD 2

2
(L − 1) σ 2f Lβ k=2 |AFn (θk )| σ f 1
σ02 = M  + + (13c)
3R 3R 2RSchip
Progress In Electromagnetics Research B, Vol. 57, 2014 145

It should be noted that σ02 represents the background noise resulting from three signals; namely the
self-interference from the non-aligned paths, the multiple access interference caused by interfering users
and thermal noise generated at the antenna array. This shows clearly that the MAI term is affected
by the beam pattern and this MAI and the self-interference are modeled as a net increase in the noise
power.
Given the Gaussian nature of z I and z Q , the final decision variable Z = |z|2 is a central chi-square
distributed with two degrees of freedom. Its PDF can be expressed as [8]
µ ¶
1 −z
fZ|H1 (z|H1 ) = exp ; z≥0 (14)
M 2σ 20 (1 + M ν) M 2σ 20 (1 + M ν)
The probability density function of the decision variable Z under hypothesis H0 can be obtained
from Equation (14) to be
µ ¶
1 −z
fZ|H0 (z|H0 ) = exp ; z≥0 (15)
M 2σ 20 M 2σ 20
From (14) we can observe that the use of a smart antenna will improve the SNR significantly (maybe
a little less than the maximum gain 10 log10 (M ) dB [13]) as compared to a single antenna case.

3. ODV-CFAR AUTOMATIC CENSORING PROCESSOR

In the serial search strategy, the PN code acquisition of DS-CDMA signals is considered as a binary
hypothesis testing problem since each cell is classified as one of the two states; H1 representing a correct
phase and H0 for incorrect phase case. Due to receiving multiple replicas of the transmitted PN codes
in multipath fading radio communication environments, usually there exist more than one state H1 , in
which the locally generated and the received codes may become synchronous, and hence satisfying the
presence of multiple H1 cells in the reference window (sliding window) used to estimate the background
noise level given by (13c) (a contribution of three main terms: thermal noise, MAI interference caused
by other users, and the self-interference caused by the nonaligned paths) as shown in Figure 3.
Sliding window
70

: Cell Under Test

60
: Background noise
50 : Synchro cells

40
Power

30

20

10

0
0 10 20 30 40 50 60
Delay cells

Figure 3. PN acquisition in presence of corrupted cells in the reference window due to multipath.

Based on the assumption that the communication channel is modeled as an L tapped delay line
that correspond to the L resolvable multipath. The multipath intensity profile (MIP) is assumed to
be either uniform or exponentially decaying with the decay rate µ. When the total fading power in all
the resolvable paths is normalized to unity, the average fading power in each resolvable path can be
expressed as [29]

1
£ 2 ¤ L ; µ=0
E αkl = 1 − e−µ (16)

 e −(l−1)µ
; µ =
6 0; l = 1, 2, . . . , L
1 − e−µL
146 Berbra, Barkat, and Anou

Hence, each resolvable path contributes two (02) H1 cells because the search step size is Tc /2. We note
that in real situations the parameter L is not only unknown but can also vary with time. Consequently,
if the number of cells to be censored is not properly selected, the system may undergo significant
performance degradation. It is well known that the performance of the conventional CA-CFAR detector,
which is suboptimum in a homogeneous environment and is equivalent to the optimum Neyman-Pearson
detector as the number of reference cells increases, is serious degraded in the presence of multipath.
However, to overcome this problem, many alternative CFAR processors have been proposed in the
literature, and in particular order statistics (OS) based CFAR processor proved to be more robust [14–
21]. In OS-CFAR processing, the references cells are rank ordered in an ascending order according to
their magnitude. The higher cells are thus assumed to contain noise plus interference, which may be
discarded if the exact number is known as this is not the case in real situations, since they unnecessarily
raise the threshold which may cause a serious degradation in the probability of detection. The OS-CFAR
processor considered in [4] estimates the background noise power from kth smallest cell in the reference
window. In the censored mean level detector (CMLD) CFAR considered in [15, 16] some largest range
cells are censored from the reference window and the threshold is computed from the remaining cells.
The CMLD is robust against multipath interference as long as the number of censored cells exceeds
the number of multipath interference. A generalization of the OS-CFAR and the CMLD known as
the trimmed mean TM-CFAR detector has been studied in [1]. All the conventional ordered statistics
detectors (OS-CFAR, CMLD and TM-CFAR) have been known to yield good performance as long as the
cells containing multipath signals are properly discarded. However, in the previous acquisition systems,
the number of censored cells is preset for the radios environments, while in practical situations, the
number of corrupted cells depending on the parameter of the model L is not only unknown but can also
vary with time as explained earlier. For this reason, we propose to use in an adaptive thresholding scheme
for maintaining the probability of false alarm constant while the probability of detection is robust.
The most important feature of the proposed system is its ability to dynamically estimate the number
of multipath signals that may lie in the reference window by using an automatic iterative backward
censoring technique based on ordered data variability (ODV) index. The ODV-CFAR dynamically
selects a group of reference cells to use in forming the threshold. This can be summarized as shown in
Figure 4.
The decision variables at the output of the square law detector are sent serially in a shift register
of length Nc + 1. The first cell Z0 corresponds to the cell under test and the remaining cells correspond

Z0 Z1 Z2 ... Z Nc

Ranking Device
Z(1) <
_ Z(2) ... <
_ Z(Nc)
H1 to tracking

Background Noise
Automatic k Level Estimator
Censoring Nc-k
H0
Block Q= Σ Z(j)
j=1
Code phase
Updated by Tc /2
Design Pfc Tk Q

Look-up Table Local PN code

Design Pfa For T k Selection generator

Figure 4. ODV-CFAR automatic censoring processor.

Progress In Electromagnetics Research B, Vol. 57, 2014 147

to the reference window which stores the output of the previous code phases. The reference cells are
ranked in ascending order according to their magnitude to form the order statistics
Z (1) ≤ Z (2) . . . ≤ Z(Nc ) (17)
The indices in parentheses indicate the rank order number, Z(1) and Z(Nc ) denote the minimum and
the maximum value respectively. Then the unwanted cells are censored, where the number is determined
by an automatic censoring block as will be described next in Section 3.1. After censoring the unwanted
cells, an estimate of the background noise level is formed from the remaining ordered data cells as
NX
c−k
Q= Z (j) (18)
j=1

where Nc is the number of reference cells and k is the number of cells censored from the upper end
by the censoring block. This estimate is then multiplied by a constant Tk which is selected so that to
achieve a desired false alarm probability Pfa . Implemented as a look-up table, the threshold multiplier
is then used to yield an adaptive threshold Tk Q. Finally, the test cell Z0 is compared with the adaptive
threshold Tk Q in order to make a decision about synchronization according to the following testing
hypothesis:
H1
Z0 ≷ Tk Q (19)
H0
If the threshold is exceeded, then a correct cell has been found and tracking is performed. Otherwise, the
local PN code generator is automatically adjusted to the next offset position, and the test is repeated.

3.1. Description of the Censoring Block Algorithm

The proposed censoring algorithm uses the variability index [18]; it is an iterative backward method
which automatically estimates the number of largest samples to be discarded. The basic idea is
to consider that the p lowest cells represent the initial estimation of the background noise level,
provided that p ≥ 12, next at each kth step (k = 0, . . . , N c − p − 1), we form the ordered subset
Es (k) = [Z(1), Z(2), . . . , Z(p), Z(N c − k)] and an ODV statistic corresponding to a shape parameter
of this population Es (k) is computed as follows [18]
σp + (Z(N c − k))2
Vk = (20)
(µp + Z (N c − k))2
where
p
X p
X
µp = Z(i) and σp = [Z(i)]2 (21)
i=0 i=0
then Vk is compared with a threshold Sk to decide whether the cells used in Vk are from a homogeneous
or a nonhomogeneous population using the following test

H1
Vk ≷ Sk (22)
H0

If H1 in (22) is true, then the subset Es (k) = [Z(1), Z(2), . . . , Z(p), Z(N c − k)] is assumed to
be nonhomogeneous, the ordered statistic Z(N c − k) is rejected and the test is performed again with
k incremented by one. This is continued until the test of (22) is not true that is we have hypothesis
H0 and the censoring algorithm stops. Next, the obtained value of k is used to yield the statistic Q
given in (18) by censoring the k highest cells and select the corresponding scaling factor Tk such that
Pfa is maintained constant at the design value. In each test, the corresponding thresholds Sk are fixed
such that a probability of false censoring Pfc is maintained. We define this probability representing the
probability of test hypothesis error in a homogenous population at the kth step by
Pf c = Prob {Vk > S k / [Z (1) , . . . , Z (p) , Z (Nc − k)] Homogeneous} (23)
148 Berbra, Barkat, and Anou

Since an analytic closed expression for the PDF of Vk is not available, we resort to Monte Carlo
simulations to estimate Pfc . The ODV-based censoring algorithm can be summarized as follows:
Step 1: Set k = 0;
Step 2: Form the ordered subset Es (k) = [Z(1), Z(2), . . . , Z(p), Z(N c − k)];
Step 3: Compute the shape parameter Vk of the subset Es (k);
Step 4: Perform the ODV-based hypothesis test given by (22);
Step 5: If H1 is a true, repeat steps 2, 3 and 4 for k = 1, 2, . . . until H0 is true or k = Nc − p;
Step 6: if H0 is a true, obtain the number of corrupted cells k.
Table 1 gives some values of the parameter Sk computed using Monte Carlo simulations. In practical
systems with programmable signal processing capability this values are stored in a look-up table.

Table 1. ODV thresholds for a Pf c = 10−2 and different size Nc of the reference window.

(Nc, p) Sk
S0 S1 S2 S3 S4 S5 S6 S7
(16, 12) 0.356 0.247 0.199 0.173 - - - -
(24, 16) 0.331 0.235 0.190 0.162 0.143 0.131 0.123 0.117

3.2. Performance Analysis of the ODV-CFAR Detector

In the proposed censoring processor, the number of censored cells k is obtained by an automatic censoring
block and the remaining ones are used to form the statistic
NX
c−k
Q= Z (j) (24)
j=1
then the cell under test Z0 is compared with the adaptive threshold Tk Q in order to make a decision
according to the decision rule
H1
Z0 ≷ Tk Q (25)
H0
The probability of false alarm in this case is given by
Z∞ Z∞
Pfa = P {Z0 > Tk Q/H0 } = fZ0 (Z0 /H0 ) fQ (q/H0 ) dz0 dq (26)
0 Tk Q
Substituting fZ0 (z0 /H0 ) from (15) into (26), we obtain
Z∞ µ ¶ µ ¶
−T k q Tk
Pf a = exp fQ (q/H0 ) dq , ΦQ (27)
M 2σ 20 M 2σ 20
0
where ΦQ (·) denotes the moment generating function (MGF) of Q. Equation (27) will be used to derive
a closed form expression for Pfa in terms of Tk . As shown in Appendix B, the probability of false alarm
Pfa is given by [30]
  Nc −k
 N k−1 NcX
X −m−1 µ ¶µ ¶ 
Pfa =
c  1 − Nc − 1 Nc − m − 1
(−1)j
1  (28)
 (Nc − k) 1 + Tk m j Tk + m + j + 1 
m=0 j=0

As a special case, when k = 0 (i.e., without censoring) the expression of the Pfa is reduced to
· ¸Nc
1
Pfa = (29)
1 + Tk
which is just the probability of false alarm of the CA-CFAR processor in a homogenous environment as
expected, and which used widely in the literature to determine the scaling factor Tk .
Progress In Electromagnetics Research B, Vol. 57, 2014 149

4. SIMULATION RESULTS AND DISCUSSIONS

In this section, we evaluate the performance of the proposed adaptive thresholding acquisition scheme
with smart antennas by means of computer simulations for various parameters. To demonstrate the
functionality of this communication system, a design probability of false alarm Pfa = 10−3 is used. The
multipath intensity profile (MIP) is assumed to be uniform (µ = 0) (except when it is indicated to have
other values) as a result the Rayleigh fading channel power is set to σf = 1/L, and the correlation length
integer of the dwell time interval is set at the value of R = 128. The adaptive ODV-CFAR thresholding is
also compared with other CFAR detectors such as CA-CFAR [1] and OS-CFAR [4] detectors in different
situations. For the OS-CFAR processor simulations, the cell representing the threshold is selected equal
to k = 0.75 N c [31] while the p lowest cells parameter in the OVD-CFAR detector is selected equal to
16. We first assess the capabilities of the ODV-CFAR processor in estimating the mean power level
of the background noise used in setting the adaptive threshold. In Figure 5, we show the probabilities
of detection Pd as a function of the SNR/chip for the proposed ODV-CFAR and OS-CFAR detectors
in the absence of interferences. The optimum Neyman-Pearson (N-P) detector is also presented for
comparison purposes. For a number of antenna elements M = 3 and in the absence of interferences,
we observe that both detectors have comparable probabilities of detection which are almost equal to
the probability of detection of the optimum Neyman-Pearson detector. There is a slight improvement
for the ODV-CFAR detector over the OS-CFAR detector and some CFAR loss relative to the optimal
N-P detector. Note that in the absence of interferences, the proposed system is reduced to CA-CFAR
processing, which is known to be an appropriate processor in such a homogeneous background radio
environment.
1 1
0.9 ODV-CFAR, µ = 1
0.9
ODV-CFAR, µ = 0.5
0.8 Neyman-pearson 0.8 ODV-CFAR, µ = 0
Probability of Detection, PD

Probability of Detection, PD

CFAR Loss ODV-CFAR

0.7 CA-CFAR, µ = 1
OS-CFAR 0.7
CA-CFAR, µ = 0.5
0.6 0.6 CA-CFAR, µ = 0
0.5 0.5

0.4 0.4

0.3 0.3 P fa =0.001

R=128
0.2 P fa =0.001 0.2 P fc =0.01
R=128 Nc=24
0.1 P fc =0.01 0.1
Nc=24
0 0
-30 -25 -20 -15 -10 -5 0 5 10 -30 -25 -20 -15 -10 -5 0 5 10
SNR/chip (dB) SNR/chip (dB)

Figure 5. Probabilities of detection of the op- Figure 6. Probabilities of detection of the ODV-
timal N-P, ODV-CFAR and OS-CFAR detectors CFAR and CA-CFAR detectors with r = 1 and
in a homogenous background (no interferences in the decay rate µ as a parameter; Pfa = 10−3 ,
the reference window); Pfa = 10−3 , Nc = 24 and N c = 24, M = 3.
M = 3.

For a number of antenna elements M = 3, Figure 6 shows the effects of µ on the detection
probability of the proposed system ODV-CFAR and the CA-CFAR processor in the presence of one
interference (r = 1) in the reference window. We observe a degradation of the performance of the CA-
CFAR processor as the parameter µ decreases. Note that µ = 0 is the worst case situation for a uniform
multipath intensity profile. We can also observe that varying µ has no effect on the performance of
the ODV-CFAR detector because the interference is successfully removed by the censoring algorithm.
This robustness against interference of the proposed algorithm over the CA-CFAR detector is clearly
observed which represents a desirable feature in many practical situations.
For the same number of antenna elements M = 3 and in the presence of r interferences in the
reference window, we observe from Figure 7 that probability of detection of the CA-CFAR processor
based on the arithmetic mean of the reference cells is seriously reduced as the number of interferences
150 Berbra, Barkat, and Anou

increases as expected, while the detection performance of the ODV-CFAR processor based on order
statistics of the output references cells and automatic censoring maintains its robustness in the sense
that that no excessive degradation of the performance occurs.
As the number of interferences r is further increased to r = 7, we now observe from Figure 8 a
performance degradation of the OS-CFAR detector even though it is an order statistics based detector,
which presents a serious limitation due to the fact that the number of interferences exceeds a known
integer threshold which is defined to be Nc −k, where k = 3N c/4; that is an interference has been selected
for background noise level estimation resulting in an overestimated threshold and thus degradation of
the performance.
In Figure 9, we present the detection performance of the ODV-CFAR processor for r = 3 interfering
signals while the number of antenna elements is increased from M = 1 to M = 6. We observe that
the probability of detection increases as expected. However, the rate of increase decreases with the
number of antennas. This enhancement in the detection performance is due to the attractive features of
smart antennas in the enhancement of the signal to noise ratio by reducing the power of the noise after

1 1
ODV-CFAR, r=0
0.9 0.9 ODV-CFAR, r=7
ODV-CFAR, r=1
OS-CFAR, r=6
ODV-CFAR, r=2
0.8 0.8 OS-CFAR, r=7
ODV-CFAR, r=3
Probability of Detection, PD

CA-CFAR, r=7

Proability of Detection, PD
0.7 CA-CFAR, r=0 0.7
CA-CFAR, r=1
0.6 CA-CFAR, r=2 0.6
CA-CFAR, r=3
0.5 0.5

0.4 0.4
P fa =0.001
0.3 Nc=24
0.3
P fa =0.001 R=128
0.2 R=128 P fc =0.01
0.2
Nc=24
0.1 0.1

0 0
-30 -25 -20 -15 -10 -5 0 5 10 -30 -25 -20 -15 -10 -5 0 5 10
SNR/chip (dB) SNR/chip (dB)

Figure 7. Probabilities of detection of the ODV- Figure 8. Probabilities of detection of the ODV-
CFAR and CA-CFAR detectors with the number CFAR, CA-CFAR and OS-CFAR detectors in
of interferences r as a parameter; Pfa = 10−3 , the presence of r interferences in the reference
N c = 24, M = 3. window; Pfa = 10−3 , N c = 24, M = 3.
1 1
0.9 0.9
0.8 0.8
Probability of Detection, PD

Probability of Detection, PD

0.7 0.7
0.6 0.6
0.5 0.5
Pfa = 0.001
0.4 M=6 0.4 R= 128
M=5 Nc=24
0.3 M=4 0.3 Pfc =0.01
r=3
0.2 M=3 0.2 ODV-CFAR, M=5
M=2 ODV-CFAR, M=1
0.1 M=1 0.1 CA-CFAR, M=5
CA-CFAR, M=1
0 0
-30 -25 -20 -15 -10 -5 0 5 10 -30 -25 -20 -15 -10 -5 0 5 10
SNR/chip (dB) SNR/chip (dB)

Figure 9. Detection performance of the proposed Figure 10. Probability of detection versus
system with the number of antennas M as a SNR/chip of the ODV-CFAR and CA-CFAR
parameter; Pf a = 10−3 , N c = 24 and r = 3. processors in the presence of r = 3 interferences
for different number of antennas M ; Pfa = 10−3
and N c = 24.
Progress In Electromagnetics Research B, Vol. 57, 2014 151

the sum of the different branches of the antenna array, and thus consolidating the signal in a specific
direction, which is the direction of the interest user. We also observe that when the number of antennas
is greater than M = 4, we start to reach a saturation level. On other hand, a slight enhancement of
performance is attained with a high complexity in the signal processing unit which increases the cost of
the hardware implementation.
Figure 10 shows the detection probability without smart-antenna (M = 1) and with five elements
smart antenna (M = 5) while the number of interfering signals is still r = 3. These results represent
an upper bound of the system performance because this simulation is conducted under the assumption
that the LMS weights converge to the optimum weights. In practice, these weights do not match the
actual values leading to a slightly reduction in detection performance.
In conclusion, we observe from the simulation results and as expected the use of smart antennas
improves the system performance. For example, for a probability of detection P d = 0.8 and with
an M = 5 antenna elements we have a 6.73 dB improvement in SNR/chip as compared to the single
antenna element (M = 1) case. Therefore, when an acquisition DS-CDMA employs the proposed PN
code acquisition with a smart antenna, any user can transmit a signal with a smaller power than with
a single antenna case.

5. CONCLUSION

In this paper, we considered the problem of serial PN code acquisition using smart antennas and
adaptive thresholding CFAR based on ordered data variability index. An automatic censoring scheme
of an unknown number of interferences samples in Rayleigh fading multipath communication channels
was proposed. We called this new censoring CFAR processing ODV-CFAR. The detector employed
ordered statistics to determine and censor the unwanted samples in the reference window. A closed
form expression for the probability of false alarm was obtained while the performance of the proposed
system in terms of the probability of detection was assessed by means of Monte Carlo simulations. The
simulation results showed that the adaptive thresholding CFAR scheme proposed with smart antennas
enhanced seriously the robustness of PN code acquisition as compared to both the CA-CFAR and
OS-CFAR processors in multipath and MAI environments. A drawback, however, is the additional
computational load compared to conventional CFAR processors, but with the existence of powerful and
high-speed digital signal processors (DSP and FPGA) such computational load increase would not be
a major concern. Nevertheless, the simulation results showed that this novel approach of using smart
antenna with ODV-CFAR adaptive thresholding was able to produce robust performance regardless of
its surrounding environment.

ACKNOWLEDGMENT

The authors are grateful to the anonymous reviewers for their constructive and valuable comments that
led to the enhancement of the quality of the paper.

APPENDIX A.

In this appendix, we derive the statistics of the self-interference, the multiple access interference MAI
and the thermal noise random variables. The in-phase signal component in (5) due to the user of interest
is given by [28]
I α1l cos (∅1l )
YD,l = [∆ 1 RR (j, N + 1) + (T c − ∆1 )RR (j, N )] (A1)
RTc
In Equation (A1), ∆1 is a random variable uniformly distributed in [0, Tc ] and RR (j, N ) is the code
partial autocorrelation function of the desired user defined as
R−1
X
RR (j, N ) = C1,s+j C1,s+j+N (A2)
s=0
152 Berbra, Barkat, and Anou

The in-phase multiple access interference term can also be expressed as

√
I βαkl cos (∅kl ) h (k) (k)
i
YMAI ,k,l = ∆k RR (j, N + 1) + (T c − ∆k )RR (j, N ) (A3)
RTc
(k)
where β = PI /PR is the power of the interfering user to the desired signal ratio and RR (j, N + 1) is
the partial crosscorrelation between the received sequence and the locally generated code of the desired
user, which can be expressed as
R−1
X
(k)
RR (j, N ) = Ck,s+j C1,s+j+N (A4)
s=0
Finally, the noise term is determined by
Z RTc µ ¶
I 1 I Tc
Nm = √ nm (t)C1 t − j dt (A5)
P R RTc 0 2
The quadrature phase components can be obtained from the in-phase components if we replace cos(·)
by sin(·) in all the previous equations.
Note that since the acquisition system can acquire only one path at any time, the other (L − 1)
path signals of the interest user inflict self interference to the aligned path. However, the term of the
user of interest can be viewed
 as 
L−1
X¡ L−1
X¡
 I ¢ 3 ¢
Y + Y I  = αkl cos (∅kl ) + I
YD,k (A6)
 D,l D,k 
4
k=0 k=0
k6=l k6=l
Since αkl are assumed to be independent and identically distributed (I.I.D) Rayleigh random variables
with a probability density function (PDF) given by (2) and ∅kl are uniformly distributed in [0, 2π),
then the first term in (A6) is a Gaussian RV [8] with mean zero and variance equal to
"µ ¶2 #
3 9σf2
E αkl cos (∅kl ) = (A7)
4 32
From (A1), the second term in (A6) is modelled as a Gaussian random variable with mean zero. We
can show that the variance of each element of (L − 1) path is
h 2
i σf2
E (Y ID,l ) = (A8)
3R
The multiple access interference term given by (A3) can also be approximated as a Gaussian random
variable with zero mean and variance equal to
¡ I ¢ βσf2
Var YMAI ,k,l = (A9)
3R
Finally, the in-phase noise component given by (A5) is also a Gaussian random variable with mean zero
and variance
h¡ ¢ i 1
I 2
E Nm = (A10)
2RSchip
where Schip = PR Tc /N0 represents the signal to noise ratio per chip SNR/chip.
Similarly, the statistics of the quadrature-phase components can also obtained following the above
approach.
When the lth resolvable path of the received signal is aligned with the local PN code (Hypothesis
H1 ), the in-phase and quadrature-phase components of term 1 in Equation (5) are Gaussian RVs with
mean zero and variance
 2 
 I L−1
X¡ ¢  9σf2 (L − 1) σ 2f
 
E YD,l + I  
YD,k   = + (A11)
32 3R
k=0
k6=l
Progress In Electromagnetics Research B, Vol. 57, 2014 153

I
YMAI ,k given by (8a) are Gaussian RVs with mean zero and variance
h¡ ¢2 i Lβσ 2f
I
E YMAI ,k = (A12)
3R
The term 3 is also a Gaussian RV with mean zero and variance given by (A10):
³ 2´ 1
I
E Nm = ; m = 1, 2, . . . , M. (A13)
2RSchip

APPENDIX B.

In this appendix, we present an outline of the proof of Equation (28). A formula relates the threshold
multiplier Tk to the desired probability of false alarm Pfa after obtaining the appropriate density of
Q. For a population of N c (I.I.D) random variables {q1 , q2 , . . . , qN c }, let {Z 1 , Z2 , . . . , ZNc −k be the
remaining set of variables upon censoring the k largest samples of {qi ; i = 1; N c. Since qi are I.I.D.,
it follows that Zi , i = 1, . . . , Nc − k are also I.I.D. From [8], the ith rank ordered random variable has
probability density function
Nc !
gi (z) = [FZ (z)]i−1 [1 − F Z (z)]Nc −i fZ (z) (B1)
(Nc − i)! (i − 1)!
where FZ (z) and fZ (z) are the cumulative probability distribution and the probability density
respectively, common to the I.I.D. N c random variables prior to ranking given by
µ ¶ µ ¶
1 −z −z
fZ (z) = exp and FZ (z) = 1 − exp (B2)
M 2σ 20 M 2σ 20 M 2σ 20
Since a particular Zi is equally likely to have any rank from 1 to N c − k, its probability density function
hk (z) is obtained by averaging (B1) over the uniform distribution of the rank i of Zi to yield

NX
c −k
1
hk (z) = gl (z) (B3)
Nc − k
l=1

Using (B1) and after some mathematical manipulations, we obtain the expression of the common
probability density function of N c − k RVs after censoring k samples as [30]
( k−1 µ ¶ )
Nc X Nc −1
hk (z) = fZ (z) 1 − Fz (z)Nc −m−1 [1 − F Z (z)]m (B4)
(Nc − k) m
m=0

Then using Laplace transform of equation of (B4), the MGF ΦQ (·) is found to be
  Nc −k
 N −m−1µ
k−1 NcX
X ¶µ ¶ 
c  1 Nc −1 Nc− m−1 j 1 
ΦQ (ω) = − (−1) (B5)
(Nc −k) 1+M 2σ 20 ω m j M 2σ 20 ω+m+j +1 
m=0 j=0
±
Setting ω=Tk M 2σ 20 and substituting Equation (B5) in (27) the probability of false alarm is obtained
as
  Nc −k
 −m−1 µ
k−1 NcX
X ¶µ ¶ 
Nc  1 Nc −1 Nc− m−1 1 
pf a = − (−1)j (B6)
 (Nc − k) 1 + Tk m j Tk + m + j + 1 
m=0 j=0
154 Berbra, Barkat, and Anou

REFERENCES

1. Sofwan, A. and M. Barkat, “PN code acquisition using smart sntennas and adaptive thresholding
trimmed-mean CFAR processing for CDMA communication,” Spring World Congress on
Engineering and Technology, SCET 2012, 1–4, China, May 2012.
2. Polydoros, A. and C. L. Weber, “A unified approach to serial search spread-spectrum code
acquisition — Part I: General theory,” IEEE Transactions on Communications, Vol. 32, No. 5,
542–549, May 1984.
3. Linatti, J., “On the threshold setting principles on code acquisition of DS-SS signals,” IEEE Journal
on Selected Areas in Communications, Vol. 18, No. 1, 62–72, Jan. 2000.
4. Kim, C. J., H. J. Lee, and H. S. Lee, “Adaptive acquisition of PN sequences for DSSS
communications,” IEEE Transactions on Communications, Vol. 46, No. 8, 993–996, Aug. 1998.
5. Kim, C. J., “Adaptive acquisition of PN code in multipath fading mobile channels,” Electronics
Letters, Vol. 38, No. 2, 135–137, Jan. 2002.
6. Benkrinah, S. and M. Barkat, “An adaptive acquisition using order statistic CFAR in DS-CDMA
serial search for a multipath Rayleigh fading channel,” Proceedings of the Third IEEE International
Conference on Systems, Signals and Devices, Tunisia, 2005.
7. Finn, H. M. and R. S. Johnson, “Adaptive detection mode with threshold control as a function of
spatially sampled clutter level estimates,” RCA Review, Vol. 29, 414–464, Sep. 1968.
8. Barkat, M., Signal Detection and Estimation, 2nd Edition, Artech House, Boston, MA, 2005.
9. Kedia, D., “Comparative analysis of peak correlation characteristics of non-orthogonal spreading
codes for wireless systems,” International Journal of Distributed and Parallel Systems (IJDPS),
Vol. 3, No. 3, 63–74, May 2012.
10. Zhang, X., X. Gao, and Z. Wang, “Blind paralind multiuser detection for smart antenna CDMA
system over multipath fading channel,” Progress In Electromagnetics Research, Vol. 89, 23–38,
2009.
11. Godara, L. C., Smart Antennas, 1st Edition, CRC Press LLC, New York, NY, 2004.
12. Wang, B. and H. M. Kwon, “PN code acquisition using smart antenna for spread-spectrum wireless
communications — Part I,” IEEE Transactions on Vehicular Technology, Vol. 52, No. 1, 142–149,
Jan. 2003.
13. Wang, B. and H. M. Kwon, “PN code acquisition for DS-CDMA systems employing smart antennas
— Part II,” IEEE Transactions on Wireless Communications, Vol. 2, No. 1, 108–117, Jan. 2003.
14. Gandhi, P. P. and S. A. Kassam, “Analysis of CFAR processors in nonhomogeneous background,”
IEEE Transactions on Aerospace and Electronic Systems, Vol. 24, No. 4, 427–445, Jul. 1988.
15. Alhariqi, N., M. Barkat, and A. Sofwan, “Serial PN acquisition using smart antenna and censored
mean level CFAR adaptive thresholding for a DS/CDMA mobile communication,” 14th IEEE
International Conference on High Performance Computing and Communications (HPCC 2012),
1193–1198, Liverpool, UK, Jun. 2012.
16. Rickard, J. T. and G. M. Dillard, “Adaptive detection algorithms for multiple-target situations,”
IEEE Transactions on Aerospace and Electronic Systems, Vol. 13, No. 4, 338–343, Jul. 1977.
17. Ritcey, J. A., “Performance analysis of the censored mean-level detector,” IEEE Transactions on
Aerospace and Electronic Systems, Vol. 22, No. 4, 443–454, Jul. 1986.
18. Smith, M. E. and P. K. Varshney, “Intelligent CFAR processor based on data variability,” IEEE
Transactions on Aerospace and Electronic Systems, Vol. 36, No. 3, 837–847, Jul. 2000.
19. Himonas, S. D. and M. Barkat, “Automatic censored CFAR detection for nonhomogeneous
environments,” IEEE Transactions on Aerospace and Electronic Systems, Vol. 28, No. 1, 286–304,
Jan. 1992.
20. Barboy, B., A. Lomes, and E. Perkalski, “Cell-averaging CFAR for multiple-target situations,” IEE
Proceedings — Radar, Sonar and Navigation, Vol. 133, No. 2, 176–186, 1986.
21. Barkat, M., S. D. Himonas, and P. K. Varshney, “CFAR detection for multiple target situations,”
IEE Proceedings — Radar, Sonar and Navigation, Vol. 136, No. 5, 193–209, 1989.
Progress In Electromagnetics Research B, Vol. 57, 2014 155

22. Prastitis, L. A., J. Frank, and S. D. Himonas, “Automatic censored cell averaging CFAR detector in
nonhomogeneous clutter,” 1992 International Radar Conference, 218–221, Brighton, UK, Oct. 1992.
23. Lehtomäki, J., J. Vartiainen, M. J. Juntti, and H. Saarnisaari, “CFAR outlier detection with
forward methods,” IEEE Transactions on Signal Processing, Vol. 55, No. 9, 4702–4706, 2007.
24. Liu, P.-Z. and C.-Z. Han, “Distributed automatic censored cell-averaging CFAR detector,” Acta
Automatica Sinica, Vol. 35, No. 7, 903–910, 2009.
25. Proakis, J. G., Digital Communications, 3rd Edition, McGraw-Hill, New York, 1995.
26. Song, Y. S., H. M. Kwon, and B. J. Min, “Computationally efficient smart antennas for CDMA
wireless communications,” IEEE Transactions on Vehicular Technology, Vol. 50, No. 6, 1613–1628,
2001.
27. Buehrer, R. M., Code Division Multiple Access (CDMA), 1st Edition, Morgan & Claypool,
Blacksburg, Virginia, US, 2006.
28. Yang, L.-L. and L. Hanzo, “Serial acquisition of DS-CDMA signals in multipath fading mobile
channels,” IEEE Transactions on Vehicular Technology, Vol. 50, No. 2, 617–628, Mar. 2001.
29. Shin, O. S. and K. B. Lee, “Utilization of multipath for spread-spectrum code acquisition in
frequency-selective Rayleigh fading channels,” IEEE Transactions on Communication, Vol. 49,
No. 4, 734–743, Apr. 2001.
30. Berbra, K., M. Barkat, and B. Atrouz, “Analysis of the CMLD in K-distributed clutter for fully
correlated/uncorrelated texture,” Proceedings of the 1999 International Radar Conference, Brest,
France, May 1999.
31. Rohling, H., “Radar CFAR thresholding in clutter and multiple target situations,” IEEE
Transactions on Aerospace and Electronic Systems, Vol. 19, No. 4, 608–621, Jul. 1983.