IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO.

6, JUNE 2001 1103

Adaptive Narrowband Interference Rejection

in DS-CDMA Systems: A Scheme of Parallel
Interference Cancellers
Wen-Sheng Hou, Li-Mei Chen, and Bor-Sen Chen, Senior Member, IEEE

Abstract—In the presence of narrowband interference (NBI), a good performance. However, if the interfering signal is strong
the performance of the direct-sequence code-division multiple ac- enough, the reliable communication becomes impossible even
cess communication system can be improved by using various in- with the advantage of spectrum spreading. Besides, due to the
terference cancellation schemes. In this paper, a scheme of parallel
interference cancellers (PICs), which includes a set of NBI cancel- electromagnetic environment in the real world, the processing
lation filters and a set of multiple access interference (MAI) can- gain is often insufficient to ensure the desired interference rejec-
cellation filters, is developed to improve the signal-to-interference tion. Therefore, the presence of narrowband interference (NBI)
ratio (SIR) and handle the MAI effect. In order to avoid perfor- in DS/SS systems is an inevitable problem. In these cases, the
mance deterioration due to unreliable initial detection, a robust interference immunity can be improved significantly by using
coefficient is introduced in our proposed design criterion. A re-
cursive least-square algorithm is employed to update the tap coeffi-
signal processing techniques which can complement the spread
cients of the PICs. In other words, no training or retraining period spectrum modulation.
is needed in our proposed PICs. Finally, the SIR performance of Several methods have been proposed to suppress the inter-
the proposed detector is analyzed and compared to that of the con- ference in DS/SS transmission systems [5]. A number of these
ventional linear detector. methods have utilized transversal suppression filters to remove a
Index Terms—CDMA, multiuser, PIC. large portion of the interference prior to despreading [6]–[8]. In
[6], the prewhitening transversal filters are designed according to
the Wiener filtering and the maximum entropy methods, in which
I. INTRODUCTION several past samples of the received signal are used to estimate
the current value of the interference. In [7], a two-side transversal
T HE CDMA system implemented by the direct-sequence
spread spectrum (DS/SS) technique is the most promising
multiplexing technology for cellular telecommunications ser-
filter is introduced so that both the past and future samples of the
incoming signal are used to predict the interference.
vices, such as personal communications, mobile telephony, and The weakness of these interference rejection approaches lies
indoor wireless networks. The advantages of DS/SS for these in ignoring the distortion of the SS signal due to the whitening.
services include superior operation in multipath environments, This weakness has been addressed in [8], where an adaptive re-
flexibility in the allocation of channels, the ability to operate ceiver is introduced by Iltis et al. In their study, a linear predic-
asynchronously, privacy, and increased capacity in burst or tion filter is cascaded with a matched filter to reduce the distor-
fading channels [3]. One of the most attractive features of direct- tion of the SS signal. First, an estimate of the transmitted signal is
sequence code-division multiple access (DS-CDMA) systems is subtracted from the received signal and an estimated interference
the capability of sharing bandwidth with narrowband communi- is obtained. Then, this estimated interference is used to modify
cation systems without performance degradation of any one of the estimate of the transmitted signal. In [9] and [15], an interfer-
the systems. ence suppression filter and a correlator have been combined into
It is well known that the processing gain of a spread spec- a single transversal structure. The transversal filter employs both
trum system will provide the system with a sufficient capability the code spreading and the interference estimation.
of interference rejection [1], [2], [4]. The higher the processing All the detectors mentioned above suppress NBI but do not
gain is, the better interference rejection the system has [4], [5]. take the multiple access interference (MAI) effect into consid-
For a moderate level of narrowband interference, the processing eration. Even though the CDMA system uses the orthogonal
gain of the spread spectrum system often provides sufficient ca- spread codes and transmits the signals through a synchronous
pability of interference rejection; hence, a simple despreading channel, the MAI effect is induced owing to the NBI suppres-
correlator [an optimum design in the case of additive white sion. A minimum mean square error (MMSE) linear detector
Gaussian noise (AWGN) channel] can be employed to achieve has been presented in [10], where both NBI and MAI are taken
into account. This technique is based on the linear MMSE al-
gorithms [11], [13]–[16] for multiuser detection. Although the
Manuscript received August 16, 1999; revised May 5, 2000 and November 10,
2000. This work was supported by the National Science Council under Contract MMSE linear detector can suppress both the MAI and NBI, it
NSC 87-2218-E-007-032. may result in noise enhancement and has some resistance loss
The authors are with the Department of Electrical Engineering, National for the near–far effect.
Tsing-Hua University, 300, Hsin Chu, Taiwan, R.O.C. (e-mail: bschen@moti.
ee.nthu.edu.tw). This paper concentrates on developing an adaptive detector
Publisher Item Identifier S 0733-8716(01)01904-7. to reduce the effects of both NBI and MAI and utilizes the pro-
0733–8716/01$10.00 © 2001 IEEE
1104 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 6, JUNE 2001

Fig. 1. Block diagram of a DS-CDMA transmission system.

cessing gain with less noise enhancement. We propose a scheme signals. Each user is assigned a different pseudonoise (PN)
of parallel interference cancellers (PICs) which consists of a set sequence, i.e., signature waveform, which directly modulates
of NBI cancellation filters and a set of MAI cancellation fil- the source signal. Then, the received signal can be modeled as
ters to suppress the NBI and cancel the MAI simultaneously.
For practical consideration, a robust coefficient is introduced in (1)
our proposed design criterion to avoid performance deteriora- where denotes the NBI signal, denotes the channel
tion due to the unreliable initial decision. An initial decision noise, and
is made by the NBI cancellation filters, and the result is then
passed through the MAI cancellation filters to make the final de-
cision. Actually, the proposed NBI cancellation filters not only (2)
suppress the NBI but also cancel the MAI. The weighting of
MAI cancellation performed by NBI cancellation filters is con- is the spread spectrum signal of the users, where denotes
trolled by the robust coefficient. the th user’s energy, is the th transmitted
According to the performance analysis, the proposed adap- symbol of the th user, is the signature waveform, and
tive detector significantly improves the signal-to-interference is the symbol period. The th signature waveform is given by
ratio (SIR) performance of the DS-CDMA systems over the con-
ventional linear detector. From the simulation results, the pro- (3)
posed adaptive detector also shows its superior ability to cancel
the NBI and MAI simultaneously and has better bit-error-rate
(BER) performance. where , for , is the th element
The remainder of the paper is organized as follows. The of the th user’s spreading sequence, is the chip waveform,
system description is addressed in Section II. An adaptive is the chip interval, and is the processing gain.
detector design is given in Section III. The proposed PICs is Without loss of generality, the signature waveforms are normal-
derived by the minimum-variance unbiased (MVU) and MMSE ized as follows:
design methods herein. For practice, an RLS algorithm is used
to realize the PICs. In Section IV, the performance analysis for for (4)
SIR improvement is presented. Simulation results are presented
in Section V. Finally, the conclusions are given in Section VI. For practical purposes, the “one-shot” detector is considered,
that is, the decision is made whenever a symbol period of the
signal , as shown in Fig. 1, is received. In the demodulation
II. SYSTEM DESCRIPTION
of the DS-CDMA systems, two kinds of interference, i.e., the
Consider a baseband synchronous DS-CDMA system wideband MAI due to the other users and the narrowband
depicted in Fig. 1, wherein users simultaneously and interference, must be minimized or cancelled to improve the
synchronously transmit the binary phase-shift keying (BPSK) system performance.
HOU et al.: ADAPTIVE NARROWBAND INTERFERENCE REJECTION IN DS-CDMA SYSTEMS 1105

Fig. 2. Proposed detector: a scheme of parallel interference cancellers.

III. ADAPTIVE DETECTOR DESIGN signal consists of the unpredictable spread spectrum signal, the
channel noise, and the predictable NBI, the NBI can be sub-
This paper concentrates on developing an adaptive detector
tracted from the received signal by using a prediction error filter
design for DS-CDMA systems such that the NBI is suppressed
or a transversal filter with two-sided taps [7]. The whitened
and the MAI is cancelled with less noise enhancement. As
signal is then despread by the desired user’s signature. In the
shown in Fig. 2, our proposed nonlinear detector is different
transform domain (frequency domain), a notch filter can be
from the decision-feedback equalizer (DFE). In DFE, the
used as an equalizer of the NBI channel [5]. The performance
decision results fed into the feedback filter result in error
is improved by the NBI whitening filter. However, a drawback
propagation whenever the decision is incorrect. However, only
of these two conventional methods is that the MAI is induced
the received signal is utilized, hence, there is no error
due to the whitening filters or/and the nonorthogonal spreading
propagation in our proposed adaptive detector. Moreover, no
signatures.
training or retraining period is needed in our proposed method.
In this paper, the PICs are developed to handle the NBI
The reason will be revealed in the following.
problem together with the MAI problem. As shown in Fig. 2,
A. Parallel Interference Cancellers (PICs) the proposed PICs consist of a set of NBI cancellation filters
and a set of MAI cancellation filters. In the first stage
The proposed detector is depicted in Figs. 1 and 2. As shown of the PICs, the NBI cancellation filters act as a combination
in Fig. 1, the received signal is passed through a chip- of an NBI whitening filter and a despreading correlator. In the
matched filter and then sampled at chip interval . The second stage of the PICs, the MAI cancellation filters act as
sampled signal is blocked as an -vector , given by an MAI canceller. Hence, the PICs can suppress the NBI and
cancel the MAI via a parallel structure by an optimal criterion.
Assume that the th user is the desired user. In the first stage
(5) of the PICs as shown in Fig. 2, the block received signal
is passed through the th NBI suppression filter , then
(6)
(7)
where is the normalized
signature sequence vector of the th user, is the sampled
vector of the NBI which is assumed to be wide-sense sta-
tionary with zero mean and covariance matrix , and
is the sampled vector of the white Gaussian channel noise (8)
with zero mean and covariance matrix , where is an iden-
tity matrix. It is assumed that , , and are mutu- is the output of the th NBI cancellation filter, where is
ally independent. a (time-varying) vector for the th user, i.e.,
The conventional strategy for the NBI rejection is focused
on whitening the received signal . Since the received (9)
1106 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 6, JUNE 2001

It is noted that the first term on the right-hand side of (8) denotes where denotes the variance operation, and is a ro-
the desired signal, the second term denotes the MAI, and the bust coefficient which weights the penalty of the initial decision
third term denotes the NBI. errors.
If the NBI cancellation filters suppress the NBI completely, The cost function represents the sum of the variance of
the output signal of the th NBI cancellation filter con- the initial estimation and the variance of the final esti-
tains only the desired signal and the MAI. Then, an initial deci- mation , for . Hence, the design criterion
sion is made as , given by is to minimize with respect to and , for
.
(10) Remark 1: The conventional strategy to overcome the NBI
where denotes the sign function. and MAI problem is to whiten the received signal . Then, its
In the second stage, the MAI cancellation filters utilize the MVU design criterion is to minimize the following cost function
initial decisions , for , to cancel the MAI.
For the th desired user, the initial decisions , for (17)
, are passed through the MAI can-
cellation filters , where the subscript denotes the th subject to for (18)
desired user and denotes the input of the
th MAI cancellation filter from the th initial decision. The de- Then
cision is made in the second stage of PICs as follows: (19)
(11) is the reconstruction symbol, which is just the initial decision of
then our proposed PICs.
Remark 2: In our proposed design method, not only the
(12) NBI whitening (cancellation) filters are utilized but the
MAI cancellation filters are also introduced. Moreover,
where a robust coefficient is introduced to avoid the incorrect
initial decisions and eliminate the MAI effect which may result
from the NBI whitening and/or the nonorthogonal signature
(13) waveforms.
and The cost function can be represented as a function of
and , given by

(14)

We have proposed a set of NBI cancellation filters to suppress

the NBI at the first stage of the PICs. Since these NBI whitening (20)
filters may induce MAI and/or the spreading signature may be
nonothonognal, the second stage of the PICs is then proposed to subject to for (21)
cancel the MAI. In the next subsections, a robust design criterion
is proposed to derive the tap coefficients of the PICs, i.e., the tap The Lagrangian multipliers approach is utilized to solve this
coefficients of the NBI cancellation filter and the MAI constrained optimization problem. Thus
cancellation filter , for , are derived via the
MVU method and MMSE method in sequence.

B. MVU Design Method

We have proposed a two-stage scheme of PICs for the
DS-CDMA systems. In the first stage, a set of NBI cancellation
filters are developed to make the initial decisions. In the second
stage, a set of MAI cancellation filters are developed to cancel (22)
the MAI which is induced by the NBI cancellation filters and
make the final decisions. If the initial decisions are incorrect, where denotes the expectation operation, and , for
then the final decisions are wrong as well. To overcome this , are the Lagrangian multipliers. Take the partial deriva-
difficulty, we introduce a robust coefficient and define a tives of with respect to and and let
constrained cost function as
and (23)
(15)
Then, the optimal NBI cancellation filters and the optimal MAI
subject to for (16) cancellation filters are derived in the following theorem [under
HOU et al.: ADAPTIVE NARROWBAND INTERFERENCE REJECTION IN DS-CDMA SYSTEMS 1107

the assumption of perfect decision, i.e., the decisions are the MMSE design method [3], and compare the MMSE method
correct]. with the MVU method.
Theorem 1 (MVU Design Method): According to the design Define two error vectors as
criterion in (15), the optimal PICs are derived as [21]

..
. (31)
(24)

(25)
..
. (32)

..
. where . According to the same
reasons mentioned in Section III-B, the MSE cost function is
(26) defined as

.. (33)
.
where is also the robust coefficient. Take the partial
derivative with respect to and and let
and

and (34)

Then, the optimal NBI cancellation filters and the MAI cancel-
lation filters are derived in the following theorem.
(27) Theorem 2 (MMSE Design Method): According to the de-
sign criterion in (33), the optimal PICs are derived as
where

(28) (35)

and
..
. (36)

(29) where

.. (37)
.

Remark 3: The difference between the MVU method and

and MMSE method is a constant scale . There-
fore, these two design methods have the same performance for
(30) all phase-shift keying (PSK) signals.

D. Adaptive Implementation
Since the matrices , , and in Theorems 1 and
2 cannot be easily obtained, and the NBI may be time varying,
C. MMSE Design Method the optimization of the NBI cancellation filters and the
The MMSE method has been used widely to handle the prob- MAI cancellation filters must be performed adaptively.
lems of equalization, deconvolution, estimation, detection, etc. In this subsection, an RLS algorithm is employed to search the
In this subsection, we investigate the PICs in Section III-A via filters’ tap coefficients.
1108 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 6, JUNE 2001

From (7) and (11), output of the NBI cancellation filters and
and output of the MAI cancellation filters are written as
Final decision: (45)

.. respectively.
.
Remark 4: The adaptation rule specified in (40)–(42) differs
from the conventional RLS algorithm, since the distortion due
to the NBI rejection is compensated by the proposed MAI can-
.. cellation filters. Moreover, no training or retraining period is
.
needed. In the other words, we propose a blind RLS adaptation
algorithm together with the scheme of PICs.
Remark 5: In the optimization strategy, we assume that the
.. .. NBI is wide-sense stationary; however, in overlay applications,
. the NBI often happens to be a data-like signal, i.e., a nonsta-
.
tionary signal with period correlation, unless the bit rate of the
NBI is an integer multiple of that of the CDMA system [22]. Al-
.. .. though the RLS algorithm provides the NBI correlation by time
. . average, it would produce noise enhancement, as compared to
the true correlation.
(38)
IV. DISCUSSION OF ROBUST COEFFICIENT
where and
A. Performance Analysis
. It can be shortened as
In this section, the SIR of our proposed detector is compared
to that of the conventional linear detector [10] which is a special
(39)
case of our proposed detector (i.e., only the NBI cancellation
filters are employed but the MAI cancellation filters are elimi-
where , is the known input matrix, and nated).
is the state vector to be estimated. Let SIR denote the SIR of the initial estimation , and
An adaptive solution to Theorems 1 and 2 is summarized in SIR denote the SIR of the final estimation . When
the following theorem. and the MAI cancellation filters are eliminated, the proposed
Theorem 3 (Adaptive Filtering): An RLS algorithm pro- detector is reduced to the conventional linear detector [10]. Let
posed to update the tap coefficients and estimate the us denote SIR as the SIR of the output of the conventional
symbol vectors is derived as linear detector.
According to the inverse lemma [10]

(40) (46)
(41)
(42) Then, (25) and (27) can be simplified as

where is the forgetting factor, and , , are (47)

represented as follows:
and

.. (48)
.
respectively. Substituting into (8), two SIRs (i.e., SIR
and SIR ) are obtained as follows

(43)
SIR (49)
Proof: See Appendix I.
After the tap coefficients of the PICs are estimated via the
proposed RLS algorithm, the final estimation and the final deci-
sion of the transmitted signal are obtained by the proposed PICs, (50)
given by

Final estimation: (44)

HOU et al.: ADAPTIVE NARROWBAND INTERFERENCE REJECTION IN DS-CDMA SYSTEMS 1109

Method 1: From extensive simulations, we have found that

SIR (51) if the SIR does not exceed SIR by 2.5 dB, the initial decision
is reliable. Thus, in order to clarify how to select an ade-
(52) quate , a design procedure can be outlined as follows.
Step 1) Choose a large to let SIR approach SIR , with
The signal-to-noise ratio (SNR) SIR at the output of the con- less erroneous initial symbol estimation as in the
ventional linear detector is written as previous analysis.
Step 2) Decrease the value of so that SIR is not larger
than SIR by 2.5 dB.
SIR (53) Step 3) Increase the value of if SIR exceeds SIR by
2.5 dB (a large error initial decision is made). Oth-
where erwise, decrease the value of .
Method 2: If the initial decision is reliable, is the
(54) best selection. In this situation, the NBI cancellation filter sup-
presses the NBI only, and the MAI cancellation filters cancel
the MAI. However, the output of NBI cancellation filters con-
We compare the performance of the proposed detector to that of tains the overall MAI, and the initial decision may be unreliable.
the conventional linear detector as Hence, we can increase the value of to join the MAI cancel-
lation in the NBI cancellation filters for improving the initial
SIR
(55) decision.
SIR
From extensive simulations, a rule of thumb for the selection
Substituting (30) into the denominator of (55), we ob- of the robust coefficient is given in the following proposition.
tain (56) as shown at the bottom of the page. Sine Proposition 1: Based on our simulation results, a selection
is positive definitive, of always provides a good performance.
(56) can be written as The simulation result for the selection of is given in Figs. 4
and 9. Its demonstration is given in Section V.
SIR
SIR V. SIMULATION RESULTS
(57) In this section, two simulation examples are presented to il-
Hence, the proposed detector always performs better than the lustrate the performance of the proposed adaptive detector in
conventional linear detector. DS-CDMA systems. Five conventional detectors for NBI rejec-
tion are also simulated to be compared to our proposed detector.
B. Choice of Robust Cofficient The five conventional detectors are given as follows (hereafter
In our proposed design criteria, the cost function has an en- referred to as 1–5).
ergy tradeoff between the initial estimation and the final 1) LPF (linear prediction filter) suppression receivers [8],
estimation . In our proposed PICs, the robust coefficient which suppress NBI by whitening the received signal (non-
controls the cancellation weighting of NBI and MAI between parametric method).
the NBI and MAI cancellation filters. When , the NBI 2) LPF suppression receivers, wherein an NBI model is given
cancellation filters perform NBI suppression only, and the (parametric method) [8].
MAI cancellation filters perform MAI cancellation only. As 3) The matched receiver with PN (pseudonoise) scrambling
increases, the NBI cancellation filters perform both NBI and (scramblers and descramblers), as shown in Fig. 3(a). The
MAI cancellation, and the MAI cancellation filters perform the lengths of the PN codes used to encode the users’ BPSK signal
residual MAI cancellation. Hence, the selection of in (15) are all 2047 [18].
and (33) is very important. An appropriate selection of can 4) The matched receiver with an LPF and a cascade of PN
give a better BER performance. In practice, we cannot get a descrambling, as shown in Fig. 3(b).
prior knowledge of . Since the NBI may be time varying in 5) The adaptive detector [10] by using linear filters.
the CDMA systems, it is impossible to get an optimal value of In the following simulations, a Butterworth spectrum Gaussian
. The following two methods are proposed for the adequate jammer is used to imitate the NBI. It is generated by a Butter-
choice of . worth filter of order eight driven by a white Gaussian noise. The

SIR
(56)
SIR
1110 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 6, JUNE 2001

Fig. 3. Matched receivers with PN scrambling.

A. Simulation I
In this example, a synchronous DS-CDMA system with five
active users is considered. All active users have the same signal
power, i.e., dB; jammer-to-signal ratio (JSR
) is 12 dB. The Walsh codes are used as spreading
codes, and , i.e., the chip number in a symbol interval
is 16. Fig. 4(a) shows the ratio of SIR over SIR with respect
to , where SIR is the SIR of the output of the proposed MAI
cancellation filters and SIR is the SIR of the output of the pro-
posed NBI cancellation filters. SIR represents the performance
of the proposed detector and SIR represents the performance
of the initial decision of the proposed detector.
The scenario revealed in Fig. 4(a) can be explained as fol-
lows. If , the NBI cancellation filters suppress the NBI
only, and contains the overall MAI. In this situation, the
Fig. 4. Simulation results for with JSR = 12 dB in Simulation I. initial decision is unreliable and the performance is poor
[see Fig. 4(b)]. When the value of increases, the NBI cancel-
filter is a bandpass one with center frequency at 1 radian and lation filters gradually perform the MAI cancellation apart from
bandwidth 0.25 radian. performing the NBI cancellation. Therefore, the initial decision
The tap lengths of the proposed NBI and MAI cancellation is becoming correct as the SIR increases. When is larger and
filters are chosen as and , respectively, for all users. The larger, the NBI cancellation filters gradually lose the ability of
tap lengths of LPFs and the adaptive linear filters of the conven- NBI suppression; then the SIR performance degrades and a
tional detectors 1–5 are all chosen as . In order to illustrate the larger amount of noise is enhanced.
BER performance of the detectors, the average value of BERs Fig. 4(b) shows the BER performance versus of the above
is defined as system by Monte Carlo simulation. The BER performance of
the initial decision is inversely proportional to . On the
BER other hand, the BER performance of the final decision is
very good when is between 0.2 and 0.5. It is noted that the
BER (58) BER of the initial decision is the BER of the conventional
linear detector [10] as . We conclude that the proposed
where BER is the th user’s BER. The forgetting factor is adaptive detector achieves a superior BER performance in the
chosen as 0.999 for the adaptive algorithms. final decision if the initial decision is reliable.
HOU et al.: ADAPTIVE NARROWBAND INTERFERENCE REJECTION IN DS-CDMA SYSTEMS 1111

^ (n) versus both and JSR.

Fig. 5. BER performance of the final decision B Fig. 7. BER performance versus both the number of users and E =N . The
solid line denotes the linear filter, the dashed line denotes the proposed detector.

Fig. 6. Comparison of BER among different detectors with = 0:4 in

Simulation I. Fig. 8. Convergence and tracking capability of the proposed detector.

Fig. 5 shows the BER performance of the final decision of detector 3 is poor if JSR is high. This discussion is also con-
versus both and JSR with dB. From Figs. 4 and firmed by our simulation results, as shown in Fig. 6.
5, it is concluded by the simulations that a good performance The detector 4 can cancel the jammer; hence, it has a better
performance than that of detectors 1–3. However, owing to the
is achieved as for a good balance between the correct
same reason that the LPF distorts the spread spectrum signal, it
initial decision and the SIR enhancement.
is not good enough to combat the interference in CDMA sys-
Fig. 6 shows the comparison of BER performance to JSR
tems.
among the five conventional detectors and the proposed detector
The adaptive detector 5 which uses linear filters not only sup-
by Monte Carlo simulation. The signal power is fixed at presses the NBI but also cancels the MAI. However, this de-
18 dB. It is shown that the performances of detectors 1 and 2 are tector is just a special case of our proposed detector for ,
poor since the LPFs only handle the NBI suppression. There- and is so regardless of the MAI cancellation filters. Our pro-
fore, if the spectrum of the spread signal is not flat enough, the posed PICs suppress the NBI and cancel the MAI simultane-
LPF seriously distorts the spread spectrum signal. Apart from ously. Moreover, a robust coefficient is introduced to ade-
the above reason, the LPF receivers do not consider the MAI; quately tune the burdens of the NBI and MAI cancellation filters
hence, their performances are worse. such that the proposed detector has a superior performance over
The performance of detector 3 is better than that of detectors 1 other conventional detectors, as shown in Fig. 6. This JSR is
and 2 because the descrambler spreads the jammer’s spectrum, fixed at JSR dB, the system performance versus both the
and then a matched filter with despreading sequence is utilized number of users and is shown in Fig. 7.
to achieve a better performance. If the JSR is low, the perfor- Fig. 8 shows the convergence/tracking rate of the proposed
mance of detector 3 is good. However, since the jammer is not adaptive detector in a dynamic environment. At time ,
cancelled but only spread as a white-like noise, the performance the center frequency of the jammer changes from 1 to 2 radians.
1112 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 6, JUNE 2001

Fig. 9. Simulation results for in Simulation II. Fig. 11. BER performance versus the near–far effect.

Fig. 10. Comparison of BER among different detectors in Simulation II. Fig. 12. BER performance versus the near–far effect.

It is presented that the RLS algorithm can track the changing adaptive detector 5 can handle the MAI problem. However, our
of environment more rapidly than the least-mean-square (LMS) proposed detector has a better performance than that of detector
algorithm (step size 0.002; see Appendix II). In this case, each 5. The reason is that our proposed detector consists of a set of
data point in the plot is the average of 100 simulations. NBI cancellation filters to suppress the NBI, and a set of MAI
cancellation filters to cancel the MAI. Fig. 11 shows the BER
B. Simulation II performance versus the number of users and in the case
of JSR dB.
In this simulation, the nonorthogonal codes are used as the
spreading codes to further observe the ability of the MAI can-
cellation of our proposed detector. In this case, , , C. Simulation III
dB, JSR dB, and all active users have the The last simulation addresses the sensitivity of the proposed
same signal power. Because the spreading codes are nonorthog- detector to the near–far effect. For this, the transmitter power
onal, a large quantity of MAI is presented in this CDMA system. of reference user (user 1) is kept constant, while the transmitter
Fig. 9 shows the ratio of SIR over SIR with respect to . power of other users changes from 5 to 5 dB with respect
Although there is a larger quantity of MAI in this case, our pro- to the transmitter power of the reference user. Figs. 12 and 13
posed detector can efficiently suppress NBI and MAI. show the BER performance for the orthogonal code system
Fig. 10 shows the comparison of the BER performance to JSR and nonorthogonal code system, respectively, in the case of
of the proposed detector with the five conventional detectors. dB and JSR dB. It is evident from the
A robust coefficient is chosen according to the result BER of the reference user’s results that the proposed detector
of Fig. 9. Since a large quantity of MAI is presented in this has good near–far resistance, whereas the conventional MF
CDMA system, detectors 1–4, which are not designed to handle detector and single-user detector are severely affected by near–
the MAI, have poor performances even though JSR is low. The far effect.
HOU et al.: ADAPTIVE NARROWBAND INTERFERENCE REJECTION IN DS-CDMA SYSTEMS 1113

where is the observation interval. The optimiza-

tion of the tap-coefficient vector can be obtained by min-
imizing the cost function. Take the partial derivation of
with respect to and let

(A.2)

After rearranging

(A.3)

Then, the optimal estimate is given by

Fig. 13. BER performance versus both the number of users and E =N . The (A.4)
solid line denotes the linear filter, the dashed line denotes the proposed detector.
The correlation matrix is defined as

VI. CONCLUSION (A.5)

In this paper, a novel adaptive detector which consists of a set where

of NBI cancellation filters and a set of MAI cancellation filters
is developed to mitigate the NBI and MAI in CDMA systems (A.6)
by using a scheme of parallel interference cancellers. We have
shown that our proposed method performs much better than the (A.7)
conventional interference rejection methods since the distortion
resulting from the NBI rejection is taken into consideration.
Because a robust coefficient is introduced in our proposed de- Since , according to (A.4), we get
sign criterion, the performance is improved due to the more re-
(A.8)
liable initial decision. In the first stage of PICs, the NBI cancel-
lation filters are employed for whitening the NBI. Then, in the and
second stage of the PICs, the MAI cancellation filters are em-
ployed to cancel the MAI. However, because the optimization (A.9)
of the NBI and MAI cancellation filters is performed simultane-
ously, the burdens of the NBI and MAI cancellations are shared Then, the correlation matrix in (A.5) can be expressed as
in them. In other words, the NBI cancellation filter may cancel the following recursive algorithm:
the MAI if necessary. Hence, the proposed adaptive detector is
(A.10)
robust and blind (in a less strict sense, since the spreading codes
are a priori known). The following matrix inverse lemma is useful for the derivation.
We have shown that the proposed adaptive detector can sup- Lemma 1 [17]: Let , , be positive-definite matrices,
press the NBI and cancel the MAI simultaneously. It achieves a related by . The inverse matrix can
significant improvement over the conventional detectors. More- be expressed as follows: .
over, a larger number of active users in the CDMA system with
narrowband signals can simultaneously share the channel band- Since the correlation matrix is positive definite and
width by using our proposed method. therefore nonsingular, the matrix inversion lemma can be ap-
plied to modify the recursive equation (A.10). Let ,
APPENDIX I , , and in
DERIVATION OF (40)–(42) Lemma 1; a recursive equation for the inversion of the corre-
lation matrix is obtained as
By using the abbreviation notations in (38) and (39), the cost
function in (22) can be reexpressed as

(A.11)
Let

(A.1) (A.12)
1114 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 19, NO. 6, JUNE 2001

then [11] U. Madhow and M. Honig, “MMSE interference suppression for direct-
sequence spread-spectrum CDMA,” IEEE Trans. Commun., vol. 42, pp.
3178–3188, Dec. 1994.
[12] E. Eleftheriou and D. D. Falconer, “Trackikng properties and steady-state
performance of RLS adaptive filter algorithm,” IEEE Trans. Acoust.,
Speech, Signal Processing, vol. ASSP-34, pp. 1097–1109, Oct. 1986.
[13] M. Abdulrahman and A. U. H. Sheikh, “Decision feedback equalization
for CDMA in indoor wireless communications,” IEEE J. Select. Areas
(A.13) Commun., vol. 12, pp. 698–706, May 1994.
[14] S. L. Miller, “An adaptive direct-sequence code-division multiple-access
According to (A.12) and (A.13), we can rewrite (A.11) as fol- receiver for multiuser interference rejection,” IEEE Trans. Commun.,
lows: vol. 43, pp. 1556–1565, Feb./Mar./Apr. 1995.
[15] C. N. Pateros and G. J. Saulnier, “An adaptive correlator receiver
(A.14) for direct-sequence spread-spectrum communications,” IEEE Trans.
Commun., vol. 44, pp. 1543–1552, Nov. 1996.
[16] P. B. Rapajic and B. S. Vucetic, “Adaptive receiver structures for asyn-
chronous CDMA systems,” IEEE J. Select. Areas Commun., vol. 12, pp.
APPENDIX II 685–697, May 1994.
DERIVATION OF THE LMS ALGORITHM [17] J. M. Mendel, Lessons in Digital Estimation Theory. Englewood
Cliffs, NJ: Prentice-Hall, 1987.
The cost function can be expressed as [18] D. G. Yee, “The design and implementation of a CDMA direct sequence
spread-spectrum transceiver,” M.S. thesis, Univ. California, Berkeley,
Dec. 1996.
[19] M. L. Honig, U. Madhow, and S. Verdu, “Blind adaptive multiuser de-
(A.15) tection,” IEEE Trans. Inform. Theory, pp. 944–960, July 1995.
[20] M. K. Tsatsanis and Z. Xu, “Performance analysis of minimum variance
Taking the gradient of with respect to gives CDMA receivers,” IEEE Trans. Commun., vol. 46, pp. 3014–3022, Nov.
1998.
(A.16) [21] S. M. Kay, Statistical Fundamentals of Signal Processing: Estimation
Theory. Englewood Cliffs, NJ: Prentice-Hall, 1993.
[22] H. V. Poor and X. Wang, “Code-aided interference suppression for
Multiplying (A.16) by , and let , we can get DS/CDMA communications—Part I: Interference suppression capa-
bility,” IEEE Trans. Commun., vol. 45, pp. 1101–1111, Sept. 1997.
(A.17)
since ; hence
Wen-Sheng Hou was born in Taiwan, R.O.C., on
(A.18) November 8, 1967. He received the B.S. degree from
the Department of Electrical Engineering, Tatung
Therefore, the adaptive rule is Institute of Technology, Taiwan, R.O.C., in 1991,
and the M.S. degree from the Department of Elec-
(A.19) trical Engineering, National Tsing-Hua University,
Hsin-Chu, Taiwan, in 1993. He is currently working
toward the Ph.D. degree in electrical engineering at
National Tsing-Hua University.
His research interests are in signal processing and
(A.20) communication.

where denotes the step size.

Li-Mei Chen was born in Taiwan, R.O.C., in 1969.
REFERENCES She received the B.S. degree from the Department of
[1] J. G. Proakis, Digital Communications. New York: McGraw-Hill, Electrical Engineering, National Tsing-Hua Univer-
1983. sity, Taiwan, R.O.C., in 1991, and the M.S. degree
[2] B. Widrow and S. D. Stearns, Adaptive Signal Processing. Englewood from the Department of Electrical Engineering, Na-
Cliffs, NJ: Prentice-Hall, 1985. tional Taiwan University, Hsin-Chu, Taiwan, in 1994.
[3] W. S. Hou and B. S. Chen, “Adaptive detection in asynchronous code- She is currently working toward the Ph.D. degree in
division multiple-access system in multipath fading channels,” IEEE electrical engineering at National Tsing-Hua Univer-
Trans. Commun., vol. 48, pp. 863–874, May 2000. sity.
[4] R. L. Pickholtz, D. L. Schilling, and L. B. Milstein, “Theory of spread Her research interests are in digital signal
spectrum communications: A tutorial,” IEEE Trans. Commun., vol. processing and CDMA communication.
COMM-30, pp. 855–884, May 1982.
[5] L. B. Milstein, “Interference rejection techniques in spread-spectrum
communications,” Proc. IEEE, vol. 76, pp. 657–671, June 1988.
[6] F. M. Hsu and A. A. Giordano, “Digital whitening techniques for im- Bor-Sen Chen (M’82–SM’89) received the B.S.
proving spread spectrum communications performance in the presence degree from Tatung Institute of Technology, Taiwan,
of narrowband jamming and interference,” IEEE Trans. Commun., vol. R.O.C., in 1970; the M.S. degree from National
COMM-26, pp. 209–216, Feb. 1978. Central University, Taiwan, in 1973; and the Ph.D.
[7] L. M. Li and L. B. Milstein, “Rejection of narrow-band interference degree from the University of Southern California,
in PN spread-spectrum systems using transversal filters,” IEEE Trans. Los Angeles, in 1982.
Commun., vol. COMM-30, pp. 925–928, May 1982. He was a Lecturer, Associate Professor, and
[8] R. A. Iltis and L. B. Milstein, “Performance analysis of narrow-band
Professor at Tatung Institute of Technology from
interference rejection techniques in DS spread-spectrum systems,” IEEE
1973 to 1987. He is now a Professor at National
Trans. Commun., vol. COMM-32, pp. 1169–1177, Nov. 1984.
[9] G. J. Saulnier, P. K. Das, and L. B. Milstein, “An adaptive digital sup- Tsing-Hua University, Hsin-Chu, Taiwan. His current
pression filter for direct-sequence spread-spectrum communications,” research interests include control signal processing
IEEE J. Select. Areas Commun., vol. SAC-3, pp. 676–686, Sept. 1985. and communication.
[10] H. V. Poor and X. Wang, “Code-aided interference suppression for Dr. Chen received the Distinguished Research Award from the National Sci-
DS/CDMA communications—Part II: Parallel blind adaptive implemen- ence Council of Taiwan four times. He now holds a Chair of the Foundation of
tations,” IEEE Trans. Commun., vol. 45, pp. 1112–1122, Sept. 1997. Outstanding Scholar.