422 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO.

2, MARCH 2005

Accurate Bit-Error Rate Evaluation for Synchronous

MC-CDMA Over Nakagami- -Fading Channels m
Using Moment Generating Functions
Qinghua Shi and Matti Latva-Aho, Member, IEEE

AbstractIn the bit-error rate (BER) analysis of code-division nature of multicarrier modulation makes it especially attractive
multiple-access (CDMA) systems, a Gaussian approximation is for broadband communications [3]. Secondly, with advanced
widely used to tackle the multiple access interference (MAI), signal processing techniques, CDMA has the potential to pro-
although it does not always offer satisfactory accuracy. This paper
investigates the BER performance of synchronous multicarrier vide larger capacity than other multiple access schemes. Some
(MC) CDMA systems over Nakagami- -fading channels in a other important advantages of MC-CDMA include: 1) it is band-
different way. We present an accurate and unified BER analysis width efficient [4] because of the use of orthogonal frequency di-
for synchronous MC-CDMA systems. To facilitate our analysis, vision multiplexing (OFDM) [5]; 2) modulation/demodulation
we assume a synchronous uplink, whose BER performance can can be implemented efficiently by IFFT/FFT algorithms [6]; and
be intuitively viewed as a lower BER bound of the more realistic
asynchronous MC-CDMA. The basic idea is that, by using the (3) direct frequency diversity in MC-CDMA allows us to com-
GaussChebyshev quadrature (GCQ) rule to perform inverse bine all the received signal energy scattered in the frequency
Laplace transform, an accurate BER can be numerically obtained domain [7], [8], whereas a RAKE receiver in direct-sequence
from the moment generating function (MAG) of the output de- (DS)-CDMA usually exploits only part of the received multi-
cision variable at a receiver, without any assumption about the path energy. It has been shown that MC-CDMA can outperform
MAI distribution. First, signals on all subcarriers of MC-CDMA
systems are assumed to experience independent fading. Two DS-CDMA in both the downlink [7], [8] and uplink [9] envi-
standard diversity combining techniques, equal gain combining ronments.
(EGC) and maximal ratio combining (MRC), are employed. The Bit-error rate (BER) is one of the most important perfor-
BER performance in both downlink and synchronous uplink is mance measures for communication systems and the BER anal-
analyzed. We then consider a more general system model, in which
ysis of various systems has been studied extensively. To ana-
signals on different subcarriers undergo correlated fading. The
asymptotic (error floor) performance of downlink MC-CDMA lyze the BER performance of MC-CDMA systems, including
with MRC is studied. In particular, we investigate the effects of MC-CDMA [2], [9], MC-DS-CDMA [10], [11], and multitone
spreading sequences and the delay spread of the channel on the CDMA [12], the multiple access interference (MAI) is com-
system performance. Numerical examples are provided to show monly assumed to be Gaussian distributed. However, the ac-
the main results of this paper. The accuracy of the GCQ and MGF
curacy of this approach depends on the system configuration,
based solution is verified by different approaches such as Monte
Carlo integration and the exact residue method. In addition, the especially on the number of users and their powers. It is well
accuracy of the commonly used Gaussian approximation is also known that the Gaussian approximation is not accurate when
examined. the number of users is small or the power deviation among dif-
Index TermsGaussChebyshev quadrature rule (GCQ), mo- ferent users is significant. Therefore, an accurate BER analysis
ment generating function, multicarrier code-division multiple-ac- without such a Gaussian approximation is always desirable.
cess (MC-CDMA), Nakagami- fading.
To avoid the Gaussian approximation of the MAI, the BER can
be calculated in transform domains. Two widely used transforms
I. INTRODUCTION are the Fourier and Laplace transforms, corresponding to the
characteristic function (CF) and moment generating function
A S A combination of multicarrier (MC) modulation [1] and
code-division multiple-access (CDMA), MC-CDMA [2]
benefits from both techniques. First, the parallel transmission
(MGF), respectively. The basic idea is that first the CF or MGF
of the decision variable is derived, then an associated inverse
transform is performed to calculate the BER. Since the decision
Manuscript received April 14, 2002; revised December 1, 2003; accepted
December 1, 2003. The editor coordinating the review of this paper and ap- variable can be completely and exactly characterized by its CF
proving it for publication is Qi Bi. This work was supported by Nokia, Elek- or MGF, the BER can be accurately evaluated via numerical
trobit, the Academy of Finland, the Finnish National Technology Agency, and integration techniques. For this reason, CF- and MGF-based
by the Finnish Defense Forces. This paper was presented in part at the Inter-
national Zurich Seminar on Broadband Communications, Zurich, Switzerland, methods have received considerable attention. The CF method
February 1921, 2002. was used to study the BER performance of DS-CDMA with
Q. Shi was with the Centre for Wireless Communications, University of Oulu, random sequences in Rayleigh- [13] and Nakagami-fading [14]
Oulu FIN-90014, Finland, and the Department of Electronic Engineering, City
University of Hong Kong, Kowloon, Hong Kong. He is now with the Positioning channels. In [15], a saddlepoint integration-based MGF approach
and Wireless Technology Centre, Nanyang Technological University, 637553 was proposed to compute error probabilities due to intersymbol
Singapore (e-mail: [email protected]). and cochannel interference. This approach has been applied to
M. Latva-Aho is with the Centre for Wireless Communications, University
of Oulu, Oulu FIN-90014, Finland. study the performance of DS-CDMA systems over Rician- [16]
Digital Object Identifier 10.1109/TWC.2004.842984 and Nakagami-fading [17] channels, and to analyze the BER
1536-1276/$20.00 2005 IEEE
SHI AND LATVA-AHO: ACCURATE BER EVALUATION FOR SYNCHRONOUS MC-CDMA 423

performance of MC-DS-CDMA systems [18]. In [19] and [20],

a GaussChebyshev quadrature (GCQ) rule-based MGF method
was developed to compute the BER over fading channels, and
recently it has been applied in various applications [21][23].
In this paper, our goal is to present an accurate and unified BER
analysis of synchronous MC-CDMA over Nakagami- -fading
channels using the GCQ based MGF method. Instead of approx-
imating the MAI by a Gaussian random variable (RV), we derive
an exact statics of the decision variable by means of the MGF
formulation. The BER is then computed by GCQ with satisfac-
tory accuracy. Both downlink and uplink are considered, but we
assume a synchronous uplink to facilitate our analysis, which
can be intuitively viewed as a lower BER bound of the more
practical asynchronous MC-CDMA. Note that the synchronous
uplink, albeit an idealized model, is useful in the sense that it
makes the exact BER analysis mathematically tractable and
allows us to gain insight into the performance of a more realistic
asynchronous system, and in some special cases synchronous
or quasisynchronous uplink is still possible. Other contribu-
tions of this work include that a general Nakagami- (for any Fig. 1. MC-CDMA transmitter and receiver of user k .
0.5)-fading channel model is studied, and both indepen-
dent and correlated fading models for different subcarriers are where
considered. ;
The organization of this paper is as follows. Section II de- power of data bits identical for all users;
scribes the system model, including the transmitter, the channel, the th data bit of user ;
and the receiver. In Section III, the GCQ-based MGF approach rectangular pulse defined in ;
is elaborated. In Section IV, we present the BER analysis of bit duration;
synchronous MC-CDMA under the assumption of independent the th subcarrier (angular) frequency;
fading for different subcarriers. Both downlink and uplink with random carrier phase of user ;
equal gain combining (EGC) or maximum ratio combining radio (angular) frequency;
(MRC) are considered. In Section V, with a correlated-fading an integer determining the subcarrier spacing.
channel model, the asymptotic (also known as error floor) per-
formance of downlink MC-CDMA with MRC is investigated. B. Channel Model
Section VI provides some numerical examples and discussions, A slowly varying fading channel, whose parameters are un-
followed by conclusions summarized in Section VII. changed over one bit duration, is assumed. Since Nakagami dis-
tribution [24] is a versatile yet relatively simple statistical model
II. SYSTEM MODEL and it is well suited to characteristic a radio mobile channel [25],
A. Transmitter we use a Nakagami- distributed RV to represent the ampli-
tude of channel gains. By properly selecting the number of sub-
Consider an MC-CDMA system [2] with users and sub- carriers, we assume that signals on each subcarrier experience
carriers. A synchronous uplink is assumed. We will only de- flat fading. The channel gain for the th subcarrier of user
scribe the uplink, but it is straightforward to tailor the presenta- can be modeled by where is Nakagami- dis-
tion to the downlink case. Fig. 1 illustrates the block diagrams tributed with ( denotes expectation) and
of the transmitter and receiver of user . It should be noted that uniformly distributed over . Specifically, has the
Fig. 1 is shown for illustration purposes only. In practice, mul- following probability density function (pdf)
ticarrier modulation/demodulation can be implemented by in-
verse fast Fourier transform (IFFT)/FFT algorithms. In addi-
tion, if the data rate is high, a serial-to-parallel converter can be
used before the multicarrier modulation [9]. Assume that binary (2)
phase-shift keying (BPSK) modulation and binary spreading se-
quences (denoted by for user ) are employed. The where is a parameter which reflects the severity of
transmitted signal of user may be expressed as channel fading and is the gamma function [26]. Letting
is gamma distributed with a pdf given by

(3)

(1) For MC-CDMA systems, an important issue to consider is the

correlation between subcarriers. This is due to the fact that an
424 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 2, MARCH 2005

MC-CDMA system may be subject to correlated fading for dif- where we define
ferent subcarriers if the subcarrier spacing is less than the coher-
ence bandwidth of the channel. In this paper, both independent
and correlated fading-channel models are considered. In Sec-
tion IV, we assume independent fading for different subcarriers
that is valid if the subcarrier spacing of an MC-CDMA system is
larger than the coherence bandwidth of the channel [2],
whereas in Section V, we assume correlated fading for different
subcarriers that is suited for the standard MC-CDMA system
[7][9] in which the subcarrier spacing is minimized
to maintain high bandwidth efficiency. Although, for the stan-
dard MC-CDMA system, the correlated fading model is more
Assuming is transmitted and omitting a factor
appropriate, the independent fading channel model is still valu-
, we have
able because it produces BER results that can serve as a bench-
mark for an MC-CDMA system over a more realistic (corre-
lated) fading channel. For example, [27] and [28] show that
the correlation among subcarriers deteriorates the BER perfor- (8)
mance of an MC-CDMA system in the uplink. Therefore, the
independent fading channel model will lead to BER results that where is uniformly distributed over ,
can be regarded as a lower BER bound for the correlated fading and the AWGN term on the th subcarrier has zero mean and
model. variance

C. Receiver (9)
All users are assumed to be synchronized throughout this
paper. The received signal at the base station is given by where is the bit energy and is referred to as
the signal-to-noise (SNR) ratio.

III. BASIC METHODS

The GCQ based MGF approach is elaborated in this section.
(4) A Monte Carlo integration method is also provided to verify the
MGF approach.
where and is an additive white Gaussian
noise (AWGN) with zero mean and double-sided power spectral A. MGF Approach
density . The MGF and CF of the decision variable are defined as
A coherent correlation receiver with EGC or MRC is em-
ployed. Perfect channel estimation for the desired user is as-
sumed. Without loss of generality, we consider the decision vari- MGF
able of the zeroth data bit of user CF (10)

Clearly, the MGF and CF correspond to the two-sided Laplace

(5) transform and inverse Fourier transform, respectively. As stated
in [29, p. 196], in the region of convergence of (where the
two-sided Laplace transform of exists), one can get the MGF
where for EGC and for MRC. The com-
from the CF with a variable change .
ponent of the decision variable on the th subcarrier can be ex-
As the Fourier transform often leads to a simpler analysis, we
pressed as
will start from the CF formulation and then obtain the corre-
sponding MGF.
(6) Theoretically, the BER of user can be acquired by the fol-
lowing formula [19], [20]:

where and are the desired signal and the AWGN term on
the th subcarrier, respectively, and is the MAI from (11)
user and on the th subcarrier. Correspondingly, the decision
variable can be written as where the parameter is chosen to ensure the convergence
of the above integral. The remaining problem is how to calculate
(7) this complex contour integral.
SHI AND LATVA-AHO: ACCURATE BER EVALUATION FOR SYNCHRONOUS MC-CDMA 425

1) Exact Solution (Residue Calculation): An exact solution IV. BER FOR INDEPENDENT FADING
is possible if residues of can be calculated. According
In this section, independent and identically distributed (i.i.d.)
to [19, eq. (6)]
fading for different subcarriers is assumed. The independence
assumption, which can be guaranteed by a proper selection of
(12) , enables us to treat the components of the decision variable
on each subcarrier separately and independently.

where the residues are calculated and summed over the poles A. Downlink
whose real parts are positive. In some special cases where
In a downlink, signals from different users are transmitted
is simple enough, an exact BER can be derived via (12),
through the same channel, the subscript of the channel pa-
but in general, it is rather difficult as explained in [19] and [20].
rameter is, therefore, omitted. For simplicity, we define
2) Numerical Solution (GCQ): Numerical integration tech-
niques can be used to solve the inverse Laplace transform
problem. There are two numerical methods available in the
(16)
literature. One is based on the saddlepoint integration together
with the trapezoidal rule [15][18], another is the GCQ-based
approach [19][23]. In this paper, we adopt the second ap- 1) EGC : When EGC is employed
proach, which requires us to evaluate at a limited
number of complex points
(17)

The AWGN has zero mean and variance

(13)
(18)
where is the number of nodes used for GCQ. The BER of user
can be computed by Since the AWGN is independent of the desired signal and the
MAI, the CF of the decision variable conditioned on
can be expressed as
(14)

where the error term will vanish as gets large enough, i.e.,
a convergent will be observed when increasing . Note (19)
that in (14), GCQ with an even number of nodes is applied,
but only half of the nodes are used in the BER computation with
because is an even function and an odd
function with respect to . Finally, it should be pointed out that (20)
the parameter determines the convergence speed of the GCQ
approach and an optimum value of is generally difficult to
obtain, but different values of produce the same convergent and invoking (59), we get
. Our numerical experience indicates that is
usually a good choice.

B. Monte Carlo Integration Method

(21)
From (5)(9), we know that the decision variable is
a Gaussian RV, with mean and variance ,
conditioned on or, equivalently, conditioned on where is the hypergeometric function [26]. The condi-
tional MGF of is then given by
, and . Accordingly, the
conditional BER of user is

(15)

The conditioning on , and can be

(22)
removed by Monte Carlo integration.
426 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 2, MARCH 2005

Applying the GCQ rule described by (13) and (14) gives the B. Uplink
conditional BER . As are i.i.d. RVs taking In an uplink, it is reasonable to assume that all users expe-
on values from with equal probabilities, the uncondi- rience i.i.d. fading. Moreover, i.i.d. fading for each subcarrier
tional BER is is also assumed. These assumptions greatly simply the CF and
MGF derivations. Particularly, we can treat each subcarrier and
(23) each user independently.
1) EGC : The desired signal on the th subcarrier
and the MAI on the th subcarrier from user are given by

Since different users may have different BERs, an appropriate (30)

performance measure for the whole system is the BER averaged
over all users, i.e., Since the AWGN with variance given by (18), the desired
signal component , and the MAI component are
mutually independent, we have the following CF expression:

which is used throughout this paper.

2) MRC : When MRC is employed
(31)
(24)
where and are given by (20) and (59), respec-
The AWGN term has zero mean and variance given by (9). tively. Applying (57) gives the conditional CF of
Since, given and and are condi-
tionally independent, it follows that the conditional CF of
can be expressed by
With BPSK modulation and binary spreading sequences, the
above equation reduces to
(25)
(32)
where
where the conditioning on is automatically removed.
Equation (32) indicates that the MAI is not associated with
spreading sequences used if binary spreading sequences are
(26)
employed. In other words, the code orthogonality among users
Averaging (25) over , whose pdf is given by (3), yields is lost in this case. Finally, the MGF of is given by

(27)

The CF of the decision variable conditioned on is

(33)

is readily obtained by using the GCQ method.

2) MRC : The desired signal on the th subcarrier
(28)
and the MAI on the th subcarrier from user are
and the corresponding MGF is
(34)
and the AWGN on the th subcarrier has zero mean and variance
given by (9). Note that , and are conditionally
(29) independent, given . It follows that

It is straightforward to apply the GCQ rule to compute the con-

ditional BER of the user . Then, can be ob- (35)
tained by using (23).
SHI AND LATVA-AHO: ACCURATE BER EVALUATION FOR SYNCHRONOUS MC-CDMA 427

with frequencies and can be modeled in the frequency do-

main by two complex RVs and , where and
are Nakagami- distributed with
and are random phases uniformly distributed over .
(36) From [24, eq. (126)] or [33, eq. (1)], the joint pdf of and
is given by
and from (57) we have

(37)
(40)
where, again, BPSK modulation and binary spreading se-
quences are assumed, and the conditioning on is automat- where is the th order modified Bessel function of the
ically removed. Letting , the CF of conditioned first kind [26] and is the correlation coefficient between
on can be written as and .
If Rayleigh fading is considered, we let and denote
two corresponding Rayleigh-fading envelopes. According to
[34, eqs. (1.5)(22)], the joint pdf of and is

(38)

Averaging (38) over (return to the MGF formulation) results (41)

in
with , and

(42)

As (40) reduces to (41) under the condition , the relation

(39)
(43)
where . This integral has a closed-form solution for
the case of [30]. In general, it can be evaluated by holds. Another parameter of interest is the covariance of and
using numerical integration techniques such as the Gauss-La- , which is given by [24, eq. (127)]
guerre quadrature. The interested reader is referred to [31], [32]
(44)
for details. In particular, [31] provides the associated Laguerre
polynomials and the weights explicitly. Letting , we define an covariance matrix
Due to the assumption of independent fading for different by
subcarriers, the MGF of is given by .
We can use the GCQ method presented in Section III to evaluate
the BER.
(45)
V. BER FOR CORRELATED FADING
In the previous section, each subcarrier is assumed to un- B. Error Floor of Downlink MC-CDMA With MRC
dergo independent fading. This assumption does not hold uni-
When a correlated fading channel is considered, it is often
versally because it requires , with which the MC-CDMA
difficult to use the MGF method to study the BER performance
system is not bandwidth efficient. In general, MC-CDMA sys-
of an MC-CDMA system. The main problem is that the MGF
tems are subject to correlated fading for different subcarriers
of the decision variable is in general rather difficult to acquire
because the frequency spacing between adjacent subcarriers of
because we should deal with all the subcarriers jointly, whereas
an MC-CDMA system is usually minimized (i.e., ) to
for the independent fading channel model each subcarrier can
maintain high bandwidth efficiency. In this section, we will in-
be handled independently. Nevertheless, in a special case when
vestigate the BER performance of MC-CDMA over a correlated
downlink MC-CDMA with MRC is under consideration, the
Nakagami- -fading channel.
MGF of the decision variable can be obtained because of the
well-behaved multivariate gamma distribution. In fact, MRC
A. Correlated Channel Model diversity reception with correlated branches in Nakagami-
An omnidirectional antenna is considered. The channel is as- fading is well studied in [35], [36]. We extend these results
sumed to have an exponential multipath intensity profile with a to an MC-CDMA system that is more complicated due to the
delay spread . The channel gains on two different subcarrier presence of the MAI. To simplify our analysis, the AWGN is
428 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 2, MARCH 2005

neglected, since here we are interested in the asymptotic BER According to (12), the conditional BER of user is given by
(also known as error floor) performance which is still of great
importance for MAI limited MC-CDMA systems. (53)
With coherent detection and MRC, the decision variable can
be written as
2) Numerical Solution (for Any ): It is straight-
(46) forward to apply the GCQ method described by (13) and (14) to
compute the conditional BER . As the eigenvalues
are conditioned on through the matrix defined
where is defined in (16). The decision variable consists
by (48) and (16), the unconditional BER of user can be com-
of two kinds of RVs: and . Starting from [36] and
puted by averaging over as is described in
[37], we can derive the MGF of conditioned on (see
(23).
Appendix B for details)
We observe that the symmetric Toeplitz covariance matrix
describes the correlated Nakagami- -fading channel, while
(47) the diagonal matrix and the covariance matrix charac-
terize the MAI. This simplicity enables us to treat the MAI and
where are the eigenvalues of the matrix . is a diag- the Nakagami- fading strictly. The limitation of the proposed
onal matrix given by approach is that it is appropriate for systems with relatively
small number of users, because its computational complexity
(48)
increases exponentially with the number of users. However, it
From [36, eq. (10)], the matrix is related to the covariance is still useful in that it allows us to gain insight into the per-
matrix by formance of downlink MC-CDMA with MRC when different
spreading sequences, different correlation coefficients among
(49) subcarriers, and different fading severities are considered. Fur-
thermore, for the case of small number of users, the Gaussian
When independent fading among subcarriers is assumed, the approximation does not offer good accuracy and simulation is
covariance matrix as well as the matrix reduces to ( rather time consuming.
is assumed to be large enough) where is an identity
matrix. The eigenvalues of the matrix become VI. NUMERICAL RESULTS
. Substituting these eigenvalues into (47) leads to (29)
where is assumed. This is what we expect, since Numerical examples are provided in this section. We
the independent fading channel model is only a special case of consider three families of spreading sequences, namely,
the correlated fading channel model. WalshHadamard, Gold, and orthogonal Gold codes. The
Given the conditional MGF , we can calcu- WalshHadamard codes are generated from the Hadamard
late the BER conditioned on via two methods. matrix. The Gold codes of length 31 are generated from two
1) Closed-Form Solution (for Integer ): When the fading preferred maximal length sequences 45 and 67 in octal. The
parameter is an integer, we are able to derive a BER expres- orthogonal Gold codes [38] of length 32 are derived from their
sion in closed form. As in [36], (47) can be expressed as a sum corresponding Gold codes. Thirty-two subcarriers
of partial fractions are employed unless otherwise specified. We choose the fol-
lowing three values of the fading parameter to investigate the
performance of MC-CDMA in Nakagami- fading: ,
(50)
the most severe fading; , Rayleigh fading; and ,
an example of less severe (than Rayleigh) fading. In the case
with the coefficient determined by of independent fading, results of [2] are used for the Gaussian
approximation approach. Specifically, in the downlink we use
[2, eq. (19)] for EGC and [2, eq. (20)] for MRC, while in the
uplink we use [2, eq. (16)] for EGC and [2, eq. (18)] for MRC.
The independent fading channel model is considered first. In
(51)
the downlink, WalshHadamard codes are used to separate dif-
ferent users. In the uplink, all binary spreading sequences pro-
To obtain the coefficient explicitly, we need to compute the duce the same BER performance because the code orthogonality
higher order differentiation of a product function, which can be among users is destroyed. In Figs. 2 and 3, we compare the
done with a recursive algorithm provided in [36]. BER results obtained by the Monte Carlo integration, the MGF
Note that (47) contains poles of order at . method, and the Gaussian approximation. Fig. 2 shows that in
After some manipulations (Appendix C), we obtain the fol- the downlink EGC performs much better than MRC and there
lowing residue expression: are appreciable differences between the BER results from the
Gaussian approximation and from other methods. Fig. 3 shows
(52) that in the uplink MRC outperforms EGC, and the Gaussian ap-
proximation is not accurate enough for MRC but offers good
SHI AND LATVA-AHO: ACCURATE BER EVALUATION FOR SYNCHRONOUS MC-CDMA 429

Fig. 2. Downlink BER performance, evaluated by the Monte Carlo integration, Fig. 4. Downlink BER performance versus SNR for synchronous MC-CDMA
the MGF method, and the Gaussian approximation, versus SNR for synchronous m
m
with EGC over a Nakagami- fading channel. WalshHadamard codes are
MC-CDMA over a Rayleigh ( = 1)-fading channel. WalshHadamard codes employed (N = 32).
are employed (N = 32).

Fig. 5. Downlink BER performance versus SNR for synchronous MC-CDMA

Fig. 3. Uplink BER performance, evaluated by the Monte Carlo integration,
m
with MRC over a Nakagami- fading channel. WalshHadamard codes are
employed (N = 32).
the MGF method, and the Gaussian approximation, versus SNR for synchronous
m
MC-CDMA over a Rayleigh ( = 1)-fading channel (N = 32).

accuracy for EGC. In both Figs. 2 and 3 perfect agreement is

observed between the Monte Carlo integration and the GCQ
based MGF method, whereas the Gaussian approximation may
produce large discrepancies. These results demonstrate the ex-
cellent accuracy of the GCQ based MGF method.
Figs. 4 and 5 show the downlink BER performance of
MC-CDMA with EGC and MRC in Nakagami- fading.
Different values of the fading parameter are used to inves-
tigate their effect on the BER performance. Clearly, the BER
performance changes considerably with the fading severity
parameter . Compared with MRC, EGC appears to be more
sensitive to the change of .
The uplink BER results of MC-CDMA with EGC and MRC Fig. 6. Uplink BER performance versus SNR for synchronous MC-CDMA
in Nakagami- fading are displayed in Figs. 6 and 7. Similar to m
with EGC over a Nakagami- fading channel (N = 32).
the downlink case, the fading parameter influences the BER
performance significantly, especially when the number of user accuracy of the GCQ based MGF method. Fig. 8 also clearly re-
is small. Also, we can see that the fading parameter affects veals that increasing the value of the fading parameter (means
the BER performance more apparently for EGC than for MRC. less severe fading) yields better performance, as expected.
The correlated fading channel model is considered in the fol- In Fig. 9, we study the error floor performance of downlink
lowing. In Fig. 8, we compare the BER results obtained by MC-CDMA with MRC over a correlated Nakagami- fading
the residue calculation and the GCQ based approaches. Perfect channel when different families of spreading sequences are
agreement is observed. This again demonstrates the very good employed. It is evident that, under different fading conditions,
430 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 2, MARCH 2005

Fig. 7. Uplink BER performance versus SNR for synchronous MC-CDMA

m
with MRC over a Nakagami- fading channel (N = 32).

Fig. 9. Downlink error floor performance versus the number of users

for synchronous MC-CDMA with MRC over a correlated Nakagami-m
Fig. 8. Downlink error floor performance, evaluated by the residue and GCQ fading channel when different families of spreading sequences are employed
methods, versus the number of users for synchronous MC-CDMA with MRC (T = 0 :1 T ) .
m
over a correlated Nakagami- -fading channel. WalshHadamard codes, eight
subcarriers (N = 8), and T = 0:1T are used.
VII. CONCLUSION
orthogonal sequences (orthogonal Gold and WalshHadamard Using the GCQ based MGF approach, the BER performance
codes) always perform better than nonorthogonal sequences of synchronous MC-CDMA with EGC or MRC over a Nak-
(Gold sequences) and the WalshHadamard codes are the best agami- -fading channel was studied, without any approxima-
among these spreading sequences. The differences among them tion for the MAI. Both downlink and synchronous uplink, inde-
are significant when the number of user is relatively small pendent fading and correlated fading were considered. Numer-
or fading is not severe, but these differences decrease as the ical results show that the GCQ based MGF approach provides
number of users increases or fading becomes more severe. satisfactory accuracy and the fading parameter can greatly
Moreover, we see from Fig. 9 that the WalshHadamard codes affect the BER performance.
perform better than the orthogonal Gold sequences regardless When independent fading for different subcarriers is as-
of the severity of fading. This observation suggests that for sumed, it was found that EGC performs better than MRC in the
downlink MC-CDMA over frequency selective fading channels downlink but performs worse than MRC in the synchronous
WalshHadamard codes can keep more orthogonality (produce uplink. As expected, the Gaussian approximation is generally
less MAI) than the orthogonal Gold sequences. not accurate enough, especially when the number of users is
The effect of correlation between subcarriers on the error small. However, for the case of MC-CDMA with EGC in the
floor performance of downlink MC-CDMA with MRC over a synchronous uplink, the Gaussian approximation gives good
Nakagami- channel is investigated in Fig. 10, where the av- accuracy. Compared to MC-CDMA with MRC, MC-CDMA
erage error floor versus the number of users for different delay with EGC appears to be more sensitive to the change of the
spreads is shown. The delay spread of the channel is used as a fading parameter in both the downlink and synchronous
measure of the correlation between subcarriers. We can see that uplink.
the error floor performance improves as the delay spread of the The error floor performance of downlink MC-CDMA with
channel decreases (equivalently, the correlation between subcar- MRC over a correlated Nakagami- -fading channel was also
riers increases), regardless of the values of the fading parameter investigated. It was found that, regardless of the severity of
. This indicates that the correlation between subcarriers helps fading, WalshHadamard codes outperform both Gold and
to reduce the MAI for downlink MC-CDMA with MRC. orthogonal Gold codes, and the correlation between subcarriers
SHI AND LATVA-AHO: ACCURATE BER EVALUATION FOR SYNCHRONOUS MC-CDMA 431

where is the zeroth order Bessel function of the first kind

[26]. Averaging (56) over and invoking the integral formula
[39, p. 486, eq. (11.4.28)], we obtain

(57)

If (i.e., is a Rayleigh RV), (57) reduces to

(58)

corresponding to the CF of a Gaussian RV with zero mean and

variance , which is well known [40, p. 41, eq. (2-1-99)].
Compared to [14, eq. (16)], (57) is expressed by a confluent
hypergeometric function in a more compact form. More impor-
tantly, [14, eq. (16)] is only valid for an integer , whereas (57)
is valid for any real number .
For convenience, we list the CF of available in [41, p. 1735,
Table II]

(59)

APPENDIX II
MGF OF
It is well known that are gamma distributed pro-
Fig. 10. Downlink error floor performance versus the number of users for vided that are Nakagami- distributed [36], [37]. Further,
m
synchronous MC-CDMA with MRC over a correlated Nakagami- -fading
N
channel with different delay spreads. Thirty-two subcarriers ( = 32) and
a set of correlated gamma RVs can be represented by their
Walsh-Hadamard codes are used. joint CF [36], [37] as

is beneficial to the BER performance of downlink MC-CDMA

with MRC.
(60)

where denotes the determinant of a matrix, is an iden-

APPENDIX I tity matrix, is a diagonal matrix,
CF OF and the matrix is related to the covariance matrix by (49).
We derive the CF of where is Nakagami- dis- It follows that
tributed with and uniformly distributed over
. From the definition (10), the conditional CF of ,
given and , can be expressed as (61)
(54)
where and
Averaging (54) over yields
(62)

(55) with defined by (48) and (16).

Let . By replacing in (61)
where the integral representation [26, p. 376, eq. (9.6.16)] of with , the MGF of the decision variable conditioned on
the zeroth order modified Bessel function of the first kind is given by
is applied. Using the relation [26, p. 375, eq. (9.6.3)], (55) be-
(63)
comes
Assuming that are eigenvalues of , then
(56) are eigenvalues of . We obtain (47).
432 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 4, NO. 2, MARCH 2005

APPENDIX III REFERENCES

RESIDUE CALCULATION
[1] J. A. C. Bingham, Multicarrier modulation for data transmission: An
The residue of at the pole of order idea whose time has come, IEEE Commun. Mag., vol. 28, no. 5, pp.
, can be calculated by 514, May 1990.
[2] N. Yee, J. P. Linnartz, and G. Fettweis, Multi-carrier CDMA in in-
door wireless radio networks, IEICE Trans. Commun., vol. E77-B, pp.
900904, July 1994.
[3] N. Morinaga, M. Nakagawa, and R. Kohno, New concepts and tech-
nologies for achieving highly reliable and high-capacity multimedia
wireless communications systems, IEEE Commun. Mag., vol. 35, no.
1, pp. 3440, Jan. 1997.
[4] S. B. Slimane, Bandwidth efficiency of MC-CDMA signals, Electron.
Lett., vol. 35, pp. 17971798, Oct. 14, 1999.
[5] R. W. Chang, Synthesis of band-limited orthogonal signals for
multichannel data transmission, Bell Syst. Technol. J., vol. 45, pp.
17751796, Dec. 1966.
(64) [6] S. B. Weinstein and P. M. Ebert, Data transmission by frequency-di-
vision multiplexing using the discrete Fourier transform, IEEE Trans.
Commun. Technol., vol. COM-19, no. 5, pp. 628634, Oct. 1971.
where we use (50) and define [7] S. Hara and R. Prasad, Design and performance of multicarrier CDMA
system in frequency-selective Rayleigh fading channels, IEEE Trans.
Veh. Technol., vol. 48, no. 5, pp. 15841595, Sep. 1999.
[8] , Overview of multicarrier CDMA, IEEE Commun. Mag., vol. 35,
(65) no. 12, pp. 126133, Dec. 1997.
[9] X. Gui and T. S. Ng, Performance of asynchronous orthogonal mul-
ticarrier CDMA system in frequency selective fading channel, IEEE
Note that we only need to handle the terms Trans. Commun., vol. 47, no. 7, pp. 10841091, July 1999.
[10] E. A. Sourour and M. Nakagawa, Performance of orthogonal multicar-
rier CDMA in a multipath fading channel, IEEE Trans. Commun., vol.
(66) 44, no. 3, pp. 356367, Mar. 1996.
[11] S. Kondo and L. B. Milstein, Performance of multicarrier DS CDMA
systems, IEEE Trans. Commun., vol. 44, no. 2, pp. 238246, Feb. 1996.
in (64) because other terms vanish when we substitute [12] L. Vandendorpe, Multitone spread spectrum multiple access communi-
cations system in a multipath Rician fading channel, IEEE Trans. Veh.
into (64). Therefore, Technol., vol. 44, no. 2, pp. 327337, May 1995.
[13] J. Cheng and N. C. Beaulieu, Accurate DS-CDMA bit-error probability
calculation in Rayleigh fading, IEEE Trans. Wireless Commun., vol. 1,
no. 1, pp. 315, Jan. 2002.
[14] , Precise bit error rate calculation for asynchronous DS-CDMA
in Nakagami fading, in Proc. IEEE GLOBECOM, San Francisco, CA,
(67) Nov. 2000, pp. 980984.
[15] C. W. Helstorm, Calculating error probabilities for intersymbol and
cochannel interference, IEEE Trans. Commun., vol. COM-34, no. 5,
pp. 430435, May 1986.
Using the formula for higher order differentiation of a product [16] D. Liu, C. L. Despins, and W. A. Krzymien, Low-complexity per-
[26, p. 12, eq. (3.3.8)], we have formance evaluation of binary and quaternary DS-SSMA over Rician
fading channels via the characteristic function method, Wireless
Personal Commun., vol. 7, pp. 257273, Aug. 1998.
[17] S. W. OH and K. H. Li, Performance evaluation for forward-link cel-
lular DS-CDMA over frequency-selective Nakagami multipath fading
channels, Wireless Personal Commun., vol. 18, pp. 275287, Sep. 2001.
[18] B. Smida, C. L. Despins, and G. Y. Delisle, MC-CDMA performance
evaluation over a multipath fading channel using the characteristic func-
(68) tion method, IEEE Trans. Commun., vol. 49, no. 8, pp. 13251328,
Aug. 2001.
[19] E. Biglieri et al., Computing error probabilities over fading channels: A
It is easy to check that the terms in (68) become zero. unified approach, Eur. Trans. Telecommun., vol. 9, pp. 1525, Jan.Feb.
On the other hand, the terms satisfying also vanish 1998.
when we substitute into (68). Hence, we only need [20] E. Biglieri et al., Simple method for evaluating error probabilities,
Electron. Lett., vol. 32, pp. 191192, Feb. 1, 1996.
to consider the term for the residue calculation. It
[21] R. E. Ziemer et al., Effects of carrier tracking in RAKE reception of
follows that wide-band DSSS in Rician fading, IEEE Trans. Microwave Theory
Tech., vol. 47, no. 6, pp. 681686, June 1999.
[22] R. E. Ziemer and T. B. Welch, Equal-gain combining of multichannel
(69) DPSK in Doppler-spread Ricean fading, IEEE Trans. Veh. Technol., vol.
49, no. 5, pp. 18461855, Sep. 2000.
[23] G. Taricco and E. Biglieri, Exact pairwise error probability of space-
Substituting (69) into (67), we obtain (52). time codes, IEEE Trans. Inf. Theory, vol. 48, no. 2, pp. 510513, Feb.
2002.

ACKNOWLEDGMENT
m
[24] M. Nakagami, The -distributionA general formular of intensity
distribution of rapid fading, in Statistical Methods in Radio Wave Prop-
agation, W. C. Hoffman, Ed. New York: Pergamon, 1960.
The authors would like to thank the anonymous reviewers for [25] W. R. Braun and U. Dersch, A physical mobile radio channel model,
their constructive and helpful suggestions and comments. IEEE Trans. Veh. Technol., vol. 40, no. 2, pp. 472482, May 1991.
SHI AND LATVA-AHO: ACCURATE BER EVALUATION FOR SYNCHRONOUS MC-CDMA 433

[26] M. Abramowitz and I. A. Stegun, Handbook of Mathematical Functions [41] A. Annamalai, C. Tellambura, and V. K. Bhargava, Equal-gain diversity
with Formulas, Graphs, and Mathematical Tables, 4th ed. New York: receiver performance in wireless channels, IEEE Trans. Commun., vol.
Dover, 1970. 48, no. 10, pp. 17321745, Oct. 2000.
[27] J. Park et al., Performance of MC-CDMA systems in nonindependent
Rayleigh fading, in IEEE Proc. Int. Conf. Communications, vol. 1, Van-
couver, BC, Canada, June 610, 1999, pp. 506510.
[28] T. Kim et al., Performance of an MC-CDMA system with frequency
offsets in correlated fading, IEEE Proc. Int. Conf. Communications, vol. Qinghua Shi received the B.S. and M.S. degrees
2, pp. 10951099, June 2000. in electronic engineering from Nanjing University,
[29] E. Lukacs, Characteristic Functions, 2nd ed. London, U.K.: Griffin, Nanjing, China, in 1991 and 1994, respectively,
1970. and the Ph.D. degree in wireless engineering from
[30] Q. Shi and M. Latva-aho, Exact bit error rate calculations for syn- Southeast University, Nanjing, in 2000.
chronous MC-CDMA over a Rayleigh fading channel, IEEE Commun. From 1994 to 1996, he was a Research Engineer
Lett., vol. 6, no. 7, pp. 276278, July 2002. with the State Key Laboratory of Milimeter Waves,
[31] E. W. Weisstein. Laguerre-Gauss Quadrature. [Online]. Available: Nanjing, China. From 2000 to 2002, he was a Re-
http://mathworld.wolfram.com/Laguerre-GaussQuadrature.html; search Scientist with the Centre for Wireless Com-
http://mathworld.wolfram.com/LaguerrePolynomial.html munication, University of Oulu, Oulu, Finland. He is
[32] V. I. Krylov, Approximate Calculation of Integrals. New York: currently a Research Fellow with the Department of
Macmillan, 1962. Electronic Engineering, City University of Hong Kong, Hong Kong. His re-
[33] C. C. Tan and N. C. Beaulieu, Infinite series representations of search interests include spread spectrum, CDMA, multicarrier, spacetime mod-
m
the bivariate Rayleigh and Nakagami- distributions, IEEE Trans. ulation/coding.
Commun., vol. 45, no. 10, pp. 11591161, Oct. 1997.
[34] W. C. Jakes Jr., Ed., Microwave Mobile Communications. New York:
Wiley, 1974.
[35] P. Lombardo, G. Fedele, and M. M. Rao, MRC performance for bi-
nary signals in Nakagami fading with general branch correlation, IEEE Matti Latva-Aho (S96M98) received the M.Sc.
Trans. Commun., vol. 47, no. 1, pp. 4452, Jan. 1999. degree in electrical engineering, Lic.Tech., and
[36] Q. T. Zhang, Maximal-ratio combining over Nakagami fading with an Dr.Tech. degrees from the University of Oulu, Oulu,
arbitrary branch covariance matrix, IEEE Trans. Veh. Technol., vol. 48, Finland, in 1992, 1996, and 1998, respectively.
no. 4, pp. 11411150, July 1999. From 1992 to 1993, he was a Research Engineer
[37] , Exact analysis of postdetection combining for DPSK and NFSK with Nokia Mobile Phones, Oulu. From 1994 to
systems over arbitrary correlated Nakagami channels, IEEE Trans. 1998, he was a Research Scientist with the Telecom-
Commun., vol. 46, no. 11, pp. 14591467, Nov. 1998. munication Laboratory and Centre for Wireless
[38] B. M. Popovic, Efficient despreaders for multicode CDMA systems, Communications , University of Oulu. He was the
in Proc. Int. Conf. Universal Personal Communications (ICUPC), San Director of Centre for Wireless Communications,
Diego, CA, Oct. 1216, 1997, pp. 516520. University of Oulu, from 1998 to 2003. Since 2000,
[39] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Prod- he has been a Professor of digital transmission techniques at the Telecommuni-
ucts, 2nd ed. San Diego, CA: Academic, 1965. cations Laboratory. He has published more than 70 conference or journal papers
[40] J. G. Proakis, Digital Communications. NewYork: McGraw-Hill, in the field of CDMA communications. His research interests include future
1995. broadband wireless communication systems and related transceiver algorithms.