Analysis of the Performance Degradation in Uplink

MC-CDMA Systems with Doppler Induced

Frequency Offsets
Ivan Cosovic1 , Michael Schnell1 , Miroslav Dukic2 , Andreas Springer3
1
German Aerospace Center (DLR), Institute of Communications and Navigation,
82234 Wessling, Germany, Phone: +49-8153-282853,
Email: ivan.cosovic@dlr.de / michael.schnell@dlr.de
2
Department of Communications, Faculty of Electrical Engineering, University of Belgrade,
11120 Belgrade, Serbia and Montenegro, Phone: +381-11-3218-305,
Email: dukic@etf.bg.ac.yu
3
Institute for Communication and Information Engineering, University of Linz,
4040 Linz, Austria, Phone: +43-732-2468-9725,
Email: a.springer@icie.jku.at

Abstract— In this paper, uplink Time Division Duplex (TDD)
Multi-Carrier Code-Division Multiple-Access (MC-CDMA) sys-
tems applying pre-equalization to combat channel impairments
are considered. Especially, the performance degradation caused
by the Doppler induced frequency offsets among the different
mobile users is analyzed in detail. Based on this analysis different
frequency interleaving schemes are proposed and their potential
to reduce the performance degradation is investigated. It turns
out, that a frequency interleaver which is allowed to skip a few,
well-defined subcarriers is capable of reducing the influence of
Doppler induced frequency offsets significantly causing a very
little loss in bandwidth efficiency.
I. I NTRODUCTION
Multi-Carrier Code-Division Multiple-Access (MC-
CDMA) [1], [2] is well recognized as a very promising
candidate for the air interface of future mobile wireless
downlink systems. However, when considering MC-CDMA
for uplink transmission some problems arise. In particular,
channel estimation and equalization as well as time and
frequency synchronization are more difficult and require
more complex algorithms. One of the possibilities to
mainly overcome the problems of channel estimation and
equalization is to apply channel pre-equalization at the
MC-CDMA transmitter as described in [3], [4]. In this
paper, the influence of Doppler induced frequency offsets
on the performance of uplink time division duplex (TDD)
MC-CDMA applying pre-equalization at the transmitter
is analyzed in detail. Additionally, countermeasures based
on thoroughly designed frequency interleaving schemes
are proposed. Sensitivity of uplink MC-CDMA systems to
frequency offsets has been considered in several papers [5],
[6], but both, an in-depth analysis of the influence of
misaligned users on each other and proposals for reducing
the resulting performance degradation are not yet available.
Throughout the paper, a slowly varying time-variant channel
is assumed. Thus, channel pre-equalization techniques are
applicable to the uplink TDD/MC-CDMA transmission which
are based on the channel estimation results from the downlink
transmission [3], [4]. With TDD mode the effects of the
channel on the uplink signal can be pre-equalized at the mobile
station based on the channel estimation from the downlink
frame.
The desired user is assumed to be synchronized to the
receiver frequency, while all other users are misaligned in
frequency. Different frequency interleaving schemes are in-
vestigated and their performances are analyzed. A novel skip-
carrier interleaver is introduced which significantly improves
the performance of an uplink MC-CDMA system in the
presence of frequency offsets between the different users.
The paper is organized as follows. The TDD/MC-CDMA
uplink transmission system considered throughout this paper
is described in Section II. In Section III, both the influence
of misaligned users on each other is analyzed and frequency
interleaving schemes are investigated. The performance of the
uplink TDD/MC-CDMA system in the presence of Doppler
induced frequency offsets and the performance gains achieved
through the proposed skip-carrier interleaver are presented in
Section IV. Finally, Section V summarizes the results.
II. T RANSMISSION SYSTEM
A time synchronous uplink MC-CDMA system with K
active users is considered. Fig. 1 depicts the block diagram
of the considered uplink transmitter of mobile user k, k =
0, . . . , K − 1. The uplink transmitter is based on the MC-
CDMA system concept with M -Modification as proposed
in [2]. Note, the M -Modification allows each user to transmit
M data symbols simultaneously using an additional Orthog-
onal Frequency-Division Multiplexing (OFDM) component

IEEE Communications Society 0-7803-8533-0/04/$20.00 (c) 2004 IEEE

within the MC-CDMA system. Thus, each MC-CDMA sym- c (k ) G0( k ) ( k )
d 0( k ) s0( k ) v0
bol of user k consists of M complex data symbols d(k) =

INTERLEAVER
USER (k) x S/P x
.

FREQUENCY
(k) (k) . .

CYCLIC
(d0 , . . . , dM −1 ) obtained from the symbol mapper. An MC- (k )

PREFIX
d . . .

IFFT
S/P . P/S
CDMA symbol is formed in the following way. After serial to
(k )
c (k ) (k )
. ( k ) G N -1( k )
c .
parallel conversion (S/P) each symbol is spread with the same d M -1 s Nc -1 v N c -1
(k) (k)
x S/P x
user specific spreading sequence c(k) = (c0 , . . . , cL−1 ) of
length L. It is presumed that Walsh-Hadamard (WH) orthog-
Fig. 1. MC-CDMA uplink transmitter of mobile user k.
onal spreading sequences are used for spreading. Note, the
(k)
duration of a data symbol dm , m = 0, . . . , M −1, is T = LTc ,
while the duration of a spreading chip is Tc . After another
The variance of the additive white Gaussian noise (AWGN) is
S/P and frequency interleaving the resulting Nc = M L chips
(k) (k) σ 2 , and K is the number of active users.
s(k) = (s0 , . . . , sNc −1 )T are pre-equalized with an Nc × Nc The received signal is influenced by frequency selective
diagonal channel pre-equalization matrix G(k) to obtain the fading of the time-variant multipath channel, AWGN, and
(k) (k)
Nc pre-equalized chips v(k) = (v0 , . . . , vNc −1 )T according multiple-access interference (MAI) caused mainly by the
to Doppler induced frequency offsets among the K users. As
v(k) = G(k) s(k) . (1) (n)
a result, the received symbol rm , m = 0, . . . , M − 1, of a
(k) certain user k = n, after guard interval removal, fast Fourier
The diagonal elements Gi , i = 0,. . .,Nc − 1, of the pre-
transform (FFT), and despreading, can be represented as
equalization matrix are calculated from the channel state infor-
mation derived from downlink channel estimation. The chan- L−1

K−1 c −1

N
(n) (k) (k) (k)
(k) (k)
(n)
nel fading on each subcarrier is determined using pilot symbols rm = cl c(i div M ) ρi ejθi gi ejγi
interleaved in time and frequency and applying appropriate l=0 k=0 i=0
(k)
interpolation [2]. Finally, the pre-equalized chips v(k) are ·d(i mod M ) S(lM +m−i −(k) )+η (3)
modulated on the Nc subcarriers using the inverse fast Fourier
transform (IFFT). After that, parallel to serial conversion (P/S) for the case of block interleaving, where
is performed and a guard interval, that exceeds the delay sin(πx) Nc −1

spread of the multipath channel, is added as cyclic prefix. S(x) = e−jπ Nc x . (4)
Nc sin( N1c πx)
Three different frequency interleaving schemes are investi-
gated. First, a block interleaver which ensures the maximum In Equation (3), ’(.div.)’ denotes integer division and ’(.mod.)’
frequency separation between the L spread chips of each data modulo operation. Moreover, (k) (−0.5 < (k) < 0.5) is the
(k)
symbol dm , m = 0, . . . , M − 1. Second, a random interleaver frequency offset normalized to the subcarrier spacing, and
which performs random interleaving of the Nc input chips. η represents the AWGN with variance σ 2 . Note, due to the
And finally, the novel skip-carrier interleaver which is identical Doppler induced frequency offsets the subcarriers of different
to the block interleaver except for the fact that a few well- users do not coincide anymore, thus, causing MAI.
defined subcarriers are left empty.
Channel pre-equalization can be done in numerous ways [3]. III. I NFLUENCE OF FREQUENCY OFFSETS
Since investigation of channel pre-equalization is out of the Since the focus of the investigation is on the influence of
scope of this paper, perfect channel knowledge acquired from misaligned users on the desired user, it is assumed that the
the downlink transmission is presumed. The pre-equalization frequency offset of the desired user k = n is (n) = 0.
(n)
method applied is constrained quasi-MMSE pre-equalization. The desired signal part Rm in Equation (3) is
When applying power constrained pre-equalization, the trans-
L−1
mitted power for the system with pre-equalization is kept the 1
(n) (n)
(n) (n)
(n)
Rm = ρlM +m ejθlM +m glM +m ejγlM +m d(n)
m , (5)
same as in the case without pre-equalization. The complex L
(k) l=0
elements Gi , i = 0, . . . , Nc − 1, of the diagonal channel
(n)
pre-equalization matrix G(k) are given by [3] whereas the MAI part Im is given by
(k)
Gi
(k) (k)
= gi ejγi
L−1

K−1 c −1

N
(n) (k) (k) (k)
(k) (k)
(n)
  1 Im = cl c(i div M ) ρi ejθi gi ejγi
2
l=0 k=0 i=0
(k)∗  k=n
Hi  Nc  (k)
=  −1  ,(2) ·d(i mod M ) S(lM +m−i −(k) ). (6)
(K − 1)|H |2 + σ 2 LN 
(k) c (k) 2
i |Ht |
 2
(k) 2 2
t=0 (K−1)|Ht | +σ L Using ’E{.}’ to denote expectation, the exact analytical ex-
2
pression of the MAI variance σMAI,m of the mth data symbol
where superscript ’*’ denotes complex conjugation. The com-
can be written as
plex channel fading coefficient of the kth user on the ith sub-
(k) (k) (k) (k)
2 (n) 2
carrier is Hi and can be represented as Hi = ρi ejθi . σMAI,m = E{|Im | }. (7)

IEEE Communications Society 0-7803-8533-0/04/$20.00 (c) 2004 IEEE

 1−ε (k) ε (k)
approximated by
2
σ̂MAI,m ≈
Power spectrum of individual subcarriers

L−1 2

K−1 
(n) (k) (k) 
E{|S( (k) 2 
)| }E  cl cl ρlM +m glM +m 
(k)

k=0 l=0
k=n

K−1
desired user, user (n)
interfering user, user (k), k≠n + E{|S(1 − (k) )|2 }
k=0
k=n
 L−1 2

(n) (k) (k) 
·E  cl cl ρlM +m−1 glM +m−1  ,
(k)
(10)
l=0
Frequency when m > 0, and by
L−1 
Fig. 2. Illustration of the influence of the two closest subcarriers of an 
K−1   (n) (k) (k) (k) 2
interfering user on the specific subcarrier of the desired user;
(k) > 0.
2
σ̂MAI,0≈ E{|S((k) )|2 }E  cl cl ρlM glM  +
k=0 l=0
k=n
L−1 2

K−1   (n) (k) (k) 
E{|S(1− )| }E  cl cl−1 ρlM −1 glM −1  , (11)

(k) 2 (k)

However, further evaluation of Equation (7) is very difficult. k=0 l=1

2 k=n
Therefore, an upper bound σ̂MAI,m of the MAI variance is
considered in the sequel. This upper bound can be calculated when m = 0. The MAI influence in Equation (10) is a result
from Equation (6) as only of the non-ideal channel pre-equalization represented
(k) (k) (k) (k)
by ρlM +m glM +m and ρlM +m−1 glM +m−1 , while spreading
K−1

M −1 L−1


 2
code c(k) and despreading code c(n) are synchronized and,
2  (k) 
σ̂MAI,m= E S(pM +m−t− ) therefore, do not cause additional MAI. Note, orthogonal WH
k=0 t=0 p=−L+1 spreading codes are used for spreading. In Equation (11), the
k=n
 L−1 2
MAI influence is two-fold. Besides the influence from the

(n) (k) (k)  non-ideal pre-equalization, there is an additional influence in

·E  cl cl−p ρ(l−p)M +t g(l−p)M +t  ,(8)
(k)
l=0
0
l−p<L
(k)
where it is presumed, that E{|dt |2 } = 1, t = 0, . . . , M − 1.
Analysis of |S(pM+m−t−(k) )|2 shows, that this expression
has its main contribution for the case
−1 < (pM + m − t − (k) ) < 1.
This inequality is fulfilled only for the two shifted subcarriers
of the interfering users which are closest to the specific
subcarrier of the desired user.
For simplicity and without loss of generality it is assumed
that (k) > 0, k = 0, . . . , K − 1, k
= n. In this case, the
inequality given in Equation (9) is fulfilled when pM +m−
t = 0 or pM +m−t = 1. These two cases correspond to the
two subcarriers of interfering users shifted by (k) and 1 −
(k) , respectively, compared to the observed subcarrier of the
desired user. Fig. 2 illustrates the influence of the two closest
subcarriers of an interfering user on the specific subcarrier of
the desired user. For the illustration in Fig. 2 it is assumed, that
the frequency offset of the interfering user is (k) > 0 and that
the influence of the fading channel is ideally pre-equalized.
When applying the condition given in Equation (9) to
Equation (8), it can be shown, that the influence of MAI is not
equally distributed over all data symbols. The analysis shows,
that the upper bound of the MAI variance for this case can be
IEEE Communications Society
the second part of Equation (11) caused by unsynchronized
spreading codes, shifted by one chip. The additional signal
degradation due to the unsynchronized spreading codes in
(n) (k) (k) (k)
Equation (11) is represented by cl cl−1 ρlM −1 glM −1 .
The results from Equations (10) and (11) can be used
to develop a frequency interleaving scheme which is more
resistant to frequency offsets than existing ones, as shown in
the following.
(9)
An important property of the block interleaver lies in the fact
that for (k) > 0 only for symbol m = 0 the MAI influence
is two-fold, while the other M − 1 symbols show an ’inner-
protection’ property and are affected in a less destructive way
by MAI. Analogously, in the case (k) < 0 the MAI influence
is two-fold only when m = M −1. From this considerations, it
can be concluded that in a realistic environment, where some
users have positive and others have negative frequency offsets,
only the symbols m = 0 and m = M −1 suffer from two-fold
MAI influence, while the other M − 2 symbols profit from the
’inner-protection’ property.
The MAI influence for the case of block interleaving is
illustrated in Fig. 3 for parameters L = 4, M = 4, K = 2. It
is assumed, that user k = 0 is the desired user, the frequency
offset of the second user k = 1 is (1) > 0, and that the
fading channel is ideally pre-equalized. From Fig. 3 it can
be seen, that two different types of MAI influence exist.
There is influence from synchronized chips and influence from
unsynchronized chips. The ’inner-protection’ property of the
block interleaver is reflected in these two different influences.
0-7803-8533-0/04/$20.00 (c) 2004 IEEE
 (0) (0)
cd
(0) (0)
c1d1
(0) (0)
cd
(0) (0)
cd cd
(0) (0) (0) (0)
cd cd
(0) (0)
cd
(0) (0)
cd
(0) (0) 35
1 0 1 2 1 3 2 0 2 1 2 2 2 3 3 0
(1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1) (1)
cd cd cd cd cd cd cd cd cd 30

Error Distribution in %
0 3 1 0 1 1 1 2 1 3 2 0 2 1 2 2 2 3

25

20
Power spectrum of individual subcarriers

15

10

... ... 5

0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
desired user, user (0)
interfering user, user (1) Data Symbol
influence from sync. chip
influence from unsync. chip Fig. 5. Distribution of errors over data symbols for the case of block
interleaving and a maximal Doppler shift of 30%; desired user uses c(14) ,
Et /N0 = 14 dB, L = 16, K = 16, and M = 16.

Frequency
IV. S IMULATIONS R ESULTS
Fig. 3. Desired and interfering user’s signal power spectrum of individual
subcarriers for the case of block interleaving; L = 4, M = 4, K = 2, and
In this section, several simulation results are given that

(1) > 0. illustrate the performance improvements obtained by the pro-
posed countermeasures for reducing the influence of frequency
offsets.
f The underlying mobile radio channel for the simulations
is based on the assumption of independent Rayleigh fading
t ... on each subcarrier. This assumption is justified by the fact,
(k ) (k )
c d
0 0
... c d (k )
0
(k )
M- 2
(k ) (k )
c d
0 M -1 1
(k ) (k )
c d 0
... (k )
c d
1
(k )
M-2
(k )
c d
1
(k )
M -1 cL( k- )1d 0( k ) ... cL( k-)1d M( k -) 2 c(Lk- )1 d M( k )- 1 that frequency interleaving is applied leading to a spreading
(a) Block interleaver of the chips of one data symbol over the whole transmission
f
skipped skipped skipped bandwidth. QPSK modulation is applied and no channel
carrier carrier carrier coding is considered. The spreading is performed with WH
t X X ... X codes of length L = 16. The number of simultaneously
(k)
c d
0
(k )
0
... c0( k ) dM(k -) 2 c d (k)
1
(k )
0
... c1( k ) d M( k-) 2 c (k)
L- 1 d (k)
0
... c L( k- )1d M( k -) 2 transmitted symbols is M = 16 which results in Nc = 256
(b) Skip-carrier interleaver used subcarriers. Channel pre-equalization is performed with
quasi-MMSE pre-equalization. The frequency offset of the
desired user k = n is set to (n) = 0, while the frequency
Fig. 4. Illustration of frequency allocation schemes: (a) Block interleaver
and (b) Skip-carrier interleaver. offsets (k) of the interfering users k
= n are chosen according
to a Jakes distribution.
The distribution of errors over data symbols is illustrated
in Fig. 5 for block interleaving. The desired user uses c(14)
It can be noticed, that only each M th subcarrier is influenced as spreading code, the signal-to-noise ratio (SNR) is fixed
by unsynchronized spreading code chips and, therefore, suffers at Et /N0 = 14 dB, the maximal Doppler shift is 30% of the
from larger degradation due to MAI. subcarrier spacing, and a fully-loaded system is considered
The random interleaver does not posses the ’inner- (K = 16). Note, Et /N0 represents the transmitted energy per
protection’ property and, therefore, is much more sensitive bit over the noise spectral density. The unequally distributed
to frequency offsets, as will be shown in the next section. MAI influence over data symbols is clearly visible. The data
The proposed novel skip-carrier interleaver operates on the symbols m = 0 and m = 15 suffer from approximately ten
premise, that only the two symbols m = 0 and m = M − 1 times larger degradation than other symbols. In addition, it
are influenced by two-fold MAI, when a block interleaver is can be noticed that also data symbols m = 1, 2, 13, and 14
used. Moreover, the two-fold influence on these two symbols suffer from larger degradation than the remaining symbols.
comes from each other. This leads to the conclusion, that by The reason for this lies in the fact, that MAI degradation is
not transmitting on the subcarriers that correspond to one not only due to the two closest subcarriers, like approximated
of these two symbols, the two-fold MAI influence can be in Equation (9), but is also partly caused by the more distant
mainly eliminated. Thus, for the skip-carrier interleaver block subcarriers.
interleaving is performed with the difference that each M th The different frequency interleaving schemes for TDD/MC-
chip is left empty. This results in a loss of bandwidth efficiency CDMA are compared in Fig. 6. A maximal Doppler shift of
1
by a factor of M . The frequency allocation schemes for the 30% of the subcarrier spacing is assumed and the system
block and the skip-carrier interleaver are illustrated in Fig. 4. is fully loaded (K = 16). The comparison is performed on

IEEE Communications Society 0-7803-8533-0/04/$20.00 (c) 2004 IEEE

 0 0
10 10
No Frequency Offset
Random Interleaver, worst-case
Random Interleaver, worst-case
Block Interleaver, worst-case
Block Interleaver, worst-case
Skip-Carrier Interleaver, worst case
Skip-Carrier Interleaver, worst-case
-1 -1 Random Interleaver, averaged perf.
10 Random Interleaver, averaged perf. 10
Block Interleaver, averaged perf.
Block Interleaver, averaged perf.
Skip-Carrier Interleaver, averaged perf.
Skip-Carrier Interleaver, averaged perf.
Bit Error Rate

Bit Error Rate

-2 -2
10 10

-3 -3
10 10
Maximal Doppler Shift 30%

-4 -4
10 0 10 20 10 0 10 20 30 40
5 15 50

SNR in dB Maximal Doppler Shift (%)

Fig. 6. Performance of different frequency interleavers for a maximal Doppler Fig. 7. BER performance versus subcarrier frequency offset for various
shift of 30%; L = 16, K = 16, M = 16 for block and random interleaver, frequency interleavers for a fixed Et /N0 = 14 dB; L = 16, K = 16, M = 16
M = 15 for skip-carrier interleaver. for block and random interleaver, M = 15 for skip-carrier interleaver.

35

30
the basis of bit error rate (BER) over SNR curves, where
Error Distribution in %

SNR is given in Et /N0 . As it can be seen, the skip-carrier 25

interleaver performs best, while block and especially random 20

interleaver show much larger signal degradation. For the skip- 15
carrier interleaver the subcarriers which belong to the border- 10
ing symbol m = M − 1 are skipped. Both, the worst-case
5
performance and the averaged performance are considered for
all interleavers. For the worst-case performance the spreading 0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
code of the desired user is chosen in such a way, that it Data Symbol
causes maximal possible signal degradation. For the averaged
performance, the performance is averaged by assigning all Fig. 8. Distribution of errors over data symbols for the case of skip-carrier
possible spreading codes to the desired user and averaging interleaving and a maximal Doppler shift of 30%; desired user uses c(14) ,
Et /N0 = 14 dB, L = 16, K = 16, and M = 15.
the results. The reason for the difference in the worst-case
performance and the averaged performance lies in the fact, that
unsynchronized codes, shifted by one chip, play an important
role in the MAI influence as given in Equation (11). Thus, interleaver. In addition, it can be noticed from Fig. 7, that the
when assigning different spreading codes to the desired user, performance of the block and the skip-carrier interleaver is
different influence from the spreading codes of the other active comparable for maximal Doppler shifts lower than 10%.
users will affect the desired user. However, spreading code As already noticed, the skip-carrier interleaver does not
selection does not play a vital role for the random interleaver, eliminate the influence of MAI completely. Fig. 8 illustrates
since it already averages the effect of spreading code selection the distribution of errors over symbols for the skip-carrier
by its random nature. Therefore, the worst-case performance interleaver. The desired user uses c(14) as spreading code,
and averaged performance are almost the same in the case the maximal Doppler shift is 30%, the system is fully-loaded
of random interleaver. Although it is expected that the skip- and the SNR is fixed at Et /N0 = 14 dB. It can be noticed,
carrier interleaver almost completely eliminates the influence that on data symbols m = 0 and m = 14 a considerably
of frequency offsets, Fig. 6 shows, that residual influence larger number of errors is generated than on the other data
remains. This is due to the fact, that not only the two closest symbols. This motivates to skip some more subcarriers and
subcarriers cause MAI degradation, like already illustrated in to observe, whether there is further significant performance
Fig. 5. improvement. However, compared to Fig. 5 the percentage of
Another representation of the simulation results is given in errors generated on bordering symbols is significantly smaller.
Fig. 7, where the BER is shown in dependence of the maximal Fig. 9 illustrates the improvements achieved by skipping
Doppler shift for a fixed SNR of Et /N0 = 14 dB. Again, the additional subcarriers. Again, similar as for the skip-carrier
skip-carrier interleaver outperforms both block and random interleaver, the additional skipped subcarriers are those who

IEEE Communications Society 0-7803-8533-0/04/$20.00 (c) 2004 IEEE

 -3
10
The proposed novel skip-carrier interleaver outperforms
With Frequency Offset both block and random interleaver when the maximal nor-
No Frequency Offset
malized frequency offset is larger than 10%. For a maximal
normalized frequency offset lower than 10% the performance
of skip-carrier and block interleaver is comparable and the
degradation due to frequency offsets is tolerable. The skip-
Bit Error Rate

carrier interleaver introduces a loss in bandwidth efficiency

-4
10 0
16 32 48 64 80 96
Number of Skipped Subcarriers
Fig. 9. BER performance versus the number of skipped subcarriers for a
maximal Doppler shift of 30%; desired user uses c(14) , Et /N0 = 14 dB,
L = 16, and K = 16.
belong to bordering symbols, e.g. to symbol m = M − 2 if
altogether 2L = 32 subcarriers are skipped. The additional
skipping of subcarriers leads to a further loss of bandwidth
efficiency. From Fig. 9, it can be concluded, that a good
trade-off between bandwidth efficiency and performance im-
provements is to skip L = 16 or 2L = 32 subcarriers. This
1 1
corresponds in the considered case to a loss of M = 16 and
2 2
M = 16 in bandwidth efficiency, respectively. Further skip-
ping of subcarriers no longer results in significant performance
improvements. Therefore, it is suggested not to skip more than
2L = 32 subcarriers, since the resulting additional gains no
longer justify the additional loss in bandwidth efficiency.
V. C ONCLUSIONS
In this paper, the influence of frequency offsets on the
performance of uplink TDD/MC-CDMA is considered and
countermeasures for the reduction of the effects of frequency
offsets are proposed and investigated. A scenario in which the
desired user is synchronized to the receiver, while all other
interfering users are misaligned in frequency, is considered.
It is shown, that frequency interleaving schemes have a great
impact on the system performance.
1
by a factor M . However, this loss is justified due to the large
gains obtained with the skip-carrier interleaver as illustrated
in this paper.
Further performance gains can be achieved if additional
subcarriers are skipped. However, these gains are smaller than
for the original skip-carrier interleaver. Moreover, the loss in
bandwidth efficiency is increased. Thus, there is a trade-off
112 128
between the additional performance improvements obtained
by additional subcarriers skipping and the increased loss in
bandwidth efficiency. Skipping L or 2L subcarriers seems to
be a good compromise.
VI. ACKNOWLEDGMENTS
The authors would like to acknowledge the valuable support
of A. Dammann, E. Haas, and R. Raulefs.
R EFERENCES
[1] K. Fazel and L. Papke, “On the performance of convolutionally-coded
CDMA/OFDM for mobile communication system”, in Proceedings
IEEE International Symposium on Personal, Indoor and Mobile Radio
Communications (PIMRC’93), Sept. 1993, pp. 468–472.
[2] S. Kaiser, Multi-Carrier CDMA Mobile Radio Systems - Analysis and Op-
timization of Detection, Decoding and Channel Estimation, Düsseldorf:
VDI Verlag, Fortschritt-Berichte VDI, series 10, no. 531, 1998.
[3] I. Cosovic, M. Schnell and A. Springer, “On the performance of different
channel pre-compensation techniques for uplink time division duplex
MC-CDMA”, in Proceedings IEEE Vehicular Technology Conference
(VTC’03, Fall), Oct. 2003.
[4] D.G. Jeong and M.J. Kim, “Effects of channel estimation error in
MC-CDMA/TDD systems”, in Proceedings IEEE Vehicular Technology
Conference (VTC’00, Spring), May 2000, pp. 1773–1777.
[5] T. Kim, Y. Kim, et al, “Performance of an MC-CDMA system with
frequency offsets in correlated fading”, in Proceedings IEEE International
Conference on Communications (ICC’00), June 2000, vol. 2, pp. 1095–
1099.
[6] Y. Yang, et al, “Performance of up-link MC-CDMA system with
frequency offset and imperfect channel estimation simultaneously over
nakagami fading channels”, Wireless Personal Communications, vol. 27,
no. 3, pp. 247–265, 2003.

IEEE Communications Society 0-7803-8533-0/04/$20.00 (c) 2004 IEEE