Professional Documents
Culture Documents
Abstract— In this paper, uplink Time Division Duplex (TDD) Throughout the paper, a slowly varying time-variant channel
Multi-Carrier Code-Division Multiple-Access (MC-CDMA) sys- is assumed. Thus, channel pre-equalization techniques are
tems applying pre-equalization to combat channel impairments applicable to the uplink TDD/MC-CDMA transmission which
are considered. Especially, the performance degradation caused
by the Doppler induced frequency offsets among the different are based on the channel estimation results from the downlink
mobile users is analyzed in detail. Based on this analysis different transmission [3], [4]. With TDD mode the effects of the
frequency interleaving schemes are proposed and their potential channel on the uplink signal can be pre-equalized at the mobile
to reduce the performance degradation is investigated. It turns station based on the channel estimation from the downlink
out, that a frequency interleaver which is allowed to skip a few, frame.
well-defined subcarriers is capable of reducing the influence of
Doppler induced frequency offsets significantly causing a very The desired user is assumed to be synchronized to the
little loss in bandwidth efficiency. receiver frequency, while all other users are misaligned in
frequency. Different frequency interleaving schemes are in-
I. I NTRODUCTION vestigated and their performances are analyzed. A novel skip-
carrier interleaver is introduced which significantly improves
Multi-Carrier Code-Division Multiple-Access (MC- the performance of an uplink MC-CDMA system in the
CDMA) [1], [2] is well recognized as a very promising presence of frequency offsets between the different users.
candidate for the air interface of future mobile wireless The paper is organized as follows. The TDD/MC-CDMA
downlink systems. However, when considering MC-CDMA uplink transmission system considered throughout this paper
for uplink transmission some problems arise. In particular, is described in Section II. In Section III, both the influence
channel estimation and equalization as well as time and of misaligned users on each other is analyzed and frequency
frequency synchronization are more difficult and require interleaving schemes are investigated. The performance of the
more complex algorithms. One of the possibilities to uplink TDD/MC-CDMA system in the presence of Doppler
mainly overcome the problems of channel estimation and induced frequency offsets and the performance gains achieved
equalization is to apply channel pre-equalization at the through the proposed skip-carrier interleaver are presented in
MC-CDMA transmitter as described in [3], [4]. In this Section IV. Finally, Section V summarizes the results.
paper, the influence of Doppler induced frequency offsets
on the performance of uplink time division duplex (TDD) II. T RANSMISSION SYSTEM
MC-CDMA applying pre-equalization at the transmitter A time synchronous uplink MC-CDMA system with K
is analyzed in detail. Additionally, countermeasures based active users is considered. Fig. 1 depicts the block diagram
on thoroughly designed frequency interleaving schemes of the considered uplink transmitter of mobile user k, k =
are proposed. Sensitivity of uplink MC-CDMA systems to 0, . . . , K − 1. The uplink transmitter is based on the MC-
frequency offsets has been considered in several papers [5], CDMA system concept with M -Modification as proposed
[6], but both, an in-depth analysis of the influence of in [2]. Note, the M -Modification allows each user to transmit
misaligned users on each other and proposals for reducing M data symbols simultaneously using an additional Orthog-
the resulting performance degradation are not yet available. onal Frequency-Division Multiplexing (OFDM) component
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 LN
(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)
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)
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
-3 -3
10 10
Maximal Doppler Shift 30%
-4 -4
10 0 10 20 10 0 10 20 30 40
5 15 50
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 %