OFDM-CDMA versus DS-CDMA: Performance Evaluation for Fading Channels

German Aerospace Research Establishment (DLR), Institute for Communications Technology D-82234 Oberpfa enhofen, Germany, Tel.: +49 8153 28 2805, Fax: +49 8153 28 1442

Stefan Kaiser

In code division multiple access (CDMA) systems direct sequence (DS) applying maximum ratio combining (MRC) at the receiver has achieved high interest for mobile communications. A second and rather novel technique is orthogonal frequency division multiplexing (OFDM) applying maximum-likelihood detection (MLD) at the receiver. In this paper the performance of both techniques is compared considering the mobile radio environment. The analytical results for both DS-CDMA and OFDM-CDMA are con rmed by Monte Carlo simulations. The analysis enables a performance comparison between the DS-CDMA system and the OFDM-CDMA system with respect to the demands of low complexity receivers which is important for the system design. The results show that OFDM-CDMA outperforms DS-CDMA in terms of spectral e ciency.

Abstract

simulation results in Section 5. In Section 6 a performance comparison between both techniques in terms of spectral efciency is carried out. Finally, Section 7 summarizes the results.

2 DS-CDMA with MRC

The DS-CDMA transmission scheme applying MRC is shown in Figure 1 for the down-link. The base station has a maximum user capacity of Nu users. Every user i, i = 1 : : : Nu, is assigned a spreading code vector Ci of period L. Since good correlation properties of the spreading codes are required for MRC, preferentially phased Gold codes 8] are chosen. The data bits di, di 2 f;1 1g, of each user i are
Base Station
d1 d2
Spreader Code 1 Spreader Code 2

Multiple access systems based on DS-CDMA have achieved increasing signi cance for mobile communications applications 1, 2]. A promising concept is based on DS-CDMA applying MRC at the receiver. MRC takes advantage of the channel diversity to combat the multi-path fading 3]. However, the capacity of a DS-CDMA system is limited by both multi-user interference and inter-symbol interference (ISI) in high data rate applications. OFDM-CDMA systems have been proposed lately 4]. OFDM is applied to combat the frequency selectivity of the channel using a simple one tap equalizer 5, 6]. Furthermore OFDM prevents the ISI and inter-carrier interference (ICI) by inserting a guard interval between adjacent OFDM symbols. OFDM is typically used for audio, TV and HDTV transmission over terrestrial channels 5, 7] and achieves high spectral e ciency. The aim of this study is to carry out a performance comparison between DS-CDMA and OFDM-CDMA. Both techniques are considered for the down-link (base to mobile), since for this application power e cient mobile receivers with low complexity are required. Whereas, for the up-link (mobile to base) a high complexity multi-user detector is generally accepted to cope with the presence of several users in independent fading channels. The paper is organized as follows: In Section 2 the DSCDMA transmission scheme applying MRC is presented. The OFDM-CDMA transmission scheme applying MLD is described in Section 3. An analytical performance evaluation for DS-CDMA and OFDM-CDMA is shown in Section 4. The validity of the analytical results is con rmed by

1 Introduction

Mobile User i

Despreader Code i

di

Spreader Code i

Channel

Despreader Code i

di

dNu

Spreader Code Nu

Despreader Code i

Figure 1: DS-CDMA transmission scheme with MRC multiplied by the corresponding spreading code Ci such that one bit duration Tb exactly ts one code period consisting of L chips, each chip of duration Tc . The chip streams of all users are added synchronously in the base station to yield the vector S with S=
Nu X i=1

diCi = s0 s1 : : : sL;1]

(1)

consisting of L components. In the mobile receiver MRC is realized by a RAKEreceiver. As described in 3, 9] a RAKE-receiver with diversity D has D arms, each to resolve one echo, whereby D is limited due to the hardware complexity. In each arm the received signal is delayed, de-spread with Ci and weighted with the conjugate complex channel attenuation d of the assigned echo. Finally, the RAKE-receiver adds the results ^ obtained from each arm and makes a decision on di. Note that MRC requires knowledge about the channel conditions.

slowly changing frequency selective fading channel. fDmax is The transmission scheme under study for the down-link of a the maximum Doppler frequency. OFDM-CDMA system with MLD is presented in Figure 2. The output of the OFDM modulator can be expressed as
Base Station
S0
serial to parallel Conv.

max

{ { {
Channel

d1

Spreader Code 1

Spreader Code Nu/K

x(t) =
ML K

M ;1 K ;1 K ;1 XXL X m=0 k=0 l=0

sl k m ej 2 fmL+kL=K+l t

t 2 0 T ] (3)

Spreader Code 1

where sl k m is the lth component of Sk of the mth spread data bit of one user with the assigned sub-carrier frequency
1

di
Spreader Code Nu/K

Sk

serial to parallel Conv.

Interleaver
Nc

OFDM Mod.

fmL+kL=K +l = f0 + mL + kL=K + l : T
s

(4)

Spreader Code 1

K-1

dNu
Spreader Code Nu/K

SK-1

serial to parallel Conv.

f0 is the lowest sub-carrier frequency (for simplicity f0 = 0 in the following) and the sub-carrier distance is 1=Ts due to the orthogonality. Based on the assumptions that the guard interval is greater than the multi-path spread and that the channel varies slowly compared to the symbol duration, the received signal without noise sampled at rate 1=Tc , Tc = Ts =Nc, can be written as
Maximum Likelihood Detector

Mobile User i
1

OFDM Demod.
Nc

ML parallel Deinter- K to serial leaver Conv.

Rk

n y (nTc ) = (;1)
di
^

M ;1 K ;1 K ;1 X XXL m=0 k=0 l=0

l k m sl k m e

kL=K j2 n(mL+Nc +l) (5)

Figure 2: OFDM-CDMA transmission scheme with MLD In order to reduce the complexity of the MLD, the transmitted signals of the Nu users are grouped into K blocks. Every block has a maximum user capacity of Nu =K users. With that, the data bits di are spread by a spreading code vector Ci of period L=K. For the OFDM-CDMA system orthogonal Walsh-Hadamard codes 10] are chosen for the spreading, which results in zero cross-correlation. The chip streams of all users of the kth block are added synchronously in the base station to yield the vector Sk k = 0 : : : K ; 1 with Sk =
Nu (k+1) KX

where hl k m = l k m ej l k m is the channel response with the attenuation l k m and the phase rotation l k m (the Doppler shift is also included in this expression) at the frequency fmL+kL=K +l . Without noise fhl k m sl k m g is the Discrete Fourier Transform (DFT) of fy(nTc )(;1)n =Nc g 5]. In practice the DFT can be realized by a Fast Fourier Transform (FFT). Thus, the OFDM signal can be generated by an inverse FFT. The received signal Rk after OFDM demodulation is affected by at Rayleigh fading and noise. Rk can be written as a vector of L=K components with Rk = Hk Sk + Nk (6) where the diagonal matrix Hk describes the Rayleigh fading on the sub-carriers, Sk represents the transmitted sequence and Nk is the noise vector assigned to block k. To detect the transmitted data from the received signal by exploiting the diversity D0 a MLD is applied, i.e. the most likely transmitted sequence is determined by minimizing the Euclidean distance e2 between the received and all possible transmitj ted sequences. Let Vj , j = 1 : : : 2Nu =K , denote the set of all possible transmitted sequences, then S^k = Vj if e2 = minjjRk ; Hk Vj jj2: j (7) ^ ^ From Sk we decide on di. This algorithm also requires knowledge about the channel to weight the sequences Vj with the assigned fading Hk . The complexity of this algorithm grows exponentially when the number of users increases. For that reason the total of all Nu user's signals is split into K independent blocks at the base station. Hence, instead of 2Nu possible sequences only 2Nu =K possible sequences have to be

i= Nu k+1 K

diCi = s0 k s1 k : : : sL=K ;1 k ]

(2)

consisting of L=K components, where at the receiver it results in a diversity of D0 = L=K. As indicated in Figure 2, each block can transmit M > 1 data bits per OFDM symbol and user. Therefore, M vectors Sk are grouped together and all MTb seconds ML=K chips per block are passed to the OFDM modulator. Thus the output of each serial to parallel converter is mapped onto ML=K sub-carriers. Prior to the OFDM modulation a frequency and time interleaver scrambles the outputs of all K blocks in order to achieve independent fading between adjacent sub-carriers in time. The total number of OFDM sub-carriers is Nc = ML. Hence, one OFDM symbol has the duration Ts = MTb and is preceded by a guard interval to prevent ISI and ICI. The total OFDM symbol duration is T = Ts + . By increasing M the ratio =T i.e the relative overhead due to the guard interval decreases. However, T should not be chosen too large since

systems with a realistic complexity. In this paper up to Nu =K = 8 users per block are considered. The e ects of varying K on the performance is shown in Section 5.

1 1 ; 1 2D(Nu ; 1) + DN0 + 1 ; 2 2 2 3L Eb

4 Performance Evaluation

D ;1 X j=0

;1+
j

1 !j 1 + 1 2D(Nu ; 1) + DN0 + 1 ; 2 2 2 3L Eb

(12)

In this section, the performance of DS-CDMA and OFDMCDMA in AWGN and frequency-selective Rayleigh fading channels is analytically evaluated. An approximation for the bit error probability of a DS-CDMA system with MRC is derived for Rayleigh fading channels. Whereas, for a OFDMCDMA system with MLD an upper bound for the bit error probability is de ned. In the case of DS-CDMA, the multi-user interference can be taken into account using the Gaussian approximation 11] for the synchronous case 12]. Due to the central limit theorem this approximation is su ciently good especially for a large number of users. The resulting bit error probability Pb is given by s ! 1 erfc Eb =N0 Pb = 2 (8) 1 + Itot Eb=N0 where Eb is the data bit energy and N0 =2 is the Gaussian noise spectral density. ;1 the case of random codes the inIn terference is Itot NuL 12]. However, for preferentially u2 phased Gold codes the interference will be lower, Itot NL;1 8]. Since in the OFDM-CDMA case there is no multi-user interference by using orthogonal Walsh-Hadamard codes, the bit error probability is given by (8) without interference, Itot = 0.

4.2.2 OFDM-CDMA

4.1 AWGN channel

We suppose that all sub-carriers are a ected by independent Rayleigh fading, i.e. ideal frequency/time interleaving, where all fading amplitudes are uniformly distributed. With MLD, the bit error probability Pb is bounded by 13, 14] 1 X X a(S S ) p(S ) P(S jS ) ^k k Pb 2 (13) k ^k k Sk S 2V ^
k

where V is the set of all possible transmitted sequences Sk . ^ a(Sk Sk ) is the number of bit errors that occur when the ^ sequence Sk is transmitted and Sk 6= Sk is chosen in the detector. p(Sk ) is the probability that the sequence Sk has been transmitted. Since there are 2Nu =K possible sequences per block which are assumed to be equal likely, we obtain ^ p(Sk ) = 2;Nu =K . P(Sk jSk ) represents the pairwise error ^ probability, i.e. the probability that the decoder chooses Sk when Sk was transmitted. The pairwise error probability ^ P(Sk jSk ) using the Cherno bound is given by 13] ^ P (Sk jSk )
L K ;1 Y

l=0

^ 1 + KEb jsl k ; sl k j 4LN

2 ;1

(14)

Hence the bit error probability in the multi-user case can be bounded by
P

4.2.1 DS-CDMA

4.2 Frequency selective fading channel

X N 2; Ku ;1

Considering a multi-path channel, for the single-user case the bit error probability Pb for DS-CDMA with RAKE-receiver of diversity D can be written in the closed form 3] D D ;1 D ; 1 + j X 1+ j Pb = 1 ; (9) 2 j 2 j =0

Sk Sk 2V ^

^ a(Sk Sk )

l=0

k;^ 1 + K EbjslLN sl k j 4
0

;1

In this section the performance of the above described systems will be analyzed analytically and by simulations. After the speci cation of the system-parameters, the analytical and simulation results of the DS-CDMA system with MRC r and the OFDM-CDMA system with MLD will be presented. = D+ (10) The analytical results assume independent Rayleigh fading and the results obtained by simulations consider a real fadand = Eb =N0. It is assumed that the average signal to noise ing channel typical for bad urban environments 15]. Since ratio per path p is equal for all paths so that the signal to we do not take into account channel coding, we focus on the noise ratio per data bit is = D p . To handle the multi- bit error probability of Pb = 10;3. user interference we consider the Gaussian approximation for the asynchronous case 12]. The resulting signal to noise 5.1 System-Speci cation ratio per data bit int is obtained as The considered channel is a fading channel for a typical bad 2(Nu ; 1) + N0 ;1 : (11) urban environment with a multi-path spread of Tm = 10 s. int The Doppler shift is modeled by a classical Doppler spectrum 3L Eb (Jakes spectrum) 15]. The used bandwidth is B = 1:25 MHz Replacing by int in (10) the bit error probability in the and the resulting chip duration is Tc = 0:8 s. As typical for multi-user case can be approximated with (9) as CDMA applications a carrier frequency band at f0 = 1 GHz

5 System Performance

(15)

cies up to 200Hz have to be taken into account in the mobile Eb =N0 = 20dB, the maximum number of active users is radio channel 1]. We assume that the DS-CDMA and the about 16, yielding a spectral e ciency of 0.212 bit/s/Hz. OFDM-CDMA receiver have perfect channel knowledge and are perfectly synchronized. For channel sounding and syn- 5.3 Performance of OFDM-CDMA chronization 15% of the data rate are assumed. The OFDM-CDMA system utilizes Walsh-Hadamard codes of period L=K = 4 with K = 16 blocks and of period 5.2 Performance of DS-CDMA L=K = 8 with K = 8 blocks, respectively. With that, one The DS-CDMA system utilizes preferentially phased Gold base station can supply up to 64 users. The total number codes of period L = 63 so one base station can theoreti- of OFDM sub-carriers is Nc = 512, so M = 8 data bits per cally supply up to 65 users. The achievable netto chip rate user could be transmitted per OFDM symbol. The resultis 1.0625 M-chip/s when using the BPSK modulation and ing sub-carrier distance is 2.441 kHz, hence the chip durataking into account the loss due to channel sounding and tion on each sub-carrier is Ts = 409:6 s. To combat the synchronization. Figure 3 shows the performance for di er- echoes of the channel a guard interval of = 16 s is chosen. The resulting netto chip rate is 1.0236 M-chip/s when BPSK 10 modulation is used and taking into account the loss due to analytical results, fading channel simulation results, fading channel channel sounding, synchronization and the guard interval. AWGN channel In order to reduce the correlation of the fading factor of ad10 63 users jacent sub-carriers a pseudo random frequency interleaver is 32 users applied. 10 The performance for di erent numbers of users for the 16 users OFDM-CDMA system is illustrated in Figure 5 for maximum Nu =K = 4 users per block, i.e. K = 16 blocks and code 10 period L=K = 4, corresponding to a diversity of D0 = 4. For 8
0 -1 -2

Pb

-3

r se 1u

use

10

-4

rs

10

1 us er

10

-5

10 0 2 4 6 8 10 12 14 average Eb/N0 in dB 16 18 20 10 Pb

-1

analytical results, fading channel simulation results, fading channel AWGN channel

Figure 3: Analytical and simulation results for DS-CDMA with MRC: Code period L = 63 and diversity D = 10. ent numbers of users. The considered RAKE-receiver has a diversity of D = 10. This is rather high taking into account the receiver complexity but promises a good performance as shown in Figure 4, where the performance for di erent diversities D is plotted, considering 4 active users. In Figure 3
10
0

-2

64

use

10

-3

rs

32
us ers

164

use

rs

10

-4

us

er

10

-5

fading channel, 4 users AWGN channel

10

-1

Figure 5: Analytical and simulation results for OFDML CDMA with MLD: Code period K = 4 and diversity D0 = 4. a low number of users at low bit error probabilities the simulation results are somewhat worser than the bounds, the reason for that is that the bounds assume perfect interleaving. Simulations with independent Rayleigh fading on each sub-carrier con rmed the validity of the bounds. At Pb = 10;3 and Eb=N0 = 17dB the maximum number of active users is 64 (full user capacity), yielding a spectral e ciency of 0.818 bit/s/Hz. The results take into account the additional required signal energy for the guard interval in the signal to noise ratio. The choice of diversity D0 = 4 promises a low system complexity, since the MLD has to evaluate only 16 possible sequences. Figure 6 illustrates that by increasing the diversity D0 , corresponding to a decrease of the number of blocks K, the system performance could be improved signi cantly. In the system comparison in Section 6 we consider the two cases D0 = 4 and D0 = 8. The curve in Figure 6 with D0 = 1 corresponds to conventional OFDM transmission without bandwidth spreading.

8 10 12 14 average Eb/N0 in dB

16

18

20

D=1

10 Pb

-2

D=2
D= 4

10

-3

r se 1u

D =8

16 D=

10

-4

D= 32

10

-5

Figure 4: In uence of the diversity D on the performance of DS-CDMA: Code period L = 63 and 4 active users. it can be seen that the analytical results correspond well with the simulation results, especially for a high number of users. As it is expected, the maximum number of active users the system can support at Pb = 10;3 is up to 25% of

8 10 12 14 average Eb/N0 in dB

16

18

20

Two multiple access systems based on direct sequence CDMA have been compared for the communication from conv. OFDM a base station to a mobile user over fading channels. For D=1 (K=64 10 ) the DS-CDMA system with MRC and the OFDM-CDMA D= 2 (K system with MLD an approximation and a bound, respec=32 ) D= 4 (K tively, have been derived to evaluate the performance. The 10 =16 ) analytical results are con rmed by simulation results. A perD= 8 (K =8) formance comparison of the OFDM-CDMA system and the 10 DS-CDMA system shows that for a low number of active users both systems perform similarly but with an increasing number of active users the OFDM-CDMA system outper10 0 2 4 6 8 10 12 14 16 18 20 forms the DS-CDMA system. We can conclude that OFDMaverage E /N in dB CDMA is an interesting alternative to DS-CDMA with reFigure 6: In uence of the diversity D0 (and K, resp.) on the spect to the high spectral e ciency and the demands of low performance of OFDM-CDMA: Nu = 64 active users. complex mobile receivers.
10
-1 -2

fading channel, 64 users AWGN channel

Pb

-3

1-

ers us 64

-4

-5

The OFDM-CDMA system and the DS-CDMA system are compared taking into account the resulting spectral e ciency in terms of the necessary average Eb =N0 to achieve a given bit error probability. The OFDM-CDMA system with MLD has a diversity D0 = 4 and D0 = 8, respectively. The DSCDMA system with MRC has a diversity D = 10. The comparison with respect to the spectral e ciency is illustrated in Figure 7 for the bit error probability Pb = 10;3 without
1.0 0.9 spectral efficiency in bit/s/Hz
D= 8)

6 OFDM-CDMA vs. DS-CDMA

References

0.8
DM

A(

0.7
DM

0.6 0.5 0.4 0.3 0.2 0.1 0.0 8 9 10

OF

DM

-C

-C

DM

(D

4)

DS-CDMA (D

=10)

11

Figure 7: OFDM-CDMA versus DS-CDMA: Spectral e ciency as a function of the average Eb =N0 for Pb = 10;3. channel coding. It can be seen that both systems have a similar performance for a low number of users, i.e. for a low spectral e ciency. However, when the number of users increases the OFDM-CDMA system outperforms the DS-CDMA system signi cantly. For Pb = 10;3 the OFDM-CDMA system with D0 = 4 has a spectral e ciency of about 0.8 bit/s/Hz (64 users) at Eb =N0 = 17dB whereas the DS-CDMA system has a spectral e ciency of only about 0.2 bit/s/Hz (16 users), i.e. in this case the OFDM-CDMA system can handle 4 times more users than the DS-CDMA system. Considering the OFDM-CDMA system with D0 = 8, a spectral e ciency of about 0.8 bit/s/Hz (64 users) can be achieved at Eb=N0 = 13dB. The DS-CDMA system reaches only a spectral e ciency of about 0.14 bit/s/Hz at Eb=N0 = 13dB, this is less than 1/5 of the OFDM-CDMA user capacity.

12 13 14 15 16 average Eb/N0 in dB

17

18

19

20

1] D. L. Schilling , \Spread spectrum for commercial communications," IEEE Commun. Mag., pp. 66{79, April 1991. 2] P. Jung, P. W. Baier, and A. Steil, \Advantages of CDMA and spread spectrum techniques over FDMA and TDMA in cellular mobile radio applications," IEEE Trans. Veh. Technol., vol. 42, pp. 357{364, August 1993. 3] J. G. Proakis, Digital Communications. McGraw-Hill, Inc, second ed., 1989. 4] K. Fazel, \Performance of CDMA/OFDM for mobile communication system," in IEEE, 2nd International Conference on Universal Personal Communications (ICUPC) Proc., October 1993. 5] M. Alard and R. Lassalle, \Principles of modulation and channel coding for digital broadcasting for mobile receivers," EBU Review, pp. 47{69, August 1987. 6] S. Weinstein and P. M. Ebert, \Data transmission by frequencydivision multiplexing using the discrete fourier transform," IEEE Trans. Commun., vol. 19, pp. 628{634, October 1971. 7] K. Fazel, S. Kaiser, P. Robertson, and M. J. Ruf, \A concept of a recon gurable digital-TV/HDTV transmission scheme with exible channel coding & modulation for terrestrial broadcasting," in Proc. International Workshop on HDTV '94, October 1994. 8] R. De Gaudenzi, C. Elia, and R. Viola, \Bandlimited quasisynchronous CDMA: A novel satellite access technique for mobile and personal communication systems," IEEE J. Selected Areas Commun., vol. 10, pp. 328{343, February 1992. 9] G. L. Turin, \Introduction to spread-spectrum antimultipath techniques and their application to urban digital radio," Proceedings of the IEEE, vol. 68, pp. 328{353, March 1980. 10] M. Williams and Sloane, The theory of error-correcting codes. North-Holland Mathematical Library, second ed., 1978. 11] M. B. Pursley, \Performance evaluation for phase-coded spreadspectrum multiple-accesscommunication-PartI: System analysis," IEEE Trans. Commun., vol. 25, pp. 795{799, August 1977. 12] E. Geraniotis and B. Gha ari, \Performance of binary and quaternary direct sequence spread-spectrum multiple-access systems with random signature sequences," IEEE Trans. Commun., vol. 39, pp. 713{724, May 1991. 13] D. Divsalar and M. K. Simon, \The design of trellis coded MPSK for fading channels: Performance criteria," IEEE Trans. Commun., vol. 36, pp. 1004{1012, September 1988. 14] R. G. McKay, P. J. McLane, and E. Biglieri, \Error bounds for trellis-coded MPSK on a fading mobile satellite channel," IEEE Trans. Commun., vol. 39, pp. 1750{1761, December 1991. 15] Cost-207 WG1, \Proposal on channel transfer functions to be used in GSM tests late 1986," tech. rep., CEPT Paris, September 1986.
et al.