PERFORMANCE COMPARISON
`OF DIFFERENT MULTIPLE ACCESS SCHEMES FOR THE DOWNLINK
`OF AN OFDM COMMUNICATION SYSTEM
`
`Hermann Rohling and Rainer Griinheid
`
`The authors are with the Institute of Telecommunications,
`Technical University Braunschweig, Germany
`E-Mail: rohling@ifn.ing.tu-bs.de
`
`Abstract - In this paper, a comparison of different multiple
`access schemes for the downlink of an OFDM communication
`system is performed. The access methods OFDM-TDMA,
`OFDM-FDMA, and OFDM-CDMA are described and com-
`pared with regard to flexibility and computation complexity.
`Furtheron, the performance of the different approaches is
`quantitatively analysed, comparing the resulting BER for the
`given OFDM system in a time-variant and frequency-selective
`radio channel.
`
`I. INTRODUCTION
`
`In recent years, the well-known multicarrier transmission tech-
`nique (OFDM) has not only been applied to broadcast systems
`like digital terrestrial television and radio (DTVB, DAB), but
`has also been discussed for use in mobile communications.
`Since the performance of a multi-user system highly depends
`on a suitable multiple access technique, the question arises as
`to concepts for an efficient multiple access scheme specific for
`the OFDM transmission technique.
`For single carrier systems, a variety of techniques is known,
`such as the conflict-free access methods FDMA, TDMA, and
`CDMA. In this paper, different multiple access schemes are
`described and compared for the downlink of an OFDM com-
`munication system, namely OFDM-FDMA, OFDM-TDMA,
`and OFDM-CDMA.
`The paper is organized as follows. After a description of the
`considered OFDM system and the radio channel model
`(section II), the principles of the different multiple access
`schemes are outlined (section III.A) and compared concerning
`the achievable flexibility (user data rate, reaction to radio
`channel behaviour) and computation complexity (III.B and
`III.C, respectively). Quantitative results with respect to BER
`performance in a frequency-selective radio channel with
`Rayleigh fading on each subcarrier are presented in section
`III.D. The paper ends with some conclusive remarks.
`
`II. SYSTEM PARAMETERS AND RADIO CHANNEL
`MODEL
`
`In the OFDM communication system considered throughout
`this paper, a frame structure according to Fig. 1 is assumed.
`
`The total frame consists of 128 OFDM symbols, where sym-
`bols for up- and downlink are arranged in a time division du-
`plex (TDD) structure. In the uplink, a TDMA approach is
`applied, and reference symbols are needed for each user due to
`incoherent demodulation (see below). In contrast, three differ-
`ent multiple access schemes are considered for the downlink,
`as already mentioned.
`The bandwidth of approx. 8 MHz, processed in each receiver,
`is subdivided into K=256 subcarriers which are spaced by
`AF=28 kHz. To cope with multipath propagation, a guard
`interval of length TG=5 !is is added, yielding a total symbol
`duration of T5=40.7 The subcarriers are modulated with
`DQPSK when OFDM-FDMA and TDMA are analysed,
`whereas OFDM-CDMA requires a coherent demodulation, and
`consequently a QPSK modulation is applied in this case. The
`total data rate resulting from these parameters amounts to
`12.57 Mbit/s if all subcarriers are used.
`Forward error correction is realized in the form of punctured
`convolutional codes with soft decision.
`
`.r4c
`7:,
`iiicc OFDM
`.o
`E 2 : -2 Symbols
`y
`u)0 o
`c oc for
`
`SignallingSiing
`WN CA U)
`
`OFDM Symbols for
`User Data
`(User 0 ... U-1)
`
`FDMA / TDMA / COMA
`
`Downlink
`Frame
`
`OFDM
`Symbols
`for
`User 0
`
`OFDM
`Symbols
`for
`User 1
`
`cc
`
`• • •
`
`TDMA
`
`OFDM
`Symbols
`for
`User U-1
`
`cc
`
`Uplink
`Frame
`
`128 OFDM Symbols in Total
`(Uplink + Downlink)
`
`Fig. 1: Frame structure of the considered OFDM transmission
`system
`
`In the considered system, a mobile receiver is assumed in a
`cellular environment (cell radius 250 m). The radio channel
`is chosen to be frequency -selective and time -variant. The
`power delay profile, which has been derived from measure-
`ments, is exponentially decreasing, with a maximum delay of 5
`
`0 -7803 -3659 -3/97 $10.00 ©1997 IEEE
`
`1365
`
`Page 1 of 5
`
`

`
`1.ts. As for the Doppler power spectrum, a Jakes distribution
`with a maximum doppler frequency of fmnax=70 Hz is taken
`into account (f0=2.5 GHz).
`
`and it is not fixed, but can be adapted, e.g. for each downlink
`frame. Aspects concerning an "intelligent" adaptive subcarrier
`management are addressed in section III.B.
`
`III. ACCESS SCHEMES FOR OFDM
`
`This section concisely outlines the multiple access schemes
`OFDM-FDMA, OFDM-TDMA, and OFDM-CDMA, before
`comparisons are drawn with respect to flexibility and compu-
`tation complexity. Furthermore, simulation results concerning
`the BER performance are presented and discussed.
`
`A. Description
`
`A general schematic overview of the considered multiple ac-
`cess schemes is shown in Fig. 2, where the time-frequency
`plane is depicted, and each square represents one subcarrier in
`a specific OFDM symbol. The allocation of subcarriers to the
`users is visualized by different shades of gray.
`
`f I
`
`Fig. 2: Overview of different multiple access schemes for
`OFDM: FDMA (top), TDMA (middle), CDMA (bottom)
`
`An FDMA approach for OFDM implies that an unambiguous
`subset of all carriers is assigned to each user (Fig. 2, top). This
`allocation represents a purely digital subcarrier management,
`
`1366
`
`In an OFDM-TDMA system (Fig. 2, middle), the total band-
`width (all subcarriers) is allocated to one user for the duration
`of several OFDM symbols within a downlink or uplink frame.
`The performance of an OFDM-TDMA system has been ana-
`lysed in [1].
`
`Finally, for OFDM-CDMA (see [2]), a subset of orthogonal
`codes is assigned to each user, and the information symbols are
`spread in the frequency domain over multiple subcarriers (Fig.
`2, bottom). Spreading is performed over L (out of a total of K)
`subcarriers, and in each group of K/L subcarriers, a certain
`number of M orthogonal codes (where 11/1_.) is used.
`
`B. Flexibility
`
`An efficient multiple access scheme should grant a high
`flexiblity when it comes to the allocation of time-bandwidth
`ressources. On the one hand, the behaviour of the frequency-
`selective radio channel should be taken into account, while on
`the other hand the user requirements for different and/or
`changing data rates have to be met.
`
`In OFDM-FDMA, this flexibility can be accomplished by
`suitably choosing the subcarriers associated with each user.
`Here, the fact that each user experiences a different radio
`channel can be exploited by allocating only "good" subcarriers
`with high SNR to each user. Furthermore, the number of sub-
`channels for a specific user can be varied, according to the
`required data rate.
`
`Since the OFDM-TDMA concept allocates the total bandwidth
`to a single user, a reaction to different subcarrier attenuations
`could consist of leaving out highly distorted subcarriers, see
`[1]. Concerning different/varying data rates, the number of
`OFDM symbols per user in each frame can be adapted ac-
`cordingly.
`
`Both for OFDM-FDMA and TDMA, the concept of an adap-
`tive modulation of different subcarriers, as proposed in [1], is
`conceivable to further exploit channel state information.
`
`In OFDM-CDMA, a flexibility lies in the allocation of all
`available codes to the users, depending on the required data
`rates. As OFDM-CDMA is applied using coherent modulation,
`the necessary channel estimation provides information about
`the subcarrier attenuations; this information can be used when
`performing an equalization in the receiver [7].
`
`C. Computation Complexity and Signalling
`
`The complexity which is associated with the different multiple
`access algorithms highly depends on the adaptive methods that
`
`Page 2 of 5
`
`

`
`A number of U=8 users is chosen as an example, where each
`user transmits the same data rate in the downlink. For TDMA,
`this implies an equal number of OFDM symbols in each
`downlink frame (all subcarriers used). In the case of FDMA,
`the total bandwidth is divided into 8 groups of 32 subcarriers.
`It is assumed that each user can be allocated the most suitable
`subcarriers, i.e. with highest SNR (optimal subearrier alloca-
`tion without "collisions" between users). To gain access to
`information about the radio channel for different users, the
`OFDM symbols received in the uplink frame within the TDD
`structure can be analysed.
`
`As for OFDM-CDMA, each user is allocated 32 Walsh-
`Hadamard codes. The code length L is set to 16 (with full load,
`i.e. 16 codes per group). Wiener filtering (see [31) is used as a
`detection method.
`When coherent demodulation is applied, ideal knowledge of
`the channel transfer function is assumed.
`Some of the above mentioned assumptions are summarized
`below.
`
`OFDM-TDMA:
`incoherent demodulation
`•
`no channel adaptation (all subcarriers used)
`•
`
`OFDM-FDMA:
`•
`incoherent demodulation
`channel adaptation: optimal allocation of subcarriers with
`•
`highest SNR
`
`OFDM-CDMA:
`•
`coherent demodulation
`ideal knowledge of channel transfer function
`•
`full code set used
`•
`
`In Fig. 3, Fig. 4, and Fig. 5, the performance of OFDM-
`TDMA, CDMA, and FDMA is plotted when using convolu-
`tional codes (memory length 6) with different code rates. It can
`clearly be seen that OFDM-CDMA leads to a better perform-
`ance than OFDM-TDMA, due to different demodulation
`schemes and an additional diversity gain for CDMA. A differ-
`ence of about 4 dB is observed at a BER of 10 -3 .
`From Fig. 5, it becomes obvious that FDMA outperforms both
`CDMA and TDMA, under the assumption that an ideal subcar-
`rier allocation (each user is assigned the subcarriers with high-
`est SNR) is accomplished.
`
`When comparing different access schemes, the demodulation
`process (incoherent or coherent) has to be taken into account.
`In [6], multiple access schemes are compared for coherent
`(BPSK) modulation in the coded case, where CDMA outper-
`forms TDMA at high code rates whereas at low code rates, the
`BER curves are similar to those for FDMA and TDMA. The
`latter two show identical performance because no adaptive
`subcarrier allocation is performed.
`
`are applied in each system. Generally, some sort of signalling
`information (see Fig. 1) about the subeanrier/code allocation
`has to be created at the base station and to be processed at the
`mobile receivers.
`If in the case of OFDM-TDMA, weak subearriers are left out
`or adaptive modulation/coding is applied, additional informa-
`tion has to be transmitted; however, for the procedure of leav-
`ing out subcarriers, in [1] a procedure is presented that mainly
`does without signalling. As to OFDM-CDMA, the computa-
`tion complexity is increased by a matrix multiplication in the
`transmitter/receiver for spreading and despreading, respec-
`tively. Furthermore, a channel estimation procedure is neces-
`sary for coherent demodulation.
`In the following, some comparisons are made which refer to
`the required signalling information in the downlink for a sub-
`carrier/code allocation. If a number of U users (maximum set
`to 32) and K subcarriers is considered, the necessary number
`of bits amounts to (frame length: 128 OFDM symbols):
`
`Number
`of bits
`
`7 U
`7U
`
`5K
`
`7U
`7 U
`
`5K
`
`_C
`
`ontents
`
`FDMA/TDMA/CDMA -allocation f. uplink
`start slot in uplink frame for each user
`end slot in uplink frame for each user
`DMA - allocation for downlink
`allocation table for all subcarriers
`TDMA - allocation for downlink
`start slot in downlink frame for each user
`end slot in uplink frame for each user
`CDMA - allocation for downlink
`allocation table for all codes
`
`—F
`
`_
`
`Hence, for K=256 subcarriers and user numbers of U=8, 16,
`and 32, furthermore considering DQPSK modulation and a
`code rate of 1/3 for FEC, the following overhead (relative to
`the frame length of 128) has to be used for the above men-
`tioned signalling information:
`
`Access scheme
`OFDM-FDMA
`OFDM-TDMA
`OFDM-CDMA
`
`8 users
`6.4%
`1.0%
`6.4%
`
`—
`16 users
`6.9%
`2.1%
`6.9%
`
`32 users
`7.9%
`4.1%
`7.9%
`
`Thus, it can be seen that the required overhead is higher for
`FDMA or CDMA because a user number (5 bits) has to be
`transmitted for each subcarrier (FDMA) / code (CDMA). The
`maximum overhead is below 8 % in all cases.
`
`D. Comparison of BER Performance
`
`For a quantitative comparison of the BER performance of all
`considered access schemes, the system parameters and channel
`model described in section II are used in the simulation run.
`Henceforth, only the downlink will be analysed.
`
`1367
`
`Page 3 of 5
`
`

`
`cc
`Lo
`co
`
`1E+00
`
`1E-01
`
`lE 02
`
`1E-03
`
`1E-04
`
`1E-05
`
`1E-06
`
`0
`
`4(- R=1 /3
`-Ef R=1 /2
`R=2/3
`
`12
`
`14
`
`16
`
`Fig. 3: Performance of OFDM-TDMA (incoherent demodula-
`tion) in a frequency-selective radio channel with Rayleigh
`fading on each subcarrier (all subcarriers used)
`
` 11
`
`moisomms■■••■
`MI■IMEI■16■6
`MZIMIIMM■-
`ammes■■■■■• NE&
`
`ra■
`
`12
`
`14
`
`16
`
`Fig. 4: Performance of OFDM-CDMA (coherent demodula-
`tion, ideal channel estim.) in a frequency-selective radio chan-
`nel with Rayleigh fading on each subcarrier (full code set
`used)
`
`1E+00
`
`lE 01
`
`1E 02
`
`CC
`111 1 E 03
`
`1E-04
`
`1E-05
`
`1E-06
`
`0
`
`12
`
`14
`
`16
`
`Fig. 5: Performance of OFDM-FDMA (incoherent demodula-
`tion) in a frequency-selective radio channel with Rayleigh
`fading on each subcarrier (optimal subcarrier allocation)
`
`A coherent demodulation implies that pilot symbols are in-
`serted in an OFDM symbol. This redundancy can, on the other
`hand, be used for a more powerful channel coding in the case
`of incoherent demodulation because no channel estimation is
`required here. Thus, a comparison at the same data rate leads
`to different code rates for FDMA/TDMA (incoherent) and
`CDMA (coherent).
`Clearly, the amount of pilot symbols needed for channel esti-
`mation depends on the characteristics of the radio channel
`(maximum delay, doppler frequency). If approx. 11% redun-
`dancy in the form of pilot symbols is considered as an exam-
`ple, a code rate of 2/3 can be used for incoherent demodulation
`(FDMA, TDMA) and of 3/4 for OFDM-CDMA. The corre-
`sponding results for a fixed user data rate are summarized in
`Fig. 6.
`
`1E+00
`
`1E-01
`
`1E-02
`
`1E03
`
`1E 04
`
`1E-05
`
`CC
`
`-x-TDMA, R=213
`-8-CDMA, R=3/4
`H-FDMA, R=2/3
`
`1E-06
`
`0
`
`2
`
`4
`
`6
`
`8
`SIN [dB]
`Fig. 6: Performance comparison of OFDM-TDMA, -FDMA
`and -CDMA for a fixed user data rate
`
`10
`
`12
`
`14
`
`16
`
`The performance gain of OFDM-FDMA vs. OFDM-CDMA
`amounts to 4 dB at a BER of 10-3. While OFDM-CDMA is
`still superior to OFDM-TDMA, the remaining difference is 3
`dB at a BER of 10-3 . Furthermore, taking into consideration
`that an ideal channel estimation is assumed for OFDM-
`CDMA, a performance degradation can be expected when a
`real channel estimation procedure is implemented. In [4] and
`[5], a loss of approx. 1.5 dB is reported for a channel estima-
`tion with filtering in the time and frequency domain. Conse-
`quently, this means that OFDM-TDMA performs only about
`1.5 dB worse than OFDM-CDMA under these assumptions.
`
`As a result from this comparison, it can be observed that
`OFDM-FDMA leads to the best performance for the consid-
`ered system because information about the radio channel is
`taken into account, allocating only suitable subcarriers with
`high SNR to each user. Clearly, the advantage of FDMA vs.
`CDMA depends on the system parameters and the channel
`characteristics, e.g. the possibility of an adaptation in a time-
`variant channel (frame length vs. coherence time of the chan-
`nel). Moreover, an adaptation to the radio channel for CDMA
`is also conceivable by adaptively choosing the subset of carri-
`ers over which the modulation symbols are spread for each
`user. This is subject to further investigation.
`
`1368
`
`Page 4 of 5
`
`
`
`

`
`International Symposium on Spread Spectrum Techniques and
`Applications (ISSSTA) 1996, Mainz, Germany, pp. 1366-1370
`
`[7] T. Miller, H. Rohling, R. Griinheid: "Comparison of Dif-
`ferent Detection Algorithms for OFDM-CDMA in Broadband
`Rayleigh Fading", Proc. VTC 1995, Chicago, pp. 835-838
`
`IV. CONCLUSION
`
`In this paper, the performance of different multiple access
`schemes for the downlink of an OFDM transmission system
`has been studied. The characteristics of the access methods
`OFDM-TDMA, OFDM-FDMA, and OFDM-CDMA have
`been outlined. Some remarks about computation complexity
`and flexibility with regard to data rate and radio channel ad-
`aptation have been made, before the BER performance in a
`frequency-selective and time-variant radio channel has been
`compared in the coded case (convolutional codes with soft
`decision). As a result, OFDM-FDMA, with an adaptation to
`the user-specific radio channels, leads to the best performance
`for the given parameters, if an optimal subcarrier allocation is
`assumed, followed by OFDM-CDMA and OFDM-TDMA.
`With a fixed user data rate, a gain of about 4 dB of FDMA vs.
`CDMA at a BER of 10 -3 is observed, while OFDM-TDMA
`performs approx. 3 dB worse. If a loss due to real channel
`estimation is considered, the difference to OFDM-CDMA can
`be expected to shift in favour of OFDM-TDMA and FDMA.
`Finally, it should be emphasized that the analysed TDMA and
`FDMA approach with incoherent demodulation can do without
`a channel estimation procedure, thus reducing the computation
`complexity in each mobile receiver.
`
`V. ACKNOWLEDGEMENT
`
`This work has been supported by the Deutsche Forschungs-
`gemeinschaft.
`
`REFERENCES
`
`[1] H. Rohling, R. Grtinheid: "Performance of an OFDM-
`TDMA Mobile Communication System", Proc. VTC 1996,
`Atlanta, pp. 1589-1593
`
`[2] N. Yee, J.-P. Linnartz: "Controlled Equalization of Multi-
`Carrier CDMA in an indoor Rician Fading Channel", Proc.
`VTC 1994, Stockholm, pp. 1665-1669
`
`[3] N. Yee, J.-P. Linnartz: "Wiener Filtering of Multi-Carrier
`CDMA in Rayleigh Fading Channel", Proc. PIMRC 1994, The
`Hague, pp. 1344-1347
`
`[4] M. Sandell: "Design and Analysis of Estimators for Multi-
`carrier Modulation and Ultrasonic Imaging", Ph.D. thesis,
`Lulea University of Technology, Lulea, Sweden, Sept. 1996
`
`[5] P. Mier: "TCM on Frequency-Selective Land-Mobile
`Fading Channels", Proc. of the 5th International Workshop on
`Digital Communications, Elsevier Science Publishers B.V., M.
`Luise and E. Biglieri (eds.), Pisa, Italy, Sept.1991
`
`[6] S. Kaiser: "Trade-off between Channel Coding and
`Spreading in Multi-Carrier CDMA Systems", Proc. of IEEE
`
`1369
`
`Page 5 of 5