Abstract. OFDM-CDMA also referred as MC-CDMA is a spread-spectrum transmission technique

Transcription

1 I Abstract OFDM-CDMA also referred as MC-CDMA is a spread-spectrum transmission technique introduced in Since then, numerous technical papers about OFDM-CDMA have been published. Moreover, OFDM-CDMA is considered to be a promising candidate for the air interface of the fourth generation (4G) wireless communication systems. Interleave-division multiple-access (IDMA) is a recently proposed multiple access scheme, which employs random interleavers as the only method for user separation. As a particular case of CDMA, IDMA inherits many distinguished features of well-studied CDMA. Furthermore, it allows a low-cost turbo-type multiuser detection (MUD) algorithm applicable to system with a large number of users, which is crucial for high-rate multiple access communication. The wor presented in this thesis is completely devoted to the investigation of an OFDM-IDMA scheme. The proposed scheme employs IDMA instead of CDMA in the OFDM-CDMA scheme and combines advantages of both OFDM and IDMA. With sufficient guard intervals, OFDM can completely remove intersymbol interference (ISI). IDMA with a chip-by-chip (CBC) iterative multi-user detector can overcome both crosscell and intra-cell multiple access interference (MAI) problems efficiently. We first present the basic principles of OFDM-IDMA transmitter and receiver. Comparative studies between OFDM-IDMA and OFDM-CDMA are carried out. We also compare OFDM-IDMA with IDMA with LLRC algorithm. Thans to the independent processing of ISI and MAI, OFDM-IDMA offers better performance than plain IDMA with LLRC in terms of both BER and complexity. Then we derive a simple and effective performance analysis method for OFDM- IDMA based on SNR evolution. To support large number of users and achieve high rates, we optimize the transmitted power profiles with the aid of SNR evolution technique. Several power allocation methods are presented, which are selected according to the system configurations and channel conditions. The comparisons between OFDM-IDMA and i

2 OFDM-TDMA based on BICM are carried out, which demonstrates a spectral-power efficiency advantage of non-orthogonal multiple access approaches over orthogonal ones. The proposed OFDM-IDMA scheme provides a competitive alternative to conventional OFDM-CDMA scheme for future high-rate multiuser communications over multipath fading channels. ii

3 II Acnowledgement I would lie to express my sincere appreciation and gratitude to my supervisor, Prof. LI Ping for his support and help during my studies. He has been a great source of nowledge and has inspired me with many ideas which are very useful for my research. His guidance, encouragement and patience made this thesis possible. I am very grateful to Dr. Lihai LIU, Dr. Keying WU, Dr. W. K. LEUNG, Dr. Shenghui SONG, Mr. Jun TONG, Mr. Peng WANG, Mr. Xiaojun YUAN, Mr. Qinghua GUO, Mr. Hao WU and Ms. Shuling CHE for their invaluable advice and assistance during my studies at City University of Hong Kong. I really enjoy the harmony atmosphere that has been built by all members in our group. I am indebted to my labmates as well as all of my friends here for their encouragement and support. Without them, life would have been difficult. Finally, my deepest love and gratitude is devoted to my parents. They are always a source of love, supporting and encouragement. To them, I dedicate this thesis. iii

4 Contents I Abstract i II Acnowledgement iii 1 Introduction The Development of Wireless Communication Systems Research Problems and Objectives Research Contributions Thesis Organization Preliminaries: Multipath Channel Models, Multicarrier CDMA and IDMA Multipath Channel Models Radio Channel Characteristics Channel Modeling Multicarrier Modulation Multicarrier CDMA OFDM-CDMA Transmitter OFDM-CDMA Receiver IDMA over Multipath Channels Conclusions OFDM-IDMA Motivation of OFDM-IDMA Scheme Principles of OFDM-IDMA OFDM-IDMA vs. OFDM-CDMA OFDM-CDMA with PIC Detection OFDM-CDMA with CBC Detection OFDM-IDMA with Low-rate FEC Codes iv

5 3.4 OFDM-IDMA vs. IDMA with LLRC Performance Comparison between OFDM-IDMA and IDMA with LLRC Complexity Comparison between OFDM-IDMA and IDMA with LLRC Conclusions Performance Evaluation and Optimization Based on SNR Evolution Introduction Performance Evaluation Performance Analysis Method for OFDM-IDMA Base on SNR Evolution Numerical Examples Power Allocation Transmitters with the Knowledge of Average Channel Gains Transmitters without the Knowledge of Average Channel Gains Conclusions Conclusions and Future Research Wor Conclusions Future Research Wor Appendix A: Linear Programming Method 68 Appendix B: The Power Profiles Obtained by damping Method 70 References 72 Publications 77 v

6 List of Figures 1 (a) An illustration of the IDMA-OFDM scheme. (b) An equivalent system modeling of (a) Bandwidth consuming illustration: (a) General multicarrier modulation, (b) OFDM modulation Bloc diagram of a OFDM digital communication system Bloc diagram of uplin OFDM-CDMA transmitter with K simultaneous users Bloc diagram of the detector for OFDM-CDMA system. Note, the MMSE detection bloc comprises the despreading process Transmitter and receiver structures of an IDMA scheme with K simultaneous users Transmitter and receiver structures of the OFDM-IDMA scheme with K simultaneous users, where π is the interleaver of user-. For simplicity, the above figure does not include the insertion and removing of the guard intervals between OFDM symbols Performance comparison between OFDM-IDMA (solid line) and OFDM- CDMA (dashed line) with different iteration numbers. K =16 and S= Performance comparison between OFDM-IDMA and OFDM-CDMA. CBC detection algorithm is adopted by both schemes. K = 32, S = 16, and iteration number = Performance comparison between OFDM-IDMA based on turbo-hadamard code (solid line) and OFDM-CDMA (dashed line) based on turbo code. User number K is mared in the figure vi

7 11 Performance comparison between OFDM-IDMA (solid line) and IDMA with LLRC (dashed line) in quasi-static Rayleigh fading multipath channels with different path numbers L = 1, 2, 4, 8. K = 32 and S = 16. Iteration number= Performance comparison between OFDM-IDMA (solid line) and IDMA with LLRC (dashed line) in quasi-static Rayleigh fading multipath channels with different iteration number= 1, 5, 10. The path number L = 8. K = 40 and S = Performance comparison between OFDM-IDMA (solid line) and IDMA with LLRC (dashed line) in quasi-static Rayleigh fading multipath channels with different user number K = 1, 16, 32, 40, 56. The path number L = 8. Iteration number= 10 and S = Comparison between simulation results (solid line) and corresponding evolution results (dashed line) of OFDM-IDMA in uncorrelated Rayleigh fading channel. K = 2S and sum rate = 2 bits/symbol for all curves Comparison between simulation results (solid line) and corresponding evolution results (dashed line) of OFDM-IDMA with a spreading factor S=16 in uncorrelated Rayleigh fading channel The lowest-power users BER performance of a OFDM-IDMA system after power allocation obtained by simulation (solid line) and evolution (dashed lines) over uncorrelated fading channel. Iteration number = 10 and S = The BER performance of OFDM-IDMA with sum rate = 2, 3, 5 bits/symbol in an uplin cellular scenario with path loss, log-normal shadowing and uncorrelated Rayleigh fading channel conditions vii

8 18 The required average transmitted power versus P out (G 0 ) of OFDM-IDMA with sum rate = 2, 3, 5 bits/symbol in an uplin cellular scenario with path loss, log-normal shadowing and uncorrelated Rayleigh fading channel conditions The BER performance of OFDM-IDMA with sum rate = 2, 3, 5 bits/symbol in an uplin cellular scenario with path loss, log-normal shadowing and multipath quasi-static Rayleigh fading channel conditions The required average transmitted power versus P out (G 0 ) of OFDM-IDMA with sum rate = 2, 3, 5 bits/symbol in an uplin cellular scenario with path loss, log-normal shadowing and multipath quasi-static Rayleigh fading channel conditions The comparison between OFDM-IDMA using superposition coding and OFDM-TDMA with throughput = 3 bits/symbol in uncorrelated Rayleigh fading channels The comparison between OFDM-IDMA using superposition coding and OFDM-TDMA with throughput = 3 bits/symbol in an uplin cellular scenario with path loss, log-normal shadowing and uncorrelated Rayleigh fading channel conditions Flowchart of damping power allocation method based on SNR evolution The lowest-power users BER performance of a OFDM-IDMA system after power allocation obtained by simulation (solid line) and evolution (dashed lines) over quasi-static Rayleigh fading multipath channel. Iteration number = 30 and S = 8. Safe margin T = The lowest-power users BER performance of a OFDM-IDMA system after power allocation obtained by simulation (solid line) and evolution (dashed lines) over quasi-static Rayleigh fading multipath channel. Iteration number = 30 and S = 8. Safe margin T = viii

9 List of Tables 1 Complexity comparison between OFDM-IDMA and IDMA with LLRC with with K = 32, L = 8, iteration number (it) = 10 and N c = The optimized received power profiles obtained by linear programming method The power profiles with safe margin T = 0.1 for Fig The power profiles with safe margin T = 0.12 for Fig ix

10 1 Introduction 1.1 The Development of Wireless Communication Systems Since the beginning of the 20th century, technologies have evolved remarably to provide new methods and products for wireless communications. Especially in the past two decades, wireless communication services were penetrating into our society with an explosive growth rate. As this thesis deals with a spread-spectrum multiple-access communication scheme for cellular communication systems, a brief historical review of cellular systems is provided as follows. First Generation (1G) Wireless Communication Systems: The 1G systems are characterized by the fact that they are all based on the analog technologies. Some representatives of 1G cellular systems are the Advanced Mobile Phone System (AMPS) in USA, Nordic Mobile Telephone (NMT) in Scandinavia and the Total Access Communication System (TACS) in UK [1]. They are designed to carry voice transmission only. In these systems, each user has a unique frequency band. This multiple-access technique separates the signals of different users in frequency domain, which is called frequency-division multiple-access (FDMA). With the introduction of 1G systems, the worldwide mobile maret experienced annual growth rates of 30% to 50% and there are nearly 20 million subscribers by 1990 [2]. Second Generation (2G) Wireless Communication Systems: The 2G cellular systems are all based on the digital technologies. The most popular 2G standards include three time-division multiple-access (TDMA) standards and one code-division multipleaccess (CDMA) standard. The Global System for Mobile communications (GSM) is the first operated digital cellular system based on TDMA, whose commercial operation began in 1991 in Europe. The other two TDMA based standards are Interim Standard 54 (IS-54) in North America and Pacific Digital Cellular (PDC) in Japan. The only CDMA based 2G cellular standard is Interim Standard 95 (IS-95). Since the mid 1990s, the cel- 1

11 lular communications industry has witnessed explosive growth. There are more than 600 million cellular subscribers worldwide in late 2001 [3]. Most of today s cellular systems belong to 2G. Third Generation (3G) Wireless Communication Systems: The 3G standards have been developed specially to support high-rate data services such as high-speed internet access, video stream and high quality image transmission. International Mobile Telecommunications-2000 (IMT-2000) is the global standard for 3G wireless communications, defined by a set of interdependent International Telecommunication Union (ITU) Recommendations. IMT-2000 provides a framewor for worldwide wireless access by lining the diverse systems based networs [4]. The most important 3G standards are the European and Japanese Wideband-CDMA (WCDMA), the American CDMA2000, and the Chinese Time-Division Synchronous CDMA (TD-SCDMA). All three standards are based on CDMA and operate around 2 GHz. 3G systems have recently been launched or are planned to be launched in many countries and regions. There are four companies CSL Limited, Hutchison 3G HK Limited, SmarTone 3G Limited and SUNDAY 3G (Hong Kong) Limited who have obtained 3G licenses from office of the telecommunications authority in Hong Kong in September 2001 [5]. CSL Limited launched Hong Kong s first 3G services in December, Fourth Generation (4G) Wireless Communication Systems: Even before 3G systems were being deployed, researchers started the investigations on possible techniques for 4G systems. The explicit 4G standard is still not confirmed. In June 2003, ITU approved the recommendation ITU-R M.1645 Framewor and overall objectives of the future development of IMT-2000 and systems beyond IMT-2000 [6]. The document states the capabilities as well as possible technologies for 4G systems. Important features of 4G wireless systems includes: 4G needs to support data rates of up to 100 Mbps for high mobility such as mobile access and up to 1 Gbps for low mobility such as local wireless access. 2

12 4G uses of pacet-based architectures which will offer increased system security and reliability, intersystem mobility and interoperability capabilities. 4G satisfies future requirements for universal wireless networ that will provide high data rates and a seamless interface with a wireline bacbone networ. The above features impose technical challenges on system design. There are some promising technologies (mainly related to physical layer) for 4G wireless systems such as sophisticated forward error correcting (FEC) codes (e.g., [7], [8], [9]), CDMA with improved detection algorithms (e.g., [10], [11], [12]), multicarrier modulation including orthogonal frequency-division multiplexing (OFDM) and adaptive resource management. At the moment there are many research institutions andindustrial companies which investigate the appropriateness of different techniques for the 4G air interface. In this thesis, the research wor is entirely devoted an OFDM based multicarrier multiple-access scheme for future wireless communications as introduced in the next section. 1.2 Research Problems and Objectives In principle, OFDM together with orthogonal multiple access scheme (TDMA or FDMA) can suppress intersymbol interference (ISI) as well as establish orthogonality among same-cell users, but it is sensitive to cross-cell multiple-access-interference (MAI). On the other hand nonorthogonal multiple access scheme CDMA alleviates the cross-cell MAI problem by the spread-spectrum technique. OFDM-CDMA combines advantages of both OFDM and CDMA, in particular the simple treatment of ISI and mitigation of cross-cell MAI. OFDM-CDMA is a promising technique for the implementation of the physical layer in 4G cellular systems. However compared with counterpart orthogonal multiple access methods such as OFDM-FDMA (also referred as OFDMA), OFDM-CDMA alleviates the cross-cell MAI problem by the spread-spectrum technique at the expense of allowing intra-cell MAI. MUD is a promising technique for the intra-cell MAI problem, but 3

13 the complexity related to MUD has been a major concern for its practical application. Interleave-division multiple-access (IDMA) with a low-cost turbo-type iterative detection algorithm can overcome both cross-cell and intra-cell MAI problem effectively. As so, we consider a hybrid scheme orthogonal frequency-division multiplexing interleavedivision multiple-access (OFDM-IDMA) in this thesis, which has been proposed in [30], [31] recently. IDMA transmitter OFDM modulator Multipath fading channel OFDM demodulator IDMA receiver (a) IDMA transmitter Equivalent parallel sub-channels (b) IDMA receiver Figure 1: (a) An illustration of the IDMA-OFDM scheme. (b) An equivalent system modeling of (a). The investigated OFDM-IDMA scheme is illustrated in Fig. 1. It consists of an IDMA layer cascaded with an OFDM layer. The output of the IDMA layer is used as the input to the OFDM layer at the transmitter and vice versa at the receiver. The OFDM modulator and demodulator form an orthogonal transform pair, so they (together with the channel) can be modeled by a set of equivalent parallel sub-channels (each corresponding to an OFDM sub-carrier). The resultant equivalent system is shown in Fig. 1(b). Based on Fig. 1(b), we expect that most IDMA principles developed in [12], [22] can be applied here. Two main objectives are set in this wor: The first goal is to evaluate the performance of the proposed OFDM-IDMA scheme in multipath fading channels. Comparative studies between OFDM-IDMA and 4

14 other existing multiuser communication systems will be carried out. The second goal is to derive performance analysis technique for OFDM-IDMA. Based on performance prediction technique, we can further improve spectral-power efficiency of OFDM-IDMA scheme by power allocation methods. 1.3 Research Contributions The focus of this thesis has been concentrated on the investigation of the OFDM- IDMA scheme in uplin scenario. The first contribution of this thesis is carrying out comparative studies between OFDM- IDMA and OFDM-CDMA. From the simulation results, we can see that OFDM-IDMA outperforms OFDM-CDMA in terms of BER performance mainly thans to the efficient chip-by-chip detection algorithm. The CBC detection algorithm can also be applied to OFDM-CDMA scheme to improve performance. With a low-rate forward error correction (FEC) code, OFDM-IDMA can achieve additional coding gain by removing spreading and devoting entire bandwidth expansion to FEC coding. We also carry out comparisons between OFDM-IDMA and IDMA with LLRC. The same research wor on comparisons between OFDM-IDMA and IDMA with LLRC have been done in [31]. We provide more simulations to confirm their conclusions. The details about these research wors are presented in Chapter 3. The second contribution is improving the power efficiency and spectrum efficiency of OFDM-IDMA scheme by power allocation methods. We first derive a fast and effective SNR evolution technique to access the performance of OFDM-IDMA systems with lowrate repetition codes. Several power allocation methods are addressed, which are selected according to the system configurations and channel conditions. When transmitters have the nowledge of average channel gains, we adopt a two-step power allocation method. In this case, OFDM-IDMA outperforms OFDM-TDMA in terms of BER performance over multipath fading channels. When the nowledge of average channel gains is not available 5

15 at the transmitters, we develop a damping method, which can set a safe margin for each user according to its own power level to overcome the unnown fluctuations of the channel. The details about these research wors are presented in Chapter Thesis Organization The remaining of the thesis is organized as follows. Chapter 2 provides preliminaries on multicarrier CDMA, IDMA and multipath channel models. We first review the history and basic principle of conventional OFDM- CDMA. Several detection strategies for OFDM-CDMA scheme are elaborated. Then a recently proposed multiple access scheme referred as IDMA is presented. In contrast to CDMA where spreading sequence are employed for user separation, distinguishes singles from different users by user-specific chip-level interleavers. We also describe radio channel characteristics and channel models for maltipath radio propagation in this part. Chapter 3 is devoted to OFDM-IDMA systems. Comprehensive comparisons between OFDM-IDMA and OFDM-CDMA are carried out. We also compare the complexity and BER performance between OFDM-IDMA and IDMA with log-lielihood ratio combining (LLRC) approach in multipath fading environment. Various simulation results of OFDM-IDMA scheme are provided. Chapter 4 develops power allocation methods to improve the power efficiency and spectrum efficiency of OFDM-IDMA scheme. We first derive performance analysis method for OFDM-IDMA. An SNR evolution technique is investigated to trac the performance of iterative detection process. Based on such technique, we optimize the system performance by power allocation among users. Several power allocation methods are presented, which are selected to use according to the total number of users and channel conditions. Finally, Chapter 5 draws the conclusions and provides some suggestions for future studies. 6

16 2 Preliminaries: Multipath Channel Models, Multicarrier CDMA and IDMA This chapter provides the preliminaries for this thesis. First a description of mobile radio channel is presented. Several channel models for multipath propagation are described. Then an introduction about multicarrier modulation and OFDM-CDMA is given. In the last part of this chapter, we introduce IDMA, a recently proposed multiple access scheme based on random interleaving. 2.1 Multipath Channel Models Radio Channel Characteristics Mobile radio channels are considered to be the most challenging channels. In cellular systems, the transmitted signal between a base station (BS) and a mobile terminal (MT) suffers from some nearly independent effects including path loss, log-normal shadowing, Doppler shift and multipath propagation. The choice of system architecture and optimization of system parameters for communications are dependent on the channel conditions. Channel fade statistics characterize the fading process of the channel. A simple and often used approach is obtained from the assumption that there is a large number of scatterers in the channel that contribute to the signal at the receiver side. The application of the central limit theorem leads to a complex-valued Gaussian process for the channel impulse response. In the absence of line of sight (LOS) or a dominant component, the process has zero mean. The amplitude of fading coefficient h is a random variable with Rayleigh distribution given by P ( h ) = h /2σ2 e h 2, (1) σ2 where 2σ 2 = E{ h 2 } is the average power. The phase of h is uniformly distribute in the 7

17 interval [0, 2π]. In this case, the channel is referred to be Rayleigh fading channel [34]. Variations of the received power due to path loss, shadowing and the average power of multipath propagation are considered to change slowly. If we have the nowledge of these variations at the receiver side, we are able to counteract them effectively by power allocation method, which will be discussed in details in Chapter Channel Modeling A. Channel Model in Time Domain A multipath propagation channel modeled by a tapped delay-line with several nonzeros taps is used in this thesis. The fading factor of each tap is Rayleigh distributed random variable. The power profile and time delay profile can be set. The channel can be expressed by channel impulse response composed of several scattered impulses received from L different paths as L 1 h(τ, t) = h l e (2πf D,lt+ϕ l ) δ(τ τ l ), (2) l=0 where 1 if τ = τ l δ(τ τ l ) = 0 otherwise (3) and h l is the impulse response of path-l with the amplitude h l, the phase ϕ l, the propagation time delay τ l and the Doppler frequency shift f D,l. As the Doppler effect is not taen into account in this thesis, the channel model is simplified to be L 1 h(τ) = h l e ϕ l δ(τ τ l ). (4) l=0 As we also assume the fading coefficients do not change within one simulation frame, this channel model is referred as quasi-static Rayleigh fading multipath channel. 8

18 B. Channel Model in Frequency Domain Simulations of OFDM systems can be carried out in time domain or more efficiently in the frequency domain. For time domain simulation, we have to carry out IFFT operation and cyclic prefix insertion at the transmitter side and FFT operation and cyclic prefix removing at the receiver side, which will be described in next section. It is more computationally efficient to simulate OFDM systems in the frequency domain. During the frequency domain simulation, we first obtain the fading coefficients of subcarriers by Fourier transformation of multipath fading coefficients. Then the transmitted symbols are multiplied by fading coefficients of subcarriers to simulate the the fading affect on subchannels. Preconditions for the frequency domain implementation are the absence of ISI, flat fading on each sub-carrier, and the time-invariance during one OFDM symbol. A proper system design approximately fulfills these preconditions. A further simplification of the channel model for simulating multi-carrier systems in frequency domain is given by using the so-called uncorrelated fading channel model. This channel model is based on the assumption that the fading on adjacent data symbols after inverse OFDM operation and de-interleaving can be considered to be uncorrelated. This assumption holds when a frequency and time interleaver with sufficient interleaving depth is employed. For a propagation scenario without LOS, the fading amplitude H(j) of subcarrier-j is generated by a Rayleigh distribution 1 and the channel model is referred to as an uncorrelated Rayleigh fading channel [15]. The advantages of the uncorrelated Rayleigh fading channel model for multicarrier systems are their simple implementation in the frequency domain and the simple reproducibility of the simulation results. In this case, all the FFT and IFFT operations are avoided in simulation process. 1 It can be proved that H(j) has a distribution of Rayleigh when there are a sufficiently large number of independent and identical distributed (i.i.d.) paths. However when there are only a few independent paths, the distribution of H(j) is not Rayleigh. 9

19 In this thesis, we will use both uncorrelated Rayleigh fading channel model and multipath quasi-static Rayleigh fading channel model for simulations. 2.2 Multicarrier Modulation The principle of multicarrier modulation is to split the transmission bandwidth into several narrowband subchannels, so called subcarriers, which are used for parallel data transmission.the serial high rate data stream is converted into several parallel low rate substreams and each substream is modulated onto a single subcarrier. The bandwidth of each is sufficiently narrow so that the frequency response characteristics of the subchannels are nearly flat. With each subchannel, we associate a carrier x n (t) = e i2πf nt, n = 0, 1, 2,..., N c 1 (5) where f n is the middle frequency in the nth subchannel. subcarrier spacing (a) General multicarrier modulation frequency saved bandwidth frequency subcarrier spacing (b) OFDM modulation Figure 2: Bandwidth consuming illustration: (a) General multicarrier modulation, (b) OFDM modulation OFDM is an efficient realization of multicarrier modulation communication in which subcarriers satisfy the orthogonality condition with the minimum frequency separation 10

20 1/T, where T is the signal duration. 2 Compared with other general muticarrier modulations, OFDM save bandwidth by allowing the subcarrier spectra to overlap, which is illustrated in Fig. 2. It is shown in Fig. 2(b) that at the middle frequency of one subcarrier, the components of other subcarriers are all zero. As so OFDM subcarriers are still mutual orthogonal, even if their spectra overlap. The modulator and demodulator in an OFDM system can be implemented by use of a ban of filters based on discrete Fourier transform (DFT). When the number of subchannels is large, say N c > 25, the modulator and demodulator in an OFDM system are efficiently implemented by use of the fast Fourier transform algorithm (FFT) to compute the DFT. Next, we describe an OFDM system in which the modulator and the demodulator are implemented based on the DFT. Input data Modulation X S/P converter... IFFT... P/S converter x Add cyclic Ext. Processing in frenquency domain Processing in time domain Multipath channel Output data Demodulation ˆX P/S converter... FFT... S/P converter ˆx Remove cyclic Ext. Figure 3: Bloc diagram of a OFDM digital communication system The basic bloc diagram of the OFDM is illustrated in Fig. 3. After symbol mapping, a serial to parallel (S/P) buffer subdivides the symbol sequence into N c substreams. Then each substream is modulated onto a subcarrier by IFFT operation. Let X = [X(0),, X(j),, X(J 1)] T be the symbol sequence, where J is the frame length and the superscript T indicates transpose. After OFDM modulation, the transmitted sequence can be expressed as x = W H X. (X is divided into blocs with length of N c for OFDM transmission.) W is DFT matrix and the superscript H indicates Hermite transpose. 2 The orthogonality among communication signals is defined as that these signals do not interfere with each other and can be transmitted together. 11

21 The (m, n)-th entry of W is W[m, n] = 1 Nc e i2πmn/n c. (6) We assume an L-path channel model with fading coefficients h = [h(0), h(1),, h(l 1)]. The output of multipath channel can be written as ˆx = h x + z, (7) where denotes the convolution and the elements {z(j)} of z are samples of additive noise. A simple way to completely avoid intersymbol interference (ISI) is to append a socalled cyclic prefix to the beginning of each OFDM symbol. Assuming the channel has maximum delay of m, the ISI in any pair of successive OFDM symbol blocs affects the first m signal samples. We add the cyclic prefix with a duration n longer than m at the transmitter side, and discard the the first n samples of each received OFDM symbol bloc. In this way, ISI is completely removed. However cyclic prefixes consume transmission energy and cause a loss in data rate of n n+n c 100%. At the receiver side, we do opposite operations of transmission process as shown in lower part of Fig. 3. After OFDM demodulation, the received signal can be expressed as where H(j) = L 1 l=0 ˆX(j) = H(j) X(j) + Z(j), (8) h(l) e i2πjl/nc is the fading coefficient of the subcarrier-j. Z(j), FFT of z(j), is a complex white Gaussian noise with variance σ 2 in each dimension. The strengths and weanesses of OFDM modulation are summarized as follows [15]: Advantages: High spectral efficiency due to minimum spacing between subcarriers to eep orthogonality among subcarriers. Simple digital realization by IFFT and FFT operations. 12

22 Low complex receiver structure due to the avoidance of ISI with a sufficient guard interval. Different coding and modulation schemes can be employed on different subcarriers which are adapted to the channel conditions. Disadvantages: Time domain signals with high pea-to average power ratio (PAPR) require high quality linear amplifier or a signal clipping and compensation schematism. Loss in spectral efficiency due to the guard interval. OFDM is more sensitive to Doppler spreads than single-carrier modulation and accurate frequency and time synchronization is required. The invention of OFDM can be dated bac in the early 1970s [19]. Since 1990, OFDM has been implemented for variety of standards and applications, such as digital audio broadcasting (DAB), terrestrial digital video broadcasting (DVB-T), asynchronous digital subcarrier lines (ADSL) and IEEE More recently, OFDM-based mobile cellular networs are being developed under IEEE and IEEE Multicarrier CDMA Multicarrier CDMA is a spread spectrum multiple access communication method based on the combination of multicarrier modulation and CDMA. In 1993, several multicarrier CDMA schemes were proposed in roughly the same time. Based on their spreading and modulation types, these schemes can be divided into three categories: multicarrier CDMA (MC-CDMA), multi-carrier direct-sequence CDMA (MC-DS-CDMA) and multitone CDMA (MT-CDMA) [20]. In this thesis, we only deals with the first ind combination, i.e., MC-CDMA, which is also referred as OFDM-CDMA. 13

23 OFDM-CDMA combines advantages of both OFDM and CDMA and provides an effective solution for multiuser communication over multipath channels in cellular environments. With sufficient guard interval, OFDM can completely remove ISI. Compared with orthogonal multiple-access schemes, CDMA provides an effective solution to the cross-cell MAI problem. Meanwhile due to its spread spectrum nature, CDMA is more robust against channel fading. As so, OFDM-CDMA is highly regarded as a promising candidate for the implementation of the physical layer in future wireless communication systems. A significant amount of research wor has been spent on OFDM-CDMA scheme, such as [14] [15] [16] [17]. A brief description about the basic priciples of OFDM-CDMA and detection strategy for uplin OFDM-CDMA proposed in [17], [18] is given as follows. To our nowledge, this is the best practical detection method for OFDM-CDMA OFDM-CDMA Transmitter The bloc diagram of uplin OFDM-CDMA transmitter with K simultaneous active users is shown in Fig. 4. d 1 FEC encoder Symbol mapping x 1 c 1 s 1 IFFT Cyclic Prefix Multipath Fading Channel 1 AWGN d 2 FEC encoder Symbol mapping x 2... c 2 s 2 IFFT Cyclic Prefix Multipath Fading Channel 2 d K FEC encoder Symbol mapping x K c K s K IFFT Cyclic Prefix Multipath Fading Channel K OFDM Modulation Figure 4: Bloc diagram of uplin OFDM-CDMA transmitter with K simultaneous users. 14

24 Let d be the data stream of user-. After channel coding and symbol mapping, we obtain the complex-valued symbol sequence x = [x (1),, x (j),, x (J)] T, where J is the frame length. x is spread by a spreading sequence c. The spreading process results in sequence s with the elements given by where c = [c (1) s (j) = c x (j) = [s (1) (j), s(2) (j),, s(s) (j)]t, (9), c(2),, c(s) ]T is the spreading sequence of user- with length S. The resulting sequence s is OFDM modulated onto N c subcarriers. The spreading operation in OFDM-CDMA is carried out in frequency domain. If S > N c, s (j) should be modulated on to more than one OFDM symbol. If S > N c, several elements of s can be modulated in parallel. By choosing S as a factor of N c (i.e., N c div S), the transmission complexity can be reduced. To simplify the derivation, we presume S = N c. This presumption is made only for purpose of simpler notation OFDM-CDMA Receiver The bloc diagram of an OFDM-CDMA detector multiuser detector based on soft parallel interference cancellation (PIC) is shown in Fig. 5. The Soft PIC detection bloc illustrated in Fig. 5 comprises a minimum mean square error (MMSE) multiuser detector bloc, K maximum ratio combining (MRC) single user detector blocs and the interferences reconstruction/cancellation process. According to (8), after OFDM demodulation the elements of received signal r can be represent as r(j) = r (j) + z(j) = H s (j) + z(j), (10) where z(j) = [z (1) (j), z (2) (j),, z (S) (j)] T are the samples of complex white Gaussian noise with variance σ 2 in each dimension. The S S diagonal fading coefficients matrix H is introduced for convenience and the diagonal element H (s) coefficient on subcarrier-s. of H is the fading 15

25 r Remove Cyclic Prefix FFT OFDM Demodulation _ MMSE MUD _ MRC SUD 1... MRC SUD K c 1 * c K * ˆx 1ˆr 1 ω 1 l 1 1 ˆω Demapping Decoder Tanh(.) H 1 xˆ K... Mapping ω K l K ˆK ω Demapping Decoder Tanh(.) Mapping H K c 1 c K rˆk r ˆ 1... rˆ K PIC MUD Figure 5: Bloc diagram of the detector for OFDM-CDMA system. Note, the MMSE detection bloc comprises the despreading process. Considering transmission in frequency selective channels, the orthogonality among different users is destroyed and MAI occurs. Many data detection techniques are introduced to suppress MAI. In general, data detection techniques can be divided into single user detection (SUD) and multiuser detection (MUD) techniques. SUD is realized by an adaptive one-tap equalizer to compensate the the phase and amplitude distortions imposed on the received signal during the transmission. SUD does not use any information about MAI. The SUD bloc in Fig. 5 employs MRC equalization. MUD techniques improve uplin OFDM-CDMA performance significantly by exploiting the nowledge about MAI. Two MUD technique are considered in Fig. 5, namely MMSE and PIC [25]. The soft PIC detection process for OFDM-CDMA system shown in Fig. 5 can be summarized as follows: Step 1. Initially, MMSE-MUD is applied to suppress MAI and detects desired signals. Note that MMSE-MUD only wors once during the whole detection process. Step 2. The corresponding interference is reconstructed and subtracted from original received signal of each user. 16

26 Step 3. Low complexity MRC-SUDs are applied to detect desired signals. Then the iteration is carried out between Step 2 and Step 3. The details of MRC-SUD, MMSE-MUD and PIC-MUD algorithms are described as follows. MRC: MRC-SUD equalization maximizes the signal-to-noise ratio (SNR) in the single-user case. However, when the system has several simultaneous active users and MAI is not suppressed, MRC performs poorly. Sice after some iterations, the received signal is almost MAI free, MRC can offer good performance with low cost in that case.the equalization coefficient are given by G (s) = H (s), and s, (11) where denotes the complex-conjugate operator. MMSE: MMSE-MUD is described by a K S matrix G MMSE = (A H A + 2σ 2 I S ) 1 A H, (12) where superscript 1 denotes the inverse of a matrix. A ia an S K matrix with elements A s, = H (s) The detected signal can be represented by a K-element vector c (s), and s. (13) ˆx(j) = G MMSE r(j) = [ˆx 1 (j), ˆx 2 (j),, ˆx K (j)]. (14) After MMSE -MUD, the detected signal ˆx is demapped. The symbol demapper outputs the real-valued soft decided bits ω. The soft-in soft-out (SISO) decoders (DECs) or referred to as a posteriori probability (APP) decoders need extrinsic log-lielihood ratios (LLRs) as inputs. We approximate soft decided coded bits ω as input LLRs for DEC-. Note, the MMSE detection bloc comprises the despreading process. (APP decoding is a standard operation [24] and so we will not discuss it in detail.) PIC: After APP decoding, we reconstruct the interferences for the soft PIC process. The LLRs of re-encoded bits l are transform into soft bits ˆω by their means. Then the 17

27 soft bits ˆω are mapped onto soft symbols such that the reliability information included in the soft bits are retained. The reconstructed symbols are spread with their specific spreading sequence c and the resulting chips are once again distorted by channel matrix H to obtain ˆr. For user-, the corresponding MAI is reconstructed by Î = ˆr i. (15) i Î is subtracted from form received singnal r. After cancelling the interference, the desired data symbols can be detected with single-user detector. 2.4 IDMA over Multipath Channels In this section, we introduce the basic principles of IDMA. Our focus is the detection algorithm for IDMA scheme in complex multipath fading channels described in [22]. IDMA is a recently proposed multiple access scheme, in which user-specific interleavers are adopted as the only mechanism for user separation. IDMA can be regarded as a particular case of chip interleaved CDMA [21]. As so, IDMA inherits many advantages of CDMA. Thans to random interleaving and chip by chip (CBC) iterative multiuser detection algorithm, the IDMA scheme is applicable to cancel MAI and ISI effectively and support systems with large numbers of users in multipath fading channels. The upper part of Fig. 6 shows the transmitter structure of the IDMA scheme with K simultaneous users. Let d be the data stream of user-. This data stream is encoded by a forward error correction (FEC) code, generating a chip sequence c. (Here, chip is used instead of bit as the FEC encoding may include spreading or repetition coding.) Then c is permutated by a user-specific interleaver- to produce chip sequence v. After quadrature phase shift eying (QPSK) symbol mapping, the symbol sequence x = [x (1),, x (j),, x (J)] T is produced, where J is the frame length. We use either superscripts Re and Im or function notations Re(.) and Im(.) to indicate real and imaginary parts, respectively. After symbol mapping, we have x (j) = x Re (j) + ix Im (j). (16) 18

28 Transmitter for user-1 d 1 d K FEC 1 Transmitter for user- FEC... c 1 c K K v 1 v K QPSK QPSK x 1 xk Multiple Access Channel ˆd 1 dˆk Decoder (DEC) Decoder (DEC) { ( c 1( j ))} e ESE { ( c1 ( j))} e DEC { e ( c ( j))} ESE { e ( c ( j))} DEC K K π 1 1 π 1 1 π K π K { e ( v ( j))} ESE 1 { ( ( ))} edec v1 j Elementary { e ( v ( j))} ESE K { e ( v ( j))} DEC K Signal Estimator (ESE) r Multiuser Detector Figure 6: Transmitter and receiver structures of an IDMA scheme with K simultaneous users. where x Re (j) and xim (j) are two coded bits from v. The mapping rule for QPSK used in this thesis is given as follows: (0, 0) (1, 1) (0, 1) (1, 1) (1, 1) ( 1, 1) (1, 0) ( 1, 1) The tapped delay-line channel model described in Section is adopted for multipath propagation. We assume an L-path channel model with fading coefficients h = [h (0), h (1),, h (L 1)] for user-, where h (l) = h (Re) (l) + ih (Im) (l). The received signal can be represented by r(j) =,l h (l) x (j l) + n(j), (17) where {n(j)} are the samples of a complex AWGN process with variance σ 2 in each 19

29 dimension. As illustrated in lower part of Fig. 6, an iterative sub-optimal receiver structure is adopted, which consists of an elementary signal estimator (ESE) and K single-user APP decoders (DECs). The multiple access and coding constraints are considered separately in the ESE and DECs. In the iterative detection process, the ESE and DECs exchange extrinsic information in a turbo-type manner [7]. The outputs of the ESE and DECs are LLRs about {x Re (j), xim (j)} as defined below: e ( x Re (j) ) ( ( ) p y x Re (j) = +1) log. (18) p (y x Re (j) = 1) These LLRs are further distinguished by subscripts, i.e., e ESE (x Re (j)) and e DEC(x Re (j)), depending on whether they are generated by the ESE or DECs. For the ESE, y in (18) denotes the received channel output. For the DECs, y in (18) is the deinterleaved version of the outputs of the ESE. A global turbo-type iterative process is applied to exchange the LLRs generated by the ESE and DECs, as detailed below. We focus on detecting x Re (j) after receiving r. (x Im (j) can be handled in a similar same way.) The CBC detection algorithm includes two parts: A. ESE part Now we concentrate on the detection of x Re (j) for user- from path-l and rewrite (17) as r(j + l) = h (l)x (j) + ζ,l (j), (19) where ζ,l (j) consists of the MAI from other users, ISI of the multipath propagation and the noise. Denote the conjugate of h (l) by h (l). We generate where r(j + l) = h (l)r(j + l) = h (l) 2 x (j) + ζ,l (j), (20) ζ,l (j) = h (l)ζ,l (j). (21) By the central limit theorem, ζ,l (j) can be approximated as a Gaussian variable. This approximation is used by ESE to generate LLR for x (j). The phase shift due to h (l) is 20

30 cancelled out in (20), which means that r Im (j + l) is not a function of x Re (j). Therefore we get ( e ESE x Re (j) ) ( ) p ( r Re (j + l) x Re (j) = +1) = log l p ( r Re (j + l) x Re (j) = 1) ) exp ( ( rre (j+l) E( ζ,l (j)) h (l) 2 ) 2 2Var( ζ,l (j)) = log ) exp ( ( r Re (j+l) E( ζ,l (j))+ h (l) 2 ) 2 2Var( ζ,l (j)) ( ζre ) = 2 h (l) 2 rre (j + l) E (j) ( ζre ), Var (j) e ESE ( x Re (j) ) = l=0 (22) L 1 ( e ESE x Re (j) ), (23) l where we assume that the distortion components in the received samples from different paths are uncorrelated, so that the LLR values based on individual chips can be directly summed. ( ζre ) ( ζre ) Then we consider the calculations related to E (j) and Var (j) in (22). We introduce a definition of covariance matrix to mae the derivation more concise. Let α be a complex random variable and define its covariance matrix as Cov(α) = Var(α Re ) E(α Im α Im ) E(α Re )E(α Im ) E(α Im α Im ) E(α Re )E(α Im ) Var(α Im ). (24) According to (21), we have ) E ( ζ,l (j) ) Var ( ζ (j) = h (l)e (ζ,l(j)), = R T (l)cov (ζ (j)) R (l), (25) where R (l) = h Re h Im (l) him (l) (l) hre (l). By (19), we have E (ζ,l (j)) = E (r(j + l)) h (l)e (x (j)), Cov (ζ,l (j)) = Cov (r(j + l)) R (l)cov (x (j)) R T (l). (26) 21

31 In (26), mean and variance of the received signal are estimated as follows E (r(j)) = h (l)e (x (j l)),,l Cov (r(j)) =,l R (l)cov (x (j l)) R T (l) + σ 2 I, (27) where I is a 2 2 identity matrix. B. DEC part The DECs in Fig. 6 carry out APP decoding using the output of the ESE as the input. With QPSK signaling, outputs of the DEC- are the extrinsic LLRs for {x Re (j)} and {x Im (j)}. Then extrinsic LLRs for x (j) are used to generate the following statistics ) ) ( E (x (j)) = tanh Cov (x (j)) = e DEC(x Re 2 (j)) 1 ( E ( x Re (j))) 2 ( + i tanh e DEC(x Im 2 (j)) 0 1 ( E ( x Im (j))) 2 0, (28), where we have assumed that the extrinsic LLRs for the real and imaginary parts of x (j) are uncorrelated, and thus the off-diagonal entries of Cov (x (j)) are zeros. As discussed above, {E (x (j))} and {Cov (x (j))} will be used by the ESE to update the interference mean and variance used in the next iteration by (25), (26) and (27). Initially, we set E (x (j)) = 0 and Cov (x (j)) = I for all, j, implying no information feedbac from DECs. Since in the detection process ESE combines L LLR values, one for each path, in RAKE manner, the detection algorithm is referred to as IDMA with LLR combining (LLRC) in [22]. The total detection complexity of IDMA with LLRC is O(KL), which will be discussed in details in Section Conclusions In this chapter, we provide some preliminaries for this thesis. We first provide a description of mobile radio channel. Several channel models for multipath propagation are 22

32 described. These channel models will be adopted for simulations in Chapter 3 and Chapter 4. An introduction about the principles of OFDM modulation is presented. We have reviewed the basics of OFDM-CDMA systems and described a detection algorithm for OFDM-CDMA based on PIC. The basic architecture of IDMA and its detection algorithm in maltipath fading channel are also presented. By reviewing these techniques, we are inspired to investigate a spread-spectrum multiple access communication scheme based on OFDM and IDMA, which will be elaborated in Chapter 3. 23

33 3 OFDM-IDMA 3.1 Motivation of OFDM-IDMA Scheme OFDM-CDMA is widely regarded as a promising candidate for the implementation of physical layer in the fourth generation (4G) wireless communication systems. A significant amount of research wor has been spent on the OFDM-CDMA scheme, such as [14], [15], [16], [17]. However, there are some disadvantages and difficulties related to OFDM-CDMA systems. The OFDM-CDMA scheme often employs mutual orthogonal codes to distinguish users, and one major problem with the OFDM-CDMA scheme in frequency selective channel is the distortion of orthogonality among users (especially in uplin transmission scenario), which leads to serious MAI problem. MUD is a promising technique for the MAI problem, but the complexity related to MUD has been a major concern for its practical application. The maximum a posteriori (MAP) multiuser detector has exponential complexity with the number of users K [27]. Other linear multiuser detectors for conventional OFDM-CDMA system, e.g., the linear MMSE detector (as shown in Chapter 2) and the decorrelator, usually have quadratic complexity with the number of users K. The quadratic complexity is mainly due to the operations involved in resolving the correlation between spreading sequences. When K is large, the it is computationally prohibitive for practical implementation. In OFDM-CDMA systems, spreading sequences are employed to distinguish signals from different users. From a coding theory point of view, it is not a wise choice to use spreading sequences for user separation, since the spreading operation results in bandwidth expansion without coding gain. The theoretical analysis in [28], [29] shows that the capacity of multiple access channel can only be approached, when entire bandwidth expansion is devoted to FEC coding. The idea of a hybrid communication scheme combining OFDM and IDMA has recently been proposed and studied in [30] [31]. The OFDM-IDMA scheme, which em- 24