Abstract of thesis entitled

A Study of Channel Estimation for OFDM Systems and System Capacity for MIMO-OFDM Systems

Submitted by Zhou Wen For the degree of Doctor of Philosophy at the university of Hong Kong in July 2010

This thesis concerns about two issues for the next generation of wireless communications, namely, the channel estimation for orthogonal frequency-division multiplexing (OFDM) systems and the multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) system capacity. For channel estimation for OFDM systems over quasi-static fading channels having resolvable mulitipath number L, a novel fast linear minimum mean square error (LMMSE) channel estimation method is proposed and investigated. The proposed algorithm deploys Fourier transform (FFT) and the computational complexity is therefore significantly reduced to O(Nplog2(Np)), as compared to that of O(Np3) for the conventional LMMSE method, where the notation O() is the BachmannLandau function and Np is the number of pilots for an OFDM symbol. The normalized mean square errors (NMSE) are derived in closed-form expressions. Numerical results show that the NMSE is marginally the same with that of the conventional LMMSE for signal to noise ratio (SNR) ranges from 0 dB to

25 dB. For channel estimation for OFDM systems over fast fading and dispersive channels, a novel channel estimation and data detection method is proposed to reduce the inter-carrier interference (ICI). A new pilot pattern composed of the comb-type and the grouped pilot pattern is proposed. A closed-form expression for channel estimation mean square error (MSE) has been derived. For SNR = 15 dB, normalized Doppler shift of 0.06, and L = 6, both computer simulation and numerical results have consistently shown that the ICI is reduced by 70.6% and 43.2%, respectively for channel estimation MSE and bit error rate (BER). The pilot number per OFDM symbol is also reduced significantly by 92.3%, as compared to the comb-type pilot pattern. A closed-form mathematic expression has been proposed for the capacity of the closed-loop MIMO-OFDM systems with imperfect feedback channel. The lower threshold of feedback SNR is derived. For L = 6, numerical results show that the lower threshold of feedback SNR is proportional to antenna numbers N and system SNR. The increasing rate of the feedback SNR threshold increases from 0.82 to 1.01 when N increases from 2 to 16. The variance and mean of OFDM system capacity over Rayleigh channels and Ricean channels have been respectively investigated that the closed-form expression for the capacity variance has been proposed. The resultant system capacity variances over the two channels are respectively evaluated by numerical method and also verified by computer simulation. The joint probability density function (PDF) of two arbitrary correlated Ricean random variables has also been derived in an integral form. Numerical results reveal that the variance of OFDM system is proportional to SNR and inversely proportional to L for the two channels respectively. For the same two respective channels, the variance

marginally increases with a linear rate of 0.166 bit2/dB and 0.125 bit2/dB, when L = 2 and SNR ranges from 0 dB to 15 dB. The variance is reduced from 1.75 bit2 to 1.30 bit2 and from 1.48 bit2 to 1.26 bit2, when SNR = 10 dB and L ranges from 2 to 4. (Total words: 495)

A Study of Channel Estimation for OFDM Systems and System Capacity for MIMO-OFDM Systems

by

Zhou Wen
B. Eng., M. Eng., USTC, P. R. China

A thesis submitted in partial fulfillment of the requirements for the Degree of Doctor of Philosophy at the university of Hong Kong in July 2010.

Declaration
I declare that this thesis represents my own work, except where due acknowledgement is made, and that it has not been previously included in a thesis, dissertation or report submitted t to this University or any other institution for a degree, diploma, or other qualifications.

Signature: ___________ Zhou Wen

Acknowledgements
I would like to take this opportunity to express my gratitude to all the people who have ever helped me in the thesis writing and the course of the research. My sincere and hearty thanks and appreciations go firstly to my supervisor, Dr. W.H. Lam, whose suggestions and encouragement have given me much insight into the research work. It has been a great privilege and joy to study under his guidance and supervision. His insightful observation and effective feedback inspired me during the research. Furthermore, it is my honor to benefit from his personality and diligence, which I will treasure my whole life. I also gratefully acknowledge Prof. V.O.K. Li, Prof. G.L. Li, Prof. Y.C. Wu, Prof. S.C. Chan, Prof. T.S. Ng and Prof. Agnes S.L. Lam for their interesting courses and helpful discussions. I would like to thank the office staff and technical staff from the EEE department for their helpful administrative and facility supports. Especially, Ms. Julie Hungs readiness to help students is very impressive. I also appreciate the HKSAR government for the studentship support to the study in the University of Hong Kong. I am extremely grateful to all my friends and classmates who have kindly provided me assistance and companionship in the process of preparing this thesis: Dr. Zhi Zhang, Dr. Zhiqiang Chen, Dr. Mingxiang Xiao, Dr. Fei Mai, Mr. Xueyong Liu, Mr. Xiaoguang Dai, Ms. Ziyun Shao, Mr. Ka-Chung Leung, Mr. Peng Zhang, Dr. Yanhui Geng, Ms. Qiong Sun, Mr. Haoling Xiahou, Mr. Zhibo Ni, Mr. Jun Zhang, Mr. Xiaolei Sun, Mr. Chengwen Xing. They have made the life during the past four years an enjoyable and memorable experience. Finally, I wish to express my hearty gratitude to my parents, for their encouragements and love in all my endeavors.
ii

Contents
Declarations......................................................................................................................... i Acknowledgements............................................................................................................. ii Contents.............................................................................................................................. iii List of Figures.....................................................................................................................vii
Chapter 1: Introduction...................................................................................................... 1

1.1 Research motivation................................................................................................ 4 1.2 Organization and contributions of the thesis........................................................... 5

Chapter 2: OFDM systems and MIMO systems .............................................................. 9

2.1 Wireless Channel................................................................................................... 10 2.1.1 Large scale propagation ............................................................................. 11 2.1.2 Small scale propagation ............................................................................. 13 2.1.3 Typical wireless channel models................................................................ 17 2.2 OFDM systems ..................................................................................................... 20 2.2.1 Basic principles and characteristics for OFDM systems ........................... 21 2.2.2 Peak-to-Average (PAR) of OFDM systems ............................................... 30 2.2.3 Channel estimation for OFDM systems..................................................... 33 2.2.4 Synchronization of OFDM systems ........................................................... 38 2.2.5 Advantages and disadvantages of OFDM systems .................................... 39 2.3 MIMO systems...................................................................................................... 40 2.3.1 Basic MIMO system model ....................................................................... 40 2.3.2 Functions of MIMO systems...................................................................... 42 2.3.3 Overview of Space Time codes.................................................................. 45 2.3.4 Capacity of MIMO systems ....................................................................... 52
iii

2.4 MIMO-OFDM systems......................................................................................... 54 2.5 Summary ............................................................................................................... 56

Chapter 3: Channel estimation for OFDM systems over quasi-static fading channels ............................................................................................................................................. 57

3.1 Introduction........................................................................................................... 58 3.2 System Model ....................................................................................................... 61 3.3 The Proposed Fast LMMSE Algorithm ................................................................ 63 3.3.1 Properties of the channel correlation matrix in frequency domain ............ 63 3.3.2 The proposed fast LMMSE channel estimation algorithm ........................ 65 3.3.3 Computational complexity comparison between the proposed method and the conventional LMMSE method ...................................................................... 69 3.4 Analysis of the Mean Square Error (MSE) of the Proposed Fast LMMSE Algorithm .................................................................................................................... 70 3.4.1 MSE analysis of the conventional LMMSE algorithm.............................. 71 3.4.2 MSE analysis for the proposed fast LMMSE algorithm............................ 72 3.5 Numerical and Simulation Results........................................................................ 75 3.6 Conclusion ............................................................................................................ 81
Chapter 4: Channel estimation and data detection for OFDM systems over fast fading channels.................................................................................................................. 87

4.1 Introduction........................................................................................................... 88 4.2 System Model ....................................................................................................... 91 4.3 The Proposed Channel Estimation and Data Detection ........................................ 92 4.3.1. The proposed pilot pattern ........................................................................ 92

iv

4.3.2. Channel Estimation and data detection for the first M1 OFDM symbols of each block ........................................................................................................... 94 4.3.3. Channel estimation and data detection for the last M2 OFDM symbols of each block ........................................................................................................... 95 4.3.4. Summary of the proposed channel estimation and data detection ............ 98 4.4. Analysis of MSE of the proposed channel estimation method ............................ 99 4.4.1. MSE analysis of channel estimation for the first M1 OFDM symbols ... 100 4.4.2. MSE analysis of channel estimation for the last M2 OFDM symbols .... 103 4.4.3 MSE analysis of channel estimation for one OFDM block ..................... 105 4.5 Numerical and Simulation Results...................................................................... 106 4.6. Conclusion ......................................................................................................... 112
Chapter 5: MIMO-OFDM system capacity with imperfect feedback channel ......... 118

5.1 The open-loop and closed-loop capacity for MIMO Systems ............................ 119 5.1.1 MIMO system model ............................................................................... 119 5.1.2 MIMO system capacity............................................................................ 120 5.1.3 Numerical Results and discussion............................................................ 124 5.2 The closed-loop capacity with imperfect feedback channel for MIMO-OFDM systems ...................................................................................................................... 127 5.2.1 System Model .......................................................................................... 128 5.2.2 Closed-Loop Capacity and Feedback SNR for MIMO-OFDM Systems 130 5.2.3 Numerical Results .................................................................................... 136 5.3 Summary ............................................................................................................. 142
Chapter 6: Capacity of OFDM systems over time and frequency selective fading
v

channels............................................................................................................................ 144

6.1 Introduction......................................................................................................... 145 6.2 OFDM System Model......................................................................................... 147 6.3 OFDM System Capacity ..................................................................................... 148 6.3.1 OFDM system capacity over Rayleigh fading channels .......................... 148 6.3.2 OFDM system capacity over Ricean fading channels ............................. 153 6.4 Numerical and Simulation Results...................................................................... 157 6.5 Conclusion .......................................................................................................... 161
Chapter 7: Conclusions and future works .................................................................... 167

7.1 Conclusions......................................................................................................... 167 7.2 Future works ....................................................................................................... 169 APPENDIX A: The derivation of the rank of channel frequency autocorrelation matrix
RHH in Chapter 3............................................................................................................... 170

APPENDIX B: The derivation of equation (3-20) in Chapter 3....................................... 171 APPENDIX C: The derivation of the joint PDF of two arbitrary correlated Ricean random variables ............................................................................................................................ 173 Appendix D: List of Abbreviations................................................................................... 176 REFERENCES.................................................................................................................. 179 Publications....................................................................................................................... 191

vi

List of Figures
Fig 1.1: Organization of the thesis. ....................................................................................... 6 Fig 2.1: Path Loss, shadowing and multipath versus distance............................................ 11 Fig 2.2: The Doppler power spectrum function expressed by (2-4). .................................. 14 Fig 2.3: The multi-path effect between the transmitter and the receiver in wireless communication........................................................................................................ 14 Fig 2.4: Time varying impulse response of a wireless channel, for the path number N = 3, 4, and 5........................................................................................................................ 15 Fig 2.5: Four kinds of small scale propagations. ................................................................ 16 Fig 2.6: PDFs for Rayleigh fading with the variance 2 = 0.5, 2, and 5, respectively........ 17 Fig 2.7: PDFs for Ricean fading with Ricean factor Kr = 0 dB, 10 dB, and 20 dB, respectively.............................................................................................................. 18 Fig 2.8: PDFs for Nagakami-m fading with m = 0.5, 1, and 10.......................................... 19 Fig 2.9: The continuous OFDM system model................................................................... 22 Fig 2.10: The waveform of Gk ( w) ....................................................................................... 24 Fig 2.11: Equivalent transmitter for OFDM systems.......................................................... 25 Fig 2.12: Equivalent receiver for OFDM systems. ............................................................. 25 Fig 2.13: CP for an OFDM symbol..................................................................................... 26 Fig 2.14: SNRloss versus CP length. .................................................................................. 27 Fig 2.15: The inter-symbol interference of OFDM systems without CP. ........................... 28 Fig 2.16: Extraction of the data in frequency domain......................................................... 28 Fig 2.17: The discrete baseband OFDM system model. ..................................................... 29 Fig 2.18: The output power versus the input power for a power amplifier......................... 31 Fig 2.19: The power spectrum comparison between the input signal and the output signal passing through an amplifier. .................................................................................. 32 Fig 2.20: Two kinds of pilot patterns (black dot: pilot, white dot: user data). .................... 34 Fig 2.21: Pilot-aided channel estimation for OFDM systems............................................. 35 Fig 2.22: The basic MIMO system model. ......................................................................... 40 Fig 2.23: Received signal after diversity operation. ........................................................... 43
vii

Fig 2.24: Diversity-multiplexing tradeoff, d*(r) versus r. .................................................. 44 Fig 2.25: The Alamouti STBC diagram for 22 MIMO systems. ...................................... 46 Fig 2.26: V-BLAST system diagram................................................................................... 49 Fig 2.27: The baseband MIMO-OFDM system model....................................................... 55 Fig 3.1: Baseband OFDM system. ...................................................................................... 61 Fig 3.2: Channel estimation based on comb-type pilots. .................................................... 62 Fig 3.3: The first row of the channel autocorrelation matrix R Hp Hp , A . ............................. 82

Fig 3.4: The first row of the LMMSE matrix R Hp Hp R Hp Hp + I with different SNRs. SNR
................................................................................................................................. 83 Fig 3.5: Normalized Mean square error (NMSE) of channel estimation of LMMSE algorithm versus that of the proposed fast LMMSE algorithm by computer simulation and numerical method. .......................................................................... 83 Fig 3.6: NMSE of LMMSE algorithm with matched SNR and mismatched SNRs versus SNR, by simulation and numerical method, respectively. ...................................... 84 Fig 3.7: NMSE of the proposed fast LMMSE algorithm with matched SNR and mismatched SNRs versus SNR, by simulation and numerical method, respectively. ................................................................................................................................. 84 Fig 3.8: Bit error rate (BER) of the LS, LMMSE, the proposed fast LMMSE and perfect channel estimation versus SNR............................................................................... 85 Fig 3.9: BER comparison between LMMSE channel estimation with matched SNR and LMMSE channel estimation with designed SNRs.................................................. 85 Fig 3.10: BER comparison between the proposed fast LMMSE channel estimation with estimated SNR and the proposed fast LMMSE channel estimation with designed SNRs. ...................................................................................................................... 86 Fig 4.1: Pilot pattern (gray circle: user data, black circle: pilot). ..................................... 114 Fig 4.2: The normalized mean square error (NMSE) of channel estimation for the first M1 OFDM symbols, for f d T = 0.01, 0.06 and 0.1, respectively. ................................. 114 Fig 4.3: The NMSE of channel estimation based on equi-spaced and grouped pilot pattern,
viii

for the polynomial order Q =1, 2, 3 and the normalized Doppler shift f d T = 0.01 and 0.1, respectively..................................................................................................... 115 Fig 4.4: The NMSE of channel estimation based on grouped pilot pattern, for c =1, 2, 3 and the normalized Doppler shift f d T = 0.01 and 0.1, respectively.............................. 115 Fig 4.5: The NMSE of channel estimation based on grouped pilot pattern, for the number of pilot groups N group =18, 36, 72 and the normalized Doppler shift f d T = 0.01 and 0.1, respectively..................................................................................................... 116 Fig 4.6: The NMSE of channel estimation for the proposed algorithm and LS algorithm by numerical method and simulation at f d T = 0.01 and 0.06, respectively. ............... 116 Fig 4.7: Bit error ratio (BER) of LS, the proposed algorithm and the algorithm in [29], for normalized Doppler shift f d T = 0.01 and 0.06, respectively. ................................ 117 Fig 5.1: The eigenmode transmission of MIMO systems. ................................................ 123 Fig 5.2: The MIMO system open-loop capacity versus the number of transmitter antennas NT , for the number of receiver antennas N R = 1 ...................................... 124 Fig 5.3: The MIMO system open-loop capacity versus the number of receiver antennas N R , for the number of transmitter antennas NT = 1 ....................................................... 124 Fig 5.4: The capacities of the N by 1 MISO system, the 1 by N SIMO system, and the N by N MIMO system as a function of N, for SNR = 5 dB........................................... 125 Fig 5.5: The open-loop and closed-loop capacity for MIMO systems, versus SNR. ....... 126 Fig 5.6: The closed-loop MIMO-OFDM system model. .................................................. 128 Fig 5.7: The open-loop and closed-loop system capacity for MIMO-OFDM systems having different transmitter antenna and receiver antenna numbers. ................... 138 Fig 5.8: The capacity gain of the closed-loop capacity with imperfect feedback over that of the open-loop capacity versus feedback channel SNR, for NT = N R = 4 . .............. 140 Fig 5.9: The capacity gain of closed-loop capacity with imperfect feedback over that of the open-loop capacity versus feedback channel SNR, for system SNR = 10 dB...... 140 Fig 5.10: The lower threshold of the feedback SNR versus the MIMO-OFDM system SNR,
ix

for different antenna pairs. .................................................................................... 142 Fig 6.1: The PDF of the capacity at a certain subcarrier, fCi ,k ( x) in (6-36), for SNR = 0 dB, 5 dB, 10 dB, and 20 dB, respectively.................................................................... 162 Fig 6.2: The joint PDF of of | H (i, k1 ) |2 and | H (i, k2 ) |2 , f|H (i ,k )|2 ,|H (i ,k
1 2 2 )|

( x, y ) ,for the coefficient

of equation (6-23), = 0.61................................................................................... 162 Fig 6.3: The coefficient of equation (6-24), , versus different subcarrier gap between
k1 and k2 . ................................................................................................................ 163

Fig 6.4: The variance of OFDM system capacity for the number of channel paths L = 2, 4, and 8, over the Rayleigh fading channel............................................................... 163 Fig 6.5 The variance of OFDM system capacity versus the CP of an OFDM symbol in unit of sample point, over the Rayleigh fading channel............................................... 164 Fig 6.6: The variance of OFDM system capacity versus the number of subcarriers of one OFDM symbol, for Rayleigh fading channels. ..................................................... 164 Fig 6.7: The variance of OFDM system capacity over Ricean fading channels for L = 2, 4, 8, respectively........................................................................................................ 165 Fig 6.8: The mean value of OFDM system capacity for Rayleigh fading channel and Ricean fading channel, by numerical method. ...................................................... 165 Fig 6.9: The variance of OFDM system capacity for Rayleigh fading channel and Ricean fading channel, by computer simulation and numerical method. ......................... 166

Chapter 1: Introduction
The research on wireless communication systems with high data rate, high spectrum efficiency and reliable performance is a hot spot. There are several advanced communication technologies or protocols proposed recently, including Orthogonal frequency division multiplexing (OFDM) [1], multiple input multiple output (MIMO) [2], Ultra-Wideband (UWB) technology [3], cognitive radio [4], World Interoperability for Microwave Access (WiMAX) [92], and 3GPP Long Term Evolution (LTE) [92], [93]. OFDM is an efficient high data rate transmission technique for wireless communication. OFDM presents advantages of high spectrum efficiency, simple and efficient implementation by using the fast Fourier Transform (FFT) and the inverse Fast Fourier Transform (IFFT), mitigration of inter-symbol interference (ISI) by inserting cyclic prefix (CP) and robustness to frequency selective fading channel. MIMO is the use of multiple antennas at both the transmitter and receiver to improve communication performance. It is one of several forms of smart antenna technology. MIMO technology has attracted attention in wireless communications, because it increases in data throughput without additional bandwidth or transmit power. It achieves this by higher spectral efficiency and link reliability or diversity. The combination of MIMO with OFDM technique is a promising technique for the next generation wireless communication. A new protocol draft employing the MIMO-OFDM as the physical layer technology, IEEE 802.11n, as an amendment to IEEE 802.11 standards has been proposed [53]. Wireless LAN technology

has seen rapid advancements and MIMO-OFDM has gradually been adopted in its standards. The following table shows the existing IEEE 802.11 WLAN protocols.

Table 1.1 Existing 802.11 WLAN Standards

IEEE Protocol Name Standard Approved Available Bandwidth Frequency Band of Operation Non-Overlapping Channels (US) Data Rate per Channel 802.11b 802.11a 802.11g June 2003 802.11n

Sept. 1999 Sept. 1999

Released in 2009 83.5/580 83.5 MHz 580 MHz 83.5 MHz MHz 2.4 GHz 3 5 GHz 24 2.4 GHz 3 2.4/5 GHz 3/24 1600 Mbps DSSS, CCK, OFDM, MIMO

111 Mbps 654 Mbps 154 Mbps DSSS, CCK DSSS, CCK, OFDM

Modulation Type

OFDM

UWB is a technology for transmitting data spread over a large bandwidth (usually larger than 500 MHz) that shares among users. UWB was traditionally applied in non-cooperative radar imaging. Most recent applications include sensor data collection, precision locating, and tracking applications. The concept of cognitive radio was first proposed by Dr. J. Mitola and Prof. G. Q. Maguire [4] in 1999 and was an extension to the concept of software radio. Cognitive radio is an intelligent communication system that could detect and track the communication environments. It would adjust the transmitter and the receivers parameters adaptively according to the changes of environment

parameters such as the mobile velocity of the user, so that the system stability could be

ensured, the system performance could remain a good condition, and the spectrum efficiency could be improved. WiMAX is a telecommunications protocol that provides fully mobile Internet access. The name "WiMAX" was created by the WiMAX Forum, which was founded in 2001. The forum refers to WiMAX as a standards-based technology enabling the delivery of last mile wireless broadband access as an alternative to cable and digital subscriber line (DSL). The basis of WiMAX is IEEE 802.16 standard which is sometimes referred to as WiMAX equivalently. The current WiMAX revision is based on IEEE 802.16e, which was approved in December 2005. The physical layer of WiMAX adopts a lot of advanced technologies such as scalable orthogonal frequency division multiplexing access (OFDMA), MIMO, adaptive antenna array and so on. Current WiMAX that is based on the IEEE 802.16e protocol belongs to 3G family. Future WiMAX is based on IEEE 802.16m, which has been submitted to the International Telecommunication Union (ITU) for International Mobile Telecommunication Advanced (IMT-Advanced) standardization. Future WiMAX, or the proposed WiMAX release 2, is considered as a candidate of 4G family. LTE is the latest standard in the mobile communication systems. The current generation of mobile communication system is collectively known as 3G. Although LTE is often referred to as 4G, the first released LTE is actually a 3.9G technology as it does not completely meet the 4G requirements. The main advantages of LTE include high throughput, low latency, plug and play, a simple architecture resulting in low power consumption, supporting seamless passing by base stations with former wireless networks such as Global System for Mobile Communications

(GSM), Universal Mobile Telecommunications System (UMTS), and CDMA2000. LTE also adopts OFDM and MIMO technologies in the physical layer. It uses a 2 by 2 MIMO system as the basic configuration, that is, both the base station and the mobile end equip 2 antennas. The next step for LTE evolution is LTE Advanced and is currently being standardized by 3rd Generation Partnership Project (3GPP) organization. The thesis studies two issues: channel estimation for OFDM systems and MIMO-OFDM system capacity. The chapter is organized as follows. Section 1.1 describes the research motivation. Section 1.2 provides the thesis contributions and the overall organization of the thesis.

1.1 Research motivation

Earlier OFDM systems such as the digital audio broadcasting (DAB) system in Europe does not require channel estimation module. It only uses DPSK demodulation for the sake of reducing the complexity of the receiver. However, with increasing demands of high data transmission rate and reliable communication quality, channel estimation has become a necessary part in the OFDM system. For example, the digital video broadcasting (DVB) system adopts the channel estimation module. In a broadband wireless environment, the channel is often time varying and frequency selective, which distorts the transmitted signal significantly, so that accurate and real time channel estimation is the challenging topic in the OFDM system. Channel state information can be used for the detection of the received signal, improving the capacity of the system throughput by adjusting the modulation at the

transmitter through the feedback. Therefore, one issue of the thesis studies channel estimation for OFDM systems over time varying and frequency dispersive fading channels. With the increasing number of mobile phone users and higher demands for wireless services, future communication systems should have higher system capacity. MIMO technique is a breakthrough of improving system capacity. Telatar [46] and Foschini [47] have firstly formulated the system capacity of the MIMO systems assuming independent and identically distributed fading at different antennas. They have proved that the MIMO system capacity for n transmitter antennas and n receiver antennas increases linearly with n at a fixed transmitter power. That is, MIMO systems can improve the system capacity significantly without increasing the system bandwidth. A number of MIMO techniques known as layered space time architectures or Bell Laboratories layered space time (BLAST) architectures [5][8] have been proposed. Many studies on MIMO system capacity have been conducted. Since the combination of MIMO with OFDM is a trend, a lot of research work has been done on the MIMO-OFDM system capacity. However, the research on MIMO-OFDM system capacity with imperfect feedback channel is not mature and corresponding work is not much. Therefore, the second issue of the thesis is to study the MIMO-OFDM system capacity with imperfect feedback channels.

1.2 Organization and contributions of the thesis

Firstly, we briefly describe the organization of the thesis. Chapter 1 gives a general

introduction, research motivation, organization, and contributions of the thesis. Chapter 2 describes the basic OFDM system model and MIMO system model. The wireless channel model, the principles of OFDM and MIMO systems, the combination of MIMO and OFDM, that is, MIMO-OFDM system is also introduced in Chapter 2. Next, as depicted in Fig 1.1, the thesis begins with the first issue, that is, channel estimation. The proposed channel estimation method in Chapter 3 is based on quasi-static fading channels and that in Chapter 4 is based on fast fading channels. Then, the thesis switches to next issue, that is system capacity. In Chapter 5, the MIMO-OFDM system capacity with imperfect feedback channel is investigated. In Chapter 6, the capacity variances for OFDM systems over Rayleigh and Ricean fading channels are derived, respectively. Finally, Chapter 7 concludes the thesis and discusses future research works.

Fig 1.1: Organization of the thesis.

Secondly, the major contributions of this thesis are summarized as follows.

A fast linear minimum mean square error (LMMSE) channel estimation method for

OFDM systems over slow fading channels has been proposed. Unlike the conventional method, the channel state information is not needed in advance. Almost the same performance with the conventional LMMSE channel estimation in terms of the normalized mean square error (NMSE) of channel estimation and bit error rate (BER) could be achieved for the proposed method. The computational complexity can be reduced significantly since the proposed method replaces the inverse operation with FFT operation. (Chapter 3)

A new pilot pattern and corresponding channel estimation method and data

detection for OFDM systems over fast fading channels have been proposed. The proposed channel estimation and data detection based on the proposed pilot pattern can eliminate inter-carrier interference (ICI) effect effectively. And the number of required pilots is also reduced significantly, compared with the conventional least square (LS) method. MSE analysis for the channel estimation based on the grouped pilot pattern is provided, too. (Chapter 4)

The closed-loop MIMO-OFDM system capacity with imperfect feedback channel

has been formulated. We use the feedback SNR to measure the closed-loop capacity. Since low feedback SNR may not yield positive gain of closed-loop capacity over

open-loop capacity, there exists a lower threshold of feedback SNR. The lower thresholds for different antenna pairs are further investigated by numerical method. (Chapter 5)

The variances of OFDM system capacity over Rayleigh fading channels and Ricean

fading channels have been derived. The effects of SNR, the number of channel paths, power profile of fading channel, the delay of the channel on the variance have been thoroughly investigated, for both multipath Rayleigh channels and multipath Ricean channels. The joint PDF of two arbitrary correlated Ricean random variables has been provided in an integral form. (Chapter 6)

Chapter 2: OFDM systems and MIMO systems

Since the thesis studies the channel estimation for OFDM systems and system capacity for MIMO-OFDM systems, this chapter briefly introduces the background of OFDM systems and MIMO systems. OFDM is an effective technology which provides high spectrum efficiency, high data transmission rate and is robust to multi-path fading [1]. OFDM has already been widely put into practice in DAB system, DVB system and WLAN. MIMO systems which employ multi-element antenna arrays at the transmitter and receiver ends are capable of high data rate transmission. A number of MIMO techniques known as layered space time architectures or BLAST architectures have been proposed [5]-[7], [9]. As OFDM technique can mitigate the ISI and transform the frequency selective fading channel into a set of flat fading channels, the combination of MMO with OFDM technique is a trend for future wireless communication. A new protocol draft, IEEE 802.11n, as an amendment to IEEE 802.11 standards has been proposed and investigated [53]. The draft proposes MIMO-OFDM as the physical layer technique and is to be approved by IEEE. The chapter is organized as follows. Section 2.1 provides an introduction to wireless channel in communication systems. Section 2.2 describes the basic OFDM system model, principles, and related key technologies in OFDM systems. Section 2.3 introduces the basic MIMO system model and briefly overviews the existing MIMO systems. Section 2.4 presents the MIMO-OFDM system model. The last section 2.5 summarizes the chapter.

2.1 Wireless Channel

Since there exist reflections, scattering, and diffraction in the transmission of electromagnetic wave, the spatial environments such as the landscape of a city, obstructions and so on will make complicated impacts on the transmission of electromagnetic wave. There are two kinds of propagation, including the large-scale propagation and the small scale propagation. The signal variations due to path loss or shadowing occur over relatively large distances, this variation is referred to as the large scale propagation effects. Path loss is a major component in the analysis and design of the link budget of a telecommunication system. The small scale propagation refers to the phenomena that the amplitude of the received wireless signal varies very fast in a short time period or a short distance. The sources of small scale propagation include the Doppler shift effect and multi-path effect. We begin with the introduction of the large scale propagation.

10

2.1.1 Large scale propagation

Fig 2.1: Path Loss, shadowing and multipath versus distance. Fig 2.1 plots the ratio of the received-to-transmit power in dB versus the distance for the combined effects of path loss, shadowing, and multipath. Observe that the free space path loss is linearly proportional to the log-distance between the transmitter and receiver. The shadowing loss has slower variations compared to that of the multipath effect. The large scale path loss model has many successful types including the free space path loss model, the Hata model, and the Okumura model and so on. Most of them are obtained by a combination of the analytical and empirical methods. We next briefly introduce the free space path loss model as an example of path loss effect.

Free space path loss model: An example of path loss effect

Consider a signal transmitted through free space to a receiver located at distance d from the transmitter. It is assumed that there does not exist any obstructions between the

11

transmitter and receiver. The channel model with this kind of transmission method is called a LOS channel, and the corresponding received signal is named the LOS signal. The average received power PR expressed in dB form is given by PR = PT + 10 log10 (Gl ) + 20 log10 () 20 log10 (4) 20 log10 (d) dB (2-1) where Gl is the product of the transmit and receive antenna gains in the LOS direction, PT is the transmitter signal power, d is the distance between the transmitter and the receiver, is the wave length in unit of meter.

Shadow fading

The signal fading due to shadowing from obstacles affecting the wave propagation is referred to as the shadow fading. A signal will typically experience random variation due to blockage from objects in the transmission path. Reflecting surfaces and
scattering objects may produce random variation of the received signal, too. Thus, a

model for the random attenuation due to these effects is also needed. Statistic models have been used to characterize this attenuation since the parameters such as the locations, sizes of the blocking objects or scattering objects are generally unknown. The most common model for this additional attenuation is the log-normal shadowing fading. This model has been demonstrated to accurately model the shadow fading in both outdoor and indoor radio propagation environments. variable X with a log-normal distribution expressed by
p ( x) =

The path loss is a random

(10 log10 x u x ) 2 exp , x > 0 2 2 x 2 x x

(2-2)

12

where = 10/ ln10, x is the

mean value of xdB, xdB = 10 log10 x in unit of dB, and

xis the standard deviation of xdB , also in unit of dB.

2.1.2 Small scale propagation

The small scale fading of a signal is a more rapid fluctuation, which is usually caused by constructive and destructive interference between two or more versions of the same signal (multi-path effect) or Doppler effect due to moving terminals or surroundings objects. We briefly introduce the multi-path effect and the Doppler shift effect.
2.1.2.1 Doppler shift effect

The Doppler shift effect, named after Austrian physicist Christian Doppler who proposed it in 1842, is the change in frequency of a wave for an observer moving relative to the source of the wave. When the mobile handset moves away from the base station, the carrier frequency is decreased and rather than vice versa. The Doppler shift between the base station and the mobile handset, fd, is expressed as fd = v

cos( )

(2-3)

Where v is the velocity of the mobile handset, is the carrier wave length, is the angle between the velocity of the mobile handset in the radial direction of the mobile handset and the base station. Assuming that the arrival angles of the signal is uniformly distributed in the range (-, ), the power spectrum density function (PSD) is given by [36] Pav f f c 1 , fc fd f fc + fd fd fd
2

1 2

S( f ) =

(2-4)

13

where Pav is the average power of the received signal and fc is the carrier frequency.

S( f )

fc fd

fc

fc + fd

Fig 2.2: The Doppler power spectrum function expressed by (2-4).

Fig 2.2 shows the classic U-shape power spectrum. Note that no components fall outside the interval [fc - fd, fc + fd] and the power of the transmitted signal spreads between (fc - fd) Hz and (fc + fd) Hz.
2.1.2.2 Multi-path effect

Fig 2.3: The multi-path effect between the transmitter and the receiver in wireless communication. Multi-path effect refers to the propagation phenomenon that results in radio signals
14

reaching the receiving antenna by two or more paths. Causes of multi-path include atmospheric ducting, ionospheric reflection and refraction, and reflection. Fig 2.3 shows the multi-path effect in wireless communication.

(a )
t = t0

t1

t1 + 2

(b)
t = t0
t1
t1 + 3

(c )

t = t0

t1

t1 + 4

Fig 2.4: Time varying impulse response of a wireless channel, for the path number N = 3, 4, and 5.

Assume that the transmitted signal is an impulse function with a very short pulse width expressed by S (t ) = a0 (t ) and the signal then pass through a wireless channel. At the receiver, the received signal is given by S0 (t ) = ai S [t i (t ) ]
i =1 N

(2-5)

where ai is the magnitude of the i-th path and i is the excess delay of the i-th path, N is the number of resolvable paths. As in Fig 2.4, the received signal is a series of pulses and the number of pulses is the number of channel paths. Note that usually the amplitude ai is time

15

varying and it is a function of time t.

Fig 2.5: Four kinds of small scale propagations.

Thus, based on Doppler shift effect and multi-path effect, T. S. Rappaport divided wireless fading channels into four categories, as shown in Fig 2.5. The first one is the fast fading channel, referred to the kind of channel when the coherence time of the channel, which is a function of Doppler shift, is small relative to the symbol duration of the transmitted signal. The second is the slow fading channel, referred to the kind of channel when the coherence time of the channel is large relative to the symbol duration of the transmitted signal. The third one is the flat fading channel, referred to the kind of channel when the coherence bandwidth of the channel, which is a function of the maximum delay of the channel, is larger than the bandwidth of the signal. The fourth one is the frequency selective fading channel, referred to the channel when the coherence bandwidth of the

16

channel is less than the bandwidth of the signal.

2.1.3 Typical wireless channel models

There are mainly three kinds of wireless channel models, as described as follows.
(1) Rayleigh fading channel
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
2 = 0.5 2 =2 2 =5

10

Fig 2.6: PDFs for Rayleigh fading with the variance 2 = 0.5, 2, and 5, respectively.

Rayleigh fading is a reasonable model when there are many objects in the environment that scatter the radio signal before it arrives at the receiver. The envelope of the channel impulse response is Rayleigh distributed, as expressed by
r2 f (r ) = 2 exp 2 , 2
r r0

(2-6)

where r is the amplitude of the channel impulse response. Fig 2.6 depicts several PDFs for Rayleigh fading channel with the variance 2 = 0.5, 2, and 5. The PDF curve with larger

17

variance has a broader spreading range. Usually, the normalized autocorrelation function of a Rayleigh fading channel is a 0-th order Bessel function of the first kind, as expressed by R( ) = J 0 (2 f d ) (2-7)

Where fd is the Doppler shift and J 0 () is the 0-th order Bessel function of the first kind. A well known simulation model for the Rayleigh fading channel is the Jakes Model, firstly proposed by William C. Jakes in 1975.
(2) Ricean fading channel
6 Ricean factor Kr = 0 dB 5 Ricean factor Kr = 10 dB Ricean factor Kr = 20 dB 4

0.5

1.5

2.5

3.5

4.5

Fig 2.7: PDFs for Ricean fading with Ricean factor Kr = 0 dB, 10 dB, and 20 dB, respectively.

Ricean fading occurs when one of the channel paths, typically a line of sight signal (LOS), is much stronger than the others. In a Ricean fading channel, the amplitude gain is

18

characterized by a Ricean distribution, which is formulated by

f (r ) =

r 2 + kd 2 rkd exp I0 2 2 2 2
r

(2-8)

Where r is the amplitude of the channel impulse response, 2 2 is the power of the non-LOS multi-path components, kd2 is the power of the LOS component, I 0 () is the modified Bessel function the 0-th order. Fig 2.7 shows the PDFs for Ricean fading with Ricean factor Kr = 0 dB, 10 dB, and 20 dB, respectively. The Ricean factor Kr is defined by
2 Kr = kd / 2 2 .

(2-9)

(3) Nakagami fading channel

3 m=1 m = 0.5 m = 10

2.5

1.5

0.5

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

Fig 2.8: PDFs for Nagakami-m fading with m = 0.5, 1, and 10.

Both Rayleigh and Ricean distributions can be obtained by using mathematics to

19

describe the physical properties of the channel models. However, a few channels can not be characterized well by the two previous models. So a more general fading distribution was created whose parameters could be adjusted to characterize more channel models. The distribution is the Nakagami fading distribution, or Nakagami-m fading distribution, first proposed by M. Nakagami, in 1960 [10]. The Nakagami fading distribution is given by
2 m m 2 m 1 mr 2 / P r e p ( r ) = ( m) P 0
r0 r<0

(2-10)

Where r is the amplitude of the channel impulse response, P is the average power of the received signal, () is the Gamma function, defined as
(n) = x n 1e x dx
0

(2-11)

Fig 2.8 depicts the PDFs for Nagakami-m fading with m =0.5, 1, and 10. The total signal power P is assumed to be 1. It is worth pointing out that the PDF is unilaterally Guassian distributed when m = 0.5 and the PDF is Rayleigh distributed when m = 1.

2.2 OFDM systems

OFDM technology can be traced back to the middle 60s, at that time, R. W. Chang published a paper [95] about how to synthesize the transmission of a band limited signal on multi-channel. Chang introduced a theory that multiple data streams can be transmitted through a linear band limited multi-channel without the ICI and ISI. Weinstein and Ebert made major contribution to OFDM in 1971. In that year they published a journal paper

20

about how to modulate/demodulate band signal by Discrete Fourier Transformation (DCT) [1]. To suppress ISI, they proposed the empty guard interval between two adjacent symbols, but the orthogonality between two subcarriers over a frequency selective channel can not be ensured. Another major contribution was made by A. Peled and A. Ruiz [11], who introduced the concept of CP in 1980, which ensured the orthogonality among subcarriers of an OFDM symbol. The CP is copied from the end of the OFDM symbol and it is transmitted followed by each OFDM symbol. When the length of CP is larger than that of the impulse response of the fading channel, ICI could be avoided.
2.2.1 Basic principles and characteristics for OFDM systems

X i ,0 e j0t e j1t X i , N 1 e jN 1t
(a) The transmitter for continuous OFDM systems.

X i ,1

. . .

xi(t)

21

 TTs ()
e j 0t

X i ,0

x i( t )
e j1t

T ()
Ts

X i ,1

. . .

T ()
Ts

X i , N 1

e j N 1t
(b) The receiver for continuous OFDM systems.

(c) The frequency division for OFDM systems. Fig 2.9: The continuous OFDM system model.

As shown in Fig 2.9(a), the transmitted signal xi(t) for OFDM systems is given by

22

xi (t ) = X i ,k g k (t iT )
k =0

N 1

(2-12)

where
e j 2 f k t g k (t ) = 0
t [0, T )

others

(2-13)

and fk = f0 + k , k = 0,1,..., N 1. T

(2-14)

where X i ,k denotes the transmitted signal of the k-th subcarrier at the i-th OFDM symbol interval, T is one OFDM symbol duration excluding the CP length, N is the number of OFDM subcarriers, fk is the frequency of the k-th subcarrier, fo is the lowest frequency which corresponds to the first subcarrier. For the baseband OFDM system, the lowest frequency fo = 0.

The orthogonality between two subcarriers of an OFDM symbol

The orthogonality between two arbitrary subcarriers is described in the following.

Two arbitrary subcarriers are orthogonal in the time domain since the following

equation holds.

(t ) gl* (t )dt = T (k l )

(2-15)

Two arbitrary subcarriers are also orthogonal in the frequency domain.

The Fourier transform of gk(t) is expressed by sin( wT ) jwT / 2 e Gk ( w) = 2 ( w 2f k ) W (2-16)

23

where denotes convolution. Fig 2.10 shows the waveform of Gk(w).

Gk ( w)

2 f k

Fig 2.10: The waveform of Gk ( w) .

The transmitted signals for the k1-th subcarrier and k2-th subcarrier in frequency domain are therefore X i ,k1 Gk1 ( w) and X i ,k2 Gk2 ( w) , respectively, and satisfy

X
=0

i ,k1

Gk (w) X i ,k2 Gl (w)* dw sin(w 2 fk1 )T sin(w 2 fk 2 )T jT ( fk1 fk 2 ) e dw w 2 fk1 w 2 fk 2

(2-17)

= 4 X i ,k1 X i ,k2

Thus, two arbitrary subcarriers are also orthogonal in the frequency domain.

Using FFT/IFFT to implement the transmitter modulation and receiver demodulation

When sampling an OFDM symbol at a sample rate N/Ts and assuming f0=0then the transmitted signal can be expressed as
xi ,m =

N 1 k =0

X i , k g k ( t iT )

t =(i+

m ) Ts N

, m0,1,..., N-1.

(2-18)

24

Equation (2-18) can be further derived as

xi , m = e
j 2 f 0T m N m j 2 k N 1 N X e i ,k = N IDFTN { X i .k } , m0,1,..., N-1. k =0

(2-19)

After sampling, the transmitter shown in Fig 2.9(a) is equivalent to the following diagram.
X i ,0 X i ,1

xi ,m

X i , N 1

Fig 2.11: Equivalent transmitter for OFDM systems.

And the receiver shown in Fig 2.9(b) is also equivalent to the following diagram.

Fig 2.12: Equivalent receiver for OFDM systems.

Since FFT and IFFT are easy to implement, the computational complexity can be reduced significantly compared to other systems.

Cyclic Prefix
A. Peled and A. Ruiz [11] firstly introduced the concept of cyclic prefix in 1980, which not only eliminates ISI, but also ensures the orthogonality between two subcarriers of an OFDM symbol. Cyclic prefix is a duplicate of last part of an OFDM symbol, shown as

25

following.

C y c lic p re fix (C P )

A n O F D M sym bol

tim e

Fig 2.13: CP for an OFDM symbol.

The insertion of CP has two effects. On one hand it eliminates the interference between OFDM symbols, however, on the other hand it also leads to the loss of SNR since it is necessary to provide the transmitter more power.

(1) Loss of SNR:

At the receiver the SNR loss is given by SNRloss = 10 log10 (1 ) (2-20)

where = TCP / T , TCP is the CP length and T is the OFDM symbol length. The longer the CP is, the greater SNRloss is, as shown in Fig 2.14. Observe that the SNRloss increases with the increase of CP length and when < 0.2 , SNRloss < 1 dB .

26

10

Loss of SNR(dB)

0.1

0.01

0.1

Relative length of CP

0.2

0.3

0.4

Fig 2.14: SNRloss versus CP length.

(2) Eliminate ISI:

Recall that a signal x(t) passes through the channel with impulse response h(t), the received signal is given by r (t ) = h(t ) * x(t ) (2-21)

If the channel has multiple paths, there will be ISI effect. The formula above in discrete form is written as rm = hi xm i
i =0 L 1

(2-22)

where L is the number of paths. Fig 2.15 shows the ISI effect for OFDM systems without CP insertion. We set L = 2.

27

(n-1)-th frame

n-th frame

(n+1)-th frame

h0 ... Sn1,N 2 Sn1,N 1 Sn,0 h1

Sn ,1

...

... Sn1,N 2 Sn1,N 1 Sn,0

+ +

Sn,N 1

Sn+1,0 ... Sn,N 1 Sn+1,0 ...

Sn,1

...

h2

... Sn1,N 2 Sn1,N 1 Sn,0 ... rn 1, N 2 rn 1, N 1 rn,0

rn,1

Sn,1 rn, N 1

...
rn +1,0

Sn,N 1

Sn+1,0 ...

...

...

Fig 2.15: The inter-symbol interference of OFDM systems without CP. Without CP insertion, ISI will occur for OFDM systems. However, the insertion of CP could avoid ISI among OFDM symbols since the channel delay will be absorbed if the CP length is equal to or larger than the maximum delay of the channel. Considering the CP insertion, at the receiver the received signal should first remove CP and then perform FFT, shown in Fig 2.16.

Yi ,k

Fig 2.16: Extraction of the data in frequency domain.

28

Summary of OFDM system operation

This section summarized the process of OFDM systems. Fig 2.17 depicts the discrete baseband OFDM system model. The baseband transmitted OFDM signals can be written as xi (t ) = X i ,n exp ( jwi t ) ,
i =0 N 1

0t T

(2-23)

where wi = 2i/T is the central frequency of the n-th subchannel and xi(t) is the i-th corresponding transmitted symbol in time domain, T is the time duration of one OFDM symbol excluding CP. Assuming the period of the sampling clock is Tsamp, then the discrete time sequence at the transmitter is xi ,n = xi (nTsamp )

= X i ,n exp ( jwi nTsamp ) = IDFTN ( X i ,n )

i =0

N 1

(2-24)

X i ,0
X i ,1 X i , N 1
. . . ID F T

x i ,0 x i ,1
. . . I n s e rt C P . . .
P a r r a lle l to S e r ia l tr a n s f o r m

f ilte r in g

x i , N 1

tra n s m itte r channel R e c e iv e r Yi ,0

Y i ,1 Yi , N 1
. . . DFT . . . CP
re m o v in g

H (t, f)
AW GN n o is e

. . .

S/P

f ilte r in g

Fig 2.17: The discrete baseband OFDM system model.

29

The time sequence xi,n is then put into the CP insertion module. After parallel to serial transformation and filtering, it is sent to the receiver via the wireless fading channel. The received signal after DFT operation, Yi,k, is given by Yi ,k = X i , k H i , k + Wi , k (2-25)

2 where wi,k denotes the AGWN with zero mean and variance w , Hi,k is the frequency

response of the radio channel at the k-th subcarrier of the i-th OFDM symbol. If the channel frequency response Hi,k is known at the receiver, the estimated transmitted signal Xi,k, X i ,k , is given by
X i ,k = Yi , k / H i ,k , k = 0,1,..., N 1.

(2-26)

2.2.2 Peak-to-Average (PAR) of OFDM systems

The Peak-to-Average (PAR) is a major problem of OFDM systems [81]-[83]. As the transmitted signal is a summation of N independent subcarriers, the envelope of the transmitted signal is not a constant. The PAR of OFDM systems is defined as
max n (| xn |2 ) . PAR=10 lg E (| xn |2 )

(2-27)

where xn is the transmitted OFDM signal in time domain. Since the power amplifier has a limited linear range, the transmitted signal of OFDM systems will distort when passing through a power amplifier. The power spectrum of the transmitted signal will extend to a larger range so that the performance of OFDM systems will be reduced correspondingly.

30

Output power

P_out

Input power P_in

Fig 2.18: The output power versus the input power for a power amplifier.

Fig 2.18 shows the output versus the input power for a general power amplifier. Observe that when the input signal power is less than a value P_in, the output signal is linearly proportional to the input signal. However, when the input power exceeds P_in, the output signal will be compressed. To illustrate the spectrum leak of the input signal, we give an example. Fig 2.19(a) is a typical power spectrum of band-limited transmitted signal. Fig 2.19(b) is the power spectrum of the output signal passing through an amplifier. The energy leak effect is very obvious by comparing the two subplots.

31

Spectrum at Input to Amplifier 10 0 -10 Log Magnitude (dB) -20 -30 -40 -50 -60 -4 -3 -2 -1 0 1 Normalized Frequency (f/fsym) 2 3 4

(a) The power spectrum of a band-limited input signal.

Spectrum at Output of Amplifier 10 0 -10 Log Magnitude (dB) -20 SPECTRAL REGROWTH -30 -40 -50 -60 -4 -3 -2 -1 0 1 ) Normalized Frequency (f/fsym 2 3 4

(b) The power spectrum of the output signal passing through an amplifier. Fig 2.19: The power spectrum comparison between the input signal and the output signal passing through an amplifier. There are several methods proposed to deal with the PAR problem of OFDM systems, including clipping [82], Error-Control precoding, peak cancellation, PAR reduction codes

32

[83], symbol scrambling codes, and so on.

2.2.3 Channel estimation for OFDM systems

For wide-band wireless communication systems, the channel is time varying and dispersive fading, which will distort the transmitted signal. Thus, the accurate and real-time estimation of channel is a challenging task in OFDM systems. The present channel estimation methods generally can be divided into two kinds. One kind is based on the pilots and the other is blind channel estimation which does not use pilots. Blind channel estimation methods do not use pilots and have higher spectral efficiency. However, they often suffer from high computation complexity and low convergence speed since they often need a large amount of receiving data to obtain some statistical information such as cyclostationarity induced by the cyclic prefix. Blind channel estimation methods are not suitable for applications with fast varying fading channels. There are two classical pilot patterns for the pilot aided channel estimation methods, which are the block-type pattern and the comb-type pattern, as the following Fig 2.20 shows.

33

(a) The block-type pilot pattern.

(b) The comb-type pilot pattern. Fig 2.20: Two kinds of pilot patterns (black dot: pilot, white dot: user data).

The block-type pilot pattern is depicted in Fig 2.20(a). The pilots are inserted into all the subcarriers of one OFDM symbol with a certain period. The block-type can be adopted in slow fading channel, that is, the channel is static within a certain period of OFDM symbols.

34

The comb-type pilot pattern is plotted in Fig 2.20(b). It refers to that the pilots are inserted at some specific subcarriers in each OFDM symbol. The comb-type is preferable in fast varying fading channels, that is, the channel varies over two adjacent OFDM symbols but remains static within one OFDM symbol. The comb-type pilot arrangement based channel estimation has been shown to be more applicable since it can track fast varying fading channels in comparison with the block-type one. When the fading channel can not be viewed as static within an OFDM symbol, ICI occurs. The comb-type pilot pattern can not eleminate ICI and the grouped pilot pattern was proposed by Song and Lim [29] to suppress ICI. The grouped pilot pattern will be discussed in Chapter 4 in detail.

Y (k )

Yp ( k )

H p (k )

H (k )

X p (k )
Fig 2.21: Pilot-aided channel estimation for OFDM systems.

Fig 2.21 depicts the basic pilot-aided channel estimation processing for OFDM systems. Firstly, the pilot signal Yp (k) is extracted from the received signal after FFT operation Y(k), where k is the index of subcarrier of one OFDM symbol. Secondly, the channel estimator performs channel frequency response estimation at pilot subcarriers. There are some channel estimation methods for this part such as LS, MMSE estimator and so on. Thirdly, once the channel frequency response estimation at pilot subcarriers, H p (k ) , is obtained, the

35

estimator performs interpolation to obtain channel frequency response estimation at all subcarriers. There are linear interpolation method, second-order polynomial interpolation method [15], DFT based interpolation method and so on. Next, we briefly describe the LS estimator and the MMSE estimator.

Least square (LS) estimator

It is assumed that the channel frequency response at pilot subcarriers is

H p = [ H p (0) H p (1)

H p ( N p 1)]T

(2-28)

where Np is the number of pilots and the notation ()T denotes transpose. The extracted pilot signal vector is given by
Yp = [Yp (0)Yp (1)

Yp ( N p 1)]T

(2-29)

Then the vector Yp can be further expressed as

Yp = X p H p + Wp
where

(2-30)

X p (0) Xp = 0

X p ( N p 1)
0

Wp is the Guassian noise vector at pilots positions and it is given by Wp = [Wp(0) Wp(1)Wp(Np-1)]T
= [H (0) H (1) Let H p p p (2-31)

( N 1)]T denote the estimated channel response. According H p p

the LS criterion, e p = (Yp H p X p ) H (Yp H p X p ) should achieve the minimum. By differential operation to ep, we have that

36

H H p ,ls = Qls X p Yp

(2-32)

is enable ep to achieve the minimum value, where the matrix QLS is given by
1 QLS = ( X H p Xp)

(2-33)

Therefore the frequency response estimation at pilot subcarriers is given by

H p ,ls Yp (0) Yp (1) = X p 1Yp = X p (0) X p (1) Yp ( N p 1) X p ( N p 1)
T

(2-34)

Minimum Mean Square Error (MMSE) estimator

Minimum mean square error (MMSE) estimator has a better estimation accuracy compared to LS estimator in terms of MSE. At the same MSE, the MMSE estimator has about 5 to 12 dB SNR gain over that of LS estimator. However, the main shortcoming of the MMSE estimator is the higher computation complexity. With the increase of operation points, i.e., the number of subcarriers, its computational complexity will increase exponentially. The literature [84] also proposed the LMMSE estimator through linear approximation to the MMSE estimator, using the transmitted signal correlation matrix in frequency domain. The MMSE estimation in frequency domain for the pilot subcarriers, H p , mmse , is given by
1 H 1 2 H p , mmse = RH p H p ,ls RH p ,ls H p ,ls H p ,ls = RH p H p RH p H p + w ( X p X p )

H p ,ls

(2-35)

is the channel frequency response at pilots positions using LS algorithm and it where H p ,ls
2 is expressed in (2-34). w is the variance of the AWGN noise, the cross correlation

matrices in (2-35) are given by RH p H p = E {H p H H p },

37

H , RH p H p ,ls = E H p H p ,ls RH p ,ls H p ,ls

p ,ls

{ = E {H

H . H p ,ls

(2-36)

where E() denotes expectation and the notation ()H denotes Hermitian transpose, Hp is given by formula (2-34).

2.2.4 Synchronization of OFDM systems

The synchronization of the received signal is an important topic in telecommunications. OFDM system is very sensitive to frequency offset and phase noise, so synchronization is the key problem for OFDM systems. Lack of necessary synchronizer, the orthogonality among sub channels of an OFDM symbol will be destroyed, which will introduce ICI. The synchronization of OFDM system includes timing and frequency synchronization. Timing synchronization includes symbol synchronization and sample clock

synchronization. The aim of symbol synchronization is to seek for an appropriate FFT window. If a FFT window can not be found correctly, ISI will be introduced so that OFDM system performance will be reduced. The aim of sampling clock synchronization is to synchronize the local sample clock at the receiver and the sender clock at the transmitter. If there is an error between the transmitter clock and the receiver clock, the ICI will occur, and it will also affect the symbol synchronization so that worsens the system performance. For single carrier communication systems, the carrier frequency offset only produce a phase rotation of the transmitted signal. And the phase rotation can be recovered by channel equalization. However, for OFDM systems, the frequency asynchrony between the transmitter carrier and the receiver carrier will destroy the orthogonality between the sub
38

carriers within one OFDM symbol. ICI could also be introduced due to the carrier frequency offset, which could seriously decrease the system performance. Thus, frequency synchronization is an indispensable part in OFDM systems. There are a lot of synchronization algorithms, which can be divided into three kinds, that is, CP based method [87], Preamble based method [85], [86], and pilots based method [88].

2.2.5 Advantages and disadvantages of OFDM systems

Generally speaking, the followings are the advantages of OFDM systems.
High spectrum efficiency. Through CP insertion, ISI can be eliminated. OFDM system adopts FFT/IFFT to implement modulation and demodulation so that

the computational complexity is low.

The receiver only needs to perform simple equalization. High system capacity and high data rate.

And the disadvantages of OFDM system include:

The amplitude of OFDM varies greatly and has large PAR, which will result in the

distortion of signal. So it is necessary that the RF power amplifier has a wide linear range to ensure that distortion of the transmitted signal is reduced as far as possible.
OFDM system is sensitive to frequency excursion, so it has strict requirement about

frequency synchronization.

39

2.3 MIMO systems

MIMO is an important breakthrough in wireless communication. MIMO technique was first proposed by Marconi in 1908 and used to cancel the channel fading in order to enhance the system performance. However, the signal processing costs were very high at that time so that MIMO was not given proper due. Until the early 1990s, people began to study MIMO technique again as the signal processing has been further developed.

2.3.1 Basic MIMO system model

Fig 2.22: The basic MIMO system model.

Fig 2.22 shows the basic MIMO system model with NT transmitter antennas and NR receiver antennas, where H denotes the fading channel. MIMO systems can be divided into two categories, shown as follows.

2.3.1.1 MIMO system with flat fading channels

When the fading channel is flat, that is, there is only one resolvable path, the impulse response between the n-th transmitter antenna and the m-th receiver antenna, hm,n, is

40

expressed by hm ,n (t , ) = hm, n (t ) (t ) Thus, the MIMO system channel matrix is given by (2-37)

H = (hm, n ) N R NT = (am.n + jbm, n ) N R NT

(2-38)

where am,n and bm,n are Gaussian random variables with zero mean and unity variance. The received signal Y, is given by
Y = HX + w

(2-39)
T

where the received signal Y = y1

x1 x2
T

y2 w2

yN R , the transmitted signal X = wN R .

T

x NT w1 , the AWGN noise w =

2.3.1.2 MIMO system with frequency selective fading channels

When the number of channel resolvable paths is larger than 1, the channel is frequency selective. The impulse response between the n-th transmitter antenna and the m-th receiver antenna, hm,n, is expressed by

,n ,l (t ) ( l ) hm ,n (t , ) = hm
l =0

L 1

(2-40)

where L is the number of resolvable paths, l is the delay of the l-th path. And the received signal Y at n-th time slot, is given by

Y = H l X n l + w n
l =0

L 1

(2-41)
T

where the received signal Y = y1

y2

yN R , the transmitted signal X =

41

 x1

x2

x NT w1 , the AWGN noise w =

w2

wN R , the channel matrix of

,n ,l ) N R NT . the l-th path, H l = (hm

2.3.2 Functions of MIMO systems

MIMO systems can provide two types of gains, that is, spatial multiplexing gain and diversity gain. For spatial multiplexing, a high rate signal is split into multiple lower rate streams and each stream is transmitted from a different transmit antenna. The receiver could separate these streams, provided that these data streams arrive at the receiver antenna array with low correlation coefficient between each antenna pair. Spatial multiplexing technique is very powerful for increasing channel capacity. On the other hand, MIMO systems can be designed to realize diversity in order to deal with the signal fading. The basic idea of diversity is to provide the receiver several independent fading versions of the same signal, so that the probability of simultaneously experiencing deep fading for all versions can be reduced. Fig 2.23 depicts a received signal after diversity operation. Observe that the combined signal is enhanced in SNR, compared with the 1-th branch signal and the 2-th branch signal.

42

Signal after diversity

Signal of 1-th branch

Signal of 2-th branch

Fig 2.23: Received signal after diversity operation.

In summary, a MIMO system could provide two types of gains: diversity gain and spatial multiplexing gain. To increase a kind of gain will reduce the second gain with no doubt. There is a fundamental trade off between how much of each type of gain any MIMO system scheme could extract: higer spatial multiplexing gain comes at the price of sacrificing diversity gain. L. Zeng and N. C. Tse [89] provided an optimal tradeoff curve between diversity gain and multiplexing gain. To formulate their result, we firstly give the following definition. Definition 1: A MIMO scheme{C(SNR)} is said to achieve spatial multiplexing gain r and diversity gain d if the data rate

43

R( SNR) =r SNR log( SNR ) lim and the average error probability lim log Pe ( SNR) =d log( SNR)

(2-42)

SNR

(2-43)

So we have the following result presented in Theorem 1 [89]. Theorem 1: The optimal tradeoff curve d*(r) is given by the piecewise-linear function connecting the points (k, d*(k)), k = 0,1,..., min(NR, NT), where d*(k) = (NR-k)(NT-k). (2-44)

where d is the diversity gain defined in (2-43), r is the multiplexing gain defined in (2-42). NR is the number of receiver antennas and NT is the number of transmitter antennas.

Fig 2.24: Diversity-multiplexing tradeoff, d*(r) versus r.

44

Fig 2.24 further plots the diversity-multiplexing tradeoff curve. Observe that the optimal
* = min(NR, NT), which means that the maximum tradeoff curve intersects the r axis at rmax * achievable spatial multiplexing gain rmax is the total number of degrees of freedom provided

by the channel, as suggested by the ergodic capacity result. And at this point, no positive
* , the data rate approaches the diversity gain could be achieved. That is, as r approaches rmax

ergodic capacity and there is not protection against the randomness in the fading channel.

2.3.3 Overview of Space Time codes

Space time codes for MIMO systems are divided into three kinds, including the space time block code (STBC) [9], space time trellis code (STTC) [12], and layered space time code (LSTC) [6]-[7].

Spacetime trellis codes (STTCs) distribute a trellis code over multiple antennas and multiple time-slots and provide both coding gain and diversity gain. Spacetime block codes (STBCs) act on a block of data at a time and provide only diversity gain, but are much less complex in implementation compared with the STTCs. Both the STBC systems and STTC systems are designed to achieve the diversity gain.

The LSTC systems are used to improve the data transmission rate, that is, to acquire the spatial multiplexing gain. LSTC systems were first proposed by G. J. Foschini [6] in 1996. The Bell laboratories layered space time (BLAST) system was exploited by G. J. Foschini at Lucent Technologies' Bell Laboratories. The system is the earliest MIMO system in history. Multiple antennas are equipped at both the transmitter and the receiver in order to

45

exploit the many different paths in a highly scattering wireless environment. By careful allocation of the data to be transmitted to the transmitting antennas, multiple data streams were transmitted simultaneously and independently within a single frequency band. Their results showed that the capacity of the system grew directly in line with the relatively small value of the number of transmitter antennas and the number of receiver antennas. However, the performance of anti-fading is not good enough. Next, we introduce two classic space time codes for MIMO systems, that is, STBC and V-BLAST systems.

2.3.3.1 Space time block code (STBC) for MIMO systems

s1

s
T ran sm itter A n ten n a 1

* 2

s2
s 1*
T ra n sm itter A n ten n a 2

h 12
h1 1
R e ce iv er A n ten n a 1 AW GN n o ise w 1

h 21 h 22
R e ceiv er A n ten n a 2

+
Y1
C h an n e l estim ato r 1

+
h1 1
C o m b in atio n

h 12 h 22

Y2
C h an n e l estim ato r 2

AW GN n o ise w 2

h1 1

h 21

h 21

S 1

S2

h 12

h 22

M a x im u m L ik elih o o d d etec tio n

s1

s2

Fig 2.25: The Alamouti STBC diagram for 22 MIMO systems.

46

STBC was first proposed by Alamouti in 1998 and was further developed by V. Tarohk et. al [91] according to the general orthogonal design principle. Since the column of the coding matrix is orthogonal to the row of the coding matrix at the transmitter, the receiver can perform maximum likelihood (ML) detection with low computational complexity. Fig 2.25 shows the Alamouti STBC diagram, for NR = NT = 2, where NR is the number of receiver antennas, NT is the number of transmitter antennas. At time t, two transmitted signal S1 and S2 are sent out. At the next time t + T, the conjugate versions of the previous signals are sent out. The superscript ()* in Fig 2.25 denotes conjugate. We use matrix to represent the transmitted signal.
s X= 1 s2
* s2 * s1

(2-45)

Let X1 denote the first row of the matrix X and X2 denote the second row of the matrix X. We have
X1 i X 2 = 0

(2-46)

where (i) denotes inner product of two vectors. That is, the two rows are orthogonal. Next, the fading channel is assumed to be flat and static within two adjacent symbols, that is, hij(t) = hij(t +T), for i = j = 1, 2. At time t, the received signals Y1(t) and Y2(t) are given by Y1 (t ) = h11s1 + h21s2 + w1 (t ) Y2 (t ) = h21s1 + h22 s2 + w2 (t ) At time t + T, the received signals Y1(t+T) and Y2(t+T) are given by
* * Y1 (t + T ) = h11s2 + h21s1 + w1 (t + T )

(2-47)

47

* * Y2 (t + T ) = h21s2 + h22 s1 + w2 (t + T )

(2-48)

It is assumed that the channel estimators in Fig 2.25 are perfect. The output of the channel
and s2 , and estimator hij, i, j = 1, 2, is then put into the combiner. The combiner generates s1 , s2 are given by s1
* * = h11 s1 Y1 (t ) + h21Y1* (t + T ) + h12 Y2 (t ) + h22Y2* (t + T ) * * * * (h11s1 + h21s2 + w1 (t )) + h21[h11 ( s2 ) + h21 (t + T )] + s1 + w1 = h11 * * * * (h12 s1 + h22 s2 + w2 (t )) + h22 [h12 ( s2 ) + h22 (t + T )] h12 s1 + w2 * * * * (t + T ) + h12 w2 (t ) + h22 w2 (t + T ) w1 (t ) + h21w1 = (| h11 |2 + | h21 |2 + | h12 |2 + | h22 |2 ) s1 + h11

= (| h11 |2 + | h21 |2 + | h12 |2 + | h22 |2 ) s1 + w

(2-49)
* * = h11Y1* (t + T ) + h21 s2 Y1 (t ) h12Y2* (t + T ) + h22 Y2 (t ) * * * * = (| h11 |2 + | h21 |2 + | h12 |2 + | h22 |2 ) s2 h11w1 (t + T ) + h21 (t ) + h22 w1 (t + T ) h12 w2 w2 (t + T )

= (| h11 |2 + | h21 |2 + | h12 |2 + | h22 |2 ) s2 + w

(2-50)

, s2 , are put into the maximum likelihood Finally, the two signals of each branch, s1
detector to generate the estimated transmitted signal, that is, s1 and s2 .

48

2.3.3.2 V-BLAST systems

1 1 TX RX
Serial to parallel channel estimation and data detection

RX

TX

RX NT

TX NR

RX RX

Fig 2.26: V-BLAST system diagram.

The vertical Bell Laboratories layered space time (V-BLAST) architecture [7] is a typical example of spatial multiplexing technique. As in Fig 2.26, in V-BLAST systems, each data stream is extracted by a transmitter antenna independently and there is no coding cross transmitter antennas. The fading channel is flat. Thus, the received signal is given by
Y = HX + w

(2-51)
T

where the received signal Y = y1

x1 x2
T

y2

yN R , the transmitted signal X = w2 wN R , the channel matrix H is

T

x NT , the AWGN noise w = w1

given by
H = (hm,n ) N R NT = (am.n + jbm,n ) N R NT .

(2-52)

The aim of data detection for V-BLAST systems is to detect the transmitted signal X from the received signal Y. The algorithms of data detection for V-BLAST systems are divided

49

into four kinds. The followings give a brief overview of the four kinds of algorithms.

Maximum Likelihood (ML) detection

for ML detection is given by The estimated signal x = arg min Y HX x

x

(2-53)

The ML detection is to find the optimal x to satisfy formula (2-53) and the computational complexity is very high although it can achieve the optimal performance in terms of BER.
Zero Forcing (ZF) detection

for ZF detection is given by The estimated signal x = H+Y x (2-54)

where H + = (H H H)1 H H is the pseudo inverse of H. ZF algorithm is very simple and easy to implement, however, the performance is worse than ML algorithm.
MMSE detection

Firstly, the estimated channel matrix G MMSE is acquired by minimizing the cost function
e(H ) = E{|| X HY ||2 } = E{( X HY) H ( X HY)}

(2-55)

By differentiating above formula and making the first order derivative zero, we have the estimated channel matrix is given by
2 2 G MMSE = (H H H + w / X I ) 1 H H

(2-56)

2 is is the power of the transmitted signal X. Therefore, the detected signal X where X

given by
=G X MMSE Y

(2-57)

50

where GMMSE is expressed in (2-56). MMSE detection minimizes the overall error due to noise and mutual interference among different sub streams and achieves better performance in comparison with ZF detection. However, the computational complexity is higher than that of ZF detection.
Successive Interference Cancellation (SIC) detection

The key idea of SIC detection is to distinguish the transmitted signal of each sub stream by successively canceling the interferences between two sub streams. The SIC algorithm can be summarized as follows. Step 1: Initialization: let Y = Y and G1 be the first row of the matrix GMMSE or H+, where
GMMSE is given by (2-56) and H+ is the pseudo inverse of H. Note that if G1 is

chosen from GMMSE, the SIC algorithm is also referred to as MMSE-SIC detection. Similarly, if G1 chosen from H+, the SIC algorithm is referred to as ZF-SIC detection. Step 2: For i = 1, the first sub stream signal is detected by X i = G i Yi Then, the detected signal after decision is given by (2-58)

= arg min || X X || X i i
X C

(2-59)

where C is the set of all code words. Step 3: Update the received signal Yi.
Yi +1 = Yi H i X i

(2-60)

where Hi is the i-th column of the channel matrix H.

51

Step 3: Let i = i+1 and return to Step 2 in order to detect the signal of the next sub stream until all sub streams are detected. In addition, the performance of SIC algorithm is a litter better than MMSE algorithm in terms of BER, since the computational complexity is also a little higher than MMSE algorithm.

2.3.4 Capacity of MIMO systems

In MIMO systems, a transmitter sends multiple streams by multiple transmit antennas. The transmit streams go through a matrix channel which consists of NTNR paths between the NT transmit antennas at the transmitter and NR receive antennas at the receiver. To derive the capacity of MIMO systems, we use the basic MIMO system model depicted in Fig 2.22 and assume that the channel fading is flat. The received signal Y, is given by
Y = HX + w

(2-61)
yN R , the transmitted signal X = w2 wN R , H is defined by formula
T T

where the received signal Y = y1

x1 x2
T

y2

x NT w1 , the AWGN noise w =

(2-38). The MIMO system capacity can be expressed by [46]

C= P log 2 det I N R + HR XX H H 2 R XX ,tr( R XX ) = P NT w
max

(2-62)

where RXX is the autocorrelation matrix of X, P is the total transmitter signal power, I N R is
2 an NR by NR identity matrix, w is the power of AWGN noise. And MIMO system capacity

52

is divided into two categories based on whether the transmitter has the channel state information (CSI).

(1) Open loop capacity for MIMO systems

When the transmitter can not has CSI and only the receiver has full CSI, the optimum power allocation is the equal power allocation on each data substream to achieve the maximum system capacity. On this condition, the MIMO system has the open loop capacity, which is further derived from formula (2-62) and expressed by
SNR Copen = E log det I N R + HH H M

(2-63)

2 . Using singular value decomposition of matrix, where the signal to noise ratio SNR = P / w

the formula (2-63) can be equivalently written as [46] Nmin SNR log 1+ i Copen = E i =1 NT
H min where N min = min( NT , N R ) , {i }iN =1 are the eigenvalues of the matrix HH .

(2-64)

(2) Close loop capacity for MIMO systems

When both the transmitter and the receiver have full CSI, the optimal power allocation is the water-filling allocation, i.e., the total power is allocated to those substreams with high SNRs and not allocated to those substreams with low SNRs. On this condition, the MIMO systems achieve the maximum capacity, that is, the close loop system capacity, expressed by [50]

53

Cclose and u satisfies

Nmin u + = E log i 2 i =1 w

(2-65)

N min

2 w 0, u =P i i =1

(2-66)

min where ( x) + denotes max(0, x), N min = min( NT , N R ) , {i }iN =1 are the eigenvalues of the

matrix HH H .

2.4 MIMO-OFDM systems

Since OFDM technique can mitigate the ISI and transform the frequency selective fading channel into a set of flat fading channels, the combination of MMO with OFDM technique to combat the frequency selective fading of the channel is a trend for wireless communication future. A new protocol draft, IEEE 802.11n, as an amendment to IEEE 802.11 standards has been proposed and investigated [53]. The draft proposes MIMO-OFDM as the physical layer technique and is to be approved by IEEE. OFDM can be extended to MIMO systems by performing the IDFT/DFT and CP operations at each of the transmitter and receiver antennas. The baseband MIMO-OFDM system model is depicted as Fig 2.27.

54

x1

y1

xNT

yN R

Fig 2.27: The baseband MIMO-OFDM system model.

The MIMO-OFDM system is equipped with NT transmitter antennas and NR receiver antennas. It is assumed that the number of subcarriers for one OFDM symbol is N. Each data substream xi is put into serial to parallel module and performs IFFT transform. Then, after adding CP for each substream, the user data are sent out by transmitter antennas. At the receiver, the received signal after FFT operation at the k-th subcarrier, yk, is given by

y k = H k x k + w k , k = 0,1,..., N 1.
where H k is the N R NT channel yk2 for frequency response matrix at the

(2-67) k-th

subcarrier, y k = [ y1 k OFDM symbol xk2

i ykN R ]T , yk is the received signal at the k-th subcarrier of an

the

i-th

receiver

antenna,

()T

denote

transpose.

x k = [ x1 k

i xkNT ]T and xk is the transmitted signal at the k-th subcarrier of an OFDM

symbol for the i-th transmitter antenna, N is the number of subcarriers of an OFDM symbol. w k = [ w1 k
2 wk i wkNt ]T and wk is the additive white Gaussian noise at the k-th

subcarrier and the i-th receiver antenna in the frequency domain.

55

2.5 Summary
This chapter introduces the background knowledge on OFDM systems and MIMO systems. First, we describe the traits of wireless channels and provide the commonly used wireless channels, including Rayleigh fading channel, Ricean fading channel, and Nakagami fading channel. Second, the baseband OFDM system model has been depicted and the processing of OFDM system has been given in detail. Third, we introduce the basic MIMO system model, briefly overview the classic MIMO systems, and discuss two kinds of capacity of MIMO systems, that is, the open loop capacity and the close loop capacity. Finally, the baseband MIMO-OFDM system model has been shown and discussed correspondingly.

56

Chapter 3: Channel estimation for OFDM systems over quasi-static fading channels
In this chapter, a fast LMMSE channel estimation method has been proposed for OFDM systems. In comparison with the conventional LMMSE channel estimation, the proposed channel estimation method does not require the statistic knowledge of the channel in advance and avoids the inverse operation of a large dimension matrix by using the fast FFT operation. Therefore, the computational complexity can be reduced significantly. The NMSE of the proposed method and the conventional LMMSE estimation have been derived. Numerical results show that the NMSE of the proposed method is very close to that of the conventional LMMSE method, which is also verified by computer simulation. In addition, computer simulation shows that the performance of the proposed method is almost the same with that of the conventional LMMSE method in terms of bit error rate (BER). The chapter is organized as follows. Section 3.1 is the introduction to channel estimation for OFDM systems. Section 3.2 describes the OFDM system model. Section 3.3 describes the proposed fast LMMSE channel estimation. We analyze the mean square error (MSE) of the proposed fast LMMSE channel estimation and the MSE of the conventional LMMSE channel estimation in section 3.4. The simulation results and numerical results of the proposed algorithm are discussed in section 3.5 followed by conclusion in Section 3.6.
57

3.1 Introduction
Channel estimation plays an important part in OFDM systems. It can be employed for the purpose of detecting received signal, improving the capacity of orthogonal frequency division multiple access (OFDMA) systems by cross-layer design [13] and improving the system performance in terms of bit error rate (BER) [14][16]. The present channel estimation methods generally can be divided into two kinds. One kind is based on the pilots [17][19], [30] and the other is blind channel estimation [20], [21], [31] which does not use pilots. Blind channel estimation methods avoid the use of pilots and have higher spectral efficiency. However, they often suffer from high computation complexity and low convergence speed since they often need a large amount of receiving data to obtain some statistical information such as cyclostationarity induced by the cyclic prefix. Therefore, blind channel estimation methods are not suitable for applications with fast varying fading channels. And most practical communication systems such as World Interoperability for Microwave Access (WiMAX) system adopt pilot assisted channel estimation, so this chapter studies the first kind. For the pilot-aided channel estimation methods, there are two classical pilot patterns, which are the block-type pattern and the comb-type pattern [15]. The block-type refers to that the pilots are inserted into all the subcarriers of one OFDM symbol with a certain period. The block-type can be adopted in slow fading channel, i.e., the channel is stationary within a certain period of OFDM symbols. The comb-type refers to that the pilots are inserted at some specific subcarriers in each OFDM symbol. The comb-type is
58

preferable in fast varying fading channels, i.e., the channel varies over two adjacent OFDM symbols but remains stationary within one OFDM symbol. The comb-type pilot arrangement based channel estimation has been shown as more applicable since it can track fast varying fading channels, compared with the block-type one [15], [25]. The channel estimation based on comb-type pilot arrangement is often performed by two steps. Firstly, it estimates the channel frequency response on all pilot subcarriers, by LS method, LMMSE method and so on. Secondly, it obtains the channel estimates on all subcarriers by interpolation, including data subcarriers and pilot subcarriers in one OFDM symbol. There are several interpolation methods including linear interpolation method, second order polynomial interpolation method, phase compensated interpolation [15] and so on. In [22], the LMMSE channel estimation method based on channel autocorrelation matrix in frequency domain has been proposed. To reduce the computational complexity of LMMSE estimation, a low rank approximation to LMMSE estimation has been proposed by singular value decomposition [17]. The drawback of LMMSE channel estimation [17], [22] is that it requires the knowledge of channel autocorrelation matrix in frequency domain and the signal to noise ratio (SNR). Though the system can be designed for fixed SNR and channel frequency autocorrelation matrix, the performance of the OFDM system will degrade significantly due to the mismatched system parameters. In [23], a channel estimation exploiting channel correlation both in time and frequency domain has been proposed. Similarly, it needs to know the channel autocorrelation matrix in frequency domain, the Doppler shift and SNR in advance. Mismatched parameters of

59

the Doppler shift and the delay spread will degrade the performance of the system [24]. It is noted that the channel estimation methods proposed in [17], [22][24] can be adopted in either the block-type pilot pattern or the comb-type pilot pattern. When the assumption that the channel is time-invariant within one OFDM symbol is not valid due to high Doppler shift or synchronization error, the ICI has to be considered. Some channel estimation and signal detection methods have been proposed to compensate the ICI effect [28], [29]. In [28], a new equalization technique to suppress ICI in LMMSE sense has been proposed. Meanwhile, the authors reduced the complexity of channel estimator by using the energy distribution information of the channel frequency matrix. In [29], the authors proposed a new pilot pattern, i.e., the grouped and equi-spaced pilot pattern and corresponding channel estimation and signal detection to suppress ICI. In this chapter, the OFDM system framework based on comb-type pilot arrangement is adopted and we assume that the channel remains stationary within one OFDM symbol and therefore there is no ICI effect. We propose a fast LMMSE channel estimation method. The proposed method has three advantages over the conventional LMMSE method. Firstly, the proposed method does not require the knowledge of channel autocorrelation matrix and SNR in advance but can achieve almost the same performance with the conventional LMMSE channel estimation in terms of the NMSE of channel estimation and BER. Secondly, the proposed method need only fast FFT operation instead of the inversion operation of a large dimensional matrix. Therefore, the computational complexity can be reduced significantly, compared with the conventional LMMSE method. Thirdly, the

60

proposed method can track the changes of channel parameters, i.e., the channel autocorrelation matrix and SNR. However, the conventional LMMSE method can not track the channel. Once the channel parameters change, the performance of the conventional LMMSE method will degrade due to the parameter mismatch.

3.2 System Model

The OFDM system model with pilot signal (i.e., training sequence) assisted is shown in Fig 3.1.
X (i, k )

x(i, n)

X (i, k )

Y (i, k )

H (i, k )

Yp (i, k )

Fig 3.1: Baseband OFDM system.

For N subcarriers in the OFDM system, the transmitted signal x(i, n) in time domain after IFFT is given by

61

x(i, n) = IFFTN [ X (i, k )] =

1 N 1 X (i, k ) exp{ j 2 nk / N} N k =0

(3-1)

where X (i, k ) denotes the transmitted signal in frequency domain at the k-th subcarrier in the i-th OFDM symbol. The comb-type pilot pattern [15] is adopted in this chapter. The pilot subcarriers are equi-spaced inserted into each OFDM symbol. It is assumed that the number of the total pilot subcarriers is N p and the inserting gap is R . Each OFDM symbol is composed of the pilot subcarriers and the data subcarriers. It is assumed that the index of the first pilot subcarrier is k0 . Therefore, the set of the indexes of pilot subcarriers, , can be written as

= {k | k = mR + k0 , m = 0,1,..., N p 1}

(3-2)

where k0 [0, R) . The received signal Y (i, k ) in frequency domain after FFT can be written as
Y (i, k ) = X (i, k ) H (i, k ) + W (i, k )

(3-3)

2 where W (i, k ) denotes the AGWN with zero mean and variance w , H (i, k ) is the frequency

response of the radio channel at the k-th subcarrier of the i-th OFDM symbol. Then, the received pilot signal Yp (i, k ) is extracted from Y (i, k ) to perform channel estimation.
H p (i, k )

H (i, k )

Yp (i, k )

X p (i, k )

Fig 3.2: Channel estimation based on comb-type pilots.

62

As shown in Fig 3.2, the channel estimator firstly performs channel frequency response estimation at pilot subcarriers. There are some channel estimation methods for this part such as LS, LMMSE estimator [15] and so on. Next, once the channel frequency response estimation at pilot subcarriers, H p (i, k ) , is obtained, the estimator performs interpolation to obtain channel frequency response estimation at all subcarriers. There are linear interpolation method [15], second-order polynomial interpolation method [15], DFT based interpolation method [26] and so on. In our system model, the linear interpolation method is adopted. After channel estimation, maximum likelihood detection is performed to obtain the estimated frequency signal X (i, k ) . The X (i, k ) is given by
X (i, k ) = arg min | Y (i, k ) H (i, k ) S |2
S

(3-4)

where S s and s is the set containing all constellation points, which depends on modulation method, i.e., the signal mapper. For instance, if QPSK modulation is adopted, the set
s ={ 1 1 1 1 (1 + j ), (1 j ), (1 + j ), (1 j )} . Finally, the estimated 2 2 2 2

frequency signal X (i, k ) passes through the signal demapper to obtain the received bit sequence.

3.3 The Proposed Fast LMMSE Algorithm

3.3.1 Properties of the channel correlation matrix in frequency domain
The channel impulse response in time domain can be expressed as

63

h(i, n) = hl (i ) (n l )
l =0

L 1

(3-5)

where hl (i) is the complex gain of the l-th path in the i-th OFDM symbol period, () is the Kronecker delta function, l is the delay of the l-th path in unit of sample point, L is the number of resolvable paths. Assume that different paths hl (i ) are independent from each
2 other and the power of the l-th path is l2 . The channel is normalized so that h = 2 = 1 . l
l

The channel response in frequency domain H (i, k ) is the FFT of h(i, n) and it is given by

H (i, k ) = FFTN (h(i, n)) = h(i, m) exp{ j 2 mk / N }

m=0

N 1

(3-6)

where FFTN (i) denotes N points FFT operation. The channel autocorrelation matrix in frequency domain can be expressed as RHH (m, n) = E[ H (i, m) H * (i, n)]
N 1 N 1 = E h(i, k ) exp{ j 2 km / N } h* (i, k ) exp{ j 2 kn / N } k =0 k =0

= E{| h(i, k ) |2 }exp{ j 2 k (m n) / N }

k =0

N 1

(3-7)

= l2 exp{ j 2 l (m n) / N }
l =0

L 1

where E (i) denotes expectation. Denote the vector form of the channel autocorrelation matrix by R HH and we have R HH = [ RHH (i, j )]N N . It is easy to find that the matrix, R HH , is a circulant matrix. Therefore, as in [27], the eigenvalues of R HH are given by

[0 1

N 1 ] = [FFTN ( RHH (0,0) RHH (0,1)

RHH (0, N 1))].

(3-8)

The formula (3-8) can be equivalently written as

k = RHH (0, n) exp{ j 2 nk / N }, k = 0,1,..., N 1.

n =0

N 1

(3-9)

64

We can easily obtain from (3-7) and (3-9) that the number of non-zero eigenvalues of R HH is equal to the total number of resolvable paths, L (see Appendix A). It is known by us that the rank of a square matrix is the number of its nonzero eigenvalues. Therefore the rank of R HH is L and R HH is a singular matrix since L < N . The matrix R HH does not have the inverse matrix and has only the Moore-Penrose inverse matrix. However, the rank of
2 is N (see Appendix A), where I is an N by N identity matrix. the matrix R HH + w 2 is not singular and has the inverse matrix. Therefore, the matrix R HH + w

3.3.2 The proposed fast LMMSE channel estimation algorithm

Let
H p (i ) = [ H p (i, 0) H p (i,1)

H p (i, N p 1)]T

(3-10)

denote the channel frequency response at pilot subcarriers of the i-th OFDM symbol and
Yp (i ) = [Yp (i, 0) Yp (i,1)

Yp (i, N p 1)]T

(3-11)

denote the vector of received signal at pilot subcarriers of the i-th OFDM symbol after FFT. Denote the pilot signal of the i-th OFDM symbol by X p (i, j ), j = 0,1,..., N p 1 . The channel estimate at pilot subcarriers based on LS criterion is given by
H p ,ls (i ) = [ H p ,ls (i, 0) H p ,ls (i,1)
Y (i, 0) Yp (i,1) = p X p (i, 0) X p (i,1) H p ,ls (i, N p 1)]T
T

Yp (i, N p 1) . X p (i, N p 1)

(3-12)

The LMMSE estimator at pilot subcarriers is given by [17]

H p ,lmmse (i ) = [ H p ,lmmse (i, 0) H p ,lmmse (i,1)
1

H p ,lmmse (i, N p 1)]

= R Hp Hp R Hp Hp + I H p ,ls (i ) SNR
65

(3-13)

where R Hp Hp is channel autocorrelation matrix at pilot subcarriers and is defined by

H R Hp Hp = E{H p H H p } , where () denotes Hermitian transpose. It is easy to verify that the
2 matrix R Hp Hp is circulant, the rank of R Hp Hp is equal to L and the rank of R Hp Hp + w is

2 equal to Np. The signal to noise ratio (SNR) is defined by SNR = E | X p (k ) |2 / w and

= E | X p (k ) |2 E |1/ X p (k ) |2 is a constant depending on the signal constellation. For

16QAM modulation = 17 / 9 and for QPSK and BPSK modulation = 1 . If the channel autocorrelation matrix
R Hp Hp
1

and

SNR

are

known

in

advance, R Hp Hp R Hp Hp + I needs to be calculated only once. However, the SNR

autocorrelation matrix R Hp Hp and SNR are often unknown in advance and time varying. Therefore the LMMSE channel estimator becomes unavailable in practice. To solve the problem, we propose the fast LMMSE channel estimation algorithm. The algorithm can be divided into three steps. The first step is to obtain the estimate of channel autocorrelation matrix R Hp Hp , R Hp Hp . Firstly, we obtain the LS channel estimation at pilot subcarriers in time domain, hp.ls (i, k ) and it is given by
hp.ls (i, k ) =
1 Np
N p 1 n =0

p ,ls

(i, n) exp{ j 2 nk / N p }, k = 0,1,..., N p 1.

(3-14)

Secondly, the most significant taps (MST) algorithm [32] has been proposed to obtain the refined channel estimation in time domain. The MST algorithm deals with each OFDM symbol by reserving the most significant L paths in terms of power and setting the other taps to be zero. The algorithm can reduce the influence of AWGN and other interferences, significantly, compared with the LS method. Other interferences may result from the

66

interference from other paths. However, the algorithm may choose the wrong paths and omit the right paths because of the influence of AWGN and other interference. Thus, we will improve the algorithm of [32] by processing several adjacent OFDM symbols jointly. We calculate the average power of each tap for NMST adjacent OFDM symbols, PLS (k ) and it is given by PLS (k ) = 1 N MST
N MST 1

i =0

| hp ,ls (i, k ) |2 , k = 0,1,..., N p 1.

(3-15)

Then we choose the L most significant taps from PLS (k ) and reserve the indexes of them into a set . Finally, the refined channel estimation in time domain, hp , MST , is given by h (i, k ), if k hp , MST (i, k ) = p ,ls , k = 0,1,..., N p 1, i = 0,1,..., N MST 1. 0, if k Denote the first row of the matrix R Hp Hp by A . Then A can be given from (3-7) by
A = N p IFFTN p [ PMST ]

(3-16)

(3-17)

where PMST is a 1 by Np vector with each entry

P ( k ), if k PMST (k ) = LS , k = 0,1,..., N p 1. 0, if k

(3-18)

Since the matrix R Hp Hp is circulant, R Hp Hp can be acquired by circle shift of A . The second step is to obtain the estimate of SNR. The estimate of SNR, SNR , is given by
SNR =

P P
k k LS

MST

(k ) .

( k ) PMST ( k )
k

(3-19)

The third step is to obtain the estimate of the matrix R Hp Hp R Hp Hp + I , SNR R Hp Hp R Hp Hp + I . We refer to the matrix R Hp Hp R Hp Hp + I as the LMMSE SNR SNR
67
1 1

matrix in this chapter. Since R Hp Hp

is a circulant matrix and R Hp Hp + I is a SNR

1

circulant matrix, the product of R Hp Hp and R Hp Hp + I is also a circulant matrix. SNR

Therefore, we need only compute the estimate of the first row of the LMMSE matrix. Denote the first row of LMMSE matrix by B . The estimate of B , B , is given by (see Appendix B) PMST (0) PMST (1) B = IFFTN p P (0) + PMST (1) + MST N p SNR N p SNR PMST ( N p 1) PMST ( N p 1) + N p SNR

(3-20)

where IFFTN p (i) denotes N p points IFFT operation. Therefore the estimated LMMSE

matrix R Hp Hp R Hp Hp + I can be obtained from circle shift of B . The channel SNR

estimation in frequency domain at pilot subcarriers for the i-th OFDM symbol can be given by

H p , fast lmmse (i ) = R Hp Hp R Hp Hp + I H p ,ls (i ), i = 0,1,..., N MST 1. SNR

(3-21)

The proposed fast LMMSE algorithm avoids the matrix inverse operation and can be very efficient since the algorithm only uses the FFT and circle shift operation. The proposed fast LMMSE algorithm can be summarized as follows. Step 1: Obtain the LS channel estimation of pilot signal in time domain, hp.ls (i, k ) , by formula (3-14). Step 2: Calculate the average power of each tap for NMST OFDM symbols, PLS (k ) , by formula (3-15). Then, choose the L most significant taps from PLS (k ) and
68

reserve it as PMST (k ) , by formula (3-18). Step 3: Obtain the estimate of SNR, SNR , by formula (3-19). Step 4: Obtain the estimate of the first row of the LMMSE matrix, B , by formula (3-20).

Step 5: Obtain the estimation of the LMMSE matrix, R Hp Hp R Hp Hp + I , by SNR

circle shift of B . Then, the channel estimation in frequency domain at pilot subcarriers can be obtained by formula (3-21). It is noted that the estimation of the LMMSE matrix requires only Np points FFT operation and circle shifting operation, which reduce the computational complexity significantly compared with the conventional LMMSE estimator since it requires the inverse operation of a large dimension matrix.

3.3.3 Computational complexity comparison between the proposed method and the conventional LMMSE method
For an n by n matrix, its inverse requires (n3-n)/3 operations is the most commonly used algorithm, that is, Gaussian Elimination is adopted. The usual way to count operations is to count one for each "division" (by a pivot) and one for each "multiply-subtract" when eliminating an entry of the matrix. Since addition and subtraction consume much less computational time compared with division or multiplication, we do not consider addition and subtraction as one computation. It is assumed that a complex division or multiplication has computational complexity O(1) and counts one computation. Thus,

69

the inverse of an n by n matrix requires (n3-n)/3 computations and has computational complexity O(n3). The LMMSE matrix needs a inverse of an Np by Np matrix and a multiplication of two Np by Np matrices, therefore the number of computations for the conventional LMMSE method is (Np3- Np)/3+ Np3 = (4 Np3- Np)/3. Hence the conventional method to compute the LMMSE matrix has complexity O(Np3). For the proposed method, Step 1 requires a Np-point FFT operation; Step 2 requires Np+1 computations; Step 3 requires 1 computations; Step 4 requires a Np-point FFT operation; Step 5 does not require computation. As a n-point FFT requires nlog2(n) computations if the most commonly used algorithm, i.e., Cooley-Tukey algorithm [94] is adopted. Therefore, the proposed method requires 2Nplog2(Np)+ Np+2 computations and hence has complexity O(Nplog2(Np)). Therefore, the proposed fast LMMSE algorithm has less computation complexity compared with the conventional LMMSE algorithm.

3.4 Analysis of the Mean Square Error (MSE) of the Proposed Fast LMMSE Algorithm
In this section, we will present the mean square error (MSE) of the proposed fast LMMSE algorithm. Firstly, we present the MSE of LMMSE algorithm for comparison. We study two cases. One case is the MSE analysis for matched SNR, i.e., the designed SNR is equal to the true SNR and the other one is the MSE analysis for mismatched SNR. Secondly, we present the MSE of the proposed fast LMMSE algorithm. Similarly, we study two cases. One is for matched SNR and the other is for mismatched SNR.

70

3.4.1 MSE analysis of the conventional LMMSE algorithm

Denote the MSE of LMMSE algorithm by MSE ( SNR, SNRdesign ) , where SNR is the true SNR and SNRdesign is the designed SNR. (i) MSE analysis for matched SNR The MSE of LMMSE algorithm at pilot subcarriers for matched SNR can be derived as [34]

MSE ( SNR, SNR )

1 = Np
N p 1 k =0

E|H

p ,lmmse

(i, k ) H p (i, k ) |2

(3-22)

= 1 A R Hp Hp

+ I AH SNR

where A is the first row of the matrix R Hp Hp and () H denotes Hermitian transpose. (ii) MSE analysis for mismatched SNR The MSE of LMMSE algorithm on pilot subcarriers for mismatched SNR can be derived as [34]

MSE ( SNR, SNRdesign )

= 1 Np
N p 1 k =0

E|H

p ,lmmse

(i, k ) H p (i, k ) |2 I R Hp Hp + I R Hp Hp SNR

1

= 1 + A R Hp Hp + SNRdesign 2 A R Hp Hp

SNRdesign

I AH

I AH SNR design

(3-23)

where A is the first row of the matrix R Hp Hp and () H denotes Hermitian transpose.

71

3.4.2 MSE analysis for the proposed fast LMMSE algorithm

Let us denote the MSE of the proposed fast LMMSE algorithm by MSE ( SNR, SNR ) , where SNR is the true SNR and SNR is the estimated SNR or the designed SNR. (i) MSE for matched SNR The MSE of the proposed fast LMMSE algorithm is given by

MSE ( SNR, SNR )

1 = E Np
N p 1 k =0

|H

2 ( i , k ) H ( i , k ) | p , fast lmmse p

2 = E | H p , fast lmmse (i, 0) H p (i, 0) |

N p 1 1 = E | Np k =0

N p 1 l =0

(l ) exp{ j N
N p 1 k =0

lk}H p ,ls (i, k ) H p (i, 0) |2 p

N p 1 1 = E | (l ) Np l =0

exp{ j N

lk}H p ,ls (i, k ) H p (i, 0) |2 p

N p 1 = E | (l )hp ,ls (i, l ) H p (i, 0) |2 l =0

N p 1 N p 1 = E | (l )hp ,ls (i, l ) hp (i, j ) |2 j =0 l =0

(3-24)

where (l ) =

PMST (l ) PMST (l ) +

, l=0,1,, Np-1. If the number of the chosen OFDM

N p SNR

symbol to obtain the estimated average power for each tap, NMST, is large, we can replace (l ) with E ( (l )) in equation (3-24). Then, equation (3-24) can be further derived as

72

MSE ( SNR, SNR)

N p 1 N p 1 = E | E[ (l )]hp ,ls (i, l ) hp (i, j ) |2 j =0 l =0 2 N p 1 N p 1 E (| hp , MST (l ) | ) E | hp ,ls (i, l ) hp (i, j ) |2 . l =0 E (| h j =0 (l ) |2 ) + , p MST N p SNR

(3-25) If the improved MST algorithm chooses L ( L L ) paths, where L is number of resolvable paths of the dispersive channel, and the chosen L paths contain all the L channel paths without omission. Then equation (3-25) can be further written as

MSE ( SNR, SNR )

2 N p 1 N p 1 E (| hp , MST ( j ) | ) hp ,ls (i, j ) hp (i, j ) |2 = E | 2 j = 0 E (| h j =0 p , MST ( j ) | ) + N SNR p

1 L 1 SNR N p 1 ) + ( L L ) = 1 + | 1 ( l ) |2 ( l2 + 1 SNR N p l =0 + SNR N SNR N p p

2

L 1 1 2 1 ( l ) l2 SNR N p l =0

1 L 1 L 1 1 1 2 2 SNR = 1 + | 1 ( l ) | ( l + 2 1 ( l ) l2 ) + ( L L ) SNR N p l =0 l =0 1 + SNR N p SNR SNR

(3-26) where l is the channel delay of the l-th resolvable path, l2 is the power of the l-th

73

1 l2 + SNR N p , if i ; 1 2 + + l SNR N SNR N p p , = { l : l = 0,1,..., L 1} . path, 1 (i) = 1 SNR , if i . 1 + SNR SNR

(ii) MSE for mismatched SNR Similarly, the MSE of the proposed fast LMMSE algorithm for mismatched SNR is given by

MSE ( SNR, SNR)

N p 1 N p 1 = E | (l )hp ,ls (i, l ) hp (i, j ) |2 j =0 l =0

1 L 1 L 1 1 1 SNR 2 ( l ) l2 = 1 + | 2 ( l ) |2 ( l2 + ) + ( L L ) 2 1 SNR N p SNR N p l =0 l =0 + SNR SNR (3-27) where (l ) = PMST (l ) PMST (l ) + resolvable path,

, l=0,1,, Np-1. l is the channel delay of the l-th

N p SNR

l2

is

the

power

of

the

l-th

path,

1 2 + l SNR N p , if i ; 1 2 + + l SNR N p SNR N p , = { l : l = 0,1,..., L 1} . 2 (i ) = 1 SNR , if i . 1 + SNR SNR

74

It is noted that since the channel is assumed to be normalized, the MSE of the proposed fast LMMSE algorithm and the MSE of the conventional LMMSE are equal to their NMSE, respectively. In addition, for the sake of performance comparison between the above analysis of NMSE and the NMSE obtained by computer simulation, we define the NMSE obtained by simulation as follows.

NMSEsimu =

|H
i =0 j =0

K 1 N p 1

(i, j ) H p (i, j ) |2 (3-28)

p

|H
i =0 j =0

K 1 N p 1

(i, j ) |

where H p (i, j ) denotes the channel estimate at the j-th pilot subcarrier in the i-th OFDM symbol, obtained by LMMSE algorithm or the proposed fast LMMSE algorithm, K denotes the number of OFDM symbols in the simulation.

3.5 Numerical and Simulation Results

Both computer simulation and numerical method have been deployed to investigate the performance of the proposed fast LMMSE algorithm for channel estimation. In the simulation, we employ the channel model of COST207 [35] having 6 paths, i.e., L=6, and the maximum delay spread of 2.5 s . The channel power intensity profile is listed in TABLE 3.1. The number of subcarriers of the OFDM system, N, is equal to 2048 and the CP length is equal to 128 sample points. The bandwidth of the system is 20 MHz so that one OFDM symbol period Ts = 102.4 s and the CP period TCP = 6.4 s > 2.5 s . The number of total pilots N p is equal to 128 and the pilot gap R is 16. The transmitted signal is BPSK modulated and the Doppler shift is 100 Hz.
75

a. Channel autocorrelation matrix under different SNRs Fig 3.3 shows the magnitude of the first row of the channel autocorrelation matrix R Hp Hp , A. Since the channel autocorrelation matrix is circulant, it is enough to show the first row of the channel autocorrelation matrix. Observe that the magnitude of A varies approximately periodically and the period is 13 pilot subcarriers. Since the channel power intensity profile is negative exponential distributed, the period of the first row of the channel autocorrelation matrix is decided by the delay of the second path. The delay of the second path is 0.5 s , i.e., 10 sample points. According to equation (3-7), the period is Np/1=128/10=12.8. It is noted that the parameter N should be replaced by Np in equation (3-7). Therefore, the period is about 13, as shown in Fig 3.3. Fig 3.4 shows the magnitude

of the first row of the LMMSE matrix R Hp Hp R Hp Hp + I with SNR of 5 dB, 10 dB SNR
and 20 dB, respectively. Since the LMMSE matrix is also circulant, it is sufficient to depict the first row of the LMMSE matrix. Observe that the value of the first row of the LMMSE matrix is symmetry and the center point is 64. Since the first row of the LMMSE matrix is an NP-point FFT expressed by (B-9), the magnitude of the first row is symmetric. The first row of the LMMSE matrix is approximately periodic and the period is about 13 pilot subcarriers. Observe that the value of the first row of the LMMSE matrix varies insignificantly when SNR changes from 5 dB to 20 dB. In addition, the local maximum values of the curves correspond to strong correlation between pilot subcarriers and the local minimum values correspond to weak correlation between pilot subcarriers. b. Normalized mean square error (NMSE) comparison of channel estimation between
76

LMMSE algorithm and the proposed fast LMMSE algorithm Fig 3.5 shows the NMSE of channel estimation of LMMSE algorithm versus that of the proposed fast LMMSE algorithm by computer simulation and numerical method, respectively. The numerical results of LMMSE algorithm and the proposed fast LMMSE algorithm are obtained by equation (3-22) and (3-26), respectively. The simulation results are obtained by equation (3-28). We replace H p in (3-28) with H p ,lmmse for LMMSE algorithm and replace H p with H p , fast lmmse for the proposed LMMSE algorithm, respectively. For the proposed fast LMMSE algorithm, the number of OFDM symbols chosen to obtain the average power of each tap, NMST, is 20 and the number of chosen paths, L , is 10. The number of OFDM symbols in the simulation, K, is 5000, for both LMMSE algorithm and the proposed fast LMMSE algorithm. Observe that the NMSE of the proposed fast LMMSE algorithm is very close to that of LMMSE algorithm in theory over the SNR range from 0 dB to 25 dB. In addition, for LMMSE algorithm the numerical result is verified by the simulation. For the proposed fast LMMSE algorithm, the simulation result approaches the numerical result well, except that the simulation result is a little higher than the numerical result at low SNR. Observe that both the proposed fast LMMSE algorithm and LMMSE algorithm are superior to LS algorithm. For instance, the LMMSE algorithm has about 16 dB gain over the LS algorithm, at the same MSE over the SNR range from 0 dB to 25 dB. Fig 3.6 shows the NMSE of LMMSE algorithm with matched SNR and mismatched SNRs versus SNR, by simulation and numerical method, respectively. Firstly, we give a

77

necessary illustration of the curves obtained by numerical method. For the curves with matched SNR, we use equation (3-22) to calculate the MSEs under different SNRs, by numerical method. For the curves with mismatched SNRs, i.e., designed SNRs, we use equation (3-23) to obtain the results, by numerical method. Secondly, for the curves with mismatched SNRs obtained by computer simulation, we use the designed SNR (predetermined and invariable) instead of the true SNR in equation (3-13) to obtain the channel estimation of pilot subcarriers. Observe that the analysis results are verified by computer simulation well, for the designed SNR of 5 dB, 10 dB and 20 dB, respectively. For the case of the designed SNR of 5 dB, the MSE approaches the curve of matched SNR well within the range from 0 dB to about 10 dB. However, when the SNR increases, a MSE floor of about 2 103 occurs. Similar trend can be found for the case of designed SNR of 10 dB. Observe that the curve of designed 20 dB approaches the curve with matched SNR well within the SNR range from 0 dB to 25 dB. Therefore, if we only know the channel autocorrelation matrix R Hp Hp and do not know the SNR, the above results suggest that we use a higher designed SNR in equation (3-13) when performing channel estimation. Fig 3.7 shows the NMSE of the proposed fast LMMSE algorithm with matched SNR and mismatched SNRs versus SNR, by simulation and numerical method respectively. Firstly, we give a brief illustration of the curves obtained by numerical method. For the curve with matched SNR, we use equation (3-26) to obtain the results. For the curves with mismatched SNRs, i.e., designed SNR, we use equation (3-27) to obtain the numerical

78

results. To verify the numerical results, we perform computer simulation for each case with different designed SNR. In the computer simulation, the step 3 in the proposed fast LMMSE algorithm is modified by letting the estimated SNR, SNR , be the designed SNR. For instance, if we choose the designed SNR to be 10 dB, SNR will be set to be 10 dB in the step 3 of the proposed fast LMMSE algorithm instead of using formula (3-19) to obtain SNR . For the computer simulation, the number of OFDM symbols chosen to obtain the average power of each tap, NMST, is 20 and the number of chosen paths, L , is 10. The number of OFDM symbols in the simulation, K, is 5000. Observe that the analysis results are verified by computer simulation well, for the designed SNR of 5 dB, 10 dB and 20 dB, respectively. For the case of the designed SNR of 5 dB, the MSE approaches the curve of matched SNR well within the range from 0 dB to about 10 dB. However, when the SNR increases, a MSE floor of about 2 103 occurs. Similar trend can be found for the case of designed SNR of 10 dB. Observe that the curve of designed 20 dB approaches the curve of matched SNR well within the SNR range from 0 dB to 25 dB. c. Bit error rate (BER) comparison between LMMSE algorithm and the proposed fast LMMSE algorithm Fig 3.8 shows the BER of LS, LMMSE, the proposed fast LMMSE and perfect channel estimation, respectively. We adopt linear interpolation to obtain the channel frequency response at all subcarriers after the channel frequency response at pilot subcarriers is obtained by LS, LMMSE, and the proposed fast LMMSE estimator. Once the channel frequency response is obtained, we use maximum likelihood detection to obtain the

79

estimated signal X (i, k ) . In addition, the perfect channel estimation refers to that the channel frequency response is known by the receiver in advance. Observe that the BER of LMMSE estimator is very close to that of the proposed fast LMMSE estimator over the SNR range from 0 dB to 25 dB. And they are about 1 dB worse than the perfect channel estimator, over the SNR ranging from 0 dB to 25 dB. The LMMSE estimator and the proposed LMMSE estimator are about 3-4 dB better than the LS estimator at the same BER over the SNR ranging from 0 dB to 25 dB. Fig 3.9 shows the BER performance of the LMMSE channel estimation with matched SNR and the LMMSE channel estimation with designed SNRs. The LMMSE channel estimator with designed SNR refers to that we use a predetermined and unchanged SNR in equation (3-13) instead of the true SNR. Observe that the BER of the LMMSE with designed SNR of 5 dB, 10 dB and 20 dB are almost overlapped with each other within the lower SNR range from 0 dB to 15 dB. However, when SNR increases from 15 dB to 25 dB, the BER of the LMMSE estimator with higher designed SNR is better than that of the lower designed SNR. The results are consistent with the NMSEs in Fig 3.4. Therefore, a design for higher SNR is preferable as for mismatch in SNR. Fig 3.10 shows the BER of the proposed fast LMMSE estimator with estimated SNR and the proposed fast LMMSE estimator with designed SNRs. It is noted that the proposed fast LMMSE estimator with estimated SNR refers to our proposed algorithm summarized in section 3. The proposed fast LMMSE estimator with designed SNR refers to that we modify the step 3 of the proposed algorithm by using a predetermined and unchanged

80

SNR instead of using formula (3-19) to obtain the estimated SNR. Observe that the BER of the proposed fast LMMSE estimator with designed SNR of 5 dB, 10 dB and 20 dB are almost overlapped with each other within the lower SNR range from 0 dB to 15 dB. However, when SNR increases from 15 dB to 25 dB, the BER of the proposed fast LMMSE estimator with higher designed SNR is better than that of the lower designed SNR. Thus, a design for higher SNR is preferable as for mismatch in SNR.

3.6 Conclusion
In this chapter, a fast LMMSE channel estimation method has been proposed and thoroughly investigated for OFDM systems. Since the conventional LMMSE channel estimation requires the channel statistics, i.e., the channel autocorrelation matrix in frequency domain and SNR, which are often unavailable in practical systems, the application of the conventional LMMSE channel estimation is limited. Our proposed method can efficiently estimate the channel autocorrelation matrix by the improved MST algorithm and calculate the LMMSE matrix by Kumars fast algorithm and exploiting the property of the channel autocorrelation matrix so that the computation complexity can be reduced significantly. We present the MSE analysis for the proposed method and the conventional LMMSE method and investigate the MSE thoroughly under two cases, i.e., the matched SNR and the mismatched SNR. Numerical results and computer simulation show that a design for higher SNR is preferable as for mismatch in SNR.

81

TABLE 3.1. CHANNEL POWER INTENSITY PROFILE Tap 1 2 3 4 5 6

the magnitude of the first row of channel autocorrelation matrix 1.2 1.1 1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2

Delay (us) 0 0.5 1.0 1.5 2.0 2.5

Gain (dB) 0.0 -6.0 -12.0 -18.0 -24.0 -30.0

Doppler Spectrum Clarke [36] Clarke Clarke Clarke Clarke Clarke

20 40 60 80 100 120 the index of the first row of channel autocorrelation matrix

140

Fig 3.3: The first row of the channel autocorrelation matrix R Hp Hp , A .

82

the magnitude of the first row of the LMMSE matrix

0.05 0.045 0.04 0.035 0.03 0.025 0.02 0.015 0.01 0.005 0

SNR = 5 dB SNR = 10 dB SNR = 20 dB

20

40

60

80

100

120

140

the index of the first row of the LMMSE matrix

Fig 3.4: The first row of the LMMSE matrix R Hp Hp R Hp Hp + I with different SNRs. SNR
10
0

10

-1

NMSE

10

-2

10

-3

10

-4

LS, simulation the proposed fast LMMSE, simulation the proposed fast LMMSE, numerical method LMMSE, simulation LMMSE, numerical method 5 10 SNR (dB) 15 20 25

Fig 3.5: Normalized Mean square error (NMSE) of channel estimation of LMMSE algorithm versus that of the proposed fast LMMSE algorithm by computer simulation and numerical method.
83

10

-1

10

-2

NMSE
-3

10

10

-4

LMMSE,matched SNR,numerical method LMMSE,SNR_design = 5 dB,numerical method LMMSE,SNR_design = 10 dB,numerical method LMMSE,SNR_design = 20 dB,numerical method LMMSE,SNR_design = 5 dB,simulation LMMSE,SNR_design = 10 dB,simulation LMMSE,SNR_design = 20 dB,simulation
0 5 10 15 20 25

SNR (dB)

Fig 3.6: NMSE of LMMSE algorithm with matched SNR and mismatched SNRs versus SNR, by simulation and numerical method, respectively.
-1

10

10

-2

NMSE 10
-3

10

-4

the proposed fast the proposed fast the proposed fast the proposed fast the proposed fast the proposed fast the proposed fast 0 5

LMMSE, matched SNR, numerical method LMMSE,SNR_design = 5 dB,numerical method LMMSE,SNR_design = 10 dB,numerical method LMMSE,SNR_design = 20 dB,numerical method LMMSE,SNR_design = 5 dB,simulation LMMSE,SNR_design = 10 dB,simulation LMMSE,SNR_design = 20 dB,simulation 10 SNR (dB) 15 20 25

Fig 3.7: NMSE of the proposed fast LMMSE algorithm with matched SNR and mismatched SNRs versus SNR, by simulation and numerical method, respectively.

84

10

10

-1

LS the proposed fast LMMSE algorithm LMMSE perfect channel estimation

BER

10

-2

10

-3

10

-4

10

15

20

25

SNR (dB)

Fig 3.8: Bit error rate (BER) of the LS, LMMSE, the proposed fast LMMSE and perfect channel estimation versus SNR.
10
0

10

-1

LMMSE,matched SNR LMMSE,SNR_design= 5 dB LMMSE,SNR_design= 10 dB LMMSE,SNR_design= 20 dB

BER

10

-2

10

-3

10

-4

10 SNR (dB)

15

20

25

Fig 3.9: BER comparison between LMMSE channel estimation with matched SNR and LMMSE channel estimation with designed SNRs.

85

10

10

-1

BER

10

-2

10

-3

10

-4

the proposed fast the proposed fast the proposed fast the proposed fast 0 5

LMMSE,estimated SNR LMMSE,SNR_design= 5 dB LMMSE,SNR_design= 10 dB LMMSE,SNR_design= 20 dB 10 SNR (dB) 15 20 25

Fig 3.10: BER comparison between the proposed fast LMMSE channel estimation with estimated SNR and the proposed fast LMMSE channel estimation with designed SNRs.

86

Chapter 4: Channel estimation and data detection for OFDM systems over fast fading channels
The Doppler shift effect of fast fading channels will result in the ICI and degrades the performance of OFDM systems. In this chapter, a novel pilot pattern and corresponding channel estimation and data detection method for OFDM systems with fast fading channels have been proposed and investigated. The proposed pilot pattern is composed of two classical pilot patterns, which are the comb-type pilot pattern and the grouped pilot pattern. The proposed pilot pattern can reduce the required pilot number significantly compared with the grouped pilot pattern. We present the MSE analysis of channel estimation based on the proposed pilot pattern and investigate the effects of various parameters in the proposed pilot pattern on the MSE of channel estimation thoroughly by numerical method. Computer simulation and numerical results show that the proposed channel estimation and data detection based on the proposed pilot pattern can eliminate the ICI effectively while maintaining much lower pilot ratio per OFDM symbol than the grouped and equi-spaced pilot pattern. The chapter is organized as follows. Section 4.1 introduces existing channel estimation methods over fast fading channels. Section 4.2 describes the baseband OFDM system model. Section 4.3 describes the proposed channel estimation and data detection. We analyze the MSE of the proposed channel estimation method in section 4.4. The simulation results and numerical results of the proposed algorithm are discussed in section 4.5
87

followed by conclusion in Section 4.6.

4.1 Introduction
OFDM is sensitive to Doppler shift, time synchronization error and carrier frequency synchronization error, which can introduce the ICI and destroy the orthogonality between subcarriers in one OFDM symbol so that the system performance will degrade in terms of BER, especially at high SNR. Firstly, when the OFDM symbol duration is much smaller than the channel coherence time, the fading channel can be viewed as stationary within one OFDM symbol and there is no ICI effect. Pilots assisted channel estimation methods based on this assumption have been thoroughly investigated [17], [22]-[26], [37]. Ref. [37] has proved that the isolated and equi-spaced pilot pattern, i.e., the comb-type pilot pattern, is optimum for the time invariant channel estimation in terms of the MSE of channel estimates. The channel estimator assisted by comb-type pilot pattern is often performed in the frequency domain, which consists of two steps. The first step is to obtain the channel estimates on all pilot subcarriers by LS method, MMSE method and so on. The second step is to obtain the channel estimates on all subcarriers. Ref. [22] proposed a LMMSE channel estimation method based on channel autocorrelation matrix in frequency domain. However, the complexity of LMMSE estimator is high. To reduce the computational complexity of LMMSE estimator, a low rank approximation to LMMSE estimation has been proposed by singular value decomposition [17]. In [23], a MMSE channel estimator has been derived,

88

which make full use of the channel correlations both in time and frequency domain and can improve the OFDM system performance significantly. Ref. [30] proposed a LS channel estimation technique based on the comb-type pilot arrangement. The proposed LS channel estimator is conducted in the time domain, which is different from the conventional LS channel estimator conducted in the frequency domain [15]. Secondly, when the OFDM symbol duration is larger than or comparable to the channel coherence time, the fading channel can not be viewed as stationary within one OFDM symbol and time variation of the channel will result in ICI. Several channel estimation and data detection methods taking ICI into consideration have been proposed and investigated [29], [38]-[40]. In [39], the authors modeled the fading channel as polynomials and proposed a new MIMO-OFDM technique based on the weighting factor optimization for group transmission to reduce the ICI effect. The proposed scheme has advantages of not requiring pilot signals or tracking channel variation. However, the scheme is based on the assumption that the channel coefficients between adjacent subcarriers are approximately identical, therefore it will not be suitable for the OFDM systems where the subcarrier group gap is larger than the channel coherent bandwidth. In [38], the authors proposed a novel pilot-based channel estimation. They used linear channel interpolation to handle the rapid variation of channel impulse response of each path within one OFDM symbol and reduced the unknown parameters significantly. However, the coefficients to perform channel interpolation require the knowledge of Doppler shift. If the Doppler shift is not estimated accurately, the accuracy of the coefficients will degrade and hence the

89

performance of channel estimation will degrade, too. The authors of [29] proposed a new pilot pattern, i.e., the grouped and equi-spaced pilot pattern and corresponding channel estimation and signal detection to suppress ICI. However, the number of required pilots of one OFDM symbol is too large and the proposed scheme in [29] can only be available in OFDM systems with small maximum delay spread of the channel. In this chapter, we assume that the fading channel varies very fast so that it can not be deemed to be stationary within one OFDM symbol. Therefore, the ICI effect due to high Doppler shift has to be considered. To eliminate the effect of ICI, we propose a new pilot pattern and corresponding channel estimation method and data detection. The proposed pilot pattern consists of two classical pilot patterns, which are the comb-type pilot pattern [15] and the grouped pilot pattern [29]. The proposed channel estimation and data detection based on the proposed pilot pattern can eliminate ICI effect effectively. In comparison with the LS algorithm, the proposed algorithm achieves superior performance in terms of the MSE and BER, especially at high Doppler shift. In comparison with the algorithm of [29], the proposed algorithm can achieve almost the same performance with it in terms of BER while reducing the number of pilots significantly. In addition, we present the MSE analysis for the MST algorithm based on comb-type pilot pattern when the ICI effect is considered. Although Ref. [29] proposed the grouped and equi-spaced pilot pattern and corresponding channel estimation, it did not present the MSE analysis of channel estimation. To compensate this gap, we present the MSE analysis for the channel estimation based on the grouped pilot pattern.

90

4.2 System Model

We adopt the OFDM system model in [44] without loss of generality. For N carriers in the OFDM system, the transmitted signal x(i, n) in time domain after IFFT is given by
x(i, n) = IFFTN [ X (i, k )] = 1 N

X (i, k ) exp{ j 2 nk / N }
k =0

N 1

(4-1)

where X(i, k) denotes the transmitted signal in frequency domain at the k-th subcarrier in the i-th OFDM symbol. The system transmits data each block by each block. One data block consists of M OFDM symbols. The first M 1 OFDM symbols in one block adopt the comb-type pilot pattern [15]. The last M2 OFDM symbols adopt the grouped pilot pattern as shown in Fig 4.1. The value of X(i, k) (including the pilots) is set according to the modulation constellation. We assume that the data X(i, k) is normalized, i.e., |X(i, k)|2 = 1. For instance, X(i, k) = 1 and X (i, k ) = (1 1) / 2 for BPSK and QPSK, respectively. It is assumed that E[ X (i, k )] = 0, where X(i, k) is the user data (not pilot) and E (i) denotes expectation. A cyclic prefix (CP) is inserted into each OFDM symbol prior to the transmission and the CP is removed before the FFT process at the receiver. The received signal Y(i, k) in frequency domain after FFT can be written as [29]
Y (i , k ) = X (i , k ) g (i , k , k ) +

j=0, j k

N 1

g ( i , k , j ) X ( i , j ) +W ( i , k )

(4-2)

where W (i, k ) denotes the additive white Gaussian noise (AWGN) with zero mean and
2 variance W ,

2 2 j n( pq) j ql 1 N 1 L1 N g(i, p, q) = h(i, n,l )e e N N n=0 l =0

(4-3)

91

where h(i, n, l) is the channel impulse response in time domain of the l-th path at the n-th sample point within the i-th symbol time, l is the channel delay of the l-th path, L is the number of resolvable paths. It is assumed that different paths h(i, n, l) are independent and the power of the l-th path is l2 . The channel is normalized so that h2 = l2 = 1 . The
l =0 L 1

channel autocorrelation function is given by

* E h(i1 , n1 , l1 ) h (i2 , n2 , l2 )

= l2 ( l1 l2 ) J 0 {2 f d Tsym [(i1 i2 ) + (n1 n2 ) / N sym ]} 1

(4-4)

where (i) is the Kronecker delta function, J 0 (i) is the zero order Bessel function of the first kind, fd is the maximum Doppler shift, Tsym is the during time of an OFDM symbol including the CP, Nsym is the length of an OFDM symbol including the CP in unit of sample point, (i)* denotes conjugate. In addition, the item ICI.
j = 0, j k

N 1

g (i, k , j ) X (i, j ) in

(4-3) is the

4.3 The Proposed Channel Estimation and Data Detection

4.3.1. The proposed pilot pattern

As shown in Fig 4.1, the OFDM system transmits signal each block by each block. One data block consists of M (M = M1+M2) OFDM symbols. The black circle denotes pilot subcarrier and the gray circle denotes user data subcarrier. The first M 1 OFDM symbols in one block adopt the comb-type pilot pattern [15]. Namely, the pilot subcarriers are equi-spaced inserted into each OFDM symbol. It is assumed that the number of the total pilot subcarriers in each OFDM symbol is N P1 and the inserting gap is R . Therefore, the set of the indexes of pilot subcarriers,

can

be

written

92

as = {k | k = mR + k0 , m = 0,1,..., N P1 1} ,where k0 [0, R) and k0 is the index of the first pilot subcarrier. The last M2 OFDM symbols in one block adopt the grouped pilot pattern [29]. It is assumed that the number of pilot groups in each OFDM symbol is N group and one group consists of 2c+1 pilot subcarriers. And it is assumed that the set of index of the center pilot in each group is = {P(0), P(1),..., P( N group 1)} . The benefits of the proposed pilot pattern are explained as follows. Firstly, if we only use the comb-type pilot pattern, the ICI effect due to high Doppler shift can not be compensated or eliminated. The comb-type pilot pattern is optimal for channel estimation in MSE sense. However, the prerequisite is that the channel fading can be viewed as stationary between two adjacent OFDM symbols. If the prerequisite does not hold, the comb-type pilot pattern is not optimal for channel estimation. Secondly, if we only use the grouped pilot pattern, the pilot occupancy ratio is too large although the pilot pattern can eliminate ICI effect significantly. Therefore, we combine the two classical pilot patterns. The MST algorithm is adopted to find the channel multi-paths and then the channel delay information is used to reduce the required pilot number for the grouped pilot pattern. So there are two benefits of the combined pilot pattern. For one thing, it can reduce the required pilot number significantly compared with the grouped pilot pattern. For another, it can also eliminate the ICI effect significantly by setting appropriate parameters for the length of the two blocks, i.e., M1 and M2.

93

4.3.2. Channel Estimation and data detection for the first M1 OFDM symbols of each block

Since the first M1 OFDM symbols adopt the comb-type pilot pattern, we can obtain the LS channel estimate at pilot subcarriers in time domain for the i-th OFDM symbol, hp.ls (i, n) and it is given by 1 hp.ls (i, n) = N P1
N P 1 1 m=0

p ,ls

(i, mR + k0 )e j 2 n ( mR + k0 ) / N , n = 0,1,..., N P1 1, i = 0,1,..., M 1 1. (4-5)

where H p ,ls (i, k ) =

Y (i, k ) , k , = {k | k = mR + k0 , m = 0,1,..., N P1 1} , Y (i, k ) is the i-th X (i, k )

received OFDM symbol after fast FFT operation, X (i, k ) is the i-th transmitted OFDM symbol. Next, the MST algorithm [32] has been proposed to get the refined channel estimate in time domain. The MST algorithm deals with each OFDM symbol by reserving the most significant L paths in terms of power and setting the other taps to be zero. The algorithm can reduce the influence of AWGN and other interference significantly, compared with the LS method. However, the algorithm may choose the wrong paths and omit the right paths because of the AWGN and other interference. Thus, we will improve the algorithm of [32] by processing several adjacent OFDM symbols jointly. Firstly, we calculate the average power of each tap for the M1 adjacent OFDM symbols, PLS (n) and it is given by 1 M1 1 | hp ,ls (i, n) |2 , n = 0,1,..., N P1 1. PLS (n) = M 1 i =0 (4-6)

Next, we choose the L most significant taps from PLS (n) and reserve the indexes of them
94

into a set , i.e., = { l : l = 0,1,..., L 1} , where l is the delay of the l-th path in unit of sample point. Finally, the refined channel estimate in time domain, hp , MST ,is given by
h (i, n), if n hp , MST (i, n) = p ,ls , n = 0,1,..., N P1 1, i = 0,1,..., M 1 1. 0, if k

(4-7)

Then, we can obtain the refined channel estimate in frequency domain, H MST , and it is given by
H MST (i, k ) =
N P 1 1

n=0

hp , MST (i, n)e

2 nk N

, k = 0,1,..., N 1 , i = 0,1,..., M 1 1.

(4-8)

So the detected data X (i, k ) is given by X (i, k ) = Y (i, k ) , i = 0,1,..., M 1 1, k = 0,1,..., N 1. H MST (i, k ) (4-9)

4.3.3. Channel estimation and data detection for the last M2 OFDM symbols of each block

We use the polynomial of Q order to model the channel [29] shown as follows.
h(i, n, l ) = ai ,l , q n q + e(i, n, l )
q =0 Q

(4-10)

= h (i, n, l ) + e(i, n, l )

where ai ,l ,q is the polynomial coefficient and e(i, n, l ) is the approximation error,

h (i, n, l ) = ai ,l ,q n q . It is noted that we use Taylor series of h(i, n, l ) around 0. Therefore
q =0 Q

the approximation error can be expressed by

e(i, n, l ) = nQ +1 d (Q +1) h(i, n0 , l ) (Q + 1)! dnQ +1

(4-11)

95

where n0 [0, n] . Since the ICI terms which do not significantly affect Y(i, k) can be discarded [41], i.e., g(i, k, j) = 0 if| j k |> c , where 2c is the number of dominant ICI terms. We can rewrite the received signal Y(i, k) as follows.
Y (i , k ) = X (i , k ) g (i , k , k ) + = =

j=0, j k

N 1

g ( i , k , j ) X ( i , j ) +W ( i , k )

j=k c

k +c

g ( i , k , j ) X ( i , j ) +W ( i , k ) ) i , n , l , k +

(4-12)
i , n ,l , k

h (i , n ,
n=0 l=0

N 1 L 1

e ( i , n , l )
n=0 l=0

N 1 L 1

+W ( i , k )

a
n=0 l=0 q=0

N 1 L 1

i ,l , q

n q i , n ,l , k + W ( i , k )

where

i ,n,l ,k
N 1 L 1 n=0 l =0

1 = N

q = k c

k +c

j 2 n ( k q ) / N

e j 2 q l / N X (i, q)

(4-13)

and W (i, k ) = e(i, n, l )i ,n ,l ,k +W (i, k ) . Equation (4-12) can be written as the vector form, shown as follows. Y (i, k ) = i ,k K i ai + W (i, k ), i = M 1 , M 1 + 1,..., M 1 + M 2 1. where
i ,k = i ,0, k
T Ki = K i ,0

(4-14)

i ,1,k
KT i ,1

i , N 1,k i ,n ,0,k 1 LN , i ,n ,k =
KT i , N 1
T NL( Q +1) L

i ,n,1,k
0 0
T ( Q +1) L1

i ,n , L 1,k 1 L ,

, K i ,n

n = 0 0

0 0 , n L(Q +1) L

n = 1 n
a i ,l = ai ,l ,0

nQ
ai ,l ,1

1( Q +1)

T , ai = ai ,0 T ( Q +1)1

aT i ,1

aT i , L 1

ai ,l ,Q

, (i)T denotes transpose operation. Since the number of

the unknown parameters ai is (Q+1)L, the number of pilot groups, Ngroup, should satisfy Ngroup (Q+1)L, where L is the number of resolvable paths. When the index k in equation

96

(4-14) is chosen from the index set , where = {P(0), P(1),..., P( N group 1)} , we can stack the signal Y(i, k) (k = P(0), P(4-1),,P(Ngroup-1)) together and obtain

Y(i ) = i K i ai + W(i ), i = M 1 , M 1 + 1,..., M 1 + M 2 1.

where

(4-15)

Y(i ) = Y (i, P(0)) Y (i, P (1))

T i = i , P (0)

Y (i, P( N group 1))

T N group LN

T N group 1

T i , P (1)

T i , P ( N group 1)

,
T N group 1

W(i ) = W (i, P(0)) W (i, P(1))

W (i, P ( N group 1))

i can be given by Therefore, the estimated coefficients a i = (i K i ) + Y(i ) a (4-16)

where (i) + denotes moore-penrose inverse operation. Then, we can obtain the estimated channel impulse response h (i, n, l ) from (4-10) and the estimated g (i, p, q) from (4-3). Next, we perform data detection. Similar to the detection method in [29], we perform the following steps. Step 1: Initialization of the detected data X (i, k ) , i = M 1 , M 1 + 1,..., M 1 + M 2 1 . Let X [0] (i, k ) = Y (i, k ) / g (i, k , k ), k = 0,1,..., N 1 , and k ,where is the set of indexes of the pilots in one OFDM symbol and = {P(0)-c, P(0)-c+1,, P(0)+c, , P(Ngroup)-c, P(Ngroup)-c+1, P(Ngroup)+c}. Step 2: At the j-th iteration (j = 1,2,), we update X [ j ] (i, k ) by computing Y (i, k ) X [ j +1] (i, k ) =

q =k c ,q k

k +c

g (i, k , q) X [ j ] (i, q ) (4-17)

g (i, k , k )

97

Step 3: When

k = 0, k

| X

N 1

[ j +1]

(i, k ) X [ j ] (i, k ) | < , where is a tolerance threshold,

[ j +1]

k = 0, k

N 1

|X

(i, k ) |

end the detection. Otherwise, go back to step 2 and increase the iteration number j = j+1.

4.3.4. Summary of the proposed channel estimation and data detection

This section summarizes the proposed channel estimation and data detection. Firstly, the OFDM system transmits signal each block by each block. One data block consists of M OFDM symbols. The first M1 OFDM symbols in one block adopt the comb-type pilot pattern and the last M2 OFDM symbols in one block adopt the grouped pilot pattern. The proposed pilot pattern for channel estimation and data detection is shown as Fig 4.1. Secondly, the whole processing of channel estimation and data detection for one OFDM block can be summarized as follows. Step 1: For the first M1 OFDM symbols in one block, we can obtain the LS channel estimate at pilot subcarriers in time domain for the i-th OFDM symbol, hp.ls (i, n) , i = 0,1,, M1-1, from formula (4-5). Step 2: Calculate the average power of each tap for the first M1 adjacent OFDM symbols, PLS (n) , from formula (4-6). Step 3: Choose the L most significant taps from PLS (n) and reserve the indexes of them into a set , i.e., = { l : l = 0,1,..., L 1} . Then we can obtain the

98

refined channel estimate in time domain, hp , MST , from formula (4-7). Step 4: Obtain the refined channel estimate in frequency domain, H MST , from formula (4-8) and perform data detection using formula (4-9). Step 5: For the last M2 OFDM symbols in one block, we can obtain the estimated i , i= M1, M1+1,,M-1,by calculating formula (4-16). It is noted coefficients a that the required channel delay of each path l is chosen from the set in the i . calculation of the estimated coefficients a Step 6: Obtain the estimated channel impulse response h (i, n, l ) from (4-10) and the estimated g (i, p, q ) from (4-3). Step 7: Perform data detection for the last M2 OFDM symbols by using the three steps proposed in section 4.3.3.

4.4. Analysis of MSE of the proposed channel estimation method

In this section, we will present the MSE of the proposed channel estimation method. Firstly, we present the MSE of channel estimation for the first M1 OFDM symbols based on comb-type pilot pattern in one block; secondly, we present the MSE of channel estimation for the last M2 OFDM symbols based on grouped distributed pilot pattern in one block; thirdly, we present the MSE of channel estimation for one OFDM block by combining the analysis results of the two previous parts. Then, we clarify the channel impulse response representation to avoid confusing. In the

99

first M1 OFDM symbols, we did not consider the ICI effect and use the LS estimation. And the channel is viewed as static in an OFDM symbol. Thus, we use h(i, n) to stand for the channel impulse response, where i stands for the index of OFDM symbol, n stands for the index of time axis in unit of sample point, for each symbol. And hp , MST (i, n) is the MST estimation of h(i, n) . But, for the last M2 OFDM symbols, ICI is considered and the channel can not viewed as static within an OFDM symbol. Thus, we use h(i, n, l) to stand for the channel impulse response, where i still stands for the index of OFDM symbol, n is the index of sample point, l stands for the l-path channel delay at time n.

4.4.1. MSE analysis of channel estimation for the first M1 OFDM symbols
Denote the MSE of channel estimation for the first M1 OFDM symbols by MSE , M1 . It is assumed that the joint MST algorithm formulated by (4-6) has chosen the L paths correctly, i.e., the set of indexes of the chosen paths = { l : l = 0,1,..., L 1} and each path delay l is correct. Then we have

MSE , M =
1

N 1 1 N 1 p1 2 | ( , , ) ( , ) | E h i n m h i m p MST , N n =0 m =0

(4-18) that the and received the item

To signal

further

derive

formula

(4-18),

we

recall

Y (i , k ) = X (i , k ) g (i , k , k ) +

j=0, j k

N 1

g ( i , k , j ) X ( i , j ) +W ( i , k )

j = 0, j k

N 1

g (i, k , j ) X (i, j ) is

the ICI item. Since the ICI consists of a large number of interferences,

2 , can be derived as we can model the item as an AWGN noise W1. The variance of W1, W 1

100

2 W = E |
1

g (i, k , j ) X (i, j ) |2 j =0, j k N 1 N 1 = E g (i, k , j1 ) X (i, j1 ) g * (i, k , j2 ) X * (i, j2 ) j2 = 0, j2 k j1 =0, j1 k

N 1

(4-19)

It is assumed that different subcarriers are independent, i.e., E [ X (i, k1 ) X (i, k2 ) ] = 0 , where k1 k2 . Therefore, formula (4-19) can be further derived as
2 W =
1

j = 0, j k N 1

E ( g (i, k , j ) g (i, k , j ) ) E ( X (i, j ) X

* *

N 1

(i, j ) )

= =

j = 0, j k N 1

E ( g (i, k , j ) g (i, k , j ) ) h(i, n, l ) e

n =0 l =0
2 l

1 E j = 0, j k N
N 1

N 1 L 1

2 n ( k j) N

2 j l N

1 N

h* (i, n, l ) e
n =0 l =0

N 1 L 1

2 n ( k j) N

2 j l N

( n1 n2 )( k j ) (n1 n2 ) j 2N J 2 f T e = 0 d sym N j = 0, j k n1 = 0 n2 = 0 l = 0 sym 1 N 1 N 1 N 1 (n n ) j 2 ( n1 n2 )( k j) = 2 J 0 2 f d Tsym 1 2 e N N j=0, j k n1 =0 n2 =0 N sym 1 N 1 N 1 2 f d Tsym (n1 n2 ) = 1 2 J0 N n1 =0 n2 =0 N sym

1 N2

N 1 N 1 L 1

(4-20)

Therefore, the total noise power consisting of the AWGN noise W and the equivalent
2 , is given by AWGN noise W1 resulting from ICI, W 2 2 2 W = W + W

(4-21)

Next, we will further derive formula (4-5). Recall that the LS channel estimate at pilot subcarriers in time domain for the i-th OFDM symbol, hp.ls (i, n) and it is given by

101

1 hp.ls (i, n) = N P1 = = 1 N P1 1 N P1
1 N P1

N P 1 1 m=0

p ,ls

(i, mR + k0 )e j 2 n ( mR + k0 ) / N

N P1 1 m=0 N P1 1 m=0

Y (i, mR + k0 ) j 2 n ( mR + k0 ) / N e X (i, mR + k0 )
0 0

g (i, mR + k , mR + k ) +

0 0

W (i, mR + k0 ) j 2 n ( mR + k0 ) / N e X (i, mR + k0 )

N P 1 1 m=0

g (i, mR + k , mR + k ) +

W (i, mR + k0 ) j 2 n ( mR + k0 ) / N e X (i, mR + k0 )
N P 1 1 m=0

1 = N P1

N P 1 1 L 1 m=0

2 j ( l n )( mR + k0 ) 1 h(i, l )e N + N P1 l =0

W (i, mR + k0 ) j 2 n ( mR + k0 ) / N e (4-22) X (i, mR + k0 )

(i, ) + w (i, n), if n h l = w (i, n), if n (i, ) = 1 where h l N 1 h(i, m, l ) , w (i, n) = N P1 m=0
N 1 N P1 1 m=0

W (i, mR + k0 ) j 2 n ( mR + k0 ) / N , = {l: l = e X (i, mR + k0 ) 1 2 W . Since we have N P1

0, 1,, L-1}. It is easy to verify that the variance of w (i, n) is

assumed that the MST algorithm has chosen the correct L paths, the formula (4-18) can be further derived as

MSE , M

N 1 1 N 1 p1 | h(i, n, m) hp , MST (i, m) |2 = E N n=0 m=0 N 1 L 1 1 (i, ) |2 + L 2 = E | h(i, n, l ) h l W N n=0 l =0 N P1

2 = 2 N = 2 N2

2 f d Tsym (n n) L 2 1 J 0 W + N N n = 0 n = 0 l = 0 sym P1
N 1 N 1 L 1 2 l

(4-23)

1 J
n = 0 n = 0

N 1 N 1

2 f d Tsym (n n) L 2 W + N N sym P1

102

4.4.2. MSE analysis of channel estimation for the last M2 OFDM symbols
Denote the MSE of channel estimation for the last M2 OFDM symbols by MSE , M 2 . Similar to section A, it is assumed that the joint MST algorithm formulated by (4-6) has chosen the L paths correctly, i.e., the set of indexes of the chosen

paths = { l : l = 0,1,..., L 1} and each path delay l is correct. Then we have

MSE , M =
2

1 N 1 L 1 E | h(i, n, l ) h (i, n, l ) |2 N n =0 l =0
N 1 L 1 n =0 l =0

(4-24)

To further deriver formula (4-24), we model the item e(i, n, l )i , n ,l , k in formula (4-12) as an AWGN noise W2, since the item is composed of a large number of interferences. The
2 , can be derived as variance of W2, W 2

2 W2

N 1 L 1 = E | e(i, n, l )i ,n ,l ,k |2 n=0 l =0 N 1 L 1 N 1 L 1 = E e(i, n1 , l1 )i , n1 ,l1 ,k e* (i, n2 , l2 ) *i , n2 ,l2 ,k n2 = 0 l2 = 0 n1 =0 l1 =0 = E ( e(i, n1 , l )e (i, n2 , l ) ) i ,n1 ,l ,k

* n1 = 0 n2 = 0 l = 0 N 1 N 1 L 1 * i , n2 ,l , k

(4-25)

d ( Q +1) h(i, n01 , l ) d (Q +1) h* (i, n02 , l ) * E i ,n1 ,l ,k i , n2 ,l ,k 2 Q +1 Q +1 dn dn n1 = 0 n2 = 0 l = 0 [ (Q + 1)!]

N 1 N 1 L 1 Q +1 n1Q +1n2

where ()* denotes conjugate, n01 [0, n1 ] , n02 [0, n2 ] . As in [45], for a wide sense stationary random process x(t), we have
(2 n ) Rx( n ) ( ) = (1) n Rx ( )

(4-26)

where x ( n ) is the n-th order derivative of x(t) with respect to t , Rx ( ) is the autocorrelation
(2 n ) function of x(t), Rx ( ) is the 2n-th order derivative of the function Rx ( ) with respect to .

To further derive formula (4-25), we assume that and n01 is uniformly distributed in the

103

range [0, n1 ] and n02 is

also

uniformly

distributed

in

the

range [0, n2 ] .

Namely, P(n01 = n) = 1/(n1 + 1) , n [0, n1 ] and

P (n02 = n) = 1/(n2 + 1) , n [0, n2 ] ,

where P() is the probability function. Then, the formula (4-25) can be further derived as

2 W2

Q +1 i ,n1 ,l ,k i ,n2 ,l ,k l (1)Q +1 n1Q +1n2 = 2 (n1 + 1)(n2 + 1) n1 = 0 n2 = 0 l = 0 n01 = 0 n02 = 0 [ (Q + 1)!] N 1 N 1 L 1 n1 n2 * 2

2( Q +1)

(2( Q +1)) J0 ( (n01 n02 ) ) (4-27)

where =

2 f d Tsym N sym

(2( Q +1)) , J0 () is the 2(Q+1) order derivative of the zero order Bessel

function of the first kind, i ,n ,l , k is defined by equation (4-13). It is noted that the
2 item i ,n1 ,l ,k *i ,n2 ,l ,k is not dependent on the variable k so that W is not dependent on k, either. 2

So the total noise power consisting of the AWGN noise W and the equivalent AWGN noise
2 W2 , W , is given by 2 2 2 W = W + W

(4-28)

Next, we will further derive formula (4-16). Recall that the estimated polynomial i can be given by coefficients a i = (i K i ) + Y(i ) a = ai + (i K i ) + W(i ) = ai + Bi W(i )
T where Bi = (i K i ) + . The matrix B can be written as Bi = Bi ,0

(4-29)
T

BT i ,1

BT i , L 1 , where

each item Bi ,l (l=0,1,,L-1) is a (Q + 1) N group matrix. The true coefficients ai can be

T written as ai = ai ,0

aT i ,1

aT i , L 1

and ai ,l = ai ,l ,0 ( Q +1) L1
i = T a i ,0 a

ai ,l ,1
T a i ,1

ai ,l ,Q

T ( Q +1)1 T

. The and

i can be written as estimated coefficients a

i ,l = i ,l ,0 a a i ,l ,1 a i , l , Q a
T ( Q +1)1

T a i , L 1

( Q +1) L1

. Therefore, the estimated polynomial coefficients for

i ,l , can be given by the l-th path, a

104

 i ,l = ai ,l + Bi ,l W(i ) a The estimated channel impulse response h (i, n, l ) is given by

i ,l h (i, n, l ) = n a = nai ,l + n Bi ,l W(i ) = h (i, n, l ) + n Bi ,l W(i )

(4-30)

(4-31)

where

n = 1 n

nQ

1( Q +1)

and

h (i, n, l ) = ai ,l ,q n q = n ai ,l
q =0

Note

that n and h (i, n, l ) are consistent with the definitions in formula (4-14) and (4-10), respectively. Then, the formula (4-24) can be further derived as

MSE , M =
2

1 N 1 L 1 E | h(i, n, l ) h (i, n, l ) |2 N n =0 l =0 N 1 L 1 1 = E | h (i, n, l ) + e(i, n, l ) h (i, n, l ) n Bi ,l W(i ) |2 N n=0 l =0

= =

1 N 1 N

{E (| e(i, n, l ) | ) + E (| B
2 n=0 l =0 n N 1 L 1

N 1 L 1

i ,l

W(i ) |2 )

(4-32)

Q +1 2( Q +1) 2 2( Q +1) l (1) n (2( Q +1)) 2 2 + J 0 || B || ( ) 0 n i ,l W 2 n=0 l =0 [(Q + 1)!] 2 N 1 L 1 Q +1 2( Q +1) N 1 (1) (2( Q +1)) J0 = ( 0 ) n2(Q +1) + W || n Bi ,l ||2 2 N n =0 l =0 N [ (Q + 1)!] n=0

where =

2 f d Tsym N sym

(2( Q +1)) , J0 () is the 2(Q+1) order derivative of the zero order Bessel

function of the first kind, J 0 () . || x || is the Euclidean norm of a vector x and it is defined by
|| x ||=

| x |
i =0 i

n 1

, where x = [ x0

x1

xn 1 ]T .

4.4.3 MSE analysis of channel estimation for one OFDM block

Since we have performed the MSE analysis of channel estimation for the comb-type pilot pattern and the grouped pilot pattern, we can combine their results to obtain the MSE
105

of channel estimation for one OFDM block. Denote the MSE of channel estimation for one OFDM block by MSE , M and it is given by

MSE , M =

M1 M MSE , M1 + 2 MSE , M 2 M M

(4-33)

where MSE , M1 and MSE , M 2 are given by formula (4-23) and (4-32), respectively. It is noted that the MSE of channel estimation is the NMSE of channel estimation since we have assumed that the channel is normalized. In addition, for the sake of performance comparison, we define the NMSE of channel estimation obtained by computer simulation as

NMSE =

| h(i, n, ) h(i, n, ) |
i =0 n =0 l =0 K 1 N 1 L 1 i =0 n =0 l =0 l l

K 1 N 1 L 1

| h(i, n, ) |
l

(4-34)

where h (i, n, l ) is the estimated channel impulse response, h(i, n, l ) is the true channel impulse response, K is the number of OFDM symbols in simulation.

4.5 Numerical and Simulation Results

Both computer simulation and numerical method have been deployed to investigate the performance of the proposed channel estimation and data detection. For computer simulation, we employ the channel model of COST207 [35] having 6 numbers of paths and the maximum delay spread of 5 s. The channel power intensity profile is listed in Table 4.1. The number of the subcarriers of the OFDM system, N, is equal to 2048 and the CP length NCP is equal to 128 sample points. So the length of an OFDM symbol including

106

the CP in unit of sample point, Nsym , is equal to 2176. The bandwidth of the system is 20 MHz so that the symbol duration Tsym = 108.8 s, the CP duration TCP = 6.4 s >5 s and the symbol duration excluding the CP, T, is 102.4 s. One OFDM block contains 25 OFDM symbols, in which the first 5 OFDM symbols adopt comb-type pilot pattern and the last 20 OFDM symbols adopt grouped pilot pattern. Namely, M = 25, M1 = 5 and M2 = 20. The transmitted signal is QPSK modulated.
a. The normalized mean square error (NMSE) of channel estimation for the first M1 OFDM symbols based on comb-type pilot pattern

Fig 4.2 shows the NMSE of channel estimation for the first M1 OFDM symbols based on comb-type pilot pattern, for fdT = 0.01, 0.06 and 0.1, respectively. The numerical results are obtained by formula (4-23). The number of the total pilot subcarriers in each OFDM symbol, Np1 = 128 and the number of resolvable paths of the channel, L = 6. Observe that the NMSE increases when the normalized Doppler shift, fdT, is increased. High normalized Doppler shift will result in NMSE floors due to the ICI effect. For instance, a NMSE floor of about 1.210-2 occurs when the normalized Doppler shift is equal to 0.06.

b. The normalized mean square error (NMSE) of channel estimation for the last M2 OFDM symbols based on grouped pilot pattern

In this subsection, we will investigate the NMSE of channel estimation for the last M2 OFDM symbols based on grouped pilot pattern thoroughly. We will show the effects of some important parameters such as the polynomial order Q, the number of the dominant

107

ICI items 2c and the number of pilot groups Ngroup. The following results are obtained by numerical method, according to the formula (4-32). In addition, the pilots are equi-spaced and grouped inserted into each OFDM symbol.
b. 1 The effect of the polynomial order Q

Fig 4.3 shows the effect of the polynomial order Q =1, 2, 3 and the normalized Doppler shift fdT = 0.01 and 0.1, respectively. The number of dominant ICI items 2c is set to be 2, i.e., c =1. The number of pilots Np2 = (2c+1)(Q+1)L, where L is the number of resolvable paths of the channel and L = 6. Observe that the NMSE of channel estimation is reduced when the polynomial order Q is increased at a fixed normalized Doppler shift value. For instance, the curve with Q = 2 and fdT = 0.01 has about 2.2 dB gain over the curve with Q = 1 and fdT = 0.01 within the whole SNR range from 0 dB to 25 dB. Observe that high fdT and low Q will result in higher NMSE compared with the case with low fdT and high Q, especially at high SNR. For instance, the NMSE of the curve with Q = 1 and fdT = 0.1 is 0.0116 at SNR = 20 dB and the NMSE of the curve with Q = 1 and fdT = 0.01 is about 0.067 at SNR = 20 dB. The above numerical results suggest that 1st order polynomial approximation is enough for fdT = 0.01 and increasing polynomial order will reduce the NMSE of channel estimation due to more pilots for fdT = 0.01. When fdT is large and ICI effect is significant, 1st order polynomial is not good enough and 2nd order polynomial approximation is a good choice.

b. 2 The effect of the number of the dominant ICI items 2c

108

Fig 4.4 shows the effect of the number of dominant ICI items under different Doppler shift values. The polynomial order Q is set to be 2. Observe that the three curves with c = 1, 2 and 3 almost overlap with each other when fdT = 0.01. Thus, we can conclude that when the normalized Doppler shift fdT is low and the ICI effect is not significant, increasing the number of dominant ICI items, 2c will not reduce the NMSE of channel estimation significantly. Observe that when fdT is 0.1, the NMSE of the curve with large c is a little lower than the NMSE of the curve with small c, especially at high SNR. For instance, the NMSE of the curve with c = 3 is 0.0011 at SNR = 25 dB and the NMSE of the curve with
c = 1 is 1.810-3 at SNR = 25 dB. The above numerical results suggest that a good choice

on the number of dominant ICI items is 2c = 2, i.e., c1 for both fdT = 0.01 and fdT = 0.1.

b. 3 The effect of the number of pilot groups Ngroup

Fig 4.5 shows the effect of the number of pilot groups Ngroup at the normalized Doppler shift fdT = 0.01 and 0.1, respectively. The polynomial order Q is equal to 2 and c = 1. Observe that the NMSE reduces with the increase of the number of pilot groups at a fixed normalized Doppler shift value. For instance, the curve with Ngroup = 36 and fdT = 0.01 has about 3.6 dB gain over the curve with Ngroup = 18 and fdT = 0.01 within the whole SNR range from 0 dB to 25 dB. Observe that the NMSE of the curve with fdT = 0.01 is almost the same with the NMSE of the curve with fdT = 0.1 within a low SNR range, at a fixed group number Ngroup. And the NMSE of the curve with fdT = 0.1 is larger than that of the curve with fdT = 0.01 within a high SNR range. For instance, the curve with fdT = 0.01 and

109

Ngroup = 18 almost overlaps with the curve with fdT = 0.1 and Ngroup = 18 when SNR ranges

from 0 dB to 15 dB. However, the curve with fdT = 0.01 and Ngroup = 18 is better than the curve with fdT = 0.1 and Ngroup = 18 over the SNR range from 15 dB to 25 dB.

c. NMSE comparison between the proposed algorithm and LS algorithm

Fig 4.6 shows the NMSE of channel estimation for the proposed algorithm and LS algorithm by numerical method and simulation at fdT = 0.01 and 0.06, respectively. In the computer simulation, the number of the most significant taps L is set to be 8. The polynomial order Q = 1, c =1, and Ngroup = 16, for both computer simulation and numerical method. Observe that the computer simulation is well verified by the numerical result. And the NMSE of channel estimation for the proposed algorithm is smaller than that of LS channel estimation at the same normalized Doppler shift. For instance, the curve of the proposed algorithm has about 3.2 dB gain over the curve of LS method over the SNR range from 0 dB to 25 dB when fdT = 0.01. Thus, the proposed algorithm can suppress ICI effect significantly in comparison with LS algorithm. Observe that increased normalized Doppler shift results in increased NMSE for both the proposed algorithm and LS algorithm due to ICI effect, especially at high SNR range. In addition, the NMSE of the proposed algorithm is less influenced by ICI effect, compared with the LS algorithm.

d. The bit error ratio (BER) comparison of LS, the proposed algorithm and the algorithm in [29]

110

Fig 4.7 shows the BER of LS, the proposed algorithm and the algorithm in [29], for normalized Doppler shift fdT = 0.01 and 0.06, respectively. In the computer simulation, the number of the most significant taps L is set to be 8. The polynomial order Q = 1, c =1, and
Ngroup = 16. Observe that the BER of LS, the proposed algorithm and the algorithm in [29]

are almost the same over the whole SNR range from 0 dB to 25 dB when the normalized Doppler shift fdT is 0.01. Observe that the proposed algorithm can effectively eliminate the ICI effect in comparison with the LS method when the normalized Doppler shift fdT is 0.06. In addition, our proposed algorithm achieves almost the same performance with the algorithm in [29] over the whole SNR range from 0 dB to 25 dB when fdT = 0.06. However, the proposed pilot pattern for the proposed algorithm uses fewer pilots than the equi-spaced and grouped pilot pattern proposed in [29].

e. Pilot ratio comparison of LS, the proposed algorithm and the algorithm in [29]

Table 4.2 shows the pilot ratio of LS, the proposed algorithm and the algorithm in [29], respectively. For the LS algorithm, the corresponding pilot pattern is the comb-type. According to the Karhunen-Loeve (KL) theorem [43], the number of pilots for each OFDM symbol is not less than the CP length in unit of sample points. For the proposed algorithm, we adopt the proposed pilot pattern shown in Fig 4.1. For the algorithm proposed in [29], the corresponding pilot pattern is equi-spaced and grouped. In addition, we further give an example to compare the pilot ratio of the three methods and the parameters are given by our computer simulation. It can be found that the pilot ratio of our

111

proposed algorithm is the least among the three algorithms. For the equi-spaced and grouped pilot pattern in [29], the pilot ratio is 37.5%, which is intolerable in practical system. Thus, we can conclude that our proposed algorithm has two advantages. One is that it can eliminate the ICI effect effectively in comparison with LS method; the second is that it can achieve almost the same performance with the algorithm in [29] in terms of BER while maintaining comparatively lower pilot ratio.

4.6. Conclusion
In this chapter, a novel pilot pattern and corresponding channel estimation and data detection method of compensating the effect of ICI due to fast fading channels for OFDM systems have been proposed and investigated. The proposed pilot pattern consists of two classical pilot patterns, which are the comb-type pilot pattern and the grouped pilot pattern. Accordingly, each OFDM data block consists of two parts of OFDM symbols. The first part of OFDM symbols adopts the comb-type pilot pattern and the second part of OFDM symbols adopts the grouped pilot pattern. We use the joint MST channel estimation for the comb-type pilot pattern and similar channel estimation of [29] for the grouped pilot pattern. The MSE of our proposed channel estimation has been derived and thoroughly investigated. Computer simulation and numerical results show that the proposed channel estimation and data detection can eliminate the ICI effect effectively in comparison with LS method and achieve almost the same performance with the algorithm in [29] in terms of BER while maintaining comparatively lower pilot ratio.

112

Table 4.1. Channel power intensity profile Tap 1 2 3 4 5 6 Delay (us) 0 1.0 2.0 3.0 4.0 5.0 Gain (dB) 0.0 -6.0 -12.0 -18.0 -24.0 -30.0 Doppler Spectrum Clarke [36] Clarke Clarke Clarke Clarke Clarke

Table 4.2. Pilot ratio of LS, the proposed algorithm and the algorithm in [29] algorithm LS The proposed algorithm Pilot ratio per OFDM symbol N CP / N
M 1 N CP M 2 (Q + 1)(2c + 1) L + M N M N

Example (our simulation)1 6.25% 3.13%

The algorithm in [29]

(Q + 1)(2c + 1) N CP / N

37.5%

Note that the parameters used in the simulation are NCP = 128, N = 2048, M = 25, M1 = 5, M2 =20, L = 6, Q = 1, c = 1.
113

Frequency

...

... ... ... ... ... ... ... ... ... ... ... ...
M2 one OFDM block

Fig 4.1: Pilot pattern (gray circle: user data, black circle: pilot).
10
-1

... ... ... ... ... ... ... ...

M1

...

2c+1

... ...
2c+1 Time

10 NMSE 10

-2

-3

fdT = 0.01 fdT = 0.06 fdT = 0.1 10

-4

10 SNR (dB)

15

20

25

Fig 4.2: The normalized mean square error (NMSE) of channel estimation for the first M1 OFDM symbols, for f d T = 0.01, 0.06 and 0.1, respectively.

114

10

10

-1

NMSE

10

-2

fdT = 0.01, Q = 1, c =1 fdT = 0.01, Q = 2, c =1 fdT = 0.01, Q = 3, c =1 fdT = 0.1, Q = 1, c =1 fdT = 0.1, Q = 2, c =1 fdT = 0.1, Q = 3, c =1 5 10 SNR (dB) 15 20 25

10

-3

10

-4

Fig 4.3: The NMSE of channel estimation based on equi-spaced and grouped pilot pattern, for the polynomial order Q =1, 2, 3 and the normalized Doppler shift f d T = 0.01 and 0.1, respectively.
10
0

10

-1

NMSE

10

-2

fdT = 0.01, Q =2, c = 1 fdT = 0.01, Q =2, c = 2 10

-3

fdT = 0.01, Q =2, c = 3 fdT = 0.1, Q =2, c = 1 fdT = 0.1, Q =2, c = 2 fdT = 0.1, Q =2, c = 3 5 10 SNR (dB) 15 20 25

10

-4

Fig 4.4: The NMSE of channel estimation based on grouped pilot pattern, for c =1, 2, 3 and the normalized Doppler shift f d T = 0.01 and 0.1, respectively.

115

10

10

-1

NMSE

10

-2

fdT = 0.01, Ngroup = 18 fdT = 0.01, Ngroup = 36 10

-3

fdT = 0.01, Ngroup = 72 fdT = 0.1, Ngroup = 18 fdT = 0.1, Ngroup = 36 fdT = 0.1, Ngroup = 72 0 5 10 SNR (dB) 15 20 25

10

-4

Fig 4.5: The NMSE of channel estimation based on grouped pilot pattern, for the number of pilot groups N group =18, 36, 72 and the normalized Doppler shift f d T = 0.01 and 0.1, respectively.
10
1

fdT = 0.01, proposed, numerical method fdT = 0.01, proposed, simulation fdT = 0.06, proposed, numerical method fdT = 0.06, proposed, simulation fdT = 0.01, LS, numerical method fdT = 0.01, LS, simulation fdT = 0.06, LS, numerical method

10

NMSE

10

-1

fdT = 0.06, LS, simulation

10

-2

10

-3

10 SNR (dB)

15

20

25

Fig 4.6: The NMSE of channel estimation for the proposed algorithm and LS algorithm by numerical method and simulation at f d T = 0.01 and 0.06, respectively.

116

10

LS, fdT = 0.01 LS, fdT = 0.06 proposed, fdT = 0.01 10

-1

proposed, fdT = 0.06 ref [3], fdT = 0.01 ref [3], fdT = 0.06

BER 10
-2

10

-3

10 SNR (dB)

15

20

25

Fig 4.7: Bit error ratio (BER) of LS, the proposed algorithm and the algorithm in [29], for normalized Doppler shift f d T = 0.01 and 0.06, respectively.

117

Chapter 5: MIMO-OFDM system capacity with imperfect feedback channel

In this chapter, the capacity for MIMO-OFDM with imperfect feedback channel is investigated. The closed-loop MIMO-OFDM capacity with imperfect feedback has been derived. A system capacity indicator, namely, the feedback SNR, that reflects the relationship between closed-loop MIMO-OFDM capacity and feedback channel, is proposed. Based on the derived closed-loop MIMO-OFDM capacity, the lower threshold of feedback SNR is defined and also further investigated by numerical method. The chapter is organized as follows. Section 5.1 introduces the classic theory about MIMO system capacity. Section 5.2 investigates the closed-loop capacity with imperfect feedback channel for MIMO-OFDM systems. Section 5.3 is the summary of this chapter.

118

5.1 The open-loop and closed-loop capacity for MIMO Systems

MIMO systems which employ multi-element antenna arrays at the transmitter and receiver ends are capable of high data rate transmission [46], [47]. A number of MIMO techniques known as layered space time architectures or BLAST architectures [48][51] have been proposed. MIMO systems capacity have been intensively investigated [46], [47], [50][52] over the last decades. Foschini [47] and Telatar [46] have firstly formulated the system capacity of the MIMO systems assuming independent and identically distributed fading at different antennas. They have proved that the MIMO system capacity for n transmitter antennas and n receiver antennas increases linearly with n at a fixed transmitter power. This section briefly reviews the basic theory of MIMO system capacity.

5.1.1 MIMO system model We adopt the basic MIMO system model depicted by Fig 2.4. The MIMO system is equipped with NT transmitter antennas and NR receiver antennas. It is assumed that the fading channel is flat. Then, the MIMO system channel matrix is given by H = (hm,n ) N R NT = (am.n + jbm,n ) N R NT

(5-1)

where am,n and bm,n are Gaussian random variables with zero mean and unity variance. The transmitted signal is X = x1
x2 x NT , satisfying E ( XX H ) P .
T

(5-2)

That is, the total transmitted signal power is equal to or less than P. And the received

119

signal Y, is given by
Y = HX + w

(5-3)
x2 x NT , the AWGN noise w
T

where Y = y1 = w1
w2

y2
T

yN R x1 , X =

wN R .

5.1.2 MIMO system capacity It is well known that MIMO system capacity is defined by
C = max I ( X, Y )
p(X)

(5-4)

Where p(X) is the probability density function (PDF) of the transmitted signal X, I(X,Y) is the mutual information between X and Y. The mutual information I(X,Y) can be further written as I ( X, Y) = H (Y) H (w ) (5-5)

where H(Y) and H(w) are the entropies of the received signal Y and AWGN noise w. And the autocorrelation matrix of the received signal Y can be written as

R YY = E (YY H ) =

P 2 HR XX H H + w I NR , NT

(5-6)

2 is the variance of the AWGN noise, RXX is the autocorrelation matrix of X, I N R is where w

an NR by NR identity matrix, ( )H denotes the Hermitian transpose. So the mutual information can be further derived as [46], [47]
P I ( X, Y ) = log 2 det I N R + HR XX H H 2 NT w

(5-7)

120

Then, the MIMO system capacity can be further expressed by

P log 2 det I N R + HR XX H H 2 R XX ,tr( R XX ) = P NT w
max

C = max I ( X, Y) =

(5-8)

5.1.2.1 Open-loop capacity for MIMO systems When the transmitter does not know the channel state information (CSI), it can only set the covariance matrix of the transmitted signal to be an identity matrix, that is, Rxx = IN
T

(5-9)

Meanwhile, the input signal X should be Guassian distributed to achieve the capacity of the MIMO systems, which is given by
P Copen = log 2 det I N R + HH H 2 NT w

(5-10)

Formula (5-10) can be equivalently written as

m P Copen = log 2 1 + i 2 i =1 NT w

(5-11)

where m = min(NR, NT), i is the eigenvalues of the matrix HHH, i = 1,2,,m. 5.1.2.2 Closed-loop capacity for MIMO systems When the transmitter has the CSI by feedback channel, it could transmit the signal by the eigenmode to maximize the channel capacity. Since the channel matrix H can be factored as H = UVH, the formula (5-12) can be further expressed by (5-12)

121

C=

P log 2 det I N R + UV H R XX V H U H . 2 R XX ,tr( R XX ) = P NT w

max

(5-13)

As det (I + AB) = det (I + BA), we have

C= P log 2 det I N R + V H R XX V H 2 R XX ,tr( R XX ) = P NT w
max

(5-14)

Let G = VH Rxx V, we have that G should satisfy the following conditions to achieve the closed-loop MIMO system capacity [46], [47], that is,
G is a diagonal matrix. The i-th diagonal element of G is

Gi = max(0, u - 2 w /i), i = 1,2,,m, while u satisfying

(5-15)

G
i =1

=P

(5-16)

where m = min (NR, NT), P is the overall power of the transmitted signal. Only if the above two conditions are met, the closed-loop capacity of MIMO system is given by
m Cclosed = log 2 1 + i2 Gi i =1 w

(5-17)

where Gi is given by (5-15). Since the MIMO closed-loop capacity is expressed by (5-17), the transmitter should transmit the signal according the eigenmode in order to achieve the closed-loop capacity. The eigenmode referes to that the transmitted signal X multiplies two matrixes before transmission, that is, X = VG1/2 X . (5-18)

122

where V is obtained by SVD of H, G is a diagonal matrix with its i-th element given by (5-17). The eigenmode of transmission is depicted as follows.

Fig 5.1: The eigenmode transmission of MIMO systems.

It is worth noting that some reference books often put a matrix UH at the receiver. However, since det (I + UA) = det (I + A) det (U) = det (I + A), that is, a unitary matrix does not influence the MIMO system capacity, we can omit the unitary matrix UH. In addition, the transmitted signal X should satisfy RXX = IN , and it is easy to find that the transferred
T

signal X satisfies R XX = VGV H .

123

5.1.3 Numerical Results and discussion

4.5

MIMO system capacity (bit/s/Hz)

SNR = 5 dB SNR = 10 dB SNR = 15 dB

3.5

2.5

1.5 2

10

12

14

16

The number of transmitter antennas, NT

Fig 5.2: The MIMO system open-loop capacity versus the number of transmitter antennas NT , for the number of receiver antennas N R = 1 .
9

MIMO system capacity (bit/s/Hz)

SNR = 5 dB SNR = 10 dB SNR = 15 dB

4 6 8 10 12 14 16

2 2

The number of receiver antennas, NR

Fig 5.3: The MIMO system open-loop capacity versus the number of receiver antennas N R , for the number of transmitter antennas NT = 1 .

124

18 16 14 12 10 8 6 4 2 0

system capacity (bit/s/Hz)

MISO SIMO MIMO

10

The number of antennas, N

Fig 5.4: The capacities of the N by 1 MISO system, the 1 by N SIMO system, and the N by N MIMO system as a function of N, for SNR = 5 dB. Fig 5.2 shows the MIMO system open-loop capacity versus the number of transmitter antennas NT, for the number of receiver antennas NR = 1. The open-loop capacity is evaluated by formula (5-11). Observe that the capacity does not increase significantly when increasing the number of transmitter antennas, at a fixed SNR. For instance, at SNR = 5 dB, the capacity for NT = 2 and NT = 6 is 1.91 bit/s/Hz and 1.98 bit/s/Hz, respectively. The above results imply that the MIMO system capacity does not increase with the increase of the number of transmitter antennas. Fig 5.3 depicts the MIMO system capacity versus the number of receiver antennas NR, for the number of transmitter antennas NT = 1. Observe that the capacity increases almost linearly with the increase of NR, at a fixed SNR. For instance, at SNR = 10 dB, the capacity increases from 4.15 bit/s/Hz to 6.27 bit/s/Hz,

125

when NR is increased from 2 to 8. Fig 5.4 shows the capacities for MISO, SIMO, and MIMO systems as a function of the number of antennas N, respectively. Observe that the capacity for MISO system does not vary as N increases. The capacity slope of MIMO system is larger than that of SIMO system and both of them are nearly linear. So it is concluded that MIMO system capacity is linearly increased with the number of antennas N when NR = NT = N, and it is uncorrelated with the number of transmitter antennas when the number of receiver antennas NR =1.

35

NT = 2, NR = 2, open loop
30

MIMO system capacity (bit/s/Hz)

NT = 2, NR = 2, close loop NT = 6, NR = 6, open loop NT = 6, NR = 6, close loop

25

20

15

10

0 -5

10

15

20

SNR (dB)

Fig 5.5: The open-loop and closed-loop capacity for MIMO systems, versus SNR.

Fig 5.5 describes the open-loop and closed-loop capacity for MIMO systems, for two antenna pairs, which is, (2, 2) and (6, 6). The open-loop capacity and closed-loop capacity are evaluated by formula (5-11) and formula (5-17), respectively. Observe that both the open-loop capacity and closed-loop capacity are increased with increasing SNR. For SNR

126

is larger than 5 dB, the capacity increases almost linearly with increasing SNR, for the pairs (2, 2) and (6, 6). For a fixed antenna pair, the capacity gain of closed-loop capacity over open-loop capacity is reduced when increasing SNR. The results imply that water filling algorithm of increasing MIMO system capacity should be adopted at lower SNR.

5.2 The closed-loop capacity with imperfect feedback channel for MIMO-OFDM systems
Since OFDM technique can mitigate the ISI and transform the frequency selective fading channel into a set of flat fading channels [1], the combination of MMO with OFDM technique has become a promising candidate for the next generation wireless communication. A new protocol draft employing the MIMO-OFDM for the physical layer, IEEE 802.11n, as an amendment to IEEE 802.11 standards has been proposed [53]. A few studies on MIMO-OFDM system capacity have been carried out [54][58]. A MIMO broadband fading channel model is introduced in [57] which investigates the effect of physical parameters on the system capacity such as the amount of delay spread, cluster angle spread and total angle spread. Intarapanich et al. [58] shows that when the tap gain correlation is greater than a certain threshold, the capacity advantage of the broadband system vanishes. In this section, we study the MIMO-OFDM system capacity with imperfect feedback since most articles focus on MIMO capacity over perfect feedback channel and research

127

outputs on MIMO-OFDM system capacity with imperfect feedback or MIMO system capacity with imperfect feedback are not many. Next, we introduce the MIMO-OFDM system model adopted in this section. 5.2.1 System Model

Fig 5.6: The closed-loop MIMO-OFDM system model.

Fig 5.6 depicts the closed-loop MIMO-OFDM system model. The MIMO-OFDM system is equipped with NT transmitter antennas and NR receiver antennas. Assumed that there is no time delay in the feedback channel from the receiver to the transmitter and the fading channel is ergodic and stationary. The received signal y k at the k-th subcarrier of an OFDM symbol is give by y k = H k x k + w k , k = 0,1,..., N 1; where H k is the NRNT channel frequency response matrix at the (5-19) k-th

128

0 subcarrier, y k = [ yk

y1 k

i ykN R 1 ]T , yk is the received signal at the k-th subcarrier of an

OFDM symbol for the i-th receiver antenna, the symbol ()T denotes transpose operation,
0 x k = [ xk

x1 k

i xkNT 1 ]T and xk is the transmitted signal at the k-th subcarrier of an

OFDM symbol for the i-th transmitter antenna, N is the number of subcarriers of an
0 OFDM symbol, w k = [ wk

w1 k

i wkNT 1 ]T and wk is the additive white Gaussian noise at

the k-th subcarrier in the i-th receiver antenna in the frequency domain. Assume that the elements of the vector w k are independently and identically distributed (i.i.d.) and satisfy
H 2 E (w k1 w k ) = w I N R (k1 k2 ) 2

(5-20)

where E () denotes expectation, I N R is the NRNR identity matrix, (i) is the Kronecker delta
2 function and w is the variance of the AWGN noise. Denote the l-th tap of the MIMO

fading channel response in time domain by H l . The NRNT complex random matrix H l can be written as [57]
2 H l = R1/ l H w ,l , l = 0,1,..., L 1;

(5-21)

where R l is the N R N R correlation matrix of H l , H w,l denotes the NRNT i.i.d. matrix with CN (0,1) entries and L is the number of resolvable paths. Assume that there is no spatial correlation at the receiver and at the transmitter so that R l = l2 I N R , where IN is an
R

NRNR identity matrix. Without loss of generality, assuming that the channel is normalized so that l2 = 1 , the channel impulse response matrix in equation (5-21) can be further
l =0 L 1

written as H l = H w,l , l = 0,1,..., L 1. (5-22)

The channel frequency response matrix at the k-th subcarrier H k is therefore given by

129

H k = H l exp{ j 2 k / N }
l =0

L 1

(5-23)

The SNR for MIMO-OFDM systems is defined as SNR = P 2 N w (5-24)

2 where P is the maximum overall transmitter power, w is the variance of the AWGN noise

of the channel H. 5.2.2 Closed-Loop Capacity and Feedback SNR for MIMO-OFDM Systems 5.2.2.1 The open-loop capacity and the closed-loop capacity for MIMO-OFDM systems For MIMO-OFDM systems under an average transmitter power constraint P, the system capacity is given by [57] 1 1 H det E log + I H H NR N 2 N w

C=

(5-25)

2 is the variance of the AWGN noise as where the symbol () H denotes Hermitian transpose, w

in the equation (5-25) above, H = diag {H k }k =0 , I N R N is an N R N N R N identity matrix,

N 1

the NT N NT N matrix satisfies the power constraint, i.e., Tr ( ) P , and Tr () denotes

the trace operation and P is the maximum overall transmitter power. The matrix is a block diagonal matrix and it is given by = diag { k }k =0
N 1

(5-26)

where the NT NT matrix k is the covariance matrix of the transmitting signal x k in

130

frequency domain and it is defined by

H k = E ( xk xk ) , k = 0,1,..., N 1.

(5-27)

1) The open-loop capacity for MIMO-OFDM systems If the channel state information is unknown by the transmitter, i.e., there is no feedback channel, the optimum power allocation is obtained according the equal power allocation, i.e., the total power is allocated uniformly on all space-frequency subcarriers. So the matrix k =
P I N , where I NT is an NT NT identity matrix. From the above equation NT N T

(5-25), the open-loop capacity for MIMO-OFDM systems is given by [57]

Copen =

1 N

log E det I
k =0

N 1

NR N

SNR H Hk Hk NT

(5-28)

where SNR is the MIMO-OFDM system signal to noise ratio defined by (5-24). Using singular value decomposition of matrix, the open-loop capacity can be written as [60]
Copen = SNR 1 N 1 Nmin E log 1 + k ,i N k =0 i =1 NT

(5-29)

H min where N min = min( NT , N R ) , {k ,i }iN =1 are the eigenvalues of the matrix H k H k . For each k,

H k is a random matrix with elements following i.i.d. complex Gaussian distribution with

zero mean and variance 1. So the probability density function (PDF) of the nonzero
H is given by [61] eigenvalue of the matrix H k H k

1 p(k ,i ) = N min

N min 1

k =0

2 k! max N min (k ,i ) LN kN,imax Nmin e k ,i , k ,i 0. k (k + N max N min )!

(5-30)

max N min where N min = min( NT , N R ) , N max = max( NT , N R ) , the function LN ( x) is the k

generalized k-th order Laguerre polynomial and it is defined by

131

max N min LN ( x) = k

1 x Nmax Nmin d k x Nmax N min + k e x e x , k = 0,1,..., N min 1. k! dx k

(5-31)

H , k ,i , is independent of Note that the PDF of the nonzero eigenvalues of the matrix H k H k

the index k and i. Using (5-30), the formula (5-29) can be written as a close form expression and it is given by
+ 2 SNR Nmin s! N max N min Nmax Nmin e d Copen = log 1 + L ( ) k 0 NT s =0 ( s + N max N min )!

(5-32)

The above formula (5-32) is the open-loop capacity for MIMO-OFDM systems equipped with NT transmitter antennas and NR receiver antennas.

2) The closed-loop capacity for MIMO-OFDM systems with perfect CSI When the CSI is perfectly known by the transmitter, the optimal power allocation is obtained by the water-filling allocation, i.e., the total power is allocated to those subchannels with high SNRs and not allocated to those subchannels with low SNRs. The covariance matrix of the transmitting signal x k is given by

k = Vk H Q k Vk

(5-33)

where Vk is an NT by NT unitary matrix extracted from the singular value decomposition (SVD) of the matrix H k . The matrix H k can be factored by SVD as

H k = U k k Vk H

(5-34)

where U k is an NR by NR unitarty matrix, Vk is an NT by NT unitarty matrix, and k is an NR by NT matrix satisfying

k = diag

k ,1 , k ,2 ,..., k , N

min

(5-35)

132

H min Notice that {k ,i }iN =1 are also the eigenvalues of the Wishart matrix H k H k . Notationally the

matrix k is written as a diagonal matrix, even though it is not square when N R NT . The elements k ,1 , k ,2 ..., k , Nmin are written along the main diagonal, and rows or columns of zeros are appended necessary to obtain the proper dimension for k if it is not square. The matrix Q k in equation (5-33) is an NT by NT diagonal matrix and is given by

Q k = diag ( Qk ,i )i =1 .
NT

(5-36)

Each diagonal element of Q k is given by 2 Qk ,i = max 0, u w k ,i where u satisfies

2 w max 0, u k ,i k = 0 i =1 N 1 N min

(5-37)

=P

(5-38)

The closed-loop capacity for MIMO-OFDM systems with perfect CSI is therefore given by [60]
Cclose
+ N 1 N min k ,i u 1 = E log 2 N k =0 i =1 w

(5-39)

where u is given by equation (5-38), the symbol ( x) + denotes max(0, x).

5.2.2.2 The closed-loop capacity for MIMO-OFDM systems with feedback channel
In a practical MIMO-OFDM system with feedback channel, the feedback channel may be lossy and cause imperfect feedback of the channel state information to the transmitter. In this section, the closed-loop capacity for MIMO-OFDM systems with imperfect feedback channel is presented. As shown in Fig 5.6, the receiver performs channel

133

estimation and obtains full channel frequency response matrix H, that is, the channel estimation is assumed perfect. Then, the receiver sends H by a feedback channel and the transmitter obtains H . The distortion between H and H is defined as

=
2 d

E || H H ||2 2 E (|| H ||
2 2

(5-40)

where || ||2 denotes the Frobenius norm. We model the distortion as an AWGN noise with variance 2 d and the feedback SNR is defined as SNR feed = E (|| H ||2 2 ) /( NNT N R )
2 = 1/ d

2 d

(5-41)

When the transmitter acquires H , it decomposes the received H by singular value decomposition (SVD) as
H k = U k k Vk , k = 0,1,, N-1.

(5-42)

where H = diag H k

{ }

N 1 k =0

. Then, the transmitter adopts the eigenmode transmission by the

following linear transformation expressed by

1/ 2 x k = Vk D k x k , k = 0,1,..., N 1.

(5-43)

where Vk is obtained by formula (5-42), Dk is a diagonal matrix with its element 2 Dk ,ii = max 0, u w k ,i and u satisfies
2 w max 0, u k ,i k = 0 i =1 N 1 N min

(5-44)

=P of
k

(5-45) given by
k

where

k ,i

is

the

i-th

diagonal
min

element

(5-42)

and k = diag

k ,1 , k ,2 ,..., k , N

) . Consequently, the received signal y at the k-th

134

subcarrier of an OFDM symbol is give by

y k = H k x k + wk
2 = H k Vk D1/ k x k + w k , k = 0,1,..., N 1.

(5-46)

2 Let Ek = H k Vk D1/ , we have that the MIMO-OFDM system capacity Cclosed at the k

eigenmode transmission satisfies [62]

Cclosed Cworst = 1 N

E
k =0

N 1

Ek

SNR H I E E log det + 2 N k k R NT

(5-47)

And the equality holds if the noise is AWGN. Since the equivalent noise is also wk expressed by (5-46), thus the equality in (5-47) holds. So the entire power allocation procedure of the closed-loop MIMO-OFDM system with feedback channel is summarized as follows. Step 1: The receiver retrieves perfect CSI, i.e., the channel frequency response matrix H and sends it to the transmitter. Step 2: The transmitter receives the distorted channel matrix H and performs SVD to get the eigenvector Vk , expressed by (5-42). Then, Dk is obtained by water filling algorithm expressed by (5-44) and (5-45). Finally, the linear transformation expressed by (5-43) is adopted and the transmitted signal x k is sent out. 5.2.2.3 Lower thresholds of feedback SNR for the closed-loop MIMO-OFDM system with feedback channel When the transmitter has the perfect CSI, the closed-loop capacity of MIMO-OFDM system is larger than the open-loop capacity of MIMO-OFDM system. However, when the transmitter has imperfect CSI, i.e., the feedback channel is distorted by noise in practice,

135

the inaccurate CSI may mislead the transmitter to perform unmatched eigenmode transmission as expressed in (5-43). When the noise power of the feedback channel is large, the mismatched closed-loop capacity of MIMO-OFDM systems is inferior to the open-loop capacity of MIMO-OFDM system. Thus there exists a lower threshold of feedback SNR to guarantee that the mismatched closed-loop capacity of MIMO-OFDM systems is larger than or equal to the open-loop capacity of MIMO-OFDM system. To further examine the dependence of the capacity, the system capacity with feedback channel

( NT , N R , SNR, SNR feed ) , where NT is the number of transmitter = Cclosed is written as Cclosed
antennas, NR is the number of receiver antennas, SNR is the system signal to noise ratio, and SNRfeed is the feedback SNR defined by (5-40). Similarly, we

write Copen = Copen ( NT , N R , SNR) . The lower threshold of feedback SNR is defined as
( NT , N R , SNR, SNR feed ) = Copen ( NT , N R , SNR ) SNR feed ,lower ( NT , N R , SNR ) = arg Eq Cclosed .
SNR feed

(5-48)
5.2.3 Numerical Results Numerical method has been employed to investigate the MIMO-OFDM system capacity

with feedback channels. We employed the channel model of COST207 [35] having 6 numbers of paths and the maximum delay spread of 5 us. The number of the subcarriers of the MIMO-OFDM system, N, is 1024 and the CP length is equal to 128 sample points. So the length of an OFDM symbol including the CP in unit of sample point is equal to 1152. The bandwidth of MIMO-OFDM system is 20 MHz so that the symbol duration is 57.6 s and the CP duration, TCP = 6.4s>5s.
136

1) The open-loop capacity and the closed-loop capacity for MIMO-OFDM systems Fig 5.7 shows the open-loop and closed-loop system capacity of MIMO-OFDM systems having different transmitter antenna and receiver antenna numbers versus system SNR. The closed-loop capacity of MIMO-OFDM systems is evaluated using equation (5-39) by Monte simulation. The open-loop capacity of MIMO-OFDM systems is evaluated using equation (5-32) by numerical method. Observe that both the open-loop capacity and the closed-loop capacity for MIMO-OFDM systems increase with increasing SNR and number of antennas. The closed-loop capacity gain over the open-loop capacity with large antenna numbers is more substantial than the case for small antenna numbers, especially at low SNR. The results imply that the use of water-filling algorithm is able to increase the system capacity for NT = NR 4 and SNR 15 dB, reaching marginally the maximum closed-loop capacity.

137

The capacity of MIMO-OFDM systems (bit/s/Hz)

90

NT = 2, NR = 2,open loop
80

NT = 4, NR = 4,open loop NT = 8, NR = 8,open loop NT = 16, NR = 16,open loop NT = 2, NR = 2,close loop NT = 4, NR = 4,close loop NT = 8, NR = 8,close loop NT = 16, NR = 16,close loop

70

60

50

40

30

20

10

10

12

14

16

18

20

System SNR (dB)

Fig 5.7: The open-loop and closed-loop system capacity for MIMO-OFDM systems having different transmitter antenna and receiver antenna numbers.

2) The closed-loop capacity with imperfect feedback channel for MIMO-OFDM systems Fig 5.8 shows the capacity gain of the closed-loop capacity with imperfect feedback over the open-loop capacity versus the feedback channel SNR, SNRfeed, for NT = NR = 4. Observe that the capacity gain increases with the increase of SNRfeed for each capacity curve at a fixed MIMO-OFDM system SNR. Low SNRfeed yields negative capacity gain, especially at high system SNR. For instance, the capacity gain is negative when the system SNR = 10 dB and SNRfeed is lower than 7.0 dB. High SNRfeed yields positive capacity gain, especially at low system SNR. However, for further increasing the SNRfeed the capacity

138

gain approaches an asymptotic level and therefore will not be further increased significantly. For instance, when SNRfeed = 20 dB and system SNR = 15 dB, the capacity gain achieves about 0.22 bit/s/Hz. Further increase of SNRfeed only marginally increases the capacity gain, approaching the asymptotical level of about 0.23 bit/s/Hz. Fig 5.9 depicts the capacity gain of closed-loop capacity with imperfect feedback over open-loop capacity versus the feedback channel SNR, for system SNR = 10 dB. Observe that the capacity gain increases with the increase of the feedback channel SNR, at a fixed antenna pair. Also the slope of the capacity gain at a fixed SNRfeed increases with the number of antenna pair. Note that the lower thresholds of SNRfeed for different numbers of antenna pairs increase with the number of antenna pairs gradually. For instance, the lower threshold of SNRfeed for (2, 2) antenna pair is about 6.5 dB, while the lower threshold for (4, 4) pair is about 6.8 dB.

139

The capacity gain with feedback channel (bit/s/Hz)

0.8

0.6

System SNR = 5 dB System SNR = 10 dB System SNR = 15 dB System SNR = 20 dB System SNR = 25 dB

0.4

0.2

-0.2

-0.4 -5

10

15

20

25

30

Feedback channel SNR (dB)

Fig 5.8: The capacity gain of the closed-loop capacity with imperfect feedback over that of the open-loop capacity versus feedback channel SNR, for NT = N R = 4 .
The capacity gain with feedback channel (bit/s/Hz)

2.5

NT = 2, NR = 2
2 1.5 1 0.5 0 -0.5 -1 -1.5 -5

NT = 4, NR = 4 NT = 8, NR = 8 NT = 12, NR = 12 NT = 16, NR = 16

10

15

20

25

30

Feedback channel SNR (dB)

Fig 5.9: The capacity gain of closed-loop capacity with imperfect feedback over that of the open-loop capacity versus feedback channel SNR, for system SNR = 10 dB.

140

3) The lower thresholds of the feedback channel SNR Fig 5.10 shows the feedback SNR lower threshold versus the MIMO-OFDM system SNR, for different antenna pairs. Observe that the lower threshold of feedback SNR at a fixed antenna number almost linearly increases with the increase of the system SNR. For instance, for the antenna pair (8, 8), the lower threshold increases from 3.4 dB to 21.5 dB for the system SNR increases from 5 dB to 25 dB. At a fixed system SNR, the lower threshold of feedback SNR increases when the number of antennas is increased. For instance, for the system SNR = 10 dB, the lower threshold increases from about 6.2 dB to 9.3 dB when the number of antennas is increased from 2 to 12. In addition, the lower threshold gap between small number of antennas and large number of antennas is reduced gradually when the system SNR increases. For instance, the lower threshold gap between the antenna pair (2, 2) and (12, 12) is reduced from 4.2 dB to 1.0 dB when the system SNR is increased from 5 dB to 25 dB.

141

The lower threshold of feedback channel SNR (dB)

25

NT = 2, NR = 2 NT = 4, NR = 4
20

NT = 8, NR = 8 NT = 12, NR = 12 NT = 16, NR = 16

15

10

10

15

20

25

System SNR (dB)

Fig 5.10: The lower threshold of the feedback SNR versus the MIMO-OFDM system SNR, for different antenna pairs.

5.3 Summary
In this chapter, we first introduce the open-loop capacity and the closed-loop capacity for MIMO systems. Numerical results for MIMO systems capacity reveal that the MIMO system capacity linearly increases with the antenna number when the number of transmitter antennas equals to the number of receiver antennas. Water filling algorithm had better be adopted in low SNR region since the capacity gain of closed-loop over the open-loop is significant at low SNR. Then, we study MIMO-OFDM system capacity with imperfect feedback. The closed-loop capacity for MIMO-OFDM system with imperfect

142

feedback has been derived. An indicator of the feedback SNR, has been proposed to describe the relationship between the feedback channel and the closed-loop MIMO-OFDM capacity. The lower threshold of feedback SNR has been defined and investigated by numerical method. The numerical results reveal the followings. First, lower feedback SNR yields negative capacity gain, especially at high system SNR. Higher feedback SNR yields positive capacity gain, especially at low system SNR. Second, the lower threshold of feedback SNR is proportional to the number of antennas and also proportional to the MIMO-OFDM system SNR.

143

Chapter 6: Capacity of OFDM systems over time and frequency selective fading channels
In this chapter, the variance and mean value of OFDM system capacity over Rayleigh fading channels and Ricean fading channels have been thoroughly investigated. The mathematical expressions of the capacity variance, as the function of the number of the subcarriers of an OFDM symbol, the SNR and the power profile of the channel, for both Rayleigh fading channels and Ricean fading channels have been derived. The system capacity variances over Rayleigh and Ricean fading channels have been evaluated by computer simulation and verified by numerical method. The chapter is organized as follows. Section 6.1 introduces the previous research work about OFDM system capacity. Section 6.2 describes the OFDM system model. Section 6.3 provides the variances of OFDM system capacity over the multipath Rayleigh fading channel and the multipath Ricean fading channel. The numerical and simulation results are provided and discussed in Section 6.4 followed by conclusion in Section 6.5.

144

6.1 Introduction
A few studies on the capacity of OFDM systems have been conducted [65][69]. A. Clark et al. [65] derive the formulas of the mean and variance of OFDM system capacity. They show that as the number of subcarriers for a power limited OFDM system approaches infinity, the capacity approaches a constant value, which equals the capacity of the infinite-bandwidth Gaussian channel. H. Blcskei et al. [57] provide the expressions for the ergodic capacity and the outage capacity of OFDM-based spatial multiplexing systems, assuming that the channel is unknown at the transmitter and perfectly known at the receiver. By introducing a physically motivated MIMO broadband fading channel model, the influence of physical parameters such as the amount of delay spread, cluster angle spread, and total angle spread have been studied [57]. C. R. N. Athaudage et al. [66] analyzed the capacity of OFDM systems with carrier frequency offset (CFO) in frequency selective Nakagami-m fading channels. The authors of [66] provided an analytical technique of evaluating the capacity of an OFDM system with CFO. J. Y. Yun et al. [69] propose a rate-t/k code of canceling the ICI for OFDM systems. They derive the capacity lower bound of OFDM systems employing the proposed code and design the optimal code to maximize the lower bound. The simulation results reveal that the propose rate-2/3 code can improve the system capacity and BER while causing a minor increase in the implementation complexity. In this chapter, the variance of OFDM system capacity is thoroughly investigated. The

145

exact expressions for the variance of OFDM system capacity over Rayleigh fading channels and Ricean fading channels have been derived. Although A. Clark et al. [65] has derived the mean and the variance of OFDM system capacity over Rayleigh fading channels, however, the channel they used has an infinite number of resolvable paths with a decaying exponential power profile and a certain root mean square delay. In addition, they does not show the effects of some important system parameters such as the number of subcarriers of one OFDM symbol, the CP length, on the system capacity. The channel used in this chapter is the discrete multipath fading channel with finite number of paths. The exact expressions for the variance of OFDM system capacity based on the discrete multipath Rayleigh fading channel and the discrete multipath Ricean fading channel have been derived. The effects of system parameters, such as the number of subcarriers of an OFDM symbol, the number of multipath of the fading channel, the CP length, the system SNR, on the system capacity have been thoroughly investigated. In addition, although A. A. Abu-Dayya et al. [71] have provided the joint PDF of two correlated Ricean random variables, they make the assumption that the envelopes of two correlated Ricean random variables are the same and the correlation coefficient is real. J. R. Mendes and M. D. Yacoub [72] have provided a general bivariate Ricean model and derived the joint normalized PDF of two correlated Ricean random variables, however, the expression of the joint PDF given by them is in a series form. This chapter presents the joint PDF two correlated Ricean random variables without limitations of [71] and the expression is in an integral form.

146

The rest chapter is organized as follows. The OFDM system model is described in section II. The expression for the variance of OFDM system capacity is presented in section III. The numerical and simulation results are provided and discussed in section IV, followed by the conclusion in section V.

6.2 OFDM System Model

We adopt the OFDM system model in [1] without loss of generality. For N carriers in the OFDM system, the transmitted signal x(i, n)in time domain after IFFT is given by
x(i, n) = IFFTN [ X (i, k )] = 1 N

X (i, k ) e
k =0

N 1

j 2 nk / N

(6-1)

where X(i, k) denotes the transmitted signal in frequency domain at the k-th subcarrier in the i-th OFDM symbol. A CP is inserted into each OFDM symbol prior to the transmission and the CP is removed before the FFT process at the receiver. The received signal Y(i, k) in frequency domain after FFT can be written as
Y (i, k ) = X (i, k ) H (i, k ) + W (i, k ), k = 0,1,..., N 1.

(6-2)

where W(i ,k) denotes the additive white Gaussian noise (AWGN) with zero mean and
2 , H(i, k) is the channel response in frequency domain at the k-th subcarrier of variance W

the i-th OFDM symbol and it is formulated by

147

H (i, k ) = DFTN (h(i, n)) = h(i, n)e

n=0 L 1 l =0 N 1 j 2 kn N

(6-3) , k = 0,1,..., N 1.

= h(i, l )e

2 k l N

And h(i, l) is the channel response in time domain at the l-th path within the i-th OFDM symbol period, l is the l-th path delay, L is the number of total resolvable paths of the fading channel. In addition, the SNR for OFDM systems is defined by SNR =
2 X 2 W

(6-4)

2 2 where W is the AWGN noise power, X = E (| X (i, k ) |2 ) .

6.3 OFDM System Capacity

In this section, the mean value and the variance of OFDM system capacity are derived. Two cases based on the channel type are studied, which are Rayleigh fading channel and Ricean fading channel.
6.3.1 OFDM system capacity over Rayleigh fading channels

This subsection presents OFDM system capacity over Rayleigh fading channels. First, we present the correlation properties of the channel frequency response coefficients for further derivation of the variance of OFDM system capacity. Second, we derive the formulas of the mean and variance of OFDM system capacity.

148

6.3.1.1 Correlation properties of the channel frequency response coefficients

It is assumed that the paths h(i, l) (l = 0,1,...,L-1) are statistically independent and each path h(i, l) follows the circulant complex Gaussian distribution with zero mean and l2 variance. Thus, the envelope of each path is Rayleigh distributed. The channel is
2 normalized so that h = l2 = 1 . The channel autocorrelation function is given by l =0 L 1

* E h(i1 , l1 )h (i2 , l2 )

( l1 l2 ) J 0{2 f d Tsym (i1 i2 )} = l2 1

(6-5)

where (i) is the Kronecker delta function, J 0 (i) is the zero order Bessel function of the first kind, fd is the maximum Doppler shift, Tsym is the time duration of an OFDM symbol including the CP, (i)* denotes conjugate. An exponentially decaying power profile is used and expressed by

l2 = exp ( l )

(6-6)

where l2 is the power of the l-th path and l is the l-th path delay, is a decaying factor. Each path h(i, l) of the fading channel can be further expressed by h(i, l ) = hRe (i, l ) + jhIm (i, l ) (6-7)

where hRe(i, l) is the real part of h(i, l), hIm(i, l) is the imaginary part of h(i, l). Since h(i, l) follows the circulant complex Gaussian distribution with zero mean and l2 variance, we have that both hRe(i, l) and hIm(i, l) are Gaussian distributed with zero mean and l2 / 2 variance; hRe(i, l) is independent on hIm(i, l). Next, the frequency response of the channel, H(i, k), can be further expressed by H (i, k ) = H Re (i, k ) + jH Im (i, k ) (6-8)

149

where HRe(i, k) is the real part of H(i, k), HIm(i, k) is the imaginary part of H(i, k). The correlation between the coefficient HRe(i, k1) and HRe(i, k1) is derived as
E [ H Re (i, k1 ) H Re (i, k2 ) ] L 1 2 k1 l 2 k1 l hRe (i, l ) cos + hIm (i, l ) sin N N 0 l = = E L 1 2 k2 l 2 k2 l + hIm (i, l ) sin hRe (i, l ) cos N N l =0 L 1 1 2 k1 l 2 k2 l 2 k1 l 2 k2 l = l2 cos + sin cos sin N N N l =0 2 N = 1 L 1 2 2 (k1 k2 ) l l cos 2 l =0 N

(6-9)

Similar to the derivation of (6-9), we have

E [ H Im (i, k1 ) H Im (i, k2 ) ] = E [ H Re (i, k1 ) H Im (i, k2 ) ] = 1 L 1 2 2 (k1 k2 ) l l cos N 2 l =0 1 L 1 2 2 (k1 k2 ) l l sin N 2 l =0

(6-10)

(6-11)

From (6-11), we immediately have

E [ H Re (i, k1 ) H Im (i, k2 )] = E [ H Re (i, k2 ) H Im (i, k1 )]

(6-12)

The formulas from (6-9) to (6-12) are the bases of the following derivation of mean value and variance of OFDM system capacity.
6.3.1.2 The mean value of OFDM system capacity

The capacity of the k-th subcarrier within an OFDM symbol is given by

Ci ,k = log 2 (1 + SNR | H (i, k ) |2 )

(6-13)

where SNR is the signal to noise ratio of the OFDM system, given by formula (6-4). Therefore, the mean value of OFDM system capacity for Rayleigh fading channels is

150

given by [65]

C = E

1 N

C
k =0

N 1

i ,k

(6-14)

2 = E log 2 (1 + SNR | H (i, k ) | )

=e

1 SNR

1 Ei bit SNR

where Ei() is the exponential integral function and it is defined as Ei( z ) =

+

et dt t

(6-15)

6.3.1.3 The variance of OFDM system capacity

The variance of OFDM system capacity is expressed as

2 2 1 N 1 1 N 1 = E Ci ,k E Ci ,k N k = 0 N k =0 1 N 1 1 N 1 N 1 2 = 2 E (Ci2,k ) + 2 E Ci ,k1 Ci ,k2 C N k =0 N k1 =0,k1 k2 k2 =0 2 C

(6-16)

Since |H(i, k)|2 follows the exponential distribution with variance 1, that is, the probability density function (PDF) of |H(i, k)|2 is expressed as
f|H ( i ,k )|2 ( x ) = e x , x 0

(6-17)

The first term in formula (6-16) is further derived as 1 N2

E (C
k =0

N 1

2 i ,k

)=

1 + 2 ( log 2 (1 + SNR x) ) f|H (i ,k )|2 ( x)dx 0 N 1 + 2 = ( log 2 (1 + SNR x) ) e x dx N 0

(6-18)

Next, we further derive the second term of formula (6-16). As in [70], for two 21 Gaussian random vectors Xc and Xs, given by

151

X c = [ X c1

X c 2 ]T , X s = [ X s1

X s 2 ]T

(6-19)

with zero mean and covariance matrices Kcc, Kss and cross-covariance matix Kcs. That is,

12 K cc = cc 1 2 12 K ss = ss 1 2 0 K cs = sc 1 2
meanwhile satisfying

cc 1 2 2 2 ss 1 2 22 cs 1 2
0
(6-20)

cc = ss , cs = sc

(6-21)

, where ()T denotes conjugate, cc, ss, cs and sc are the correlation coefficients of (Xc1, Xc2),
2 (Xs1, Xs2), (Xc1, Xs2), (Xs1, Xc2) , respectively. And 12 is the variance of Xc1, 2 is the variance

of Xc2, we have that the joint PDF of the bivariate r1 and r2, is given by

f r1 ,r2 ( x, y ) =

2 2 x xy y cc + cs 1 1 , I exp + 0 2 2 2 2 2 2 2 2 2 4 12 2 (1 cc ) cs 1 cc cs 1 2 2(1 cc cs ) 1 2

for x, y 0. (6-22) where r1 = X c21 + X s21 , r2 = X c22 + X s22 , I 0 () is the modified zero order Bessel function of the first kind. Let Xc1 = HRe(i, k1), Xc2 = HRe(i, k2), Xs1 = HIm(i, k1), and Xs2 = HIm(i, k2), we have
2 12 = 2 = 0.5

cc = ss = l2 cos
l =0

L 1

2 (k1 k2 ) l N

and

cs = sc = l2 sin
l =0

L 1

2 (k1 k2 ) l . Since the two 21 Gaussian random vectors, [HRe(i, N

k1) HRe(i, k2)]T and [HIm(i, k1) HIm(i, k2)]T, satisfy the conditions (6-20) and (6-21), the joint

152

PDF of |H(i, k1)|2 and |H(i, k2)|2 is given by

f|H (i ,k )|2 ,|H (i ,k
1

2 2 )|

( x, y ) =

x + y 2 1 exp I0 1 1 1

xy , for x, y 0.

(6-23)

where

= l21 l22 cos

l1= 0 l 2 = 0

L 1 L 1

2 (k1 k2 ) ( l1 l 2 ) . N

(6-24)

Therefore, the correlation between the capacity of subcarrier k1, Ck1 and the capacity of subcarrier k2, Ck2 , is given by

E Ci ,k1 Ci ,k2 =

log 2 (1 + SNR x) log 2 (1 + SNR y ) f|H (i ,k )|2 ,|H ( i ,k

1

2 2 )|

( x, y )dxdy . (6-25)
2 2 )|

where SNR is the system signal to noise ratio, defined by (6-4). f|H (i ,k )|2 ,|H ( i , k
1

( x, y ) is the

joint PDF of |H(i, k1)|2 and |H(i, k2)|2, expressed by (6-23). Therefore, the variance of
2 is finally obtained by substituting above formulas (6-14), (6-18) OFDM system capacity C

and (6-25) into (6-16).

6.3.2 OFDM system capacity over Ricean fading channels

This subsection provides the mean value and variance of OFDM system capacity over Ricean fading channels. First, the correlation coefficient between two arbitrary channel frequency responses is presented for further derivation of the variance of OFDM system capacity. Second, the mean and variance of OFDM system capacity are derived.
6.3.2.1 The correlation coefficient between two channel frequency responses

It is assumed that the paths h(i, l) (l = 0,1,...,L-1) are statistically independent and each path h(i, l) is given by

153

h(i, l ) = Al + i ,l

(6-26)

where Al is the LOS for the l-th path, i,l is the scattering signal following circulant complex Guassian distribution with zero mean and l2 variance. Therefore, the envelope of each path is Ricean distributed, expressed by

x 2 + | Al |2 2 x | Al 2x fl ( x) = 2 exp I l l2 0 l2

(6-27)

where I 0 () is the 0-th modified Bessel function of the first kind. The channel is normalized
2 so that h = l2 = 1 , where l2 =| Al |2 + l2 is the total power of the l-th path, composed l =0 L 1

of the LOS signal and Non-LOS signal power. The Ricean K-factor for the l-th path is K l =| Al |2 / l2 and the Ricean K-factor for the multipath Ricean fading channel is defined by
K = | Al |2 / l2
l =0 l =0 L 1 L 1

(6-28)

Then, the frequency response of the channel H(i, k) can be expressed by

H (i, k ) = FFTN (h(i, n)) = h(i, l )e j 2 k l / N
l =0 L 1

= Al e
l =0

L 1

j 2 k l / N

+ i ,l e
l =0

L 1

(6-29)
j 2 k l / N

= a (i, k ) + b(i, k )

where a(i, k) is the LOS term of H(i, k) and b(i, k) is the Non-LOS term of H(i, k). And it is easy to verify that the variance of b(i, k) is l2 and
l =0 L 1

1 N

| a(i, k ) | = | A | . Thus, the

2 2 l k l
1 2

correlation coefficient between two arbitrary frequency responses, k ,k , is derived as

154

2 k ,k = E H (i, k1 ) E ( H (i, k2 ) ) H (i, k2 ) E ( H (i, k2 ) ) / l *

1 2

L 1

l =0

= E i ,l e j 2 k1 l / N i ,l e j 2 k2 l / N / l2 l l l =0
L 1

(6-30)

= l2 e j 2 ( k1 k2 ) l / N / l2
l =0 l =0

L 1

L 1

6.3.2.2 The mean value of OFDM system capacity

The capacity of the k-th subcarrier within an OFDM symbol is also given by equation (6-13). From (6-29), since the envelope of the channel frequency response |H(i, k)| follows Ricean distribution, the PDF of |H(i, k)|, f|H ( i ,k )| ( x ) , is given by

2 2 2x x + | a(i, k ) | 2 x | a(i, k ) | , x 0. f|H (i ,k )| ( x) = L 1 exp I 0 L 1 L 1 2 2 2 l l l l =0 l =0 l =0

(6-31)

Therefore, the mean value of OFDM system capacity over Ricean fading channels, C , is given by
2 =E C log 2 (1 + SNR | H (i, k ) | )

1 N

k =0

N 1

log 2 (1 + SNR x 2 ) f|H (i ,k )| ( x)dx

(6-32)

6.3.2.3 The variance of OFDM system capacity

The variance of OFDM system capacity for Ricean fading channels is also given by equation (6-16). Note that C , instead of C , should be used in (6-16). The first term in (6-16) is further derived as
1 N2

E (C
k =0

N 1

2 i ,k

)=

1 N2

( log
+ k =0 0

N 1

(1 + SNR x 2 ) ) f| H (i ,k )| ( x)dx
2

(6-33)

155

where f|H(i, k)| is given by formula (6-31). Next, we further derive the second term of formula (6-16). Since |H(i, k1)| and |H(i, k2)| are two Ricean random variables with correlation coefficient k ,k given by (6-30), the joint PDF of |H(i, k1)| and |H(i, k2)| is
1 2

given by (Appendix C) f|H (i ,k1 )|,|H (i ,k2 )| ( x1 , x2 ) = 2 x1 x2

L 1 2 ) l2 (1 r1,2 l =0
2

| a(i, k ) |2 + | a (i, k ) |2 2r | a (i, k )a(i, k ) | cos( w + ) 1 2 1,2 1 2 1,2 1 2 exp L 1 2 2 (1 r1,2 ) l l =0 2 L 1 2 (1 r1,2 ) l2 l =0 | x | a(i, k ) | e j1 1 1 2 2 2 x1 + x2 2 x1 x2 r1,2 cos( w1,2 ) j 2 j d exp I + x a i k e e | ( , ) | 0 2 2 L 1 0 2 2 j ( w1,2 +1 ) j (1 r1,2 ) l x r a i k e e | ( , ) | 2 l =0 2 1,2 j ( w ) x1r1,2 | a(i, k1 ) | e 1,2 2 |
1 2 1 2

(6-34)

where r1,2 is the envelope of k ,k , w1,2 is the angle of k ,k , 1 is the angle of a(i, k1), 2 is the angle of a(i, k2),. So the correlation between the capacity of subcarrier k1, Ck1 and the capacity of subcarrier k2, Ck2 , is expressed by

E Ci ,k1 Ci ,k2 =

log 2 (1 + SNR x 2 ) log 2 (1 + SNR y 2 ) f|H (i ,k1 )|,|H ( i ,k2 )| ( x, y )dxdy

(6-35)

where the function f|H (i ,k1 )|,|H ( i ,k2 )| () is given by equation (6-34). Then the variance of OFDM
2 system capacity C for Ricean fading channels is finally obtained by substituting above

formulas (6-32), (6-33) and (6-35) into (6-16).

156

6.4 Numerical and Simulation Results

Both numerical method and computer simulation have been employed to investigate the OFDM system capacity. For computer simulation, the number of the subcarriers of the OFDM system, N, is 512. Both Rayleigh fading channel and Ricean fading channel are constructed for simulation, respectively. For the multipath Rayleigh fading channel, Jakes model [65] is applied to construct a Rayleigh fading channel for each path. The decaying factor in equation (6-6) is 0.15. For the multipath Ricean fading channel, the scattering signal in formula (6-26) adopts Jakes model and the decaying factor is also 0.15. The line of sight signal Al in equation (6-26) only exists in the first path, that is, Al = 0, for l = 1, 2,,L-1. In addition, the bandwidth of OFDM system is 20 MHz. The Doppler shift is 100 Hz.

A. OFDM system capacity over the Rayleigh fading channel Fig 6.1 depicts The PDF of the capacity at a subcarrier, fCi ,k ( x) , for SNR = 0 dB, 5 dB, 10 dB, and 20 dB, respectively. The PDF of Ci,k can be derived from (6-13) and it is given by
fCi ,k ( x) = 1 2x ln 2 2 x exp SNR SNR .

(6-36)

Observe that the PDF for SNR = 0 dB decreases monotonically and the PDFs for SNR = 10 dB and SNR = 20 dB have their maximum values of 0.282 when x = 3.3 and 0.257 when x = 6.6, respectively. The values of PDFs at x = 0 are nonzero and the PDF value for

157

x = 0 is decreased when SNR is increased. Fig 6.2 depicts the joint PDF of of |H(i, k1)|2 and |H(i, k2)|2, f|H (i ,k )|2 ,|H (i ,k
1 2 2 )|

( x, y ) ,for the coefficient of equation (6-23), = 0.61. Observe

that the joint PDF decrease rapidly from 0.3 to nearly zero when increasing x and y from 0 to 3. Fig 6.3 depicts the coefficient of equation (6-24), , versus different subcarrier gap between k1 and k2. The CP length is 64, the number of paths is 8, and the delay of each path is uniformly distributed over the CP length. Observe that the coefficient varies periodically from 0 to 0.28 and the period is 64. Since the coefficient is composed of a few is cosine
min

waves
i

and

the

minimum

frequency

among

the

cosine

waves

i , j = 0,1,..., L 1

j ) / N = 8 / 512 , the frequency of is equal to the minimum frequency

among the cosine waves. That is, the period of is 512/8 = 64, in unit of sample point. Fig 6.4 depicts the variance of OFDM system capacity for the number of channel paths L = 2, 4, and 8, by computer simulation and numerical method. The CP length is 64 and the delay of each path is uniformly distributed over the CP length. The variance of OFDM system capacity by numerical method is calculated by equation (6-16). Observe that the numerical results marginally overlap with the computer simulation and hence are well verified by that of computer simulation. The variance of OFDM system capacity increases with the increase of SNR linearly from 0 dB to 15 dB at a fixed number of channel paths. However, the capacity variance only marginally increases from 20 dB to 25 dB, especially when L is large. Observe that the variance of OFDM system capacity decreases when increasing the number of channel paths L, at a fixed SNR.

158

Fig 6.5 depicts the variance of OFDM system capacity versus the CP of an OFDM symbol in unit of sample point. The number of resolvable paths L is 4. Observe that the variance for SNR = 10 dB increases from 0.54 to 1.69 when increasing the CP from 16 to 128. However, further increase of CP only marginally increases the variance of OFDM system capacity. Similar results are found for the other two cases of SNR = 15 dB and SNR = 20 dB, respectively. Fig 6.6 shows the variance of OFDM system capacity versus the number of subcarriers of one OFDM symbol. The variance is evaluated by computer simulation. The CP length is 64 and the number of resolvable paths for the channel, L = 4. The delay of each path is uniformly distributed over the CP length. Observe that the variance of OFDM system capacity does not vary significantly for the number of subcarriers of one OFDM symbol N = 256, 512, 1024, and 2048.

B. OFDM system capacity over the Ricean fading channel Fig 6.7 depicts the variance of OFDM system capacity over the Ricean fading channel, for the number of resolvable paths L = 2, 4, 8, respectively. The CP length is 64 and the Ricean factor K in (6-28) is 0 dB. The variance of OFDM system capacity by numerical method is calculated by equation (6-16). Observe that the numerical results are well verified by computer simulation results. The variance increases for the increase of SNR and decreases with the increase of the number of resolvable paths L. For instance, for SNR = 15 dB, the variance with L = 2 is 1.46 and the variance with L = 8 is 0.9.

159

C. The comparison of OFDM system capacity between the Rayleigh fading channel and the Ricean fading channel Fig 6.8 shows the mean value of OFDM system capacity for Rayleigh fading channel and Ricean fading channel, calculated by equation (6-14) and (6-32), respectively. The Ricean factor K defined in (6-28) is set to be 0 dB, 10 dB, and , respectively. Observe that the OFDM system capacity increases nearly linearly with the increase of SNR, for both Rayleigh fading channel and Ricean fading channel. The OFDM system capacity with Ricean fading channel is larger than that of Rayleigh fading channel, especially at large Ricean factor K. The system capacity with Ricean fading channel increases with the increase of K until K goes to infinite. Fig 6.9 depicts the variance of OFDM system capacity for Rayleigh fading channel and Ricean fading channel, by numerical method and computer simulation. The numerical results are calculated by equation (6-16). Observe that the numerical results are well verified by simulation results. The capacity variance with Rayleigh fading channel is larger than that of Ricean fading channel. The capacity variance with Ricean fading channel decreases with the increase of Ricean factor K. In addition, the variance with Ricean fading channel increases nearly linearly at low SNR, and increases to an asymptotic level at high SNR. Further increase of SNR only marginally increases the capacity variance. For instance, for the variance curve with Ricean factor K = 10 dB, the variance increases almost linearly from 0.078 to 0.33 when SNR increases from 0 dB to 10 dB. However, for

160

further increasing SNR the capacity variance approaches an asymptotic level of 0.40.

6.5 Conclusion
The variance of OFDM system capacity under the Rayleigh fading channel and the Ricean fading channel with finite paths have been derived and thoroughly investigated. The numerical and simulation results reveal the follows. First, the variance of OFDM system capacity increase almost linearly with the increase of SNR and it decreases with the increase of the multipath number of the channel, for both Rayleigh fading channels and Ricean fading channels. Second, the capacity variance with Rayleigh fading channels increases when increasing the CP length at a fixed multipath number. However, the variance increases to a asymptotical level at a certain value of CP and further increase of CP only marginally increases the system capacity variance. Third, the capacity mean value with Ricean fading channel is larger than that of Rayleigh fading channel, especially at large Ricean factor. The capacity variance with Rayleigh fading channel is larger than that of Ricean fading channel. In addition, the joint probability density function of two arbitrary correlated Ricean random variables has been presented in an integral form.

161

0.7 SNR = SNR = SNR = SNR = 0dB 5 dB 10dB 20dB

0.6

Probability density

0.5

0.4

0.3

0.2

0.1

10

The capacity of a subcarrier, x

Fig 6.1: The PDF of the capacity at a certain subcarrier, fCi ,k ( x) in (6-36), for SNR = 0 dB, 5 dB, 10 dB, and 20 dB, respectively.

0.3 0.2 0.1 0 0 1 2 30 1 2 3

Fig 6.2: The joint PDF of of | H (i, k1 ) |2 and | H (i, k2 ) |2 , f|H (i ,k )|2 ,|H (i ,k
1

2 2 )|

( x, y ) ,for the coefficient

of equation (6-23), = 0.61.

162

1 0.9 0.8 coefficient r 0.7 0.6 0.5 0.4 0.3 0.2 0 100 200 300 |k 1-k 2| 400 500 600

Fig 6.3: The coefficient of equation (6-24), , versus different subcarrier gap between
k1 and k2 .

3.5 The variance of OFDM system capacity 3 2.5 2 1.5 1 0.5 0 0 L= L= L= L= L= L= 2, 4, 8, 2, 4, 8, simulation simulation simulation numerical method numerical method numerical method

10 SNR (dB)

15

20

25

Fig 6.4: The variance of OFDM system capacity for the number of channel paths L = 2, 4, and 8, over the Rayleigh fading channel.

163

3 The variance of OFDM system capacity

2.5

1.5 SNR = SNR = SNR = SNR = SNR = SNR = 10 dB, 15 dB, 20 dB, 10 dB, 15 dB, 20 dB, simulation simulation simulation numerical method numerical method numerical method

0.5

50 100 150 200 The CP length of an OFDM symbol in unit of sample point

Fig 6.5 The variance of OFDM system capacity versus the CP of an OFDM symbol in unit of sample point, over the Rayleigh fading channel.
2.2 The variance of OFDM system capacity 2.1 2 1.9 1.8 1.7 1.6 1.5 1.4 1.3 0 500 1000 1500 2000 The number of subcarriers of an OFDM symbol, N 2500 SNR = 10 dB SNR = 15 dB SNR = 20 dB

Fig 6.6: The variance of OFDM system capacity versus the number of subcarriers of one OFDM symbol, for Rayleigh fading channels.

164

3 L= L= L= L= L= L= 2, 4, 8, 2, 4, 8, Simulation Simulation Simulation Numerical method Numerical method Numerical method

The variance of OFDM system capacity

2.5

1.5

0.5

10 SNR (dB)

15

20

25

Fig 6.7: The variance of OFDM system capacity over Ricean fading channels for L = 2, 4, 8, respectively.
The mean value of OFDM system capacity (bit/s/Hz) 9 8 7 6 5 4 3 2 1 0 Ricean channel, K = 0 dB Ricean channel, K = 10 dB Ricean channel, K = Rayleigh channel

10 SNR (dB)

15

20

25

Fig 6.8: The mean value of OFDM system capacity for Rayleigh fading channel and Ricean fading channel, by numerical method.

165

2.5

The variance of OFDM system capacity

1.5

0.5

Ricean channel, K = 0 dB, Simulation Ricean channel, K = 10 dB, Simulation Ricean channel, K = 20 dB, Simulation Rayleigh channel, Simulation Ricean channel, K = 0 dB, Numerical method Ricean channel, K = 10 dB, Numerical method Ricean channel, K = 20 dB, Numerical method Rayleigh channel, Numerical method

10

15

20

25

SNR (dB)

Fig 6.9: The variance of OFDM system capacity for Rayleigh fading channel and Ricean fading channel, by computer simulation and numerical method.

166

Chapter 7: Conclusions and future works

7.1 Conclusions
The thesis addresses the problems of channel estimation and system capacity for OFDM-based systems. The major conclusions are summarized as follows. 1. A fast LMMSE channel estimation method has been proposed and thoroughly investigated for OFDM systems over slow fading channels. Our proposed method can marginally achieve the same performance with the convention method, in terms of NMSE and BER. The use of the improved MST algorithm and Kumars fast algorithm in the calculation of channel autocorrelation matrix has been proposed, so that the computation complexity can be reduced significantly. 2. The MSE analysis for conventional LMMSE channel estimation method has been presented. And we also provide the MSE for the proposed fast LMMSE channel estimation method, for both the matched SNR and unmatched SNR. 3. To eliminate the effect of ICI due to Doppler shift in fast fading channels, we propose a new pilot pattern and corresponding channel estimation method and data detection for OFDM systems. The proposed pilot pattern consists of two classical pilot patterns, which are the comb-type pilot pattern and the grouped pilot pattern. Computer simulation shows that the proposed channel estimation and data detection based on the proposed pilot pattern can eliminate ICI effect effectively. Compared with the
167

algorithm of [29], the proposed algorithm achieves almost the same BER performance while reducing the number of pilots significantly. 4. The MSE analysis for the MST algorithm based on comb-type pilot pattern has been presented, considering ICI effect. MSE analysis of channel estimation based on the grouped and equi-spaced pilot pattern has also been provided. 5. The closed-loop MIMO-OFDM capacity with imperfect feedback has been derived. A system capacity indicator, namely, the feedback SNR, that reflects the gain of closed-loop capacity over that of open-loop capacity, has been proposed. The lower thresholds of the feedback SNR has been provided and investigated by numerical method. Numerical results show that the lower threshold of feedback SNR is proportional to the number of antennas and also proportional to the MIMO-OFDM system SNR. 6. The variance and mean value of OFDM system capacity over Rayleigh fading channels and Ricean fading channels have been derived and thoroughly investigated. The system capacity variances over Rayleigh and Ricean fading channels have been evaluated by computer simulation and verified by numerical method. The results show that the variance of OFDM system capacity increase almost linearly with the increase of SNR and it decreases with the increase of the multipath number of the channel, for both Rayleigh fading channels and Ricean fading channels. The capacity variance with Rayleigh fading channel is larger than that of Ricean fading channel provided that the SNR, the channel delay, and the non-LOS power decaying factor are the same.

168

7. The joint probability density function of two arbitrary correlated Ricean random variables has been presented in an integral form.

7.2 Future works

Two possible topics for future researches are given by the followings. 1. The proposed channel estimation methods can be further extended to MIMO-OFDM systems. Since the pilot design for MIMO-OFDM systems should satisfy that the pilots for different transmitter antennas should be orthogonal between each other, the channel estimation will be more complicated, considering the orthogonal pilot design. 2. The research on MIMO-OFDM system capacity with imperfect feedback is still not immature, since the proof for the lower thresholds of feedback SNR has not been provided. We will further consider the derivation of lower thresholds of feedback SNR so that the results in Chapter 5 could be verified.

169

APPENDIX A: The derivation of the rank of channel frequency autocorrelation matrix RHH in Chapter 3
In this appendix, we will prove that the rank of RHH is equal to L and the rank
2 is equal to N. We can obtain from (3-7) and (3-9) that of R HH + w

k = RHH (0, n) exp{ j 2 nk / N }

n =0

N 1

= l2 exp{ j 2 l n / N }exp{ j 2 nk / N }
n =0 l =0

N 1 L 1

0, for k 0, for k = L 1 2 = N l , for k N , for k . l =0

(A-1)

where = { l | l = 0,1,..., L 1} , l is the delay of the l-th path, L is the number of resolvable paths. Thus, the number of non-zero eigenvalues of R HH is equal to L. Denote the
2 by k , k = 0,1,..., N 1 . We can obtain that eigenvalues of the matrix R HH + w

[ 0 1

N 1 ]
RHH (0, N 1))]

2 = [FFTN ( RHH (0, 0) + w RHH (0,1) 2 2 = [0 + w 1 + w 2 ]. N 1 + w

(A-2)

2 is N and the rank of Therefore the number of nonzero eigenvalues of the matrix R HH + w 2 is N. the matrix R HH + w

170

APPENDIX B: The derivation of equation (3-20) in Chapter 3

In this appendix, we will show the derivation of (3-20).
1

Since

the

matrix R Hp Hp +

I can be obtained by I is circulant, the inverse matrix R Hp Hp + SNR SNR

SNR I by C and we have

Kumars fast algorithm [22]. Denote the first row of R Hp Hp +

C = [ RH p H p (0, 0) +

SNR

RH p H p (0, 1)

RH p H p (0, N p 1)].

(B-1)

Kumars fast algorithm can be summarized as follows. Step 1: Compute N p points FFT of the vector C and we obtain
D = [d 0 d1

d N p 1 ] = FFTN p (C).

(B-2)

Step 2: E can be obtained from (B-2)

E = [1/ d 0 1/ d1

1/ d N p 1 ].
1

(B-3)

Step 3: Denote the first row of the matrix R Hp Hp + I by F and F can be given SNR
by computing N p points IFFT of the vector E
F = IFFTNp (E).

(B-4)

The above three steps can be combined as

F = IFFTN p (1 [diag{FFTN p (C)}]1 )

(B-5)

where 1 = [1 1
1

1]1 N p and diag{i} denotes diagonalization operation. The matrix

I can be acquired from the 1 by Np vector F by circle shift. Denote the R Hp Hp + SNR

171

 first row of the matrix R Hp Hp R Hp Hp + I SNR

matrix R Hp Hp + I SNR
1

by B , the first column of the

by G . It follows that
N p 1 i =0

B( j ) =

A(i)G((i j ) mod N

), j = 0,1,..., N p 1.

(B-6)

where B(i ) , A(i ) and G (i ) are the i-th elements of the vector B , A and G , respectively. A is the first row of the matrix R Hp Hp . Since G = F H and G (i ) = G * ( N p i ) , where (i)* denote conjugate, (i) H denotes Hermitian transpose, equation (B-6) can be equivalently written as B( j ) = Or equivalently,
N p 1 i =0

A(i) F (( j i) mod N

), j = 0,1,..., N p 1.

(B-7)

B= AF

(B-8)

where denotes circulant convolution, F (i ) is the i-th entry of the vector F . Using the property of DFT, (B-8) can be written as
B = AF = IFFTN p {FFTN p [ A 1 [diag{FFTN p (F )}]1}.

(B-9)

Using equation (3-17), (B-1) and (B-5), equation (B-9) can be further written as
PMST (0) PMST (1) B = IFFTN p P (0) + PMST (1) + MST N p SNR N p SNR PMST ( N p 1) . (B-10) PMST ( N p 1) + N p SNR

172

APPENDIX C: The derivation of the joint PDF of two arbitrary correlated Ricean random variables
In this appendix, we derive the joint PDF of two arbitrary correlated Ricean random variables. Consider two circulant complex Gaussian random variables X1, X2 with mean
2 mX1, mX2 and variance 2 1 , 2 , respectively. The complex random variables X1, X2 can be

further written as X1 = A1 + jB1, X2 = A2 + jB2 The covariances between Ai and Bj are as follows. Cov (A1, B1) = Cov (A2, B2) = 0 Cov (A1, B2) = - Cov (A2, B1) = u212/2 Cov (A1, A2) = Cov (B1, B2) = u112/2 The correlation coefficient between x1 and x2 is given by (C-2) (C-1)

* E ( X 1 m X 1 ) ( X 2 m X 2 ) = u + ju = 1 2

1 2

(C-3)

where ()* denotes conjugate. Then, the joint PDF for A1, A2, B1, and B2 is expressed as [72]

173

f A1 , A2 , B1 , B2 (a1 , a2 , b1 , b2 ) =

1 2 2 ) 12 2 (1 u u2
2 2 1

(a1 ma1 ) 2 / 12 2 2 + (a2 ma 2 ) / 2 + (b m ) 2 / 2 1 1 b1 1 exp 2u1 (a1 ma1 )(a2 ma 2 ) 2 2 (1 u1 u2 ) +2u (b m )(b m ) 1 1 2 b1 b2 + (b m ) 2 / 2 / 2 b2 2 +2u2 (a1 ma1 )(b2 mb 2 ) 1 2 2 ( )( ) u a m b m 2 2 2 1 1 a b (C-4) where ma1, ma2, mb1, and mb2 are the mean values of A1, A2, B1, and B2, respectively. Let
2 R1 = A12 + B12 , R2 = A22 + B2 , 1 = tan 1 ( A1 / B1 ) , and 2 = tan 1 ( A2 / B2 ) ,

(C-5)

we have the Jacobian determinant of (A1, A2, B1, B2) with respect to (R1, R2, 1, 2) is R1R2. Thus, the joint PDF of the phases and amplitudes, f R1 , R2 ,1 ,2 (r1 , r2 , 1 , 2 ) , is given by
f R1 , R2 ,1 ,2 (r1 , r2 , 1 , 2 ) = r1r2 f A1 , A2 , B1 , B2 (a1 (r1 , r2 , 1 , 2 ), a2 (r1 , r2 , 1 , 2 ), b1 (r1 , r2 , 1 , 2 ), b2 (r1 , r2 , 1 , 2 )) 2 2 2 2 2 mX mX 1mX 2 cos( 2 1 ) 1 / 1 + mX 2 / 2 r1r2 1 2 = 2 exp 2 2 2 (1 ) 1 2 1 2
2 r 2 / 12 + r22 / 2 2r1r2 cos(2 1 ) / 1 2 exp 1 1 2 2 2 mX 1r1 / 12 cos(1 1 ) + mX 2 r2 / 2 cos(2 2 ) exp 2 1 r2 mX 1 / 1 2 cos( 2 1 ) r1mX 2 / 1 2 cos(1 + 2 )

(C-6) where mX 1 and mX 2 are the envelopes of mX1 and mX2, 1 and 2 are the angles of mX1, mX2, respectively. Therefore, the joint PDF of two arbitrary correlated Ricean random variables R1, R2 is give by

174

f R1 , R2 (r1 , r2 ) = r1r2

f R1 , R2 ,1 , 2 (r1 , r2 , 1 , 2 )d1 2

2 2 2 2 mX 1 / 1 + m X 2 / 2 2 / 1 2 m X 1m X 2 cos( 2 1 ) = 2 exp 2 2 2 2 (1 ) 1 2 1 2 r 2 / 12 + r22 / 2 2r1r2 cos(2 1 ) / 1 2 exp 1 1 2 2 2 mX 1r1 / 12 cos(1 1 ) + mX 2 r2 / 2 cos( 2 2 ) exp d12 2 1 + / cos( ) / cos( ) r m r m 2 X1 1 2 2 1 1 X2 1 2 1 2

2 2 2 2 mX 2r1r2 1 / 1 + m X 2 / 2 2 / 1 2 m X 1m X 2 cos( 2 1 ) exp 2 2 2 2 (1 ) 1 2 1 2 2 r 2 / 12 + r22 / 2 2r1r2 cos( ) / 1 2 exp 1 0 1 2

2 2 I0 r2 mX 2 e j ( 1 ) / 1 2 r1mX 1e j ( 2 ) / 1 2 | d | r1mX 1e j1 / 12 + r2 mX 2 e j ( 2 ) / 2 2 1

(C-7)

175

Appendix D: List of Abbreviations

AWGN BER BLAST CFO CP CSI DAB DFT DOA DSL DVB DVB-T FFT GSM ICI IFFT IMT-Advanced ISI ITU additive white Gaussian noise bit error rate Bell Laboratories Layered Space Time carrier frequency offset cyclic prefix channel state information digital audio broadcasting Discrete Fourier Transform direction of arrival digital subscriber line digital video broadcasting digital video broadcasting terrestrial Fast Fourier Transform Global System for Mobile Communications inter-carrier interference inverse Fast Fourier Transform International Mobile Telecommunication Advanced inter-symbol interference International Telecommunication Union

176

KL LMMSE LOS LS LST LSTC LTE MIMO MIMO-OFDM

Karhunen-Loeve linear minimum mean square error line of sight signal least square Layered Space Time Layered Space Time Code 3GPP Long Term Evolution multiple-input multiple-output multiple-input multiple-output orthogonal frequency

division multiplexing MISO ML MMSE MST NLOS NMSE OFDM OFDMA PSD RV SIC multiple input single output maximum likelihood minimum mean square error most significant taps non line of sight normalized mean square errors Orthogonal Frequency Division Multiplexing Orthogonal Frequency Division Multiple Access power spectrum density random variable successive interference cancellation

177

SINR SNR STBC STC STTC SVD ULA UMTS UWB VA V-BLAST WLAN WiMAX ZF 3G 3GPP

signal to interference and noise ratio signal to noise ratio Space Time Block Code Space Time Code Space Time Trellis Code singular value decomposition uniform linear array Universal Mobile Telecommunications System Ultra-Wideband Viterbi algorithm Vertical Bell Laboratories Layered Space Time Wireless Local Area Network World Interoperability for Microwave Access zero forcing 3rd Generations 3rd Generation Partnership Project

178

REFERENCES
[1] S. B. Weinstein and P. M. Ebert, Data transmission by frequency-division multiplexing using the discrete Fourier transform, IEEE Trans. Commun., vol. 19, no. 5, pp. 628-634, 1971. [2] A. Goldsmith, S. A. Jafar, N. Jindal, and S. Vishwanath, Capacity Limits of MIMO Channels, IEEE J. Select. Areas Commun., vol. 21, no. 5, pp. 684-702, 2003. [3] Y. Liu and G. B. Giannakis, Ultra-wideband communications: an idea whose time has come, IEEE Signal Processing Magazine, vol. 21, no. 6, pp. 26-54, 2004. [4] J. Mitola and G. Q. Maguire, Cognitive radio: making software radios more personal, IEEE Personal Communications, vol. 6, no. 4, pp. 13-18, 1999. [5] G. G. Raleigh and J. M. Cioffi, Spatio-temporal coding for wireless communication, IEEE Trans. Commun., vol. 46, pp. 357-366, March 1998. [6] G. J. Foschini, Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas, Bell Labs Tech. J., pp. 41-59, 1996. [7] P. W. Wolniansky, G. J. Foschini, G. D. Golden, and R. A. Valenzuela, V-BLAST: An architecture for realizing very high data rates over the rich-scattering wireless channel, ISSSE, Pisa, Italy, 1998. [8] G. J. Foschini, G. D. Golden, R. A. Valenzuela, and P. W. Wolnianski, "Simplified processing for high spectral efficiency wireless communication employing multi-element arrays," IEEE J. Select. Areas Commun., vol. 17, no. 11, pp. 1841-1852,

179

Nov. 1999. [9] S. M. Alamouti, A simple transmit diversity technique for wireless

communications,IEEE J. Sel. Areas Commun., vol. 16, no. 8, pp. 1451-1458, 1998. [10] M. Nakagami, The m-distribution: a general formula of intensity distribution of rapid fading. Statistical Methods in Radio Wave Propagation: Proceedings of a Symposium held in 1958, pp. 3-36, Permagon Press, 1960. [11] A. Peled and A. Ruiz, Frequency domain data transmission using reduced computational complexity algorithms, in Proc. IEEE ICASSP, vol. 5, pp. 964-967, 1980. [12] S. M. Alamouti, V. Tarokh, and P. Poon, Trellis-coded modulation and transmit diversity: design criteria and performance evaluation, in Proc. IEEE ICUPC, vol. 1, pp. 703-707, 1998. [13] David S. W. Hui, Vincent K. N. Lau, and W. H. Lam, Cross layer designs for OFDMA wireless systems with heterogeneous delay requirements, IEEE Trans. Wireless Commun., vol. 6, no. 8, pp. 2872-2880, 2007. [14] S. Coleri, M. Ergen, A. Puri and A. Bahai, Channel estimation techniques based on pilot arrangement in OFDM systems, IEEE Trans. Broadcast., vol. 48, no. 3, pp. 223-229, 2002. [15] M. -H. Hsieh and C. -H. Wei, Channel estimation for OFDM systems based on comb-type pilot arrangement in frequency selective fading channels, IEEE Trans. Consum. Electron., vol. 44, no. 1, pp. 217-225, 1998.

180

[16] Y. H. Zeng, W. H. Lam, and T. S. Ng, Semiblind channel estimation and equalization for MIMO space-time coded OFDM, IEEE Trans. Circuits Syst., vol. 53, no. 2, pp. 463-474, 2006. [17] O. Edfors, M. Sandell, J. J. van de Beek, S. K. Wilson, and P. O. Brjesson, OFDM channel estimation by singular value decomposition, IEEE Trans. Commun., vol. 46, no. 7, pp. 931-939, July 1998. [18] O. Simeone, Y. Bar-Ness, and U. Spagnolini, Pilot-based channel estimation for OFDM systems by tracking the delay subspace, IEEE Trans. Wireless Commun., vol. 3, no. 1, pp. 315-325, 2004. [19] Y. Zhao and A. Huang, A novel channel estimation method for OFDM mobile communication systems based on pilot signals and transform domain processing, in Proc. 47th Vehicular Technology Conf., vol. 3, 1997, pp. 2089-2093. [20] R. Lin and A. P. Petropulu, Linear precoding assisted blind channel estimation for OFDM systems, IEEE Trans. Veh. Technol., vol. 54, no. 4, pp. 983-994, 2005. [21] X. D. Cai and A. N. Akansu, A subspace method for blind channel identification in OFDM systems, in Proc. IEEE International Conference on Communications, New Orleans, LA, July 2000, pp. 929-933. [22] J. -J. van de Beek, O. Edfors, M. Sandell, S. K. Wilson and P. O. Brjesson, On channel estimation in OFDM systems, in Proc. IEEE Vehicular Technology Conf., July 1995, pp. 815-819. [23] Y. Li, L. J. Cimini, and N. R. Sollenberger, Robust channel estimation for OFDM

181

systems with rapid dispersive fading channels, IEEE Trans. Commun., vol. 46, no. 7, pp. 902-915, July 1998. [24] M. Morelli and U. Mengali, A comparison of pilot-aided channel estimation methods for OFDM systems, IEEE Trans. Signal Process., vol. 49, no. 12, pp. 3065-3073, December 2001. [25] S. Coleri, M. Ergen, A. Puri and A. Bahai, A study of channel estimation in OFDM systems, in Proc. 56th IEEE Vehicular Technology Conf., 2002, pp. 894-898. [26] M. J. Fernndez-Getino Garca, J. M. Pez-Borrallo, and S. Zazo, DFT-based channel estimation in 2D-pilot-symbol-aided OFDM wireless systems. In Proc. IEEE Vehicular Technology Conf., 2001, pp. 810-814. [27] A. Bttcher and S. M. Grudsky, Spectral properties of banded Toeplitz matrices, SIAM, 2005. [28] Chi Kuo and Jin-Fu Chang, Equalization and channel estimation for OFDM systems in time-varying multipath channels, in Proc. IEEE International symposium on PIMRC, 2004, vol. 1, pp. 474-478. [29] W. -G. Song and J. -T. Lim, Channel estimation and signal detection for MIMO-OFDM with time varying channels, IEEE Commun. Lett., vol. 10, no. 7, pp. 540-542, July 2006. [30] J. C. Lin, Least-square channel estimation for mobile OFDM communication on time-varying frequency-selective fading channels, IEEE Trans. Veh. Technol., vol. 57, no. 6, pp. 3538-3550, 2008.

182

[31] X. G. Doukopoulos and G. V. Moustakides, Blind adaptive channel estimation in OFDM systems, IEEE Trans. Wireless Commun., vol. 5, no. 7, pp. 1716-1725, 2006. [32] H. Minn and V. K. Bhargava, An investigation into time-domain approach for OFDM channel estimation, IEEE Trans. Broadcast., vol. 46, no. 4, pp. 240-248, 2000. [33] R. Kumar, A fast algorithm for solving a Toeplitz system of equations, IEEE Trans. Acoust. Speech. Signal processing, vol. 33, no. 1, pp. 254-267, 1985. [34] S. Haykin, Adaptive filter theory: fourth edition, Publishing House of Electronics Industry, 2002. [35] COST 207 (under the direction of M. Failly), COST 207: Digital and mobile radio communications, Commission of the European Communities, EUR 12160, pp. 140-145, September 1988. [36] T. S. Rappaport, Wireless communications principles and practice, Publishing House of Electronics Industry, 2002. [37] R. Negi and J. Cioffi, Pilot tone selection for channel estimation in a mobile OFDM systems, IEEE Trans. Consum. Electron., vol. 44, no. 3, pp. 1122-1128, 1998. [38] W. -G. Song and J. -T. Lim, Pilot-symbol aided channel estimation for OFDM with fast fading channels, IEEE Trans. Broadcasting, vol. 49, pp. 398-402, Dec. 2003. [39] K. W. Park and Y. S. Cho, An MIMO-OFDM technique for high-speed mobile channels, IEEE Commun. Lett., vol. 9, no. 7, July 2005. [40] S. Y. Park and C. G. Kang, Performance of pilot-assisted channel estimation for OFDM system under time-varying multi-path Rayleigh fading with frequency offset

183

compensation, in Proc. IEEE Vehicular Technology Conf, 2000, pp. 1245-1248. [41] W. G. Jeon, K. H. Chang, and Y. S. Cho, An equalization technique for orthogonal frequency division multiplexing systems in time-variant multipath channels, IEEE Trans. Commun., vol. 47, pp. 27-32, 1999. [42] T. K. Moon and W. C. Stirling, Mathematical methods and algorithms for signal processing, Prentice Hall, 2000. [43] H. Stark and J. W. Woods, Probability and random processes with applications to signal processing, Prentice Hall, 3rd edition, 2002. [44] J.-W. Choi and Y. -H. Lee, Optimum pilot pattern for channel estimation in OFDM systems, IEEE Trans. Wireless Commun., vol. 4, no. 5, pp. 2083-2088, Sep 2005. [45] M. D. Scrinath, P. K. Rajaseharan, and R. Viswanathan, Introduction to statistical signal processing with applications, Prentice Hall, Englewood Cliffs, 1996. [46] I. E. Telatar, Capacity of Multi-Antenna Gaussian channels, AT&T Bell Laboratories, BL0112170-950615-07TM, 1995. [47] G. J. Foschini and M. J. Gans, On the limits of wireless commnunications in a fading environment when using multiple antennas, in Proc. Personal Wireless Commun 98, vol. 6, pp. 315-335, March 1998. [48] D. Shiu, G. J. Foschini, M. J. Gans, and J. M. Kahn, Fading correlation and its effect on the capacity of multi-element antenna systems, IEEE Trans. Commun., vol. 48, pp. 502-513, 2000. [49] A. Gorokhov, Capacity of multiple-antenna Rayleigh channel with a limited transmit

184

diversity, in Proc. IEEE Int. Symp. on Information Theory, 2000. [50] C.-N. Chuah, D. N. C. Tse, J. M. Kahn, and R. A. Valenzuela, Capacity scaling in MIMO wireless systems under correlated fading, IEEE Trans. Inf. Theory, vol. 48, no. 3, pp. 637-650, 2002. [51] S. L. Loyka, Channel capacity of MIMO architecture using the exponential correlation matrix, IEEE Commnu. Lett., vol. 5, no. 9, pp. 369-371, 2001. [52] M. Chiani, M. Z. Win, and A. Zanella, On the capacity of spatially correlated MIMO Rayleigh-fading channels, IEEE Trans. Inf. Theory, vol. 49, no. 10, 2003. [53] Y. Xiao, IEEE 802.11n: enhancements for higher throughput in wireless LANs, IEEE Trans. Wireless Commun., vol. 12, no. 6, pp. 82-91, 2005. [54] R. Zhang, Y. C. liang, R. Narasimhan, and J. M. Cioffi, Approaching MIMO-OFDM capacity with per-antenna power and rate feedback, IEEE J. Select. Areas Commun., vol. 25, no. 7, pp. 1284-1297, 2007. [55] M. Borgmann and H. Blcskei, On the capacity of noncoherent wideband MIMO-OFDM systems, In Proc. Int. Symp. Inf. Theory (ISIT), pp. 651-655, 2005. [56] P. L. Kafle, A. B. Sesay, and J. McRory, Capacity of MIMO-OFDM systems in spatially correlated indoor fading channels, IET Commun., vol. 1, no. 3, pp. 514-519, 2007. [57] H. Blcskei, D. Gesbert, and A. J. Paulraj, On the capacity of OFDM-based spatial multiplexing systems, IEEE Trans. Commun., vol. 50, no. 2, pp. 225-234, 2002. [58] A. Intarapanich, P. L. Kafle, R. J. Davies, and A. B. Sesay, Effect of tap gain

185

correlation on capacity of OFDM MIMO systems, Electron. Lett., vol. 40, no. 1, pp. 86-88, 2004. [59] T. M. Cover and J. A. Thomas, Elements of Information Theory. New York: Wiley, 1991. [60] J. B. Wang and K. Yao, Capacity scaling in OFDM based spatial multiplexing systems, in Proc. IEEE Vehicular Technology Conf., 2002, pp. 28-32. [61] A. T. James, Distributions of matrix variates and latent roots derived from normal samples, Annals of Mathematical Statistics, pp. 475-501, 1964. [62] B. Hassibi and B. M. Hochwald, How much training is needed in multiple-antenna wireless links?, IEEE Trans. Inf. Theory, vol. 49, no. 4, pp. 951-963, 2003. [63] D. Samardzija and N. Mandayam, Pilot-assisted estimation of MIMO fading channel response and achievable data rates, IEEE Trans. Signal process., vol. 51, no. 11, pp. 2882-2890, March 2003. [64] Doufexi, S. Armour, M. Butler, A. Nix, D. Bull, J. Mcgeehan, and P. Karlsson, A comparison of the HIPERLAN/2 and IEEE 802.11a wireless LAN standards, IEEE Commnu. Mag., pp. 172-180, 2002. [65] A. Clark, P. J. Smith, and D. P. Taylor, Instantaneous capacity of OFDM on

Rayleigh-Fading channels, IEEE Trans. Inf. Theory, vol. 53, no. 1, pp. 355-361, 2007. [66] C. R. N. Athaudage, M. Saito, and J. Evans, Capacity of OFDM systems in Nakagami-m fading channels: the role of channel frequency selectivity, in Proc. IEEE PIMRC, Sep. 2008, pp. 1-4.

186

[67] G. Thomas, OFDM capacity enhancement by selective channel use, in Proc. IEEE MILCOM, Oct. 2009, pp. 1-6. [68] I. Bergel and S. Benedetto, Bounds on the capacity of OFDM under spread frequency selective fading channels, in Proc. IEEE EEEI, Dec. 2008, pp. 755-759. [69] J. Y. Yun, S.-Y. Chung, and Y. H. Lee, Design of ICI canceling codes for OFDM systems based on capacity maximization, IEEE Signal Process. Lett., vol. 14, no. 3, pp. 169-172, 2007. [70] R. K. Mallik, On multivariate Rayleigh and exponential distributions, IEEE Trans. Inf. Theory, vol. 49, no. 6, pp. 1499-1515, 2003. [71] A. A. Abu-Dayya and N. C. Beaulieu, Switched diversity on microcellular Ricean channels, IEEE Trans. Veh. Technol., vol. 43, no. 4, pp. 970-976, 1994. [72] J. R. Mendes and M. D. Yacoub, A general bivariate Ricean model and its statistics, IEEE Trans. Veh. Technol., vol. 56, no. 2, pp. 404-415, 2007. [73] O. Kallenberg, Foundations of Modern Probability, New York: Springer-Verlag, 1997. [74] H. V. Poor, An introduction to signal detection and estimation, Springer Verlag, New York, second Edition, 1994. [75] R. M. Gray, Topelitz and circulant matrices: a review, Foundations and Trends in Communications and Information Theoty, vol. 2, pp. 155-239, 2006. [76] X. Liang, Orthogonal designs with maximal rates, IEEE Trans. Inf. Theory, vol. 49, pp. 2468-2503, 2003. [77] S. Boyd and L. Vandenberghe, Convex optimization, Cambridge University Press,

187

2004. [78] E. K. P. Chong and S. H. Zak, An introduction to optimization, New York: Wiley, 1996. [79] Y. Wang and X. Dong, Frequency-domain channel estimation for SC-FDE in UWB communications, IEEE Trans. Commun., vol. 54, pp. 2155-2163, 2006. [80] Peled A. and Ruiz A., Frequency domain data transmission using reduced computational complexity algorithms, In Proc. IEEE Int. Conf. Acoust., Speech, Signal Processing (ICASSP), pp. 964-967, Denver, CO, 1980. [81] H. Ochiai and H. Imai, On the Distribution of the Peak-to-Average Power Ratio in OFDM Signals, IEEE Trans. Commun., vol. 49, no. 2, pp. 282-289, Feb. 2001. [82] H. Ochiai and H. Imai, Performance of the deliberate clipping with adaptive symbol selection for strictly band-limited OFDM systems, IEEE Journal on Selected Areas in Communications, vol. 18, no. 11, pp.2270-2277, Nov. 2000. [83] D. Wulich and L. Goldfeld, Reduction of peak factor in orthogonal multicarrier modulation by amplitude limiting and coding, IEEE Trans. Commun., vol. 47, no. 1, pp. 18-21, Jan. 1999. [84] O. Edfors, M. Sandell, and V. D. Beek, OFDM channel estimation by singular value decomposition, IEEE Trans. commun., vol. 46, no. 7, pp. 931-939,1998. [85] T. M. Schmidl and D. C. Cox, Robust frequency and timing synchronization for OFDM, IEEE Trans. Commnun., vol. 45, pp. 1613-1621, 1997. [86] M. Morelli and U. Mengali, An improved frequency offset estimator for OFDM applications, IEEE Commun. Lett., vol. 3, pp. 75-77, 1999.

188

[87] J. J. van de Beek, M. Sandell, and P. O. Borjesson, ML estimation of time and frequency offset in OFDM systems, IEEE Trans. Signal Process., vol. 45, pp. 1800-1805, 1997. [88] L. Wen, L. Jianhua, and G. Jun, A new pilot assited frequency synchronization for wireless OFDM systems, In Proc. IEEE Int. Conf. Acoust., Speech, Signal Processing (ICASSP03), 2003. [89] L. Zeng and N. C. Tse, Diversity and multiplexing: A fundamental tradeoff in Multiple-antenna channels, IEEE Trans. Inf. Theory, vol. 49, no. 5, pp. 1073-1096, 2003. [90] H. Yang, G. Li, and L. Tang, Diversity-multiplexing tradeoff performance of linear dispersive codes, IET Commun., vol. 2, no. 10, pp. 1289-1292, 2008. [91] V. Tarokh, H. Jafarkhani, and A. R. Calderbank, Spacetime block codes from orthogonal designs, IEEE Trans. on Inf. Theory, vol. 45, no. 5, pp. 744765, July 1999. [92] K. Fazel and S. Kaiser, Multi-Carrier and Spread Spectrum Systems: From OFDM and MC-CDMA to LTE and WiMAX, 2nd Edition, John Wiley & Sons, 2008. [93] H. Ekstrm, A. Furuskr, J. Karlsson, M. Meyer, S. Parkvall, J. Torsner, and M. Wahlqvist, Technical Solutions for the 3G Long-Term Evolution, IEEE Commun. Mag., vol. 44, no. 3, pp. 3845, 2006. [94] James W. Cooley and John W. Tukey, "An algorithm for the machine calculation of complex Fourier series," Math. Comput. vol. 19, pp. 297301, 1965.

189

[95] R. W. Chang, Synthesis of band-limited orthogonal signals for multi-channel data transmission, Bell System Technical Journal, vol. 46, pp. 1775-1796, 1966.

190

Publications
Journal papers: [1] W. Zhou and W. H. Lam, A fast LMMSE channel estimation method for OFDM systems, EURASIP Journal on Wireless Communications and Networking, vol. 2009, Article ID 752895, 13 pages, 2009. [2] W. Zhou and W. H. Lam, Channel Estimation and Data Detection for OFDM systems over Fast Fading and Dispersive Channels , IEEE Transactions on Vehicular Technology, vol. 59, no. 3, pp. 1381-1392, 2010. [3] W. Zhou, X. Y. Liu, and W. H. Lam, Capacity of OFDM systems over time and frequency selective fading channels, submitted to IEEE Transactions on Vehicular Technology. [4] W. Zhou, W. H. Lam, MIMO-OFDM system Capacity with imperfect feedback channel, submitted to IET Communications.

Conference papers: [1] W. Zhou and W. H. Lam, A novel method of Doppler shift estimation for OFDM systems, IEEE Military Communication Conference (MILCOM 2008), pp. 1-7, San Deigo, USA, November 2008. [2] W. Zhou and W. H. Lam, Channel Estimation and Data Detection for OFDM systems over Fast Fading Channels, IEEE International Symposium on Personal, Indoor and Mobile Communications (PIMRC 2009), pp. 3109-3113, Tokyo, Japan, September 2009.

191