Dis Sera Tat Ion

UNIVERSITY OF SOUTHAMPTON Faculty of Engineering, Science and Mathematics School of Electronics and Computer Science
Comparative Study of Different types of Coded Modulation schemes using EXIT chart and BER Characteristics
by Cherian Danny Joseph
23th September 2010
A dissertation submitted in partial fullment of the degree of MSc in Wireless Communication by examination and dissertation
SUPERVISOR: Dr. Robert G . Maunder
Abstract
This report, extensively studies on the topic Coded Modulation(CM) a scheme which combines both the coding as well as the modulation in together to get a bandwidth efcient scheme. This project contributes a genuine comparative study on different Coded Modulation schemes such as Trellis Coded Modulation (TCM), Bit interleaved Coded Modulation (BICM),Turbo Trellis Coded Modulation (TTCM) and Bit Interleaved Coded Modulation with Iterative Decoding (BICM-ID) in context of 8 level Phase Shift Keying(8PSK) over the Gaussian and uncorrelated Rayleigh channels. Here, the comparison are done in terms of decoding complexity, the bandwidth efciency , the coding gain and the frame length by using the study tools such as EXtrinsic Information Transfer Charts(EXIT charts) and BER curve characteristics.
Acknowledgments
Firstly, I would like to thank my supervisor Dr. Robert. G. Maunder for his considerable support and guidance all throughout the project. The tutorial section at the beginning of the project was very useful in understanding the concepts of my project. He was tremendously helpful when ever I faced any technical issues. The weekly meeting arranged helped me to improve the quality of my project a lot. It was a real pleasure working under him. I am also grateful Professor Lajos Hanzo and Dr Lie Liang Yang for the wonderful lectures as well as the suggestions given during the course of Personal Multimedia Communication which provided me a excellent base to start my Project. I would like to thank all my class mates for their valuable supports and various helps throughout the completion of MSc project. Meanwhile, I would also thank University of Southampton who let me study this wonderful course. Finally, I would like to dedicate this report to my parents for their unending love, care and support that they have given me all throughout my life.
ii
Contents
Abstract Acknowledgement List of Figures List of Tables 1 Introduction 1.1 1.2 A Historical Background on Coded Modulation . . . . . . . . . . . . . . . . . Motivation and Organisation of this project . . . . . . . . . . . . . . . . . . . 1.2.1 1.2.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Chapter Organisation . . . . . . . . . . . . . . . . . . . . . . . . . .
i ii vi x 1 2 3 3 4 5 5 7 8 8 10 11 13
2 Background Literature and Review 2.1 2.2 2.3 2.4 2.5 Modulation Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mapping Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-PSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Convolutional Code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 2.6 State and Trellis Diagram . . . . . . . . . . . . . . . . . . . . . . . .
Viterbi Algorithm -(VA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
2.6.1 2.6.2 2.7 2.8
Virterbi Hard Decoding . . . . . . . . . . . . . . . . . . . . . . . . . . Virterbi Soft Decoding . . . . . . . . . . . . . . . . . . . . . . . . . .
13 14 14 15 15 18 20 20 21 23 23 26 27 28 28 31 31 32 35 36 37 38 38 38 40 40 42 44 50
Puncturing Convolution codes . . . . . . . . . . . . . . . . . . . . . . . . . . Trellis Coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.8.1 2.8.2 TCM Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . TCM encoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.9
Turbo Trellis coded Modulation . . . . . . . . . . . . . . . . . . . . . . . . . 2.9.1 2.9.2 TTCM Encoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . TTCM Decoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.10 Bit-interleaved Coded modulation . . . . . . . . . . . . . . . . . . . . . . . . 2.10.1 BICM Encoder and Decoder . . . . . . . . . . . . . . . . . . . . . . . 2.11 Bit-Interleaved Coded Modulation Using Iterative Decoding . . . . . . . . . . 2.11.1 BICM-ID Transmitter and Receiver . . . . . . . . . . . . . . . . . . . 2.12 BCJR Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.12.1 Log-BCJR algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.12.2 Symbol based MAP Algorithm . . . . . . . . . . . . . . . . . . . . . . 2.13 Exit Charts Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.13.1 EXIT chart Generation . . . . . . . . . . . . . . . . . . . . . . . . . . 2.13.2 Area Properties of EXIT chart . . . . . . . . . . . . . . . . . . . . . . 2.13.3 EXIT band charts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.14 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Results And Discussion 3.1 3.2 3.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Study of Individual Coded Modulation Schemes . . . . . . . . . . . . . . . . . 3.3.1 3.3.2 3.3.3 3.3.4 Uncoded QPSK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BICM Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . BICM-ID Simulation Results . . . . . . . . . . . . . . . . . . . . . . TCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
3.3.5 3.4
TTCM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51 57 57 59 59 60 62 62 66 68 68 68 69 69 70 71 71 72 72 73 77 83 83 86 90 91 91
Comparative Study of the Coded Modulation . . . . . . . . . . . . . . . . . . 3.4.1 3.4.2 3.4.3 3.4.4 3.4.5 Performance Over AWGN Channel . . . . . . . . . . . . . . . . . . . Performance over Uncorrelated Narrowband Rayleigh Fading Channels Effect of Block Length on Coded Modulation . . . . . . . . . . . . . . Coding Gain versus Complexity . . . . . . . . . . . . . . . . . . . . . Area/Capacity vs SNR . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5 3.6
Case Study:7 PSK over AWGN channel . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4 Management and Planning 4.1 4.2 4.3 4.4 4.5 4.6 4.7 Initial Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Available Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Project Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Initial and Final Gantt chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . Management Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5 Conclusions and Future Work 5.1 5.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A Publishable Paper B Matlab codes B.1 Original BCJR algorithm developed By Rob . . . . . . . . . . . . . . . . . . . B.2 Symbol based BCJR algorithm for TCM decoders . . . . . . . . . . . . . . . . B.3 2/3 rate Convolutional enocder . . . . . . . . . . . . . . . . . . . . . . . . . . B.4 BICM-ID modulation for 8 PSK using Natural Mapping . . . . . . . . . . . . B.5 Soft Demodulation for 8 PSK . . . . . . . . . . . . . . . . . . . . . . . . . . . v
B.6 EXIT band chart for inner code of BICM-ID(demapper) . . . . . . . . . . . .
93
vi
List of Figures
1.1 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9
Factors affecting the design of Channel/Modulation Scheme [1] . . . . . . . . Modulation and Demodulation of QPSK . . . . . . . . . . . . . . . . . . . .
2 6 6 7 8 9 9 10 12 16 17 18 19 21 22
Constellation Diagram for 4 and 8-PSK . . . . . . . . . . . . . . . . . . . . . DCMC Channel Capacity for different Constellations under AWGN channel . . Bit Mapping relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-PSK constellation diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 7-PSK and 8-PSK capacity compariosn . . . . . . . . . . . . . . . . . . . . . Convolutional Encoder of rate 1/2 . . . . . . . . . . . . . . . . . . . . . . . . State Diagram and One Stage of Trellis . . . . . . . . . . . . . . . . . . . . . Constellation Diagram for 4 and 8-PSK . . . . . . . . . . . . . . . . . . . . .
2.10 8 PSK set partitioning [2],Ungerboeck . . . . . . . . . . . . . . . . . . . . . . 2.11 Encoder and trellis for the eight-state 8PSK . . . . . . . . . . . . . . . . . . . 2.12 RSC encoder and signal mapper forming the TCM encoder . . . . . . . . . . . 2.13 TTCM encoder Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.14 TTCM Decoder Schematic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.15 Paaskes non systematic convolutional encoder, with bit based interleaves and a 8 PSK modulator forming a BICM encoder,employing the Gray Mapping . . . 2.16 BICM decoder . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.17 Paaskes non systematic convolutional encoder . . . . . . . . . . . . . . . . . 2.18 Trellis Diagram for Paaskess eight state convolutional encoder . . . . . . . . . vii
24 24 25 25
2.19 Reciver of BICM-ID with iteration . . . . . . . . . . . . . . . . . . . . . . . . 2.20 A portion of Trellis diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.21 Exit chart of a 8-PSK De-mapper used in BICM-ID employing Natural Mapping 2.22 Scheme for generating the inner code EXIT charts . . . . . . . . . . . . . . . . 2.23 Scheme for generating the outer code EXIT charts . . . . . . . . . . . . . . . . 2.24 Convolutional Decoders EXIT chart for various memory and Generator Polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
27 29 32 33 33
34 35
2.25 Trajectories for iteractive decoding when Eb /N0 = 5 dB . . . . . . . . . . . . . 2.26 Mutal Infomration trajectories and EXIT band chart of TTCM , having a 1000 bit random Interleavers at Eb /N0 = 2.5 dB . . . . . . . . . . . . . . . . . . . . 3.1 3.2 System Overview for different coded Modulation Schemes [1] . . . . . . . . . BER performance of uncoded QPSK for a Frame length of 1000 bits over AWGN channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 BER performance of uncoded QPSK for a Frame length of 1000 bits over Rayleigh channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 BER performance of BICM employing 8 PSK and for various Frame length over AWGN channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.5 BER performance of BICM employing 8 PSK and for various Frame length over Rayleigh channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.6 3.7 3.8 BER performance of BICM for various Constraint length . . . . . . . . . . . . Externisic infomration transfer chart for the outer-code with different memeory Externisic infomration transfer chart for the Demodulater/Demmaper for various SNR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.9 BICM-ID performance using different number of iterations for a frame length of 3000 information bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.10 EXIT band Chart trajectories for BICM-ID at a SNR =4.5 dB for various Frame lengths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
36 39
41
41
42
43 43 45
45
46
47 48
3.11 BER performance comparison for different Frame length under AWGN channel
3.12 BER performance comparison for different Frame length under Rayleigh channel 49 viii
3.13 BER performance comparison for different code memory over AWGN channel 3.14 BER performance comparison for Different frame length under AWGN Channel
49 50
3.15 BER performance comparison for Different frame length under Rayleigh Channel 51 3.16 EXIT band chart for TCM decoder for various interleaving frame length at SNR = 2.5 dB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.17 EXIT band chart for TTCM decoder for various interleaving frame length at SNR = 2.5 dB . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.18 TTCM BER performance using different iterations . . . . . . . . . . . . . . . 3.19 EXIT band chart trajectories for various Frame lengths at SNR=2.5 dB . . . . . 3.20 BER performance comparison TTCM for Different frame length under AWGN Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.21 BER performance comparison for TTCM for Different frame length under Rayleigh Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.22 BER performance of Coded modulation employing 8 PSK and using a Frame length 2000 information bits . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.23 BER performance of Coded modulation employing 8 PSK and using a Frame length 2000 information bits . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.24 Effects of block length on Coded Modulation . . . . . . . . . . . . . . . . . . 3.25 Coding gain at a BER of 104 against the decoding complexity when compared to the uncoded QPSK under AWGN channel . . . . . . . . . . . . . . . . . . . 3.26 Coding gain at a BER of 104 decoding complexity when compared to the uncoded QPSK under Rayleigh channel . . . . . . . . . . . . . . . . . . . . . . . 3.27 Area beneath EXIT function and DCMC Capacity plots in AWGN channel . . 3.28 BER comparison of BICM using 7-PSK and 8-PSK having a frame length 1000 bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.29 EXIT Chart Comparison for de mapper of 7-PSK and 8-PSK at an SNR=0 dB . 3.30 BER comparison of BICM-ID using 7-PSK and 8-PSK having a frame length 1000 bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.31 EXIT Chart Comparison for de mapper of 7-PSK and 8-PSK at an SNR=2.5 dB 64 65 63 63 61 62 61 58 59 58 56 56 53 54 55 52
ix
3.32 BER comparison of TCM using 7-PSK and 8-PSK having a frame length 1500 bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.33 BER comparison of TTCM using 7-PSK and 8-PSK having a frame length 1500 bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 4.2 Purposed Gantt chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Actual Gantt chart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 70 70 65
Chapter
Introduction
Channel Coding or Forward Error Correction(FEC) is the technique developed to combat the effects of the channel impairment and help the receiver in its decision making process.The design of a good FEC depends on the various factors and can be illustrated in the Figure 1.1. It is feasible to design a good coding schemes which are capable of reducing the Bit Error ratio(BER) specically for a given transmission channel. However, this implies there should be further investments in terms of required complexity , the coding/interleaving delays and effective throughput. As time progressed there were different solutions developed for different code features. For example, considering the case of wireless scenario where the power is a important constrain, naturally the power reduction is an extremely important factor. On the other hand, channel coding as well a modulation schemes can be joined together to get a high rate channel coding schemes in collaboration with multilevel/phase modulation schemes. In the project the main objective is to study such a scheme where coding gain can be achieved without any bandwidth expansion and can be named as Coded Modulation. This project studies a variety of coded modulation assisted system and will be investigating their propagation in wireless environments.
1.1. A HISTORICAL BACKGROUND ON CODED MODULATION
Implementational complexity Channel charactristics
Coding/interleaving delay
Bandwidth
Coding/Modulation Scheme
Bit Error Rate
Eective Througput
Coding rate
Coding Gain
Figure 1.1: Factors affecting the design of Channel/Modulation Scheme [1]
1.1 A Historical Background on Coded Modulation

The history of channel coding started with the pioneer works of Shannon [3] in the year 1948. He predicted a reliable communication can be achieved with the aid of the channel coding by adding redundant information bits to the transmitted signal. However, he didnt proposed any channel coding scheme for real time implementation. This gave overall motivation for Hamming [4] and Golay [5], to come with rst practical error control schemes known to be block codes. Later, Convolution codes was introduced by Elias [6] in 1955 and a number of decoding algorithm were developed by Fano [7] and Massey [8]. A breakthrough in convolutional decoding was the invention of the maximum likelihood decoding sequence developed by Viterbi [9]. In 1970s there were successful implementation of convolutional codes in deep space probes. In 1974 a more complex Maximum A Posteriori(MAP) [10] algorithm was developed and was capable of achieving minimum BER. As time progressed and the evolutional of mobile communication system which has both constrains in power and bandwidth-limited scenarios , in 1987 Urgerbock came with a new bandwidth-efcient scheme which employed Set-Partitioning (SP) [11] for signal labelling and this scheme came to be known as Trellis Coded Modulation [2].TCM combined both the convolutional encoder and multilevel/phase signal sets constituting a bandwidth efcient scheme
1.2. MOTIVATION AND ORGANISATION OF THIS PROJECT
which is suitable for mobile/wireless communications [12]. Lately Zehavi [13] and Caire [14]came up with powerful coded modulation scheme which utilised bit based interleaving in conjunction with Gray signal labelling came to be know as Bit-Interleaved Coded Modulation. Another milestone in history of error correcting codes was the invention of Turbo codes [15] by Berrou, Glavieux and Thitimajshina in 1993 which used MAP [10] as its decoding algorithm and was capable of approaching the Shannon limit. As a results this turbo codes were utilised in modern third generation(3G) mobile phones as their Standardised error control scheme [16]. However, Turbo codes had a disadvantage of having a low coding rate and hence the requirement for high bandwidth is inseparable. Therefore, there was many researches going on to nd a better efcient method how the bandwidth could be saved. A higher spectral efcient binary turbo codes codes knows as BICM-bases Turbo Coded Modulation [17] was introduced in 1994. A more recent and better Bandwidth efcient Turbo Trellis coded modulation(TTCM) [1] scheme was introduced which has same properties as of Turbo codes but, by adding new puncturing mechanism the bandwidth utilisation is reduced here. In 1998 , Lie [18] proposed a new iterative joint decoding and demodulation assisted BICM known to be Bit Interleaved Coded Modulation with Iterative decoding(BICM-ID) which uses Signal Partitioning as its signal labelling scheme. The main focus of this project is to study and compare the performance of TCM, BICM, TTCM and BICM-ID schemes with the uncoded QPSK using Gray mapping scheme.
1.2 Motivation and Organisation of this project

1.2.1 Motivation
The radio spectrum available is a scarce resource and they are extremely costly. Therefore, how efcient the bandwidth could be exploited to accommodate the ever-increasing trafc demands is the main question. The coded modulation is capable of achieving the substantial coding gain by expanding the multi-point in the symbol mapping keeping the bandwidth the same. The fundamental objective of the project is to study this novel schemes and trying to evaluate the performance in terms of decoding complexity, the bandwidth efciency , the coding gain and the frame length for all the four coded modulation by the aid of the BER characteristics and
1.2. MOTIVATION AND ORGANISATION OF THIS PROJECT
EXIT chart techniques. Its interesting to note that the EXIT chart have been invented since all of the coded modulation were. So we can use this techniques to to make a good comparison that were not featured in the original coded modulation papers.
1.2.2 Chapter Organisation

Chapter 2: Four different coded modulation such as TCM, BICM, TTCM and BICMID are introduced. Each of these coded modulation is studied separately in terms of the overall coding structure, signal labelling types both Gray and Set-Partitioning, interleaving and mainly the BCJR Log based decoding philosophy. The BCJR algorithm for both the bit based and symbols based algorithm are also highlighted. Chapter 3: The above said coded modulation performance are studied separately when communicating over the AWGN and the narrowband channels with aid of the EXIT charts and BER curves. Extra care has been taken to make the comparative study genuine in terms of the complexity of structure, interteaver length and block length. Chapter 4: This section explains the management and planning section. The main task explained here is the Initial project scope, the available system resource for example softwares used for the simulation, the project tasks which describes the how difcult the given task where, what all Risks that was encountered during these project, the Initial and Final Gantt chart and Finally the management techniques used for the successful completion of this project. Chapter 5: The major ndings are summarised here. Appendix A: A publishable paper in the format of IEEE for this project had been developed and shown here. Appendix B: Important Matlab codes are given here. Having presented an overview of the project, lets discuss on detailed discourse on coded modulation in the coming chapters.
Chapter
Background Literature and Review

2.1 Modulation Scheme
During transmission of the signal, modulation is a process that is used to convey the information. Modulation can be done either in digital or in the analog domain. Modulation are done with the aid of the carrier wave, which is usually high frequency wave to convey the information of much lower frequency input signal.The main types of Digital modulation are Binary Phase Shift keying (BPSK), Frequency shift keying(FSK), Quadrature amplitude Shit keying(QAM) etc. The modulation and demodulation can be expressed in complex notation i. e) in-phase and Quadrature components which are real and imaginary parts of the signals. The complex notation are convenient to represent since in phase and quadrature component parts of the signal behave in the same way as the real and the imaginary parts of the complex numbers. Let the the transmitted signal be x(t) be expressed as shown below. x(t) = xi (t) + j.xq (t) The transmitted signal y(t) is obtained by just taking the real part of complex carrier part (c t)) of the signal is: y(t) = Re{x(t). exp(jc t) } = xi (t).cos(c t) + xq (t).sin(c t)
2.1. MODULATION SCHEME

xi (t) xi (t)
N LO 90
Carrier Generation
y(t)
y (t)
90
xq (t)
xq (t)
a) QPSK Modulation
b) QPSK demodulation
Figure 2.1: Modulation and Demodulation of QPSK
The Figure 2.1 represents the generation and detection of the QPSK modulation and these signals are converted to discrete signal level and are mapped to the in-phase and quadrature phase components by assigning them to the particular points and are known to be constellation digram or constellation patterns.
d Q d Q
I d
d 8-PSK
QPSK
Figure 2.2: Constellation Diagram for 4 and 8-PSK
In this project the fundamental schemes used are of Quadrature Phase Shift Keying (QPSK) and that of 8-Phase Shift Keying. The constellation diagram of the following modulation schemes are given in the Figure 2.2 QPSK can be considered as the special case of the M ary PSK where the phase of the carrier takes on one of the available values, namely i = 2(i1)/M where i = 1, 2, 3, 4 , M . Signalling interval duration for one of the M possible signals are given as
2.2. CAPACITY
si (t) =
( ) 2E 2 cos 2fc t + (i 1) i = 1, 2, 3, 4 , M T M
(2.0)
where E is the energy per symbol and fc is the carrier frequency.
2.2 Capacity
Capacity of the given channel is an upper bound of the achievable throughput which can be express as the Shannons capacity formulae given as C = log2 (1 + SN R)bits/channeluse Bp
Bmax =
(2.0)
Here Equation 2.2 refers to the Gaussian distributed input into the channel and cab be assumed as Continuous-input Continuous-output Memoryless Channel (CCMC) capacity.
AWGN channel Capacity 7 6 5 bits/Channel use 4 3 2 1 0 10 Shannon BPSK QPSK 8PSK 16QAM
5 SNR(dB)
10
15
20
Figure 2.3: DCMC Channel Capacity for different Constellations under AWGN channel
The Discrete-input Continuous-output Memoryless Channel (DCMC) considers the main effect by limiting the channel by log2 M (this shows how many bits per symbols are used for respective modulations) when the channel SNR is increasing where M is the number of points in the constellation diagram.The Channel capacity for different modulation under AWGN can be illustrated in the Figure 2.3.
2.3. MAPPING SCHEMES
2.3 Mapping Schemes

The bit mapping is a procedure which shows, how the bit are mapped to corresponding symbol before the transmission. There are two major classication of bit-mapping methods, they are Gray Mapping and Natural Mapping( set partitioning). In Gray Mapping the bit mapping is done in such a way that the Hamming Distance between each of the adjacent constellation point is one. Gray Mapping was a successful mapping scheme for non-iterative schemes but when an iterative schemes are used then Gray mapping cannot minimise the number of bit error. Therefore when an iterative schemes are used we mainly uses Natural mapping schemes. The illustration of mapping schemes are shown in the Figure 2.4 . Much more detailed analysis of natural Mapping scheme can be seen in the section 2.8.1.
bits
00
Symbols
0
bits
00
Symbols
0
01
01
11
10
10
11
Gray Mapping
Figure 2.4: Bit Mapping relation
Natural Mapping
2.4 7-PSK
The 7-PSK is much similar toward the 8 PSK modulation system except that the 7 -PSK system has a 7 constellation points and a point at the origin. The modulated signal for a 7 PSK can be expressed as follows:
2.4. 7-PSK
Constellation points of 7 PSK 2
Quadrature (AU)
2 2
0 1 Inphase (AU)
Figure 2.5: 7-PSK constellation diagram
Comparison of Capacity Curve between 7 and 8 PSK schemes 3 8PSK 7PSK
2.5
Bits / channel use
1.5
0.5
0 10
5 SNR(dB)
10
15
20
Figure 2.6: 7-PSK and 8-PSK capacity compariosn
2.5. CONVOLUTIONAL CODE
10
si (t) =
( 2E cos 2fc t +
T
2 (i 7
1)
(1
7)
(m = 8) i 7) of the transmitted
where E is the energy in each respective waveforms when 1 signal.
The Constellation points for the 7 PSK is shown FIgure 2.5 here we can see that the conventional difference between 8 PSK as here there will a point in centre of the constellation digram. The 7-PSK can achieve a capacity much higher than that of the 8-PSK and can be be capacitive efcient modulation scheme. The capacity comparison of 8-PSK and 7-PSK is shown in the gure 2.6
V (1)
+ S0
S0 S0
+ S1
V S1 S1
V(2)
Figure 2.7: Convolutional Encoder of rate 1/2
2.5 Convolutional Code

In general , a rate k/n convolution encoder has M-element shift register, k per input information bit and n output coded bits which are given by the linear combination of the content of the register and the input information bits. Generally a convolution encoder of rate 1/n is used, one of the most widely used are the binary convolutional codes .The n generator polynomial is described by the specic connections to the register stage. Upon clocking the shift register output moves to next state and so and so. The generator polynomial are constituted by a binary pattern, indicating the presence or absence
11
of a specic link from a particular shift register stage. For example by referring the gure 2.7 , the generator polynomial is constituted by: [ ] [ ] g1 = [1 0 1] , g2 = [1 1 1]
2.5.1 State and Trellis Diagram

The main technique for characterising the operation of the state machine, as the Convolution encoder (CC) encoder is just to refer the state diagram of the Figure 2.8 . The diagram fully describes the operation of the encoder , how the state is been transferred from one state to the other state and which all path it can choose when its is going from one path to the other. Now, considering the case of the above said convolutional encoder there are two bits in the two shift registers at any time, there are four possible states and these state transition are governed by the incoming bit U . The constraint imposed by the respective encoder restrict them to only two legitimate state transition depending on the nature of the input. Another simple way of representing the encoder is to portray its trellis diagram, which is given in Figure 2.8. On the left side there are four state are portrayed.State transition are governed by the incoming bits and a state transition due to the logical zero is indicated in the gure 2.8
+ + Let the input to the convolutional encoder be denoted as U,next state be S0 , S1 and Output
be V, then the State Transition equation can be written as:

+ S0 = U + S 1 = S0
V1 = U S1
V2 = U S0 S1
Since here the total number of memory elements(m) are 2 there are 2m = 22 = 4 states(S0 , S1 ). These states will change accordingly to the input information word U to get the output(Vi ) and
+ the next states(S0 , S1 ) respectively as demonstrated in the Table 2.1
12
Initial State S0 S1 00 00 01 01 10 10 11 11
Information U 0 1 0 1 0 1 0 1
Final State + + S0 S1 00 10 00 10 01 11 01 11
Output V1 V2 ] 00 11 11 00 10 01 01 10
Table 2.1: State Transition Table
00 0 10 01 11
Figure 2.8: State Diagram and One Stage of Trellis
2.6. VITERBI ALGORITHM -(VA)
13
2.6 Viterbi Algorithm -(VA)

Viterbi Algorithm was developed by Andrew Viterbi [9], it is established that the the algorithm calculates the maximum likelihood code sequence from the received data. The Viterbi algorithm has found its application in number of application such as GSM phones, space probes etc. The Virtebi algorithm can be done either by hard or soft decoding which is explained in later subsections
2.6.1 Virterbi Hard Decoding

The state si (k) denotes a state in trellis in the stage i,Each state Si (k) in the trellis is associated in a state metric or branch metric M (si (k)) and a path in the trellis y k . So in short Viterbi algorithm
k states At time i, the most likely paths per state yi ( the closest to received sequence) will be
eventually coincide at the some time (i-l) . The beginning of Virterbi, the algorithm operates from the zero state, after that will compare the output of the received signal with respect to the encoded sequence of the trellis on the basis of the Hamming distance(HD), its nothing but the number of different bit position between two binary sequence. Now, if the received symbol 10, the associated hamming distance is one with respect to both 00 and 11 of the encoded sequence. During this stage decoder is unable to express any preference to the whether it was 00 or 11 was more. These Hamming distance are known as the context of the Viterbi decoding and known to be the branch metric. Now proceeding to next symbol it will agin compute the hamming distance of all possible four legitimate paths and the received signal. These distance will yield to the new branch metric associated with with second trellis stage. By now the encoded symbol of two original input bits have been reside. Now the obtained branch metric is added to previous branch metric to obtain the path metric . A low Hamming distance can indicates a high similarity between the received sequence and the encoded sequence concerned, which is characteristic of the most likely encoded sequence, since the probability of a high number of error is exponentially decreasing with numbers of error.
2.7. PUNCTURING CONVOLUTION CODES
14
2.6.2 Virterbi Soft Decoding

In the hard decision Viterbi decoding, based on the location of the received coded symbol, the coded bit was estimated if the received symbol is greater than zero, the received coded bit is 1; if the received symbol is less than or equal to zero, the received coded bit is 0. In Soft decision decoding, rather than estimating the coded bit and nding the Hamming distance, the distance between the received symbol and the probable transmitted symbol is found out. This is done as eight level condence for example, scale +4 indicates the highest possible condence concurring the demodulators decision for a binary 1 and -4 for the lowest possible condence. In fact, if the demodulator output -4 , the low condence of in a logical implies a high probability of a binary zero. Bearing this eight level condence scale in min, the received bits can be decoded.
2.7 Puncturing Convolution codes

Puncturing is a process by systematically deleting or not sending, some output bits of a highrate-encoder. Since, the trellis structure of the low rate ermines the same, the rate of information bits per symbol sequence does not change. As a result, the putout sequence belong to the higher rate punctured convolution codes (PC) codes. One of the goals of puncturing is that the same decoder can be used for a verity of high rate codes. One way to achieve decoding of a PC code is using the viterbi algorithm described earlier by the insertion of the deleted symbols in the position that were not send. This process is known to be depuncturing. The deleted symbols are marked by some kind of special ag. The Viterbi Algorithm as mentioned previously works on the maximum likelihood sequence of received data may not guarantee the minimised symbol error therefore, much more decoding algorithm known as BCJR algorithm can be used which will to nd the individual probability of the incoming bit and can be proved much better than VA. The detailed algorithm can seen in the section 2.12.
2.8. TRELLIS CODED MODULATION
15
2.8 Trellis Coded Modulation

The basic idea of Trellis Coded Modulation [2] is that, instead of sending the symbol m formed after the respective modulation is done, a extra parity bit is introduced by doubling the number of constellation points while trying to maintain the same effective throughput. For example if there are two information bits for 4-level Phase Shift Keying(PSK), a parity bit is being introduced by scaling the original constellation points to eight, i.e by making it to 8 PSK. As a result the redundant bit can be absorbed by the expanded constellation diagram, instead of increasing the signalling rate of the system(bandwidth). Ungerboeck [2], in his paper fully describes how to employ the TCM schemes in redundant non-binary modulation(symbol based) with the combination of a nite state Forward Error Correction(FEC) encoder, which selects the coded signal sequence. The extra bits formed by corresponding convolution encoder will restrict the possible state transformation among the consecutive phasor to a certain legitimate constellation. The receiver tries to decode the incoming noisy signal by a trellis based soft-decision maximum-lilkelihod detector and tries to map it to the each of the legitimate phasor sequence by the restrictions imposed by the convolution encoder. The term Trellis is used to describe this scheme is because the overall operation can be described by a corresponding state transition diagram similar to that of binary convolution encoder. The only difference in TCM is that, here trellis branches are labelled with respect to the redundant non-binary modulated phasors.
2.8.1 TCM Principle

The illustration of TCM principle is done by using the example of a eight-state trellis code for 8 PSK, since this is almost simple and will help in better understanding. When a transmission of two bit/symbol by a coded 8 PSK is used, the suggestion is to use a 2/3 rate convolution coder with a Maximum free Hamming Distance(HD) for a respective constraint length [19] and a Gray coding as the mapping scheme. Yet there are problems when this approach is used. As Ungerboeck [2] noted in his paper encoder should be design to maximising the free Euclidean Distance(ED) but, when Gray code mapping is used it does not translate the
16
large HD into larger ED and the permeation of binary output may have signicant inuence on the ED. This is the good time to explain the concepts of Euclidean Distance(ED) and Hamming distance(HD). Euclidian distance in a signal constellation is the distance between different points in the constellation diagram with respect to reference point. In Figure 2.9 do, d1, and d2 represents the minimum squared Euclidian Distance dmin . Just like real numbers have the concept of distance, so do the binary numbers. Lets compare two binary numbers, say 1011 and 0100 , the hamming distance between this numbers is 4. Thus, the Hamming distance is distance obtained by comparing the respective binary number and adding them.
010(2) 2 11 00 11 00
1 01 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 d0 = 2 1 0 1 0 1 0 1 0 1 0 1 0 1 0 111111111 000000000 1 0 1 0 1 0
d0 = 2 0
1 011(3)0 1 100(4) 0 1 101(5)0
1 0
1 0001(1)
d0
d1
d0 = 2sin(/8) d1 = 2 d2 = 2
d2
1 0000(0) 1 0111(7)
1 0
110(6) (B2,B1,B0)
4-PSK Signal Set
8-PSK Signal Set
Figure 2.9: Constellation Diagram for 4 and 8-PSK
Ungerboeck [2], in his paper proposed an new approach know as set partitioning which aims more directly at maximum free ED. The mapping follows a special partitioning of the signal set into subset with increasing the minimum distance 0 signals. This can be illustrated in the Figure 2.10. In short designing the encoder, Ungerboeck summarised the following rules that were to be applied to the assigned channel signals 1. Transmission originating, or merging into any of the same state should receive signals from subset BO or B1 or should have distance of at-least 1 = 2 between them. 2. Parallel Transmission should receive the signals form the subset C0 or C1 or C2 or C3 3. ALL 8 PSK signals are used in trellis diagram with equal probability. 1 2 between the
17
0 = 2sin(/8) = 0.765
B0
SUBSET B1 1 = 1.414
0 C0
1 C1
C2
SUBSET C3
2 = 2
Figure 2.10: 8 PSK set partitioning [2],Ungerboeck
In the Figure 2.11 illustrates encoder and trellis state diagram for 8-state 8PSK. Owing to the limitation imposed by the encoder, there are only a limited set of state transitions associated with the certain phasor sequence is only possible. Now, the above mentioned rules will be much clearer with this example. For example, let the correct path be all zero path and what will be the shortest distance between the two paths which diverges and then remerges, that is given by minimum squared distance free Euclidian distance of the code as seen form trellis in the Figure 2.11
d2 ree = d2 + d2 + d2 1 0 1 f
(2.-2)
where df ree is the minimum distance between any sequence in the trellis and known to be free Euclidian distance.

u2 u1 s+ 2 s2 s2 s+ 1 s1 s1 s+ 0 s0 s0 c2 c1
18
c0
0426 1537 4062 5173 2604 3715 6240 7351
11111111 00000000 11111111 00000000 111111111 000000000 00000000 11111111 1 0 1 0 1 0 1 0 11111111 00000000 11111111 00000000 00000000 11111111 11111111 00000000 11111111 00000000 00000000 11111111 11111111 11111111 00000000 00000000 00000000 11111111 11111111 00000000 00000000 00000000 11111111 11111111 11111111 00000000 11111111 00000000 00000000 11111111 1 0 1 0 1 0 1 0 11111111 00000000 00000000 00000000 11111111 11111111 11111111 00000000 00000000 00000000 11111111 11111111 11111111 00000000 11111111 00000000 00000000 11111111 6 11111111 00000000 00000000 00000000 11111111 11111111 11111111 00000000 00000000 00000000 11111111 11111111 1 0 1 0 1 0 1 0 11111111 00000000 11111111 00000000 00000000 11111111 11111111 00000000 00000000 00000000 11111111 11111111 11111111 00000000 00000000 00000000 11111111 11111111 6 11111111 00000000 11111111 00000000 00000000 11111111 11111111 00000000 00000000 111111111 000000000 00000000 11111111 11111111 1 0 1 0 1 0 1 0 11111111 00000000 00000000 111111111 000000000 00000000 11111111 11111111 11111111 00000000 11111111 00000000 111111111 000000000 00000000 11111111 00000000 11111111 00000000 111111111 000000000 00000000 11111111 11111111 11111111 00000000 111111111 00000000 000000000 00000000 11111111 11111111 7 11111111 00000000 11111111 00000000 111111111 000000000 00000000 11111111 1 0 1 0 1 0 1 0 00000000 111111111 000000000 00000000 11111111 11111111 111111111 11111111 00000000 000000000 00000000 11111111 11111111 00000000 111111111 000000000 00000000 11111111 00000000 111111111 000000000 00000000 11111111 11111111 00000000 111111111 000000000 00000000 11111111 11111111 1 0 1 0 1 0 1 0 111111111 11111111 00000000 000000000 00000000 11111111 111111111 00000000 000000000 00000000 11111111 11111111 00000000 111111111 000000000 00000000 11111111 11111111 11111111 00000000 111111111 000000000 00000000 11111111 11111111 00000000 111111111 11111111 00000000 000000000 00000000 11111111 1 0 1 0 1 0 1 0 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 11111111 11111111 00000000 11111111 00000000 11111111 00000000 11111111 00000000 1 0 1 0 1 0 1 0 11111111 00000000
Ungerbock eight State PSK Code
Figure 2.11: Encoder and trellis for the eight-state 8PSK
2.8.2 TCM encoder

TCM encoder is a specic encoder that is been selected from the family of Recursive Systematic codes [2], which only attaches one parity bit to that of the information bits i.e only m out of m
bits are RSC encoded. Hence, the branches that can diverge from and merge from is 2m , the total number of parallel transmission associated with each of 2m branches are m < m, 2mm .
The Constraint length (K 1) will dene the numbers of shift registers in the encoder. Figure 2.12 shows the generalised 8-PSK TCM encoder, which has a been designed to achieve a high FED over the AWGN channel. Its a systematic encoder which attaches a extra parity bit for the original 2 bit information word. The resultant 3-bit word are mapped to one of the 23 =8 possible point of an 8PSK modulator. The connections between the incoming bit and the modulo-2 adder is what is known to be generator polynomial. The coefcient of these generator polynomial in general is dened as:

8 PSK Mapper
010
19
u2 u1
c2 c1
011
001 000
100 c0 s2 s1 s0 101 110
Tx Symbols
111
Figure 2.12: RSC encoder and signal mapper forming the TCM encoder
j j j j g j (D) = gK .Dk + gK1 .DK + + g1 .D + g0
(2.-2)
j where D represents delay of each one register stages. The coefcient gi takes the value
1 when there is connection while, it will be zero when there is no connection at the specic encoder stage. Hence, the generator polynomial for the Figure 2.12 in the binary format is: g 0 (D) = 1001 g 1 (D) = 0100 g 2 (D) = 0010 Here g 0 represents the feedback polynomial and g j (D) for j 1 is the generator polynomial
associated with the j th information bit. The generator polynomial in octal representation is: [ ] g(D) = 11 08 04 (2.-2)
Let (S2 , S1 , S0 ) denotes the each state of the Convolutional code is shown in the Figure2.11 and the respective cord words(C2 , C1 , C0 ) generated can be seen from the table 2.2 and the Trellis diagram for the eight sate is given Figure 2.11. The conventional TCM decoders are demodulated with the aid of the modied Viterbi Algorithm (VA) [20]. The fundamental principle of VA is maximum likelihood sequence algorithm which may not guarantee the minimised symbol error. Hence, the decoding of TCM will be
2.9. TURBO TRELLIS CODED MODULATION
20
States Information Word (S2 , S2 , S1 00 01 10 11 000 000 010 100 110 001 001 011 101 111 010 000 010 100 110 011 001 011 101 111 100 000 010 100 110 101 001 011 101 111 110 000 010 100 110 111 001 011 101 111 CodeWords=(C2 , C1 , C0 )
Table 2.2: Codeword Table for the 2/3 rate TCM encoder
using a modied MAP algorithm known as Symbol based MAP algorithm this will discussed later in the section 2.12.
2.9 Turbo Trellis coded Modulation

Turbo codes [15] was a major milestone in the forward error correction codes which can even achieve an excellent bit error rates at low SNR. The original proposal was for the BPSK scheme but were soon successful with multilevel coded as well. Robertson [21] soon introduced the concept of the Turbo Trellis Coded Modulation employing two TCM codes as parallel concatenation of two recursive TCM encoder, and adapted puncturing mechanism to avoid the obvious disadvantage of the rate loss.
2.9.1 TTCM Encoder

TTCM encoder are the parallel concatenation of two TCM encoders made of Ungerboeck encoder and a signal mapper as shown in the gure 2.13. The rst TCM encoder normally operates with the original bit sequence (a1 , a2 ), while the second encoder works with the interleaved version of the input bit sequence (b1 , b2 ). Ungerboeck encoder is the same encoder explained in the previous section of TCM, having a code rate of 2/3 and a generator matrix of g(d) = [11 08 04] in octal representation. The 8 -P SK signal Mapper translates the encoded
21
a2 a1 TCM Encoder c d Tx Signal
b2 b1 TCM Encoder
Figure 2.13: TTCM encoder Structure
bits to the complex symbols by aid of the natural Mapping method (SP). Theses complex signal are then alternatively selected, an effective puncturing mechanism that punctures the parity bits. Here,lets explain how selector selects the complex signal coming from the each of the encoder. Let rst encoder produces the symbols c = (0, 2, 4, 5, 1, 6) while, the second encoder output is d = (0, 3, 6, 4, 0, 7) now the Tx signal will be of the symbols (0, 3, 4, 4, 1, 7) i.e each symbols is selected alternatively from each encoders.
2.9.2 TTCM Decoder

TTCM Decoder is much similar to that of binary turbo codes [15], except the difference in the nature of the information passed from one decoder to other decoder respectively and the treatment of the very rst decoding step and schematic of the decoder is shown in the Figure 2.14 Here, the main concern is how the symbol-based non binary TTCM scheme is being done. In symbol-based non binary scheme the systematic bit as well as the parity bits are transmitted together as in the form of complex enveloped symbol and cannot be separated from the extrinsic components, since the noise and the fading that effect the parity components will also affects the corresponding systematic components. Therefore , here in this case the symbol-based information can be spilt into two components: 1. the a-priori component of the non binary symbol provided by the alternative decoders.

y2 e y2 p y2 a y1 e y1 p y1 a a2 e b1
e
22
BCJR Symbol Based Decoder-1
cc
a2 a 1 b2
e
a1 a
a1 e
Input
a b2
a b1 z2 a z2
e
z2 p z1 a
BCJR Symbol Based Decoder-2
dc
z1 e
z1 p
Figure 2.14: TTCM Decoder Schematic
2. the inseparable extrinsic information as well as the systematic components of the nonbinary symbol Now, we can concentrate the working of the TTCM decoder. The received symbols are separated into two different symbol to make sure that upper decoder receives only the symbols encoded by the upper encoder vice versa for the seconder decoder as well and this can be described as the rst step in the decoding. After this, each decoder produces its symbol based probabilities and generates the a-priori and extrinsic information based on Log-Based BCJR algorithms The decoders then provides the corresponding a posteriori (yp ,zp ) which is subtracted with incoming a-priori (aa ,ba ) information to make sure that each of the decoder doesnt receive the same information more than once. The extrinsic information are then interleaved/deinterleaved by the random inderleavers to become the a-priori information and made to iterate between them. During the nal decoding the a posteriori information are de interleaved from the decoder-2 and uses the Hard decision for selecting the maximum a- posteriori probability associated with the information word.
2.10. BIT-INTERLEAVED CODED MODULATION
23
2.10 Bit-interleaved Coded modulation

Bit-interleaved Coded Modulation (BICM) was the idea proposed by Zehavi [13] in order to improve the diversity of the code in Rayleigh channel. The design of the coded modulation schemes are affected by several factors such as High Free Euclidian Distance which is desired for the AWGN channel, while its was interested to note that a high Effective Code Length and a high minimum product distance were the main factors effecting the fading channel. The diversity of the code can be dened as length of the shortest error path [22] and one should be aware that the shortest error distance are not necessarily be the minimum distance error. Unfortunately there was no TCM codes available which can compensate the above said difculties. In order to solve the problem Zehavi came with an idea to render the codes diversity equal to that smallest number of different bits, employing the bit-based interleaving.In short, the bit based interleaves main purpose was to: To maximise the diversity and to disperse the bursty error introduced by the fading channel To render the bit with respect to the Transmitted symbol uncorrelated or independent with each other
2.10.1 BICM Encoder and Decoder

The BICM encoder uses non-systematic eight state code of a rate 2/3 using a free bit-based hamming distance of four [1].The purposed model is illustrated in the gure 2.15. Let a be the encoding sequence, before the encoding of the sequence starts all three shift register are initialised as zero. After the bits are encoded, the each encoded bits will be interleaved by three individual parallel random interleavers of the length equal to each incoming coded bits resulting in a binary vector of C2 , C1 , C0 . These group of three bits are then mapped to the 8-PSK signal set according to that of Gray Mapping. These mapping signals points are just digitally pulse shaped signals then transmitted through the channel. The BICM decoder is implemented as shown in the gure 2.16. At the receiver, the faded noisy signal will be demodulated into six bit metric associated with the three bit position, from each received symbols.
24
a2
S0
b2 b1
Interleaver-1 Interleaver-2 Interleaver-3
c2 c1 c0 Tx Signal
8 PSK Mapper
a1
S2
S1
b0
Figure 2.15: Paaskes non systematic convolutional encoder, with bit based interleaves and a 8 PSK modulator forming a BICM encoder,employing the Gray Mapping
a b2 a2 p a1
p
Deinterleaver-1 Deinterleaver-2 Deinterleaver-3
c2 e c1 e c0 e Rx Signal
BCJR Decoder
a b1 a b0
8 PSK DeMapper
Figure 2.16: BICM decoder
These bit metric are then de-interleaved by the three independent bit de-interleavers to form the estimate code words. Then the BCJR decoder is invoked for selecting the best possible estimate of the original information bits To understand the how the BICM coding is done, we can use the Paaskes eight state convolutional code is used. The generator polynomial for a rate -k/n code there are k generator polynomials, each having n coefcients i.e) gi =(g 0 , g 1 , , g n ), i k will be the generator
polynomial for the ith information bits. The generator polynomial for encoder shown in the gure 2.17 in octal representation is: [ ] [ ] g1 = 4 2 6 g2 = 1 4 7
(2.-2)
A two-bit information namely a2 , a1 are encoded in each cycle in order to form a code word bi where i (0,1 2). The encoder has three shift register and namely S0 , S1 and S2 as given in
+ + + the gure. Let S0 , S1 and S2 represents the next states of the three memory bits. The encoded
code word for the convolutional code is given as :
25
a2
+ S0
S0 S0
b2 b1
a1
+ S2
S2 S2
+ S1
S1 S1
b0
Figure 2.17: Paaskes non systematic convolutional encoder

0 0 00 01 1 1 10 11 2 2
Figure 2.18: Trellis Diagram for Paaskess eight state convolutional encoder
+ S0 = a2 + S1 = S2 + S2 = a1
(2.-1) (2.0) (2.1)
2.11. BIT-INTERLEAVED CODED MODULATION USING ITERATIVE DECODING
26
States Information words (S2 , S2 , S1 00 01 10 11 000 000 110 101 011 001 110 000 011 101 010 101 011 000 110 011 011 101 110 000 100 100 001 010 111 101 010 100 111 001 110 001 111 100 010 111 111 001 010 100 Codeword=(C2 , C1 , C0 )
Table 2.3: Codeword Table (b2, b1 , b0 )for the Paaskess eight state convolutional encoder
b 2 = a 2 S1 b 1 = a 1 S0 b 0 = a 1 S0 S1 S2 a 2
(2.2) (2.3) (2.4)
The eight possible states corresponding to state0 to State7 is shown in the Figure 2.18. The trellis diagram gives the all possible transition of the encoder for the Figure2.17. The Table 2.3 illustrates the all possible combination of the code word for the given set of the information word a2 , a1 with help of the the trellis diagram to denote the next states.
2.11 Bit-Interleaved Coded Modulation Using Iterative Decoding

As noted previously Bit-Interleaved Modulation was purposed to increase the diversity of the Ungerboeck TCM scheme under the Rayleigh channel. Li [23, 18] suggested an new scheme of Bit-Interleaved Coded Modulation using Iterative Decoding which employed Set-Partitioning signal labelling system as that of Ungerboeck. Introduction of soft-decision feedback from the decoders output to the demapper/demodulator input to iterate between them is advanta-
2.11. BIT-INTERLEAVED CODED MODULATION USING ITERATIVE DECODING
27
geous of the fact that, it improves the reliability of the soft information passed to the demapper/demodulator.
2.11.1 BICM-ID Transmitter and Receiver

A BICM-IDs Transmitter is a serial concatenation of convolution encoder, parallel bit interaleaver and a high order modulator. Here, for the sake of making the comparative study more realistic we are using a rate 2/3 code and of a 8 PSK Modulation. The BICM-IDs transmitter is similar to that transmitter explained in the section 2.10.1. BICM-IDs receiver is almost similar to that of BICMs receiver except the fact, the iterative process is used in order to achieve global optimum through a step-by step local search. Iterative decoding is recent success in Forward error correcting codes and can achieve a rate equivalent to that Shannon capacity.
a b2 e b2 a2 a b1
BCJR
1 1
c2 e c2 a
1
1 2
c1 e c1 a
DEMAPPER
Rx Signal
b1
2 a1 a b0 e b0
1 3
c0 e c0 a
Figure 2.19: Reciver of BICM-ID with iteration
The Figure 2.19 shows the receiver of the BICM-ID, the receiver uses a sub optimal, iterative method to calculate the decoded bits. At the initial step , received signal is demodulated and can use equal likelihood assumption to generate bit metrics ce {0, 1, 2} which is interleaved by i corresponding de-interleavers to become the a priori information ba {0, 1, 2} to the Logi based BCJR decoders to generate the a posteriori bit probablities for the information and the coded word. On the second pass the extrinsic, a posteriori vectors be {0, 1, 2} are interleaved as the a i priori information to the demodulators assuming that all the bits are independent of each other
2.12. BCJR ALGORITHM
28
(by a design of a good interleaver) and will again iterate the above said steps until the nal step is reached. The nally ap {0, 1} are decoded by using the hard decision to decode the incoming i message sequence.
2.12 BCJR Algorithm

The BCJR algorithm was developed by L.Bahi, J.Coocke, F Jelinek and J.Raviv in 1974. The algorithm is suitable for exitimating bit/symbol probabilities for a nite-state Markov source transmitted through a discrete memoryless channel. The further details of the BCJR algorithm can be seen from the paper [10].
2.12.1 Log-BCJR algorithm

Lets rst nd how the soft information can be expressed by Logarithmic Likelihood Ratios(LLRs). The LLR of the received data say b is dened as the log ratio of the probabilities of the bit taking two possible values 0 and 1and denoted as L(b). The LLR can be represented as follows: L(b) = ln P (b = 0|yk ) P (b = 1|yk ) (2.4)
where yk is the received signal. As from the equation 2.12.1, if the LLR is positive, that means the probability of logic zero can be higher than the probability of logic zero while, when LLR is negative the probability of logic one is higher. The Log based BCJR algorithm is selected due to following advantages, one is that since the calculation are done in the log domain it can avoid unnecessary numerical overows and second is that multiplication and division in the log domain is addition and substation by the properties of the logarithm. The ultimate aim of the Log-BCJR algorithm is to calculate the extrinsic LLR from the corresponding decoded sequence. The calculation of extrinsic LLR say ye lead to the calculation of the three internal variable such as , and . The (T ) value represents the conditional probabilities that corresponds to each transition in trellis, here in the case of coded modulation schemes, the can be sub grouped into
29
two, a priori transition probabilities y and the channel transition probablities c . The values are the forward recursion corresponding to each state in each step in the respective trellis. The values are the backward recursion corresponding to each state in each step in the respective trellis.
sk1
0 1 2
sk
Q1
Figure 2.20: A portion of Trellis diagram
Finally, the above said three variables can used to calculate probability of specic transition in the trellis and can be used to denote as . The calculation of extrinsic information is y is the joint probability of the corresponding to , and . Now, its time to deal how the calculation of above said variables are dealt with with the help of the Figure 2.20 and diveded into for parts. 1. calcuation: With the aid of the Figure 2.20,the transition probability t (p, q) or the branch metric for the branch Sk1 = p to Sk = q can be calculated as: t (p, q) = P (Sk = q|Sk1 = p) = P (yt = x(p,q) ) where P (yt = x(p,q) ) is the apriori probability of the message symbol yt . (2.4)
30
2. calculation: The value of probabilities are calculating starting from the beginning of the trellis and proceed forward through the trellis. The t (p) for all given states p (0, Q 1) and calculated as :
Q1
k (q) =
p=0
k1 (p)(p, q)
(2.4)
3. Calculation: The value of probabilities are calculated starting from the end and working backward through the trellis. The backward recursion can be done in the similar way as that of . k1 (p) =
Q1
q=0
k (p)(p, q)
(2.4)
4. y calucation: The value of y can be calculated as: y =

(p,q)Sx
k (q).t (p, q).k1 (q)
(2.4)
5. Finally, The extrinsic LLRs of the uncoded bits are calculated with aid of Jacobin logarithm. The Jacobian logarithm can be dened as: f (1 , 2 ) = ln(e1 + e2 ) = max{1 , 2 } + ln(1 + e|1 2 | ) = max (1 , 2 ) (2.5) (2.6) (2.7)
In actual practice ln(1 + e|1 2 | ) is implemented by the a aid of Look-Up Table. This version of Log Bases BCJR which is realised by looking the Look-Up-Table is known to be Approx-Log-Bases BCJR algorithm.
2.13. EXIT CHARTS ANALYSIS
31
2.12.2 Symbol based MAP Algorithm

Till, now we have just seen how binary based MAP algorithm works. This section will explain working of non-binary based MAP algorithm or symbol bases MAP decoding algorithm [21]. The algorithm works just same as that of binary MAP algorithm except in the face that it differs its calculation of (T ). As noted previously, can be sub grouped into two a priori transition probabilities y and channel transition probablities c . Here, the main difference in calculation of non-binary based MAP is the calculation of the channel transition probabilities, the calculation of the c is as follows: Lets consider the received noisy signal be yk and transmitted noiseless signal be xk then the channel transition probablities c is given as: c (T ) = |yk sk |2 2 2n (2.7)
where 2 = N0 /2 is the noise variance , and N0 is the noises Power Spectral Density(PSD) The priori transition probabilities y are added with obtained channel transition probablities c in order to get the nal (T ). One thing we should note here is that the information symbols in most cases are independent and eqi probable. However, if the information symbols have some priori knowledge about the incoming symbols, this can used as the a priori probabilities. These a priori probabilities used to improve after each iteration until they converge. Rest all the calculation of , beta and are much similar to binary MAP algorithm. The non-binary MAP algorithm can also be evaluated in the same logarithmic domain to reduce the multiplicative complexity especially the overow and to mitigate the computational complexity. The transformation of non-binary MAP algorithm are same as explained above.
2.13 Exit Charts Analysis

EXtrinsic Information Transfer(EXIT) [24] chart analysis is powerful that is used to check the convergence of the iterated decoders. BER chart is one of most powerful tool to analysis the performance how good the decoder is, but it was not able to explain in detail how the decoder
32
converges when an iterative decoding is done. EXIT chart measures the Mutual Information (MI) that is exchanged between the constituent decoder in a iterative process. The EXIT chart is expressed in terms of the Log Likelihood ratio of both apriori information Ia and extrinsic information Ie .
Exit Chart Generation For BICMID Demapper 1 0.9 0.8 0.7 0.6 Ie 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 Ia 0.6 SNR =6 dB SNR =5 dB SNR =4 dB SNR =3 dB SNR=2 dB SNR=1dB SNR =0 dB 0.7 0.8 0.9 1 Increase in SNR
Figure 2.21: Exit chart of a 8-PSK De-mapper used in BICM-ID employing Natural Mapping
To elaborate the exit chart lets, consider the BICM-IDs demaper EXIT curve plot. EXIT curve are obtained by as a function of Ia and Ib for a various ranges of SNR. Note that when Ia = 0 it correspond to the rst iteration of decoding when all the a priori values are equal to zero the extrinsic information are almost less reliable an can results in high probability of decoding errors. In contrast, as the a priori increases the extrinsic information are getting more reliable and condent. Another important aspect is that as SNR values are getting strong, however irrespective of a priori information, the extrinsic are getting more and more reliable.
2.13.1 EXIT chart Generation

Figure 2.22 and 2.23 explains schematics by how the EXIT chart simulation are carried out. Figure 2.22 shows how the EXIT charts are being obtained for inner code/demmaper. As mentioned above BICM-ID is using an serial Concatenated codes, which is having an outer convolutional encoder and inner 8-PSK demmaper. Figure 2.22 demonstrates how the EXIT chart curve for the inner code.
33
Generate N Random Bits
Inner Code/ Mapper
Channel
Ia
Generate LLRs Measure MI
aa
Inner Decoder/ Demapper
bc
Ie
ae
Figure 2.22: Scheme for generating the inner code EXIT charts
a
Generate N Bits Outer Code
bc
Ia ba
Generate LLRs
Outer Decoder
be
Measure MI
Ie
Figure 2.23: Scheme for generating the outer code EXIT charts
34
Usually rst step in EXIT generation is to generate N random bits and to form vector of binary input say a. After that it will be passed though either inner code or outer code which is needed accordingly. If we have to genearte the EXIT chart for Inner code then output of the inner code b are passed through the channel having a xed SNR. The incoming coded signal are decoded and made to iterate for a known predened a priori information ba and will be measuring the mutual information for the respective extrinsic information. In the similar manner we will be generating the EXIT chart for the outer code , but the difference is that the here encoded bits will not send through the channel rather it will be producing the a priori informations from the encoded outer code itself . As in the case of Inner code this will iterated for the known values of the a priori and measures the mutual information . The gure 2.21 shows the EXIT chart curve for the de-mapper, the overall performance can be explained by the two values: one is Ia =0 known as no -priori information and other is known to be perfect a-priori information (Ia =1). Different Eb /N0 (dB) values causes the line to shift up and down.
Comparison of Effect of Constraint Length 1 0.9 0.8 0.7 0.6 Ie 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 Ia 0.6 0.7 0.8 0.9 1 K=7 G=[103,30,66] K=4 G=[11, 02, 04] K=5 G=[23, 04 ,16] K=6 G=[45, 26, 34]
Figure 2.24: Convolutional Decoders EXIT chart for various memory and Generator Polynomials
The outer codes EXIT function is obtained as shown in the gure 2.23 and here when considering the case of BICM-ID, the outer code is a convolutional encoder. The gure show 2.24 convolutional encoders EXIT chart of rate 2/3 having different memory and its interesting
35
to note that the curve cross at the single point (0.5,0.66). Thus, for an arbitrary code rates R the point turns out to be at (0.5,R)[10]. Figure 2.25 shows the extrinsic iterative decoding trajectory that are connected through the interleavers, the extrinsic outputs of the demapper becomes the the a priori information to the decoders and the extrinsic output of the decoder become becomes the a priori inputs to demapper. This exchange of the extrinsic information is what is known to be Extrinsic information transfer chart (EXIT chart) and can be plotted into a single diagram by combing both the demapper and decoder transfer characteristics.
Trajecteries for iterative decoding at Eb/N0=5dB 1 0.9 0.8 0.7 0.6 I(ae) 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 I(aa) 0.6 0.7 0.8 0.9 1
Figure 2.25: Trajectories for iteractive decoding when Eb /N0 = 5 dB
2.13.2 Area Properties of EXIT chart

A relevant property of the EXIT chart is its area beneath the EXIT function [25]. Lets denote the Areaouter for the outer code and Areainner for the inner code then, Areaouter beneath EXIT chart function for the APP Outer Decoder having a rate of Router is given by: Aouter = Router . (2.7)
36
The effective throughput in bits of the source information per channel symbols can be expressed as: < Aouter Rinner log2 (M ) where M is the number of constellation points . Now, when comsidering the case where Router = 1 then we can show that Ainner Rinner log2 (M ) (2.7)
(2.7)
Its mandatory to keep the EXIT chart tunnel to be opened for achieving iterative decoding convergence. EXIT chart tunnel can be only be created if the area beneath the EXIT outer Aouter function is less than the inner EXIT function Ainner , thus we have Aouter < Ainner . So in general <C (2.7)
2.13.3 EXIT band charts

Exit Band Chart 1 0.9 avg(I ) 0.8 0.7 0.6 IE 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 IA 0.6 0.7 0.8 0.9 1 avg(IE ) std(IE )
2 2
avg(IE ) std(IE )
1 1
avg(IE )
2
Figure 2.26: Mutal Infomration trajectories and EXIT band chart of TTCM , having a 1000 bit random Interleavers at Eb /N0 = 2.5 dB
EXIT band chart describes the generalised mutual information transfer characteristic and
2.14. CONCLUSION
37
the probabilistic convergence behaviour of the iterative decoding [26]. The mutual information between the information bit and the corresponding extrinsic information are obtained in the same manner how the EXIT charts are calculated. The difference between the EXIT band chart and EXIT chart is that here the simulation are kept running for various channel realisation and the interleavers used here. Thus we can obtain the various mutual information of the output extrinsic information from the constituent decoders. Then, for each Eb /No we can represent the output extrinsic information as the average extrinsic information (avg(IE ) )and as (avg(IE std IE )) The obtained EXIT band chart with the trajectories can be illustrated in the Figure 2.26. From the Figure 2.26 its quite evident that trajectories for different channel realisation t the EXIT band chart quite well. The width of the std deviation curve depends on the information frame length, as Frame length increases the EXIT band chart shrink to the ordinary EXIT charts.
2.14 Conclusion
This chapter outlines the main principles of TCM, TTCM, BICM and BICM-ID. The conceptional differences of theses coded modulation was studied in terms of the signalling labellingphilosophy, coding structure and decoding philosophy as well. The BCJR algorithm for both the bit based as well as the symbols based was studied separately here. The concepts of EXIT chart analysis is introduced here and ways how to generate this EXIT chart is also outlined here.
Chapter
Results And Discussion

3.1 Introduction
Having completed the background literature review on the principles of TCM, TTCM, BICM, BICM-ID, this section will evaluate the performance of each of above said coded modulation for the transmission over AWGN and the narrow band fading channels. The simulation carried out in this section will be using a 2/3 rate encoder and a 8 PSK Signal Mapping of various memory length. Section3.3 will be studying the individual coded modulation schemes. Section 3.4 will be a comparative study of coded modulation and just evaluating which one is the best scheme. Finally, a case study is been done using the 7-PSK which is a capacity efcient modulation when compared to 8-PSK.
3.2 System Overview

The schematic for the coded modulation is illustrated in the Figure 3.1. The source here will be producing some random information bits, which is then encoded by the one of the respective encoders and consecutively interleaved by random-interleavers . The interleaved bits/symbols are then modulated according to symbol mapping rule for each of corresponding modulation schemes.The channel discussed here for the coded modulation schemes are that of the AWGN and Rayleigh distributed at fading. The relationship between the AWGN and Rayleigh fading
38
3.2. SYSTEM OVERVIEW

TCM
39
Source
TCM /TTCM BICM Enocder
Interleaver
8 PSK Mapper
Channel BICMID Interleaver
Destination
TCM/ TTCM BICM Decoder
Deinterleaver
8 PSK Demapper
TCM
Figure 3.1: System Overview for different coded Modulation Schemes [1]
channel can be expressed as:
yt = t xt + nt
(3.0)
where xt is the transmitted discrete signal and yt is received signal. t is the Rayleigh2 distrubuted fading having an expected squared value of E(t )=1 and nt is the complex AWGN
having a noise variance of N0 /2 per dimension. For an AWGN channel t = 1. The receiver side consist of demodulator/demaper followed by a de-interleaver and will be having one of the TCM, TTCM or BICM decoders which is explained in the previous chapter. When considering the BICM-ID schemes the decoder output is interleaved and then feedback to the demmaper input as illustrated in the Figure 3.1. The study of coded modulation schemes are with Ungerbocks TCM, Robersons TTCM, Zehavis BICM and Lis BICM-ID. To make this little more elaborate the following generator polynomials are used to study and stimulate the performance of the TCM/ TTCM schemes are given in the upper part Table 3.1. These are RSC codes which will be adding one parity bits to information bits. Hence, the corresponding coding rate for a 2m+1 ary PSK is R =
m . m+1
The
3.3. STUDY OF INDIVIDUAL CODED MODULATION SCHEMES
40
number of decoding staged associated with the constraint length K is 2K1 . Lower part of the Table 3.1 shows the respective generator polynomial for the BICM/BICMID schemes. As noted previously they are non-systematic codes having maximum hamming distance. Here, as TCM/TTCM only one parity bit is added to the information bits and hence, the coding rate and bandwidth efciency remains the same as that of the TCM/TTCM. Decoders utilises the Log based- Maximum A Posteriori (Log-MAP) algorithm for its soft decision. Log-MAP algorithm was chooses since, they are the numerical stable version and to reduce the complexity as well as the numerical problem. Scheme TCM/TTCM State 8 16 32 64 8 16 BICM/BICM-ID 32 64 g1 11 23 45 103 04 01 07 02 14 03 15 06 g2 02 04 16 30 02 04 01 05 06 10 06 15 g3 04 16 34 66 06 07 04 07 16 17 15 17
Table 3.1: Generator polynomial for Ungerbock TCM coded with best maximum distance [2] as well as the Paaskes [19] generator polynomials for BICM codes. Both the codes are expressed in octal representation form.
3.3 Study of Individual Coded Modulation Schemes

3.3.1 Uncoded QPSK
The simulation results for the uncoded QPSK under the AWGN and Rayleigh channel are presented here. This is used as reference to compare the performance of the coded modulation scheme in terms of the coding gains. The Figure3.2 and Figure 3.3 just portrays the performance of the uncoded QPSK over the AWGN and Rayleigh channel.
41
QPSK performance under AWGN Channel 10

1
10
10 BER 10
10
10
5 Eb/No(dB)
10
Figure 3.2: BER performance of uncoded QPSK for a Frame length of 1000 bits over AWGN channel
10
0
BER Performance Under Rayleigh Channel
10
10 BER 10
10
10
10
15
20 Eb/No(dB)
25
30
35
40
Figure 3.3: BER performance of uncoded QPSK for a Frame length of 1000 bits over Rayleigh channel
42
3.3.2 BICM Simulation results

The simulation results for BICM with 8-PSK and Gray mapping is employed here. The interleavers are of 3 parallel independent bit random interleavers .The Figure 3.4 and Figure 3.5 gives the BER performance of BICM under AWGN and Rayleigh Channel. The simulation for both the channels are run separately for 3 parallel 100 bits and 3 parallel 1000 bits interleavers. The length of the frame length does not depends on the performance and is almost equal to the BER plot for the low Eb /No but as Eb /No goes on increasing the there is slight variation in the overall performance. At 104 BER there is almost a 0.8 dB gain for the 3 parallel 1000 bits interleavers. So we can conclude that the BICM performance does not signicantly depend on the overall frame length. It is almost free from turbo cliff(the region where the BER curve starts to fall abruptly). The Figure 3.6 shows the effect of constraint length on the BER characteristics and its much evident that the higher the code memory the performance will be starting on decaying rst but as the Eb /N0 increases the the performance goes on improving.
10
0
BER Performance of BICM under AWGN Channel 3 Parrellel 1000 bit Interleaver 3 Parrellel 100 bit Interleaver
10
10 BER 10
10
5 Eb/No(dB)
10
Figure 3.4: BER performance of BICM employing 8 PSK and for various Frame length over AWGN channel
43
10
BER Performance of BICM under Rayleigh Channel 3 Parrellel 100 Bit Interleaver 3 Parrellel 1000 Bit Interleaver
10
10 BER 10
10
10
8 Eb/No(dB)
10
12
14
16
Figure 3.5: BER performance of BICM employing 8 PSK and for various Frame length over Rayleigh channel
10
Effect of Code Memory (Constraint length) Code memory=3 Code memory=4 Code memory=5 Code memory=6
10
10
10 BER
10
10
10
10
5 6 Eb/No(dB)
10
11
Figure 3.6: BER performance of BICM for various Constraint length
44
3.3.3 BICM-ID Simulation Results

The simulation are carried out in the same way as that of the BICM except the fact that here Signal Partitioning labelling system is used to order to cop up with the Iterative decoding as Gray mapping doesnt inuence the iterative decoding. Figure 3.7 show the transfer characteristic for BICM-ID outer decoder with different memory elements. The simulations are carried for the outer code EXIT function as mentioned in previous chapter. Its remarkable that all the points converges to same points (0.5,0.67). The area beneath the EXIT function for a optimum outer decoder as noted previously should match to the coding rate of the system i.e, 2/3 is the coding rate of the system used here. If we refer to the Figure 3.7 the area beneath the EXIT function for different memory gives the same Areaouter equivalent to the coding rate. This shows the condence of my simulation is correct. The Figure 3.8 gives the inner codes EXIT chart. In the case of the BICM-ID the outer code is serially concatenated with inner code which is the demodulater/demapper .Maintaining an open tunnel is a necessary condition in order to achieve iterative decoding convergence for an innitesimally low probability of error. An open EXIT chart can only be created if the Areaouter is less than that of Areainner . Figure 3.8 shows the performance of the demodulators EXIT chat analysis for various SNR values, and its much evident that as the SNR values increases the the curve of the respective SNR values curve just moves up as the a priori information gets more an more stronger.
45
Comparison of Effect of Constraint Length 1 0.9 0.8 0.7 0.6 Ie 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 Ia 0.6 0.7 0.8 0.9 1 K=7 & Area beneath the Exit Function= 0.6683 K=4 & Area beneath the Exit Function= 0.66643 K=5 & Area beneath the Exit Function= 0.6588 K=6 & Area beneath the Exit Function= 0.66566
Figure 3.7: Externisic infomration transfer chart for the outer-code with different memeory
EXIT Chart for 8 PSK Demapper 1 0.9 0.8 0.7 0.6 Ie 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 Ia 0.6 0.7 SNR =6 dB SNR =5 dB SNR =4 dB SNR =3 dB SNR=2 dB SNR=1dB SNR =0 dB 0.8 0.9 1
Figure 3.8: Externisic infomration transfer chart for the Demodulater/Demmaper for various SNR
46
In gure 3.9, shows the effect of iteration and how much iteration is required to converge to Error Free feedback EFF (just an Analytic bound to make the comparison easy). Simulation are carried out for 3000 information bits, and the results after three iteration/pass is found to be converging to EFF. More, iteration does not not signicantly changes the performance of the BER curve i.e) as it can be seen from the Figure 3.9 after three iterations the improvement that we get is only a gain of about 0.1 dB. The point where the BER curve starts decreasing is known to be threshold Eb /N0 and the threshold Eb /N0 for the BICM-ID is about 3.5 dB. However and interesting phenomena know as error oor we can observe here due to presence of low weight code words can be at an Eb /N0 = 4.5 dB.
10
0
BICMID and the effects of Iterations 1 iterations 2 iterations 4 iterations 6 iterations 8 iterations 16 iterations
10
10 BER
10
10
10
Error Free Feedback bound region
10
6 Eb/N0(dB)
10
12
Figure 3.9: BICM-ID performance using different number of iterations for a frame length of 3000 information bits
The convergence between the demmaper and decoder can be explained easier with the help of the EXIT band chart trajectories, as these trajectories will explain how easily the mutual information are passed between to and for between each decoder until they converges. The gure 3.10 shows the comparison of EXIT band chart trajectories for a 3 parallel bit random interleavers of length of 3000 and 300 at an SNR =4.5 dB. The number of iteration required for the convergence of the decoders is only three, which is clearly evident from the Figure 3.10. Exit band graph width widen as well as become narrow according to increase in the frame length.
47
BICMID 3000 Frames/Bit 1 0.9 0.8 0.7 0.6 I(ae) 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 I(aa) BICMID 300 Frames/Bit 1 0.9 0.8 0.7 0.6 I(ae) 0.5 0.4 0.3 0.2 0.1 0 0 0.2 0.4 I(aa) 0.6 0.8 1 0.6 0.8 1
Figure 3.10: EXIT band Chart trajectories for BICM-ID at a SNR =4.5 dB for various Frame lengths
48
Figure 3.11 and Figure 3.12 shows the effect of the Frame length on the performance of the code over both AWGN and Rayleigh Channel. It can be seen from the gure that performance of the BICM-ID code is very remarkable for a system with the longer frame length. However, even for short frame length it can give good results but suffers a overall degradation of about 2 db at a BER of 104 for both AWGN and for narrow band at fading channels. Note that larger block length can lead to the earlier convergence, in term of low SNR and number of iterations used but, at higher SNR values the effect of the frame length is insignicant.
10
0
AWGN Channel 3 Parrellel Interleavers of Length 100 3 Parrellel Interleavers of Length 1000
10
10 BER
10
10
10
10
4 5 E /N (dB)
B 0
Figure 3.11: BER performance comparison for different Frame length under AWGN channel
49
10
Rayleigh Fadding Channel 3 Parallel interleaver of Lenght 1000 bit 3 Parallel interleaver of Lenght 100 bit
10
10 BER 10
10
10
8 E /N (dB)
b 0
10
12
14
16
Figure 3.12: BER performance comparison for different Frame length under Rayleigh channel
10
0
Effect of code memory of BICMID Code memory =3 Code memory =6 Code memory =4
10
10 BER
10
10
10
10
3 Eb/N0(dB)
Figure 3.13: BER performance comparison for different code memory over AWGN channel
50
In Figure 3.13 shows the effect of the code memory. The lower code memory = 3 performs much better performance at the low SNR. The lower-complexity coded performs much better at the low SNR because there will be low multiplicative errors in that regions. As the complexity is increased further more to m=4 and 6 i.e) the higher complexity codes can provide better asymptotic performance due to the improved df ree distance.
3.3.4 TCM
The BER curve for the TCM is obtained by using eight state Ungerboeck code as mentioned in the table 3.1 and 8-PSK signal mapping as its modulation scheme. The BER curve characteristic of TCM in both AWGN and Rayleigh channel is demonstrated in the FIgure 3.14 and Figure 3.15 . From these gure we can see that the TCM performance can be independent of the the total frame length for the both the Rayleigh and for thes AWGN channel. TCM performs mush better in the AWGN channel but its performance all most degrades when they are perform in the uncorrelated Rayleigh channel.
10
0
TCM performance under AWGN for various Frame Length using PSK,8 states 500 I nformati on bi ts 5000 I nformati on bi ts
10
10 BER
10
10
10
10
4 5 Eb/N0(dB)
Figure 3.14: BER performance comparison for Different frame length under AWGN Channel

BER Performance Comparison of TCM for 8 PSK over uncorrelated Rayleigh Fading 500 I nformati on bi ts 5000 I nformati onbi ts
51
10
10
BER
10
10
10
8 10 Eb/No(dB)
12
14
16
18
Figure 3.15: BER performance comparison for Different frame length under Rayleigh Channel
3.3.5 TTCM
First of all lets discuss the EXIT chart analysis of TTCM decoders which are nothing but two TCM decoders which are arranged parallel to each other. The generation is done for the inner code since, the TCM decoder here acts as the inner code for TTCM. The generator polynomial used for this simulation is given in table 3.1 having a eight state. Here its only necessary to generate the EXIT chart for any one of the decoders and EXIT chart for the second decoder can be obtained by just swapping the axis of EXIT chart, this can be illustrated in Figure 3.17. The Figure 3.16 illustrates the EXIT band chart for various frame length. As we can see from the gure, the EXIT band chart with 5000 frame length per bits, the bands of the EXIT charts are almost merging with actual EXIT chart but when the frame length is reduced we can make the difference in the bands since this points towards the fact EXIT band chart property depends on the overall Frame length of the code. The EXIT chart points when the a priori probability reaches its maximum then the amount of extrinsic information it produces will also be maximum. Thats the reason why the point of (Ia , Ie ) is coinciding with (1,1) point in the EXIT chart.
52
8 PSK500 information bits 1
0.8
0.6 IE 0.4
0.2
Area=0.6133343
0.2
0.4 IA
0.6
0.8
8 PSK5000 information bits 1
0.8
0.6 IE 0.4
0.2
Area=0.61534
0.2
0.4 IA
0.6
0.8
Figure 3.16: EXIT band chart for TCM decoder for various interleaving frame length at SNR = 2.5 dB
53
8 PSK , 500 bits/Frame 1 0.9 0.8 0.7 0.6 IE1,IA2 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 IA1,IE2 0.6 0.7 0.8 0.9 1
8 PSK , 5000 bits/Frame 1 0.9 0.8 0.7 0.6 IE1,IA2 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 IA1,IE2 0.6 0.7 0.8 0.9 1
Figure 3.17: EXIT band chart for TTCM decoder for various interleaving frame length at SNR = 2.5 dB

TTCM, 8 PSK 1 Iteration 2 Iterations 4 Iterations 6 Iterations 8 Iterations
54
10
10
10
10 BER
10
10
10
10
6 Eb/No(dB)
10
12
Figure 3.18: TTCM BER performance using different iterations
Figure 3.18 shows the performance tot TTCM for various number vs decoding iterations. It can be seen that as the number of iterations are increased in the decoder performs signicantly better. However, after the six iterations there is only the improvement achieving by using any further iterations is less signicant. For example, it can be seen that in the Figure 3.18 after the six iterations the improvement that we are getting is only about 0.2 dB. Hence the usual preferred iterations are between four and eight according to the overall complexity required per decoder. The Turbo cliff is almost starting at Eb /N0 = 2.5 dB and will be lasting till 4 dB. The Figure 3.19 shows the EXIT band chart trajectories of TTCM decoders for various frame length such as 100, 500, and 5000, this actually shows how the mutual information are exchanged between each other until they converges. From these gures we can see band graph are different for different frame lengths and also trajectories show the different channel realisation at the same SNR value its self. The different channel realisation trajectories will t the EXIT band chart quite well as seen from the Figures. The threshold Eb /N0 for the TTCM is 2.5 db and from the EXIT chart given in the Figure 3.19 the EXIT chart tunnel is opened and the decoder can exchange mutual information smoothly. This can gives the condence that the simulations are correct.
55
TTCM,100 bits Frame Length 1 0.9 0.8 0.7 0.6 IA2,IE1 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 IA1,IE2 0.6 0.7 0.8 0.9 1
Figure 3.19: EXIT band chart trajectories for various Frame lengths at SNR=2.5 dB
56
10
BER Performance of TTCM for various interleaver length,8 Iterations 8 state PSK 50 bitrandom interleaver 500 bitrandom Interleaver 5000 bitrandom Interleaver
10
10
BER
10
10
10
10
4 Eb/N0(dB)
Figure 3.20: BER performance comparison TTCM for Different frame length under AWGN Channel
10
BER performance for TTCM for various interleaver length 8 iterations,8 PSK 50 bit random interleaver 500 bit random interleaver 500 bit random interleaver
10
BER
10
10
10
5 Eb/N0(dB)
10
15
Figure 3.21: BER performance comparison for TTCM for Different frame length under Rayleigh Channel
3.4. COMPARATIVE STUDY OF THE CODED MODULATION
57
The Figure 3.20 and 3.20 refers the effect of the frame length under both AWGN channel and Rayleigh channel. Its shows how dramatically the overall performance of the TTCM depends on the Frame length that is used in the system. So in short the TTCM performs better if the frame length is much higher in length.
3.4 Comparative Study of the Coded Modulation

This section deals with the BER characteristic of TCM,TTCM,BICM and BICM-ID under AWGN and uncorrelated Rayleigh Channel. In this section complexity of the coded modulation is compared with overall coding gain obtained from the uncoded 4 -PSK modulation scheme. The complexity is compared in terms of the number of decoding stages as well as the number of decoding iteration. For TCM or BICM code the complexity is equal to the number of decoding stages namely to S = 2K1 where K is the constrain length.The TTCMs overall complexity is proportional to 2 S t where t is the number of iterations, here the overall complexity is doubled since, TTCM invoke two TCM codes. As in the case of the BICM-ID the decoder uses only one decoder but during every iterations demodulator /de-mapper is invoked .However, the overall complexity of de mapper is insignicant when compared to overall complexity of decoder used. Hence, the complexity of BICM-ID with t iteration and using S state component is proportional to t S [27].
3.4.1 Performance Over AWGN Channel

Figure 3.22 demonstrates the BER characteristic of coded modulation under AWGN channel while trying to maintain the overall decoding complexity of all the coded modulation the same in terms of the number of trellis states. The simulation uses 64-State TCM, 64 State BICM, 8 State TTCM with four iterations and 8 State BICM-ID along with 8 Iteration with a overall frame lenght of 2000 information bits. At a BER 104 TTCM outperforms all other schemes while maintaing same complexity, TTCM achieves 1 dB better performance than BICM-ID, 1.5 dB better than TCM and BICM performance degrades with almost 1.8 dB lower Eb /N0 .

AWGN Channel BICM 64 State BICMID 8 State 8 Iterations TTCM 8 State 4 Iterations TCM 64 State
58
10
10
10
10 BER 10
10
10
10
6 Eb/No(dB)
10
12
Figure 3.22: BER performance of Coded modulation employing 8 PSK and using a Frame length 2000 information bits
10
Rayleight Fading Channel TCM 64 State BICM 64 State TTCM 8 State 4 Iterations BICMID 8 State 8 Iterations
10
10 BER 10
10
8 Eb/No(dB)
10
12
14
16
Figure 3.23: BER performance of Coded modulation employing 8 PSK and using a Frame length 2000 information bits
59
3.4.2 Performance over Uncorrelated Narrowband Rayleigh Fading Channels

The uncorrelated Rayleigh fading channels means that, using an innite-length interleaver over narrowband Rayleigh Fading channels. Figure 3.23 shows the performance of 8 State TTCM using four iterations ,16-state BICM-ID employing four iterations 64-state TCM and 64 state BICM,. The states where selected in a such a way that all the coded modulation did have similar complexity, As it can be seen from the Figure 3.23 TTCM performs best, followed by BICM-ID, BICM and TCM. At BER of 104 TTCM performs about 1.8 dB better in terms of the required Eb /No value than BICM-ID, 2.7 dB better than BICM and 5 db better than TCM. As it is evident from the Figure 3.23, the Error ow of TTCM was lower than the associated Error Free Feedback(EFF) bound of the BICM-ID. One the other hand, the BERs of TTCM and BICM-ID was almost identical at Eb /No =7 dB
3.4.3 Effect of Block Length on Coded Modulation

0
AWGN Channel BICMID 3000 Frames/Bit BICMID 1500 Frames/Bit BICMID 300 Frames/Bits TTCM 3000 BIts/Frame TTCM 1500 Bits/Frame TTCM 300 Bits/Frame TCM 64 State BICM 64 State
10
10
10
10 BER
10
10
10
10
4 5 Eb/N0(dB)
Figure 3.24: Effects of block length on Coded Modulation
60
This section compares the interleaving effect of block length on the coded Modulation. It is much evident form the Figure3.24 that high interleaving block is appropriate for the iterative TTCM and BICM-ID schemes. Here BICM-ID used for the simulations are 16 state with 4 iteration while TTCM uses 8 state with 4 iterations. At a BER of 104 the 300 bits/Frame performed worst with almost 1 dB degradation when compared over the 1500 bits/Frame. When 3000 frame length was used there was only slight improvement in its SNR performance. in short, BICM-ID scheme has more advantage when use it is used over a larger frame length. The TTCM performance improves when the frame length is increased and on the whole, here TTCM exhibit best performance.
3.4.4 Coding Gain versus Complexity

This section investigate the coding gain of the coded modulation schemes utilising an 8 PSK scheme versus the complexity at a BER of 104 . The coding gain is measured by comparing to that of the uncoded QPSK scheme which exhibit a BER of 104 = 8.5 dB and 34.8 dB for transmission over the AWGN and uncorrelated Rayleigh channel respectively. The overall complexity of TCM, BICM ,BICM-ID,TTCM is given by the S-State component and number of iteration respectively. Figure 3.25 and Figure 3.26 repersents the coding gain versus the Decoding complex for the 8 PSK transmission over AWGN. At a Low Decoding complexity as of 8, the non-iterative decoding TCM exhibits the highest decoding gain for AWGN channel as it can be seen. By contrast, the BICM purposed will be performing much better in uncorrelated Rayleigh channels. For iterative decoding schemes such as BICM-ID and TTCM different combination of t and S yields different performance at the same Decoding complexity. The coding gain of BICM-ID which uses t.S = 8 8 is always better than that of t.S = 4 16 since, the states which uses S=16 and t=4 cannot reach its ideal performance at t=4. In general, the coding gain of TTCM is the highest for the both the channels.
61
5 4.5 4 Coding Gain(dB) 3.5 3 2.5 2 1.5 1 TCM BICM TTCM BICMID
10
15
20
25
30 35 40 Relative Complexity
45
50
55
60
65
Figure 3.25: Coding gain at a BER of 104 against the decoding complexity when compared to the uncoded QPSK under AWGN channel
28 27 26 25 24 23 22 21 20 19 TCM BICM BICMID TTCM Coding Gain(dB)
10
20
30 40 Relative Complexity
50
60
70
Figure 3.26: Coding gain at a BER of 104 decoding complexity when compared to the uncoded QPSK under Rayleigh channel
3.5. CASE STUDY:7 PSK OVER AWGN CHANNEL
62
3.4.5 Area/Capacity vs SNR

Area and Capacity plot 3
2.5
2 Normal Scale
1.5
1 Area beneath EXIT FunctionTTCM Actual Capacity Curve 8PSK Area beneath EXIT FunctionBICMID
0.5
0 10
0 SNR(dB)
10
15
Figure 3.27: Area beneath EXIT function and DCMC Capacity plots in AWGN channel
The area/capacity plots are obtained by generating the area beneath the EXIT function for the inner codes. Here, the maximum capacity it can achieve is three because we are taking the reference signal as 8 PSK. The Figure 3.27 gives the comparison of the DCMC channel capacity with the TCM/TTCM and BICM/BICM-ID. Its evident from the gure BICM/BICMID can achieve a almost equal to 3 bits/symbol which is the desired DCMC capacity it can reach at an SNR equivalent to 12 dB, while TTCM suffer a loss and can only achieve a capacity of 2 bits/symbol.
3.5 Case Study:7 PSK over AWGN channel

The 7-PSK is a capacity efcient modulation scheme when compared to that the 8 PSK scheme, so this section compares the performance of the coded modulation by means of 8-PSK and 7PSK and tries to nd out which one is the best modulation scheme over the AWGN and Rayleigh channel.
63
10 10 10 10 BER 10 10 10 10
Comparison of BICM for 7PSK and 8PSK 8PSK 7PSK
6 Eb/N0(dB)
10
12
Figure 3.28: BER comparison of BICM using 7-PSK and 8-PSK having a frame length 1000 bits
EXIT chart comparison of 8PSK and 7PSK mapper 1 0.9 0.8 0.7 0.6 I(ae) 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 I(aa) 0.6 0.7 0.8 0.9 1 7 PSK 8 PSK
Figure 3.29: EXIT Chart Comparison for de mapper of 7-PSK and 8-PSK at an SNR=0 dB
64
The Figure 3.28 shows the BER performance of both 7-PSK and 8-PSK having a frame length of 1000 bits/frame. The mapping scheme used here is that of Gray Mapping. As we can see from the Figure the performance of the both the 7 PSK and 8 PSK are almost same. There is no relative advantage of using the 7 PSK in the case of BICM.
10
0
BER Compariosn 7PSK 8PSK
10
10 BER
10
10
10
10
5 6 Eb/N0(dB)
10
Figure 3.30: BER comparison of BICM-ID using 7-PSK and 8-PSK having a frame length 1000 bits
The Figure 3.29 portrays the EXIT chart comparison of the respective decoder performance when using both 7 and 8-PSK respectively. Both the EXIT chart almost merges which each other showing that there is no difference at all while the 7-PSK or 8-PSK demapper are been used. The simulation are carried out at an SN R = 0 dB for an frame length of 1500 bits The Figure 3.30 illustrates the BER performance comparison of 7 and 8 PSK using the BICM-ID coding scheme. Here a total frame length of 1000 bits used while the constraint length for the modulation scheme is 4 while maintaing a total iterations of eight. The performance of the 7-PSK out performs the 8 PSK by maintaing 0.5 dB better performance than 8 PSK.
65
EXIT chart comparison between 7 PSK and 8 PSK using TCM decoder 1 0.9 0.8 0.7 0.6 Ie 0.5 0.4 0.3 0.2 0.1 0 0 0.1 0.2 0.3 0.4 0.5 Ia 0.6 0.7 0.8 0.9 1 8 PSK 7 PSK
Figure 3.31: EXIT Chart Comparison for de mapper of 7-PSK and 8-PSK at an SNR=2.5 dB
10
0
Comparison of TCM with 7 PSK and 8PSK 7PSK 8PSK
10
10 BER
10
10
10
10
5 6 Eb/N0(dB)
10
Figure 3.32: BER comparison of TCM using 7-PSK and 8-PSK having a frame length 1500 bits
3.6. CONCLUSION
66
The EXIT chart comparison for TCM decoder employing the 7 and 8 PSK are illustrated in the Figure 3.31. The EXIT chart was studied for interleaver length of 1500 bit/frame at an SN R = 2.5 dB which is threshold SNR for the TTCM. The EXIT charts revels that the performance of the both the decoders are almost identical to each other and merges with each other.
10
0
Comparison of TTCM using 7PSK and 8PSK 7PSK 8PSK
10
10 BER
10
10
10
10
3 Eb/N0(dB)
Figure 3.33: BER comparison of TTCM using 7-PSK and 8-PSK having a frame length 1500 bits
The Figure 3.32 shows the comparison of TCM using 7-PSK and 8-PSK under the AWGN channel. The 7-PSK clearly outperforms the 8-PSK at BER of 104 with an 1 dB superior performance while using a frame length of 1500 bits. The Figure 3.33 describes the comparison study of the TTCM under AWGN channel and here, the 7-PSK almost outperforms the 8PSK during low Eb /N0 and will lead to earlier convergence to the error oor. The performance of 8-PSK improves as the Eb /N0 increases and has much superior performance here.
3.6 Conclusion
This chapter studies the individual and comparative study of the four coded modulation for transmission over the AWGN and uncorrelated Raleigh channel. Each of the four coded modulation was studied separately in term of their interleaving frame length, coded memory elements and
3.6. CONCLUSION
67
with the aid of the EXIT charts. The following conclusion were obtained from the simulation that was carried out, for a given complexity TCM performed better than BICM in AWGN channel while the TCMs performance was worse than BICM in uncorrelated channel. However, when considering the iterative coding schemes BICM-ID almost outperformed both the TCM and BICM over the both the channel maintaining the same complexity. TTCM was the clear winner and has almost shown its dominance in its comparative study. Both the BICM/ID and TTCM had its share of advantage and disadvantages, as BICM/BICM-ID matched the theoretical channel capacity while TTCM suffer a loss in channel capacity. TTCM EXIT tunnel function found to be opened at a much lower SNR than BICM/ID. At last a case study was done with 7-PSK over the transmission through AWGN channel and found that TTCM, TCM and BICMID had a superior performance than 8-PSK. But BICM had an identical performance to that of 8-PSK.
Chapter
Management and Planning

This chapter will explain how I managed to complete my design works , simulations and the completion of the writing spanned through a sixteen week period.
4.1 Initial Scope

The project marked its beginning, after the rst discussion with my supervisor. We both agreed to do the project on the topic of the Coded Modulation - a bandwidth efcient coding scheme. The main objective of this project was to do a comparative study on the coded Modulation schemes such as TCM, TTCM, BICM and BICM-ID. Each of the above said coded modulation schemes was to study separately and had to evaluate their performance with the aid of the BER curve plots and the EXIT chart techniques.
4.2 Available Resources

There were numbers of resources available used to make the project a grant success. The background research was carried by IEEE papers for the related topics and can be seen in the Bibliography section. Here I would like to thank Dr Lie Liang Yangs nine month report which gave me much understanding for the topic Turbo Coding. A number of internet sites such as www.complextoreal.com, http : //en.wikipedia.org etc, also helped to create a very good
68
4.3. PROJECT TASKS
69
backup of the theory when ever I faced difculty. The matlab is the software used for simulation purposes and the tutorial support provided by the matlab was splendid. Dr Rob had already published this turbo code in his site http : //users.ecs.soton.ac.uk/rm/, the BCJR algorithm developed by him acted as the one of the fundamental codes in my project. Later I had to made necessary changes to it. The fundamental coded which I developed can be seen in Appendix section followed by Robs code. I found Lyceum much helpful whenever i had run the BER characteristics to the range of 107 to understand the error oors. Additionally Latex and X f ig helped me to make this report in world class standards.
4.3 Project Tasks

As mentioned in my Initial Scope section I had to study all coded modulation and to evaluate their performance by EXIT chart and BER curve characteristics and then had to compare the performance in term of complexity, block length and coding gains. During the start of the project I faced difculty in developing the EXIT chart coded since, I havent personally worked on EXIT charts before, but the guidance provided by Dr Rob helped me to jump this hurdle.
4.4 Risk
I literally faced difculty by the time limit and over crowding of the Lyceum as I had to do a number of iterative decoding, my computer resource was insufcient. Another risk factor which I faced was being self motivated. In the beginning when ever I faced difculty in getting the output I will be depressed and my complete day will be wasted thinking about that. In order to avoid that I had spend myself some pleasure time such as playing tennis to make me free some all troublesome and to keep me motivated.
4.5. INITIAL AND FINAL GANTT CHART
70
4.5 Initial and Final Gantt chart

The Figure 4.1 and 4.2 gives the actual time table and purposed timetable of the project planing. Purposed Gantt chart was submitted for the dissertation brief as a forthcoming timetable for my project. I almost was able to go accordingly and was able to nish my simulation before a week. That gave me better chance of organising my results I have generated so far and was able to do a good analysis of my results. The actual Gantt chart can be seen from the Figure 4.2.
Figure 4.1: Purposed Gantt chart
Figure 4.2: Actual Gantt chart
4.6. MANAGEMENT TECHNIQUES
71
4.6 Management Techniques

During this project I found many things quite interesting, especially the use of the Log Book which helped me to write down notes, referencing during my write-up and a better way to organising things. The new MScIS kept me updated with all task what my supervisor has asked me to do for the following weeks. The meeting arranged by Dr. Robert both individual and group meeting helped me to keep well focused on my aim.
4.7 Conclusion
This chapter explains the management and planning section. The main task explained here is the Initial project scope, the available system resource for example softwares used for the simulation, the project tasks which describes the how difcult the given task where, what all Risks that was encountered during these project, the Initial and Final Gantt chart and Finally the management techniques used for the successful completion of this project.
Chapter
Conclusions and Future Work

In this concluding chapter, the main nding of this project are presented. This will be followed by further studies which I could not be completed due to time constrain.
5.1 Conclusions
This project investigated the application and the performance of the coded modulation schemes over wireless fading channels using 8-PSK modulation scheme and 2/3 rate convolutional code. The four major coded modulation such as TCM, TTCM, BICM and BICM-ID was studied separately as well as a comparative study was also have been done in terms of overall complexity and coding gains. The major ndings are listed below: TCM performs better than BICM in AWGN channel showed a 2.6 dB gain while, TCM performs worse than BICM over uncorrelated Rayleigh Channel with a gain of 3.8 dB. TTCM performs best over a variety of interleavers irrespective of the interleaver length. BICM-ID performs best when they have a long interleaver length. TTCM outperformed all other three coded modulation schemes. BICM/BICM-ID can match the theoretical Discrete-input Continuos output Memoryless Channels Capacity of 3 bit/channel use. 72
5.2. FUTURE WORK
73
TCM/TTCM suffers a loss in its Channel capacity and can only match up to 2 bits/channel use. 7-PSK modulation scheme which had better capacity efciency than 8-PSK was studied and it showed improvements in its BER performance of coded modulation other than BICM scheme.
5.2 Future Work

This project focused mainly on coded modulation for the transmission over AWGN and narrow band Rayleigh channel, but actually the mobile radio channel are widely a dispersive channels and can be characterised by band limited linear lters [28]. Inter-Symbol Interference(ISI) come to play when the modulation bandwidth exceeds the coherence bandwidth of the channel. In short this signals with ISI can cause transmitted signal to spread or dispersed in time domain. So this have to equalised at the receiver to recover the information. Therefore, I would like to further extend my studies to Decision Feedback Equaliser in order to study the coded Modulation in Wide-band channels. Another important study which I would like to make is about the Space Time Trellis Coding (STTC) [29] as, it is the one of the most widely used technology to improve the capacity and reliability of the data transmission over the MIMO channel.
Bibliography
[1] S. X. NG, Coded Modulation Schemes For Wireless Channels. PhD thesis, Department of Electronics and Computer Science, University of Southampton, October 2002. [vii, viii, 2, 3, 23, 39] [2] G. UngerBoeck, Channel coding with multilevel/phase signals, IEEE Transaction on Information Theory, vol. IT-28, pp. 5567, January 1982. [vii, 2, 15, 16, 17, 18, 40] [3] C. Shannon, A mathematical theory of communication, Bell System Technical Journal, pp. 379427, 1948. [2] [4] R. Hamming, Error detecting and error correcting codes, Bell System Technical Journal, vol. 29, pp. 147160, 1950. [2] [5] M. Golay, Notes on digital coding, Proceedings of the IEEE, vol. 37, p. 657, 1949. [2] [6] P. Elias, Coding for noisy channels, IRE Convention Record, pp. 3747, 1955. [2] [7] R.Fano, A heuristic discussion of probabilistic coding, IEEE Transactions on Information Theory, vol. IT-9, pp. 6474, April 1963. [2] [8] J.Massey, Threshold Decoding. Cambridge,MA,USA:MIT Press, 1963. [2] [9] A. Viterbi, Error bounds for convolutional codes and an asymptotically optimum decoding algorithm, IEEE Transactions on Information Theory, vol. IT-13, pp. 260269, April 1967. [2, 13] 74
BIBLIOGRAPHY
75
[10] F. J. L. R. Bahl, J. Cocke and J. Raviv, Optimal decoding of linear codes for minimising symbol error rate, IEEE Transaction on Information Theory, vol. 20, pp. 284287, March 1974. [2, 3, 28] [11] D. Divsalar and M. Simon, The design of trellis coded mpsk for fading channel:set partitioning for optimum code design, IEEE Transactions on Communications, vol. 36, pp. 10311021, September 1988. [2] [12] C. Schlegel, Trellis Coding. The institute of Electrical and Electronics Engineers Inc,New YorkIEEE Press, 1997. [3] [13] E.Zehavi, 8-psk trellis codes for rayleigh fading channel, IEEE Transaction on Communication, vol. 40, pp. 873883, May 1992. [3, 23] [14] G.Caire, G. Taricco, and E.Bigheri, Bit-interleaved coded modulation, IEEE Transactions on Information Theory, vol. 44, pp. 927946, May 1998. [3] [15] A. C.Baerrou, Near shannon limit error-correction coding and decoding:turbo-codes, Proc ICC93, pp. 10641070, May 1993. [3, 20, 21] [16] R. Steele and L. Hanzo, Mobile Radio Communication : Second and Third Generation Cellular and WATM System. John Wiley and Sons, 1999. [3] [17] A. G. S. L Goff and C.Baerrou, Turbo-codes and high spectral efciency modulation, Proceedings of the IEEE International Conference on Communication, pp. 645649, 1994. [3] [18] X. Li and J. A.Ritcey, Trellis coded modulation with bit interleaving and iterative decoding, IEEE Journal on Selected Areas In Communication, vol. 17, April 1999. [3, 26] [19] E.Paaske, Short binary convolutional codes with maximal free distance for rates 2/3 and 3/4, IEEE Transaction on Information Theory, vol. IT-29, pp. 683689, 1974. [15, 40] [20] G. D. Forney, The viterbi alogorithm, Proceedings of the IEEE, vol. 61, pp. 268277, March 1973. [19]
BIBLIOGRAPHY
76
[21] P. Robertson, Bandwidth-efcient turbo trellis-coded modulation using punctured component codes, IEE Journal on Selected Areas In Communication, vol. 16, pp. 206218, February 1998. [20, 31] [22] J. Cavers and P.Ho, Analysis of error performance of trellis-coded modulation in rayleighfading, IEEE Transaction on Communication, vol. 40, pp. 7483, January 1992. [23] [23] X. Li and J. A.Ritcey, Bit-interleaved coded modulation with iterative decoding, IEEE Communication Letters, vol. 1, November 1997. [26] [24] S. ten Brink, Convergence behaviour of iteratively decoded parallel concatenated codes, IEEE Transactions on Communications, vol. 49, pp. 17271737, October 2001. [31] [25] R. G. Maunder, Irregular Variable Lenght Coding. PhD thesis, School of Electronic and Computer Science, University of Southampton, December 2007. [35] [26] J. W. Lee and R. E. Blahut, Generalised exit chart and ber analysis of generalised exit chart and ber analysis of generalised exit chart and ber analysis of nite-length turbo codes, Global Telecommunications Conference, 2003. GLOBECOM 03, vol. 4, pp. 2067 2072, December 2003. [37] [27] T. H. L. L. Hanzo and B. L. Yeap, Turbo Coding, Turbo Equalisation and Space-Time Coding. Wiley IEEE Press, 1 ed. [57] [28] R. Steele and L. Hanzo, Mobile Radio Communication : Second and Third Generation Cellular and WATM System. John Wiley and Sons, 2nd ed., 1999. [73] [29] N. V. Tarokh and A. R. Calderbank, Space time codes for high data rate wireless communication performance criterion and code construction, IEEE Transaction on Information Theory, vol. 44, pp. 744765, March 1998. [73]
Appendix
Publishable Paper
77
Comparative Study of Coded Modulation schemes using EXIT chart and BER characteristics
Cherian Danny Joseph and Robert G. Maunder School of Electronics and Computer Science, University of Southampton, SO17 1BJ, United Kingdom
AbstractThis paper , extensively studies on the topic Coded Modulation(CM) a scheme which combines both the coding as well as the modulation in together to get a bandwidth efcient scheme. This project contributes a genuine comparative study on different Coded Modulation schemes such as Trellis Coded Modulation (TCM), Bit interleaved Coded Modulation (BICM),Turbo Trellis Coded Modulation (TTCM) and Bit Interleaved Coded Modulation with Iterative Decoding (BICM-ID) in context of 8 level Phase Shift Keying(8PSK) over the Gaussian and uncorrelated Rayleigh channels. Here, the comparison are done in terms of decoding complexity, the bandwidth efciency, the coding gain and the frame length by using the study tools such as EXtrinsic Information Transfer Charts(EXIT charts) and BER curve characteristics.
I. I NTRODUCTION The radio spectrum is a scarce resource. Therefore, the main objective is to efciently exploit the available spectrum. The error correction codes always add the redundancy by increasing the number of coded symbols. Since, the channel is band limited the transmitted symbol rate should be xed which could result in lower rate information transmission rate. Now, the question is how can the channel be utilized in an efcient way still by adding the redundant bit. The key therefore is to combine the coding as well as modulation into a single unit and hence the name Coded Modulation. Ungerboeck [1], in his paper fully describes how to employ the TCM schemes in redundant non-binary modulation(symbol based) with the combination of a nite state Forward Error Correction(FEC) encoder, which selects the coded signal sequence. The extra bits formed by corresponding convolution encoder will restrict the possible state transformation among the consecutive phasor to a certain legitimate constellation. The receiver tries to decode the incoming noisy signal by a trellis based soft-decision maximum-lilkelihod detector and tries to map it to the each of the legitimate phasor sequence by the restrictions imposed by the convolution encoder. The term Trellis is used to describe this scheme is because the overall operation can be described by a corresponding state transition diagram similar to that of binary convolution encoder. The only difference in TCM is that, here trellis branches are labelled with respect to the redundant non-binary modulated phasors. paper proposed an new approach know as set partitioning [1] which aims more directly at maximum free Euclidian Distance(ED). Turbo codes [2] was a major milestone in the forward error correction codes which can even achieve an excellent bit error rates at low SNR. The original proposal was for
the BPSK scheme but were soon successful with multilevel coded as well. Robertson [3] soon introduced the concept of the Turbo Trellis Coded Modulation (TTCM) employing two TCM codes as parallel concatenation of two recursive TCM encoder, and adapted puncturing mechanism to avoid the obvious disadvantage of the rate loss. Bit-interleaved Coded Modulation (BICM) was a idea proposed by Zehavi [4] in order to improve the diversity of the code in Rayleigh channel. The design of the coded modulation schemes are affected by several factors such as High Free Euclidian Distance which is desired for the AWGN channel, while its was interested to note that a high Effective Code Length and a high minimum product distance were the main factors effecting the fading channel. The diversity of the code can be dened as length of the shortest error path [5] and one should be aware that the shortest error distance are not necessarily be the minimum distance error. Unfortunately there was no TCM codes available which can compensate the above said difculties. In order to solve the problem Zehavi came with an idea to render the codes diversity equal to that smallest number of different bits, employing the bit-based interleaving. Bit-Interleaved Modulation(BICM) was purposed to increase the diversity of the Ungerboeck TCM scheme under the Rayleigh channel. Li [6], [7] suggested a new iterative scheme of Bit-Interleaved Coded Modulation(BICM-ID) using Iterative Decoding which employed Set-Partitioning signal labelling system as that of Ungerboeck. Introduction of softdecision feedback from the decoders output to the demapper/demodulator input to iterate between them is advantageous of the fact that, it improves the reliability of the soft information passed to the demapper/demodulator. EXtrinsic Information Transfer(EXIT) [8] chart analysis is powerful that is used to check the convergence of the iterated decoders. BER chart is one of most powerful tool to analysis the performance how good the decoder is, but it was not able to explain in detail how the decoder converges when an iterative decoding is done. EXIT chart measures the Mutual Information (MI) that is exchanged between the constituent decoder in a iterative process. II. S YSTEM M ODEL The schematic for the coded modulation is illustrated in the Figure 1. The source here will be producing some random information bits, which is then encoded by the one of the respective encoders and consecutively interleaved by randominterleavers . The interleaved bits/symbols are then modulated
according to symbol rule for each of corresponding modulation schemes.The channel discussed here for the coded modulation schemes are that of the AWGN and Rayleigh distributed at fading.
TCM
Source
TCM /TTCM BICM Enocder
Interleaver
8 PSK Mapper
Channel BICMID Interleaver
Destination
TCM/ TTCM BICM Decoder
Deinterleaver
8 PSK Demapper
TCM
BICM-ID. To make this little more elaborate the following generator polynomials are used to study and stimulate the performance of the TCM/ TTCM schemes are given in the upper part Table I. These are RSC codes which will be adding one parity bits to information bits. Hence, the corresponding m coding rate for a 2m+1 ary PSK is R = m+1 . The number of decoding staged associated with the constraint length k is 2K1 . Lower part of the Table I shows the respective generator polynomial for the BICM/BICM-ID schemes. As noted previously they are non-systematic codes having maximum hamming distance. Here, as TCM/TTCM only one parity bit is added to the information bits and hence, the coding rate and bandwidth efciency remains the same as that of the TCM/TTCM. Decoders utilises the Log based- Maximum A Posteriori (Log-MAP) [11] algorithm for its soft decision. Log-MAP algorithm was chooses since, they are the numerical stable version and to reduce the complexity as well as the numerical problem.
10
0
Fig. 1.
System Overview for different coded Modulation Schemes [9]
AWGN Channel BICM 64 State BICMID 8 State 8 Iterations TTCM 8 State 4 Iterations TCM 64 State
The relationship between the AWGN and Rayleigh fading channel can be expressed as: yt = t xt + nt (1)
10
10
where xt is the transmitted discrete signal and yt is received signal. t is the Rayleigh-distrubuted fading having an 2 expected squared value of E(t )=1 and nt is the complex AWGN having a noise variance of N0 /2 per dimension. For an AWGN channel t = 1. The receiver side consist of demodulator/demaper followed by a de-interleaver and will be having one of the TCM, TTCM or BICM decoders which explained in the previous chapter. When considering the BICM-ID schemes the decoder output is is interleaved and then feedback to the demmaper input as illustrated in the Figure 1.
Scheme TCM/TTCM State 8 16 32 64 8 16 BICM/BICM-ID 32 64 g1 11 23 45 103 04 01 07 02 14 03 15 06 g2 02 04 16 30 02 04 01 05 06 10 06 15 g3 04 16 34 66 06 07 04 07 16 17 15 17
10 BER 10
10
10
10
6 Eb/No(dB)
10
12
Fig. 2. BER performance of Coded modulation employing 8 PSK and using a Frame length 2000 information bits
III. R ESULTS AND D ISCUSSION This section deals with the BER characteristic of TCM,TTCM,BICM and BICM-ID under AWGN and uncorrelated Rayleigh Channel. In this section complexity of the coded modulation is compared with overall coding gain obtained from the uncoded 4 -PSK modulation scheme. The complexity is compared in terms of the number of decoding stages as well as the number of decoding iteration. For TCM or BICM code the complexity is equal to the number of decoding stages namely to S = 2K1 where K is the constrain length.The TTCMs overall complexity is proportional to 2 S t where t is the number of iterations, here the overall complexity is doubled since, TTCM invoke two TCM codes. As in the case of the BICM-ID the decoder uses only one decoder but during every iterations demodulator /de-mapper is invoked .However, the overall complexity of de mapper is insignicant when compared to overall complexity of decoder
TABLE I G ENERATOR POLYNOMIAL FOR U NGERBOCK TCM CODED WITH BEST MAXIMUM DISTANCE [1] AS WELL AS THE PAASKE S [10] GENERATOR POLYNOMIALS FOR BICM CODES . B OTH THE CODES ARE EXPRESSED IN OCTAL REPRESENTATION FORM .
The study of coded modulation schemes are with Ungerbocks TCM, Robersons TTCM, Zehavis BICM and Lis
used. Hence, the complexity of BICM-ID with t iteration and using S state component is proportional to t S. A. Performance Over AWGN Channel Figure 2 demonstrates the BER characteristic of coded modulation under AWGN channel while trying to maintain the overall decoding complexity of all the coded modulation the same in terms of the number of trellis states. The simulation uses 64-State TCM, 64 State BICM, 8 State TTCM with four iterations and 8 State BICM-ID along with 8 Iteration with a overall frame length of 2000 information bits. At a BER 104 TTCM outperforms all other schemes while maintaing same complexity, TTCM achieves 1 dB better performance than BICM-ID, 1.5 dB better than TCM and BICM performance degrades with almost 1.8 dB lower Eb /N0 . B. Performance over Uncorrelated Narrowband Rayleigh Fading Channels The uncorrelated Rayleigh fading channels means that, using an innite-length interleaver over narrowband Rayleigh Fading channels. Figure 3 shows the performance of 8 State TTCM using four iterations,16-state BICM-ID employing four iterations 64-state TCM and 64 state BICM,. The states where selected in a such a way that all the coded modulation did have similar complexity, As it can be seen from the Figure 3 TTCM performs best, followed by BICM-ID, BICM and TCM. At BER of 104 TTCM performs about 1.8 dB better in terms of the required Eb /No value than BICM-ID, 2.7 dB better than BICM and 5 db better than TCM. As it is evident from the Figure 3, the Error ow of TTCM was lower than the associated Error Free Feedback(EFF) bound of the BICM-ID. One the other hand, the BERs of TTCM and BICM-ID was almost identical at Eb /No =7 dB
10
0
that high interleaving block is appropriate for the iterative TTCM and BICM-ID schemes. Here BICM-ID used for the simulations are 16 state with 4 iteration while TTCM uses 8 state with 4 iterations. At a BER of 104 the 300 bits/Frame performed worst with almost 1 dB degradation when compared over the 1500 bits/Frame. When 3000 frame length was used there was only slight improvement in its SNR performance. in short, BICM-ID scheme has more advantage when use it is used over a larger frame length. The TTCM performance improves when the frame length is increased and on the whole, here TTCM exhibit best performance.
10
0
AWGN Channel BICMID 3000 Frames/Bit BICMID 1500 Frames/Bit BICMID 300 Frames/Bits TTCM 3000 BIts/Frame TTCM 1500 Bits/Frame TTCM 300 Bits/Frame TCM 64 State BICM 64 State
10
10
10 BER
10
10
10
10
4 5 Eb/N0(dB)
Fig. 4.
Effects of block length on Coded Modulation
D. Study using EXIT chart anaylsis Firstly, we will be studying the EXIT band chart characteristic [12] of BICM-ID schemes. Since, BICM-ID is a serial concatenated code having the outer code will be of a 2/3 rate convolutional encoder of code memory equal to three and the inner code is of a De-mapper.
BICMID 3000 Frames/Bit 1 0.9
Rayleight Fading Channel TCM 64 State BICM 64 State TTCM 8 State 4 Iterations BICMID 8 State 8 Iterations
10
10 BER
0.8 0.7
10
0.6 I(ae) 0.5 0.4 0.3

0 2 4 6 8 Eb/No(dB) 10 12 14 16
10
0.2 0.1 0 0 0.2 0.4 I(a )

a
Fig. 3. BER performance of Coded modulation employing 8 PSK and using a Frame length 2000 information bits Fig. 5.
0.6
0.8
EXIT band Chart trajectories for BICM-ID at a SNR =4.5 dB
C. Effect of Block Length on Coded Modulation This section compares the interleaving effect of block length on the coded Modulation. It is much evident form the Figure4
Figure 5 shows the effect of the Frame length on the performance of the code over both AWGN and Rayleigh Channel. It can be seen from the gure that performance of the
BICM-ID code is very remarkable for a system with the longer frame length. However, even for short frame length it can give good results but suffers a overall degradation of about 2 db at a BER of 104 for both AWGN and narrow band at fading channels. Note that larger block length can lead to the earlier convergence, in term of low SNR and number of iterations used but, at higher SNR values the effect of the frame length is insignicant. Now, we can discusses in detailed about the TTCM EXIT band chart characteristics, the TTCM is nothing but a parallel concatenated code of two TCM decoder. The coding rate is of 2/3 and coding memory of three is used here.
TTCM,5000 bits Frame Length 1 0.9 0.8 0.7 0.6 IA2,IE1 0.5 0.4 0.3 0.2
5 4.5 4 Coding Gain(dB) 3.5 3 2.5 2 1.5 1 TCM BICM TTCM BICMID
10
15
20
25
30 35 40 Relative Complexity
45
50
55
60
65
Fig. 7. Coding gain at a BER of 104 against the decoding complexity when compared to the uncoded QPSK under AWGN channel
28 27 26 25 24 23 22 21 TCM BICM BICMID TTCM Coding Gain(dB)
0.1 0 0 0.1 0.2 0.3 0.4 0.5 IA1,IE2 0.6 0.7 0.8 0.9 1
Fig. 6.
EXIT band chart trajectories for TTCM at SNR=2.5 dB
EXIT band chart trajectories [12] of TTCM decoders for the frame length of 5000 bits are used here for simulations. This actually shows how the mutual information are exchanged between each other until they converges. From these gures we can see band graph are different for different frame lengths and also trajectories show the different channel realisation at the same SNR value its self. The different channel realisation trajectories will t the EXIT band chart quite well as seen from the Figures. The threshold Eb /N0 for the TTCM is 2.5 db and from the EXIT chart given in the Figure 6 the EXIT chart tunnel is opened and the decoder can exchange mutual information smoothly. This can gives the condence that the simulations are correct. E. Coding Gain Vs Decoding Complexity This section investigate the coding gain of the coded modulation schemes utilising an 8 PSK scheme versus the complexity at a BER of 104 . The coding gain is measured by comparing to that of the uncoded QPSK scheme which exhibit a BER of 104 =8.5 dB and 34.8 dB for transmission over the AWGN and uncorrelated Rayleigh channel respectively. The overall complexity of TCM, BICM ,BICM-ID,TTCM is given by the S-State component and number of iteration respectively. Figure 7 and Figure 8 repersents the coding gain versus the Decoding complex for the 8 PSK transmission over AWGN. At a Low Decoding complexity as of 8, the non-iterative decoding TCM exhibits the highest decoding gain for AWGN channel as it can be seen. By contrast, the BICM purposed will be performing much better in uncorrelated Rayleigh channels.
20 19
10
20
30 40 Relative Complexity
50
60
70
Fig. 8. Coding gain at a BER of 104 decoding complexity when compared to the uncoded QPSK under Rayleigh channel
For iterative decoding schemes such as BICM-ID and TTCM different combination of t and S yields different performance at the same Decoding complexity. The coding gain of BICM-ID which uses t.S = 8 8 is always better than that of t.S = 4 16 since, the states which uses S=16 and t=4 cannot reach its ideal performance at t=4. In general, the coding gain of TTCM is the highest for the both the channels.
F. Area/Capacity vs SNR The area/capacity plots are obtained by generating the area beneath the EXIT function for the inner codes. Here, the maximum capacity it can achieve is three because we are taking the reference signal as 8 PSK. The Figure 9 gives the comparison of the DCMC channel capacity with the TCM/TTCM and BICM/BICM-ID. Its evident from the gure BICM/BICMID can achieve a almost equal to 3 bits/symbol which is the desired DCMC capacity it can reach at an SNR equivalent to 12 dB, while TTCM suffer a loss and can only achieve a capacity of 2 bits/symbol.
Area and Capacity plot 3
2.5
2 Normal Scale
1.5
[11] F. J. L. R. Bahl, J. Cocke and J. Raviv, Optimal decoding of linear codes for minimising symbol error rate, IEEE Transaction on Information Theory, vol. 20, no. 3, pp. 284287, March 1974. [12] J. W. Lee and R. E. Blahut, Generalised exit chart and ber analysis of generalised exit chart and ber analysis of generalised exit chart and ber analysis of nite-length turbo codes, Global Telecommunications Conference, 2003. GLOBECOM 03, vol. 4, pp. 2067 2072, December 2003.
1 Area beneath EXIT FunctionTTCM Actual Capacity Curve 8PSK Area beneath EXIT FunctionBICMID
0.5
0 10
0 SNR(dB)
10
15
Fig. 9. Area beneath EXIT function and DCMC Capacity plots in AWGN channel
IV. C ONCLUSION This paper studies comparative study of the four coded modulation for transmission over the AWGN and uncorrelated Raleigh channel. Each of the four coded modulation was studied separately in term of their interleaving frame length, coded memory elements and with the aid of the EXIT charts. The following conclusion were obtained from the simulation that was carried out, for a given complexity TCM performed better than BICM in AWGN channel while the TCMs performance was worse than BICM in uncorrelated channel. However, when considering the iterative coding schemes BICM-ID almost outperformed both the TCM and BICM over the both the channel maintaining the same complexity. TTCM was the clear winner and has almost shown its dominance in its comparative study. Both the BICM/ID and TTCM had its share of advantage and disadvantages, as BICM/BICM-ID matched the theoretical channel capacity while TTCM suffer a loss in channel capacity. TTCM EXIT tunnel function found to be opened at a much lower SNR than BICM/ID. R EFERENCES
[1] G. UngerBoeck, Channel coding with multilevel/phase signals, IEEE Transaction on Information Theory, vol. IT-28, pp. 5567, January 1982. [2] A. C.Baerrou, Near shannon limit error-correction coding and decoding:turbo-codes, Proc ICC93, pp. 10641070, May 1993. [3] P. Robertson, Bandwidth-efcient turbo trellis-coded modulation using punctured component codes, IEE Journal on Selected Areas In Communication, vol. 16, pp. 206218, February 1998. [4] E.Zehavi, 8-psk trellis codes for rayleigh fading channel, IEEE Transaction on Communication, vol. 40, pp. 873883, May 1992. [5] J. Cavers and P.Ho, Analysis of error performance of trellis-coded modulation in rayleigh-fading, IEEE Transaction on Communication, vol. 40, pp. 7483, January 1992. [6] X. Li and J. A.Ritcey, Bit-interleaved coded modulation with iterative decoding, IEEE Communication Letters, vol. 1, November 1997. [7] , Trellis coded modulation with bit interleaving and iterative decoding, IEEE Journal on Selected Areas In Communication, vol. 17, April 1999. [8] S. ten Brink, Convergence behaviour of iteratively decoded parallel concatenated codes, IEEE Transactions on Communications, vol. 49, no. 10, pp. 17271737, October 2001. [9] S. X. NG, Coded modulation schemes for wireless channels, Ph.D. dissertation, Department of Electronics and Computer Science, University of Southampton, October 2002. [10] E.Paaske, Short binary convolutional codes with maximal free distance for rates 2/3 and 3/4, IEEE Transaction on Information Theory, vol. IT-29, pp. 683689, 1974.
Appendix
Matlab codes
B.1 Original BCJR algorithm developed By Rob
% BCJR algorithm for a unity-rate convolutional code having 1 memory element, % a generator polynomial of [1,0] and a feedback polynomial of [1,1]. % For more information, see Section 1.3.2.2 of Robs % thesis (http://eprints.ecs.soton.ac.uk/14980) or the BCJR paper % (http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1055186). % Copyright (C) 2008 Robert G. Maunder
% This program is free software: you can redistribute it and/or modify it % under the terms of the GNU General Public License as published by the % Free Software Foundation, either version 3 of the License, or (at your % option) any later version.
% This program is distributed in the hope that it will be useful, but % WITHOUT ANY WARRANTY; without even the implied warranty of % MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. % Public License for more details. See the GNU General
% The GNU General Public License can be seen at http://www.gnu.org/licenses/.
function extrinsic_uncoded_llrs = bcjr_decoder(apriori_uncoded_llrs, apriori_encoded_llrs)
if length(apriori_uncoded_llrs) = length(apriori_encoded_llrs) error(LLR sequences must have the same length); end
% All calculations are performed in the logarithmic domain in order to
83
B.1. ORIGINAL BCJR ALGORITHM DEVELOPED BY ROB

% avoid numerical issues. These occur in the normal domain, because some of % the confidences can get smaller than the smallest number the computer can % store. See Section 1.3.2.4 of Robs thesis for more information on this. % % A multiplication of two confidences is achieved using the addition of the % corresponding log-confidences. If A = log(a) and B = log(b), then % log(a*b) = A+B (Equation 1.17 in Robs thesis). % % An addition of two confidences is achieved using the Jacobian logarithm % of the corresponding log-confidences. The Jacobian logarithm is defined % in the jac.m file. If A = log(a) and B = log(b), then % log(a+b) = max(A,B) + log(1+exp(-abs(A-B))) (Equation 1.19 in Robs % thesis).
84
% Matrix to describe the trellis % Each row describes one transition in the trellis % Each state is allocated an index 1,2,3,... Note that this list starts % from 1 rather than 0. % FromState, ToState, 1, 2, 1, 2, UncodedBit, 0, 1, 1, 0, EncodedBit 0; 1; 0; 1];
transitions = [1, 1, 2, 2,
% Find the largest state index in the transitions matrix % In this example, we have two states since the code has one memory element state_count = max(max(transitions(:,1)),max(transitions(:,2)));
% Calculate the a priori transition log-confidences by adding the % log-confidences associated with each corresponding bit value. This is % similar to Equation 1.12 in Robs thesis or Equation 9 in the BCJR paper. gammas_uncoded=zeros(size(transitions,1),length(apriori_uncoded_llrs)); for bit_index = 1:length(apriori_uncoded_llrs) for transition_index = 1:size(transitions,1) if transitions(transition_index, 3)==0 gammas_uncoded(transition_index, bit_index) = apriori_uncoded_llrs(bit_index); end end end
% Calculate the a priori transition log-confidences by adding the % log-confidences associated with each corresponding bit value. This is % similar to Equation 1.12 in Robs thesis or Equation 9 in the BCJR paper. gammas_encoded=zeros(size(transitions,1),length(apriori_uncoded_llrs));
B.1. ORIGINAL BCJR ALGORITHM DEVELOPED BY ROB

for bit_index = 1:length(apriori_uncoded_llrs) for transition_index = 1:size(transitions,1) if transitions(transition_index, 4)==0 gammas_encoded(transition_index, bit_index) = apriori_encoded_llrs(bit_index); end end end
85
% Forward recursion to calculate state log-confidences. This is similar to % Equation 1.13 in Robs thesis or Equations 5 and 6 in the BCJR paper. alphas=zeros(state_count,length(apriori_uncoded_llrs)); alphas=alphas-inf; alphas(1,1)=0; % We know that this is the first state for bit_index = 2:length(apriori_uncoded_llrs) for transition_index = 1:size(transitions,1) alphas(transitions(transition_index,2),bit_index) = jac(alphas(transitions(transition_index,2),bit_index),alphas(transitions(transition_index,1), bit_index-1) + gammas_uncoded(transition_index, bit_index-1) + gammas_encoded(transition_index, bit_index-1)); end end
% Backwards recursion to calculate state log-confidences. This is similar % to Equation 1.14 in Robs thesis or Equations 7 and 8 in the BCJR paper. betas=zeros(state_count,length(apriori_uncoded_llrs)); betas=betas-inf; for state_index = 1:state_count betas(state_index,length(apriori_uncoded_llrs))=0; % The final state could be any one of these end for bit_index = length(apriori_uncoded_llrs)-1:-1:1 for transition_index = 1:size(transitions,1) betas(transitions(transition_index,1),bit_index) = jac(betas(transitions(transition_index,1),bit_index),betas(transitions(transition_index,2), bit_index+1) + gammas_uncoded(transition_index, bit_index+1) + gammas_encoded(transition_index, bit_index+1)); end end
% Calculate a posteriori transition log-confidences. This is similar to % Equation 1.15 in Robs thesis or Equation 4 in the BCJR paper. deltas=zeros(size(transitions,1),length(apriori_uncoded_llrs)); for bit_index = 1:length(apriori_uncoded_llrs) for transition_index = 1:size(transitions,1) deltas(transition_index, bit_index) =
B.2. SYMBOL BASED BCJR ALGORITHM FOR TCM DECODERS
86
alphas(transitions(transition_index,1),bit_index) + gammas_encoded(transition_index, bit_index) + betas(transitions(transition_index,2),bit_index); end end
% Calculate the extrinsic LLRs. This is similar to Equation 1.16 in % Robs thesis. extrinsic_uncoded_llrs = zeros(1,length(apriori_uncoded_llrs)); for bit_index = 1:length(apriori_uncoded_llrs) prob0=-inf; prob1=-inf; for transition_index = 1:size(transitions,1) if transitions(transition_index,3)==0 prob0 = jac(prob0, deltas(transition_index,bit_index)); else prob1 = jac(prob1, deltas(transition_index,bit_index)); end end extrinsic_uncoded_llrs(bit_index) = prob0-prob1; end end
B.2 Symbol based BCJR algorithm for TCM decoders

%BCJR algorithm for TCM decoder- A symbol based log -MAP algorithm-%function %Cherian Danny Joseph
function extrinsic_uncoded_llrs = tcmdecoder(apriori_uncoded_llrs,rx, channel,N0,k)
% All calculations are performed in the logarithmic domain in order to % avoid numerical issues. These occur in the normal domain, because some of % the confidences can get smaller than the smallest number the computer can % store.
% Matrix to describe the trellis % Each row describes one transition in the trellis % Each state is allocated an index 1,2,3,... Note that this list starts % from 1 rather than 0.
% transitions = [
FromState, 1,
ToState, 1,
UncodedBit2, 0,
UncodedBit1 0,
Symbol 1+0i;

1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 8, 2, 3, 4, 5, 6, 7, 8, 2, 1, 4, 3, 3, 6, 5, 8, 3, 4, 1, 2, 7, 8, 5, 6, 4, 3, 2, 1, 8, 7, 6, 5, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0.00+1i; -1.00+0i; -0.00-1i; 0.7071+0.7071i; -0.7071 + 0.7071i; -0.7071-0.707i; 0.7071-0.707i; 1+0i; 0.00+1i; -1.00+0i; -0.00-1i; 0.7071+0.7071i; -0.7071 + 0.7071i; -0.7071-0.707i; 0.7071-0.707i; 1+0i; 0.00+1i; -1.00+0i; -0.00-1i; 0.7071+0.7071i; -0.7071 + 0.7071i; -0.7071-0.707i; 0.7071-0.707i; 1+0i; 0.00+1i; -1.00+0i; -0.00-1i; 0.7071+0.7071i; -0.7071 + 0.7071i; -0.7071-0.707i; 0.7071-0.707i;]
87
% Find the largest state index in the transitions matrix state_count = max(max(transitions(:,1)),max(transitions(:,2)));
% Calculate the a priori transition log-confidences by adding the % log-confidences associated with each corresponding bit value.
gammas1=zeros(size(transitions,1),length(rx));
%Calculate the symbol based probablities for symbol_index=1:length(rx) for transition_index=1:size(transitions,1)

gammas1(transition_index,symbol_index)= -(abs(rx(symbol_index)-channel*transitions(transition_index,5))2)/N0; end end
88
reshapellrs=reshape(apriori_uncoded_llrs,2, length(apriori_uncoded_llrs)/2);
gammas2=zeros(size(transitions,1),length(rx)); for bit_index=1:length(rx) for transition_index=1:size(transitions,1) if transitions(transition_index, 3)==0 gammas2(transition_index,bit_index)= gammas2(transition_index,bit_index)+reshapellrs(2,bit_index); end if transitions(transition_index, 4)==0 gammas2(transition_index,bit_index)= gammas2(transition_index,bit_index)+reshapellrs(1,bit_index); end end end gammas=gammas1+gammas2;
% Forward recursion to calculate state log-confidences.
alphas=zeros(state_count,length(apriori_uncoded_llrs)/2); alphas=alphas-inf; alphas(1,1)=0; % We know that this is the first state
for state_index = 2:state_count % We know that this is *not* the first state (a log-confidence of % minus infinity is equivalent to a confidence of 0) alphas(state_index,1)=-inf; end for bit_index = 2:length(apriori_uncoded_llrs)/2 for transition_index = 1:size(transitions,1) alphas(transitions(transition_index,2),bit_index) = jac(alphas(transitions(transition_index,2),bit_index), alphas(transitions(transition_index,1),bit_index-1) + gammas(transition_index, bit_index-1)); end end
% Backwards recursion to calculate state log-confidences.

betas=zeros(state_count,length(apriori_uncoded_llrs)/2); betas=betas-inf; for state_index = 1:state_count % The final state could be any one of these betas(state_index,length(apriori_uncoded_llrs)/2)=0; end for bit_index = length(apriori_uncoded_llrs)/2-1:-1:1 for transition_index = 1:size(transitions,1) betas(transitions(transition_index,1),bit_index) = jac(betas(transitions(transition_index,1),bit_index), betas(transitions(transition_index,2),bit_index+1) + gammas(transition_index, bit_index+1)); end end
89
% Calculate a posteriori transition log-confidences.
deltas=zeros(size(transitions,1),length(apriori_uncoded_llrs)/2); for bit_index = 1:length(apriori_uncoded_llrs)/2 for transition_index = 1:size(transitions,1) deltas(transition_index, bit_index) = alphas(transitions(transition_index,1),bit_index) + gammas(transition_index, bit_index) + betas(transitions(transition_index,2),bit_index); end end
% Calculate the a posteriori LLRs.
aposteriori_uncoded2_llrs=zeros(1,length(apriori_uncoded_llrs)/2); for bit_index = 1:length(apriori_uncoded_llrs)/2 prob0=-inf; prob1=-inf; for transition_index = 1:size(transitions,1) if transitions(transition_index,3)==0 prob0 = jac(prob0, deltas(transition_index,bit_index)); else prob1 = jac(prob1, deltas(transition_index,bit_index)); end end aposteriori_uncoded2_llrs(bit_index) = prob0-prob1; end
% Calculate the a posteriori LLRs.
B.3. 2/3 RATE CONVOLUTIONAL ENOCDER
90
aposteriori_uncoded1_llrs=zeros(1,length(apriori_uncoded_llrs)/2); for bit_index = 1:length(apriori_uncoded_llrs)/2 prob0=-inf; prob1=-inf; for transition_index = 1:size(transitions,1) if transitions(transition_index,4)==0 prob0 = jac(prob0, deltas(transition_index,bit_index)); else prob1 = jac(prob1, deltas(transition_index,bit_index)); end end aposteriori_uncoded1_llrs(bit_index) = prob0-prob1; end
aposteriori_uncoded=[aposteriori_uncoded1_llrs; aposteriori_uncoded2_llrs]; extrinsic_uncoded_llrs=reshape(aposteriori_uncoded,1, length(apriori_uncoded_llrs));
end
B.3 2/3 rate Convolutional enocder

% Convolutional encoder for BICM-ID for a rate 2/3 and code memory=3 function [enocdedbits] = convolutional_encoder(u1,u0,k)
%Initialisation of Shift registers s0=0;s1=0;s2=0;
for bit_index=1:length(u1);
% Calculation of encoded bits encoded3_bits(bit_index)=mod((u0(bit_index)+s1),2); encoded2_bits(bit_index)=mod(u1(bit_index)+s0,2); encoded1_bits(bit_index)=mod(s0+u1(bit_index)+u0(bit_index)+s1+s2,2);
%Next states of convolutional encoder if (bit_index =length(u1)) s0=u0(bit_index); s1=s2; s2=u1(bit_index); end
B.4. BICM-ID MODULATION FOR 8 PSK USING NATURAL MAPPING
91
%reshaping of the en bmat=[encoded3_bits;encoded2_bits;encoded1_bits]; enocdedbits=reshape(bmat,1,3*length(encoded1_bits));
B.4 BICM-ID modulation for 8 PSK using Natural Mapping

%%% Modulation function for BICM-ID %Cherian Danny Joseph
function tx = modulate(bits)
%Arrangement of bit labels by natural bit mapping bit_labels = [0,0,0; 0,0,1; 0, 1,0; 0,1,1;1,0,0; 1,0,1; 1 1,0; 1 1,1];
k = size(bit_labels,2); M = 2k; for bit_count=0:M-1 constellation_points(bit_count+1,1)=cos((2*pi*bit_count)/8)+1i*sin((2*pi*bit_count)/8); end
if isequal(size(constellation_points),[M,1]) || isequal(size(bit_labels),[M,k]) || isequal(size(bits),[1,k]) error(wrong dimensions); end
for symbol_index = 1:M if isequal(bit_labels(symbol_index,:),bits) tx = constellation_points(symbol_index); end end end
B.5 Soft Demodulation for 8 PSK

%%%%%%Soft demodulation function for BICM-ID %Cherian Danny Joseph function [extrinsic_llrs3,extrinsic_llrs2,extrinsic_llrs1] = soft_demodulate(apriori_llrs, rx, channel, N0)
%Arrangement of bit labels bit_labels = [0,0,0; 0,0,1; 0 1,0; 0 1,1;1,0,0; 1,0,1; 1 1,0; 1 1,1];
%bit mapping
B.5. SOFT DEMODULATION FOR 8 PSK

k = size(bit_labels,2); M = 2k; for bit_count=0:M-1 constellation_points(bit_count+1,1)=cos((2*pi*bit_count)/8)+1i*sin((2*pi*bit_count)/8); end
92
%error checking if isequal(size(constellation_points),[M,1]) || isequal(size(bit_labels),[M,k]) || isequal(size(apriori_llrs),[1,k]) error(wrong dimensions); end
%Calculation of symbol probablites symbol_probabilities = -abs(rx-channel*constellation_points).2/N0;
%soft demdmodulated LLR -o/ps extrinsic_llrs = zeros(size(apriori_llrs)); for bit_index = 1:k p0 = -inf; p1 = -inf;
symbol_probabilities2 = symbol_probabilities; for symbol_index = 1:M
for bit_index2 = 1:k if bit_index2 = bit_index if apriori_llrs(bit_index2) = inf if bit_labels(symbol_index, bit_index2) == 0 symbol_probabilities2(symbol_index) = symbol_probabilities2(symbol_index) + apriori_llrs(bit_index2); end else if bit_labels(symbol_index, bit_index2) == 1 symbol_probabilities2(symbol_index) = -inf; end end end end
if bit_labels(symbol_index,bit_index) == 0 p0 = jac(p0, symbol_probabilities2(symbol_index)); else p1 = jac(p1, symbol_probabilities2(symbol_index)); end
B.6. EXIT BAND CHART FOR INNER CODE OF BICM-ID(DEMAPPER)
93
end
extrinsic_llrs(bit_index) = p0 - p1;
end
%reshaping---the calculated extrinsic infomrations mat=zeros; mat=reshape(extrinsic_llrs,3,1); extrinsic_llrs3=mat(1,:); extrinsic_llrs2=mat(2,:); extrinsic_llrs1=mat(3,:);
end
B.6 EXIT band chart for inner code of BICM-ID(demapper)

% Copyright (C) 2010 Cherian Danny Joseph % EXIT inner band chart script for BICM-ID
clear all;
% number of bits to encode bit_count=999;totalbit=bit_count;
% Number of a priori mutual informations to consider IA_count=11;
% Use the histrogram method for measuring mutual informations histogram=0;
%choose how mnay frames to simulate frame_count = 100;
% Channel SNR in dB SNR =0;
%To calculate the area area1=0;
% Noise variance N0 = 1/10(SNR/10);

channel=1;
94
% Calculate the MIs to use for the a priori LLRs IAs = (0:(IA_count-1))/(IA_count-1); IAs(1,11)=0.999; IE_means = zeros(1,IA_count); IE_stds = zeros(1,IA_count);
for IA_index = 1:IA_count
IEs = zeros(1,frame_count);
area= zeros(1,frame_count);
for frame_index = 1:frame_count % Generate some random bits bits = round(rand(1,bit_count));
% s/p conversion b_mat=reshape(bits,3,(bit_count)/3); u_1=b_mat(1,:); u_2=b_mat(2,:); u_3=b_mat(3,:);
for bit_index=0:totalbit/3-1 tx(bit_index+1)=modulate(bits(3*bit_index+1:3*(bit_index+1))); end
%Txm the modulated symbols rx=tx+sqrt(N0/2)*(randn(1,length(tx))+1i*randn(1,length(tx)));
apriori_llrs = generate_llrs(bits, IAs(IA_index));
for bit_index1=0: (length(rx))-1 [apriori_encoded3_llrs(bit_index1+1),apriori_encoded2_llrs(bit_index1+1), apriori_encoded1_llrs(bit_index1+1)]= soft_demodulate(apriori_llrs(3*bit_index1+1:3*(bit_index1+1)),rx(bit_index1+1),1,N0); end
apri_mat=[apriori_encoded3_llrs;apriori_encoded2_llrs;apriori_encoded1_llrs];
apriori_uncoded_llrs=reshape(apri_mat,1,bit_count);

if histogram
95
IEs(frame_index) = measure_mutual_information_histogram(apriori_uncoded_llrs, bits); else IEs(frame_index) = (measure_mutual_information_averaging(apriori_encoded3_llrs) +measure_mutual_information_averaging(apriori_encoded2_llrs) +measure_mutual_information_averaging(apriori_encoded1_llrs))/3; end
end % Store the mean and standard deviation of the result IE_means(IA_index) = mean(IEs); IE_stds(IA_index) = std(IEs); %area1=area1+area(frame_index);
end
% Create a figure to plot the results. %figure; %axis square; hold on; xlabel(I(a_a)); ylabel(I(a_e)); xlim([0,1]); ylim([0,1]);
% Plot the EXIT function for component decoder 1 hold on; plot(IAs,IE_means); plot(IAs,IE_means+IE_stds,--); plot(IAs,IE_means-IE_stds,--);

Dis Sera Tat Ion

Uploaded by

Copyright:

Available Formats

Dis Sera Tat Ion

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Dis Sera Tat Ion

Uploaded by

Copyright:

Available Formats

UNIVERSITY OF SOUTHAMPTON Faculty of Engineering, Science and Mathematics School of Electronics and Computer Science

23th September 2010

SUPERVISOR: Dr. Robert G . Maunder

Viterbi Algorithm -(VA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

2.6.1 2.6.2 2.7 2.8

Virterbi Hard Decoding . . . . . . . . . . . . . . . . . . . . . . . . . . Virterbi Soft Decoding . . . . . . . . . . . . . . . . . . . . . . . . . .

Case Study:7 PSK over AWGN channel . . . . . . . . . . . . . . . . . . . . . Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 Conclusions and Future Work 5.1 5.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

B.6 EXIT band chart for inner code of BICM-ID(demapper) . . . . . . . . . . . .

1.1. A HISTORICAL BACKGROUND ON CODED MODULATION

Implementational complexity Channel charactristics

Bit Error Rate

Figure 1.1: Factors affecting the design of Channel/Modulation Scheme [1]

1.1 A Historical Background on Coded Modulation

1.2. MOTIVATION AND ORGANISATION OF THIS PROJECT

1.2 Motivation and Organisation of this project

1.2. MOTIVATION AND ORGANISATION OF THIS PROJECT

1.2.2 Chapter Organisation

Background Literature and Review

2.1. MODULATION SCHEME

Figure 2.1: Modulation and Demodulation of QPSK

Figure 2.2: Constellation Diagram for 4 and 8-PSK

where E is the energy per symbol and fc is the carrier frequency.

2.3. MAPPING SCHEMES

2.3 Mapping Schemes

Constellation points of 7 PSK 2

Figure 2.5: 7-PSK constellation diagram

Comparison of Capacity Curve between 7 and 8 PSK schemes 3 8PSK 7PSK

Bits / channel use

Figure 2.6: 7-PSK and 8-PSK capacity compariosn

2.5. CONVOLUTIONAL CODE

where E is the energy in each respective waveforms when 1 signal.

Figure 2.7: Convolutional Encoder of rate 1/2

2.5 Convolutional Code

2.5. CONVOLUTIONAL CODE

2.5.1 State and Trellis Diagram

be V, then the State Transition equation can be written as:

2.5. CONVOLUTIONAL CODE

Table 2.1: State Transition Table

2.6. VITERBI ALGORITHM -(VA)

2.6 Viterbi Algorithm -(VA)

2.6.1 Virterbi Hard Decoding

2.7. PUNCTURING CONVOLUTION CODES

2.6.2 Virterbi Soft Decoding

2.7 Puncturing Convolution codes

2.8. TRELLIS CODED MODULATION

2.8 Trellis Coded Modulation

2.8.1 TCM Principle

2.8. TRELLIS CODED MODULATION

1 011(3)0 1 100(4) 0 1 101(5)0

4-PSK Signal Set

8-PSK Signal Set

Figure 2.9: Constellation Diagram for 4 and 8-PSK

2.8. TRELLIS CODED MODULATION

Figure 2.10: 8 PSK set partitioning [2],Ungerboeck

2.8. TRELLIS CODED MODULATION

0426 1537 4062 5173 2604 3715 6240 7351

Figure 2.11: Encoder and trellis for the eight-state 8PSK

2.8.2 TCM encoder

2.8. TRELLIS CODED MODULATION