Citation
Efficient communications for nonsymmetric information sources with application to picture transmission

Material Information

Title:
Efficient communications for nonsymmetric information sources with application to picture transmission
Creator:
Emami, Shahriar ( Dissertant )
Miller, Scott L. ( Thesis advisor )
Couch, Leon W. ( Reviewer )
Childers, Donald G. ( Reviewer )
Najafi, Fazil ( Reviewer )
Sigmon, Kermit ( Reviewer )
Phillips, Winfred M. ( Degree grantor )
Place of Publication:
Gainesville, Fla.
Publisher:
University of Florida
Publication Date:
Copyright Date:
1993
Language:
English
Physical Description:
vii, 159 leaves : ill., photos ; 29 cm.

Subjects

Subjects / Keywords:
Alphabets ( jstor )
Bandwidth ( jstor )
BCH codes ( jstor )
Constellations ( jstor )
Decryption ( jstor )
Error rates ( jstor )
Polynomials ( jstor )
Signals ( jstor )
Statistics ( jstor )
Supernova remnants ( jstor )
Digital communications ( lcsh )
Digital modulation ( lcsh )
Dissertations, Academic -- Electrical Engineering -- UF ( lcsh )
Electrical Engineering thesis Ph. D ( lcsh )
Modulation (Electronics) ( lcsh )
Signal processing -- Digital techniques ( lcsh )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Abstract:
This dissertation is concerned with issues related to nonsymmetric information sources. Signal design, alphabet size selection and decoding of information from these sources are among the topics covered in this dissertation. although the techniques presented here are applicable to any nonsymmetric source, the emphasis is placed on video sources. Initially a model for the statistics is placed on video sources. Initially a model for the statistics of DPCM (Differential Pulse Code Modulation) of video signals is derived and it is shown that DPCM of video signals results in a nonsymmetric source. The problem of signal selection for nonsymmetric sources in two dimensions is considered. Iterative methods for finding the minimum error signal (and minimum cost) constellation subject to an average (or a peak) power constraint are presented. Even though efficient techniques for source coding, channel coding and signal design exists, it is not known how the choice of alphabet size affects a communication system. Image transmission systems with various alphabet sizes are compared on the basis of equal information rate, bandwidth and average power. The systems employing various alphabet sizes are analyzed and computer simulations are performed using pictures with different amount of details. An optimum procedure for decoding Markov sequences is developed and the path metric is derived. A heuristic tree searching algorithm. Two other techniques for decoding Markov sequences, a symbol-by-symbol modified MAP (MMAP) receiver using higher order statistics and a maximum signal-to-noise ratio (MSNR) receiver, are also given. The decoding procedures were applied to image communication over noisy channels. In summary, the major contributions of this dissertation were the development of signal selection methods for nonsymmetric sources, derivation of procedures for decoding of correlated sources and application of these procedures to the picture communication in noisy situations.
Thesis:
Thesis (Ph. D.)--University of Florida, 1993.
Bibliography:
Includes bibliographical references (leaves 155-158).
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Shahriar Emami.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
Copyright Shahriar Emami. Permission granted to the University of Florida to digitize, archive and distribute this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.
Resource Identifier:
001933237 ( ALEPH )
30784206 ( OCLC )
AKA9304 ( NOTIS )

Downloads

This item has the following downloads:


Full Text













EFFICIENT COMMUNICATIONS FOR NONSYMMETRIC INFORMATION
SOURCES WITH APPLICATION TO PICTURE TRANSMISSION





















By

SHAHRIAR EMAMI


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


1993
















To my parents for their love, patience and support.















ACKNOWLEDGEMENTS


I wish to express my gratitude to my advisor, Dr. Scott

L. Miller, for his encouragement, support and friendship. I am also grateful to Dr. Couch for his guidance and insight. In addition, I would like to thank Dr. Childers and Dr. Najafi

for their time and interest in serving in my committee. Special thanks also go to Dr. Sigmon who has offered his valuable assistance throughout the course of this study.


iii
















TABLE OF CONTENTS



ACKNOWLEDGEMENTS. . . . . . . .

ABSTRACT.vi

CHAPTER ONE REVIEW OF MODULATION TECHNIQUES, SOURCE AND
CHANNEL CODING.1

1.1 Introduction.1 1.2 Two-Dimensional Modulation Formats.2 1.3 Source Coding.5 1.4 DPCM.5 1.5 Transform Coding.8 1.6 Channel Coding.10 1.7 Transmission Errors in a DPCM system . . . . 15 1.8 Optimum Prediction for Noisy Channels . . . . 16 1.9 Research Objectives.18 1.10 Description of Chapters.20

CHAPTER TWO DPCM VIDEO SIGNAL: A NONSYMMETRIC
INFORMATION SOURCE.21

2.1 Introduction.21 2.2 Basics of Quantizers.21 2.3 Approaches to Quantizer Design.23 2.4 MSQE Quantizer Design.23 2.5 Analysis of DPCM Encoder.26 2.6 Results.29 2.7 Discussion.31

CHAPTER THREE SIGNAL DESIGN FOR NONSYMMETRIC SOURCES .35

3.1 Introduction.35
3.2 Maximum Likelihood Signal Design for Three
Signals.37
3.3 A Numerical Approach Based on Lagrange
Multipliers Method.48
3.4 Minimum Error Signal Selection.51 3.5 Minimum Average Cost Signal Selection . . . 58 3.6 Results.66











CHAPTER FOUR ALPHABET SIZE SELECTION
FOR VIDEO SIGNAL CODING. . . . .

4.1 Introduction .
4.2 Preliminaries.
4.3 Analysis . . . . . . . . . . . . . . . . . .
4.4 Implementation Issues. . . . . . .
4.5 Nonbinary BCH Codes. .
4.6 Results.
4.7 Summary.

CHAPTER FIVE EFFICIENT DECODING OF CORRELATED SEQUENCES


5.1 5.2 5.3
5.4 5.5


Introduction .
Optimum Decoding of Markov Sequences A Modified MAP (MMAP) Receiver .
A Minimum Cost Decoder .
A Maximum Signal-To-Noise Ratio (MSNR)


Receiver .
5.6 Redundancy in the Encoded signals
5.7 Picture Transmission over noisy
Channels .
5.8 Side Information .
59 Summary. . CHAPTER SIX CONCLUSIONS AND SUMMARY . .

6.1 Summary of the Work. . . .
6.2 Directions of Future Research. APPENDIX A. EVALUATION OF AN INTEGRAL . APPENDIX B. EVALUATION OF THE DERIVATIVES REFERENCES .

BIOGRAPHICAL SKETCH. . . . .


79

79 80 83 86
94 99
104

116


116 119
122 123


* . . . 124
* . . . 127

* . . . 130
* . . . 144
* . . . . 145

* . . . 147

* . . . 147
* . . . 150

151


153 155 159















Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

EFFICIENT COMMUNICATIONS FOR NONSYMMETRIC INFORMATION
SOURCES WITH APPLICATION TO PICTURE TRANSMISSION By

Shahriar Emami

August 1993

Chairperson: Dr. Scott L. Miller Major Department: Electrical Engineering

This dissertation is concerned with issues related to nonsymmetric information sources. Signal design, alphabet size selection and decoding of information from these sources are among the topics covered in this dissertation. Although the techniques presented here are applicable to any nonsymmetric source, the emphasis is placed on video sources. Initially a

model for the statistics of DPCM (Differential Pulse Code Modulation) of video signals is derived and it is shown that DPCM of video signals results in a nonsymmetric source.

The problem of signal selection for nonsymmetric sources in two dimensions is considered. Iterative methods for finding the minimum error signal (and minimum cost) constellation subject to an average (or a peak) power constraint are presented.









Even though efficient techniques for source coding, channel coding and signal design exists, it is not known how the choice of alphabet size affects a communication system.

Image transmission systems with various alphabet sizes are compared on the basis of equal information rate, bandwidth and average power. The systems employing various alphabet sizes are analyzed and computer simulations are performed using pictures with different amount of details.

An optimum procedure for decoding Markov sequences is developed and the path metric is derived. A heuristic tree searching algorithm is employed to obtain a suboptimum solution.

Two other techniques for decoding Markov sequences, a symbol-by-symbol modified MAP (MMAP) receiver using higher order statistics and a maximum signal-to-noise ratio (MSNR)

receiver, are also given. The decoding procedures were applied to image communication over noisy channels.

In summary, the major contributions of this dissertation were the development of signal selection methods for

nonsymmetric sources, derivation of procedures for decoding of correlated sources and application of these procedures to the picture communication in noisy situations.


vii
















CHAPTER ONE
REVIEW OF MODULATION TECHNIQUES, SOURCE AND CHANNEL CODING IN DIGITAL COMMUNICATIONS

1.1 Introduction


The goal of this chapter is to present the background necessary to follow the work presented in this dissertation,

introduce the research objectives and give a brief description of the chapters.

Since two-dimensional modulation has been utilized in this work, these formats are reviewed, their spectral efficiency are calculated and the upper bound on their spectral efficiency is given. Picture transmission is one of

the applications presented here. To familiarize the reader with the field, a number of source coding techniques have been described. Error correction over noisy channels is an

important topic and has been utilized in this thesis. The gain from channel coding has been explained and different methods of error correction have been discussed. The effects of

channel errors in a DPCM system and optimum prediction in noisy channels was presented, because the enhancement of

picture quality in DPCM systems has been addressed in this dissertation.













1.2 Two-Dimensional Modulation Formats



M-ary phase-shift keying (MPSK) and quadrature amplitude modulation (QAM) are among the most popular two dimensional formats. In MPSK the transmitted signal is given by


s(t)=Re{g(t)ect} (i.i)

where

g(t) =A ejO(c) ,

(1.2)
8( )-(i i)M i=1,2, . .,M.


In other words, in MPSK while the amplitude is maintained constant the phase of signal can take on one of the M values in a symbol interval. The MPSK for M=4 is called quadrature phase-shift keying (QPSK).

In MPSK signal points are confined to the circumference of a circle. But in QAM the transmitted signal is s(t)=Re{g(t) e jct} (1.3)

where

g(t)=x(t)+jy(t), (1.4)
















s(t)=x(t) cos( Ct)-y(t) sin(act). (1.5)

The waveforms x(t) and y(t) are x(t) =xi h(t-iT) (1.6)
i

and


y(t)=Eyi h(t-iT), (1.7)
i

where T is the symbol interval in seconds and h(t) is the pulse shape.

Let us find the spectral efficiency of MPSK and QAM with rectangular pulses. The null-to-null transmission bandwidth of MPSK and QAM is


2R (1.8)


where M=21 is the number of points in the signal constellation. The spectral efficiency is therefore given by R - 1 bits/sec (1.9)
BT2 Hz

When operating over a bandlimited channel and the overall pulse shape satisfies the raised cosine rolloff filter characteristics, the bandwidth of the modulating signal is













B- (I1+r) D (1.10)
2



where D=R/l and r is the rolloff factor. Since BT =2B, the transmiss ion bandwidth of QAM is


BT=. (1 R (1.11)


and the spectral efficiency with raised cosine filtering is given by

1og2M bits/sec (1.12)
(1+r) Hz

The spectral efficiency increases with the number of the points- in the constellation. However, one can not increase the spectral efficiency by increasing the number of points in the

signal constellation too much, because as you place more signals in the constellation the error rate increases.

For reliable communications, the information rate must be kept below the channel capacity. Therefore, the spectral efficiency is upper bounded by


q <10g2(1+-S) (1.13)
N

where SIN is the signal-to-noise power ratio.

Two dimensional formats are well suited for high speed

data transmission because of their efficient use of bandwidth. However, they require coherent detection that implies the need for synchronization circuits.













1.3 Source Coding



The purpose of source coding is to remove as much redundancy as possible from the message. Efficient coding of

messages provides the opportunity for significantly decreasing the transmission costs. Two main approaches to picture coding predictive coding and transform coding will be addressed here.



1.4 DPCM



There is considerable correlation between adjacent

samples of speech or image data, and indeed the correlation is significant even between samples that are several sampling intervals apart. The meaning of this high correlation is that, in an average sense, the signal does not change rapidly from

sample to sample so that the difference between adjacent samples should have a lower variance than the variance of the signal itself.

The predicted value is the output of the predictor

system, whose input is a quantized version of the input signal. The difference signal may also be -called the prediction error signal, since it is the amount by which the predictor fails to exactly predict the input.

Since the variance of error signal is smaller than the

variance of signal, a quantizer with a given number of levels









6

can be adjusted to give a smaller quantization error than would be possible when quantizing the input directly.



1.4.1 Optimum Prediction



We are interested in linear prediction of the form
N

R(n) =E a. x(n-j) (1.14)


which is the weighted sum of N previous samples. The weights aj are linear prediction coefficients. The filter is optimized by finding the weights that minimize prediction error in a mean squared sense

Ge2=E[ (x(n) -k(n) ) 2] (1.15)

Since mean squared error is a function of aj and

e2
-- =0 ie, . . (1.16)
3ai


is a necessary condition for minimum MSE (mean-squared error). Evaluating the derivative gives E[-2x(n) - (n) 8 (n) ] (1.17)
ai

Equating this to zero yields,

E[tx(n)-k(n)x(n-i) =0; i=1,2,.,N. (1.18) This is called the orthogonality principles which states that minimum error must be orthogonal to all data used in the









7

prediction. The expansion of this equation gives the following condition for optimum a,


N
IaRi(k-j)-=Ri(k); k=l,2,.,N
j.1.


(1.19)


*Rx,(N-1) ' (al IR.,(1)
* R,(N-2) a2 = R_ (2)

� (0Io) N) Rx .,N)


(1.20)


in matrix notation,


(1.21)


where


R.=Q (i-j[)); r==R,(i)i: i,j=l,2, . ,N (1.22) The equations are called normal equations, Yule-Walker prediction equations or Wiener-Hopf equations.

The mean squared error decreases significantly by using up to three elements in predictive coding. However, if the coefficients are not exactly matched to the statistics of a picture, the decrease in mean squared error is not significant by using three previous elements as compared to one [1].


(R,, (0) Rx (1) RXX (2) R,, ( 1) ax (0) R,, (1)

,R, ( -1) "R. (-2) ., =(N-3)









8

1.5 Transform Coding



In transform coding a picture is divided into subpictures and then each of these subpictures are transformed into a set of independent coefficients. The coefficients are then quantized and coded for the transmission. An inverse

transformation is applied to recover intensities of picture element. Much of the compression is a result of dropping coefficients from transmission that are small and coarsely quantizing the others as required by the picture quality.

It is desirable to have a transform-which compacts most

of the image energy in as few coefficients as possible. Another consideration is the ease of implementation.



1.5.1 Optimum Transform



Optimum transform (K-L transform) is explicitly known, but computationally it is very demanding. This undesirable feature has prevented any hardware implementation of the

optimum transform. It is mainly studied in simulations to obtain bounds.

The most practical transform coding techniques are based on the DCT (discrete cosine transform), which provides a good compromise between information packing ability and

computational complexity. In fact, the properties of DCT have proved of such practical value that it has become the









9

international standard for transform coding systems. In

addition to that it minimizes the blocking artifact that, results when the boundaries between the subimages become visible.



1.5.2 Size of Subpictures



Computer simulations on real pictures show that the mean square error-produced by transform coding improves with the size of subpicture. However, the improvement is not

significant as subpicture is increased beyond blocksize of 16x16. Subjective quality of pictures, however, does not

appear to improve with the size of block beyond 4x4 pixels [2].



1.5.3 Bit Allocation



One method of choosing the coefficients for transmission, is to evaluate the coefficient variances on a set of average

picture, and then discard all the coefficients whose variance is lower than a certain value. Such a scheme is called zonal filtering [3)1.

Having decided which coefficients to transmit, we must then design a quantizer for each of them. This could be done by dividing a given total number of bits among all the coefficients. In order to minimize the mean square error for









10

a given total number of bits f or Gaussian variables, the optimum assignment is done by making the average quantization

error of each coef f icient the same. This requires that the bits be assigned to the coef f clients in proportion to the logarithm of their variance.



1.6 Channel Coding



Channel coding is a method of inserting structured redundancy into the source data so that transmission errors can be identified and corrected. Block coding and convolutional coding are two important subcategories of channel coding techniques.



1.6.1 Block Codes



With block coding the source data is f irst segmented into blocks of k bits; each block can represent any of M=2 k distinct messages. The encoder transforms each message into a
k
larger block of n digits. This set of 2 coded messages is called a code block. The (n-k) digits, which the encoder adds

to each message block, are called redundant digits; they carry no new information. The ratio of data bits to total bits within a block, k/n, is called code rate. The code itself is referred to as an (nk) code.

To demonstrate the performance improvement possible with









11

channel coding, let us pick a (15,11) single error correcting code. Assume a BPSK modulation, a signal-to-noise ratio of 43,776 ( S/No=43,776) and a data rate of R=4800 b/s. Let Pub and Purn represent the bit and message error rate for the uncoded system and Pcb P,m represent the bit and message error rate for the coded system, respectively. without coding

Eb S =9.6dB (1.23)
No RNo





Pb=Q( _b ) =1.02x10-5 (1.24)
u~b N0



and


P um=l(l-Pub) 11=l.12xlO-4. (1.25)

With coding

RC=4800x (15/11) =6545 b/s Eb_ S -8.25 dB
No R.No (1.26)


Pc,b=Q( Eb) =1.36X10-4 N0



The bit error rate for the coded system is inferior to that of the uncoded system and the performance improvement due to











coding is not apparent yet.

Since the code corrects all single errors within a block of 15 bits, the message error rate for the coded system will be

15
Pcm3() Pcb(l.pb) 1-k.94x10- (1.27)


It is seen by comparing the error rates that the message error rate has improved by a f actor of 58 through the use of a block code.

Most of the research on block codes has been concentrated on a subclass of linear codes known as cyclic codes. A cyclic

code word, after any number of cyclic shifts, has the property of remaining a valid code word from the original set of code words. Cyclic codes are attractive because they can be easily implemented with feedback shift registers. The decoding methods are simple and efficient.

Examples of cyclic and related codes are BCH, ReedSolomon, Hamming, Reed-Muller, Golay,. quadratic residue and

Goppa codes. The classes form overlapping sets so that a particular code may be a BCH code and also a quadratic residue code. Recent applications of these codes to digital

communication include a (31,15) Reed-Solomon code for joint tactical information distribution system(JTIDS) and a (127,112) BCH code for INTELSAT V system [4].











1.6.2 Convolutional Codes



A convolutional encoder consists of some shift registers and modulo-2 summers. For the general case, k bits at a time

are entered into the shift register, and the code rate is k/n. The state of the encoder is dependent upon the contents of the shift registers.

Convolutional codes can be described by a code tree. It

is seen that the tree contains redundant information which can be eliminated by merging, at any level, all nodes

corresponding to the same encoder state. The redrawing of the tree with merging paths has been called a trellis by Forney.

The problem of decoding a convolutional code can be thought of as attempting to find a path through the trellis or the tree by making use of some decoding rule.

The Viterbi algorithm (5) which is shown to be a maximum

likelihood decoder for convolutional codes, involves computing a metric between the received signal and the trellis path entering each state. In the event that two paths terminating

on a given state are redundant, the one having the largest metric is stored (the surviving path) . This selection of survivor is performed on for all paths entering each of the

other states. The decoder continues in this way to advance deeper into the trellis, making decisions by eliminating the least likely paths.

The complexity of Viterbi algorithm is an exponential









14

function of the code's constraint length. For large values of

constraint length (K>>10) one might consider other decoding algorithms.

The complexity of sequential decoders is relatively

independent of constraint length, so codes with much larger constraint length can be used. Also this technique is more suitable than Viterbi algorithm for low bit error rates.

Sequential decoding (5] was first introduced by Wozencraft but the most widely used algorithm to date is due

to Fano. It is an efficient method for finding the most probable code word, given the received sequence, without searching the entire tree. The explored path is probably only

local; that is, the procedure is sub-optimum. The search is performed in a sequential manner, always operating on a single path, but the decoder can back up and change previous

decisions. Each time the decoder moves forward, a tentative decision is made. If an incorrect decision is made, subsequent extensions of the path will be wrong. The decoder will eventually be able to recognize the situation. When this happens, a substantial amount of computation is needed to recover the correct path. Backtracking and trying alternate paths continue until it finally decodes successfully.

Convolutional codes using either Viterbi or sequential decoding have the ability to utilize whatever soft-decision

information might be available to the decoder. It is not surprising that they have been used widely even though their









15

theory is not as mathematically as profound as that of the block codes. Most good convolutional codes have been found by computer search rather than algebraic construction.



1.7 Transmission Errors in a DPCM System



Differential PCM systems are affected differently by bit errors than PCM systems because the DPCM decoder loop causes an error propagation, while a PCM error does not propagate in

time. Subjectively, DPCM is more error-robust than PCM in speech coding, but less robust than PCM for image coding.

Assume a channel error changes channel input u(n) to a

wrong value v(n) . Due to linearity of the decoder filter, the correct computation of output is superposed by an error output caused by input c(n)=u(n)-v(n) to the decoder loop.

Since the decoder is an all-pole filter, there will -be an infinite sequence of error samples at the output, with

decaying amplitudes. In the case of f irst-order prediction, the effect, on a future value, at time m, is described by (3] C(m) =C(n) a (m-n); m~tn (1.28)

Transmission errors therefore propagate in the reconstructed DPCM waveform.

This kind of error smearing is perceptually desirable in speech coding where a POM error spike of large magnitude is more annoying than a low amplitude error smeared over a long duration.









16

In picture coding, on the other hand, error propagation is perceptually very undesirable, taking the form of very visible streaks or blotches with one and two dimensional predictors.



1.8 Optimum Prediction for Noisy Channels



One of the early approaches to system optimization under noisy conditions was presented by Chang and Donaldson [6]. Because of the importance of the result and the relevance to this dissertation a summary of the methods is given here.

Let ri denote the received signal, fi denote the impulse response of the DPCM decoder. The output of the decoder is therefore given by Ri=ri *f= (s.+n) * fi
(1.29)
=si * f +n, * f =xi +q +ni* f i




Let us define x'i as follows

x/ =Aj-xi :qji �f

var (Xi) =E[QI2] +2E fk E(ei1i-k]


1 k













Let us assume the channel noise is uncorrelated and the difference signal samples are statistically independent, then


var (X,) =E[Q2] +2E[QiNi] +E[Ni2] F fk2 (1.31)
k

The second term is called the mutual error and can be shown to be approximately zero if the quantizer is near optimum. The error power reduces to


var (X'1) =E[Q2] +E[Ni2] fk2. (1.32)
k

The sum can be evaluated using an identity


fk2=l+a2+a4+a6+.
k
(1.33)
1 -b
1-a2

The expression for the reconstruction error variance then becomes

var (X'i) =var (Q.) +b var (Ni) . (1.34) We now define the following quantities to relate the quantizer and noise variances to the differential signal variance, var (Qj) =kq var (Ei), (1.35)

var(N.) =Kn var(Ei) .


The DPCM prediction gain is also given by












Svar(Xi) =1+a 2-2a Pi, (1.36)
var (E.)


where P, is equal to R,,(O). Putting all this together yields


var (XI) =(kq+b kn)' var (X) (1.37)
Gp


The second term above is the dominant term because the effect of channel noise is much more destructive to the reconstruction of the image than the effect due to quantization noise,

1-2a p1+a 2
var(X,) =K, var(X) 1-a2 (1.38)


To minimize the variance of the reconstruction error we will set the derivative of this expression with respect to a to zero. The optimum value of a turns out to be


a= fi-- 1 (1.39)
Pi




1.9 Research Objectives



This dissertation is concerned with issues related to nonsymmetric information sources. To motivate the work on nonsymmetric sources, it is shown that DPCM of digitized video signals results in a nonsymmetric information source. One of the main goals is to address the problem of signal design in









19

two dimensions for nonsymmetric sources. It is desired to find an algorithmic solution to the minimum error signal constellation for average and peak power constraints. In

addition the general case where the cost function is not necessarily the error rate is discussed.

Even though efficient techniques for source coding, channel coding and signal design exists, it is not known how

the choice of alphabet size affects a communication system. We would like to compare communication systems with various alphabet sizes for the transmission of video signals on the basis of equal information rate, bandwidth and average power.

Two realistic situations will be considered, when one is operating under tight bandwidth constraint and when the constraint is somewhat loose.

System performance can be improved using standard error

correction techniques at the cost of increasing the bandwidth or reducing the information rate. However, we would like to use inherent asymmetry and redundancy in the transmitted picture to improve the reception. We will model the data as a Markov source and derive the optimum method for decoding the data. We will also find a receiver that instead of minimizing the error rate maximizes the SNR (signal-to-noise) ratio.










20

1.10 Description of Chapters



A review of background material relevant to this dissertation is given in Chapter one. Two dimensional

modulation techniques, source coding techniques f or images, standard error correcting techniques and the ef fect of channel errors on predictive system are among the topics addressed in this chapter. In Chapter Two a DPCM system will be analyzed and a model for the statistics of the source will be derived. It will be shown theoretically and empirically that DPCM of video signals produces nonsymmetric sources.

The issues of signal design are addressed in Chapter Three. Algorithmic solutions to signal design for nonsymmetric information sources under average and peak power constraints

for minimizing the error rate and average cost are presented.

The study on the role of alphabet size for nonsymmetric

sources in a communication system is given in Chapter Four. In Chapter Five various methods for decoding Markov sequences are presented. The application to the transmission of video signals over noisy channels and a comparison of the method is also given. Chapter six contains a summary of presented

approaches, conclusions and comments regarding the future research directions.















CHAPTER TWO
DPCM VIDEO SIGNAL: A NONSYMMETRIC INFORMATION SOURCE


2.1 Introduction



The purpose of this chapter is to demonstrate that DPCM (differential pulse code modulation) of pictures results in a nonsymmetric information source.

To do so, some introductory material is presented first. Since a quantizer is an important component of a DPCM system, it will be examined in some detail. Quantizers will be introduced, different criteria for the design will be mentioned and the procedure for finding an optimal quantizer (in MSQE sense) will be explained step by step. A DPCM encoder will be analyzed and a model for the resulting source statistics will be given. Eventually the model will be compared with actual picture statistics and results will be compared.



2.2 Basics of Ouantizers



Quantization is the process of rounding sample values to a finite number of discrete values. This process is not an information preserving process and the reconstructed signal is 21









22

only as good as the quantized samples allow. In other words, there remains some error, the quantization error between the original and the reconstructed waveform which is related to the parameters of the quantizer.

.Let the analog signal be modeled as a random waveform and let p(x) be the probability density function of the signal.

The process of -quantization subdivides the range of the values of x into a number of discrete intervals. If a particular sample value of the analog signal falls anywhere in a given

interval, it is assigned a single discrete value corresponding to that interval. The intervals fall between boundaries denoted by XIX21 . . .XL+I, where there are L intervals. The quantized values are denoted by 11,121 .1L and are called quantum levels or representative levels. The width of an interval is x14., -xi and is called interval's step size. If all the steps are equal and, in addition, the quantum level separations are all the same, the quantizer is said to be uniform; otherwise it is a nonuniform quantizer.

It is possible to design a quantizer for a given

probability density function and a given number of levels. The optimal quantizer is non-uniform unless the signal has a uniform pdf. If a uniform quantizer is used instead the mean

squared quantization error will be larger than that of the optimal nonuniform quantizer.









23

2.3 Approaches to Quantizer Design



The quantizers can be designed based on a mean squared error criterion. This results in overspecification of the low detailed areas of picture and consequently a small amount of granular noise but relatively poor reproduction of edges.

It has been realized for sometime that for a better picture quality, quantizers should be designed on the basis of psychovisual criteria [7). One method of designing psychovisual quantizers is to minimize a weighted mean squared quantization error, where the weights are derived from subjective experiments (8]. Such optimization would be similar to mean squared error criteria, where the density function is replaced by a weighing function.


2.4 MSOE uantizer DesiQn



An optimal quantizer is defined to be a quantizer with the smallest mean squared quantization error. It is desired to find the quantizer that minimizes the mean-squared quantization error for a given probability density function and number of levels. The mean-squared quantization error is given by











X2 L- X+1


(x-l) 2p (X) dx. (2.1)

XL

Our purpose is to choose quantum levels 1i and interval boundaries xi so that eq. (2.1) is minimized.

The above expression can be differentiated to obtain a set of necessary conditions that must hold for the optimum quantizer. By applying Leibniz's rule.we get


(2.2)
a .=-2 f i=1,2, ,
ali






as-- =[(x-i_.1) 2-(x-1)2]p(x)dx=O, i=2,.,L, (2.3)
aXi



where x,=-o and XL+1= . Equation (2.3) is equivalent to i -) (2.4)


which says that interval boundaries should fall midway between the adjacent quantum levels. Alternatively,


I i=2xi-1i-1(


(2.5)









25

Equation (2.2) is readily solved for li

xl*1
f x p(x)K
, ili.1 (2.6)
f p (x)cbx
XI



The solution of the equations for the general nonuniform quantizer is difficult. However, a procedure to obtain a solution by computer iteration has been introduced by Llyod and Max [9), [10]. For a specified probability density function and a fixed value of L, 11 is first selected arbitrarily. with

xi=0,we solve eq. (2.6) for X2. Next X2 and 11 are used in eq. (2.5) to obtain 12. The process is repeated to obtain X3 from eq. (2.6) and 13 from eq. (2.5). Continued iteration finally stops when 1L is obtained from eq. (2.5). If 1, has been correctly guessed, then 1L Will satisfy the equation with xL+l =00. If it does not, 11 is corrected to a new value and the process is repeated until 1L satisfies eq. (2.6). This

procedure satisfies conditions eq. (2.5) and eq. (2.6) which are necessary for optimality.

Max [10) used the above procedure to find the quantum levels for a zero-mean Gaussian message for quantizers up to 36 levels. Paez and Glisson [11] used the procedure to find optimum levels for signals having either a gamma density or a Laplace density for L=2,4,8,16, and 32.














2.5 Analysis of DPCM Encoder



Consider an information source with alphabet {a,, a2,., aQ}. The information source is said to be nonsymmetric if the source symbols are not equally likely. It will be shown that the output of a DPCM encoder can be viewed as a nonsymmetric information source.

Assume that in a DPCM encoder a predictor of order M is used. Let us model the quantizer as a additive noise source &(n)=&(n)+q(n) (2.7)


where e(n) and 9(n) are the input and the output of the


quantizer respectively and q(n) is the quantization noise. It has been shown that (12]

6")e~ ) (2.8)
@(n)=e~n -Ebi q(n-i)
i-i


where e(n) is the difference signal in a DPCM without the quantizer and bi are the prediction coefficients. The distribution of 0(n) is therefore given by the


M
convolution of the pdf's of e(n) and Q=Zbi q(n-i) i-1












p�=pe p. (0.

And the probability of quantum level i is determined by


PxfP(xdK (2
Xi


x (n)


e (n)


A
e (n)


e(n)


A
x (n)


(b)

Figure 2.1. Block diagram of a DPCM system. decoder.


(a) encoder and


2.9)


.10)












O'Neal [13) has shown experimentally that the pdf of 6(n)


can be approximated with a Laplacian distribution


p(x) =-1 exp(-6I .XI (2.11)
V~co a


To find the statistics of the levels an optimum quantizer must be placed in the DPCM system. If we utilize a MSQE quantizer, the statistics associated with each level will be



Pi=f exp(- YTIX1) dx=f-L exp(-F2y) dy
1(2.12) =- (exp (-r a) -exp (-V2 0)
2

where a and # are tabulated in [11).

We have also verified through simulations that the choice of a gamma distribution for e(n) and uniform pdf for Q results in a satisfactory approximation to the density of (n) . In


other words



p, (x) 8,n x)2a 2VTa)

(2.13)

A numerical method must be employed to find the statistics associated with each level.









29

2.6 Results

In this section we will compute the statistics of the quantizers for the model developed earlier and will compare them to actual source statistics for real world pictures. Two

types of quantizer were used, a MSQE quantizer and a quantizer that was found by pschovisual experiments [14].

The material used were two eight-bit pictures; a low detail picture LENNA and a high detail picture AERIAL MAP (Fig.2.2 and Fig. 2.3). Both pictures consist of 512x512 pixels.

-An optimized eight-level quantizer for the, Laplacian distribution was chosen. Model parameters can be estimated by a 2 (2.14)



Table 2.1 contains theoretical and actual source statistics. Theoretical values are seen to be reasonably close to actual source statistics.

Then a seven-level quantizer [14] that is shown to work

well with different pictures was selected. The histograms for the two pictures were prepared. The Laplacian distribution and the distribution given in eq (2.13) were compared with the histograms (Figures 2.4 - 2.5). They both seem to be a fair approximation to the histograms.

Table 2.2 gives the statistics given by equation eq. (2.13), Laplacian pdf and the picture. The model based statistics are close to the actual source statistics.































Fig. 2.2. Lenna.


Fig. 2.3. Aerial map.









31

2.7 Discussion



It was shown that the pdf of input to the quantizer can be fairly approximated by equation eq. (2.13) (which is the convolution of a gamma and a uniform pdf) or simply a Laplacian distribution. The theoretical statistics derived from the models agreed well with the actual source statistics. It is seen that DPCM of video signals does in fact produce a nonsymmetric source.












TABLE 2.1. A EIGHT LEVEL MSQE QUANTIZER IS USED. THE FIRST AND THE SECOND COLUMNS SHOW THE ACTUAL SOURCE STATISTICS FOR THE PICTURES. THE THIRD COLUMN SHOWS THE STATISTICS USING A LAPLACIAN DISTRIBUTION FOR SOURCE.



Lenna Aerial Map Model Statistics

.3324 .2800 .2549

.3574 .2795 .2549

.0946 .1416 .1510

.1060 .1390 .1510

.0337 .0578 .0744

.0331 .0573 .074

.0225 .0220 .0197

.0197 .0224 .0197


TABLE 2.2. STATISTICS FOR LENNA PICTURE. A SEVEN LEVEL QUANTIZER FOUND BY PSYCHOVISUAL EXPERIMENTS IS USED. I



Laplace Eq. (13) Actual Statistics

.5541 .5967 .6578

.1965 .1558 .1544

.1965 .1558 .1418

.0259 .0270 .0193

.0259 .0270 .0200

.0005 .0019 .0026

.0005 .0019 .0038















8.3










a~(a)
8.z 8.2

8.15

0. 1

8 ---- "
-88 -68 -48 -28 8 26 46 68 86


(a)








8.2




8.2. Tp


8.1


8.85S

8I

-88 -68 -48 -ZO 8 28 48 68 68


(b)


Figure 2.2. The histogram for LENNA picture is compared with
(a) Laplacian Distribution (b) Distribution given in eq. (2.13). Histogram is shown with a broken line.















8.12 8.1 0.68
//





8.84
8. 82 j-80 -60 -40 -28 0 20 40 68 8e


(a)





8.12





I
0.1
8.8/


9.86




0.0
8.ez ,



-88 -60 -48 -20 8 20 48 60 88





(b)




Figure 2.3. The histogram for AERIAL MAP picture is compared with (a) Laplacian Distribution (b) Distribution given in equation (13). Histogram is shown with a broken line.















CHAPTER THREE
SIGNAL SELECTION FOR NONSYMMETRIC SOURCES

3.1 Introduction


In many applications one has a bandlimited channel and

has to achieve the least error rate for. a given signal-tonoise ratio. Design of high speed modems is one example where

the designer is faced with the- problem of selecting an efficient set of signals with in-phase and quadrature components.

The objective in signal design is to find the optimum signal constellation in presence of additive white Gaussian

noise under a power constraint. Two dimensional modulation formats such as MPSK and QAM have been studied before [15).

These formats confine the signal points to a certain geometry and are not optimum in the sense of minimum error rate.

There has been a few attempts to solve the signal design problem under peak or average power constraint without constraining the signal points to a special geometry such as a circle or a certain lattice. Foschini et al [16] presented an iterative approach for signal selection. A gradient search

procedure is given that incorporates a radial contraction technique to meet the average signal power constraint.

Kernighan and Lin [17] came up with a heuristic procedure for









36

solving signal design problem under a peak power constraint.

Previous investigations on signal design have focused on signal selection for equally likely signals (15]-(17]. There

are some applications where the information source is nonsymmetric. A practical instance in which such a model proves

rewarding is in the transmission of video signals. It was demonstrated in Chapter Two that DPCM of digitized video signals results in a nonsymmetric source. In this case signals should be mapped into a two dimensional signal constellation in an optimum manner. In other words the goal is to determine

the signal constellation that minimizes the probability of error (or a given cost function) in presence of additive white Gaussian noise under an average power(or a peak power) constraint, given N signals with unequal probabilities.

To illustrate the difficulty of direct approach we will

design a ML receiver for a three signal constellation. We will also describe a numerical method that uses the Lagrange

multipliers method for optimization. These two methods are appropriate for smaller signal sets.

Then a number of iterative algorithms are developed. First a normalized conjugate gradient search algorithm and a gradient search algorithm will be presented that can be applied to signal sets of any size and with any probability

distribution. The methods presented here are applicable to the design of both MAP (maximum a posteriori) and ML (maximum likelihood) receivers. These methods are generalizations and










37

modifications to the method given in (16]. Then a gradient search method for a peak power constraint is developed. Eventually a gradient search method that finds a signal constellation for an average cost function subject to a peak or average power constraint is presented. In the end, a few examples are given and conclusions are drawn.



3.2 Maximum Likelihood Signal Design for Three Signals



Here a three signal constellation is designed for a 3 symbol source for transmission on a white Gaussian noise channel. The signal constellation is depicted in Figure 3.1. From geometrical considerations a1+a 2+a3+P1+P2+P3=2n, (3.1)


AH cosp1=BH cosa, (3.2)


AH cosa2=CH cosP2, (3.3)


BH COSP3=CH cosa3. (3.4)



Let r(t)=(rl,r2) be the received waveform and p(t) be a particular value of r(t). Suppose p(t)=(p,p2) is received in the symbol interval, p,(r,,r21mi) is the conditional joint probability density of random variables defining r(t) and s1, S2. s3 are the signal vectors denoted by A, B and C on the









38

constellation. A maximum likelihood receiver sets the message estimate to mk if for i=1,2,3, I

pr(r1=p1Ir2=p21 Mk) > p.(r1=pl, r2=p21 mi) iok. (3.5) Assuming the noise components are Gaussian and statistically independent,


P (P 1'P21 m()=-- exp_ 1 [(p1-sil) 2+ (P2-si2 ) 2] .
7N, N


(3.6)


The decision rule then becomes


-1 [ (p1_Sk) 2+ (P2_sk2) 2] >. -__ [ (p-si,) 2+ (p2_S12) 2] N0 Ik N.


itk.


(3.7) Notice that the sum of squared terms on either side of the


c


Figure 3.1. A three signal constellation


inequality are the square of the Euclidean distances between the received signal and signals si and sk. Thus the decision rule can be rewritten as:












d(pSi)>- d(p,Sk)


where d(x,y) is the Euclidean distance between x and y. Point H is the intersection of the decision regions and it is therefore on the boundaries of the decision regions. For the boundary points the above inequality changes to an equality,


d(H,A)=d(H,B)


d(H,A) =d(H, C)


(3.9)


(3.10)


AH = BH = CH,


and this results in


(3.11)


(3.12)


C(2=P2,


( 3 = P3�


(3.13)


Define P, to be the probability of word error. The probability of symbol error is one minus the probability that a symbol is correct. The probability of a correct symbol is obtained by averaging all of the conditional probabilities of a correct symbol, we have


(3.14)


Pw=1- P(C I Mi) pi.


But P(Cjmj) is the result of point p falling in the decision


i ok.


(3.8)











region Ii (Figure 3.2),


3
Pw=-1, PiffPr'(PIP2, mi)dp~dp2.
i-1.


(3.15)


2


Figure 3.2. One decision region.


By substituting the expressions for probability density function eq. (3.6) into eq. (3.15) we get


___ 3 (p1-d) 2 p
Pli-I --fexp[- No dp
S PP2


(3.16)


or equivalently











- 1ptana

0PIO P2--plta


7op-.0



-nNop f


ptanc1

f
P2--pltanp pltand3

p2--pltanp3


L2 (p1-AH) 2 p2
exp [- -P -!L2-] dp2dpl np N0 N�


(p1-AH) 2 ,dp2dp
exp [-(P- - - P2 ] dp2dpl N0 N


(p 1 -AH) 2 P2
exp - -] dp 2dp1.


By making the following changes of variables


E=22 - dp2=dEJ-N


(3.18)


u=2-1 - dpj=duVIN
071


(3.19)


the probability of error in terms of the


new variables


becomes


(3.17)


3













p-i ex[-(u- AH2


- U
--- f exp[-(u- AH2


u.0 - 0 ,

P- FI2 Ut
l~fexp[-(u- ~2)
f- 0


utana2

f
c--utanp,


tana1

f
utar4


ana3

f
utanp3


exp [-e2] de du


exp [-C2 de du exp [-e2] de du.


(3.20)


Now by definition


X2
erf(x,x2) =-f fexp (-t2) dt.
F#CX1


(3.21)


Therefore, the following is the equation for P,:


=1- , f exp[-(u- - )2]erf(-u tan1'u tana2)du





PP2 fexp-u- AH2
2 1xu-0 - -0 )2] erf(-u tanj3,1utana1)du




- IV, -- o)2]erf(-u tanP2,u tan3) du.


(3.22)









43

For a given signal to noise ratio the probability of word

error depends only on a, and a2 because a,, a2 and a3 add up to 71. To find the values of aI a2 , a3 that minimize the

probability of the word error, we need to differentiate P, with respect to al and a2 and set them equal to zero:




a P o (Uf e22-(- )2 u2tan2 )
w=- 1-fe-- (u (l+tan a,) e- du
aUa0


P2 fe x
u-O


(U (I+tan2a1) e u2tan2a-U (l+tan (a+a2) )e-u2tan2 (a +d2)) du



u-3) 22
f_ e 1-~~a (OC1+C2) )e-u'2" ") du=O,
U-0


(3.23)













a" (U_- )
---w=- p' e N./ , (u(I +tana2) e-u2tan2 2) du
aa 2 7C u-0




U=O


---- 3 e d
U-0


( 1 (l+tan2 (a+a2) ) eu't a ) +u (l+tan2a2)eutan'2 du) =0.


(3.24)

Inspection shows that above equations consist of only one type of integral which can be expressed in a computable form,


u (u-k) 2e-u du=f u e-((1+1)u2+k2-2uk) du
0 0

(3.25)

By producing a perfect square in the exponent

k21 - -(1+1)(u- it k du fu e-((ll)u2+k2-2uk) du= e ue (1+1) du
0 0

(3.26)

and using the results obtained in Appendix A we get:












k2l k2
u e-(u-k)2e-U21 du=- 1-- e + /2-n k erf( -k ) 2(1+1) 2 12_(i+i)






(3.27)

If we substitute for integrals from eq. (3.27) into eq. (3.23) and eq. (3.24) we will end up with the following set of nonlinear equations:

--- AH2 s1er( A2co~I
(Pl-P3) e + 7tA (p +p2) e NO osa1erf(- ,f2cosalI 00)



-(p2+p3)e -- ICos(S . caa)lerf(- AHEcos (.,+.2)l, -)N =0


(3.28)


A A2 - -(P1P2) eN +C (p1+P3) e NO UosN1 A21cos=


- H2sin2 (al+OC2) A 2
(P2 +P3) e NOIcOs (CC1+2) jerf (- -A o s (C1 +a 2)1'o) =0


(3.29)

Given a signal-to-noise ratio and a probability set the optimum angles can be found using eq. (3.28) and eq. (3.29).









46

We experimented with a variety of signal-to-noise ratios and

probability sets. Figure 3.3 shows the optimum a, for three probability sets over a wide signal-to-noise ratio range. In

all cases optimum a, approaches 60 degrees as signal to noise ratio goes up. Even when the optimum angle is somewhat different from 60 degrees from the performance point of view the two system are almost indistinguishable.







02 61
00 -10




Figure 3.3. a, (deg.) versus average power (dB) . (a) the top curves, source statistics {.9,.05,.051, (b) The middle curve, source statistics and (c) the bottom curve, source
statistics {.6,.2,.2j.



The presented results suggest that the three signals should be placed on the vertices of an equilateral triangle (Figure 3.4). The origin of the signal constellation can be shifted to a new location that minimizes the average energy

without affecting the probability of error. Let us define 1 to be

1= AB = BC = AC

in triangle ABC. Since ABC is an equilateral triangle, AH and











1 are related by:


1 =AHvf.


(3.30)


The point G(x,y), can be found as follows:


paveP,[ (X-I)2+Yl] +p2[ (x+!)2+y2] +p3 [X2+ (y-I3)2]
2 2 2

(3.31)


- (P1-P2) 1
2


- y=P31/.
2


Substituting x and y into (30) and solving for 1 we get

1= rk Pave"


(3.34)


where k is given by


7--


(4/3)


(3.35)


aP., =0 ax


a, =0 ay


(3.32)


(3.33)


P1 I (PI-P2-1) 2+3P32] +P2 [ (Pl-P2+I) 2+3P32] +P3 [ (pl-p2) 2+3 (P3-1) 2]













C (o, 113/2)


B(-I/2,0)


A(1/2,0)


Figure 3.4. The three signals form a equilateral triangle.


3.3 A Numerical Approach Based on Lagrange Multipliers Method


In this approach the problem of signal selection is viewed as a constrained optimization problem. The constraints are incorporated into the optimization problem by the use of Lagrange multipliers method.

Let us outline the design of a MAP receiver with three signals with unequal probabilities (Figure 3.1).

A MAP receiver decision rule is m=mk if, for i=1,2,3,


pr=(r1=P11r2=P2I k)Pk' p1(1P11-r2=p21 mi)pi i~k.


(3.36)


Upon substituting (6) into (43), we get


.G (X, y)











2 2
in (Pk) -Eoj- (pj-skj) in(pi) --LE (pj-Sij)2 ij.


(3.37)

Assume point H is the intersection of decision regions. For point H the inequality turns into an equality and the sums on either side will be the distances between H and signals si and sj, therefore

AH2 -BH2 =N' in ( 1 P2


(3.38)

Similarly we obtain

AH2-CH2=N in(P) .

P3


(3.39)

In addition to the two above MAP constraints there are four geometric constraints (eq.(3.1) through eq.(3.4)), and an additional constraint to minimize the average energy. The origin of the constellation is shifted to a location that minimizes the average power. Then the average power function evaluated at that point is set to 1/N.

p a+p2b+P3 C= Pave

(3.40)


where











a=(x-AH sinp1)2+y2 b= (X+BH sinl) 2+y2
c= (x-CH sin (3.+a2+32)) 2+ (y-AH cos31+CH cos (Pl+c2+P2)) 2
x=p1AH sin31-p2BH sina1+p3CH sin(P1+C(2+P2)
y=p3AH cosP1-p3 CH COS (P1+a2+P2) .


Now the problem is to minimize a function of nine variables under seven equality constraints. The method of Lagrange multipliers can be applied to this problem. We need to define a new cost function F of sixteen variables and take the partial derivatives with respect to the sixteen variables and set them equal to zero.

cost function=F(a11,2,a3, j$P213,AHBH, CH, X11, 2*31I,4,15, 16,A



If we eliminate Xi's among the sixteen equations we end up with a nonlinear set of equations of the other nine variables which must be solved numerically.

This method becomes quite complicated as the number of the signals in the constellation grows. It was mainly used to verify the solutions obtained by the gradient based method when N was relatively small.









51

3.4 Minimum Error Signal Selection

3.4.1 Preliminaries



Let us denote the signals in the constellation by

SIS2,.,SN where si=(xi,yi). The average power is given by N (3.41)
Pave=Y pIIsol2
n-I



Where {P, ,P2 ,-. -PN} is the probability distribution of the signal set and ls.l1 is the magnitude of s,. Following the convention of [16] we choose to set P.v =1/N and define the signal-to-noise ratio to be SNR=I0 log10( No) (3.42)
N0

where No is the one sided power spectral density of white noise.

By definition the error rate (probability of symbol error) is

N (3.43)
Pei=EPn Pr(error Isn) n-1

The union bound gives the upper bound for the conditional probability of error. For a maximum likelihood receiver we have:













Pr (errortsn) < E) (3.44)
.1=1
ion

For large values of signal-to-noise ratio the conditional probability of error will be equal to the upper bound. By using approximation for Q function when the argument is large and plugging back in (3.43) we get


p N 1 exp[- Isi-s-I121 (3.45)
7En1 i1-SnII 4N,

Similarly it can be shown that the symbol error rate for a MAP receiver is given by


(IsiSnN2+Nln( Pn))2
Pe= N Pn Isi-Sn' exp[- 4N�1Si-_Sn|2 P
n-1 ion is i-snl12 +Nln (Pn) Pi





(3.46)



3.4.2 Gradient Search Algorithms for an Average Power Constraint


In search of the minimum, the gradient of the probability of error is obtained analytically and an iterative gradient search algorithm which modifies the constellation at each iteration is used to find the optimum constellation.

In the gradient search algorithm, the iterative rule is given by












S;1 =Sk-akVFk


where


(3.48)


denotes the signal vector at the kth step of the algorithm, ak is the step size and VF is the gradient of PC. Since the signal power may change with k, the signal vector is normalized at each step of the algorithm


sk.1= .
. N(pls* (k+l) 12 +P21S (k+l) 12+.2+p.sN(k+l) 12)





(3.49)

To speed up the convergence, instead of a conventional gradient search algorithm, the Fletcher and Reeves conjugate gradient method [18] can be utilized. In this method the information about the second derivative is used indirectly. The algorithm is described by


S;+1 =.,sk+akh~k


(3.50)


where


hk=-VFk+akk-1


(3.51)


and


(3.47)













tk=(VFkV t vFk -(3.52) [vFk-J t: vFk-l


in which ak is the step size and hk is the quantity in this algorithm that combines the information from current and previous steps to define a new direction. similar to gradient search algorithm, the power may alter with k. Therefore the

signal vector needs to be normalized at each step of the algorithm.

Let us summarize the procedure for iterative methods:



1. Set k=Q and select the starting points.

2. Determine the search direction by calculating hk for the conjugate gradient method and VF for the gradient method.

3. Find the improved signals coordinates and normalize them. If the improvement is smaller than a tolerance level stop, otherwise set k=k+l and go to step 2.



3.4.3 Analytical Expressions for the Gradient Vector



Let us find the gradient vector for the ML receiver first. VF the gradient of Pe is a vector of 2N components in signal space, in which each signal occupies two dimensions. The gradient as a vector of N two dimensional vector components is represented by












92 vF=


where


(3.53)


gk=g9k,g k


gk is obtained by taking the derivative of P, with respect


to sk


gk=-E (Pk+Pi)
ivk


Sk-si ISj-Sil2
Ilsk-sexp[ 4No]


I 1 +
Ils -s-1 2N


Similarly one can find the kth component of VF for a MAP receiver


2
tl
pklSk~S exp [- 4NoI SkSi 12]


+p Skl exp [ - 112
4N0 IISk-S-i12


t,
2
t2


t2 + t2
t 2NojSk-s1l2


2Nollsk-sII2 }


(3.55)


where


(3.54)


i~k

















and


SkS Sk-Si ~ISk-si sk-sill





t2Isk-ISki+NlD n (- )
Pi





t2=lsk-si 1 -Noln (-)
Pi


3.4.4 Startina


Points


The average signal power is equal to I/N or

N
12
n-1N


Thus, individual signals must satisfy







Solving the inequality for sn we get


1 1


which states that starting points must be selected from inside


(3.56)


(3.57)






(3.58)


(3.59)


(3.60)


(3.61)


. . I .













a circle with radius







3.4.5 Gradient Search Algorithm for the Peak Power Constraint



Even though the average power constraint is used more often, in certain applications such as space communication the peak power constraint is much more realistic. Given a peak power constraint the transmitted signal points must be placed inside the circle such that the error rate is minimum. We would like to modify the iterative procedure developed earlier to accommodate the peak power constraint. All we need to do is to further modify the modified constellation at each iteration to insure that peak signal power is bounded.

Let the peak power be P=l/N. The signal set is modified using the iterative rule (gradient search or conjugate gradient search), let


Mmax {Isi*k+1) 11)}. (3.62)
1:5i N

We will modify the signal set again in the following manner













s (k+i)


2~ (k+l)
Sk.l /, j (3.63)


s. (k+i)



to meet the peak power constraint.



3.5 Minimum Average Cost Signal Selection



The communication channels are almost never error free. The probability of error (error rate) is a measure of system

performance. Various error types are usually weighted equally but in some communication system certain errors are more costly than the others. The Bayes receiver allows us to rank

the different error types [19]. To utilize the Bayes receiver, one must know the source statistics and a reasonable estimate of a cost matrix must be obtained. The Bayes receiver requires more apriori knowledge about the communication system than others but it results in a superior performance if there is no mismatch between the design and operating conditions.

The goal is to find a signal constellation that minimizes the average cost subject to a peak or average power constraint. First a workable expression for the average cost will be obtained.











3.5.1 Average Cost Function



The Bayes receiver which minimizes the average cost is given by

N N
C=>3 (3P(sjIsi)pi Lji, (3.64)
1-1 j-1


where Li is cost involved when the receiver picks si when si was actually sent.

The boundaries of the decision regions are no longer straight lines; instead they are two dimensional curves. For instance, the boundary between si and si is given by

N N
SLik P(-rlISk) Pk= F L3 k P (rlISk) Pk,3.5
k-I k-i



where conditional probability of error is p(.rlsk)= 1 ex ( S-Sk12 (.6
~No


Channel noise could cause the received signal to move to an adjacent decision region and result in an error. For large values of signal-to-noise ratio; however, errors are almost entirely the result of displacement of a received signal to an adjacent decision region from the nearest point on a boundary to the transmitted signal. Let si be the transmitted signal and sji be the point with shortest distance on the boundary









60

with sj. p(sj si) can be approximated with a Q function

P(sjJsj) =Q( llsi-sjil )
V(3.67)

substituting back into the average cost expression we get

N s- N e[SiSJ2 (3.68)
2 i id i~iPJ- ns lexp - N
2 n Lit is SJII N.



3.5.2 The Nearest Points on the Boundaries



The first step in computing the average cost function is to find the set of nearest point to each and every signal point in the constellation on the boundaries. There are N signal points in the constellation and for each one there are (N-l) of nearest points on (N-i) boundaries, therefore a total of N(N-l) of nearest points must be found.

The problem of finding sji which is the nearest point to si on the boundary between si and si can be formulated like a Lagrange multiplier problem. We would like to minimize lIs-si1 subject to a constraint (eq. (3.65)). Let us form the

auxiliary function

N N
4=:ls-si2+1(E Lik p(rIsk)pk-ELik p(rIsk)pk), (3.69) k-i k-i


and differentiate it with respect to x,y and X












- =0, ay

;aY=0.
al


Eliminating equations


(3.70)


X we will come up with the following set of


N
(XXi) E (Lik-Ljk) (x-xk) exp(- ) )Pk i - k=1 No
(Y-Yi) Y (Lik-Ljk) (Y-Yk) exp (.- 112
k-I N.


(3.71)


NIS-Sk112 NS Ski J2
E Lik exp(- S-k)p N exp(- )pk=0 (3.72)
k-1 No k-1 o



where si=(xi,yi) and s=(x,y)=sji.


3.5.3 The Gradient of the Cost Function


To evaluate gk , first the portion of cost function that depends on sk must be formed:

N J sk-sjkl112]
E Ljk PISkSjkl exp[+ Lki 1 exp lSi-Skill2
-2 Pi Isi-Skill 1 N0

gkI and gk2 the components of gk are found by taking the derivative of above expression with respect to the components














Lk Pk 1 exp[- DSk-Sk
II ks-skD N.

L p 1 exp [- ISi-Skil2]
LkiPi i Ski 11 No


1 1
jsk-S I112 N
1 1-L
+- 11 DHs2-sk~iI{ No


(3.73)


where


aXjk) + (ykYk (
ul = (Xk-Xjk) (1 - ayj)k )Xk)
)aXk aXk




aXkl aYki


l Lik PkS I S1 exp [- Sk-SJk12]


V2LkiPI 1 exp N Si-Ski 2
V2Lk P ls-ski 1 N.


1 11
ISk-Sjk


1-L N,


I 1 1
Isi-ski ll No


(3.76)


where


(aXjk) + (Yk-Yjk)
ayk


of sk


N
j.1
N
-E U2 i-i


(3.74) (3.75)


and


N j.1
N
1-1


(1 - ayJk)
ayk


(3.77)


V1 = (xk-_xjk)












v2=(xxk)-) a +a(y-Y) (- (3.78)



The gradient of the cost function depends not only on the nearest points on the boundary but also on their derivatives. Let us find the derivative of the components of sji with respect to the components of s,. Differentiating eq. (3.71) and (3.72) with respect to xi and yj we get

N ISSk12 N w-s 12
Lik W exp - I ]pk=LJk wlexp k ]Pk
k-1 N0 k-1 N,

(3.79)

where


Wl = (X-Xk) a+ (Yyk)ay


(3.80)

IS-Sk12 N Is -Sk 2
E Lik w2 exp [ ]Pk= Lk w2 exp[- ]Pk
k-1 N k-1 N.

(3.81)

where

w2= (x-xk) i+ (Y-Yk) X ay1


(3.82)










ax N
ax -1) EN (Lik-Ljk) (Y-yk) exp PkI-sU
ax1 k-1 N
N ay- zl ss
+(x-xN) (Lik-Ljk) ( -2 (y-yk)---)exp[- ISSk
k-1 ax1 N. N.
N
E (LIk-Ljk) (x-xk) exp - is-skI2
Nx ax1 N. kI
(y-yi) E (Lik-Lk) ( ax -2 (x-xk) -l S-exp[I- N
k-i 1 0 0o IV. No


(3.83)
1


where


zl = (x-xk) (-Lx)
axi


+ (Y-yk) ( ay
ax1


-yE k-i (Lik-Ljk) (Y-Yk) exp[- iss Pk
ak-1 No
N ays-ski
X-Xi) E (Lk a -2 ( o) ) exp [ NS = ( ay -1) (LikLik) (x-xk)exp []
N k-iS 12___(Y N xz Is-skll:
i) (Lik-Ljk) (l. -2 (x-xk) -L' ) exp [ ' k-1 ay, 1 IV.


12]Pk


2
-]Pk


(3.85)


where


Z2 = (X-Xk) (- x) + (Y-Yk)
ay1


(er) ay1


(3.86)


It can be shown that (refer to Appendix B)


D 2 ] Pk


2
] Pk


(3.84)












axi =0 IIxj,=0'
axi ayX



(3.87)

and


aIYJ1 =0 "~ 0
axi aYJ


(3.88)

Equations (3.75) and (3.78) imply that v2=u2=0. Contrary to the equations for finding sji, the equations for finding its derivatives are linear.



3.5.4 The Iterative Algorithm



To find the signal constellation we will start from an initial guess. An iterative rule (either gradient search or conjugate gradient search algorithm) is selected. An improved signal set is found using the iterative rule. Depending on the type of constraint (peak or average power constraint) the constellation is modified accordingly (eq.(15) or eq.(30)). The iteration is continued until a minima is reached.

The above procedure is valid under the assumption of large channel SNR. When channel SNR is not large, obtaining an analytical expression for the gradient is not feasible because of the complexity of the shape of the decision regions. An obvious solution is an exhaustive search which is









66

computationally very expensive and long. However, one can use a modified form of the above procedure to save on the

computational expenses. In the modified procedure, the average cost is computed based on the received data and the gradient based update rule is replaced with the directional set (Powell's) method in multidimensions [20) which does not require the gradient.



3.5.5 Application to DPCM System



In a DPCM system some errors are more costly than others, therefore we could benefit from utilizing a Bayes receiver. In Chapter Five an optimum method of selecting the cost matrix will be presented.



3.6 Results



In this section we present some numerical results,

elaborate on the design procedure and discuss the performance of the methods described in this chapter.

The probability of error, in general, is not a convex function of the signal set, therefore the algorithm can converge at local as well as global minima. The multi start technique can be applied to the global optimization problem

[16]. In this technique one selects an optimization technique and runs it from a number of different starting points. The









67

set of all terminating points hopefully includes the global

minimum point. our implementation of the technique is somewhat different. To come up with a set of start points, First a search procedure is implemented that finds signal sets having small error rates initiating from random start points. About

thirty start points are selected. Then we run the gradient search techniques from these starting points.

To run the algorithms we start with a small step size and monitor the changes in probability of error as a function of

iteration number. if P, changes very slowly we can proceed to increase the step size. On the other hand, we may need to

decrease the step size if P, starts to oscillate around a minima or goes unstable.

The best constellations for a number of sources with three, four, five, seven and eight signals were found. Figure 3.6 shows the best constellations for three, five and seven signals with the probability distributions given in Table I.

We compared the speed of convergence of the conjugate gradient method to that of the conventional gradient search method for a given value of step size. The normalized

conjugate gradient method converges to a solution much faster than the normalized gradient method. A set of initial start points and the optimum constellation for seven equally likely signals are shown are in Figure 3.7. The conjugate gradient method is faster about an order of magnitude.

To evaluate the gain in using a non-symmetric signal














TABLE 3. 1. VALUES OF SIGNAL-TO-NOISE RATIO AND THE PROBABILITY DISTRIBUTIONS OF NON-SYMMETRIC SOURCE.


TABLE 3.2. SOURCE STATISTICS FOR THE FOUR SIGNAL SOURCE.


source statisticsI .35, .35, .15, .15


Source Number SNR(dB) Probability Distribution

1 12.2 .8,.1 1.1

2 9 .9,.025,.025,.025,.025

3 8 _.96,.01,.01,.O01,.O1

4 15.5 .674, .142, .142, .018, .018, .003, .003






































-1 -8.5 8 8.5 I 1.5


(a)


x

-I -8.5 8 8.5 1 1.5


Figure 3.6. Best signal constellations for non-symmetric sources. Probability distributions are given in Table 3.1.
(a) Source 1.(b) Source 2. (c) Source 3. (d) source 4.


x x


8.5 -


-8.5F


-1.5


-8.5I-


-1.51
-1.5





(C)


-i b,


(d)


8.5 1 1.5


Figure 3.6--continued





























-6.5-


-1.5 -


18-1




I.
.
tS.
IV


-9.5 8 9.5 1 1.5


a 288 488 688
number of iterations


B8B 1800


Figure 3.7. (a) Seven signal constellation for an equally likely source. O's represent the start points and x's represent the final constellation. (b) The solid and dashed curves show the error rate as a function of number of iterations for gradient search and conjugate gradient method respectively.


CI 0
K
0
x 0 Ox
a 0
xx


a









72

design relative to a equally likely signal selection, a first

order Gauss-Markov source with a correlation coefficient of 0.9 was synthesized. The output was encoded using a first order DPCM encoder. A four signal constellation was designed

with the source statistics shown in Table 3.2. The performance of the non-symmetric constellation was compared with that of the well known equally likely constellation (for four equally likely signals, the best constellation is formed by the vertices of a square (16)). Figure 3.9 demonstrates a comparison between the systems in terms of output signal-tonoise ratio (output SNR). For large values of channel signalto-noise ratio (channel SNR) the two design procedures result

in an identical performance. As the channel SNR decreases the curves representing the performance of non-symmetric and

equally likely signal design separate and the difference between the two systems gets larger. The non-symmetric signal design is 3 dB (in terms of output SNR) superior to equally likely signal design for the noisiest channel considered (a channel SNR of 4 dB). It is seen that significant improvement

in performance can be obtained for noisy channels by utilizing non-symmetric signal design. The amount of improvement is a function of source statistics.

A comparison between the equally likely signals and unequal signal probabilities shows that if the signal statistics are not very different, the shape of constellation

is not appreciably different from the equally likely case. But









73

if the signal probabilities are very different from the equally likely case the shape of the constellation could be very different from the equally likely constellation. Figure

3. 6 (b) and Figure 3. 6 (c) show constellations f or two dif f erent five signal sources. The constellations for the extremely nonsymmetric source is completely different from the other one.

Generally speaking the geometry of. the constellations depends upon the power constraint. For example, the optimum

signal constellations for five equally likely signals subject to average power and peak power constraints are displayed in

Figure 3. 8 (a) and Figure 3. 8 (b) . Clearly, the choice of power constrain affects the geometry.

Simulations show that for large values of signal-to-noise ratio the average cost signal selection does not result in any improvement relative to the minimum error signal design. For

other values of signal-to-noise ratio the decision regions form unusual shapes which vary with the signal-to-noise ratio. Figure 3.10 shows the decision regions for a four signal constellation with the statistics listed in Table 3.2. Because of the complexity of the shape, obtaining an analytical expression for the gradient was not feasible and as a result the modified procedure in section 3.4.10 was utilized.

To compare the minimum error with the minimum cost signal selection, a' first order Gauss-Markov source with the correlation coefficient of 0.9 was generated. The output of

the source was encoded with a first order DPCM system. An









74

eight-level MSQE (mean-squared quantization error) optimized

quantizer was used and the output (reconstruction) signal-tonoise ratio of the two systems were evaluated over a wide range of channel SNR.

The results graphed in Figure 3.11 indicate that for large values of channel signal-to-noise ratio the two systems are almost identical. However, for smaller values of signalto-noise ratio the minimum cost design is superior to the MAP

system. In this case, the improvement in the output signal-tonoise ratio is around 0.5 dB for the intermediate values of

channel signal-to-noise ratio and 1 dB for low values of channel SNR.






























































































(b)




Figure 3.8. The best signal constellations for five equally likely signals subject to average power (a), and peak power

(b) constraints.


.4.














.2- x




.4


x x


.2




.4






















- 10 /'


8



41
3 4 5 a 7 a 9 10 11 12






Figure 3.9. Performance results for non-symmnetric signal
selection (the solid curve) and the equally likely signal design (the dashed curve) over a wide range of channel signalto-noise ratio.






















�sloquAs ;ouTqsTp Aq paXau
9aP suoTbaa UoTsT ap Dq; puP s,* Aq pa-uasaadaa 9- sTuBs
;9LU, "EP TT=UNS laUUpqo (q) pup sp s'c=uNs IaUUptqP (P) "ubTSsp

TlU6Ts -4soo uinUITUTUI 10_ SUOT6ba UOTST09P alq 0 " a DnbTd



(q)



TO 0 TOgL





oeIeeooeo@oeobIIe H. *666,46 666 .oo.iogI. XXI





IXXXXXXIXXXIXIXIXX: XIXEXIXXXXH*HXHXX IXXXXXEXIXMXIMXXXXXXXXXEXXXXXXIXXIXX
IXI IXIx!XIXEX~XIXXIXEXVXEXXXXIXI X EXXXXIXVXXXIXXXXIXXXXXXXXXXXKXIXXX 4XX X~XXXXIX! KlU! MIN XE XXXIXEXMMXIX XIXXIXXXXIXIXIXXXIKXEXXXXXXXXMXX ExXXXXXXIXI~XIXXEEXIXI XXXlXI






I.*. .:: t++ 5.1.OOOOVit4l . 4+
94+.X I KIMI 1 11 I X .I .XV .I x . N4 ''

I" MXXX II XEI X IXXI XIX II X X ix X ENIX xx XXv

I "NxX IX IX I XX I XIX I XXXI"IVI WXE "xxxxEXvXX













IIXIII . IIIIxIX.II.I





0400000000000oooox:-+D404+ M J-0�0,:�:::::::=:::::t 4iaiaai


. ++t. + Iit s


X4IIIXIIXXXXiXIXXE XXIXXIIIIXXiI~ XXIXIXlxXXXIXIXIXXvXIXXXxIXvXIXXIXIXX

YIXXIIIXXXXXXIXXI XXXXIIKKXXXXX xxxx xxxxx (~cxxix xxxixxXxxxvxxxlxix
XXXXX XXXxlxxXXIIIIXXI XXXIX!X! XIxXX XXXXIEIIIIXXXIXIxxxxxxxXXIXIIXXXX xx.Xvx~x. x x Ymmi xxxxxw wxwi~xi Xx.Xl XIXII XXIXXIIXIIXIXXXIIXIXxvXIIYXX
-xxixlxixxxxlxlxx~xlxxxxlx X I~XXXI x 1 R S
XxEIXXXIXXVXIXXIIXIXIXIXXIXI XXXXIXXX .xxxxxvvxxvlvxAvlxxxx





























5 8 7 8 9 10 11 12 13 14
choan.l SNR


Figure 3.11. Comparison between minimum cost signal selection (the solid curve) and minimum error signal design(the dashed curve) for a first order Gauss-Markov source with correlation coefficient of 0.9.
















CHAPTER FOUR
ALPHABET SIZE SELECTION FOR VIDEO SIGNAL CODING



4.1 Introduction



Previously the methods of signal selection for

nonsymmetric source were studied. Here we are going to utilize those methods in the deign of an image transmission system.

An image transmission system typically consists of three parts, a source coder, a channel coder and a modulator.

Source coding is the first step in the transfer of information from a source. The purpose of source coding is to

remove as much redundancy as possible from the source. DPCM of video signals with as few as seven quantization levels has been shown to produce pictures virtually indistinguishable from the original under standard viewing conditions [21]. Since the output of DPCM system is a nonsymmetric source, an entropy coder must be used to take advantage of that

redundancy. The well known Shannon-Fano method and the Huffman procedure are examples of entropy coding techniques [22].

In channel coding the goal is to correct errors

introduced in the channel by inserting redundancy in the data. Channel coding is accomplished by either decreasing the









80

information rate or increasing the bandwidth. Block and convolutional codes are the two major classes of error correction codes.

Modulation is the process of mapping a base band signal

into a band pass signal. To transmit the channel encoded signals one must design a signal constellation that minimizes error rate under a power constraint. This issue has been addressed in Chapter Three.

Even though efficient techniques for source coding, channel coding and signal design exist, it is not known how

the choice of alphabet size affects the performance of a communication system. In other words, given a nonsymmetric memoryless source with a known probability distribution, it is not clear what alphabet size results in smallest error rate subject to equal average power, bandwidth and the information rate. It is the purpose of this chapter to explore the

relationship between signal constellation size and a system performance measure (the error rate) for video signals under different bandwidth constraints.



4.2 Preliminaries

4.2.1 Description of the Communication Systems



The block diagrams of the system under consideration are shown in Figures 4.1-4.2. Huffman optimal source coding is used to encode the source signals.









81

The BCH codes are chosen for channel coding (If extra bandwidth is available) because codes close to any rate can be found. A q-ary BCH code is denoted by (n,kq,tq) where kq represents the number of information symbols, n represents the bolck size and tq represents the the error correction capability of the code.

The receivers are coherent in-phase/quadrature detectors (except for BPSK where only the in-phase branch is needed) and perfect carrier and symbol synchronization is assumed.

The code words for the nonbinary systems are not equally likely. A minimum error procedure is the optimum way of decoding at the receiver but a maximum likelihood procedure can be used as a sub-optimal procedure. The Berlekamp-Massey procedure will be used for decoding of the received data [23][25].



4.2.2 Communication Channel



The communication channel is modeled as a q-ary independent error channel and the source alphabet and channel output alphabet are the same. If a symbol si is transmitted there is a probability of pi that s, is received (i1j) and a probability of Pki that sk is received (ixk), and so on and so forth. Therefore, the probability that si is received correctly is












(4.1)


q
P i=I-- Pn
n=1
noi


Notice that pji depends on the modulation scheme. Here a two dimensional signal constellation is utilized. The channel matrix is


(4.2)


P= [pjj] ,


S2 � s3 �


PNN


Figure 4.1. The Model of the communication channel.


where Pji is given by


Pji N� 1 exp[- iSiI2] 'n jjsi-sill 4No It is also obvious that


Pji= Pij


1 i~q
l~j~q.


(4.3)


(4.4)













4.3 Analysis



For a fair comparison of systems employing different alphabet sizes, average power, bandwidth and information rate are held equal for all systems.

In practice the bandwidth constraint could be either very

strict or can be somewhat relaxed. Therefore two cases are considered:

I. When the bandwidth constraint is strict and can not be relaxed. The binary system against which all other systems will be compared uses source coding, BPSK modulation and no channel coding due to strict bandwidth limitations. For the

same information rate higher alphabet systems send longer pulses and require less bandwidth. Therefore these systems can use the extra bandwidth for error-correction codes (Figure

4.1).

Here nonbinary BCH codes are used for error correction.

For a nonbinary system of alphabet size q a q-ary (n,kq,) code must be found to satisfy the equal information rate: n hb (4.5)
kq hq


where hb and hq are the average message length for binary and q-ary systems respectively.




























(b)



Figure 4.2. The transmitters for systems employing (a) binary
(b) nonbinary alphabets.


In variable length coding (such as Huffman coding) a symbol decoding error propagates in the block. Therefore, it is appropriate to use word error probability as the figure of

merit. For the binary system P,, is given by Pw= 1- (l P.) 1(4.6)

where p, is the crossover probability. For a BPSK system p, is known to be equal to


po=Q (v -) (4.7)

where e is the signal-to-noise ratio.

II. When the bandwidth constraint is not very strict and

can be somewhat relaxed. Let us allow some bandwidth expansion for the binary system (Figure 4.2). As a result the nonbinary









85

systems employing higher order alphabets will enjoy even greater bandwidth expansion.




Ei Sure Huf Emn - oig BP
encoding




Figure 4.3. Transmitter for the binary system when the bandwidth can be expanded.




Let the bandwidth expansion for the binary system be percent. To satisfy the equal information rate constraint, binary BCH (n,k,t) code and q-ary BCH (nq,kq,t,) code must be found such that:



=(4.8) kq h q






The binary system makes no error unless t+l or more of the n total bits in a word are in error. The probability of t+l or more symbols error is given by

n




where p, is the crossover probability. Usually P, can be











approximated by


(t1 n-t1 Pe P "<1. (4.1 0)




Let the word error probability for the system using the q-ary alphabet be Pwq and let n =M (4.11)
nlq

where n and n q are the block lengths for the binary and the qary systems respectively and m is an integer. Then the word error rate for the q-ary system can be expressed as Pq111U Pwq M. (4.12)



4.4 Implementation Issues



As it was mentioned DPCM of video signals with as few as seven quantization levels has been shown to produce pictures virtually indistinguishable from the original. The quantizer uses seven quantization levels : a zero difference and three graded sizes, each of positive and negative differences.

In order to make results meaningful, we used pictures of different contents. Low, medium and high detail pictures were

used in the simulations. Table 4.1 gives the probabilities associated with each quantization level for three different scenes namely a low detail scene MICHAEL, a medium detail









87

NARROWS scene and a high detail scene BANKSIAS [21]. It is observed that:


P(Y0) > P(Y1) =P(Y-1) > P(Y2) =P(Y-2) > P(Y3) =P(Y-3) (4.13)


where p(yi) denotes the probability of ith difference level.

The primitive irreducible polynomials over the nonbinary fields are given in Table 4.2. The minimal polynomials and the generator polynomials of the nonbinary codes are given in Tables 4.4, 4.5, 4.6 and 4.7. Notice that GF(4) the ground field for GF(42) in itself is an extended version of GF(2). The elements of GF(22) are 0,I,A and B where B=A2. The arithmetic tables for GF(4) are provided in Table 4.3.

Alphabets of size 6 are not used in the simulations because 6 is neither a prime or power of a prime number and does not lead to BCH code implementation.












TABLE 4.1. DIFFERENCE SYMBOL PROBABILITIES.


TABLE 4.2. PRIMITIVE IRREDUCIBLE POLYNOMIALS OVER NONBINARY FIELDS.



Field Polynomial


GF(2) x2 +x +1

GF(3) x3 +2x +1

GF(4) x2 +x +A

GF(5) x2 +x +2

GF(7) x2 +x +3


Picture p (y.) P (Y+I) P (Y+2) P (Y+3)
P (Y-I) P (Y-2) P (Y-3)
Low .674 .142 .018 .003
Detail

Medium .584 .172 .032 .004
detail

High .5 .166 .064 .02
Detail















TABLE 4.3. ADDITION AND MULTIPLICATION TABLES FOR GF(4).


+ 0 1 A B

0 0 1 A B

1 1 0 B A

A A B 0 1

B B A 1 0









X 0 1 A B

0 0 0 0 0

1 0 1 A B

A 0 A B 1

B 0 B 1 A















TABLE 4.4. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(27). THE POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE
COEFFICIENTS OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS . (B) GENERATOR POLYNOMIALS FOR 3ARY CODES. COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS. THE POWERS OF a ARE GIVEN IN THE MIDDLE COLUMN.


GF(3') Minimal Polynomial


1,3,9 1 2 01

2,6,18 2 1 1 1

4,10,12 2 01 1

13 1 1

14,16,22 2 2 0 1

5,15,19 1 1 2 1

17,23,25 1 0 2 1

(a)


3-ary code Powers of a g(x)

(26,19,2) 13,14,15,16 120111201

(26,14,3) 14, 15, 16, 17, 18, 19 ,1122002000021


(b)
















TABLE 4.5. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(16). THE POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE COEFFICIENTS
OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS . (B) GENERATOR POLYNOMIALS FOR 4-ARY CODES COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS. THE POWERS OF ax ARE GIVEN IN THE MIDDLE COLUMN.



GF(4') Minimal Polynomial


0 1 1

1,4 All

2,8 Bl 11

3,12 1 B1

5 Al1

6,9 1 A1

7,13 A A 1

10 B I

11,14 BB I


4-ary Code Powers of a g(x)

(15,12,1) 0,1 A B 0 1

(15,9,2) 1,2,3,4 1 A A 1 1 B 1

(15,7,3) 0,1,2,3,4,5 A 0 B 1 B B 1 1
















TABLE 4.6. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(25). THE POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE COEFFICIENTS OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS . (B) GENERATOR POLYNOMIALS FOR 5-ARY CODES. COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS. THE POWERS OF a ARE GIVEN IN THE MIDDLE COLUMN.



GF(52) Minimal Polynomial


0 4 1

1,5 2 1 1

2,10 4 3 1
3,15 3 0 1

4,20 1 4 1

6 3 1

7,11 3 2 1

8,16 1 1 1

9,21 2 0 1

12 1 1

13,17 2 4 1

14,22 4 2 1

18 2 1

19,23 3 3 1


(a)


5-ary code Powers of a g(x)

(24,15,3) 0,1,2,3,4,5 1322210121

(24,12,4) 0,1,2,3,4,5,6,7 4102441132021













TABLE 4.7. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(49). THE POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE COEFFICIENTS
OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS . (B) GENERATOR POLYNOMIALS FOR 7-ARY CODES. COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER ON THE RIGHT COLUMNS. THE POWERS OF a ARE GIVEN IN THE MIDDLE COLUMN.



GF(7') Minimal Polynomial


1,7 3 11

2,14 2 5 1

3 66 1

4 4 01

5 53 1

6 14 1

8 4 1

9,15 63 1

10 41 1

11 54 1

12 1 01

13 3 21

1116 15 1

(a)


7-ary code Powers of a g(x)

(48,31,5) 1 through 10 235134333561123361

(48,27,6) 1 through 12 3202052111645042656531

(48,24,8) 1 through 16 3655056652662113433534431




Full Text
67
set of all terminating points hopefully includes the global
minimum point. Our implementation of the technique is somewhat
different. To come up with a set of start points, First a
search procedure is implemented that finds signal sets having
small error rates initiating from random start points. About
thirty start points are selected. Then we run the gradient
search techniques from these starting points.
To run the algorithms we start with a small step size and
monitor the changes in probability of error as a function of
iteration number, if Pe changes very slowly we can proceed to
increase the step size. On the other hand, we may need to
decrease the step size if Pe starts to oscillate around a
minima or goes unstable.
The best constellations for a number of sources with
three, four, five, seven and eight signals were found. Figure
3.6 shows the best constellations for three, five and seven
signals with the probability distributions given in Table I.
We compared the speed of convergence of the conjugate
gradient method to that of the conventional gradient search
method for a given value of step size. The normalized
conjugate gradient method converges to a solution much faster
than the normalized gradient method. A set of initial start
points and the optimum constellation for seven equally likely
signals are shown are in Figure 3.7. The conjugate gradient
method is faster about an order of magnitude.
To evaluate the gain in using a non-symmetric signal


49
ln(pk) (pj-skj)2> In(Pi)-JL£ (Pj-s^.)2 i*J.
oi-X
OJ7-1
(3.37)
Assume point H is the intersection of decision regions. For
point H the inequality turns into an equality and the sums on
either side will be the distances between H and signals s¡ and
Sj, therefore
AH2-BH2=N0 ln( )
P2
Similarly we obtain
(3.38)
AH2~CH2=N0
in( ) .
P3
(3.39)
In addition to the two above MAP constraints there are four
geometric constraints (eq.(3.1) through eq.(3.4)), and an
additional constraint to minimize the average energy. The
origin of the constellation is shifted to a location that
minimizes the average power. Then the average power function
evaluated at that point is set to 1/N.
p1a+p2b+p3c=Pave
(3.40)
where


30
Fig. 2.2. Lenna.
Fig. 2.3. Aerial map.


89
TABLE 4.3. ADDITION AND MULTIPLICATION TABLES FOR GF(4)
+
0
1
A
B
0
0
1
A
B
1
1
0
B
A
A
A
B
0
1
B
B
A
1
0
X
0
1
A
B
0
0
0
0
0
1
0
1
A
B
A
0
A
B
1
B
0
B
1
A


65
and
dx.
dx
1=0
dx.
By,
1=0,
(3.87)
p-o p-o.
dx, dy,
(3.88)
Equations (3.75) and (3.78) imply that v2=u2=0.
Contrary to the equations for finding Sj¡, the equations for
finding its derivatives are linear.
3.5.4 The Iterative Algorithm
To find the signal constellation we will start from an
initial guess. An iterative rule (either gradient search or
conjugate gradient search algorithm) is selected. An improved
signal set is found using the iterative rule. Depending on the
type of constraint (peak or average power constraint) the
constellation is modified accordingly (eq.(15) or eq.(30)).
The iteration is continued until a minima is reached.
The above procedure is valid under the assumption of
large channel SNR. When channel SNR is not large, obtaining an
analytical expression for the gradient is not feasible because
of the complexity of the shape of the decision regions. An
obvious solution is an exhaustive search which is


115
(a)
(b)
Figure 4.12. comparison of systems employing non-binary alphabets
against the binary system when the bandwidth can somewhat be
expanded. The solid curve represents the binary system, (a) ternary
(26,14,3) code and binary (15,11,1) code are used and a block is
390 bits long; (b) 4-ary (15,7,3) and binary (15,11,1) are used and
the block length is 225 bits long. A high detail picture is used.


92
TABLE 4.6. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(25) THE
POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE COEFFICIENTS
OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING ORDER ON THE
RIGHT COLUMNS .(B) GENERATOR POLYNOMIALS FOR 5-ARY CODES.
COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER ON THE RIGHT
COLUMNS. THE POWERS OF a ARE GIVEN IN THE MIDDLE COLUMN.
GF(52)
Minimal Polynomial
0
4 1
1/5
2 11
2,10
4 3 1
3,15
3 0 1
to
o
14 1
6
3 1
7,11
3 2 1
8,16
111
9,21
2 0 1
12
1 1
13,17
2 4 1
14,22
4 2 1
18
2 1
19,23
3 3 1
(a)
5-ary code
Powers of a
g(x)
(24,15,3)
0,1,2,3,4,5
1322210121
(24,12,4)
0,1,2,3,4,5,6,7
4102441132021
(b)


136
O 0.02 0.0-4- O.OS O.OS 0.1
(a)
Figure 5.7. The picture SNR vs. channel error rate, the dotted
curves represent the corrupted picture, (a) The solid curve
represents the MMAP receiver, (b) the solid curve represents
the MSNR receiver and the dashed curve represents the Sayood
and Borkenhagen method [12].


47
1 are related by:
l=AHs[3.
(3.30)
The point G(x,y), can be found as follows:
pave=Pi I U--|)2+y2] +p21 (^+-|)2+y2] +p3 [x2+(y--^)2]
(3.31)
3Pave
Bx
=0
(Pi-Pa) 1
(3.32)
dP*ve
By
=0
v-EihH
y 2
(3.33)
Substituting x and y into (30) and solving for 1 we get
1=JFP¡
(3.34)
where k is given by
(4/3)
Pi i (Pi"P2-l) 2+3p32] +P2 [ (p1-p2+l)2+3p32] +p3 [ (P!-p2)2+3 (p3-l)2]
(3.35)


141
(c)
Figure 5.11continued


80
information rate or increasing the bandwidth. Block and
convolutional codes are the two major classes of error
correction codes.
Modulation is the process of mapping a base band signal
into a band pass signal. To transmit the channel encoded
signals one must design a signal constellation that minimizes
error rate under a power constraint. This issue has been
addressed in Chapter Three.
Even though efficient techniques for source coding,
channel coding and signal design exist, it is not known how
the choice of alphabet size affects the performance of a
communication system. In other words, given a nonsymmetric
memory less source with a known probability distribution, it is
not clear what alphabet size results in smallest error rate
subject to equal average power, bandwidth and the information
rate. It is the purpose of this chapter to explore the
relationship between signal constellation size and a system
performance measure (the error rate) for video signals under
different bandwidth constraints.
4.2 Preliminaries
4.2.1 Description of the Communication Systems
The block diagrams of the system under consideration are
shown in Figures 4.1-4.2. Huffman optimal source coding is
used to encode the source signals.


32
TABLE 2.1. A EIGHT LEVEL MSQE QUANTIZER IS USED. THE FIRST AND
THE SECOND COLUMNS SHOW THE ACTUAL SOURCE STATISTICS FOR THE
PICTURES. THE THIRD COLUMN SHOWS THE STATISTICS USING A
LAPLACIAN DISTRIBUTION FOR SOURCE.
Lenna
Aerial Map
Model Statistics
.3324
.2800
.2549
.3574
.2795
.2549
.0946
.1416
.1510
.1060
. 1390
. 1510
.0337
.0578
.0744
.0331
.0573
. 074
.0225
.0220
.0197
.0197
.0224
.0197
TABLE 2.2. STATISTICS FOR LENNA PICTURE. A SEVEN LEVEL
QUANTIZER FOUND BY PSYCHOVISUAL EXPERIMENTS IS USED.
Laplace
Eq. (13)
Actual Statistics
.5541
.5967
.6578
. 1965
.1558
.1544
.1965
. 1558
. 1418
.0259
.0270
.0193
.0259
.0270
.0200
.0005
.0019
.0026
.0005
.0019
.0038


13
1.6.2 Convolutional Codes
A convolutional encoder consists of some shift registers
and modulo-2 summers. For the general case, k bits at a time
are entered into the shift register, and the code rate is k/n.
The state of the encoder is dependent upon the contents of the
shift registers.
Convolutional codes can be described by a code tree. It
is seen that the tree contains redundant information which can
be eliminated by merging, at any level, all nodes
corresponding to the same encoder state. The redrawing of the
tree with merging paths has been called a trellis by Forney.
The problem of decoding a convolutional code can be thought of
as attempting to find a path through the trellis or the tree
by making use of some decoding rule.
The Viterbi algorithm [5] which is shown to be a maximum
likelihood decoder for convolutional codes, involves computing
a metric between the received signal and the trellis path
entering each state. In the event that two paths terminating
on a given state are redundant, the one having the largest
metric is stored (the surviving path). This selection of
survivor is performed on for all paths entering each of the
other states. The decoder continues in this way to advance
deeper into the trellis, making decisions by eliminating the
least likely paths.
The complexity of Viterbi algorithm is an exponential


146
method that makes use of (M,L) algorithm was also presented.
Then, a modified MAP method for Markov sources is described.
The MMAP is in fact a special case of tree decoding where no
search is performed.
A maximal signal-to-noise (MSNR) receiver for DPCM system
is developed that minimizes the distortion power due to
channel errors and the appropriate cost matrix for this
receiver is computed.
These methods were applied to picture transmission over
noisy channels and were compared to a recent method. The SNR
graphs and the subjective examination of the enhanced pictures
demonstrate that the procedures are quite effective and are
superior to the method given in [12].
The Markov modelling the transmitted pictures was found
to improve the SNR of the received pictures. Further more, it
was observed that the gain increases with the increase in the
order of Markov model. For instance, in the case of MSNR
receiver, the third order modelling of sources was about 2 dB
superior to utilizing a first order model.
The MSNR receiver was found to be the more effective than
the MMAP receiver for a given order of Markov source by about
1 dB. However, the performance of the suboptimum tree decoder
is very close to MSNR receiver. Considering the ease of
implementation and the superiority of results, MSNR is
certainly the method of choice.


CHAPTER THREE
SIGNAL SELECTION FOR NONSYMMETRIC SOURCES
3.1 Introduction
In many applications one has a bandlimited channel and
has to achieve the least error rate for a given signal-to-
noise ratio. Design of high speed modems is one example where
the designer is faced with the problem of selecting an
efficient set of signals with in-phase and quadrature
components.
The objective in signal design is to find the optimum
signal constellation in presence of additive white Gaussian
noise under a power constraint. Two dimensional modulation
formats such as MPSK and QAM have been studied before [15].
These formats confine the signal points to a certain geometry
and are not optimum in the sense of minimum error rate.
There has been a few attempts to solve the signal design
problem under peak or average power constraint without
constraining the signal points to a special geometry such as
a circle or a certain lattice. Foschini et al [16] presented
an iterative approach for signal selection. A gradient search
procedure is given that incorporates a radial contraction
technique to meet the average signal power constraint.
Kernighan and Lin [17] came up with a heuristic procedure for
35


124
Q-ary alphabet {aj} for j=l,2,...,Q. The average cost (or loss)
in decoding to aj is given by
N
di=ELiJt^(rilvi=aJt) P (5.11)
k-1
where Ljk is the cost in deciding that aj was transmitted where
actually ak was transmitted. For each received symbol, the
Bayes receiver computes dw d2,...,dQ and decides a, was
transmitted if d, is the smallest of all dj. In other words,
the Bayes receiver chooses =a2 if d, j=l,2,...,Q, j^l
When the received sequence is a Markov sequence of order
M the present symbol depends on M previous symbols. Thus
expression (5.11) needs to be modified to reflect the
redundancy in the sequence
N
df=E hk P^iWi =ak) p(vd=ak\*=£,_!
*=i
(5.12)
5.5 A Maximal Siqnal-to-Noise Ratio (MSNR) Decoder
In this section a maximal signal-to-noise ratio (MSNR)
receiver for DPCM (differential pulse code modulation) of
pictures is developed. To do so the modified Bayes classifier
for Markov sources found earlier will be utilized and a cost


CHAPTER TWO
DPCM VIDEO SIGNAL: A NONSYMMETRIC
INFORMATION SOURCE
2.1 Introduction
The purpose of this chapter is to demonstrate that DPCM
(differential pulse code modulation) of pictures results in a
nonsymmetric information source.
To do so, some introductory material is presented first.
Since a quantizer is an important component of a DPCM system,
it will be examined in some detail. Quantizers will be
introduced, different criteria for the design will be
mentioned and the procedure for finding an optimal quantizer
(in MSQE sense) will be explained step by step. A DPCM encoder
will be analyzed and a model for the resulting source
statistics will be given. Eventually the model will be
compared with actual picture statistics and results will be
compared.
2.2 Basics of Quantizers
Quantization is the process of rounding sample values to
a finite number of discrete values. This process is not an
information preserving process and the reconstructed signal is
21


149
average power were compared. Two realistic situations are
considered, when one is operating under tight bandwidth
constraint and when the bandwidth constraint is somewhat
loose. The systems employing various alphabet sizes were
analyzed and computer simulations were performed using
pictures with different amount of details.
In the bandlimited case the ternary alphabet was seen to
be the superior system. In the other case, ternary system was
somewhat superior to the binary system for relatively small
bandwidth expansion. For larger bandwidth expansions, the
binary system outperformed all the others.
6.1.3 Decoding of Correlated Sequences
It is known that the source coding process does not
remove all the redundancy from the signal. An optimum tree
decoding method for decoding of the correlated sequence. A
practical solution to the problem using the (M,L) algorithm is
discussed. A modified MAP (MMAP) receiver for Markov sequences
and a maximal signal-to-noise ratio (MSNR) receiver for DPCM
signals were also derived.
The methods were applied to the picture transmission
problem. Considerable amount of improvement (from both
qualitative and quantitative point of view) was obtained. The
maximum SNR receiver was found to be superior to the other
methods.


95
chosen to prescribe the consecutive roots, and therefore for
the primitive codes where we choose a primitive element of
GF(qm) the block length is n=qm-l.
4.5.1 Encoding Nonbinarv BCH Codes
A BCH code or any other cyclic code can be encoded by
using the generator polynomial g(x) in the manner indicated by
basic definition of a cyclic code, namely,
v(x)=u(x) g(x) (4.15)
That is we associate a polynomial u(x) over GF(q) of degree
k-1 with the set of k information symbols to be transmitted
and multiply by degree r polynomial g(x) forming the degree
n-1 code polynomial v(x). However, this results in a
nonsystematic code structure, and therefore it is preferred
instead to form codewords using
v(x) = [xru(x) mod g(x) ] + xru (x) (4.16)
It is seen that this encoding operation places the k
information symbols in the k highest order terms of code
polynomial, while the parity check symbols are confined to the
r lowest order terms.
Encoding can be implemented with a division circuit using
linear feedback shift register. An encoding circuit using a


29
2.6 Results
In this section we will compute the statistics of the
quantizers for the model developed earlier and will compare
them to actual source statistics for real world pictures. Two
types of quantizer were used, a MSQE quantizer and a quantizer
that was found by pschovisual experiments [14].
The material used were two eight-bit pictures; a low
detail picture LENNA and a high detail picture AERIAL MAP
(Fig.2.2 and Fig. 2.3). Both pictures consist of 512x512
pixels.
An optimized eight-level quantizer for the Laplacian
distribution was chosen. Model parameters can be estimated by
o2= (2.14)
o0=< (§(n)-§(n))2> .
Table 2.1 contains theoretical and actual source statistics.
Theoretical values are seen to be reasonably close to actual
source statistics.
Then a seven-level quantizer [14] that is shown to work
well with different pictures was selected. The histograms for
the two pictures were prepared. The Laplacian distribution and
the distribution given in eq (2.13) were compared with the
histograms (Figures 2.4 2.5). They both seem to be a fair
approximation to the histograms.
Table 2.2 gives the statistics given by equation eq.
(2.13), Laplacian pdf and the picture. The model based
statistics are close to the actual source statistics.


139
Figure 5.9continued
(c)
Eventually the improvement process was judged based on
the subjective quality of the improved pictures. The encoded
version of LENNA picture is given in Figures 5.10. The
corrupted pictures, the enhanced version of them using Sayood
and Borkenhagen method [11], the MMAP and MSNR receivers for
channel error rates of p=.01 and p=.l are shown (Figures 5.10-
5.11). An examination of the pictures shows that the three
enhancement methods are effective. Further more, it is
observed that MMAP and MSNR are certainly more effective in
improving than the method given in [12]. Among our two
receivers MSNR seems to work better. The MSNR does a good job
in removing the horizontal streaks which are very unpleasant
to a viewer. The subjective examination of pictures seems to
agree well with conclusions derived from the SNR graphs.


85
systems employing higher order alphabets will enjoy even
greater bandwidth expansion.
Figure 4.3. Transmitter for the binary system when the
bandwidth can be expanded.
Let the bandwidth expansion for the binary system be /3
percent. To satisfy the equal information rate constraint,
binary BCH (n,k,t) code and q-ary BCH (nq,kq,tq) code must be
found such that:
-|=P+1
(4.8)
The binary system makes no error unless t+1 or more of
the n total bits in a word are in error. The probability of
t+1 or more symbols error is given by
E (|pe(l-Pe)(n-)
i-t+i \
(4.9)
where pe is the crossover probability. Usually Pw can be


98
5.correct the indicated errors.
Figure 4.5. The linear feedback shift register for generating
a sequence of syndrome values in Berlekamp-Massey algorithm.
The Peterson's direct method for solving coefficients of
error location polynomial requires a large number of
multiplications and divisions and it is not appropriate for
computer use either. In Berlekamp-Massey algorithm one will
find the coefficients of error location polynomial by finding
the tap gains in a linear feedback shift register [24].
For a given sequence of syndrome values, there are a
determinable number of polynomials of various lengths that
will generate the syndromes. This corresponds to the fact that
there are in general a number of error patterns that can
account for a given set of syndrome values. However, the
lowest-weight error pattern corresponds to the given syndrome.
Once the coefficients of the error location polynomial
has been determined, the roots of the polynomial must be
found. Chien has developed a recursive algorithm for finding


152
j v/2o(a+v/2ou) e"2 du= J 2o2 u e
a a
v/2o v/Jo
du +
J v/2 a a e
a
v/2o
du.
(A.5)
With a simple variable change the first integral can be
computed
J u e~u* du= J --iev dv= -e 2oi
_1
2
a
a*
2o2
(A.6)
and by definition the second integral is:
/
/O
du=&erf(--2-, ~)
2 /2
Thus
(A.7)
/
(x-a)2
2o2
x e
_ a-
dx= o2 e 22 + V2* a o erf(-, )
2 /2o
(A .8)


110
(a)
(b)
Figure 4.9. Systems with non-binary alphabets are compared against
the binary system. Word error rate is plotted versus average power
when the bandwidth is fixed. A word is considered to be one block
long. The solid curve represents binary system and the dashed curve
represents (a) a ternary system using (26,19,2) code; (b) a 4-ary
system using (15,9,2) code; (c) a 5-ary system using (24,12,4) code
for medium detail picture and (d) a 7-ary system using (48,24,8)
code. The results are given for high detail picture.


ACKNOWLEDGEMENTS
I wish to express my gratitude to my advisor, Dr. Scott
L. Miller, for his encouragement, support and friendship. I am
also grateful to Dr. Couch for his guidance and insight. In
addition, I would like to thank Dr. Childers and Dr. Najafi
for their time and interest in serving in my committee.
Special thanks also go to Dr. Sigmon who has offered his
valuable assistance throughout the course of this study.
iii


129
(a)
(b)
Figure 5.2. The average autocorrelation function for the
picture (top curve) and the encoded picture (the bottom
curve). (a) horizontal direction, (b) vertical direction.


62
of sv
y-i
N
I sk~si
-S>2*iPi, !s ,exp[-
2-1
*ki I'
llsj-s*J2
] [
+A]
¡Si-S^ll2 P<
where
(3.73)
and
i=(*k-Xjt) (1^) +(y*"yJ*> ("'^) '
(3.74)
a^i-^ju) + dxt.
(3.75)
=-n tP*-irb-iexPt-!££5r^l It -- +--1
S* [s^-s^ll2 i70
J-l
N
~E V2 LJci Pi ||Si_Sjci|| eXp[~'
N0
ISi-Sjtil'.j [ 1
l-l
N
iSiSitili2 N
(3.76)
l= (-^t)+(yk-yjk) (1-^) (3.77)
where


100
TABLE 4.8. THE BCH CODES, THE CODE RATES AND THE QUANTITY hb/hq
FOR (A) MEDIUM DETAIL IMAGE; (B) HIGH DETAIL IMAGE AND (C) LOW
DETAIL IMAGE.
Code
Code rate
hb/ha
(26,19,2)
1.36
1.38
(15,9,2)
1.66
1.66
(24,15,3)
1.6
1.71
(48,27,6)
1.77
1.78
(a)
Code
Code rate
hb/ha
(26,19,2)
1.36
1.49
(15,9,2)
1.66
1.83
(24,12,4)
2
1.94
(48,24,8)
2
2.14
(b)
Code
Code rate
hb/ha
(26,19,2)
1.36
1.3
(15,9,2)
1.66
1.51
(24,15,3)
1.6
1.54
(48,31,5)
1.54
1.58
(c)


86
approximated by
Pw~ (t+1) !
n I
. C*1
(n-t-1)!
pe (4.10)
Let the word error probability for the system using the
q-ary alphabet be Pwq and let
=m (4.11)
"g
where n and nq are the block lengths for the binary and the q-
ary systems respectively and m is an integer. Then the word
error rate for the q-ary system can be expressed as
Pq=l-(1-Pvq)a. (4.12)
4.4 Implementation Issues
As it was mentioned DPCM of video signals with as few as
seven quantization levels has been shown to produce pictures
virtually indistinguishable from the original. The quantizer
uses seven quantization levels : a zero difference and three
graded sizes, each of positive and negative differences.
In order to make results meaningful, we used pictures of
different contents. Low, medium and high detail pictures were
used in the simulations. Table 4.1 gives the probabilities
associated with each quantization level for three different
scenes namely a low detail scene MICHAEL, a medium detail


101
TABLE 4.9. PICTURE PROBABILITIES BEFORE AND AFTER CHANNEL
ENCODING. (A) 3-ARY CODE (26,19,2) FOR THE LOW DETAIL IMAGE;
(B) 4-ARY CODE (15,12,1) FOR THE LOW DETAIL IMAGE; (C) 5-ARY
CODE (24,15,3) FOR THE HIGH DETAIL IMAGE AND (D) 7-ARY CODE
(48,24,8) FOR THE HIGH DETAIL IMAGE.
before encoding
after encoding
.6903
.5943
. 1747
.2173
.1350
.1884
(a)
before encoding
after encoding
.6641
.5814
.1535
.1725
. 1392
. 1617
.0432
.0844
(b)
before encoding
after encoding
.5108
.3942
.1684
. 1802
. 1684
.1800
.0944
.1343
.0580
. 1113
(c)


97
the observed syndrome values. Therefore, decoding is viewed as
a problem of solving a set of simultaneous syndrome equations,
one where the set of unknowns include error values as well as
error locations.
Let v(x), e(x) and r(x) represent the transmitted word,
error pattern and the received word respectively where the
coefficients are elements in GF(q), we have
r (x) = v(x)+e(x) (4.17)
The error polynomial e(x) has nonzero terms only in those
positions where errors have occurred, so that if there are t
errors in the received word, we can write the syndrome values
as
c
sk=J2 YiXi' ^o'V1 mo+d-2 (4.18)
-1
where X¡ is the error location for the ith error and Y¡ is its
value. Syndrome decoding of a nonbinary code proceeds as
follows:
1. calculate syndrome values.
2.Determine the error location polynomial from syndrome
values.
3. solve for the roots of this equation (error locations).
4.calculate error values.


119
5.2 Optimum Decoding of Markov Sources
Here it will be shown that the optimal decoding of Markov
sequences in the sense of minimizing the error rate requires
tree searching. A practical solution to the search problem is
presented and a path metric for each branch is derived. The
path with maximum cumulative metric using the (M,L) heuristic
search routine is show to be a suboptimum solution to the
problem.
Let v=(v,, v2 . ., vN) and r= (r,,r2, . ., rN) denote the
transmitted and the received sequences respectively where
symbols v¡ and r/s are from the same Q-ary alphabet.
The error probability of a decoder is given by
P(E) =£P(£|£)P(x)
(5.1)
x
where P(r) is independent of the decoding rule. To minimize
P(E) one must minimize
P(E\z) =P(V*v\z)
(5.2)
Minimizing is equivalent to maximizing
By Bayes rule
(5.3)
Since the channel is memoryless


114
(a)
b UB)
(b)
Figure 4.11. A comparison of systems employing non-binary alphabets
against the binary system when the bandwidth can somewhat be
expanded. The solid curve represents the binary system, (a) ternary
(26,14,3) code and binary (15,11,1) code are used and a block is
390 bits long; (b) 4-ary (15,7,3) and binary (31,26,1) are used for
a low detail picture and block length is 465 bits long. The results
are given for a high detail picture.


74
eight-level MSQE (mean-squared quantization error) optimized
quantizer was used and the output (reconstruction) signal-to-
noise ratio of the two systems were evaluated over a wide
range of channel SNR.
The results graphed in Figure 3.11 indicate that for
large values of channel signal-to-noise ratio the two systems
are almost identical. However, for smaller values of signal-
to-noise ratio the minimum cost design is superior to the MAP
system. In this case, the improvement in the output signal-to-
noise ratio is around 0.5 dB for the intermediate values of
channel signal-to-noise ratio and 1 dB for low values of
channel SNR.


106
(a)
(b)
Figure 4.6. Systems with nonbinary alphabets are compared
against the binary system. Word error rate is plotted versus
average power when the bandwidth is fixed. A word is
considered to be one block long. The solid curve represents
binary system and the dashed curve represents (a) a ternary
system using (26,19,2) code for medium detail picture; (b) a
4-ary system using (15,9,2) code for low detail image; (c) a
5-ary system using (24,15,3) code for medium detail picture
and (d) a 7-ary system using (48,27,6) code for medium detail
picture.


143
(b)
Figure 5.12--continued
(c)


14
function of the code's constraint length. For large values of
constraint length (K>>10) one might consider other decoding
algorithms.
The complexity of sequential decoders is relatively
independent of constraint length, so codes with much larger
constraint length can be used. Also this technique is more
suitable than Viterbi algorithm for low bit error rates.
Sequential decoding [5] was first introduced by
Wozencraft but the most widely used algorithm to date is due
to Fano. It is an efficient method for finding the most
probable code word, given the received sequence, without
searching the entire tree. The explored path is probably only
local; that is, the procedure is sub-optimum. The search is
performed in a sequential manner, always operating on a single
path, but the decoder can back up and change previous
decisions. Each time the decoder moves forward, a tentative
decision is made. If an incorrect decision is made, subsequent
extensions of the path will be wrong. The decoder will
eventually be able to recognize the situation. When this
happens, a substantial amount of computation is needed to
recover the correct path. Backtracking and trying alternate
paths continue until it finally decodes successfully.
Convolutional codes using either Viterbi or sequential
decoding have the ability to utilize whatever soft-decision
information might be available to the decoder. It is not
surprising that they have been used widely even though their


15
theory is not as mathematically as profound as that of the
block codes. Most good convolutional codes have been found by
computer search rather than algebraic construction.
1.7 Transmission Errors in a DPCM System
Differential PCM systems are affected differently by bit
errors than PCM systems because the DPCM decoder loop causes
an error propagation, while a PCM error does not propagate in
time. Subjectively, DPCM is more error-robust than PCM in
speech coding, but less robust than PCM for image coding.
Assume a channel error changes channel input u(n) to a
wrong value v(n). Due to linearity of the decoder filter, the
correct computation of output is superposed by an error output
caused by input c(n)=u(n)-v(n) to the decoder loop.
Since the decoder is an all-pole filter, there will be an
infinite sequence of error samples at the output, with
decaying amplitudes. In the case of first-order prediction,
the effect, on a future value, at time m, is described by [3]
C(m)=C(n) a (m~n); mzn (1.28)
Transmission errors therefore propagate in the reconstructed
DPCM waveform.
This kind of error smearing is perceptually desirable in
speech coding where a PCM error spike of large magnitude is
more annoying than a low amplitude error smeared over a long
duration.


93
TABLE 4.7. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(49) THE
POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE COEFFICIENTS
OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING ORDER ON THE
RIGHT COLUMNS .(B) GENERATOR POLYNOMIALS FOR 7-ARY CODES.
COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER ON THE RIGHT
COLUMNS. THE POWERS OF a ARE GIVEN IN THE MIDDLE COLUMN.
GF(72)
Minimal Polynomial
1,7
3 11
2,14
2 5 1
3
6 6 1
4
4 0 1
5
5 3 1
6
14 1
8
4 1
9,15
6 3 1
10
4 11
11
5 4 1
12
10 1
13
3 2 1
16
5 1
(a)
7-ary code
Powers of a
g(x)
(48,31,5)
1 through 10
235134333561123361
(48,27,6)
1 through 12
3202052111645042656531
(48,24,8)
1 through 16
3655056652662113433534431
(b)


57
a circle with radius
1
N
PnX
3.4.5 Gradient Search Algorithm for the Peak Power Constraint
Even though the average power constraint is used more
often, in certain applications such as space communication the
peak power constraint is much more realistic. Given a peak
power constraint the transmitted signal points must be placed
inside the circle such that the error rate is minimum. We
would like to modify the iterative procedure developed earlier
to accommodate the peak power constraint. All we need to do is
to further modify the modified constellation at each iteration
to insure that peak signal power is bounded.
Let the peak power be P=l/N. The signal set is modified
using the iterative rule (gradient search or conjugate
gradient search), let
M= max {Js (k+1) l2} (3.62)
1 siiN
We will modify the signal set again in the following manner


37
modifications to the method given in [16]. Then a gradient
search method for a peak power constraint is developed.
Eventually a gradient search method that finds a signal
constellation for an average cost function subject to a peak
or average power constraint is presented. In the end, a few
examples are given and conclusions are drawn.
3.2 Maximum Likelihood Signal Design for Three Signals
Here a three signal constellation is designed for a 3
symbol source for transmission on a white Gaussian noise
channel. The signal constellation is depicted in Figure 3.1.
From geometrical considerations
1+a2+a3+P1+P2 + P3=27i,
(3.1)
AH cosP 1=BH cosax,
(3.2)
AH cosa2=CH cosP2,
(3.3)
BH cosP3=CH cosa3.
(3.4)
Let r(t) = (r,,r2) be the received waveform and p(t) be a
particular value of r(t). Suppose p(t) = (p,,p2) is received in
the symbol interval, pr(rj,r2|m¡) is the conditional joint
probability density of random variables defining r(t) and s,,
s2 s3 are the signal vectors denoted by A, B and C on the


54
[vF^]" vFk
[VF^.J c vF*.,
(3.52)
in which ak is the step size and hk is the quantity in this
algorithm that combines the information from current and
previous steps to define a new direction. Similar to gradient
search algorithm, the power may alter with k. Therefore the
signal vector needs to be normalized at each step of the
algorithm.
Let us summarize the procedure for iterative methods:
1. Set k=0 and select the starting points.
2. Determine the search direction by calculating hk for the
conjugate gradient method and VF for the gradient method.
3. Find the improved signals coordinates and normalize them.
If the improvement is smaller than a tolerance level stop,
otherwise set k=k+l and go to step 2.
3.4.3 Analytical Expressions for the Gradient Vector
Let us find the gradient vector for the ML receiver
first. VF the gradient of Pe is a vector of 2N components in
signal space, in which each signal occupies two dimensions.
The gradient as a vector of N two dimensional vector
components is represented by


69
l.S
i
0.5
8
-8.5
-1
-1.5
1.5
1
8.5
8
-0.5
-1
-1.5
-1.5 -1 -8.5 0 0.5 1 1.5
X X
.5 -1 -8.5 0 8.5 1 1.
(a)
(b)
Figure 3.6. Best signal constellations for non-symmetric
sources. Probability distributions are given in Table 3.1
(a) Source l.(b) Source 2. (c) Source 3. (d) source 4.


16
In picture coding, on the other hand, error propagation
is perceptually very undesirable, taking the form of very
visible streaks or blotches with one and two dimensional
predictors.
1.8 Optimum Prediction for Noisy Channels
One of the early approaches to system optimization under
noisy conditions was presented by Chang and Donaldson [6].
Because of the importance of the result and the relevance to
this dissertation a summary of the methods is given here.
Let r¡ denote the received signal, f¡ denote the impulse
response of the DPCM decoder. The output of the decoder is
therefore given by
£i = ri*.fi=(si+ni) *fi
=s1*fi +n1 =x +qi+ni*fi
(1.29)
Let us define x'¡ as follows
var
x/i=ici -xx =g_£ +ni fi
(Xi)=E[Oi2}+2^fkEiQiNi.k}
k
1 k
(1.30)


46
We experimented with a variety of signal-to-noise ratios and
probability sets. Figure 3.3 shows the optimum a, for three
probability sets over a wide signal-to-noise ratio range. In
all cases optimum a, approaches 60 degrees as signal to noise
ratio goes up. Even when the optimum angle is somewhat
different from 60 degrees from the performance point of view
the two system are almost indistinguishable.
Figure 3.3. a, (deg.) versus average power (dB). (a) the top
curves, source statistics {.9,.05,.05}, (b) The middle curve,
source statistics {.8,.l,.l} and (c) the bottom curve, source
statistics {.6,.2,.2}.
The presented results suggest that the three signals
should be placed on the vertices of an equilateral triangle
(Figure 3.4). The origin of the signal constellation can be
shifted to a new location that minimizes the average energy
without affecting the probability of error. Let us define 1 to
be
1= AB = BC = AC
in triangle ABC. Since ABC is an equilateral triangle, AH and


75
0.4
0.2
-0.2 -
-0.4
-0.5
0.5
(a)
0.4
0.2 -
0 -
-0.2 -
-0.4 -
-0.5
0
0.5
(b)
Figure 3.8. The best signal constellations for five equally
likely signals subject to average power (a) and peak power
(b) constraints.


61
0 dx
6
By
&
=0,
=0,
=0.
(3.70)
Eliminating X we will come up with the following set of
equations
£/r r w x
v £ (Lik~Ljk) (x~xk) exp (-
Jc-l ivo
(3.71)
N
I c I
where s¡=(x¡,y¡) and s=(x,y)=sji.
3.5.3 The Gradient of the Cost Function
To evaluate gk first the portion of cost function that
depends on sk must be formed:
4^ r 1 r 1
> Ljy Py-T. ?exp [-
\\sk-sjki N
J-l
N
X-^ X r IISv-SjhII2,
+EL*i Pi T-- ii exp [ ---]
J-X
N
gk* and gk2 the components of gk are found by taking the
derivative of above expression with respect to the components


70
X
X
X
X
-Z -1 0 1 2
(C)
X
X
X
X
X
X
X
-1.5 -1 -0.5 0 0.5 1 1.5
(d)
Figure 3.6continued


135
Figure 5.6. The picture SNR vs. channel error rate. The dashed
curve represents the MMAP receiver utilizing a first order
Markov model and the solid curve represents the Sayood and
Borkenhagen method [12].
Both MMAP and MSNR receivers improve the SNR of the
received picture a great deal. For instance the MAP receiver
improves the picture SNR by about 10 dB for a bit error rate
of p=.l (Figure 5.7(a)). The maximal SNR receiver, however,
results in somewhat higher value of SNR compared to the
modified MAP receiver for a given order of Markov source
(Figure 5.7(b)). Figure 5.8 shows that the MSNR is about 1 dB
superior to the MMAP receiver. It is also observed that the
MSNR receiver is at least 2 dB superior to method given in
[12] (Figure 5.7(b)).


150
6.2 Directions of Future Research
Here we addressed the problem of signal design for
nonsymmetric source without considering the channel effects.
One could reformulate the problem to incorporate the different
effects due to having less than a perfect channel. One could
also investigate the effects of imperfect carrier frequency
and phase estimates. On the other hand, the computational
complexity of the minimum cost signal design problem is
tremendous. It will be useful to reduce its computational cost
or come up with the simplified versions of it.
To take advantage of the redundancy in signal, we used
the conditional statistics of the signal. However, it may be
possible to utilize the redundancy in other forms. Moreover,
the use of redundancy and asymmetry in other methods of image
coding is yet to be explored.


130
Figure 5.3. The DPCM prediction error.
take advantage of the redundancy in the encoded signal by
modeling the encoded signal as a Markov source.
5.7 Picture Transmission Over Noisy Channels
5.7.1 Implementation Issues
The picture was encoded to three bit DPCM samples and was
sent over the channel. The communication channel was assumed
to be a binary symmetric channel. We experimented with noisy
channels where the error rate ranged from p=0.01 to p=0.1.
It has been known for sometime that the effect of the
transmission error can be reduced by using a somewhat smaller
prediction coefficient. Chang and Donaldson [6] found the
optimum prediction coefficient that minimizes the effect of
the channel errors. A prediction coefficient of .91 was used


CHAPTER ONE
REVIEW OF MODULATION TECHNIQUES, SOURCE AND CHANNEL CODING
IN DIGITAL COMMUNICATIONS
1.1 Introduction
The goal of this chapter is to present the background
necessary to follow the work presented in this dissertation,
introduce the research objectives and give a brief description
of the chapters.
Since two-dimensional modulation has been utilized in
this work, these formats are reviewed, their spectral
efficiency are calculated and the upper bound on their
spectral efficiency is given. Picture transmission is one of
the applications presented here. To familiarize the reader
with the field, a number of source coding techniques have been
described. Error correction over noisy channels is an
important topic and has been utilized in this thesis. The gain
from channel coding has been explained and different methods
of error correction have been discussed. The effects of
channel errors in a DPCM system and optimum prediction in
noisy channels was presented, because the enhancement of
picture quality in DPCM systems has been addressed in this
dissertation.
1


Ill
(ffl)
(C)
(d)
Figure 4.9--continued


I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
^j4n)
^Kermit Sigmon
Associate Professor of
Mathematics
This dissertation was submitted to the Graduate Faculty
of the College of Engineering and to the Graduate School and
was accepted as partial fulfillment of the requirements for
the degree of Doctor of Philosophy.
August 1993
Winfred M. Phillips
Dean, College of Engineering
Dean, Graduate School


word error me
113


51
3.4 Minimum Error Signal Selection
3.4.1 Preliminaries
Let us denote the signals in the constellation by
s,,s2, ...,sN where s¡=(x¡,y¡) The average power is given by
n (3.41)
n-1
Where {p, ,p2 ,...,pN} is the probability distribution of the
signal set and ||sn|| is the magnitude of sn. Following the
convention of (16] we choose to set P,ve =1/N and define the
signal-to-noise ratio to be
SNR=10 log10 (3.42)
Na
where N0 is the one sided power spectral density of white
noise.
By definition the error rate (probability of symbol
error) is
'A iv (3.43)
ie=E Pn Pr (error |sn)
The union bound gives the upper bound for the conditional
probability of error. For a maximum likelihood receiver we
have:


56
I
SkSl
sk-si
Isjt-SjH
and
t,=is-sf\\2+N0ln(l-£) ,
Pi
t2 = lsJt-sil2"-woln(;fi)
3.4.4 Starting Points
The average signal power is equal to 1/N or
Ep.kiH-
n-1 iv
Thus, individual signals must satisfy
Solving the inequality for sn we get
\ pnN n ^ pnN
(3.56)
(3.57)
(3.58)
(3.59)
(3.60)
(3.61)
which states that starting points must be selected from inside


28
O'Neal [13] has shown experimentally that the pdf of e(n)
can be approximated with a Laplacian distribution
p(x) =^-exp(- ^1*1) (2.11)
v/2 o
To find the statistics of the levels an optimum quantizer must
be placed in the DPCM system. If we utilize a MSQE quantizer,
the statistics associated with each level will be
Po . P
P, = f -i- exp () dx= exp (-/2y) dy
0 0^2 a (2.12)
= -i (exp(-y2a) -exp(-./2P) ,
where a and /? are tabulated in [11].
We have also verified through simulations that the choice
of a gamma distribution for e(n) and uniform pdf for Q results
in a satisfactory approximation to the density of e(n) In
other words
P(*> =
/
| 8no |x|
(2.13)
A numerical method must be employed to find the statistics
associated with each level.


91
TABLE 4.5. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(16). THE
POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE COEFFICIENTS
OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING ORDER ON THE
RIGHT COLUMNS .(B) GENERATOR POLYNOMIALS FOR 4-ARY CODES
COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER ON THE RIGHT
COLUMNS. THE POWERS OF a ARE GIVEN IN THE MIDDLE COLUMN.
GF(42)
Minimal Polynomial
0
1 1
1,4
All
2,8
B 1 1
3,12
1 B 1
5
A 1
6,9
1 A 1
7,13
A A 1
10
B 1
11,14
B B 1
(a)
4-ary Code
Powers of a
g(x)
(15,12,1)
o
H
A B 0 1
(15,9,2)
1,2,3,4
1 A A 1 1 B 1
(15,7,3)
0,1,2,3,4,5
A0B1BB11
(b)


33
Figure 2.2. The histogram for LENNA picture is compared with
(a) Laplacian Distribution (b) Distribution given in eq.
(2.13). Histogram is shown with a broken line.


10
/
a given total number of bits for Gaussian variables, the
optimum assignment is done by making the average quantization
error of each coefficient the same. This requires that the
bits be assigned to the coefficients in proportion to the
logarithm of their variance.
1.6 Channel Coding
Channel coding is a method of inserting structured
redundancy into the source data so that transmission errors
can be identified and corrected. Block coding and
convolutional coding are two important subcategories of
channel coding techniques.
1.6.1 Block Codes
With block coding the source data is first segmented into
blocks of k bits; each block can represent any of M=2k
distinct messages. The encoder transforms each message into a
larger block of n digits. This set of 2k coded messages is
called a code block. The (n-k) digits, which the encoder adds
to each message block, are called redundant digits; they carry
no new information. The ratio of data bits to total bits
within a block, k/n, is called code rate. The code itself is
referred to as an (n,k) code.
To demonstrate the performance improvement possible with


96
shift register with r=n-k stages is shown in Figure 4.4. Each
box in the circuit is a q-ary storage device. The
multiplications and additions are done in GF(q), the symbol
field of the code. The operation of the circuit is as follows:
1. Shift the k information symbols into the encoder and
simultaneously into the channel. As soon as the k
information symbols have entered the shift register,
the r=n-k symbols in the register are the check symbols.
2. Disable the feedback circuit.
3. Shift the contents of the register out and into the
channel.
Figure 4.4. The encoder for q-ary BCH codes.
4.5.2 Decoding BCH Codes
Given the set of syndrome values for the received word,
the decoding task is to find the most likely error pattern,
within the error correction limit of the code, that produces


38
constellation. A maximum likelihood receiver sets the message
estimate to mk if for i=l,2,3,
pr{r=p1, r2=p2\ mk) > pz{r1=pi, x2=p2\ i*k. (3.5)
Assuming the noise components are Gaussian and statistically
independent,
Pr(Pi-'Pal jn)=lFexp{--^ [(p1-siJ)2+(p2-si2)2]}. (3,6)
The decision rule then becomes
t(Pi-sw)2+(p2-s^)2] > -~[(p1-s)2+(p2-si)2] i*k.
JV0 iVo
(3.7)
Notice that the sum of squared terms on either side of the
c
Figure 3.1. A three signal constellation
inequality are the square of the Euclidean distances between
the received signal and signals s¡ and sk. Thus the decision
rule can be rewritten as:


133
5.7.2 Results
DPCM decoders with first, second and third order Markov
source models were implemented and compared based upon the SNR
of the reconstructed signal and subjective impression.
Considerable improvements in the SNR of the received signal
and subjective impression is obtained.
Figure 5.5(a) and 5.5(b) give a comparison of the various
orders of Markov source. It is seen that the improvements
increase with the increase in the order of Markov source. The
second order modelling of the source is clearly superior to
first order modelling. However, third order modelling of the
source is only slightly better than that obtained with the
second order modelling. The fourth and higher order modelling
of the source were not implemented because of two reasons.
Almost all of the redundancy is in the adjacent symbols and
the overhead cost becomes tremendous (the overhead associated
with the fourth dimensional statistics is about one hundred
fifty percent).
The MMAP was compared to the Viterbi decoding method
given in [12]. The MMAP with a first order Markov source is
comparable in performance to Sayood and Borkenhagen's receiver
[12] (Figure 5.6), but it has the advantage of running faster
and requiring no memory.


127
e= (^jc-yj)2 pk\i Pi-
k-0 1-0
(5.19)
The SNR can be maximized by minimizing en2 One can think of
en2 as an average cost function where
Lki=(yk-yi)2-
(5.20)
Therefore SNR is maximized by minimizing the cost function and
setting the cost according to (5.20).
5.6 Redundancy in the Encoded Signals
It has been shown that assuming an AR model for the
signals the output of the encoder contains some redundancy
[12]. Here we will examine the encoded signal and its
autocorrelation function to get a feel for the extent of
residual redundancy.
Standard 512x512 pixel eight-bit pictures such as LENNA
were encoded using a first order DPCM encoder. The average
normalized autocorrelation in horizontal and vertical
direction were found and plotted (Figure 5.2). This shows that
certain amount of correlation among the adjacent encoded
symbols exist. The DPCM prediction error signal is shown in


23
2.3 Approaches to Quantizer Design
The quantizers can be designed based on a mean squared
error criterion. This results in overspecification of the low
detailed areas of picture and consequently a small amount of
granular noise but relatively poor reproduction of edges.
It has been realized for sometime that for a better
picture quality, quantizers should be designed on the basis of
psychovisual criteria [7]. One method of designing
psychovisual quantizers is to minimize a weighted mean squared
quantization error, where the weights are derived from
subjective experiments [8]. Such optimization would be similar
to mean squared error criteria, where the density function is
replaced by a weighing function.
2.4 MSOE Quantizer Design
An optimal quantizer is defined to be a quantizer with
the smallest mean squared quantization error. It is desired to
find the quantizer that minimizes the mean-squared
quantization error for a given probability density function
and number of levels. The mean-squared quantization error is
given by


43
For a given signal to noise ratio the probability of word
error depends only on a, and a2 because a,, a2 and a3 add up to
n. To find the values of a, a2 a3 that minimize the
probability of the word error, we need to differentiate Pw
with respect to a, and a2 and set them equal to zero:
U-0
(u(l + tan2^) e
-u'tan'a
-u(l+tan2(ax+a2)) e
-uJtanJ(a1+o2)\
(3.23)


158
[36] Melsa, J. L. and Cohn, D., Decision and Estimation
Theory. Me Graw-Hill Book Company, New York 1978.
[37] Modestino, J. W., and Daut, D. G., "Combined Source-
Channel Coding of Images," IEEE Trans. Commun.. Vol. COM-
27, pp. 1644-1659, 1979.


125
matrix (or its elements Ljk) will be derived.
Let the alphabet
(Sj / 3-2 '
represent output levels
(ylfy2f ,y
in a DPCM system. Suppose the receiver makes an error and aj
is chosen where ak was actually sent. The reconstruction error
amounts to yj-yk. The error affects the following samples. The
contribution of error to following symbols will be a(yj-yk),
a2(Yj-yk) a3(yj-yk) and so on and so forth, where a is the
prediction coefficient in the DPCM system [3]. The total
distortion power due to a channel error is therefore given by
N= (yj-yk)2+a2 (yj-yk)2+a4 (yj-yk)2 +. .
(5.13)
= Wj~yk)2 [l+a2+a4+. .] =(yj-yk)2(
1 d.
If we set
Lj k=(yk-y^2 (5.i4)
the modified Bayes receiver will minimize the average cost
which is distortion power due to channel errors in this case.
Thus, the maximal SNR receiver calculates d^dj, ..., dQ
(the average loss for each symbol) and chooses the symbol with
the least average loss where average loss is


142
(d)
Figure 5.11continued
(a)
Figure 5.12. The effectiveness of the picture enhancement
process in a very noisy situation (p=.l) is demonstrated, (a)
Corrupted picture, (b) enhanced picture using method in [12],
and (c) enhanced picture using MSNR receiver.


40
region I¡ (Figure 3.2),
pw=1-'EPiffPr(Pi'P2¡ mi)dpdp2,
i-l J J
(3.15)
Figure 3.2. One decision region.
By substituting the expressions for probability density
function eq. (3.6) into eq. (3.15) we get
^//exp[~ (Pl^J>2~^idpdp2 (3-16>
1 1 PlP2
or equivalently


2
1.2 Two-Dimensional Modulation Formats
M-ary phase-shift keying (MPSK) and quadrature amplitude
modulation (QAM) are among the most popular two dimensional
formats. In MPSK the transmitted signal is given by
s(t) =Re{g( t) eJ"ct}
(1.1)
where
g(t) =A e:70(c),
(1.2)
In other words, in MPSK while the amplitude is maintained
constant the phase of signal can take on one of the M values
in a symbol interval. The MPSK for M=4 is called quadrature
phase-shift keying (QPSK).
In MPSK signal points are confined to the circumference
of a circle. But in QAM the transmitted signal is
s(t) =Re{g( t) eJct
(1.3)
where
gr( t) =x( t) +jy( t) ,
(1.4)
or


TABLE OF CONTENTS
ACKNOWLEDGEMENTS iii
ABSTRACT . vi
CHAPTER ONE REVIEW OF MODULATION TECHNIQUES, SOURCE AND
CHANNEL CODING 1
1.1 Introduction 1
1.2 Two-Dimensional Modulation Formats . . .. 2
1.3 Source Coding 5
1.4 DPCM 5
1.5 Transform Coding 8
1.6 Channel Coding 10
1.7 Transmission Errors in a DPCM system .... 15
1.8 Optimum Prediction for Noisy Channels .... 16
1.9 Research Objectives 18
1.10 Description of Chapters 20
CHAPTER TWO DPCM VIDEO SIGNAL: A NONSYMMETRIC
INFORMATION SOURCE 21
2.1 Introduction 21
2.2 Basics of Quantizers 21
2.3 Approaches to Quantizer Design 23
2.4 MSQE Quantizer Design 23
2.5 Analysis of DPCM Encoder 26
2.6 Results 29
2.7 Discussion 31
CHAPTER THREE SIGNAL DESIGN FOR NONSYMMETRIC SOURCES 35
3.1 Introduction 35
3.2 Maximum Likelihood Signal Design for Three
Signals 37
3.3 A Numerical Approach Based on Lagrange
Multipliers Method 48
3.4 Minimum Error Signal Selection 51
3.5 Minimum Average Cost Signal Selection .... 58
3.6 Results 66
iv


140
Figure 5.10. The DPCM encoded version of LENNA in the absence
of channel errors.
Figure 5.11. The effectiveness of the picture enhancement
process in a relatively noisy situation (p=.01) is
demonstrated, (a) Corrupted picture, (b) enhanced picture
using method in [12], (c) enhanced picture using MMAP receiver
and (d) enhanced picture using MSNR receiver.


53
(3.47)
where
Si (k)
s2{k)
(3.48)
sN(k)
denotes the signal vector at the kth step of the algorithm, ak
is the step size and VF is the gradient of Pc.
Since the signal power may change with k, the signal vector is
normalized at each step of the algorithm
v/i7(p1jsi (k+1) 12+p2\s2 (k+1) J2+. . +pN\s(k+l) |2)
(3.49)
To speed up the convergence, instead of a conventional
gradient search algorithm, the Fletcher and Reeves conjugate
gradient method [18] can be utilized. In this method the
information about the second derivative is used indirectly.
The algorithm is described by
(3.50)
where
hfr vFfr+tt j^£_i
(3.51)
and


48
Figure 3.4. The three signals form a equilateral triangle.
3.3 A Numerical Approach Based on Lagrange Multipliers
Method
In this approach the problem of signal selection is
viewed as a constrained optimization problem. The constraints
are incorporated into the optimization problem by the use of
Lagrange multipliers method.
Let us outline the design of a MAP receiver with three
signals with unequal probabilities (Figure 3.1).
A MAP receiver decision rule is m=mk if, for i=l,2,3,
Pi (ri=Pi' r2=P21 mk)pk> Pr(ri=Pi,r2=p2| mi)pi i*k.
(3.36)
Upon substituting (6) into (43), we get


71
e.s -
-1.5 -1 -8.5 8 8.5
1.5
(a)
(b)
Figure 3.7. (a) Seven signal constellation for an equally
likely source. 0's represent the start points and x's
represent the final constellation, (b) The solid and dashed
curves show the error rate as a function of number of
iterations for gradient search and conjugate gradient method
respectively.


To my parents
for their love, patience and support.


109
o (dB>
(a)
e (dB)
(b)
Figure 4.8. Systems with non-binary alphabets are compared for the
fixed bandwidth case. The first system is represented with a solid
curve and the second system is represented with a dashed curve (a)
7-ary and ternary systems and (b) 7-ary and 4-ary systems
respectively.


87
NARROWS scene and a high detail scene BANKSIAS [21]. It is
observed that:
p(y0) > p(y1)*p(y-1)> P(y2) *P(y.2) > p(y3)p(y.3) (4*13)
where p(y¡) denotes the probability of ith difference level.
The primitive irreducible polynomials over the nonbinary
fields are given in Table 4.2. The minimal polynomials and the
generator polynomials of the nonbinary codes are given in
Tables 4.4, 4.5, 4.6 and 4.7. Notice that GF(4) the ground
field for GF(42) in itself is an extended version of GF(2) .
The elements of GF(22) are 0,1, A and B where B=A2. The
arithmetic tables for GF(4) are provided in Table 4.3.
Alphabets of size 6 are not used in the simulations
because 6 is neither a prime or power of a prime number and
does not lead to BCH code implementation.


27
P5=Pe* Pq- (2-9)
And the probability of quantum level i is determined by
Pi= f p5(x) dx.
(2.10)
e (n)
x(n)
p
(b)
Figure 2.1. Block diagram of a DPCM system, (a) encoder and
decoder.


90
TABLE 4.4. (A) MINIMAL POLYNOMIALS FOR ELEMENTS OF GF(27).
THE POWERS OF a ARE SHOWN ON THE LEFT COLUMN AND THE
COEFFICIENTS OF MINIMAL POLYNOMIALS ARE SHOWN IN ASCENDING
ORDER ON THE RIGHT COLUMNS .(B) GENERATOR POLYNOMIALS FOR 3-
ARY CODES. COEFFICIENTS OF G(X) ARE SHOWN IN ASCENDING ORDER
ON THE RIGHT COLUMNS. THE POWERS OF a ARE GIVEN IN THE MIDDLE
COLUMN.
GF(33)
Minimal Polynomial
1/3,9
12 0 1
2,6,18
2 111
4,10,12
2 0 11
13
1 1
14,16,22
2 2 0 1
5,15,19
112 1
17,23,25
10 2 1
(a)
3-ary code
Powers of a
g(x)
(26,19,2)
13,14,15,16
20111201
(26,14,3)
14,15,16,17,18,19
1122002000021
(b)


137
Figure 5.8. The picture SNR vs. channel error rate. The solid
curve represents the MSNR receiver and the dotted curve
represents the MMAP receive.
The minimum error decoding procedure with a limited
search is expected to outperform the MMAP receiver. Table 5.2
compares the performance of a MMAP receiver with a minimum
error receiver that utilizes a (M=750, L=7) search. For
smaller error rate the improvements are around 0.5 dB but for
very noisy channels the search does not seem to be beneficial.
TABLE 5.2. THE PICTURE SNR SUBJECT TO VARIOUS CHANNEL ERROR
RATES FOR MMAP AND MIN ERROR RECEIVER. A (M=750,L=7) SEARCH
WAS PERFORMED FOR THE MINIMUM ERROR RECEIVER.
p
.01
o
to
.03
.05
t-'
o

.1
MMAP
22.23
20.22
19.44
18.41
17.52
16.69
Min. error
22.82
20.98
20.1
18.7
17.68
16.72


107
(c)
Figure 4.6--continued.
(d)


144
5.8 Side Information
The transmitter must compute and transmit higher
dimensional statistics of the picture together with the
picture. The amount of side information depends upon system
parameters. Consider a DPCM system with the following
parameters:
Number of pixels =NxN
n= length of DPCM words
q= number of quantizer levels
m= digits of accuracy in statistics
p= order of Markov model.
The increase in bit rate is equal to
R=gp log2 (10,n-l) (5.22)
The percentage of overhead is therefore given by
% overhead=
gplog2 (10m-l) 3.32 m qp
NxNxn m2 n
(5.23)
The effect of noisy statistics on the performance of the
system was studied. It turns out that the system performance
deteriorates rapidly as the channel noise increases.
Therefore, one must protect the conditional statistics against
channel noise. To reach the performance that is achievable by
utilizing noise-free statistics, the following coding strategy


78
chonntl SNR
Figure 3.11. Comparison between minimum cost signal selection
(the solid curve) and minimum error signal design(the dashed
curve) for a first order Gauss-Markov source with correlation
coefficient of 0.9.


Internet Distribution Consent Agreement
In reference to the following dissertation:
AUTHOR: Emami, Shahriar
TITLE: Efficient communications for nonsymmetric information sources with
application to picture transmission (record number: 1933237)
PUBLICATION DATE: 1993
i, Sho^hfi AY" } as copyright holder for the
aforementioned dissertation, hereby grant specific and limited archive and distribution rights to
the Board of Trustees of the University of Florida and its agents. I authorize the University of
Florida to digitize and distribute the dissertation described above for nonprofit, educational
purposes via the Internet or successive technologies.
This is a non-exclusive grant of permissions for specific off-line and on-line uses for an
indefinite term. Off-line uses shall be limited to those specifically allowed by "Fair Use" as
prescribed by the terms of United States copyright legislation (cf, Title 17, U.S. Code) as well as
to the maintenance and preservation -oftrdigital rehve-eopy: fgltKZffion allows the University
of Florida to generate image- and text-based versions as appflOpiieteJaffid to provide and enhance
access using search software.
This grant of permissions prohibits use-of the digitized versions fbr commercial use or profit.
Signature
1 rtnV &
Printed or Typed Name of Copyright'Holder/Incensee
Personal information blurred
Date of Signature
Please print, sign and return to:
Cathleen Martyniak
UF Dissertation Project
Preservation Department
University of Florida Libraries
P.O. Box 117007
Gainesville, FL 32611-7007
5/28/2008


3
s(t)=x(t) cos (o>ct) -y( t) sin(oct) .
(1.5)
The waveforms x(t) and y(t) are
x( t) =]Txi h(t-iT)
(1.6)
i
and
y(t) =52y h(t-iT),
(1.7)
i
where T is the symbol interval in seconds and h(t) is the
pulse shape.
Let us find the spectral efficiency of MPSK and QAM with
rectangular pulses. The null-to-null transmission bandwidth of
MPSK and QAM is
(1.8)
where M=2 is the number of points in the signal constellation.
The spectral efficiency is therefore given by
R 1 bi ts/sec
^ Bt 2 Hz
(1.9)
When operating over a bandlimited channel and the overall
pulse shape satisfies the raised cosine rolloff filter
characteristics, the bandwidth of the modulating signal is


145
was found empirically. Each element of statistics is coded
into a log2(10m-l) bit natural binary code. The first
(log2(10m-l))/2 bits are coded using a (5,1) repeat code and
the rest are left uncoded. Thus, the percentage of side
information is determined by
% overhead= A0. (5.24)
N2 n
The amount of overhead for a 512x512 pixel picture, a
seven level quantizer and four decimal digits of accuracy is
given in Table 5.3. Even with a third order Markov model the
amount of side information does not exceed twenty one percent
of the total picture information.
TABLE 5.3. PERCENTAGE OVERHEAD AS A FUNCTION OF ORDER OF
MARKOV SOURCE MODEL USED IN THE SYSTEM.
order of Markov source
percentage overhead
first
1 %
second
3 %
third
21 %
5.9 Summary
An optimum method for decoding correlated sequences with
the aid of Markov source modeling was derived and it was shown
to require tree searching. A suboptimum implementation of the


112
(a)
Figure 4.10. Systems with non-binary alphabets are compared against
the binary system. Word error rate is plotted versus average power
when the bandwidth is fixed. A word is considered to be one block
long. The solid curve represents binary system and the dashed curve
represents (a) a ternary system using (26,19,2) code; (b) a 4-ary
system using (15,9,2) code; (c) a 5-ary system using (24,15,3) code
and (d) a 7-ary system using (48,31,5) code. A low detail picture
is used.


I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
Scott L. Miller, Chair
Associate Professor of
Electrical Engineering
I certify that I have read this study and that in my
opinion it conforms to acceptable standards of' scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
¡A) ^AiusL
1
Leon Couch
Professor of
Electrical Engineering
I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosophy.
<&>i J
Donald G. Childers
Professor of
Electrical Engineering
I certify that I have read this study and that in my
opinion it conforms to acceptable standards of scholarly
presentation and is fully adequate, in scope and quality, as
a dissertation for the degree of Doctor of Philosoj
'Fazil Najafi v
Associate Professor of
Civil Engineering


83
4.3 Analysis
For a fair comparison of systems employing different
alphabet sizes, average power, bandwidth and information rate
are held equal for all systems.
In practice the bandwidth constraint could be either very
strict or can be somewhat relaxed. Therefore two cases are
considered:
I. When the bandwidth constraint is strict and can not be
relaxed. The binary system against which all other systems
will be compared uses source coding, BPSK modulation and no
channel coding due to strict bandwidth limitations. For the
same information rate higher alphabet systems send longer
pulses and require less bandwidth. Therefore these systems can
use the extra bandwidth for error-correction codes (Figure
4.1) .
Here nonbinary BCH codes are used for error correction.
For a nonbinary system of alphabet size q a q-ary (n,kq,tq)
code must be found to satisfy the equal information rate:
n Ab
(4.5)
where hb and hq are the average message length for binary and
q-ary systems respectively.


60
with Sj. p (sj | Sj) can be approximated with a Q function
P(S |s.)=C>(l£LJ!zi!)
(3.67)
substituting back into the average cost expression we get
C=A
2 N
n
'E'Ehi p _.1 exp[-BSi Sji¡2] (3-68)
Ji 1 ISv-S,J
3.5.2 The Nearest Points on the Boundaries
The first step in computing the average cost function is
to find the set of nearest point to each and every signal
point in the constellation on the boundaries. There are N
signal points in the constellation and for each one there are
(N-l) of nearest points on (N-l) boundaries, therefore a total
of N(N-l) of nearest points must be found.
The problem of finding Sj¡ which is the nearest point to
s¡ on the boundary between s¡ and Sj can be formulated like a
Lagrange multiplier problem. We would like to minimize ||s-s¡||
subject to a constraint (eq. (3.65)). Let us form the
auxiliary function
Jc-l k-i
and differentiate it with respect to x,y and X


156
[12] Sayood, K., and Borkenhagen, J. C., "Use of Residual
Redundancy In The Design of Joint Source/Channel
Coders," IEEE Trans.Commun.. Vol. 39, No. 6, pp.838-
846, 1991.
[13] O'Neal, J. B., "Predictive Quantizing Systems
(Differential Pulse code Modulation) for the
Transmission of Television Signals," The Bell Systems
Technical Journal. Vol. XLV, pp.689-721, 1966.
[14] Limb, J. 0., Vision Oriented Coding of Visual Signals,
Ph.D. dissertation. University of Western Australia,
1966.
[15] Arthurs, E., and Dym, H., "On Optimum Detection of
Digital Signals in Presence of White Gaussian Noise-A
geometric interpretation and a study of three basic data
transmission systems," IRE Trans, on Communications
Systems. Vol. CS-10, No.4, pp.336-372, 1962.
[16] Foschini, G.J., Gitlin, R.D., and Weinstein, S.B.,
"Optimization of Two -Dimensional Signal Constellation
in presence of Gaussian Noise," IEEE Trans. Commun..
Vol. COM-22, pp.28-38, 1974.
[17] Kernighan, B.W., and Lin, S., "Heuristic Solution of a
Signal Design Optimization Problem," The Bell Systems
Technical Journal. Vol. 52, pp. 1145-1159, 1973.
[18] Fletcher, R., and Reeves, C.M., "Function Minimization
by Conjugate Gradients," Computer Journal. Vol. 7, No. 2,
pp.149-154, 1964.
[19] Couch, II, L.W.. Digital and Analog Communications
Systems. Macmillan Publishing Co.,Inc., New York, 1990.
[20] Press, W. H., Teukolsky, S. A., and Vetterling, W. T.,
Numerical Recipes. Cambridge University Press,
Cambridge, UK, 1986.
[21] Hullett, J. L., "Impairment by Transmission errors of
Differentially Quantized Television Signals,"
Proceedings of the Symposium on Picture Bandwidth
Compression. Cambridge, MA, 1969.
[22] Thomas, J. B., An Introduction to Statistical
Communication Theory. John Wiley & Sons, Inc. New York,
1969 .
[23] Berlekamp, E. R., Algebraic Coding Theory. McGraw-Hill
Book Company,Inc., New York, 1968.


105
assumptions of equal information rate, bandwidth and average
power. Pictures of different contents were used and two cases
were considered, when the bandwidth is strictly fixed and when
the bandwidth can be somewhat expanded.
The results suggest that the nonbinary alphabets are more
appropriate for picture tranmission over noisy channels when
the bandwidth constraint is tight. This has to do with the
fact that under equal information rate the nonbinary systems
utilize error correction codes and the binary system does not.
Among the nonbinary systems, the system with ternary alphabet
performed best. On can explain this trend in the following
manner. The ternary system has the least code rate among the
nonbinary sytems. A little bit of coding is useful; but it
will not be beneficial any longer, if you overdo it.
In case two, for a modest bandwidth expansion the ternary
system is superior to the binary system above a threshold.
However, for larger bandwidth expansion factors the binary
system seems to be more efficient than the systems utilizing
nonbinary alphabets for image transmission.


19
two dimensions for nonsymmetric sources. It is desired to find
an algorithmic solution to the minimum error signal
constellation for average and peak power constraints. In
addition the general case where the cost function is not
necessarily the error rate is discussed.
Even though efficient techniques for source coding,
channel coding and signal design exists, it is not known how
the choice of alphabet size affects a communication system. We
would like to compare communication systems with various
alphabet sizes for the transmission of video signals on the
basis of equal information rate, bandwidth and average power.
Two realistic situations will be considered, when one is
operating under tight bandwidth constraint and when the
constraint is somewhat loose.
System performance can be improved using standard error
correction techniques at the cost of increasing the bandwidth
or reducing the information rate. However, we would like to
use inherent asymmetry and redundancy in the transmitted
picture to improve the reception. We will model the data as a
Markov source and derive the optimum method for decoding the
data. We will also find a receiver that instead of minimizing
the error rate maximizes the SNR (signal-to-noise) ratio.


xml record header identifier oai:www.uflib.ufl.edu.ufdc:UF0008221000001datestamp 2009-01-28setSpec [UFDC_OAI_SET]metadata oai_dc:dc xmlns:oai_dc http:www.openarchives.orgOAI2.0oai_dc xmlns:dc http:purl.orgdcelements1.1 xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.openarchives.orgOAI2.0oai_dc.xsd dc:title Efficient communications for nonsymmetric information sources with application to picture transmission dc:creator Emami, Shahriardc:publisher Shahriar Emamidc:date 1993dc:type Bookdc:identifier http://www.uflib.ufl.edu/ufdc/?b=UF00082210&v=0000130784206 (oclc)001933237 (alephbibnum)dc:source University of Florida


128
Figure 5.3. This also indicates that some correlation is
present in the encoded picture.
Although significant amount of redundancy has been
removed from the picture some structure is still present in
the encoded signal. The decoding methods given in this chapter


59
3.5.1 Average Cost Function
The Bayes receiver which minimizes the average cost is
given by
^=EEp(sil si)Pihi' (3.64)
j"i j-i
where Lj¡ is cost involved when the receiver picks Sj when s¡ was
actually sent.
The boundaries of the decision regions are no longer
straight lines; instead they are two dimensional curves. For
instance, the boundary between s¡ and s is given by
£L* ^(^|sjc)PrE% (rlsjcJp*, (3.65)
k-l k-1
where conditional probability of error is
P(rjsk) = exp(--S ) (3.66)
nN0 N0
Channel noise could cause the received signal to move to an
adjacent decision region and result in an error. For large
values of signal-to-noise ratio; however, errors are almost
entirely the result of displacement of a received signal to an
adjacent decision region from the nearest point on a boundary
to the transmitted signal. Let s¡ be the transmitted signal
and Sj¡ be the point with shortest distance on the boundary


9
international standard for transform coding systems. In
addition to that it minimizes the blocking artifact that,
results when the boundaries between the subimages become
visible.
1.5.2 Size of Subpictures
Computer simulations on real pictures show that the mean
square error produced by transform coding improves with the
size of subpicture. However, the improvement is not
significant as subpicture is increased beyond blocksize of
16x16. Subjective quality of pictures, however, does not
appear to improve with the size of block beyond 4x4 pixels
[2].
1.5.3 Bit Allocation
One method of choosing the coefficients for transmission,
is to evaluate the coefficient variances on a set of average
picture, and then discard all the coefficients whose variance
is lower than a certain value. Such a scheme is called zonal
filtering [3].
Having decided which coefficients to transmit, we must
then design a quantizer for each of them. This could be done
by dividing a given total number of bits among all the
coefficients. In order to minimize the mean square error for


CHAPTER FOUR ALPHABET SIZE SELECTION
FOR VIDEO SIGNAL CODING 79
4.1 Introduction 79
4.2 Preliminaries 80
4.3 Analysis 83
4.4 Implementation issues 86
4.5 Nonbinary BCH Codes 94
4.6 Results 99
4.7 Summary 104
CHAPTER FIVE EFFICIENT DECODING OF CORRELATED SEQUENCES 116
5.1 Introduction 116
5.2 Optimum Decoding of Markov Sequences . 119
5.3 A Modified MAP (MMAP) Receiver ...... 122
5.4 A Minimum Cost Decoder 12 3
5.5 A Maximum Signal-To-Noise Ratio (MSNR)
Receiver 124
5.6 Redundancy in the Encoded Signals 127
5.7 Picture Transmission over noisy
Channels 130
5.8 Side Information 144
5.9 Summary 145
CHAPTER SIX CONCLUSIONS AND SUMMARY 147
6.1 Summary of the Work 147
6.2 Directions of Future Research 150
APPENDIX A. EVALUATION OF AN INTEGRAL 151
APPENDIX B. EVALUATION OF THE DERIVATIVES ..... 153
REFERENCES 155
BIOGRAPHICAL SKETCH 159
v


24
x2 £-1 *H
eq2 = f (x-l1)2p{x) dx+Y, f (x-li)2p(x) dx
1~Z Xi
** f 2 1 ^
+f (x-lL)2p(x)dx.
*L
Our purpose is to choose quantum levels 1¡ and interval
boundaries x¡ so that eq.(2.1) is minimized.
The above expression can be differentiated to obtain a
set of necessary conditions that must hold for the optimum
quantizer. By applying Leibniz's rule-we get
(2.2)
92,
= -2 f (x~li)2p(x) dx=0, i=l, 2 . L,
de
dx
2- = t (x-li.1)2-(xi-li)2)p(x) dx=0, i=2,
L. (2.3)
where x,=-o and xL+1=>. Equation (2.3) is equivalent to
i=2,3 L, (2.4)
which says that interval boundaries should fall midway between
the adjacent quantum levels. Alternatively,
li=2xi-li.i, i=2,3 . L. (2.5)


CHAPTER SIX
CONCLUSIONS AND DISCUSSIONS
6.1 Summary of the Work
Some important issues related to nonsymmetric information
sources such as signal design, alphabet size selection, and
decoding of information were investigated. New methods for
signal design including a procedure that minimizes a cost
function subject to different power criteria were developed.
The role of alphabet size in a communication system for a
asymmetric source was studied. It was found that the
performance is sensitive to size of the alphabet. Subject to
a fixed bandwidth, information rate and average power
constraint, the ternary system was found to superior to all
the others. Optimum tree decoding of correlated sequences was
addressed and modified MAP (MMAP) receiver and a maximal SNR
(MSNR) receiver were derived for decoding of correlated
sources. Experimental results showed the effectiveness of the
method. Further more, they were found to be superior to a
recently reported technique.
147


138
To get a feel for the improvement process, a horizontal
line of the video signal, the corrupted version of it and the
enhanced version of the line using MMAP receiver are shown in
Figure 5.9. The enhanced signal looks almost like the
reference signal indicating that the decoding process is quite
effective.
(b)
Figure 5.9. (a) A horizontal line of video signal, (b)
corrupted version of the video line and (c) the enhanced
version of the line.


42
utana.
Pv=1^ J exp [-(u- -^-)2] J exp[-e2]
u-0 N o --utanPj
de du
ucanttj
f exp [- (u- ^-)2] f exp [-e2] de du
71 O N ..Lp,
utana,
f exp [-{u- }2] f exp[-e!]
V-o N w
de du.
Now by definition
*2
erf(x1,x2) = (exp (-t2) dt.
Therefore, the following is the equation for Pw:
(3.20)
(3.21)
Pw=l--^- f exp [- (u- -^-)2]erf(-u tanpi#u tana2) du
2v^uJ>0 N
2/rt ji
/2LZJ2
exp [- (u-, ^)2] erf(-u tanp,, u tana.) du
\ jW_
i-n
f exp [-(u-
2\/u-L0 N
Atf2
)2]eri(-u tanP2,u tana,) du.
(3.22)


BIOGRAPHICAL SKETCH
Shahriar Emami received his BSEE with honors from the
University of Florida in August 1986. He then entered Graduate
School at the University of Florida and received his MS in
December 1987. Meanwhile he worked as a graduate teaching
assistant conducting electrical engineering laboratories and
grading for graduate and undergraduate courses. Afterwards,
Shahriar entered the Ph.D. program and focused primarily on
digital communications. He has been working as a research
assistant for Dr. Scott L. Miller at the Telecommunication
Research Lab and is expected to receive his Ph.D. in August
1993.
159


45
Ju e~iu~k)2e~uil du= e
k2l
(i+D
k*
e ^y/2Â¥ k
2(1+1) 2 (1+J)3
erf (
v/Tl+IT
)
(3.27)
If we substitute for integrals from eq. (3.27) into eq.
(3.23) and eq. (3.24) we will end up with the following set of
nonlinear equations:
(Pi~p3) e
AH2
N
71
AH2
2\| Nt
(Pi+P2) e
ah2
Icosa.lerfi- |cosal|,>)
1 11 N No
-(P2+P3) e
Na
AH2,
|cos (a1+a2)|erf (-. (cos (a1+a2)|/~)
N No /
(3.28)
-sin2a2
AH2
(p1-p2)e N + -| (Pi+p3) e N '|cosa2|erf(- ^^|cosa2|,~)
^ \ V N
AH2,
-Ajf. sin2 (p2+P3)e 0 (cos (ax+a2) (erf (
AH2
N ^
(cos (a1+a2)|/)
=0
(3.29)
Given a signal-to-noise ratio and a probability set the
optimum angles can be found using eq. (3.28) and eq. (3.29).


22
only as good as the quantized samples allow. In other words,
there remains some error, the quantization error between the
original and the reconstructed waveform which is related to
the parameters of the quantizer.
Let the analog signal be modeled as a random waveform and
let p(x) be the probability density function of the signal.
The process of quantization subdivides the range of the values
of x into a number of discrete intervals. If a particular
sample value of the analog signal falls anywhere in a given
interval, it is assigned a single discrete value corresponding
to that interval. The intervals fall between boundaries
denoted by x,,x2, . .,xL+,, where there are L intervals. The
quantized values are denoted by 1,,12,...,1L and are called
quantum levels or representative levels. The width of an
interval is xi+1 -x¡ and is called interval's step size. If all
the steps are equal and, in addition, the quantum level
separations are all the same, the quantizer is said to be
uniform; otherwise it is a nonuniform quantizer.
It is possible to design a quantizer for a given
probability density function and a given number of levels. The
optimal quantizer is non-uniform unless the signal has a
uniform pdf. If a uniform quantizer is used instead the mean
squared quantization error will be larger than that of the
optimal nonuniform quantizer.


131
according to formula given in [6].
The MMAP receiver and the MSNR receiver with first,
second and third order Markov models were utilized. For the
first order Markov modelling, the decoder uses the previous
symbol along the same line. Figure 5.4 shows the neighboring
symbols used in second and third order modelling of the
source.
The codeword assignment is given in Table 5.1. Other
codeword assignments such as natural and gray level coding
assignment were also tried for purpose of comparison, but the
given assignment consistently resulted in the largest SNR for
the reconstructed signal.
The conditional probability of error is given by
p(ri|vi)=p(l-p) (5.21)
where a is the Hamming distance between r¡ and v¡ and n is the
number of bits in a DPCM word.
We used a seven-level quantizer given in [14] that was
found experimentally and is seen to give satisfactory results
for variety of pictures.
We define signal-to-noise ratio (SNR) as the ratio
between the power in the original picture to the power in the
difference between the original and the received picture.


120
P(z\y) ^PUJvj)
(5.4)
For an Mth order Markov sequence
P(lf) =JIP(vi\vi-i'vi-2 v-m^ (5.5)
i
Substituting (5.4) and (5.5) into (5.3) we get
p(y|x) =
nPir^mvJvw
Til)
/ Vi-M>
Taking the logarithm of both sides gives
(5.6)
logP(:z|z) =52 log (PirJvj) P(vi|vi.1, . ., vi.M)) -log P(n) .
i
(5.7)
One might be tempted to use Viterbi algorithm for
decoding; However, Viterbi decoding is not applicable to this
problem because it makes a decision at each node without
considering the effect of the next symbol. The next symbol
could have a drastic effect on the cumulative metric.
Therefore, the use of Viterbi decoding will not yield the
optimum path.
Since the decision metric for each sample depends not
only on the present sample but also the previous ones, a code
tree must be formed and delayed decision decoding must be
employed. The code tree consists of a number of nodes and
branches. From each node there are Q branches which extend the


77
2f
xxxxxxixxxxixxkkikxxihxxkixxxxxxxkxxixx
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
. XXXXXXXXXXXXXXXXXXKXXXKXXXXXXXKKXXXXXXX
1.5 -XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX*
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
XXXXXXXXXXXKXXKXXXXXXXXXHKXXXXXXXXXXXXK
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
..xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
1-MKXXXXXXXXXXXXXKXXXXXXXXXXXXXXXXXXXXXX
xxxxxxxxxxxxxxxxxxxaxxxxxxxxxxxxxxxxxxx
.XXXXXXXXXXXXXXXXXXf
xxxxxxxxxxkx+44 444+
XXXXX44644444+0*000
444444440004M 0 00 94-
-1.5
MM
xxxxxxxxxxxxxxxxxxx
44444XXXXXXXXXXXXXX
4+4H+44444xxxk*xxx
0.5 4444444444444444444
HIM
444444444444t44444
utXM
0004000090600000000
oooooooooooooootoo
OOOOOOODOOOOOOOOOOO
-I
-0.5
0 i
(a)
2
1.5
1
a
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
XXXIXXXXXXIXXXXXXXXXX1XXXXXXXXKXXXXXXIH
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxixixxxxxxx
XX1XXXXXXXXXXXXXXXXXXXXXXXXXX1XXXXXXXXX
-1.5
xxxxxxxxxxxixxxixxxxxxixxxixxXxxxixxxxx
xxxxxxxxxxxxxxxxxxx
5 44444444444444XXXXX
.samwmmtm
.mm
nuta*"*"
400
xxxxxxxxxxxxxxxxxxx
XXXX44I000900990940-
-0.5
0.5
(b)
Figure 3.10. The decision regions for minimum cost signal
design, (a) channel SNR=3.5 dB and (b) channel SNR=11 dB. The
signals are represented by *'s and the decision regions are
marked by distinct symbols.


26
2.5 Analysis of DPCM Encoder
Consider an information source with alphabet {a1; a2, . ,
aQ}. The information source is said to be nonsymmetric if the
source symbols are not equally likely. It will be shown that
the output of a DPCM encoder can be viewed as a nonsymmetric
information source.
Assume that in a DPCM encoder a predictor of order M is
used. Let us model the quantizer as. a additive noise source
e(n)=e(n)+q(n) (2.7)
where e(n) and §(n) are the input and the output of the
quantizer respectively and q(n) is the quantization noise. It
has been shown that [12]
e(n) =e(n) bx q(n-i) (2*8)
-i
where e(n) is the difference signal in a DPCM without the
quantizer and b¡ are the prediction coefficients. The
distribution of e(n) is therefore given by the
M
convolution of the pdf's of e(n) and £2=]^ g(n-i) ;
-1


121
node to the next time sample. The path with the largest
cumulative metric of
X) log (PUi | v) PivJ, . ., v^H))
i
will be the optimum path. Associated with each branch of the
tree is a branch metric. The branch metrics of the tree are
given by
Tirjvj) =log(P(ri|vi) Piv^v^, . ., v^)) (5.8)
Once the tree is appropriately populated, the next task
is to select a sequence that minimizes the additive metric. An
exhaustive search over all candidate sequences will provide
the optimum solution. However, the enormous computational
burden of the exhaustive search suggests the use of a non-
exhaustive search procedure that concentrates on the most
promising portions of the tree.
There are a number of tree searching algorithms. We
choose to use the (M,L) algorithm [34]. This algorithm being
a breadth-first procedure examines all the branches extending
from the current M nodes. At this point only the best M of
these branches in the sense of largest cumulative path metric
are saved. An example is presented in Figure 5.1 where the
branching factor of the tree is three. At each depth only
three branches are saved and the rest are discarded.



PAGE 2

7R P\ SDUHQWV IRU WKHLU ORYH SDWLHQFH DQG VXSSRUW

PAGE 3

$&.12:/('*(0(176 , ZLVK WR H[SUHVV P\ JUDWLWXGH WR P\ DGYLVRU 'U 6FRWW / 0LOOHU IRU KLV HQFRXUDJHPHQW VXSSRUW DQG IULHQGVKLS , DP DOVR JUDWHIXO WR 'U &RXFK IRU KLV JXLGDQFH DQG LQVLJKW ,Q DGGLWLRQ , ZRXOG OLNH WR WKDQN 'U &KLOGHUV DQG 'U 1DMDIL IRU WKHLU WLPH DQG LQWHUHVW LQ VHUYLQJ LQ P\ FRPPLWWHH 6SHFLDO WKDQNV DOVR JR WR 'U 6LJPRQ ZKR KDV RIIHUHG KLV YDOXDEOH DVVLVWDQFH WKURXJKRXW WKH FRXUVH RI WKLV VWXG\ LLL

PAGE 4

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

PAGE 5

&+$37(5 )285 $/3+$%(7 6,=( 6(/(&7,21 )25 9,'(2 6,*1$/ &2',1* ,QWURGXFWLRQ 3UHOLPLQDULHV $QDO\VLV ,PSOHPHQWDWLRQ LVVXHV 1RQELQDU\ %&+ &RGHV 5HVXOWV 6XPPDU\ &+$37(5 ),9( ()),&,(17 '(&2',1* 2) &255(/$7(' 6(48(1&(6 ,QWURGXFWLRQ 2SWLPXP 'HFRGLQJ RI 0DUNRY 6HTXHQFHV $ 0RGLILHG 0$3 00$3f 5HFHLYHU $ 0LQLPXP &RVW 'HFRGHU $ 0D[LPXP 6LJQDO7R1RLVH 5DWLR 0615f 5HFHLYHU 5HGXQGDQF\ LQ WKH (QFRGHG 6LJQDOV 3LFWXUH 7UDQVPLVVLRQ RYHU QRLV\ &KDQQHOV 6LGH ,QIRUPDWLRQ 6XPPDU\ &+$37(5 6,; &21&/86,216 $1' 6800$5< 6XPPDU\ RI WKH :RUN 'LUHFWLRQV RI )XWXUH 5HVHDUFK $33(1',; $ (9$/8$7,21 2) $1 ,17(*5$/ $33(1',; % (9$/8$7,21 2) 7+( '(5,9$7,9(6 5()(5(1&(6 %,2*5$3+,&$/ 6.(7&+ Y

PAGE 6

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f RI YLGHR VLJQDOV LV GHULYHG DQG LW LV VKRZQ WKDW '3&0 RI YLGHR VLJQDOV UHVXOWV LQ D QRQV\PPHWULF VRXUFH 7KH SUREOHP RI VLJQDO VHOHFWLRQ IRU QRQV\PPHWULF VRXUFHV LQ WZR GLPHQVLRQV LV FRQVLGHUHG ,WHUDWLYH PHWKRGV IRU ILQGLQJ WKH PLQLPXP HUURU VLJQDO DQG PLQLPXP FRVWf FRQVWHOODWLRQ VXEMHFW WR DQ DYHUDJH RU D SHDNf SRZHU FRQVWUDLQW DUH SUHVHQWHG YL

PAGE 7

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f UHFHLYHU XVLQJ KLJKHU RUGHU VWDWLVWLFV DQG D PD[LPXP VLJQDOWRQRLVH UDWLR 0615f UHFHLYHU DUH DOVR JLYHQ 7KH GHFRGLQJ SURFHGXUHV ZHUH DSSOLHG WR LPDJH FRPPXQLFDWLRQ RYHU QRLV\ FKDQQHOV ,Q VXPPDU\ WKH PDMRU FRQWULEXWLRQV RI WKLV GLVVHUWDWLRQ ZHUH WKH GHYHORSPHQW RI VLJQDO VHOHFWLRQ PHWKRGV IRU QRQV\PPHWULF VRXUFHV GHULYDWLRQ RI SURFHGXUHV IRU GHFRGLQJ RI FRUUHODWHG VRXUFHV DQG DSSOLFDWLRQ RI WKHVH SURFHGXUHV WR WKH SLFWXUH FRPPXQLFDWLRQ LQ QRLV\ VLWXDWLRQV YLL

PAGE 8

&+$37(5 21( 5(9,(: 2) 02'8/$7,21 7(&+1,48(6 6285&( $1' &+$11(/ &2',1* ,1 ',*,7$/ &20081,&$7,216 ,QWURGXFWLRQ 7KH JRDO RI WKLV FKDSWHU LV WR SUHVHQW WKH EDFNJURXQG QHFHVVDU\ WR IROORZ WKH ZRUN SUHVHQWHG LQ WKLV GLVVHUWDWLRQ LQWURGXFH WKH UHVHDUFK REMHFWLYHV DQG JLYH D EULHI GHVFULSWLRQ RI WKH FKDSWHUV 6LQFH WZRGLPHQVLRQDO PRGXODWLRQ KDV EHHQ XWLOL]HG LQ WKLV ZRUN WKHVH IRUPDWV DUH UHYLHZHG WKHLU VSHFWUDO HIILFLHQF\ DUH FDOFXODWHG DQG WKH XSSHU ERXQG RQ WKHLU VSHFWUDO HIILFLHQF\ LV JLYHQ 3LFWXUH WUDQVPLVVLRQ LV RQH RI WKH DSSOLFDWLRQV SUHVHQWHG KHUH 7R IDPLOLDUL]H WKH UHDGHU ZLWK WKH ILHOG D QXPEHU RI VRXUFH FRGLQJ WHFKQLTXHV KDYH EHHQ GHVFULEHG (UURU FRUUHFWLRQ RYHU QRLV\ FKDQQHOV LV DQ LPSRUWDQW WRSLF DQG KDV EHHQ XWLOL]HG LQ WKLV WKHVLV 7KH JDLQ IURP FKDQQHO FRGLQJ KDV EHHQ H[SODLQHG DQG GLIIHUHQW PHWKRGV RI HUURU FRUUHFWLRQ KDYH EHHQ GLVFXVVHG 7KH HIIHFWV RI FKDQQHO HUURUV LQ D '3&0 V\VWHP DQG RSWLPXP SUHGLFWLRQ LQ QRLV\ FKDQQHOV ZDV SUHVHQWHG EHFDXVH WKH HQKDQFHPHQW RI SLFWXUH TXDOLW\ LQ '3&0 V\VWHPV KDV EHHQ DGGUHVVHG LQ WKLV GLVVHUWDWLRQ

PAGE 9

7ZR'LPHQVLRQDO 0RGXODWLRQ )RUPDWV 0DU\ SKDVHVKLIW NH\LQJ 036.f DQG TXDGUDWXUH DPSOLWXGH PRGXODWLRQ 4$0f DUH DPRQJ WKH PRVW SRSXODU WZR GLPHQVLRQDO IRUPDWV ,Q 036. WKH WUDQVPLWWHG VLJQDO LV JLYHQ E\ VWf 5H^J Wf H-FW` f ZKHUH JWf $ HFf f ,Q RWKHU ZRUGV LQ 036. ZKLOH WKH DPSOLWXGH LV PDLQWDLQHG FRQVWDQW WKH SKDVH RI VLJQDO FDQ WDNH RQ RQH RI WKH 0 YDOXHV LQ D V\PERO LQWHUYDO 7KH 036. IRU 0 LV FDOOHG TXDGUDWXUH SKDVHVKLIW NH\LQJ 436.f ,Q 036. VLJQDO SRLQWV DUH FRQILQHG WR WKH FLUFXPIHUHQFH RI D FLUFOH %XW LQ 4$0 WKH WUDQVPLWWHG VLJQDO LV VWf 5H^J Wf H-f¯FW f ZKHUH JU Wf [ Wf M\ Wf f RU

PAGE 10

VWf [Wf FRV R!FWf \ Wf VLQRFWf f 7KH ZDYHIRUPV [Wf DQG \Wf DUH [ Wf @7[L KWL7f f L DQG \Wf \ KWL7f f L ZKHUH 7 LV WKH V\PERO LQWHUYDO LQ VHFRQGV DQG KWf LV WKH SXOVH VKDSH /HW XV ILQG WKH VSHFWUDO HIILFLHQF\ RI 036. DQG 4$0 ZLWK UHFWDQJXODU SXOVHV 7KH QXOOWRQXOO WUDQVPLVVLRQ EDQGZLGWK RI 036. DQG 4$0 LV f ZKHUH 0 f« LV WKH QXPEHU RI SRLQWV LQ WKH VLJQDO FRQVWHOODWLRQ 7KH VSHFWUDO HIILFLHQF\ LV WKHUHIRUH JLYHQ E\ f± B 5 B EL WVVHF A %W +] f :KHQ RSHUDWLQJ RYHU D EDQGOLPLWHG FKDQQHO DQG WKH RYHUDOO SXOVH VKDSH VDWLVILHV WKH UDLVHG FRVLQH UROORII ILOWHU FKDUDFWHULVWLFV WKH EDQGZLGWK RI WKH PRGXODWLQJ VLJQDO LV

PAGE 11

%B LUf' f ZKHUH ' 5 DQG U LV WKH UROORII IDFWRU 6LQFH %7 % WKH WUDQVPLVVLRQ EDQGZLGWK RI 4$0 LV G B Uf5 Q7 M LLLf DQG WKH VSHFWUDO HIILFLHQF\ ZLWK UDLVHG FRVLQH ILOWHULQJ LV JLYHQ E\ ORJA$7 EM WVVHF f Uf +] 7KH VSHFWUDO HIILFLHQF\ LQFUHDVHV ZLWK WKH QXPEHU RI WKH SRLQWV LQ WKH FRQVWHOODWLRQ +RZHYHU RQH FDQ QRW LQFUHDVH WKH VSHFWUDO HIILFLHQF\ E\ LQFUHDVLQJ WKH QXPEHU RI SRLQWV LQ WKH VLJQDO FRQVWHOODWLRQ WRR PXFK EHFDXVH DV \RX SODFH PRUH VLJQDOV LQ WKH FRQVWHOODWLRQ WKH HUURU UDWH LQFUHDVHV )RU UHOLDEOH FRPPXQLFDWLRQV WKH LQIRUPDWLRQ UDWH PXVW EH NHSW EHORZ WKH FKDQQHO FDSDFLW\ 7KHUHIRUH WKH VSHFWUDO HIILFLHQF\ LV XSSHU ERXQGHG E\ WLORJO! f 1 ZKHUH 61 LV WKH VLJQDOWRQRLVH SRZHU UDWLR 7ZR GLPHQVLRQDO IRUPDWV DUH ZHOO VXLWHG IRU KLJK VSHHG GDWD WUDQVPLVVLRQ EHFDXVH RI WKHLU HIILFLHQW XVH RI EDQGZLGWK +RZHYHU WKH\ UHTXLUH FRKHUHQW GHWHFWLRQ WKDW LPSOLHV WKH QHHG IRU V\QFKURQL]DWLRQ FLUFXLWV

PAGE 12

6RXUFH &RGLQJ 7KH SXUSRVH RI VRXUFH FRGLQJ LV WR UHPRYH DV PXFK UHGXQGDQF\ DV SRVVLEOH IURP WKH PHVVDJH (IILFLHQW FRGLQJ RI PHVVDJHV SURYLGHV WKH RSSRUWXQLW\ IRU VLJQLILFDQWO\ GHFUHDVLQJ WKH WUDQVPLVVLRQ FRVWV 7ZR PDLQ DSSURDFKHV WR SLFWXUH FRGLQJ SUHGLFWLYH FRGLQJ DQG WUDQVIRUP FRGLQJ ZLOO EH DGGUHVVHG KHUH '3&0 7KHUH LV FRQVLGHUDEOH FRUUHODWLRQ EHWZHHQ DGMDFHQW VDPSOHV RI VSHHFK RU LPDJH GDWD DQG LQGHHG WKH FRUUHODWLRQ LV VLJQLILFDQW HYHQ EHWZHHQ VDPSOHV WKDW DUH VHYHUDO VDPSOLQJ LQWHUYDOV DSDUW 7KH PHDQLQJ RI WKLV KLJK FRUUHODWLRQ LV WKDW LQ DQ DYHUDJH VHQVH WKH VLJQDO GRHV QRW FKDQJH UDSLGO\ IURP VDPSOH WR VDPSOH VR WKDW WKH GLIIHUHQFH EHWZHHQ DGMDFHQW VDPSOHV VKRXOG KDYH D ORZHU YDULDQFH WKDQ WKH YDULDQFH RI WKH VLJQDO LWVHOI 7KH SUHGLFWHG YDOXH LV WKH RXWSXW RI WKH SUHGLFWRU V\VWHP ZKRVH LQSXW LV D TXDQWL]HG YHUVLRQ RI WKH LQSXW VLJQDO 7KH GLIIHUHQFH VLJQDO PD\ DOVR EH FDOOHG WKH SUHGLFWLRQ HUURU VLJQDO VLQFH LW LV WKH DPRXQW E\ ZKLFK WKH SUHGLFWRU IDLOV WR H[DFWO\ SUHGLFW WKH LQSXW 6LQFH WKH YDULDQFH RI HUURU VLJQDO LV VPDOOHU WKDQ WKH YDULDQFH RI VLJQDO D TXDQWL]HU ZLWK D JLYHQ QXPEHU RI OHYHOV

PAGE 13

FDQ EH DGMXVWHG WR JLYH D VPDOOHU TXDQWL]DWLRQ HUURU WKDQ ZRXOG EH SRVVLEOH ZKHQ TXDQWL]LQJ WKH LQSXW GLUHFWO\ 2SWLPXP 3UHGLFWLRQ :H DUH LQWHUHVWHG LQ OLQHDU SUHGLFWLRQ RI WKH IRUP 1 eQf ne DM [QMf f ZKLFK LV WKH ZHLJKWHG VXP RI 1 SUHYLRXV VDPSOHV 7KH ZHLJKWV DM DUH OLQHDU SUHGLFWLRQ FRHIILFLHQWV 7KH ILOWHU LV RSWLPL]HG E\ ILQGLQJ WKH ZHLJKWV WKDW PLQLPL]H SUHGLFWLRQ HUURU LQ D PHDQ VTXDUHG VHQVH RH (>[Qf6WQfff f 6LQFH PHDQ VTXDUHG HUURU LV D IXQFWLRQ RI DM DQG GR A  1 f GD LV D QHFHVVDU\ FRQGLWLRQ IRU PLQLPXP 06( PHDQVTXDUHG HUURUf (YDOXDWLQJ WKH GHULYDWLYH JLYHV (>L[Qf eQffM"ULf @ f D (TXDWLQJ WKLV WR ]HUR \LHOGV (>^[Qf [Qf f[QLf @ L O 1 f 7KLV LV FDOOHG WKH RUWKRJRQDOLW\ SULQFLSOHV ZKLFK VWDWHV WKDW PLQLPXP HUURU PXVW EH RUWKRJRQDO WR DOO GDWD XVHG LQ WKH

PAGE 14

SUHGLFWLRQ 7KH H[SDQVLRQ RI WKLV HTXDWLRQ JLYHV WKH IROORZLQJ FRQGLWLRQ IRU RSWLPXP DM 1 e D 5ALNMf 5[[Nf N O1 f ML RU nAf Af rrrf 5MRF,1 fn D ? DL 5MRFL2f 5AGf 5AL1f D Amf U-:f 5AL1f 5A1 f 5MF[nL f D1 rrrf¬r! f LQ PDWUL[ QRWDWLRQ ‘5[[D Ur[ f ZKHUH U[[ ^5[[Gff 1 f 7KH HTXDWLRQV DUH FDOOHG QRUPDO HTXDWLRQV @

PAGE 15

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f LV H[SOLFLWO\ NQRZQ EXW FRPSXWDWLRQDOO\ LW LV YHU\ GHPDQGLQJ 7KLV XQGHVLUDEOH IHDWXUH KDV SUHYHQWHG DQ\ KDUGZDUH LPSOHPHQWDWLRQ RI WKH RSWLPXP WUDQVIRUP ,W LV PDLQO\ VWXGLHG LQ VLPXODWLRQV WR REWDLQ ERXQGV 7KH PRVW SUDFWLFDO WUDQVIRUP FRGLQJ WHFKQLTXHV DUH EDVHG RQ WKH '&7 GLVFUHWH FRVLQH WUDQVIRUPf ZKLFK SURYLGHV D JRRG FRPSURPLVH EHWZHHQ LQIRUPDWLRQ SDFNLQJ DELOLW\ DQG FRPSXWDWLRQDO FRPSOH[LW\ ,Q IDFW WKH SURSHUWLHV RI '&7 KDYH SURYHG RI VXFK SUDFWLFDO YDOXH WKDW LW KDV EHFRPH WKH

PAGE 16

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

PAGE 17

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f GLJLWV ZKLFK WKH HQFRGHU DGGV WR HDFK PHVVDJH EORFN DUH FDOOHG UHGXQGDQW GLJLWV WKH\ FDUU\ QR QHZ LQIRUPDWLRQ 7KH UDWLR RI GDWD ELWV WR WRWDO ELWV ZLWKLQ D EORFN NQ LV FDOOHG FRGH UDWH 7KH FRGH LWVHOI LV UHIHUUHG WR DV DQ QNf FRGH 7R GHPRQVWUDWH WKH SHUIRUPDQFH LPSURYHPHQW SRVVLEOH ZLWK

PAGE 18

FKDQQHO FRGLQJ OHW XV SLFN D f VLQJOH HUURU FRUUHFWLQJ FRGH $VVXPH D %36. PRGXODWLRQ D VLJQDOWRQRLVH UDWLR RI 61 f DQG D GDWD UDWH RI 5 EV /HW 3XE DQG 3XP UHSUHVHQW WKH ELW DQG PHVVDJH HUURU UDWH IRU WKH XQFRGHG V\VWHP DQG 3FE 3FP UHSUHVHQW WKH ELW DQG PHVVDJH HUURU UDWH IRU WKH FRGHG V\VWHP UHVSHFWLYHO\ ZLWKRXW FRGLQJ f SXE ? Af [O f DQG 38LD OO38WEf OO O [ f :LWK FRGLQJ L"F [ f E6 f f [O2 7KH ELW HUURU UDWH IRU WKH FRGHG V\VWHP LV LQIHULRU WR WKDW RI WKH XQFRGHG V\VWHP DQG WKH SHUIRUPDQFH LPSURYHPHQW GXH WR

PAGE 19

FRGLQJ LV QRW DSSDUHQW \HW 6LQFH WKH FRGH FRUUHFWV DOO VLQJOH HUURUV ZLWKLQ D EORFN RI ELWV WKH PHVVDJH HUURU UDWH IRU WKH FRGHG V\VWHP ZLOO EH 3FP (Af 3FO3FMEf-F [O f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f 5HHG6RORPRQ FRGH IRU MRLQW WDFWLFDO LQIRUPDWLRQ GLVWULEXWLRQ V\VWHP-7,'6f DQG D f %&+ FRGH IRU ,17(/6$7 9 V\VWHP >@

PAGE 20

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f 7KLV VHOHFWLRQ RI VXUYLYRU LV SHUIRUPHG RQ IRU DOO SDWKV HQWHULQJ HDFK RI WKH RWKHU VWDWHV 7KH GHFRGHU FRQWLQXHV LQ WKLV ZD\ WR DGYDQFH GHHSHU LQWR WKH WUHOOLV PDNLQJ GHFLVLRQV E\ HOLPLQDWLQJ WKH OHDVW OLNHO\ SDWKV 7KH FRPSOH[LW\ RI 9LWHUEL DOJRULWKP LV DQ H[SRQHQWLDO

PAGE 21

IXQFWLRQ RI WKH FRGHnV FRQVWUDLQW OHQJWK )RU ODUJH YDOXHV RI FRQVWUDLQW OHQJWK .!!f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

PAGE 22

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f WR D ZURQJ YDOXH YQf 'XH WR OLQHDULW\ RI WKH GHFRGHU ILOWHU WKH FRUUHFW FRPSXWDWLRQ RI RXWSXW LV VXSHUSRVHG E\ DQ HUURU RXWSXW FDXVHG E\ LQSXW FQf XQfYQf WR WKH GHFRGHU ORRS 6LQFH WKH GHFRGHU LV DQ DOOSROH ILOWHU WKHUH ZLOO EH DQ LQILQLWH VHTXHQFH RI HUURU VDPSOHV DW WKH RXWSXW ZLWK GHFD\LQJ DPSOLWXGHV ,Q WKH FDVH RI ILUVWRUGHU SUHGLFWLRQ WKH HIIHFW RQ D IXWXUH YDOXH DW WLPH P LV GHVFULEHG E\ >@ &Pf &Qf D PaQf P]Q f 7UDQVPLVVLRQ HUURUV WKHUHIRUH SURSDJDWH LQ WKH UHFRQVWUXFWHG '3&0 ZDYHIRUP 7KLV NLQG RI HUURU VPHDULQJ LV SHUFHSWXDOO\ GHVLUDEOH LQ VSHHFK FRGLQJ ZKHUH D 3&0 HUURU VSLNH RI ODUJH PDJQLWXGH LV PRUH DQQR\LQJ WKDQ D ORZ DPSOLWXGH HUURU VPHDUHG RYHU D ORQJ GXUDWLRQ

PAGE 23

,Q SLFWXUH FRGLQJ RQ WKH RWKHU KDQG HUURU SURSDJDWLRQ LV SHUFHSWXDOO\ YHU\ XQGHVLUDEOH WDNLQJ WKH IRUP RI YHU\ YLVLEOH VWUHDNV RU EORWFKHV ZLWK RQH DQG WZR GLPHQVLRQDO SUHGLFWRUV 2SWLPXP 3UHGLFWLRQ IRU 1RLV\ &KDQQHOV 2QH RI WKH HDUO\ DSSURDFKHV WR V\VWHP RSWLPL]DWLRQ XQGHU QRLV\ FRQGLWLRQV ZDV SUHVHQWHG E\ &KDQJ DQG 'RQDOGVRQ >@ %HFDXVH RI WKH LPSRUWDQFH RI WKH UHVXOW DQG WKH UHOHYDQFH WR WKLV GLVVHUWDWLRQ D VXPPDU\ RI WKH PHWKRGV LV JLYHQ KHUH /HW Uc GHQRWH WKH UHFHLYHG VLJQDO Ic GHQRWH WKH LPSXOVH UHVSRQVH RI WKH '3&0 GHFRGHU 7KH RXWSXW RI WKH GHFRGHU LV WKHUHIRUH JLYHQ E\ eL ULrIL VLQLf rIL VrIL Q r [ TLQLrIL f /HW XV GHILQH [nc DV IROORZV YDU [L LFL [[ JBe QL r IL ;Lf (>2L`AIN(L4L1LN` N N f

PAGE 24

/HW XV DVVXPH WKH FKDQQHO QRLVH LV XQFRUUHODWHG DQG WKH GLIIHUHQFH VLJQDO VDPSOHV DUH VWDWLVWLFDOO\ LQGHSHQGHQW WKHQ YDUL;M (>2L@ (>4L1L@ (>1L@neIN f N 7KH VHFRQG WHUP LV FDOOHG WKH PXWXDO HUURU DQG FDQ EH VKRZQ WR EH DSSUR[LPDWHO\ ]HUR LI WKH TXDQWL]HU LV QHDU RSWLPXP 7KH HUURU SRZHU UHGXFHV WR YDU¯[nLf (>4c@ (O1L`neIN f N 7KH VXP FDQ EH HYDOXDWHG XVLQJ DQ LGHQWLW\ LMW ODDD N D -E f 7KH H[SUHVVLRQ IRU WKH UHFRQVWUXFWLRQ HUURU YDULDQFH WKHQ EHFRPHV YDU;IBf YDU4Lf E YDU1f f :H QRZ GHILQH WKH IROORZLQJ TXDQWLWLHV WR UHODWH WKH TXDQWL]HU DQG QRLVH YDULDQFHV WR WKH GLIIHUHQWLDO VLJQDO YDULDQFH YDU4f NJ YDU(f YDU 1f .Q YDU(ef f 7KH '3&0 SUHGLFWLRQ JDLQ LV DOVR JLYHQ E\

PAGE 25

9-e+W!rDDS f S YDUL(M ZKHUH S LV HTXDO WR 5f±f 3XWWLQJ DOO WKLV WRJHWKHU \LHOGV WUD U < W ? YDU;Lf NENQfn f JS 7KH VHFRQG WHUP DERYH LV WKH GRPLQDQW WHUP EHFDXVH WKH HIIHFW RI FKDQQHO QRLVH LV PXFK PRUH GHVWUXFWLYH WR WKH UHFRQVWUXFWLRQ RI WKH LPDJH WKDQ WKH HIIHFW GXH WR TXDQWL]DWLRQ QRLVH YDU;nMf . YD[;Lf D 3L D f D 7R PLQLPL]H WKH YDULDQFH RI WKH UHFRQVWUXFWLRQ HUURU ZH ZLOO VHW WKH GHULYDWLYH RI WKLV H[SUHVVLRQ ZLWK UHVSHFW WR D WR ]HUR 7KH RSWLPXP YDOXH RI D WXUQV RXW WR EH D A3L 3L f 5HVHDUFK 2EMHFWLYHV 7KLV GLVVHUWDWLRQ LV FRQFHUQHG ZLWK LVVXHV UHODWHG WR QRQV\PPHWULF LQIRUPDWLRQ VRXUFHV 7R PRWLYDWH WKH ZRUN RQ QRQV\PPHWULF VRXUFHV LW LV VKRZQ WKDW '3&0 RI GLJLWL]HG YLGHR VLJQDOV UHVXOWV LQ D QRQV\PPHWULF LQIRUPDWLRQ VRXUFH 2QH RI WKH PDLQ JRDOV LV WR DGGUHVV WKH SUREOHP RI VLJQDO GHVLJQ LQ

PAGE 26

WZR GLPHQVLRQV IRU QRQV\PPHWULF VRXUFHV ,W LV GHVLUHG WR ILQG DQ DOJRULWKPLF VROXWLRQ WR WKH PLQLPXP HUURU VLJQDO FRQVWHOODWLRQ IRU DYHUDJH DQG SHDN SRZHU FRQVWUDLQWV ,Q DGGLWLRQ WKH JHQHUDO FDVH ZKHUH WKH FRVW IXQFWLRQ LV QRW QHFHVVDULO\ WKH HUURU UDWH LV GLVFXVVHG (YHQ WKRXJK HIILFLHQW WHFKQLTXHV IRU VRXUFH FRGLQJ FKDQQHO FRGLQJ DQG VLJQDO GHVLJQ H[LVWV LW LV QRW NQRZQ KRZ WKH FKRLFH RI DOSKDEHW VL]H DIIHFWV D FRPPXQLFDWLRQ V\VWHP :H ZRXOG OLNH WR FRPSDUH FRPPXQLFDWLRQ V\VWHPV ZLWK YDULRXV DOSKDEHW VL]HV IRU WKH WUDQVPLVVLRQ RI YLGHR VLJQDOV RQ WKH EDVLV RI HTXDO LQIRUPDWLRQ UDWH EDQGZLGWK DQG DYHUDJH SRZHU 7ZR UHDOLVWLF VLWXDWLRQV ZLOO EH FRQVLGHUHG ZKHQ RQH LV RSHUDWLQJ XQGHU WLJKW EDQGZLGWK FRQVWUDLQW DQG ZKHQ WKH FRQVWUDLQW LV VRPHZKDW ORRVH 6\VWHP SHUIRUPDQFH FDQ EH LPSURYHG XVLQJ VWDQGDUG HUURU FRUUHFWLRQ WHFKQLTXHV DW WKH FRVW RI LQFUHDVLQJ WKH EDQGZLGWK RU UHGXFLQJ WKH LQIRUPDWLRQ UDWH +RZHYHU ZH ZRXOG OLNH WR XVH LQKHUHQW DV\PPHWU\ DQG UHGXQGDQF\ LQ WKH WUDQVPLWWHG SLFWXUH WR LPSURYH WKH UHFHSWLRQ :H ZLOO PRGHO WKH GDWD DV D 0DUNRY VRXUFH DQG GHULYH WKH RSWLPXP PHWKRG IRU GHFRGLQJ WKH GDWD :H ZLOO DOVR ILQG D UHFHLYHU WKDW LQVWHDG RI PLQLPL]LQJ WKH HUURU UDWH PD[LPL]HV WKH 615 VLJQDOWRQRLVHf UDWLR

PAGE 27

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

PAGE 28

&+$37(5 7:2 '3&0 9,'(2 6,*1$/ $ 1216<00(75,& ,1)250$7,21 6285&( ,QWURGXFWLRQ 7KH SXUSRVH RI WKLV FKDSWHU LV WR GHPRQVWUDWH WKDW '3&0 GLIIHUHQWLDO SXOVH FRGH PRGXODWLRQf RI SLFWXUHV UHVXOWV LQ D QRQV\PPHWULF LQIRUPDWLRQ VRXUFH 7R GR VR VRPH LQWURGXFWRU\ PDWHULDO LV SUHVHQWHG ILUVW 6LQFH D TXDQWL]HU LV DQ LPSRUWDQW FRPSRQHQW RI D '3&0 V\VWHP LW ZLOO EH H[DPLQHG LQ VRPH GHWDLO 4XDQWL]HUV ZLOO EH LQWURGXFHG GLIIHUHQW FULWHULD IRU WKH GHVLJQ ZLOO EH PHQWLRQHG DQG WKH SURFHGXUH IRU ILQGLQJ DQ RSWLPDO TXDQWL]HU LQ 064( VHQVHf ZLOO EH H[SODLQHG VWHS E\ VWHS $ '3&0 HQFRGHU ZLOO EH DQDO\]HG DQG D PRGHO IRU WKH UHVXOWLQJ VRXUFH VWDWLVWLFV ZLOO EH JLYHQ (YHQWXDOO\ WKH PRGHO ZLOO EH FRPSDUHG ZLWK DFWXDO SLFWXUH VWDWLVWLFV DQG UHVXOWV ZLOO EH FRPSDUHG %DVLFV RI 4XDQWL]HUV 4XDQWL]DWLRQ LV WKH SURFHVV RI URXQGLQJ VDPSOH YDOXHV WR D ILQLWH QXPEHU RI GLVFUHWH YDOXHV 7KLV SURFHVV LV QRW DQ LQIRUPDWLRQ SUHVHUYLQJ SURFHVV DQG WKH UHFRQVWUXFWHG VLJQDO LV

PAGE 29

RQO\ DV JRRG DV WKH TXDQWL]HG VDPSOHV DOORZ ,Q RWKHU ZRUGV WKHUH UHPDLQV VRPH HUURU WKH TXDQWL]DWLRQ HUURU EHWZHHQ WKH RULJLQDO DQG WKH UHFRQVWUXFWHG ZDYHIRUP ZKLFK LV UHODWHG WR WKH SDUDPHWHUV RI WKH TXDQWL]HU /HW WKH DQDORJ VLJQDO EH PRGHOHG DV D UDQGRP ZDYHIRUP DQG OHW S[f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c DQG LV FDOOHG LQWHUYDOnV VWHS VL]H ,I DOO WKH VWHSV DUH HTXDO DQG LQ DGGLWLRQ WKH TXDQWXP OHYHO VHSDUDWLRQV DUH DOO WKH VDPH WKH TXDQWL]HU LV VDLG WR EH XQLIRUP RWKHUZLVH LW LV D QRQXQLIRUP TXDQWL]HU ,W LV SRVVLEOH WR GHVLJQ D TXDQWL]HU IRU D JLYHQ SUREDELOLW\ GHQVLW\ IXQFWLRQ DQG D JLYHQ QXPEHU RI OHYHOV 7KH RSWLPDO TXDQWL]HU LV QRQXQLIRUP XQOHVV WKH VLJQDO KDV D XQLIRUP SGI ,I D XQLIRUP TXDQWL]HU LV XVHG LQVWHDG WKH PHDQ VTXDUHG TXDQWL]DWLRQ HUURU ZLOO EH ODUJHU WKDQ WKDW RI WKH RSWLPDO QRQXQLIRUP TXDQWL]HU

PAGE 30

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

PAGE 31

[ e r+ HT I [OfS^[f G[< I [OLfS[f G[ a= ;L rr I A I [O/fS[fG[ r/ 2XU SXUSRVH LV WR FKRRVH TXDQWXP OHYHOV c DQG LQWHUYDO ERXQGDULHV [c VR WKDW HTf LV PLQLPL]HG 7KH DERYH H[SUHVVLRQ FDQ EH GLIIHUHQWLDWHG WR REWDLQ D VHW RI QHFHVVDU\ FRQGLWLRQV WKDW PXVW KROG IRU WKH RSWLPXP TXDQWL]HU %\ DSSO\LQJ /HLEQL]nV UXOHZH JHW f I [aOLfS[f G[ L O / GH G[ W [OLf[LOLffS[f G[ L / f ZKHUH [ rR DQG [/ m! (TXDWLRQ f LV HTXLYDOHQW WR L / f ZKLFK VD\V WKDW LQWHUYDO ERXQGDULHV VKRXOG IDOO PLGZD\ EHWZHHQ WKH DGMDFHQW TXDQWXP OHYHOV $OWHUQDWLYHO\ OL [LOLL L / f

PAGE 32

(TXDWLRQ f LV UHDGLO\ VROYHG IRU c r M [ S [f G[  f rL}L I S[fG[ rL 7KH VROXWLRQ RI WKH HTXDWLRQV IRU WKH JHQHUDO QRQXQLIRUP TXDQWL]HU LV GLIILFXOW +RZHYHU D SURFHGXUH WR REWDLQ D VROXWLRQ E\ FRPSXWHU LWHUDWLRQ KDV EHHQ LQWURGXFHG E\ /O\RG DQG 0D[ >@>@ )RU D VSHFLILHG SUREDELOLW\ GHQVLW\ IXQFWLRQ DQG D IL[HG YDOXH RI / LV ILUVW VHOHFWHG DUELWUDULO\ ZLWK [ RR ZH VROYH HT f IRU [ 1H[W [ DQG DUH XVHG LQ HT f WR REWDLQ 7KH SURFHVV LV UHSHDWHG WR REWDLQ [ IURP HT f DQG IURP HT f &RQWLQXHG LWHUDWLRQ ILQDOO\ VWRSV ZKHQ / LV REWDLQHG IURP HT f ,I KDV EHHQ FRUUHFWO\ JXHVVHG WKHQ / ZLOO VDWLVI\ WKH HTXDWLRQ ZLWK [/ RR LI LW GRHV QRW LV FRUUHFWHG WR D QHZ YDOXH DQG WKH SURFHVV LV UHSHDWHG XQWLO / VDWLVILHV HT f 7KLV SURFHGXUH VDWLVILHV FRQGLWLRQV HT f DQG HT f ZKLFK DUH QHFHVVDU\ IRU RSWLPDOLW\ 0D[ >@ XVHG WKH DERYH SURFHGXUH WR ILQG WKH TXDQWXP OHYHOV IRU D ]HURPHDQ *DXVVLDQ PHVVDJH IRU TXDQWL]HUV XS WR OHYHOV 3DH] DQG *OLVVRQ >@ XVHG WKH SURFHGXUH WR ILQG RSWLPXP OHYHOV IRU VLJQDOV KDYLQJ HLWKHU D JDPPD GHQVLW\ RU D /DSODFH GHQVLW\ IRU / DQG

PAGE 33

$QDO\VLV RI '3&0 (QFRGHU &RQVLGHU DQ LQIRUPDWLRQ VRXUFH ZLWK DOSKDEHW ^D D D4` 7KH LQIRUPDWLRQ VRXUFH LV VDLG WR EH QRQV\PPHWULF LI WKH VRXUFH V\PEROV DUH QRW HTXDOO\ OLNHO\ ,W ZLOO EH VKRZQ WKDW WKH RXWSXW RI D '3&0 HQFRGHU FDQ EH YLHZHG DV D QRQV\PPHWULF LQIRUPDWLRQ VRXUFH $VVXPH WKDW LQ D '3&0 HQFRGHU D SUHGLFWRU RI RUGHU 0 LV XVHG /HW XV PRGHO WKH TXDQWL]HU DV D DGGLWLYH QRLVH VRXUFH HQf HQfTQf f ZKHUH HQf DQG iQf DUH WKH LQSXW DQG WKH RXWSXW RI WKH TXDQWL]HU UHVSHFWLYHO\ DQG TQf LV WKH TXDQWL]DWLRQ QRLVH ,W KDV EHHQ VKRZQ WKDW >@ HQf HQf E[ TQLf rf L ZKHUH HQf LV WKH GLIIHUHQFH VLJQDO LQ D '3&0 ZLWKRXW WKH TXDQWL]HU DQG Ec DUH WKH SUHGLFWLRQ FRHIILFLHQWV 7KH GLVWULEXWLRQ RI HQf LV WKHUHIRUH JLYHQ E\ WKH 0 FRQYROXWLRQ RI WKH SGInV RI HQf DQG e @A JQLf 

PAGE 34

3 3Hr 3T f $QG WKH SUREDELOLW\ RI TXDQWXP OHYHO L LV GHWHUPLQHG E\ 3L I S[f G[ f H Qf [Qf S Ef )LJXUH %ORFN GLDJUDP RI D '3&0 V\VWHP Df HQFRGHU DQG GHFRGHU

PAGE 35

2n1HDO >@ KDV VKRZQ H[SHULPHQWDOO\ WKDW WKH SGI RI HQf FDQ EH DSSUR[LPDWHG ZLWK D /DSODFLDQ GLVWULEXWLRQ S[f f§AH[S Arf f Y R r 7R ILQG WKH VWDWLVWLFV RI WKH OHYHOV DQ RSWLPXP TXDQWL]HU PXVW EH SODFHG LQ WKH '3&0 V\VWHP ,I ZH XWLOL]H D 064( TXDQWL]HU WKH VWDWLVWLFV DVVRFLDWHG ZLWK HDFK OHYHO ZLOO EH 3R 3 3 I L H[S f G[ H[S \f G\ Ar r D f L H[S\Df H[S3f ZKHUH D DQG " DUH WDEXODWHG LQ >@ :H KDYH DOVR YHULILHG WKURXJK VLPXODWLRQV WKDW WKH FKRLFH RI D JDPPD GLVWULEXWLRQ IRU HQf DQG XQLIRUP SGI IRU 4 UHVXOWV LQ D VDWLVIDFWRU\ DSSUR[LPDWLRQ WR WKH GHQVLW\ RI HQf ,Q RWKHU ZRUGV 3mr! _ QR _[_ f $ QXPHULFDO PHWKRG PXVW EH HPSOR\HG WR ILQG WKH VWDWLVWLFV DVVRFLDWHG ZLWK HDFK OHYHO

PAGE 36

5HVXOWV ,Q WKLV VHFWLRQ ZH ZLOO FRPSXWH WKH VWDWLVWLFV RI WKH TXDQWL]HUV IRU WKH PRGHO GHYHORSHG HDUOLHU DQG ZLOO FRPSDUH WKHP WR DFWXDO VRXUFH VWDWLVWLFV IRU UHDO ZRUOG SLFWXUHV 7ZR W\SHV RI TXDQWL]HU ZHUH XVHG D 064( TXDQWL]HU DQG D TXDQWL]HU WKDW ZDV IRXQG E\ SVFKRYLVXDO H[SHULPHQWV >@ 7KH PDWHULDO XVHG ZHUH WZR HLJKWELW SLFWXUHV D ORZ GHWDLO SLFWXUH /(11$ DQG D KLJK GHWDLO SLFWXUH $(5,$/ 0$3 )LJ DQG )LJ f %RWK SLFWXUHV FRQVLVW RI [ SL[HOV $Q RSWLPL]HG HLJKWOHYHO TXDQWL]HU IRU WKH /DSODFLDQ GLVWULEXWLRQ ZDV FKRVHQ 0RGHO SDUDPHWHUV FDQ EH HVWLPDWHG E\ R HQf! f R iQfiQff! 7DEOH FRQWDLQV WKHRUHWLFDO DQG DFWXDO VRXUFH VWDWLVWLFV 7KHRUHWLFDO YDOXHV DUH VHHQ WR EH UHDVRQDEO\ FORVH WR DFWXDO VRXUFH VWDWLVWLFV 7KHQ D VHYHQOHYHO TXDQWL]HU >@ WKDW LV VKRZQ WR ZRUN ZHOO ZLWK GLIIHUHQW SLFWXUHV ZDV VHOHFWHG 7KH KLVWRJUDPV IRU WKH WZR SLFWXUHV ZHUH SUHSDUHG 7KH /DSODFLDQ GLVWULEXWLRQ DQG WKH GLVWULEXWLRQ JLYHQ LQ HT f ZHUH FRPSDUHG ZLWK WKH KLVWRJUDPV )LJXUHV f 7KH\ ERWK VHHP WR EH D IDLU DSSUR[LPDWLRQ WR WKH KLVWRJUDPV 7DEOH JLYHV WKH VWDWLVWLFV JLYHQ E\ HTXDWLRQ HT f /DSODFLDQ SGI DQG WKH SLFWXUH 7KH PRGHO EDVHG VWDWLVWLFV DUH FORVH WR WKH DFWXDO VRXUFH VWDWLVWLFV

PAGE 37

)LJ /HQQD )LJ $HULDO PDS

PAGE 38

'LVFXVVLRQ ,W ZDV VKRZQ WKDW WKH SGI RI LQSXW WR WKH TXDQWL]HU FDQ EH IDLUO\ DSSUR[LPDWHG E\ HTXDWLRQ HT f ZKLFK LV WKH FRQYROXWLRQ RI D JDPPD DQG D XQLIRUP SGIf RU VLPSO\ D /DSODFLDQ GLVWULEXWLRQ 7KH WKHRUHWLFDO VWDWLVWLFV GHULYHG IURP WKH PRGHOV DJUHHG ZHOO ZLWK WKH DFWXDO VRXUFH VWDWLVWLFV ,W LV VHHQ WKDW '3&0 RI YLGHR VLJQDOV GRHV LQ IDFW SURGXFH D QRQV\PPHWULF VRXUFH

PAGE 39

7$%/( $ (,*+7 /(9(/ 064( 48$17,=(5 ,6 86(' 7+( ),567 $1' 7+( 6(&21' &2/8016 6+2: 7+( $&78$/ 6285&( 67$7,67,&6 )25 7+( 3,&785(6 7+( 7+,5' &2/801 6+2:6 7+( 67$7,67,&6 86,1* $ /$3/$&,$1 ',675,%87,21 )25 6285&( /HQQD $HULDO 0DS 0RGHO 6WDWLVWLFV 7$%/( 67$7,67,&6 )25 /(11$ 3,&785( $ 6(9(1 /(9(/ 48$17,=(5 )281' %< 36<&+29,68$/ (;3(5,0(176 ,6 86(' /DSODFH (T f $FWXDO 6WDWLVWLFV

PAGE 40

)LJXUH 7KH KLVWRJUDP IRU /(11$ SLFWXUH LV FRPSDUHG ZLWK Df /DSODFLDQ 'LVWULEXWLRQ Ef 'LVWULEXWLRQ JLYHQ LQ HT f +LVWRJUDP LV VKRZQ ZLWK D EURNHQ OLQH

PAGE 41

Ef )LJXUH 7KH KLVWRJUDP IRU $(5,$/ 0$3 SLFWXUH LV FRPSDUHG ZLWK Df /DSODFLDQ 'LVWULEXWLRQ Ef 'LVWULEXWLRQ JLYHQ LQ HTXDWLRQ f +LVWRJUDP LV VKRZQ ZLWK D EURNHQ OLQH

PAGE 42

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

PAGE 43

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f LQ SUHVHQFH RI DGGLWLYH ZKLWH *DXVVLDQ QRLVH XQGHU DQ DYHUDJH SRZHURU D SHDN SRZHUf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f DQG 0/ PD[LPXP OLNHOLKRRGf UHFHLYHUV 7KHVH PHWKRGV DUH JHQHUDOL]DWLRQV DQG

PAGE 44

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f $+ FRV3 %+ FRVD[ f $+ FRVD &+ FRV3 f %+ FRV3 &+ FRVD f /HW UWf UUf EH WKH UHFHLYHG ZDYHIRUP DQG SWf EH D SDUWLFXODU YDOXH RI UWf 6XSSRVH SWf SSf LV UHFHLYHG LQ WKH V\PERO LQWHUYDO SUUMU_Pcf LV WKH FRQGLWLRQDO MRLQW SUREDELOLW\ GHQVLW\ RI UDQGRP YDULDEOHV GHILQLQJ UWf DQG V V V DUH WKH VLJQDO YHFWRUV GHQRWHG E\ $ % DQG & RQ WKH

PAGE 45

FRQVWHOODWLRQ $ PD[LPXP OLNHOLKRRG UHFHLYHU VHWV WKH PHVVDJH HVWLPDWH WR PN LI IRU L O SU^U¯ S U S? PNf ! S]^U SL [ S? LrN f $VVXPLQJ WKH QRLVH FRPSRQHQWV DUH *DXVVLDQ DQG VWDWLVWLFDOO\ LQGHSHQGHQW 3U3Ln3DO MQf O¯)H[S^A >SVL-fSVLf@` f 7KH GHFLVLRQ UXOH WKHQ EHFRPHV W3LVZfSVAf@ ! a>SV£¯fSVLf@ LrN -9 L9R f 1RWLFH WKDW WKH VXP RI VTXDUHG WHUPV RQ HLWKHU VLGH RI WKH F )LJXUH $ WKUHH VLJQDO FRQVWHOODWLRQ LQHTXDOLW\ DUH WKH VTXDUH RI WKH (XFOLGHDQ GLVWDQFHV EHWZHHQ WKH UHFHLYHG VLJQDO DQG VLJQDOV Vc DQG VN 7KXV WKH GHFLVLRQ UXOH FDQ EH UHZULWWHQ DV

PAGE 46

GSVLf! GSVNf LrN f ZKHUH G[\f LV WKH (XFOLGHDQ GLVWDQFH EHWZHHQ [ DQG \ 3RLQW + LV WKH LQWHUVHFWLRQ RI WKH GHFLVLRQ UHJLRQV DQG LW LV WKHUHIRUH RQ WKH ERXQGDULHV RI WKH GHFLVLRQ UHJLRQV )RU WKH ERXQGDU\ SRLQWV WKH DERYH LQHTXDOLW\ FKDQJHV WR DQ HTXDOLW\ G+ $f G+ %f G+ $f G+ &f f RU $+ %+ &+ f DQG WKLV UHVXOWV LQ DL 3Ln p f§ 3n D 3 f 'HILQH 3Z WR EH WKH SUREDELOLW\ RI ZRUG HUURU 7KH SUREDELOLW\ RI V\PERO HUURU LV RQH PLQXV WKH SUREDELOLW\ WKDW D V\PERO LV FRUUHFW 7KH SUREDELOLW\ RI D FRUUHFW V\PERO LV REWDLQHG E\ DYHUDJLQJ DOO RI WKH FRQGLWLRQDO SUREDELOLWLHV RI D FRUUHFW V\PERO ZH KDYH 3: a
PAGE 47

UHJLRQ ,c )LJXUH f SZ n(3LII3U3Ln3c PLfGS¯GS LO f )LJXUH 2QH GHFLVLRQ UHJLRQ %\ VXEVWLWXWLQJ WKH H[SUHVVLRQV IRU SUREDELOLW\ GHQVLW\ IXQFWLRQ HT f LQWR HT f ZH JHW AH[S>a 3OA-!aALGSf«GS ! r3O3 r RU HTXLYDOHQWO\

PAGE 48

3 3L W]0 , 3L SWDQD@ H[S > 3Df§ 3LWDQ3L 3L$LLf -S_ 1Q f GSGS 3 Q1 m 3O 3M&DQDM H[S> 3Lf§3LWDQ3M 3L$LLf @ GSGS[ 3 Q1 0 3O 3MWDQDM H[S> 33LWDQ3M 3L$+f MA@GSG3O OYR f %\ PDNLQJ WKH IROORZLQJ FKDQJHV RI YDULDEOHV H GS GHVILUR : f X A GS GXYA :! f WKH SUREDELOLW\ RI HUURU LQ WHUPV RI WKH QHZ YDULDEOHV EHFRPHV

PAGE 49

XWDQD 3Y f§A H[S >X Af@ H[S>H@ X 1 R mXWDQ3M GH GX XFDQWWM f§ I H[S > X fµAf@ I H[S >H@ GH GX 2 1 m/S XWDQD f§ I H[S >^X `@ I H[S>H@ 9R 1 Zr GH GX 1RZ E\ GHILQLWLRQ r HUI[[f f§ H[S Wf GW 7KHUHIRUH WKH IROORZLQJ LV WKH HTXDWLRQ IRU 3Z f f 3Z OA I H[S > X Af@HUIX WDQSLX WDQDf GX YAX-! 1 UW ML /=H[S > X Af§f@ HUIX WDQS X WDQDf GX ? M:B LQ r I H[S >X ?³X/ 1 $WI f@HULX WDQ3X WDQDf GX f

PAGE 50

)RU D JLYHQ VLJQDO WR QRLVH UDWLR WKH SUREDELOLW\ RI ZRUG HUURU GHSHQGV RQO\ RQ D DQG D EHFDXVH D D DQG D DGG XS WR Q 7R ILQG WKH YDOXHV RI D D D WKDW PLQLPL]H WKH SUREDELOLW\ RI WKH ZRUG HUURU ZH QHHG WR GLIIHUHQWLDWH 3Z ZLWK UHVSHFW WR D DQG D DQG VHW WKHP HTXDO WR ]HUR 8 XO WDQAf H XnWDQnD XOWDQD[Dff H X-WDQ-DRf? f

PAGE 51

-A¯ eN -7 HXBAU!DXLWDQDfHmmZmDf X GX B } X D:f AIH 9A XOWDQDDfHX:f¯ODff GX 8 7¯ X XOWDQ DDff HX:f¯f¯!XOWDQDf HX:f¯ GDf f ,QVSHFWLRQ VKRZV WKDW DERYH HTXDWLRQV FRQVLVW RI RQO\ RQH W\SH RI LQWHJUDO ZKLFK FDQ EH H[SUHVVHG LQ D FRPSXWDEOH IRUP IX H XNfHXO J OrLf XrNXNf GX f %\ SURGXFLQJ D SHUIHFW VTXDUH LQ WKH H[SRQHQW -f¯ JO-fXNXNf GX H NO rf f¯ rfX Lf GX f DQG XVLQJ WKH UHVXOWV REWDLQHG LQ $SSHQGL[ $ ZH JHW

PAGE 52

-X HaLXaNfHaXLO GX H NO L' Nr H A\g N f -f HUI f§ Y7O,7 rrf f ,I ZH VXEVWLWXWH IRU LQWHJUDOV IURP HT f LQWR HT f DQG HT f ZH ZLOO HQG XS ZLWK WKH IROORZLQJ VHW RI QRQOLQHDU HTXDWLRQV 3LaSf H $+ 1f± $+ ?_ 1W 3L3f H DK r ,FRVDOHUIL _FRVDO_m!f 1 1R 33f H 1D $+ _FRV DDf_HUI f§f§FRV DDf_af 1 1R f VLQD $+ SSfH 1r _ 3LSf H 1r n_FRVD_HUI AA_FRVD_af A ? 9 1 $+ $MI VLQDmf S3fH FRV D[Df HUI ‘ $+ 1 A FRV DDf_rrf f *LYHQ D VLJQDOWRQRLVH UDWLR DQG D SUREDELOLW\ VHW WKH RSWLPXP DQJOHV FDQ EH IRXQG XVLQJ HT f DQG HT f

PAGE 53

:H H[SHULPHQWHG ZLWK D YDULHW\ RI VLJQDOWRQRLVH UDWLRV DQG SUREDELOLW\ VHWV )LJXUH VKRZV WKH RSWLPXP D IRU WKUHH SUREDELOLW\ VHWV RYHU D ZLGH VLJQDOWRQRLVH UDWLR UDQJH ,Q DOO FDVHV RSWLPXP D DSSURDFKHV GHJUHHV DV VLJQDO WR QRLVH UDWLR JRHV XS (YHQ ZKHQ WKH RSWLPXP DQJOH LV VRPHZKDW GLIIHUHQW IURP GHJUHHV IURP WKH SHUIRUPDQFH SRLQW RI YLHZ WKH WZR V\VWHP DUH DOPRVW LQGLVWLQJXLVKDEOH )LJXUH D GHJf YHUVXV DYHUDJH SRZHU G%f Df WKH WRS FXUYHV VRXUFH VWDWLVWLFV ^` Ef 7KH PLGGOH FXUYH VRXUFH VWDWLVWLFV ^OO` DQG Ff WKH ERWWRP FXUYH VRXUFH VWDWLVWLFV ^` 7KH SUHVHQWHG UHVXOWV VXJJHVW WKDW WKH WKUHH VLJQDOV VKRXOG EH SODFHG RQ WKH YHUWLFHV RI DQ HTXLODWHUDO WULDQJOH )LJXUH f 7KH RULJLQ RI WKH VLJQDO FRQVWHOODWLRQ FDQ EH VKLIWHG WR D QHZ ORFDWLRQ WKDW PLQLPL]HV WKH DYHUDJH HQHUJ\ ZLWKRXW DIIHFWLQJ WKH SUREDELOLW\ RI HUURU /HW XV GHILQH WR EH $% %& $& LQ WULDQJOH $%& 6LQFH $%& LV DQ HTXLODWHUDO WULDQJOH $+ DQG

PAGE 54

DUH UHODWHG E\ O $+V> f 7KH SRLQW *[\f FDQ EH IRXQG DV IROORZV SDYH 3L , 8_f\@ S A_f\@ S >[\Af@ f 3DYH %[ 3L3Df f G3rYH %\ Y(LK+ \ f 6XEVWLWXWLQJ [ DQG \ LQWR f DQG VROYLQJ IRU ZH JHW -)3c f ZKHUH N LV JLYHQ E\ f 3L L 3L3Of S@ 3 > SSOfS@ S > 3Sf SOf@ f

PAGE 55

)LJXUH 7KH WKUHH VLJQDOV IRUP D HTXLODWHUDO WULDQJOH $ 1XPHULFDO $SSURDFK %DVHG RQ /DJUDQJH 0XOWLSOLHUV 0HWKRG ,Q WKLV DSSURDFK WKH SUREOHP RI VLJQDO VHOHFWLRQ LV YLHZHG DV D FRQVWUDLQHG RSWLPL]DWLRQ SUREOHP 7KH FRQVWUDLQWV DUH LQFRUSRUDWHG LQWR WKH RSWLPL]DWLRQ SUREOHP E\ WKH XVH RI /DJUDQJH PXOWLSOLHUV PHWKRG /HW XV RXWOLQH WKH GHVLJQ RI D 0$3 UHFHLYHU ZLWK WKUHH VLJQDOV ZLWK XQHTXDO SUREDELOLWLHV )LJXUH f $ 0$3 UHFHLYHU GHFLVLRQ UXOH LV P PN LI IRU L O 3L UL 3Ln U 3 PNfSN! 3UUL 3LU S_ PLfSL LrN f 8SRQ VXEVWLWXWLQJ f LQWR f ZH JHW

PAGE 56

OQSNf SMVNMf! ,Q3Lf-/e 3MVAf LrRL; 2f $VVXPH SRLQW + LV WKH LQWHUVHFWLRQ RI GHFLVLRQ UHJLRQV )RU SRLQW + WKH LQHTXDOLW\ WXUQV LQWR DQ HTXDOLW\ DQG WKH VXPV RQ HLWKHU VLGH ZLOO EH WKH GLVWDQFHV EHWZHHQ + DQG VLJQDOV Vc DQG 6M WKHUHIRUH $+%+ 1 OQ f§f 3 6LPLODUO\ ZH REWDLQ f $+a&+ 1 LQ f§f 3 f ,Q DGGLWLRQ WR WKH WZR DERYH 0$3 FRQVWUDLQWV WKHUH DUH IRXU JHRPHWULF FRQVWUDLQWV HTf WKURXJK HTff DQG DQ DGGLWLRQDO FRQVWUDLQW WR PLQLPL]H WKH DYHUDJH HQHUJ\ 7KH RULJLQ RI WKH FRQVWHOODWLRQ LV VKLIWHG WR D ORFDWLRQ WKDW PLQLPL]HV WKH DYHUDJH SRZHU 7KHQ WKH DYHUDJH SRZHU IXQFWLRQ HYDOXDWHG DW WKDW SRLQW LV VHW WR 1 SDSESF 3DYH f ZKHUH

PAGE 57

D [$+ VLQ3f \ E [%+ VLQDOf \ F [&+ VLQ 3LD3f f \$+ FRV3 &+ FRV 3LD3f f [ S$+ VLQS%+ VLQD S&¯ VLQ 3[RF3f \ S$+ FRV3S &+ FRV 3D3f 1RZ WKH SUREOHP LV WR PLQLPL]H D IXQFWLRQ RI QLQH YDULDEOHV XQGHU VHYHQ HTXDOLW\ FRQVWUDLQWV 7KH PHWKRG RI /DJUDQJH PXOWLSOLHUV FDQ EH DSSOLHG WR WKLV SUREOHP :H QHHG WR GHILQH D QHZ FRVW IXQFWLRQ ) RI VL[WHHQ YDULDEOHV DQG WDNH WKH SDUWLDO GHULYDWLYHV ZLWK UHVSHFW WR WKH VL[WHHQ YDULDEOHV DQG VHW WKHP HTXDO WR ]HUR FRVW IXQFWLRQ )DOW D D S[ 3 3 $+ %+ &+ ; ; ; ; ; $ ,I ZH HOLPLQDWH ;V DPRQJ WKH VL[WHHQ HTXDWLRQV ZH HQG XS ZLWK D QRQOLQHDU VHW RI HTXDWLRQV RI WKH RWKHU QLQH YDULDEOHV ZKLFK PXVW EH VROYHG QXPHULFDOO\ 7KLV PHWKRG EHFRPHV TXLWH FRPSOLFDWHG DV WKH QXPEHU RI WKH VLJQDOV LQ WKH FRQVWHOODWLRQ JURZV ,W ZDV PDLQO\ XVHG WR YHULI\ WKH VROXWLRQV REWDLQHG E\ WKH JUDGLHQW EDVHG PHWKRG ZKHQ 1 ZDV UHODWLYHO\ VPDOO

PAGE 58

0LQLPXP (UURU 6LJQDO 6HOHFWLRQ 3UHOLPLQDULHV /HW XV GHQRWH WKH VLJQDOV LQ WKH FRQVWHOODWLRQ E\ VV V1 ZKHUH Vc [c\cf 7KH DYHUDJH SRZHU LV JLYHQ E\ Q f Q :KHUH ^S S S1` LV WKH SUREDELOLW\ GLVWULEXWLRQ RI WKH VLJQDO VHW DQG __VQ__ LV WKH PDJQLWXGH RI VQ )ROORZLQJ WKH FRQYHQWLRQ RI @ ZH FKRRVH WR VHW 3YH 1 DQG GHILQH WKH VLJQDOWRQRLVH UDWLR WR EH 615 ORJ f 1D ZKHUH 1 LV WKH RQH VLGHG SRZHU VSHFWUDO GHQVLW\ RI ZKLWH QRLVH %\ GHILQLWLRQ WKH HUURU UDWH SUREDELOLW\ RI V\PERO HUURUf LV n$ LY f Lf¬H ( 3Q 3U HUURU _VQf 7KH XQLRQ ERXQG JLYHV WKH XSSHU ERXQG IRU WKH FRQGLWLRQDO SUREDELOLW\ RI HUURU )RU D PD[LPXP OLNHOLKRRG UHFHLYHU ZH KDYH

PAGE 59

f )RU ODUJH YDOXHV RI VLJQDOWRQRLVH UDWLR WKH FRQGLWLRQDO SUREDELOLW\ RI HUURU ZLOO EH HJXDO WR WKH XSSHU ERXQG %\ XVLQJ DSSUR[LPDWLRQ IRU 4 IXQFWLRQ ZKHQ WKH DUJXPHQW LV ODUJH DQG SOXJJLQJ EDFN LQ f ZH JHW 3H 1 1 QL LrQ H[S >fµ O6L6Q+ 1} ‘ f 6LPLODUO\ LW FDQ EH VKRZQ WKDW WKH V\PERO HUURU UDWH IRU D 0$3 UHFHLYHU LV JLYHQ E\ 1U 1 3Ha??f§OA9Q = n f¯ QL Or ;!m e WHLVM OVUVDU1OQOMcLf ‘H[S > f§ OV f¯VQOO 1DOQ aff 3[ :'VLVQ_ f *UDGLHQW 6HDUFK $OJRULWKPV IRU DQ $YHUDJH 3RZHU &RQVWUDLQW ,Q VHDUFK RI WKH PLQLPXP WKH JUDGLHQW RI WKH SUREDELOLW\ RI HUURU LV REWDLQHG DQDO\WLFDOO\ DQG DQ LWHUDWLYH JUDGLHQW VHDUFK DOJRULWKP ZKLFK PRGLILHV WKH FRQVWHOODWLRQ DW HDFK LWHUDWLRQ LV XVHG WR ILQG WKH RSWLPXP FRQVWHOODWLRQ ,Q WKH JUDGLHQW VHDUFK DOJRULWKP WKH LWHUDWLYH UXOH LV JLYHQ E\

PAGE 60

f ZKHUH 6L Nf V^Nf f V1Nf GHQRWHV WKH VLJQDO YHFWRU DW WKH NWK VWHS RI WKH DOJRULWKP DN LV WKH VWHS VL]H DQG 9) LV WKH JUDGLHQW RI 3F 6LQFH WKH VLJQDO SRZHU PD\ FKDQJH ZLWK N WKH VLJQDO YHFWRU LV QRUPDOL]HG DW HDFK VWHS RI WKH DOJRULWKP YLSMVL Nf S?V Nf S1?VNOf _f f 7R VSHHG XS WKH FRQYHUJHQFH LQVWHDG RI D FRQYHQWLRQDO JUDGLHQW VHDUFK DOJRULWKP WKH )OHWFKHU DQG 5HHYHV FRQMXJDWH JUDGLHQW PHWKRG >@ FDQ EH XWLOL]HG ,Q WKLV PHWKRG WKH LQIRUPDWLRQ DERXW WKH VHFRQG GHULYDWLYH LV XVHG LQGLUHFWO\ 7KH DOJRULWKP LV GHVFULEHG E\ f ZKHUH KIU f§Y)IUWW MAeBL f DQG

PAGE 61

>Y)A@ Y)N >9)AF Y)r f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

PAGE 62

! 9) "1 ZKHUH N NnNf f JN LV REWDLQHG E\ WDNLQJ WKH GHULYDWLYH RI 3H ZLWK UHVSHFW WR VN 3MF3f 6Na6L UH[S >_Vr6LO LrN :Q ‘@ >‘ AL OAVM 1 f 6LPLODUO\ RQH FDQ ILQG WKH NWK FRPSRQHQW RI 9) IRU D 0$3 UHFHLYHU LrN 3r=VrrLH[3> W¯ WD W ‘@ > 7f§@ t1?VNV$ 1MVN6L L U -9_V-WVL_D W_ L?\_V-WV3L96LH[3Wn f ZKHUH

PAGE 63

, 6N6O VNVL ,VMW6M+ DQG W LVf±VI??1OQOef 3L W OV-WVLOZROQILf fµ 6WDUWLQJ 3RLQWV 7KH DYHUDJH VLJQDO SRZHU LV HTXDO WR 1 RU (SNL+ Q fµLY 7KXV LQGLYLGXDO VLJQDOV PXVW VDWLVI\ 6ROYLQJ WKH LQHTXDOLW\ IRU VQ ZH JHW ? SQ1 Q A SQ1 f f f f f f ZKLFK VWDWHV WKDW VWDUWLQJ SRLQWV PXVW EH VHOHFWHG IURP LQVLGH

PAGE 64

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f OHW 0 PD[ ^-V Nf O` f VLL1 :H ZLOO PRGLI\ WKH VLJQDO VHW DJDLQ LQ WKH IROORZLQJ PDQQHU

PAGE 65

6L Nf VI“a0 6] Nf 6NrLa \I:0 VNOf \775 WR PHHW WKH SHDN SRZHU FRQVWUDLQW f 0LQLPXP $YHUDJH &RVW 6LJQDO 6HOHFWLRQ 7KH FRPPXQLFDWLRQ FKDQQHOV DUH DOPRVW QHYHU HUURU IUHH 7KH SUREDELOLW\ RI HUURU HUURU UDWHf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

PAGE 66

$YHUDJH &RVW )XQFWLRQ 7KH %D\HV UHFHLYHU ZKLFK PLQLPL]HV WKH DYHUDJH FRVW LV JLYHQ E\ A ((SVLO VLf3LKLn f ML ML ZKHUH /Mc LV FRVW LQYROYHG ZKHQ WKH UHFHLYHU SLFNV 6M ZKHQ Vc ZDV DFWXDOO\ VHQW 7KH ERXQGDULHV RI WKH GHFLVLRQ UHJLRQV DUH QR ORQJHU VWUDLJKW OLQHV LQVWHDG WKH\ DUH WZR GLPHQVLRQDO FXUYHV )RU LQVWDQFH WKH ERXQGDU\ EHWZHHQ Vc DQG V LV JLYHQ E\ e/r AA_VMFf3U(b f¬‘UOVMF-Sr f NO N ZKHUH FRQGLWLRQDO SUREDELOLW\ RI HUURU LV 3UMVNf f§H[S6 f f Q1 1 &KDQQHO QRLVH FRXOG FDXVH WKH UHFHLYHG VLJQDO WR PRYH WR DQ DGMDFHQW GHFLVLRQ UHJLRQ DQG UHVXOW LQ DQ HUURU )RU ODUJH YDOXHV RI VLJQDOWRQRLVH UDWLR KRZHYHU HUURUV DUH DOPRVW HQWLUHO\ WKH UHVXOW RI GLVSODFHPHQW RI D UHFHLYHG VLJQDO WR DQ DGMDFHQW GHFLVLRQ UHJLRQ IURP WKH QHDUHVW SRLQW RQ D ERXQGDU\ WR WKH WUDQVPLWWHG VLJQDO /HW Vc EH WKH WUDQVPLWWHG VLJQDO DQG 6Mc EH WKH SRLQW ZLWK VKRUWHVW GLVWDQFH RQ WKH ERXQGDU\

PAGE 67

ZLWK 6M S VM _ 6Mf FDQ EH DSSUR[LPDWHG ZLWK D 4 IXQFWLRQ 36 _Vf &!Oe/-]Lf f VXEVWLWXWLQJ EDFN LQWR WKH DYHUDJH FRVW H[SUHVVLRQ ZH JHW & f§$ 1 Q n(n(KL S Bf± H[S>%6L 6MLc@ f -L ,6Y67KH 1HDUHVW 3RLQWV RQ WKH %RXQGDULHV 7KH ILUVW VWHS LQ FRPSXWLQJ WKH DYHUDJH FRVW IXQFWLRQ LV WR ILQG WKH VHW RI QHDUHVW SRLQW WR HDFK DQG HYHU\ VLJQDO SRLQW LQ WKH FRQVWHOODWLRQ RQ WKH ERXQGDULHV 7KHUH DUH 1 VLJQDO SRLQWV LQ WKH FRQVWHOODWLRQ DQG IRU HDFK RQH WKHUH DUH 1Of RI QHDUHVW SRLQWV RQ 1Of ERXQGDULHV WKHUHIRUH D WRWDO RI 11Of RI QHDUHVW SRLQWV PXVW EH IRXQG 7KH SUREOHP RI ILQGLQJ 6Mc ZKLFK LV WKH QHDUHVW SRLQW WR Vc RQ WKH ERXQGDU\ EHWZHHQ Vc DQG 6M FDQ EH IRUPXODWHG OLNH D /DJUDQJH PXOWLSOLHU SUREOHP :H ZRXOG OLNH WR PLQLPL]H __VVc__ VXEMHFW WR D FRQVWUDLQW HT ff /HW XV IRUP WKH DX[LOLDU\ IXQFWLRQ Wf __VVL__$ /LN S^L?VNfSNa
PAGE 68

Wf G[ W! %\ žt f (OLPLQDWLQJ ; ZH ZLOO FRPH XS ZLWK WKH IROORZLQJ VHW RI HTXDWLRQV eU U Z [ Y e /LNa/MNf [a[Nf H[S f§ -FO LYR f 1 , Ff§ , ZKHUH Vc [c\cf DQG V [\f VML 7KH *UDGLHQW RI WKH &RVW )XQFWLRQ 7R HYDOXDWH JN ILUVW WKH SRUWLRQ RI FRVW IXQFWLRQ WKDW GHSHQGV RQ VN PXVW EH IRUPHG A U U ! /M\ 3\7 "H[S > ??VNVMNL r 1f± -O 1 ;A ; U ,,6Y6MK,, (/rL 3L 7f§ LL H[S > f§ f±@ -; 1f± JNr DQG JN WKH FRPSRQHQWV RI JN DUH IRXQG E\ WDNLQJ WKH GHULYDWLYH RI DERYH H[SUHVVLRQ ZLWK UHVSHFW WR WKH FRPSRQHQWV

PAGE 69

RI VY \L 1 , VNaVL 6!rL3L V H[S>  rNL ,n OOVMVr@ > $@ c6L6AOO 3 ZKHUH f DQG f¯L rN;MWf f°A¯f \r\-r! nAf n f mDf¯ALAMXf \M\Xf Af G[W f Q Wm3rLUELH[3WeeUAO ,Wf§  f¯6r >VAVAOO L -O 1 a( 9 /-FL 3L __6LB6MFL__ H;S>an 1 ,6L6MWLOnM > OO 1f± L6Lf¯6LWLOL 1r f O [rr-OF! AWf\N\MNf Af f ZKHUH

PAGE 70

f 7KH JUDGLHQW RI WKH FRVW IXQFWLRQ GHSHQGV QRW RQO\ RQ WKH QHDUHVW SRLQWV RQ WKH ERXQGDU\ EXW DOVR RQ WKHLU GHULYDWLYHV /HW XV ILQG WKH GHULYDWLYH RI WKH FRPSRQHQWV RI 6Mc ZLWK UHVSHFW WR WKH FRPSRQHQWV RI Vc 'LIIHUHQWLDWLQJ HT f DQG f ZLWK UHVSHFW WR [c DQG \c ZH JHW f ZKHUH f f ZKHUH Z [[ f

PAGE 71

ABf( L/LN/MN! \\rf H[S > 2-&L NO = _JJrO ‘fSN [[LfnA /LNa/MNf aJaaa \a\rf LWf H[3 Wa % rO R[L L9R R Af( ^/LNa/MNf [[NfH[S>6 ANc @SN R[L NO Za @SM -U D[L R _JJr_ L @Sr f ZKHUH f 9 A _ RBR _ f( A/LNaKNf \r\MW! H[S > @ S [[Lf ( \\rf Af H[S > %VA6MF @ 3N Aaf( /LNa/MNf [[NfH[S>Ae$@SN \9Lf ( f¯U rar-F! ! H[S > 6rN!, @ Sr MFL r\M R f ZKHUH ]8[Wf^_f\\c>fAOf ,W FDQ EH VKRZQ WKDW UHIHU WR $SSHQGL[ %f f

PAGE 72

DQG G[ G[  G[ %\  f SR SR G[ G\ f (TXDWLRQV f DQG f LPSO\ WKDW Y X &RQWUDU\ WR WKH HTXDWLRQV IRU ILQGLQJ 6Mc WKH HTXDWLRQV IRU ILQGLQJ LWV GHULYDWLYHV DUH OLQHDU 7KH ,WHUDWLYH $OJRULWKP 7R ILQG WKH VLJQDO FRQVWHOODWLRQ ZH ZLOO VWDUW IURP DQ LQLWLDO JXHVV $Q LWHUDWLYH UXOH HLWKHU JUDGLHQW VHDUFK RU FRQMXJDWH JUDGLHQW VHDUFK DOJRULWKPf LV VHOHFWHG $Q LPSURYHG VLJQDO VHW LV IRXQG XVLQJ WKH LWHUDWLYH UXOH 'HSHQGLQJ RQ WKH W\SH RI FRQVWUDLQW SHDN RU DYHUDJH SRZHU FRQVWUDLQWf WKH FRQVWHOODWLRQ LV PRGLILHG DFFRUGLQJO\ HTf RU HTff 7KH LWHUDWLRQ LV FRQWLQXHG XQWLO D PLQLPD LV UHDFKHG 7KH DERYH SURFHGXUH LV YDOLG XQGHU WKH DVVXPSWLRQ RI ODUJH FKDQQHO 615 :KHQ FKDQQHO 615 LV QRW ODUJH REWDLQLQJ DQ DQDO\WLFDO H[SUHVVLRQ IRU WKH JUDGLHQW LV QRW IHDVLEOH EHFDXVH RI WKH FRPSOH[LW\ RI WKH VKDSH RI WKH GHFLVLRQ UHJLRQV $Q REYLRXV VROXWLRQ LV DQ H[KDXVWLYH VHDUFK ZKLFK LV

PAGE 73

FRPSXWDWLRQDOO\ YHU\ H[SHQVLYH DQG ORQJ +RZHYHU RQH FDQ XVH D PRGLILHG IRUP RI WKH DERYH SURFHGXUH WR VDYH RQ WKH FRPSXWDWLRQDO H[SHQVHV ,Q WKH PRGLILHG SURFHGXUH WKH DYHUDJH FRVW LV FRPSXWHG EDVHG RQ WKH UHFHLYHG GDWD DQG WKH JUDGLHQW EDVHG XSGDWH UXOH LV UHSODFHG ZLWK WKH GLUHFWLRQDO VHW 3RZHOOnVf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

PAGE 74

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

PAGE 75

7$%/( 9$/8(6 2) 6,*1$/7212,6( 5$7,2 $1' 7+( 352%$%,/,7< ',675,%87,216 2) 1216<00(75,& 6285&( 6RXUFH 1XPEHU 615G%f 3UREDELOLW\ 'LVWULEXWLRQ 7$%/( 6285&( 67$7,67,&6 )25 7+( )285 6,*1$/ 6285&( VRXUFH VWDWLVWLFV

PAGE 76

O6 L ; ; Df Ef )LJXUH %HVW VLJQDO FRQVWHOODWLRQV IRU QRQV\PPHWULF VRXUFHV 3UREDELOLW\ GLVWULEXWLRQV DUH JLYHQ LQ 7DEOH Df 6RXUFH OEf 6RXUFH Ff 6RXUFH Gf VRXUFH

PAGE 77

; ; ; ; = &f ; ; ; ; ; ; ; Gf )LJXUH f§FRQWLQXHG

PAGE 78

HV Df Ef )LJXUH Df 6HYHQ VLJQDO FRQVWHOODWLRQ IRU DQ HTXDOO\ OLNHO\ VRXUFH nV UHSUHVHQW WKH VWDUW SRLQWV DQG [nV UHSUHVHQW WKH ILQDO FRQVWHOODWLRQ Ef 7KH VROLG DQG GDVKHG FXUYHV VKRZ WKH HUURU UDWH DV D IXQFWLRQ RI QXPEHU RI LWHUDWLRQV IRU JUDGLHQW VHDUFK DQG FRQMXJDWH JUDGLHQW PHWKRG UHVSHFWLYHO\

PAGE 79

GHVLJQ UHODWLYH WR D HTXDOO\ OLNHO\ VLJQDO VHOHFWLRQ D ILUVW RUGHU *DXVV0DUNRY VRXUFH ZLWK D FRUUHODWLRQ FRHIILFLHQW RI ZDV V\QWKHVL]HG 7KH RXWSXW ZDV HQFRGHG XVLQJ D ILUVW RUGHU '3&0 HQFRGHU $ IRXU VLJQDO FRQVWHOODWLRQ ZDV GHVLJQHG ZLWK WKH VRXUFH VWDWLVWLFV VKRZQ LQ 7DEOH 7KH SHUIRUPDQFH RI WKH QRQV\PPHWULF FRQVWHOODWLRQ ZDV FRPSDUHG ZLWK WKDW RI WKH ZHOO NQRZQ HTXDOO\ OLNHO\ FRQVWHOODWLRQ IRU IRXU HTXDOO\ OLNHO\ VLJQDOV WKH EHVW FRQVWHOODWLRQ LV IRUPHG E\ WKH YHUWLFHV RI D VTXDUH >@f )LJXUH GHPRQVWUDWHV D FRPSDULVRQ EHWZHHQ WKH V\VWHPV LQ WHUPV RI RXWSXW VLJQDOWR QRLVH UDWLR RXWSXW 615f )RU ODUJH YDOXHV RI FKDQQHO VLJQDO WRQRLVH UDWLR FKDQQHO 615f WKH WZR GHVLJQ SURFHGXUHV UHVXOW LQ DQ LGHQWLFDO SHUIRUPDQFH $V WKH FKDQQHO 615 GHFUHDVHV WKH FXUYHV UHSUHVHQWLQJ WKH SHUIRUPDQFH RI QRQV\PPHWULF DQG HTXDOO\ OLNHO\ VLJQDO GHVLJQ VHSDUDWH DQG WKH GLIIHUHQFH EHWZHHQ WKH WZR V\VWHPV JHWV ODUJHU 7KH QRQV\PPHWULF VLJQDO GHVLJQ LV G% LQ WHUPV RI RXWSXW 615f VXSHULRU WR HTXDOO\ OLNHO\ VLJQDO GHVLJQ IRU WKH QRLVLHVW FKDQQHO FRQVLGHUHG D FKDQQHO 615 RI G%f ,W LV VHHQ WKDW VLJQLILFDQW LPSURYHPHQW LQ SHUIRUPDQFH FDQ EH REWDLQHG IRU QRLV\ FKDQQHOV E\ XWLOL]LQJ QRQV\PPHWULF VLJQDO GHVLJQ 7KH DPRXQW RI LPSURYHPHQW LV D IXQFWLRQ RI VRXUFH VWDWLVWLFV $ FRPSDULVRQ EHWZHHQ WKH HTXDOO\ OLNHO\ VLJQDOV DQG XQHTXDO VLJQDO SUREDELOLWLHV VKRZV WKDW LI WKH VLJQDO VWDWLVWLFV DUH QRW YHU\ GLIIHUHQW WKH VKDSH RI FRQVWHOODWLRQ LV QRW DSSUHFLDEO\ GLIIHUHQW IURP WKH HTXDOO\ OLNHO\ FDVH %XW

PAGE 80

LI WKH VLJQDO SUREDELOLWLHV DUH YHU\ GLIIHUHQW IURP WKH HTXDOO\ OLNHO\ FDVH WKH VKDSH RI WKH FRQVWHOODWLRQ FRXOG EH YHU\ GLIIHUHQW IURP WKH HTXDOO\ OLNHO\ FRQVWHOODWLRQ )LJXUH Ef DQG )LJXUH Ff VKRZ FRQVWHOODWLRQV IRU WZR GLIIHUHQW ILYH VLJQDO VRXUFHV 7KH FRQVWHOODWLRQV IRU WKH H[WUHPHO\ QRQ V\PPHWULF VRXUFH LV FRPSOHWHO\ GLIIHUHQW IURP WKH RWKHU RQH *HQHUDOO\ VSHDNLQJ WKH JHRPHWU\ RI WKH FRQVWHOODWLRQV GHSHQGV XSRQ WKH SRZHU FRQVWUDLQW )RU H[DPSOH WKH RSWLPXP VLJQDO FRQVWHOODWLRQV IRU ILYH HTXDOO\ OLNHO\ VLJQDOV VXEMHFW WR DYHUDJH SRZHU DQG SHDN SRZHU FRQVWUDLQWV DUH GLVSOD\HG LQ )LJXUH Df DQG )LJXUH Ef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

PAGE 81

HLJKWOHYHO 064( PHDQVTXDUHG TXDQWL]DWLRQ HUURUf RSWLPL]HG TXDQWL]HU ZDV XVHG DQG WKH RXWSXW UHFRQVWUXFWLRQf VLJQDOWR QRLVH UDWLR RI WKH WZR V\VWHPV ZHUH HYDOXDWHG RYHU D ZLGH UDQJH RI FKDQQHO 615 7KH UHVXOWV JUDSKHG LQ )LJXUH LQGLFDWH WKDW IRU ODUJH YDOXHV RI FKDQQHO VLJQDOWRQRLVH UDWLR WKH WZR V\VWHPV DUH DOPRVW LGHQWLFDO +RZHYHU IRU VPDOOHU YDOXHV RI VLJQDO WRQRLVH UDWLR WKH PLQLPXP FRVW GHVLJQ LV VXSHULRU WR WKH 0$3 V\VWHP ,Q WKLV FDVH WKH LPSURYHPHQW LQ WKH RXWSXW VLJQDOWR QRLVH UDWLR LV DURXQG G% IRU WKH LQWHUPHGLDWH YDOXHV RI FKDQQHO VLJQDOWRQRLVH UDWLR DQG G% IRU ORZ YDOXHV RI FKDQQHO 615

PAGE 82

Df ‘ Ef )LJXUH 7KH EHVW VLJQDO FRQVWHOODWLRQV IRU ILYH HTXDOO\ OLNHO\ VLJQDOV VXEMHFW WR DYHUDJH SRZHU Df DQG SHDN SRZHU Ef FRQVWUDLQWV

PAGE 83

FKDQU[O 615 )LJXUH 3HUIRUPDQFH UHVXOWV IRU QRQV\PPHWULF VLJQDO VHOHFWLRQ WKH VROLG FXUYHf DQG WKH HTXDOO\ OLNHO\ VLJQDO GHVLJQ WKH GDVKHG FXUYHf RYHU D ZLGH UDQJH RI FKDQQHO VLJQDO WRQRLVH UDWLR

PAGE 84

I [[[[[[L[[[[L[[NNLN[[LK[[NL[[[[[[[N[[L[[ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; B ;;;;;;;;;;;;;;;;;;.;;;.;;;;;;;..;;;;;;; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;r [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[ ;;;;;;;;;;;.;;.;;;;;;;;;+.;;;;;;;;;;;;. [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[ [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[ 0.;;;;;;;;;;;;;.;;;;;;;;;;;;;;;;;;;;;;} [[[[[[[[[[[[[[[[[[[D[[[[[[[[[[[[[[[[[[[ ;;;;;;;;;;;;;;;;;;I [[[[[[[[[[N[ ;;;;;r 0 00 [[[[[[[[[[[[[[[[[[[ ‘;;;;;;;;;;;;;; +[[[Nr[[[ m+,0 ‘W XW;0 fµRRRRRRRRRRRRRRRWRR 2222222'22222222222 , L Df D [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[ fµm;;;,;;;;;;,;;;;;;;;;;;;;;;;;;.;;;;;;,+ [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[ [[[[[[[[[[[[[[[[[[[[[[[[[[[[[L[L[[[[[[[ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; [[[[[[[[[[[L[[[L[[[[[[L[[[L[[;[[[L[[[[[ [[[[[[[[[[[[[[[[[[[ ;;;;; VDPZPPWP PP QXWD¯rr ‘ [[[[[[[[[[[[[[[[[[[ ;;;;, Ef )LJXUH 7KH GHFLVLRQ UHJLRQV IRU PLQLPXP FRVW VLJQDO GHVLJQ Df FKDQQHO 615 G% DQG Ef FKDQQHO 615 G% 7KH VLJQDOV DUH UHSUHVHQWHG E\ rnV DQG WKH GHFLVLRQ UHJLRQV DUH PDUNHG E\ GLVWLQFW V\PEROV

PAGE 85

FKRQQWO 615 )LJXUH &RPSDULVRQ EHWZHHQ PLQLPXP FRVW VLJQDO VHOHFWLRQ WKH VROLG FXUYHf DQG PLQLPXP HUURU VLJQDO GHVLJQWKH GDVKHG FXUYHf IRU D ILUVW RUGHU *DXVV0DUNRY VRXUFH ZLWK FRUUHODWLRQ FRHIILFLHQW RI

PAGE 86

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

PAGE 87

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f IRU YLGHR VLJQDOV XQGHU GLIIHUHQW EDQGZLGWK FRQVWUDLQWV 3UHOLPLQDULHV 'HVFULSWLRQ RI WKH &RPPXQLFDWLRQ 6\VWHPV 7KH EORFN GLDJUDPV RI WKH V\VWHP XQGHU FRQVLGHUDWLRQ DUH VKRZQ LQ )LJXUHV +XIIPDQ RSWLPDO VRXUFH FRGLQJ LV XVHG WR HQFRGH WKH VRXUFH VLJQDOV

PAGE 88

7KH %&+ FRGHV DUH FKRVHQ IRU FKDQQHO FRGLQJ ,I H[WUD EDQGZLGWK LV DYDLODEOHf EHFDXVH FRGHV FORVH WR DQ\ UDWH FDQ EH IRXQG $ TDU\ %&+ FRGH LV GHQRWHG E\ QNTWTf ZKHUH NT UHSUHVHQWV WKH QXPEHU RI LQIRUPDWLRQ V\PEROV Q UHSUHVHQWV WKH EROFN VL]H DQG WT UHSUHVHQWV WKH WKH HUURU FRUUHFWLRQ FDSDELOLW\ RI WKH FRGH 7KH UHFHLYHUV DUH FRKHUHQW LQSKDVHTXDGUDWXUH GHWHFWRUV H[FHSW IRU %36. ZKHUH RQO\ WKH LnQSKDVH EUDQFK LV QHHGHGf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c LV WUDQVPLWWHG WKHUH LV D SUREDELOLW\ RI SMc WKDW 6M LV UHFHLYHG LSAMf DQG D SUREDELOLW\ RI SNL WKDW VN LV UHFHLYHG L"rNf DQG VR RQ DQG VR IRUWK 7KHUHIRUH WKH SUREDELOLW\ WKDW Vc LV UHFHLYHG FRUUHFWO\ LV

PAGE 89

3LL A -\ 3QL Q L QrL f 1RWLFH WKDW 3Mc GHSHQGV RQ WKH PRGXODWLRQ VFKHPH +HUH D WZR GLPHQVLRQDO VLJQDO FRQVWHOODWLRQ LV XWLOL]HG 7KH FKDQQHO PDWUL[ LV 3 W3LM@ f 1 )LJXUH 7KH 0RGHO RI WKH FRPPXQLFDWLRQ FKDQQHO ZKHUH 3Mc LV JLYHQ E\ 3D fµH[S >f§f§@ Q L6M6MO 1Q f ,W LV DOVR REYLRXV WKDW 3ML 3LM ,]L]T f

PAGE 90

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f +HUH QRQELQDU\ %&+ FRGHV DUH XVHG IRU HUURU FRUUHFWLRQ )RU D QRQELQDU\ V\VWHP RI DOSKDEHW VL]H T D TDU\ QNTWTf FRGH PXVW EH IRXQG WR VDWLVI\ WKH HTXDO LQIRUPDWLRQ UDWH Q B $E f ZKHUH KE DQG KT DUH WKH DYHUDJH PHVVDJH OHQJWK IRU ELQDU\ DQG TDU\ V\VWHPV UHVSHFWLYHO\

PAGE 91

Df Ef )LJXUH 7KH WUDQVPLWWHUV IRU V\VWHPV HPSOR\LQJ Df ELQDU\ Ef QRQELQDU\ DOSKDEHWV ,Q YDULDEOH OHQJWK FRGLQJ VXFK DV +XIIPDQ FRGLQJf D V\PERO GHFRGLQJ HUURU SURSDJDWHV LQ WKH EORFN 7KHUHIRUH LW LV DSSURSULDWH WR XVH ZRUG HUURU SUREDELOLW\ DV WKH ILJXUH RI PHULW )RU WKH ELQDU\ V\VWHP 3Z LV JLYHQ E\ 3Z OOSHfQ f ZKHUH SF LV WKH FURVVRYHU SUREDELOLW\ )RU D %36. V\VWHP SF LV NQRZQ WR EH HTXDO WR SH 2:Hf f ZKHUH H LV WKH VLJQDOWRQRLVH UDWLR ,, :KHQ WKH EDQGZLGWK FRQVWUDLQW LV QRW YHU\ VWULFW DQG FDQ EH VRPHZKDW UHOD[HG /HW XV DOORZ VRPH EDQGZLGWK H[SDQVLRQ IRU WKH ELQDU\ V\VWHP )LJXUH f $V D UHVXOW WKH QRQELQDU\

PAGE 92

V\VWHPV HPSOR\LQJ KLJKHU RUGHU DOSKDEHWV ZLOO HQMR\ HYHQ JUHDWHU EDQGZLGWK H[SDQVLRQ )LJXUH 7UDQVPLWWHU IRU WKH ELQDU\ V\VWHP ZKHQ WKH EDQGZLGWK FDQ EH H[SDQGHG /HW WKH EDQGZLGWK H[SDQVLRQ IRU WKH ELQDU\ V\VWHP EH SHUFHQW 7R VDWLVI\ WKH HTXDO LQIRUPDWLRQ UDWH FRQVWUDLQW ELQDU\ %&+ QNWf FRGH DQG TDU\ %&+ QTNTWTf FRGH PXVW EH IRXQG VXFK WKDW _ 3 f 7KH ELQDU\ V\VWHP PDNHV QR HUURU XQOHVV W RU PRUH RI WKH Q WRWDO ELWV LQ D ZRUG DUH LQ HUURU 7KH SUREDELOLW\ RI W RU PRUH V\PEROV HUURU LV JLYHQ E\ ( _SHO3HfQf LWL ? f ZKHUH SH LV WKH FURVVRYHU SUREDELOLW\ 8VXDOO\ 3Z FDQ EH

PAGE 93

DSSUR[LPDWHG E\ 3Za Wf Q , &r QWf SHO f /HW WKH ZRUG HUURU SUREDELOLW\ IRU WKH V\VWHP XVLQJ WKH TDU\ DOSKDEHW EH 3ZT DQG OHW f§ P f ‘ J ZKHUH Q DQG QT DUH WKH EORFN OHQJWKV IRU WKH ELQDU\ DQG WKH T DU\ V\VWHPV UHVSHFWLYHO\ DQG P LV DQ LQWHJHU 7KHQ WKH ZRUG HUURU UDWH IRU WKH TDU\ V\VWHP FDQ EH H[SUHVVHG DV 3T O3YTfD f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

PAGE 94

1$552:6 VFHQH DQG D KLJK GHWDLO VFHQH %$1.6,$6 >@ ,W LV REVHUYHG WKDW S\f ! S\frS\f! 3\f r3\f ! S\fmS\f rf ZKHUH S\cf GHQRWHV WKH SUREDELOLW\ RI LWK GLIIHUHQFH OHYHO 7KH SULPLWLYH LUUHGXFLEOH SRO\QRPLDOV RYHU WKH QRQELQDU\ ILHOGV DUH JLYHQ LQ 7DEOH 7KH PLQLPDO SRO\QRPLDOV DQG WKH JHQHUDWRU SRO\QRPLDOV RI WKH QRQELQDU\ FRGHV DUH JLYHQ LQ 7DEOHV DQG 1RWLFH WKDW *)f WKH JURXQG ILHOG IRU *)f LQ LWVHOI LV DQ H[WHQGHG YHUVLRQ RI *)f 7KH HOHPHQWV RI *)f DUH $ DQG % ZKHUH % $ 7KH DULWKPHWLF WDEOHV IRU *)f DUH SURYLGHG LQ 7DEOH $OSKDEHWV RI VL]H DUH QRW XVHG LQ WKH VLPXODWLRQV EHFDXVH LV QHLWKHU D SULPH RU SRZHU RI D SULPH QXPEHU DQG GRHV QRW OHDG WR %&+ FRGH LPSOHPHQWDWLRQ

PAGE 95

7$%/( ',))(5(1&( 6<0%2/ 352%$%,/,7,(6 3LFWXUH 3
PAGE 96

7$%/( $'',7,21 $1' 08/7,3/,&$7,21 7$%/(6 )25 *)f $ % $ % % $ $ $ % % % $ ; $ % $ % $ $ % % % $

PAGE 97

7$%/( $f 0,1,0$/ 32/<120,$/6 )25 (/(0(176 2) *)f 7+( 32:(56 2) D $5( 6+2:1 21 7+( /()7 &2/801 $1' 7+( &2()),&,(176 2) 0,1,0$/ 32/<120,$/6 $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 %f *(1(5$725 32/<120,$/6 )25 $5< &2'(6 &2()),&,(176 2) *;f $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 7+( 32:(56 2) D $5( *,9(1 ,1 7+( 0,''/( &2/801 *)f 0LQLPDO 3RO\QRPLDO Df DU\ FRGH 3RZHUV RI D J[f f f Ef

PAGE 98

7$%/( $f 0,1,0$/ 32/<120,$/6 )25 (/(0(176 2) *)f 7+( 32:(56 2) D $5( 6+2:1 21 7+( /()7 &2/801 $1' 7+( &2()),&,(176 2) 0,1,0$/ 32/<120,$/6 $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 %f *(1(5$725 32/<120,$/6 )25 $5< &2'(6 &2()),&,(176 2) *;f $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 7+( 32:(56 2) D $5( *,9(1 ,1 7+( 0,''/( &2/801 *)f 0LQLPDO 3RO\QRPLDO $OO % % $ $ $ $ % % % Df DU\ &RGH 3RZHUV RI D J[f f R + $ % f $ $ % f $%%% Ef

PAGE 99

7$%/( $f 0,1,0$/ 32/<120,$/6 )25 (/(0(176 2) *)f 7+( 32:(56 2) D $5( 6+2:1 21 7+( /()7 &2/801 $1' 7+( &2()),&,(176 2) 0,1,0$/ 32/<120,$/6 $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 %f *(1(5$725 32/<120,$/6 )25 $5< &2'(6 &2()),&,(176 2) *;f $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 7+( 32:(56 2) D $5( *,9(1 ,1 7+( 0,''/( &2/801 *)f 0LQLPDO 3RO\QRPLDO WR R Df DU\ FRGH 3RZHUV RI D J[f f f Ef

PAGE 100

7$%/( $f 0,1,0$/ 32/<120,$/6 )25 (/(0(176 2) *)f 7+( 32:(56 2) D $5( 6+2:1 21 7+( /()7 &2/801 $1' 7+( &2()),&,(176 2) 0,1,0$/ 32/<120,$/6 $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 %f *(1(5$725 32/<120,$/6 )25 $5< &2'(6 &2()),&,(176 2) *;f $5( 6+2:1 ,1 $6&(1',1* 25'(5 21 7+( 5,*+7 &2/8016 7+( 32:(56 2) D $5( *,9(1 ,1 7+( 0,''/( &2/801 *)f 0LQLPDO 3RO\QRPLDO Df DU\ FRGH 3RZHUV RI D J[f f WKURXJK f WKURXJK f WKURXJK Ef

PAGE 101

1RQELQDUY %&+ &RGHV 7KH QRQELQDU\ %&+ FRGHV DUH D VSHFLDO FODVV RI F\FOLF FRGHV WKDW FDQ EH FRQVWUXFWHG IRU DQ\ V\PERO DOSKDEHW GHILQHG RQ D ILQLWH ILHOG VD\ *)Tf ZKLFK FDQ EH D SULPH ILHOG RU VRPH H[WHQVLRQ RI D SULPH ILHOG $V D JHQHUDOL]DWLRQ RI ELQDU\ FDVH D WHUURU FRUUHFWLQJ %&+ FRGH RQ *)Tf LV D F\FOLF FRGH DQG DOO WKH FRGHZRUGV KDYH URRWV WKDW LQFOXGH W FRQVHFXWLYH SRZHUV RI VRPH HOHPHQW M FRQWDLQHG LQ *)TPf DQ H[WHQVLRQ ILHOG RI *)Tf ,W ZLOO EH FRQYHQLHQW WR GLVWLQJXLVK EHWZHHQ WKH WZR ILHOGV E\ FDOOLQJ WKH *)Tf WKH V\PERO ILHOG DQG *)TPf WKH ORFDWRU ILHOG $V ZLWK WKH ELQDU\ FRGHV %&+ FRGHV RQ *)Tf FDQ EH SULPLWLYH RU QRQSULPLWLYH GHSHQGLQJ RQ ZKHWKHU D SULPLWLYH RU QRQSULPLWLYH HOHPHQW RI *)TPf LV XVHG WR VSHFLI\ WKH FRQQHFWLYH URRWV RI FRGHZRUGV $ WHUURU FRUUHFWLQJ FRGH PD\ KDYH HLWKHU RGG RU HYHQ PLQLPXP GHVLJQ GLVWDQFH JLYHQ E\ G WO RU G W UHVSHFWLYHO\ 7KH JHQHUDWRU SRO\QRPLDO RI D %&+ FRGH RQ *)Tf LV GHILQHG DV WKH OHDVW FRPPRQ PXOWLSOH RI PLQLPDO SRO\QRPLDO RI DPR DPR DPRG WKH VHTXHQFH RI SRZHUV FDQ EHJLQ ZLWK DQ\ DUELWUDU\ SRZHU Df J [f /&0 >O!D}R [f -fD}rL [f !%Qf¯LL [f @ f 7KH EORFN OHQJWK RI WKH FRGH LV WKH RUGHU RI WKH HOHPHQW

PAGE 102

FKRVHQ WR SUHVFULEH WKH FRQVHFXWLYH URRWV DQG WKHUHIRUH IRU WKH SULPLWLYH FRGHV ZKHUH ZH FKRRVH D SULPLWLYH HOHPHQW RI *)TPf WKH EORFN OHQJWK LV Q TPO (QFRGLQJ 1RQELQDUY %&+ &RGHV $ %&+ FRGH RU DQ\ RWKHU F\FOLF FRGH FDQ EH HQFRGHG E\ XVLQJ WKH JHQHUDWRU SRO\QRPLDO J[f LQ WKH PDQQHU LQGLFDWHG E\ EDVLF GHILQLWLRQ RI D F\FOLF FRGH QDPHO\ Y[f X[f J[f f 7KDW LV ZH DVVRFLDWH D SRO\QRPLDO X[f RYHU *)Tf RI GHJUHH N ZLWK WKH VHW RI N LQIRUPDWLRQ V\PEROV WR EH WUDQVPLWWHG DQG PXOWLSO\ E\ GHJUHH U SRO\QRPLDO J[f IRUPLQJ WKH GHJUHH Q FRGH SRO\QRPLDO Y[f +RZHYHU WKLV UHVXOWV LQ D QRQV\VWHPDWLF FRGH VWUXFWXUH DQG WKHUHIRUH LW LV SUHIHUUHG LQVWHDG WR IRUP FRGHZRUGV XVLQJ Y[f >[UX[f PRG J[f @ [UX [f f ,W LV VHHQ WKDW WKLV HQFRGLQJ RSHUDWLRQ SODFHV WKH N LQIRUPDWLRQ V\PEROV LQ WKH N KLJKHVW RUGHU WHUPV RI FRGH SRO\QRPLDO ZKLOH WKH SDULW\ FKHFN V\PEROV DUH FRQILQHG WR WKH U ORZHVW RUGHU WHUPV (QFRGLQJ FDQ EH LPSOHPHQWHG ZLWK D GLYLVLRQ FLUFXLW XVLQJ OLQHDU IHHGEDFN VKLIW UHJLVWHU $Q HQFRGLQJ FLUFXLW XVLQJ D

PAGE 103

VKLIW UHJLVWHU ZLWK U QN VWDJHV LV VKRZQ LQ )LJXUH (DFK ER[ LQ WKH FLUFXLW LV D TDU\ VWRUDJH GHYLFH 7KH PXOWLSOLFDWLRQV DQG DGGLWLRQV DUH GRQH LQ *)Tf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

PAGE 104

WKH REVHUYHG V\QGURPH YDOXHV 7KHUHIRUH GHFRGLQJ LV YLHZHG DV D SUREOHP RI VROYLQJ D VHW RI VLPXOWDQHRXV V\QGURPH HTXDWLRQV RQH ZKHUH WKH VHW RI XQNQRZQV LQFOXGH HUURU YDOXHV DV ZHOO DV HUURU ORFDWLRQV /HW Y[f H[f DQG U[f UHSUHVHQW WKH WUDQVPLWWHG ZRUG HUURU SDWWHUQ DQG WKH UHFHLYHG ZRUG UHVSHFWLYHO\ ZKHUH WKH FRHIILFLHQWV DUH HOHPHQWV LQ *)Tf ZH KDYH U [f Y[fH[f f 7KH HUURU SRO\QRPLDO H[f KDV QRQ]HUR WHUPV RQO\ LQ WKRVH SRVLWLRQV ZKHUH HUURUV KDYH RFFXUUHG VR WKDW LI WKHUH DUH W HUURUV LQ WKH UHFHLYHG ZRUG ZH FDQ ZULWH WKH V\QGURPH YDOXHV DV F VN
PAGE 105

FRUUHFW WKH LQGLFDWHG HUURUV )LJXUH 7KH OLQHDU IHHGEDFN VKLIW UHJLVWHU IRU JHQHUDWLQJ D VHTXHQFH RI V\QGURPH YDOXHV LQ %HUOHNDPS0DVVH\ DOJRULWKP 7KH 3HWHUVRQn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

PAGE 106

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f EXW WKH RXWSXW VRXUFH LV QRW TXLWH DV DV\PPHWULF DV WKH LQSXW VRXUFH $IWHU HQFRGLQJ WKH ODUJHU SUREDELOLWLHV JHW VOLJKWO\ VPDOOHU DQG WKH VPDOOHU SUREDELOLWLHV JHW VOLJKWO\ ELJJHU )LJXUHV LOOXVWUDWH KRZ WKH V\VWHPV XQGHU

PAGE 107

7$%/( 7+( %&+ &2'(6 7+( &2'( 5$7(6 $1' 7+( 48$17,7< KEKT )25 $f 0(',80 '(7$,/ ,0$*( %f +,*+ '(7$,/ ,0$*( $1' &f /2: '(7$,/ ,0$*( &RGH &RGH UDWH KEKD f f f f Df &RGH &RGH UDWH KEKD f f f f Ef &RGH &RGH UDWH KEKD f f f f Ff

PAGE 108

7$%/( 3,&785( 352%$%,/,7,(6 %()25( $1' $)7(5 &+$11(/ (1&2',1* $f $5< &2'( f )25 7+( /2: '(7$,/ ,0$*( %f $5< &2'( f )25 7+( /2: '(7$,/ ,0$*( &f $5< &2'( f )25 7+( +,*+ '(7$,/ ,0$*( $1' 'f $5< &2'( f )25 7+( +,*+ '(7$,/ ,0$*( EHIRUH HQFRGLQJ DIWHU HQFRGLQJ Df EHIRUH HQFRGLQJ DIWHU HQFRGLQJ Ef EHIRUH HQFRGLQJ DIWHU HQFRGLQJ Ff

PAGE 109

7DEOH f§FRQWLQXHG EHIRUH HQFRGLQJ DIWHU HQFRGLQJ &0 2 fµ Gf

PAGE 110

FRQVLGHUDWLRQ SHUIRULQ 7KH H[SUHVVLRQV SORWWHG DUH WKH ZRUG HUURU SUREDELOLW\ :KHQ WKH EDQGZLGWK FRQVWUDLQW LV VWULFW WKH WHUQDU\ DOSKDEHW LV VHHQ WR EH VXSHULRU WR WKH ELQDU\ V\VWHP )LJXUH Dff ,W LV DOVR VHHQ WKDW WKH V\VWHPV HPSOR\LQJ WKH DOSKDEHWV RI VL]H IRXU ILYH DQG VHYHQ DUH PRUH HIILFLHQW WKDQ WKH ELQDU\ V\VWHP DERYH D FHUWDLQ FRGLQJ WKUHVKROG )LJXUHV EfGff 0RUHRYHU WKH FRGLQJ WKUHVKROG VHHPV WR EH ODUJHU IRU ODUJHU DOSKDEHWV VL]HV 7KH QRQELQDU\ V\VWHPV DUH H[SHFWHG WR RXWSHUIRUP WKH ELQDU\ V\VWHP VLQFH WKH\ XWLOL]H HUURU FRUUHFWLRQ FRGHV DQG WKH ELQDU\ V\VWHP GRHV QRW 1RQELQDU\ V\VWHPV ZHUH DOVR FRPSDUHG WR HDFK RWKHU 7KH UHVXOWV LQGLFDWH WKDW WKH WHUQDU\ V\VWHP LV VXSHULRU WR WKH RWKHU RQHV )LJXUH Df LQGLFDWHV WKDW WHUQDU\ V\VWHP RXWSHUIRUPV WKH DU\ V\VWHP EXW WKH LPSURYHPHQWV GHFUHDVH VORZO\ DV WKH VLJQDOWRQRLVH UDWLR LQFUHDVHV DU\ V\VWHP RXWSHUIRUPV V\VWHPV HPSOR\LQJ DOSKDEHWV RI VL]H IRXU DQG ILYH DERYH UHDVRQDEOH WKUHVKROGV 7KH FRPSDULVRQ EHWZHHQ WKH V\VWHPV ZLWK DU\ DQG DU\ DOSKDEHWV LV VKRZQ LQ )LJXUH Ef )RU WKH RWKHU FDVH ZKHQ WKH EDQGZLGWK FRQVWUDLQW FDQ EH VRPHZKDW UHOD[HG WKH VLWXDWLRQ LV GLIIHUHQW 7KHUH LV DOZD\V D WKUHVKROG DERYH ZKLFK WKH QRQELQDU\ LV VXSHULRU WR WKH ELQDU\ V\VWHP )RU D WZHQW\ IRXU SHUFHQW EDQGZLGWK H[SDQVLRQ WKH WHUQDU\ V\VWHP LV VWLOO VXSHULRU WR WKH ELQDU\ V\VWHP DERYH D WKUHVKROG EXW WKH LPSURYHPHQWV RYHU WKH ELQDU\ V\VWHP

PAGE 111

DUH QRW DV PXFK DV WKH IL[HG EDQGZLGWK FDVH )LJXUH Dff )RU ODUJHU EDQGZLGWK H[SDQVLRQ IDFWRUV WKH FRGLQJ WKUHVKROG EHFRPHV ODUJH )RU YHU\ ODUJH EDQGZLGWK H[SDQVLRQ IDFWRUV WKH ELQDU\ V\VWHP LV VXSHULRU WR WHUQDU\ V\VWHPV 7KH ELQDU\ V\VWHP WXUQV RXW EH VXSHULRU WR ODUJHU DOSKDEHW V\VWHPV DU\ DU\ DQG DU\f ZLWK D OLWWOH EDQGZLGWK H[SDQVLRQ EHORZ D WKUHVKROG )LJXUH Ef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

PAGE 112

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

PAGE 113

Df Ef )LJXUH 6\VWHPV ZLWK QRQELQDU\ DOSKDEHWV DUH FRPSDUHG DJDLQVW WKH ELQDU\ V\VWHP :RUG HUURU UDWH LV SORWWHG YHUVXV DYHUDJH SRZHU ZKHQ WKH EDQGZLGWK LV IL[HG $ ZRUG LV FRQVLGHUHG WR EH RQH EORFN ORQJ 7KH VROLG FXUYH UHSUHVHQWV ELQDU\ V\VWHP DQG WKH GDVKHG FXUYH UHSUHVHQWV Df D WHUQDU\ V\VWHP XVLQJ f FRGH IRU PHGLXP GHWDLO SLFWXUH Ef D DU\ V\VWHP XVLQJ f FRGH IRU ORZ GHWDLO LPDJH Ff D DU\ V\VWHP XVLQJ f FRGH IRU PHGLXP GHWDLO SLFWXUH DQG Gf D DU\ V\VWHP XVLQJ f FRGH IRU PHGLXP GHWDLO SLFWXUH

PAGE 114

Ff )LJXUH FRQWLQXHG Gf

PAGE 115

Df Ef )LJXUH $ FRPSDULVRQ RI V\VWHPV HPSOR\LQJ QRQELQDU\ DOSKDEHWV DJDLQVW WKH ELQDU\ V\VWHP ZKHQ WKH EDQGZLGWK FDQ VRPHZKDW EH H[SDQGHG 7KH VROLG FXUYH UHSUHVHQWV WKH ELQDU\ V\VWHP Df WHUQDU\ f FRGH DQG ELQDU\ f FRGH DUH XVHG DQG D EORFN LV ELWV ORQJ Ef DU\ f DQG ELQDU\ f DUH XVHG DQG WKH EORFN OHQJWK LV ELWV ORQJ 7KH UHXOWV DUH JLYHQ IRU D PHGLXP GHWDLO SLFWXUH

PAGE 116

R G%! Df H G%f Ef )LJXUH 6\VWHPV ZLWK QRQELQDU\ DOSKDEHWV DUH FRPSDUHG IRU WKH IL[HG EDQGZLGWK FDVH 7KH ILUVW V\VWHP LV UHSUHVHQWHG ZLWK D VROLG FXUYH DQG WKH VHFRQG V\VWHP LV UHSUHVHQWHG ZLWK D GDVKHG FXUYH Df DU\ DQG WHUQDU\ V\VWHPV DQG Ef DU\ DQG DU\ V\VWHPV UHVSHFWLYHO\

PAGE 117

Df Ef )LJXUH 6\VWHPV ZLWK QRQELQDU\ DOSKDEHWV DUH FRPSDUHG DJDLQVW WKH ELQDU\ V\VWHP :RUG HUURU UDWH LV SORWWHG YHUVXV DYHUDJH SRZHU ZKHQ WKH EDQGZLGWK LV IL[HG $ ZRUG LV FRQVLGHUHG WR EH RQH EORFN ORQJ 7KH VROLG FXUYH UHSUHVHQWV ELQDU\ V\VWHP DQG WKH GDVKHG FXUYH UHSUHVHQWV Df D WHUQDU\ V\VWHP XVLQJ f FRGH Ef D DU\ V\VWHP XVLQJ f FRGH Ff D DU\ V\VWHP XVLQJ f FRGH IRU PHGLXP GHWDLO SLFWXUH DQG Gf D DU\ V\VWHP XVLQJ f FRGH 7KH UHVXOWV DUH JLYHQ IRU KLJK GHWDLO SLFWXUH

PAGE 118

,OO mIIOf &f Gf )LJXUH FRQWLQXHG

PAGE 119

Df )LJXUH 6\VWHPV ZLWK QRQELQDU\ DOSKDEHWV DUH FRPSDUHG DJDLQVW WKH ELQDU\ V\VWHP :RUG HUURU UDWH LV SORWWHG YHUVXV DYHUDJH SRZHU ZKHQ WKH EDQGZLGWK LV IL[HG $ ZRUG LV FRQVLGHUHG WR EH RQH EORFN ORQJ 7KH VROLG FXUYH UHSUHVHQWV ELQDU\ V\VWHP DQG WKH GDVKHG FXUYH UHSUHVHQWV Df D WHUQDU\ V\VWHP XVLQJ f FRGH Ef D DU\ V\VWHP XVLQJ f FRGH Ff D DU\ V\VWHP XVLQJ f FRGH DQG Gf D DU\ V\VWHP XVLQJ f FRGH $ ORZ GHWDLO SLFWXUH LV XVHG

PAGE 120

ZRUG HUURU PH

PAGE 121

Df E 8%f Ef )LJXUH $ FRPSDULVRQ RI V\VWHPV HPSOR\LQJ QRQELQDU\ DOSKDEHWV DJDLQVW WKH ELQDU\ V\VWHP ZKHQ WKH EDQGZLGWK FDQ VRPHZKDW EH H[SDQGHG 7KH VROLG FXUYH UHSUHVHQWV WKH ELQDU\ V\VWHP Df WHUQDU\ f FRGH DQG ELQDU\ f FRGH DUH XVHG DQG D EORFN LV ELWV ORQJ Ef DU\ f DQG ELQDU\ f DUH XVHG IRU D ORZ GHWDLO SLFWXUH DQG EORFN OHQJWK LV ELWV ORQJ 7KH UHVXOWV DUH JLYHQ IRU D KLJK GHWDLO SLFWXUH

PAGE 122

Df Ef )LJXUH FRPSDULVRQ RI V\VWHPV HPSOR\LQJ QRQELQDU\ DOSKDEHWV DJDLQVW WKH ELQDU\ V\VWHP ZKHQ WKH EDQGZLGWK FDQ VRPHZKDW EH H[SDQGHG 7KH VROLG FXUYH UHSUHVHQWV WKH ELQDU\ V\VWHP Df WHUQDU\ f FRGH DQG ELQDU\ f FRGH DUH XVHG DQG D EORFN LV ELWV ORQJ Ef DU\ f DQG ELQDU\ f DUH XVHG DQG WKH EORFN OHQJWK LV ELWV ORQJ $ KLJK GHWDLO SLFWXUH LV XVHG

PAGE 123

&+$37(5 ),9( ()),&,(17 '(&2',1* 2) &255(/$7(' 6(48(1&(6 ,QWURGXFWLRQ 'LIIHUHQWLDO SXOVH FRGH PRGXODWLRQ '3&0f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

PAGE 124

HUURQHRXV SL[HOV ,Q WKLV PHWKRG D SL[HO LV FRPSDUHG ZLWK SL[HOV RQ WKH SUHYLRXV DQG IROORZLQJ OLQHV ,I WKH GLIIHUHQFHV DUH DERYH D WKUHVKROG DQ HUURQHRXV SL[HO LV IRXQG 7KH FRGHZRUG DVVLJQPHQW LV VXFK WKDW HUURUV ZLOO EH HLWKHU VPDOO RU ODUJH %HFDXVH RI KLJK FRUUHODWLRQ EHWZHHQ DGMDFHQW OLQHV WKH SL[HO LQ HUURU LV UHSODFHG E\ WKH FRUUHVSRQGLQJ SL[HO LQ WKH SUHYLRXV OLQH :DOVK+DGDPDUG WUDQVIRUPDWLRQ KDYH EHHQ XVHG E\ )HQZLFN 6WHHO DQG 9DQVDQML >@ ,Q WKLV PHWKRG '3&0 EORFNV DUH WUDQVIRUPHG DQG VHQW RYHU WKH FKDQQHO '3&0 EORFNV DUH WUDQVIRUPHG LQ D PDQQHU WKDW ZKHQ WKH UHFHLYHU UHFRYHUV WKH WUDQVIRUP FRHIILFLHQWV WKH SUHVHQFH RI DQ HUURU LV GHWHFWHG DQG LWV PDJQLWXGH LV NQRZQ &RUUHODWLRQ WHFKQLTXHV DUH XVHG WR ILQG WKH HUURQHRXV VHTXHQFH 3HULRGLF 3&0 XSGDWHV KDYH EHHQ XVHG E\ VRPH UHVHDUFKHUV EHIRUH $ JRRG H[DPSOH LV WKH ZRUN E\ 1JDQ DQG 6WHHOH >@ 7KH\ XVHG VHYHUDO 3&0 XSGDWHV SHU OLQH ,I WKH GLIIHUHQFH EHWZHHQ WKH 3&0 XSGDWH ZRUG DQG '3&0 RXWSXW LV JUHDWHU WKDQ D WKUHVKROG WKDW SRWLRQ RI WKH OLQH LV LQ HUURU 7KH SL[HO YDOXH LQ HUURU LV UHSODFHG E\ D FRPELQDWLRQ RI WKH DGMDFHQW SL[HOV 7KH SHUFHQWDJH LQFUHDVH LQ ELW UDWH GXH WR VLGH LQIRUPDWLRQ ZDV DERXW SHUFHQW 0DQ\ RI WKH V\VWHPV SUHVHQWHG WDNH DGYDQWDJH RI VLJQDO VWDWLVWLFV 6WHHO -D\DQW DQG 6FKPLGW >@ LQWURGXFHG D VWDWLVWLFDO EORFN SURWHFWLRQ SURFHGXUH ,Q WKLV ZRUN WKH VWDWLVWLFDO IHDWXUHV RI DQ LQSXW EORFN VXFK DV WKH DYHUDJH DQG

PAGE 125

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f DQG D PD[LPDO VLJQDOWRQRLVH UDWLR 0615f ZLOO DOVR EH SUHVHQWHG 7KH DSSOLFDWLRQ RI WKH SURSRVHG PHWKRGV WR SLFWXUH WUDQVPLVVLRQ RYHU QRLV\ FKDQQHOV ZLOO GLVFXVVHG LQ VHFWLRQ 7KH UHVXOWV ZLOO FRPSDUHG WR WKH SHUIRUPDQFH RI D UHFHQWO\ SURSRVHG WHFKQLTXH

PAGE 126

2SWLPXP 'HFRGLQJ RI 0DUNRY 6RXUFHV +HUH LW ZLOO EH VKRZQ WKDW WKH RSWLPDO GHFRGLQJ RI 0DUNRY VHTXHQFHV LQ WKH VHQVH RI PLQLPL]LQJ WKH HUURU UDWH UHTXLUHV WUHH VHDUFKLQJ $ SUDFWLFDO VROXWLRQ WR WKH VHDUFK SUREOHP LV SUHVHQWHG DQG D SDWK PHWULF IRU HDFK EUDQFK LV GHULYHG 7KH SDWK ZLWK PD[LPXP FXPXODWLYH PHWULF XVLQJ WKH 0/f KHXULVWLF VHDUFK URXWLQH LV VKRZ WR EH D VXERSWLPXP VROXWLRQ WR WKH SUREOHP /HW Y Y Y Y1f DQG U UU U1f GHQRWH WKH WUDQVPLWWHG DQG WKH UHFHLYHG VHTXHQFHV UHVSHFWLYHO\ ZKHUH V\PEROV Yc DQG UV DUH IURP WKH VDPH 4DU\ DOSKDEHW 7KH HUURU SUREDELOLW\ RI D GHFRGHU LV JLYHQ E\ 3(f e3e_ef3[f f [ ZKHUH 3Uf LV LQGHSHQGHQW RI WKH GHFRGLQJ UXOH 7R PLQLPL]H 3(f RQH PXVW PLQLPL]H 3(?]f 39rY?]f f 0LQLPL]LQJ LV HTXLYDOHQW WR PD[LPL]LQJ %\ %D\HV UXOH f 6LQFH WKH FKDQQHO LV PHPRU\OHVV

PAGE 127

3]?\f A38-YMf f )RU DQ 0WK RUGHU 0DUNRY VHTXHQFH 3OIf -,3YL?YLLnYL Y¯PA f L 6XEVWLWXWLQJ f DQG f LQWR f ZH JHW S\_[f Q3LUAPY-YZ 7LOf 9L0! 7DNLQJ WKH ORJDULWKP RI ERWK VLGHV JLYHV f ORJ3]_]f ORJ 3LU-YMf 3YL_YL YL0ff ORJ 3Qf L f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

PAGE 128

QRGH WR WKH QH[W WLPH VDPSOH 7KH SDWK ZLWK WKH ODUJHVW FXPXODWLYH PHWULF RI ;f ORJ 38L _ Yf 3LYYA+ff L ZLOO EH WKH RSWLPXP SDWK $VVRFLDWHG ZLWK HDFK EUDQFK RI WKH WUHH LV D EUDQFK PHWULF 7KH EUDQFK PHWULFV RI WKH WUHH DUH JLYHQ E\ 7LUMYMf ORJ3UL_YLf 3LYAYA YAff f 2QFH WKH WUHH LV DSSURSULDWHO\ SRSXODWHG WKH QH[W WDVN LV WR VHOHFW D VHTXHQFH WKDW PLQLPL]HV WKH DGGLWLYH PHWULF $Q H[KDXVWLYH VHDUFK RYHU DOO FDQGLGDWH VHTXHQFHV ZLOO SURYLGH WKH RSWLPXP VROXWLRQ +RZHYHU WKH HQRUPRXV FRPSXWDWLRQDO EXUGHQ RI WKH H[KDXVWLYH VHDUFK VXJJHVWV WKH XVH RI D QRQ H[KDXVWLYH VHDUFK SURFHGXUH WKDW FRQFHQWUDWHV RQ WKH PRVW SURPLVLQJ SRUWLRQV RI WKH WUHH 7KHUH DUH D QXPEHU RI WUHH VHDUFKLQJ DOJRULWKPV :H FKRRVH WR XVH WKH 0/f DOJRULWKP >@ 7KLV DOJRULWKP EHLQJ D EUHDGWKILUVW SURFHGXUH H[DPLQHV DOO WKH EUDQFKHV H[WHQGLQJ IURP WKH FXUUHQW 0 QRGHV $W WKLV SRLQW RQO\ WKH EHVW 0 RI WKHVH EUDQFKHV LQ WKH VHQVH RI ODUJHVW FXPXODWLYH SDWK PHWULF DUH VDYHG $Q H[DPSOH LV SUHVHQWHG LQ )LJXUH ZKHUH WKH EUDQFKLQJ IDFWRU RI WKH WUHH LV WKUHH $W HDFK GHSWK RQO\ WKUHH EUDQFKHV DUH VDYHG DQG WKH UHVW DUH GLVFDUGHG

PAGE 129

%HFDXVH RI WZR LQWHUHVWLQJ IHDWXUHV WKH 0/f LV IDYRUHG LQ SUDFWLFH 7KH RXWSXWV DUH UHOHDVHG LQ D V\QFKURQRXV PDQQHU DQG D IL[HG QXPEHU RI EUDQFKHV DUH YLVLWHG DW HDFK GHSWK 7KH DOJRULWKP LQ RQH IRUP RU DQRWKHU KDV EHHQ XVHG LQ VSHHFK DQG LPDJH FRGLQJ SUHYLRXVO\ >@>@ )LJXUH $ WUHH ZLWK EUDQFKLQJ IDFWRU RI WKUHH LV VKRZQ $W HDFK GHSWK RQO\ WKUHH EUDQFKHV DUH VDYHG $ 0RGLILHG 0$3 00$3f 5HFHLYHU $ 0$3 UHFHLYHU FKRRVHV e r LI SUL?YL DNf S!SUL_YL D-fSYADAf f IRU M O4 DQG M"¯N

PAGE 130

+RZHYHU LI VRPH UHGXQGDQF\ LV SUHVHQW LQ WKH HQFRGHG VLJQDO D PRUH UHDVRQDEOH FULWHULD FDQ EH IRUPXODWHG $VVXPLQJ WKDW WKH GDWD KDV D PHPRU\ RI 0 DQG Ec E0 DUH WKH SUHYLRXV GHFLVLRQV WKH PRGLILHG 0$3 UXOH ZLOO EH VHW £ DN LI 3U. Dr! 3YL DN?YLL ELLn fµ fµ fµ r YL0 ELQf ! SUL_YL DMfSYL DM_YLB ELBOI YAEAf f IRU M O 4 DQG M"¯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

PAGE 131

4DU\ DOSKDEHW ^DM` IRU M O4 7KH DYHUDJH FRVW RU ORVVf LQ GHFRGLQJ WR DM LV JLYHQ E\ 1 GL (/L-WAULOYL D-Wf 3YL D-W! f N ZKHUH /MN LV WKH FRVW LQ GHFLGLQJ WKDW DM ZDV WUDQVPLWWHG ZKHUH DFWXDOO\ DN ZDV WUDQVPLWWHG )RU HDFK UHFHLYHG V\PERO WKH %D\HV UHFHLYHU FRPSXWHV GZ GG4 DQG GHFLGHV D ZDV WUDQVPLWWHG LI G LV WKH VPDOOHVW RI DOO GM ,Q RWKHU ZRUGV WKH %D\HV UHFHLYHU FKRRVHV £ D LI GGM IRU DOO M O4 MAOfµ :KHQ WKH UHFHLYHG VHTXHQFH LV D 0DUNRY VHTXHQFH RI RUGHU 0 WKH SUHVHQW V\PERO GHSHQGV RQ 0 SUHYLRXV V\PEROV 7KXV H[SUHVVLRQ f QHHGV WR EH PRGLILHG WR UHIOHFW WKH UHGXQGDQF\ LQ WKH VHTXHQFH 1 GI ( KN 3AL:L DNf SYG DN?r eB r L f $ 0D[LPDO 6LTQDOWR1RLVH 5DWLR 0615f 'HFRGHU ,Q WKLV VHFWLRQ D PD[LPDO VLJQDOWRQRLVH UDWLR 0615f UHFHLYHU IRU '3&0 GLIIHUHQWLDO SXOVH FRGH PRGXODWLRQf RI SLFWXUHV LV GHYHORSHG 7R GR VR WKH PRGLILHG %D\HV FODVVLILHU IRU 0DUNRY VRXUFHV IRXQG HDUOLHU ZLOO EH XWLOL]HG DQG D FRVW

PAGE 132

PDWUL[ RU LWV HOHPHQWV /MNf ZLOO EH GHULYHG /HW WKH DOSKDEHW 6M n r n UHSUHVHQW RXWSXW OHYHOV \OI\I fµ fµ fµ \ LQ D '3&0 V\VWHP 6XSSRVH WKH UHFHLYHU PDNHV DQ HUURU DQG DM LV FKRVHQ ZKHUH DN ZDV DFWXDOO\ VHQW 7KH UHFRQVWUXFWLRQ HUURU DPRXQWV WR \M\N 7KH HUURU DIIHFWV WKH IROORZLQJ VDPSOHV 7KH FRQWULEXWLRQ RI HUURU WR IROORZLQJ V\PEROV ZLOO EH D\M\Nf D@ 7KH WRWDO GLVWRUWLRQ SRZHU GXH WR D FKDQQHO HUURU LV WKHUHIRUH JLYHQ E\ 1 \M\NfD \M\NfD \M\Nf f :Ma\Nf >ODD @ \M\Nff§ G ,I ZH VHW /M N \N\A Lf WKH PRGLILHG %D\HV UHFHLYHU ZLOO PLQLPL]H WKH DYHUDJH FRVW ZKLFK LV GLVWRUWLRQ SRZHU GXH WR FKDQQHO HUURUV LQ WKLV FDVH 7KXV WKH PD[LPDO 615 UHFHLYHU FDOFXODWHV GAGM G4 WKH DYHUDJH ORVV IRU HDFK V\PEROf DQG FKRRVHV WKH V\PERO ZLWK WKH OHDVW DYHUDJH ORVV ZKHUH DYHUDJH ORVV LV

PAGE 133

1 GM @& \N\Mf SU$YL DNnf SYL DN?YLL E f 7KH JLYHQ SURRI LV RQO\ DSSOLFDEOH WR ILUVW RUGHU '3&0 V\VWHPV /HW XV JLYH D PRUH JHQHUDO SURRI ,W KDV EHHQ VKRZQ WKDW WKH VLJQDOWRQRLVH UDWLR RI D '3&0 V\VWHP LV JLYHQ E\ >@ ZKHUH D LV WKH YDULDQFH RI WKH VLJQDO DH LV WKH YDULDQFH RI WKH LQSXW WR WKH TXDQWL]HU DQG f LQ ZKLFK HT UHSUHVHQWV WKH PHDQVTXDUHG TXDQWL]DWLRQ QRLVH HP LV PXWXDO HUURU WHUP DQG HF UHSUHVHQWV QRLVH GXH WR FKDQQHO QRLVH 7KH RQO\ WHUP LQ WKH 615 H[SUHVVLRQ WKDW GHSHQGV RQ WKH FKRLFH RI UHFHLYHU LV HF HF LV JLYHQ E\ f ZKHUH ']]f LV WKH GLVFUHWH V\VWHP WUDQVIHU IXQFWLRQ DQG

PAGE 134

Hm ‹ AMF\Mf SN?L 3L N f 7KH 615 FDQ EH PD[LPL]HG E\ PLQLPL]LQJ HQ 2QH FDQ WKLQN RI HQ DV DQ DYHUDJH FRVW IXQFWLRQ ZKHUH /NL \N\Lf f 7KHUHIRUH 615 LV PD[LPL]HG E\ PLQLPL]LQJ WKH FRVW IXQFWLRQ DQG VHWWLQJ WKH FRVW DFFRUGLQJ WR f 5HGXQGDQF\ LQ WKH (QFRGHG 6LJQDOV ,W KDV EHHQ VKRZQ WKDW DVVXPLQJ DQ $5 PRGHO IRU WKH VLJQDOV WKH RXWSXW RI WKH HQFRGHU FRQWDLQV VRPH UHGXQGDQF\ >@ +HUH ZH ZLOO H[DPLQH WKH HQFRGHG VLJQDO DQG LWV DXWRFRUUHODWLRQ IXQFWLRQ WR JHW D IHHO IRU WKH H[WHQW RI UHVLGXDO UHGXQGDQF\ 6WDQGDUG [ SL[HO HLJKWELW SLFWXUHV VXFK DV /(11$ ZHUH HQFRGHG XVLQJ D ILUVW RUGHU '3&0 HQFRGHU 7KH DYHUDJH QRUPDOL]HG DXWRFRUUHODWLRQ LQ KRUL]RQWDO DQG YHUWLFDO GLUHFWLRQ ZHUH IRXQG DQG SORWWHG )LJXUH f 7KLV VKRZV WKDW FHUWDLQ DPRXQW RI FRUUHODWLRQ DPRQJ WKH DGMDFHQW HQFRGHG V\PEROV H[LVW 7KH '3&0 SUHGLFWLRQ HUURU VLJQDO LV VKRZQ LQ

PAGE 135

)LJXUH 7KLV DOVR LQGLFDWHV WKDW VRPH FRUUHODWLRQ LV SUHVHQW LQ WKH HQFRGHG SLFWXUH $OWKRXJK VLJQLILFDQW DPRXQW RI UHGXQGDQF\ KDV EHHQ UHPRYHG IURP WKH SLFWXUH VRPH VWUXFWXUH LV VWLOO SUHVHQW LQ WKH HQFRGHG VLJQDO 7KH GHFRGLQJ PHWKRGV JLYHQ LQ WKLV FKDSWHU

PAGE 136

Df Ef )LJXUH 7KH DYHUDJH DXWRFRUUHODWLRQ IXQFWLRQ IRU WKH SLFWXUH WRS FXUYHf DQG WKH HQFRGHG SLFWXUH WKH ERWWRP FXUYHf Df KRUL]RQWDO GLUHFWLRQ Ef YHUWLFDO GLUHFWLRQ

PAGE 137

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

PAGE 138

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f Sf¯OSfQDf f ZKHUH D LV WKH +DPPLQJ GLVWDQFH EHWZHHQ Uc DQG Yc DQG Q LV WKH QXPEHU RI ELWV LQ D '3&0 ZRUG :H XVHG D VHYHQOHYHO TXDQWL]HU JLYHQ LQ >@ WKDW ZDV IRXQG H[SHULPHQWDOO\ DQG LV VHHQ WR JLYH VDWLVIDFWRU\ UHVXOWV IRU YDULHW\ RI SLFWXUHV :H GHILQH VLJQDOWRQRLVH UDWLR 615f DV WKH UDWLR EHWZHHQ WKH SRZHU LQ WKH RULJLQDO SLFWXUH WR WKH SRZHU LQ WKH GLIIHUHQFH EHWZHHQ WKH RULJLQDO DQG WKH UHFHLYHG SLFWXUH

PAGE 139

Df fµ 2 Ef )LJXUH 7KH VROLG FLUFOHV LQGLFDWH WKH QHLJKERULQJ V\PEROV XVHG LQ FRPSXWLQJ FRQGLWLRQDO VWDWLVWLFV Df VHFRQG RUGHU 0DUNRY PRGHO DQG Ef WKLUG RUGHU 0DUNRY PRGHO 7$%/( &2'(:25' $66,*10(17 )25 9$5,286 48$17,=(5 /(9(/6 /HYHO 3UREDELOLW\ FRGHZRUG

PAGE 140

5HVXOWV '3&0 GHFRGHUV ZLWK ILUVW VHFRQG DQG WKLUG RUGHU 0DUNRY VRXUFH PRGHOV ZHUH LPSOHPHQWHG DQG FRPSDUHG EDVHG XSRQ WKH 615 RI WKH UHFRQVWUXFWHG VLJQDO DQG VXEMHFWLYH LPSUHVVLRQ &RQVLGHUDEOH LPSURYHPHQWV LQ WKH 615 RI WKH UHFHLYHG VLJQDO DQG VXEMHFWLYH LPSUHVVLRQ LV REWDLQHG )LJXUH Df DQG Ef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f 7KH 00$3 ZDV FRPSDUHG WR WKH 9LWHUEL GHFRGLQJ PHWKRG JLYHQ LQ >@ 7KH 00$3 ZLWK D ILUVW RUGHU 0DUNRY VRXUFH LV FRPSDUDEOH LQ SHUIRUPDQFH WR 6D\RRG DQG %RUNHQKDJHQnV UHFHLYHU >@ )LJXUH f EXW LW KDV WKH DGYDQWDJH RI UXQQLQJ IDVWHU DQG UHTXLULQJ QR PHPRU\

PAGE 141

, Df Ef )LJXUH 7KH YDULDWLRQ RI SLFWXUH 615 DV D IXQFWLRQ RI FKDQQHO HUURU UDWH 7KH UHFHLYHU XWLOL]LQJ ILUVW WKH GRWWHG FXUYHf VHFRQG WKH GDVKHG FXUYHf DQG WKLUG WKH VROLG FXUYHf RUGHU 0DUNRY PRGHOOLQJ RI WKH VRXUFH Df 00$3 UHFHLYHU Ef 0615 UHFHLYHU

PAGE 142

)LJXUH 7KH SLFWXUH 615 YV FKDQQHO HUURU UDWH 7KH GDVKHG FXUYH UHSUHVHQWV WKH 00$3 UHFHLYHU XWLOL]LQJ D ILUVW RUGHU 0DUNRY PRGHO DQG WKH VROLG FXUYH UHSUHVHQWV WKH 6D\RRG DQG %RUNHQKDJHQ PHWKRG >@ %RWK 00$3 DQG 0615 UHFHLYHUV LPSURYH WKH 615 RI WKH UHFHLYHG SLFWXUH D JUHDW GHDO )RU LQVWDQFH WKH 0$3 UHFHLYHU LPSURYHV WKH SLFWXUH 615 E\ DERXW G% IRU D ELW HUURU UDWH RI S O )LJXUH Dff 7KH PD[LPDO 615 UHFHLYHU KRZHYHU UHVXOWV LQ VRPHZKDW KLJKHU YDOXH RI 615 FRPSDUHG WR WKH PRGLILHG 0$3 UHFHLYHU IRU D JLYHQ RUGHU RI 0DUNRY VRXUFH )LJXUH Eff )LJXUH VKRZV WKDW WKH 0615 LV DERXW G% VXSHULRU WR WKH 00$3 UHFHLYHU ,W LV DOVR REVHUYHG WKDW WKH 0615 UHFHLYHU LV DW OHDVW G% VXSHULRU WR PHWKRG JLYHQ LQ >@ )LJXUH Eff

PAGE 143

2 226 226 Df )LJXUH 7KH SLFWXUH 615 YV FKDQQHO HUURU UDWH WKH GRWWHG FXUYHV UHSUHVHQW WKH FRUUXSWHG SLFWXUH Df 7KH VROLG FXUYH UHSUHVHQWV WKH 00$3 UHFHLYHU Ef WKH VROLG FXUYH UHSUHVHQWV WKH 0615 UHFHLYHU DQG WKH GDVKHG FXUYH UHSUHVHQWV WKH 6D\RRG DQG %RUNHQKDJHQ PHWKRG >@

PAGE 144

)LJXUH 7KH SLFWXUH 615 YV FKDQQHO HUURU UDWH 7KH VROLG FXUYH UHSUHVHQWV WKH 0615 UHFHLYHU DQG WKH GRWWHG FXUYH UHSUHVHQWV WKH 00$3 UHFHLYH 7KH PLQLPXP HUURU GHFRGLQJ SURFHGXUH ZLWK D OLPLWHG VHDUFK LV H[SHFWHG WR RXWSHUIRUP WKH 00$3 UHFHLYHU 7DEOH FRPSDUHV WKH SHUIRUPDQFH RI D 00$3 UHFHLYHU ZLWK D PLQLPXP HUURU UHFHLYHU WKDW XWLOL]HV D 0 / f VHDUFK )RU VPDOOHU HUURU UDWH WKH LPSURYHPHQWV DUH DURXQG G% EXW IRU YHU\ QRLV\ FKDQQHOV WKH VHDUFK GRHV QRW VHHP WR EH EHQHILFLDO 7$%/( 7+( 3,&785( 615 68%-(&7 72 9$5,286 &+$11(/ (5525 5$7(6 )25 00$3 $1' 0,1 (5525 5(&(,9(5 $ 0 / f 6($5&+ :$6 3(5)250(' )25 7+( 0,1,080 (5525 5(&(,9(5 S R WR Wn R m 00$3 0LQ HUURU

PAGE 145

7R JHW D IHHO IRU WKH LPSURYHPHQW SURFHVV D KRUL]RQWDO OLQH RI WKH YLGHR VLJQDO WKH FRUUXSWHG YHUVLRQ RI LW DQG WKH HQKDQFHG YHUVLRQ RI WKH OLQH XVLQJ 00$3 UHFHLYHU DUH VKRZQ LQ )LJXUH 7KH HQKDQFHG VLJQDO ORRNV DOPRVW OLNH WKH UHIHUHQFH VLJQDO LQGLFDWLQJ WKDW WKH GHFRGLQJ SURFHVV LV TXLWH HIIHFWLYH Ef )LJXUH Df $ KRUL]RQWDO OLQH RI YLGHR VLJQDO Ef FRUUXSWHG YHUVLRQ RI WKH YLGHR OLQH DQG Ff WKH HQKDQFHG YHUVLRQ RI WKH OLQH

PAGE 146

)LJXUH f§FRQWLQXHG Ff (YHQWXDOO\ WKH LPSURYHPHQW SURFHVV ZDV MXGJHG EDVHG RQ WKH VXEMHFWLYH TXDOLW\ RI WKH LPSURYHG SLFWXUHV 7KH HQFRGHG YHUVLRQ RI /(11$ SLFWXUH LV JLYHQ LQ )LJXUHV 7KH FRUUXSWHG SLFWXUHV WKH HQKDQFHG YHUVLRQ RI WKHP XVLQJ 6D\RRG DQG %RUNHQKDJHQ PHWKRG >@ WKH 00$3 DQG 0615 UHFHLYHUV IRU FKDQQHO HUURU UDWHV RI S DQG S O DUH VKRZQ )LJXUHV f $Q H[DPLQDWLRQ RI WKH SLFWXUHV VKRZV WKDW WKH WKUHH HQKDQFHPHQW PHWKRGV DUH HIIHFWLYH )XUWKHU PRUH LW LV REVHUYHG WKDW 00$3 DQG 0615 DUH FHUWDLQO\ PRUH HIIHFWLYH LQ LPSURYLQJ WKDQ WKH PHWKRG JLYHQ LQ >@ $PRQJ RXU WZR UHFHLYHUV 0615 VHHPV WR ZRUN EHWWHU 7KH 0615 GRHV D JRRG MRE LQ UHPRYLQJ WKH KRUL]RQWDO VWUHDNV ZKLFK DUH YHU\ XQSOHDVDQW WR D YLHZHU 7KH VXEMHFWLYH H[DPLQDWLRQ RI SLFWXUHV VHHPV WR DJUHH ZHOO ZLWK FRQFOXVLRQV GHULYHG IURP WKH 615 JUDSKV

PAGE 147

)LJXUH 7KH '3&0 HQFRGHG YHUVLRQ RI /(11$ LQ WKH DEVHQFH RI FKDQQHO HUURUV )LJXUH 7KH HIIHFWLYHQHVV RI WKH SLFWXUH HQKDQFHPHQW SURFHVV LQ D UHODWLYHO\ QRLV\ VLWXDWLRQ S f LV GHPRQVWUDWHG Df &RUUXSWHG SLFWXUH Ef HQKDQFHG SLFWXUH XVLQJ PHWKRG LQ >@ Ff HQKDQFHG SLFWXUH XVLQJ 00$3 UHFHLYHU DQG Gf HQKDQFHG SLFWXUH XVLQJ 0615 UHFHLYHU

PAGE 148

Ff )LJXUH f§FRQWLQXHG

PAGE 149

Gf )LJXUH f§FRQWLQXHG Df )LJXUH 7KH HIIHFWLYHQHVV RI WKH SLFWXUH HQKDQFHPHQW SURFHVV LQ D YHU\ QRLV\ VLWXDWLRQ S Of LV GHPRQVWUDWHG Df &RUUXSWHG SLFWXUH Ef HQKDQFHG SLFWXUH XVLQJ PHWKRG LQ >@ DQG Ff HQKDQFHG SLFWXUH XVLQJ 0615 UHFHLYHU

PAGE 150

Ef )LJXUH FRQWLQXHG Ff

PAGE 151

6LGH ,QIRUPDWLRQ 7KH WUDQVPLWWHU PXVW FRPSXWH DQG WUDQVPLW KLJKHU GLPHQVLRQDO VWDWLVWLFV RI WKH SLFWXUH WRJHWKHU ZLWK WKH SLFWXUH 7KH DPRXQW RI VLGH LQIRUPDWLRQ GHSHQGV XSRQ V\VWHP SDUDPHWHUV &RQVLGHU D '3&0 V\VWHP ZLWK WKH IROORZLQJ SDUDPHWHUV 1XPEHU RI SL[HOV 1[1 Q OHQJWK RI '3&0 ZRUGV T QXPEHU RI TXDQWL]HU OHYHOV P GLJLWV RI DFFXUDF\ LQ VWDWLVWLFV S RUGHU RI 0DUNRY PRGHO 7KH LQFUHDVH LQ ELW UDWH LV HTXDO WR 5 JS ORJ QOf f 7KH SHUFHQWDJH RI RYHUKHDG LV WKHUHIRUH JLYHQ E\ b RYHUKHDG JSORJ POf B P TS 1[1[Q P Q f 7KH HIIHFW RI QRLV\ VWDWLVWLFV RQ WKH SHUIRUPDQFH RI WKH V\VWHP ZDV VWXGLHG ,W WXUQV RXW WKDW WKH V\VWHP SHUIRUPDQFH GHWHULRUDWHV UDSLGO\ DV WKH FKDQQHO QRLVH LQFUHDVHV 7KHUHIRUH RQH PXVW SURWHFW WKH FRQGLWLRQDO VWDWLVWLFV DJDLQVW FKDQQHO QRLVH 7R UHDFK WKH SHUIRUPDQFH WKDW LV DFKLHYDEOH E\ XWLOL]LQJ QRLVHIUHH VWDWLVWLFV WKH IROORZLQJ FRGLQJ VWUDWHJ\

PAGE 152

ZDV IRXQG HPSLULFDOO\ (DFK HOHPHQW RI VWDWLVWLFV LV FRGHG LQWR D ORJPOf ELW QDWXUDO ELQDU\ FRGH 7KH ILUVW ORJPOff ELWV DUH FRGHG XVLQJ D f UHSHDW FRGH DQG WKH UHVW DUH OHIW XQFRGHG 7KXV WKH SHUFHQWDJH RI VLGH LQIRUPDWLRQ LV GHWHUPLQHG E\ b RYHUKHDG $ f 1 Q 7KH DPRXQW RI RYHUKHDG IRU D [ SL[HO SLFWXUH D VHYHQ OHYHO TXDQWL]HU DQG IRXU GHFLPDO GLJLWV RI DFFXUDF\ LV JLYHQ LQ 7DEOH (YHQ ZLWK D WKLUG RUGHU 0DUNRY PRGHO WKH DPRXQW RI VLGH LQIRUPDWLRQ GRHV QRW H[FHHG WZHQW\ RQH SHUFHQW RI WKH WRWDO SLFWXUH LQIRUPDWLRQ 7$%/( 3(5&(17$*( 29(5+($' $6 $ )81&7,21 2) 25'(5 2) 0$5.29 6285&( 02'(/ 86(' ,1 7+( 6<67(0 RUGHU RI 0DUNRY VRXUFH SHUFHQWDJH RYHUKHDG ILUVW m b VHFRQG b WKLUG b 6XPPDU\ $Q RSWLPXP PHWKRG IRU GHFRGLQJ FRUUHODWHG VHTXHQFHV ZLWK WKH DLG RI 0DUNRY VRXUFH PRGHOLQJ ZDV GHULYHG DQG LW ZDV VKRZQ WR UHTXLUH WUHH VHDUFKLQJ $ VXERSWLPXP LPSOHPHQWDWLRQ RI WKH

PAGE 153

PHWKRG WKDW PDNHV XVH RI 0/f DOJRULWKP ZDV DOVR SUHVHQWHG 7KHQ D PRGLILHG 0$3 PHWKRG IRU 0DUNRY VRXUFHV LV GHVFULEHG 7KH 00$3 LV LQ IDFW D VSHFLDO FDVH RI WUHH GHFRGLQJ ZKHUH QR VHDUFK LV SHUIRUPHG $ PD[LPDO VLJQDOWRQRLVH 0615f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

PAGE 154

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f UHFHLYHU DQG D PD[LPDO 615 0615f UHFHLYHU ZHUH GHULYHG IRU GHFRGLQJ RI FRUUHODWHG VRXUFHV ([SHULPHQWDO UHVXOWV VKRZHG WKH HIIHFWLYHQHVV RI WKH PHWKRG )XUWKHU PRUH WKH\ ZHUH IRXQG WR EH VXSHULRU WR D UHFHQWO\ UHSRUWHG WHFKQLTXH

PAGE 155

6LJQDO 6HOHFWLRQ IRU 1RQVYPPHWULF 6RXUFHV *UDGLHQW EDVHG VLJQDO VHOHFWLRQ PHWKRGV IRU QRQV\PPHWULF VRXUFHV WKDW PLQLPL]H WKH V\PERO HUURU UDWH VXEMHFW WR D SHDN RU DQ DYHUDJH SRZHU FRQVWUDLQW ZHUH GHYHORSHG 7KH JUDGLHQW DQG FRQMXJDWH JUDGLHQW YHUVLRQV RI WKH DOJRULWKPV ZHUH FRPSDUHG DQG WKH FRQMXJDWH JUDGLHQW YHUVLRQ ZDV IRXQG WR EH PXFK IDVWHU 7KH JHQHUDOL]HG YHUVLRQ RI WKH DOJRULWKP WKDW ILQGV D VLJQDO VHW WKDW PLQLPL]HV D FRVW IXQFWLRQ QRW QHFHVVDULO\ WKH V\PERO HUURU UDWHf XQGHU D SRZHU FRQVWUDLQW ZDV DOVR DGGUHVVHG ,W ZDV VHHQ WKDW WKH QRQV\PPHWULF VLJQDO GHVLJQ JDLQV VLJQLILFDQW LPSURYHPHQWV RYHU WKH HJXDOO\ OLNHO\ VLJQDO VHOHFWLRQ ZKHQ WKH FKDQQHO LV QRLV\ 7KH PLQLPXP FRVW VLJQDO GHVLJQ DOVR GHPRQVWUDWHG SHUIRUPDQFH LPSURYHPHQW UHODWLYH WR PLQLPXP HUURU VLJQDO VHOHFWLRQ RYHU QRLV\ FKDQQHOV $OSKDEHW 6L]H 6HOHFWLRQ IRU 1RQVYPPHWULF 6RXUFHV (YHQ WKRXJK HIILFLHQW WHFKQLTXHV IRU VRXUFH FRGLQJ FKDQQHO FRGLQJ DQG VLJQDO GHVLJQ H[LVWV LW LV QRW NQRZQ KRZ WKH FKRLFH RI DOSKDEHW VL]H DIIHFWV D FRPPXQLFDWLRQ V\VWHP 7R LQYHVWLJDWH WKH UROH RI DOSKDEHW VL]H LQ D FRPPXQLFDWLRQ V\VWHP IRU QRQV\PPHWULF LQIRUPDWLRQ VRXUFHV FRPPXQLFDWLRQ V\VWHPV ZLWK YDULRXV DOSKDEHW VL]HV IRU WUDQVPLVVLRQ RI YLGHR VLJQDOV RQ WKH EDVLV RI HTXDO LQIRUPDWLRQ UDWH EDQGZLGWK DQG

PAGE 156

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f DOJRULWKP LV GLVFXVVHG $ PRGLILHG 0$3 00$3f UHFHLYHU IRU 0DUNRY VHTXHQFHV DQG D PD[LPDO VLJQDOWRQRLVH UDWLR 0615f UHFHLYHU IRU '3&0 VLJQDOV ZHUH DOVR GHULYHG 7KH PHWKRGV ZHUH DSSOLHG WR WKH SLFWXUH WUDQVPLVVLRQ SUREOHP &RQVLGHUDEOH DPRXQW RI LPSURYHPHQW IURP ERWK TXDOLWDWLYH DQG TXDQWLWDWLYH SRLQW RI YLHZf ZDV REWDLQHG 7KH PD[LPXP 615 UHFHLYHU ZDV IRXQG WR EH VXSHULRU WR WKH RWKHU PHWKRGV

PAGE 157

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

PAGE 158

$33(1',; $ (9$/8$7,21 2) $1 ,17(*5$/ :H ZRXOG OLNH WR HYDOXDWH WKH IROORZLQJ LQWHJUDO f¯ [Df I[ H r G[ R /HW XV PDNH WKH IROORZLQJ FKDQJH RI YDULDEOH $Of X [D $f \ 7KHUHIRUH [ DQG G[ FDQ EH H[SUHVVHG LQ WHUPV RI X DQG GX [ \D XD $rf G[ D GX 8SRQ VXEVWLWXWLQJ YDOXHV IRU [ DQG G[ LQWR WKH LQWHJUDO ZH JHW f¯ [Df f¯ -[ H G[ \RD?RXf Hf¯ GX $f B D Y-R 7KH LQWHJUDO FDQ EH EURNHQ LQWR WKH IROORZLQJ WZR LQWHJUDOV

PAGE 159

M YRDYRXf Hf¯ GX R X H D D YR Y-R GX Y D D H D YR GX $f :LWK D VLPSOH YDULDEOH FKDQJH WKH ILUVW LQWHJUDO FDQ EH FRPSXWHG X HaXr GX LHY GY sH RL B D Dr R $f DQG E\ GHILQLWLRQ WKH VHFRQG LQWHJUDO LV fµ2 GX tHUI af r 7KXV $f [Df R [ H B D G[ R H r 9r D R HUIf§f§ rrf R $ f

PAGE 160

$33(1',; % (9$/8$7,21 2) 7+( '(5,9$7,9(6 'LIIHUHQWLDWLQJ HTXDWLRQV f DQG f ZLWK UHVSHFW WR ;M DQG \M ZH JHW WKH IROORZLQJ V\VWHP RI HTXDWLRQV $ $ & & G[ ML G[ % G[ G\ s% G[ D G[M ' G[ G\ Gfµ MWL OOVVM 1 ,[;M e 8LNa/MNf \\Nf [[Nf OVVW MWL 1 ‘H[S> 1Q ‘`SN % f [Y ;a;Nf U c66MX\\L!e Bf§f§fH[3>f§ N 1 % [;Lf < /LN/MNf \\Nf U ,OV6MWO N O 1 f H[S >fµ ,3r e /LNa/Mrf 8[rfH[S VV N ‘@SN % f 1 \\Lfe ALN/MNf NO [[Nf \\ffH[S>BA@SA OOVV-,

PAGE 161

F e rrr! /¯N/MNf H[S >A @ SN %rf -FrO f¯R DQG e e LUWLWf \\MfH[S>f§ 6MFL SA r LYR % f 6LQFH WKH ULJKW KDQG VLGH RI WKH HTXDWLRQV DUH ]HUR WKH VROXWLRQ WR WKH VHW RI HTXDWLRQV %Of LV G[ML G[G[LW A r G
PAGE 162

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

PAGE 163

>@ 6D\RRG . DQG %RUNHQKDJHQ & 8VH RI 5HVLGXDO 5HGXQGDQF\ ,Q 7KH 'HVLJQ RI -RLQW 6RXUFH&KDQQHO &RGHUV ,((( 7UDQV&RPPXQ 9RO 1R SS >@ 2n1HDO % 3UHGLFWLYH 4XDQWL]LQJ 6\VWHPV 'LIIHUHQWLDO 3XOVH FRGH 0RGXODWLRQf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t 6RQV ,QF 1HZ @ %HUOHNDPS ( 5 $OJHEUDLF &RGLQJ 7KHRU\ 0F*UDZ+LOO %RRN &RPSDQ\,QF 1HZ
PAGE 164

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

PAGE 165

>@ 0HOVD / DQG &RKQ ' 'HFLVLRQ DQG (VWLPDWLRQ 7KHRU\ 0H *UDZ+LOO %RRN &RPSDQ\ 1HZ @ 0RGHVWLQR : DQG 'DXW ' * &RPELQHG 6RXUFH &KDQQHO &RGLQJ RI ,PDJHV ,((( 7UDQV &RPPXQ 9RO &20 SS

PAGE 166

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

PAGE 167

, FHUWLI\ WKDW , KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 6FRWW / 0LOOHU &KDLU $VVRFLDWH 3URIHVVRU RI (OHFWULFDO (QJLQHHULQJ , FHUWLI\ WKDW , KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RIn VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ c$f A$LXV/ /HRQ &RXFK 3URIHVVRU RI (OHFWULFDO (QJLQHHULQJ , FHUWLI\ WKDW , KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ t!L -£ 'RQDOG * &KLOGHUV 3URIHVVRU RI (OHFWULFDO (QJLQHHULQJ , FHUWLI\ WKDW , KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRM n)D]LO 1DMDIL Y $VVRFLDWH 3URIHVVRU RI &LYLO (QJLQHHULQJ

PAGE 168

, FHUWLI\ WKDW , KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWDEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ AMQf A.HUPLW 6LJPRQ $VVRFLDWH 3URIHVVRU RI 0DWKHPDWLFV 7KLV GLVVHUWDWLRQ ZDV VXEPLWWHG WR WKH *UDGXDWH )DFXOW\ RI WKH &ROOHJH RI (QJLQHHULQJ DQG WR WKH *UDGXDWH 6FKRRO DQG ZDV DFFHSWHG DV SDUWLDO IXOILOOPHQW RI WKH UHTXLUHPHQWV IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ $XJXVW :LQIUHG 0 3KLOOLSV 'HDQ &ROOHJH RI (QJLQHHULQJ 'HDQ *UDGXDWH 6FKRRO


82
Pii ^ Jy -Pni
n=i
n*i
(4.1)
Notice that Pj¡ depends on the modulation scheme. Here a two
dimensional signal constellation is utilized. The channel
matrix is
P= tPij]
(4.2)
N
Figure 4.1. The Model of the communication channel,
where Pj¡ is given by
Pa =
exp [--]
n iSj-Sjl
4 Nn
(4.3)
It is also obvious that
Pji = Pij
Izizq
(4.4)


50
a= (x-AH sinP3) 2+y2
b= (x+BH sinal) 2+y2
c= (x-CH sin (Pi+a2+P2) ) 2+ (y-AH cosP 1+CH cos (Pi+a2+P2) ) 2
x=p1AH sin$1-p2BH sina 1+p3C/ sin (Px+oc2+P2)
y=p2AH cosP1-p3 CH cos (P1+a2+P2) .
Now the problem is to minimize a function of nine
variables under seven equality constraints. The method of
Lagrange multipliers can be applied to this problem. We need
to define a new cost function F of sixteen variables and take
the partial derivatives with respect to the sixteen variables
and set them equal to zero.
cost function=F(alt a2, a3, px, P2, P3, AH, BH, CH, X2, X3, X4, X5, X6, A
If we eliminate X/s among the sixteen equations we end
up with a nonlinear set of equations of the other nine
variables which must be solved numerically.
This method becomes quite complicated as the number of
the signals in the constellation grows. It was mainly used to
verify the solutions obtained by the gradient based method
when N was relatively small.


103
consideration perforin. The expressions plotted are the word
error probability.
When the bandwidth constraint is strict, the ternary
alphabet is seen to be superior to the binary system (Figure
4.6(a)). It is also seen that the systems employing the
alphabets of size four, five and seven are more efficient than
the binary system above a certain coding threshold (Figures
4.6(b)-4.6(d)). Moreover, the coding threshold seems to be
larger for larger alphabets sizes. The nonbinary systems are
expected to outperform the binary system since they utilize
error correction codes and the binary system does not.
Nonbinary systems were also compared to each other. The
results indicate that the ternary system is superior to the
other ones Figure 4.8(a) indicates that ternary system
outperforms the 7-ary system but the improvements decrease
slowly as the signal-to-noise ratio increases. 7-ary system
outperforms systems employing alphabets of size four and five
above reasonable thresholds. The comparison between the
systems with 7-ary and 4-ary alphabets is shown in Figure
4.8(b).
For the other case when the bandwidth constraint can be
somewhat relaxed the situation is different. There is always
a threshold above which the nonbinary is superior to the
binary system. For a twenty four percent bandwidth expansion
the ternary system is still superior to the binary system
above a threshold but the improvements over the binary system


44
J^=-£k JT e"(u_/^r>a(u(i+tan2a2)e-wa)
2 u-0
du
_ -(u- /aW2)2
^fe V^0 (-u(l+tan2(a1+a2)e-uW(l+a2)) du
U-0
T J
u-0
(-u(l+tan2 (a1+a2)) e-uW(-1+2>+u(l+tan2a2) e-uW2 da) =0.
(3.24)
Inspection shows that above equations consist of only one type
of integral which can be expressed in a computable form,
fu e
(u-k)2e-u2l
g ((l*i) u2*k2-2uk)
du
(3.25)
By producing a perfect square in the exponent
J
g-((l+J)u2+k2-2uk)
du= e
k2l
(1*1)
/
-(1*1)(u-
(1+i)
du
(3.26)
and using the results obtained in Appendix A we get:


123
However, if some redundancy is present in the encoded
signal a more reasonable criteria can be formulated. Assuming
that the data has a memory of M and b¡.,,, bM., are the
previous decisions, the modified MAP rule will be, set
=ak if
P(-rK=a*> P(vi=ak\vi-i=bi-i' vi-M=bi-n) >
p(ri|vi=a.j)p(vi=aj|vi_1=.bi_lf . ., v^b^)
(5.10)
for j=l, 2, . ,Q and j?k.
The symbol-by-symbol MMAP method is in fact a special
case of tree decoding procedure for Markov sequences developed
earlier where no search is performed and the past symbol
decisions are used as an aid for the current symbol decision.
5.4 A Minimum Cost Decoder
In a video transmission system some errors might be more
costly than others. Therefore instead of minimizing the usual
error rate a different criteria must be formed and minimized.
A Bayes receiver minimizes the average cost, however, to
implement Bayes receiver one must know the source statistics
and must be willing to specify the cost associated with each
decision [36].
Let the transmitted and received symbols be from the same


CHAPTER FOUR
ALPHABET SIZE SELECTION FOR VIDEO SIGNAL CODING
4.1 Introduction
Previously the methods of signal selection for
nonsymmetric source were studied. Here we are going to utilize
those methods in the deign of an image transmission system.
An image transmission system typically consists of three
parts, a source coder, a channel coder and a modulator.
Source coding is the first step in the transfer of
information from a source. The purpose of source coding is to
remove as much redundancy as possible from the source. DPCM of
video signals with as few as seven quantization levels has
been shown to produce pictures virtually indistinguishable
from the original under standard viewing conditions [21].
Since the output of DPCM system is a nonsymmetric source, an
entropy coder must be used to take advantage of that
redundancy. The well known Shannon-Fano method and the Huffman
procedure are examples of entropy coding techniques [22].
In channel coding the goal is to correct errors
introduced in the channel by inserting redundancy in the data.
Channel coding is accomplished by either decreasing the
79


102
Table 4.9continued
before encoding
after encoding
.5
.3214
. 166
. 1543
. 166
. 1543
. 064
.1034 .
.064
. 1033
.02
.0817
CM
O

.0816
(d)


55
>1
92
VF=
?N
where
9k=(9k'9k)
(3.53)
gk is obtained by taking the derivative of Pe with respect
to sk
(Pjc+P)
Sk~Si rexp [--|s*-Sil
i*k
4Wn
] [
-i-1
l^-sj2 2N0
(3.54)
Similarly one can find the kth component of VF for a MAP
receiver
i*k
P*-Zs*-*iexP["
t . ta t2 .
] [-4+ T-5-]
&N0\sk-sA2 2Njsk-Si
i r +
4JV0|sJt-si|a t| 2i\y|sJt-sJ2
+Pi-VSiexPt'
(3.55)
where


148
6.1.1 Signal Selection for Nonsvmmetric Sources
Gradient based signal selection methods for nonsymmetric
sources that minimize the symbol error rate subject to a peak
or an average power constraint were developed. The gradient
and conjugate gradient versions of the algorithms were
compared and the conjugate gradient version was found to be
much faster. The generalized version of the algorithm that
finds a signal set that minimizes a cost function (not
necessarily the symbol error rate) under a power constraint
was also addressed. It was seen that the nonsymmetric signal
design gains significant improvements over the egually likely
signal selection when the channel is noisy. The minimum cost
signal design also demonstrated performance improvement
relative to minimum error signal selection over noisy
channels.
6.1.2 Alphabet Size Selection for Nonsvmmetric Sources
Even though efficient techniques for source coding,
channel coding and signal design exists, it is not known how
the choice of alphabet size affects a communication system. To
investigate the role of alphabet size in a communication
system for nonsymmetric information sources, communication
systems with various alphabet sizes for transmission of video
signals on the basis of equal information rate, bandwidth and


20
1.10 Description of Chapters
A review of background material relevant to this
dissertation is given in Chapter One. Two dimensional
modulation techniques, source coding techniques for images,
standard error correcting techniques and the effect of channel
errors on predictive system are among the topics addressed in
this chapter. In Chapter Two a DPCM system will be analyzed
and a model for the statistics of the source will be derived.
It will be shown theoretically and empirically that DPCM of
video signals produces nonsymmetric sources.
The issues of signal design are addressed in Chapter
Three. Algorithmic solutions to signal design for nonsymmetric
information sources under average and peak power constraints
for minimizing the error rate and average cost are presented.
The study on the role of alphabet size for nonsymmetric
sources in a communication system is given in Chapter Four. In
Chapter Five various methods for decoding Markov sequences are
presented. The application to the transmission of video
signals over noisy channels and a comparison of the method is
also given. Chapter Six contains a summary of presented
approaches, conclusions and comments regarding the future
research directions.


157
[24] Massey, J. L.,"Shift Register Synthesis and BCH
Decoding," IEEE Transactions on Information Theory.
Vol. IT-15, pp. 122-127, 1969.
[25] Chien, R. T., "Cyclic Decoding Procedures for the Bose-
Chaudhuri-Hocquenghem codes," IEEE Transactions on
Information Theory. Vol. IT-10, pp. 357-363, 1964.
[26] Huffman, D.A., "A method for the Construction of
Minimum Redundancy Codes," Proceedings of the I.R.E..
Vol. 40, pp. 1098-1101, 1952.
[27] Lippmann, R., "Influence of Channel Errors on DPCM
Picture Coding," ACTA Electron.. Vol. 19, No. 4, pp. 289-
294, 1976.
[28] Stammnitz, P., "Error Protection of 34 Mbits/s DPCM
Encoded TV Signals with Multiple Error Correcting BCH
Codes," Proc. 1980 Zurich Sem. Digital Commun.. 1980.
[29] Lippman, R.,"A Technique for Channel Error Correction
in Differential PCM Picture Transmission," Proc. Int
Conf. Commun.. Seattle, WA, 1973.
[30] Fenwick, D. M., Steele, R., and Vasanji, N., "Partial
Correction of DPCM Video Signals Using a Walsh
Corrector," Radio Electron Eng.. Vol. 48, pp. 271-276,
1978.
[31] Ngan, K. N., and Steel, R., "Enhancement of PCM and DPCM
Images Corrupted By Transmission Errors," IEEE Trans.
Commun.. Vol. COM-30, pp. 257-264, 1982.
[32] Steele, R., Jayant, N. S. and Schmidt, C. E,
"Statistical Block Protection Code for DPCM-Coded
Speech," Bell System Tech. J.. Vol. 58, No. 7, pp. 1647-
1657, 1979.
[33] Moore, C. C. and Gibson, J. D., "Self-Orthogonal
Convolutional coding for DPCM-AQB Speech Encoder,"
IEEE Trans. Commun.. Vol. COM-32, pp. 980-982, 1984.
[34] Anderson, J. B. and Bodie, J. B., "Tree Encoding of
Speech," IEEE Trans. Inform. Theory. Vol. IT-21, pp.
379-387, 1975.
[35] Modestino, J. W., Bhaskaran, V. and Anderson, J. B.,
"Tree Encoding of Images in the Presence of Channel
Errors," IEEE Trans. Inform. Theory, Vol. IT-27, pp.
677-697, 1981.


81
The BCH codes are chosen for channel coding (If extra
bandwidth is available) because codes close to any rate can be
found. A q-ary BCH code is denoted by (n,kq,tq) where kq
represents the number of information symbols, n represents the
bolck size and tq represents the the error correction
capability of the code.
The receivers are coherent in-phase/quadrature detectors
(except for BPSK where only the i'n-phase branch is needed) and
perfect carrier and symbol synchronization is assumed.
The code words for the nonbinary systems are not equally
likely. A minimum error procedure is the optimum way of
decoding at the receiver but a maximum likelihood procedure
can be used as a sub-optimal procedure. The Berlekamp-Massey
procedure will be used for decoding of the received data [23]-
[25].
4.2.2 Communication Channel
The communication channel is modeled as a q-ary
independent error channel and the source alphabet and channel
output alphabet are the same. If a symbol s¡ is transmitted
there is a probability of pj¡ that Sj is received (ip^j) and a
probability of pki that sk is received (i?*k), and so on and so
forth. Therefore, the probability that s¡ is received correctly
is


APPENDIX A
EVALUATION OF AN INTEGRAL
We would like to evaluate the following integral:
(x-a)2
fx e 22 dx.
o
Let us make the following change of variable:
(A.l)
u=
x-a
(A.2)
y/2 0
Therefore x and dx can be expressed in terms of u and du:
x=y/2a u+a (A*3)
dx=/2a du.
Upon substituting values for x and dx into the integral we get
(x-a)2
Jx e 202 dx= J y/2o(a+\/2ou) e"2 du. (A.4)
0 a
v/Jo
The integral can be broken into the following two integrals:
151


58
Si (k+1)
sf~M
Sz (k+1)
Sk*i~
yfWM
s(k+l)
y/TTR
to meet the peak power constraint.
(3.63)
3.5 Minimum Average Cost Signal Selection
The communication channels are almost never error free.
The probability of error (error rate) is a measure of system
\
performance. Various error types are usually weighted equally
but in some communication system certain errors are more
costly than the others. The Bayes receiver allows us to rank
the different error types [19]. To utilize the Bayes receiver,
one must know the source statistics and a reasonable estimate
of a cost matrix must be obtained. The Bayes receiver requires
more apriori knowledge about the communication system than
others but it results in a superior performance if there is no
mismatch between the design and operating conditions.
The goal is to find a signal constellation that minimizes
the average cost subject to a peak or average power
constraint. First a workable expression for the average cost
will be obtained.


104
are not as much as the fixed bandwidth case (Figure 4.5(a)).
For larger bandwidth expansion factors the coding threshold
becomes large. For very large bandwidth expansion factors the
binary system is superior to ternary systems.
The binary system turns out be superior to larger
alphabet systems (4-ary, 5-ary and 7-ary) with a little
bandwidth expansion below a threshold. Figure 4.7(b)
demonstrates that the binary system is superior to system with
4-ary alphabet below a certain threshold. Notice that the
threshold for these systems are much larger than the threshold
value for ternary system. Further more, the threshold value
increases as either the bandwidth expansion factor or the
alphabet size go up.
Up until here we have only examined the graphs for the
medium detail picture. The results for low and high detail
picture are given in Figures 4.9-4.12. A comparison between
the graphs obtained for different pictures indicate that there
is a slight variability among the graphs which has to do with
the amount of detail in each picture. However, the
relationship between the system performance and the alphabet
size does not vary from picture to picture.
4.7 summary
Communication systems employing alphabets of different
sizes are compared for transmission of video signals under the


126
N
dj=]C (yk-yj)2 p(rAvi=ak') p(vi=ak\vi-i=b
(5.15)
The given proof is only applicable to first order DPCM
systems. Let us give a more general proof.
It has been shown that the signal-to-noise ratio of a
DPCM system is given by [37]
where a,2 is the variance of the signal, ae2 is the variance of
the input to the quantizer and
(5.17)
in which eq represents the mean-squared quantization noise, em
is mutual error term and ec represents noise due to channel
noise. The only term in the SNR expression that depends on the
choice of receiver is ec ec is given by
(5.18)
where D(z1(z2) is the discrete system transfer function and


154
c=£ <*-**> Jc*l -o
and
£=£ (irtit) (y-yj,)exp[- Sjci 3p^.
*-1 ivo
(B. 5)
Since the right hand side of the equations are zero, the
solution to the set of equations (B.l) is
dxji
dxJ
=0
dxit
-^=,
dYj
(B 6)
and
OX,
dy
dVj
*=0
(B.7)


88
TABLE 4.1. DIFFERENCE SYMBOL PROBABILITIES.
Picture
P(Yo)
p (y+i)=
p(y.i)
p(y+2) =
p(y.2)
p(y+3) =
p(y.3)
Low
Detail
. 674
. 142
.018
.003
Medium
detail
.584
.172
. 032
.004
High
Detail
.5
. 166
.064
. 02
TABLE 4.2. PRIMITIVE IRREDUCIBLE POLYNOMIALS OVER NONBINARY
FIELDS.
Field
Polynomial
GF (2)
x2 +x +1
GF (3)
x3 +2x +1
GF (4)
x2 +x +A
GF (5)
x2 +x +2
GF (7)
x2 +x +3


7
prediction. The expansion of this equation gives the following
condition for optimum aj,
N
£ a. R^ik-j) =Rxx(k) ; k=l,2,...,N (1.19)
j-i
or
'^(0) 1^(1) ***(2) RjocIN- 1)'
(a \
ai
RjociO) R^d) R^iN-2)
a2
=
^(2)
rJW-1) R^iN-2) R^N- 3) Rjcx'i 0) ,
aN,
***(*>,
(1.20)
in matrix notation,
Rxxa=r*x (1-21)
where
rxx={Rxxd)): N (1.22)
The equations are called normal equations, Yule-Walker
prediction equations or Wiener-Hopf equations.
The mean squared error decreases significantly by using
up to three elements in predictive coding. However, if the
coefficients are not exactly matched to the statistics of a
picture, the decrease in mean squared error is not significant
by using three previous elements as compared to one [1].


8
1.5 Transform Coding
In transform coding a picture is divided into subpictures
and then each of these subpictures are transformed into a set
of independent coefficients. The coefficients are then
quantized and coded for the transmission. An inverse
transformation is applied to recover intensities of picture
element. Much of the compression is a result of dropping
coefficients from transmission that are small and coarsely
quantizing the others as required by the picture quality.
It is desirable to have a transform which compacts most
of the image energy in as few coefficients as possible.
Another consideration is the ease of implementation.
1.5.1 Optimum Transform
Optimum transform (K-L transform) is explicitly known,
but computationally it is very demanding. This undesirable
feature has prevented any hardware implementation of the
optimum transform. It is mainly studied in simulations to
obtain bounds.
The most practical transform coding techniques are based
on the DCT (discrete cosine transform), which provides a good
compromise between information packing ability and
computational complexity. In fact, the properties of DCT have
proved of such practical value that it has become the


68
TABLE 3.1. VALUES OF SIGNAL-TO-NOISE RATIO AND THE PROBABILITY
DISTRIBUTIONS OF NON-SYMMETRIC SOURCE.
Source Number
SNR(dB)
Probability Distribution
1
12.2
.8,.1 1
2
9
.9, .025,.025,.025,.025
3
8
.96,.01,.01,.01,.01
4
15.5
.674,.142,.142,.018,.018,.003,.003
TABLE 3.2. SOURCE STATISTICS FOR THE FOUR SIGNAL SOURCE.
source statistics
.35,.35,.15,.15


94
4.5 Nonbinarv BCH Codes
The nonbinary BCH codes are a special class of cyclic
codes that can be constructed for any symbol alphabet defined
on a finite field, say GF(q), which can be a prime field or
some extension of a prime field. As a generalization of binary
case a t-error correcting BCH code on GF(q) is a cyclic code
, and all the codewords have roots that include 2t consecutive
powers of some element j3 contained in GF(qm) an extension
field of GF(q). It will be convenient to distinguish between
the two fields by calling the GF(q) the symbol field and
GF(qm) the locator field. As with the binary codes, BCH codes
on GF(q) can be primitive or nonprimitive, depending on
whether a primitive or nonprimitive element of GF(qm) is used
to specify the connective roots of codewords.
A t-error correcting code may have either odd or even
minimum design distance, given by d=2t+l or d=2t+2,
respectively. The generator polynomial of a BCH code on GF(q)
is defined as the least common multiple of minimal polynomial
of amo, amo+1,..., amo+d-2 (the sequence of powers can begin with
any arbitrary power a)
g (x) =LCM [ao (x) Bn0i-i (x) ] (4.14)
The block length of the code is the order of the element


76
chanrxl SNR
Figure 3.9. Performance results for non-symmetric signal
selection (the solid curve) and the equally likely signal
design (the dashed curve) over a wide range of channel signal-
to-noise ratio.


APPENDIX B
EVALUATION OF THE DERIVATIVES
Differentiating equations (5.38) and (5.39) with respect
to Xj and yj we get the following system of equations
A-
A-
C
C
dx
ji
dx4
+B-
dx.
dy
+B-
dx
a
dxj
+D
dx
dy
dYji
dxj
dVli-
0
dyj=0
pi-0
dxj
dyJA-0
dy*
(B.l)
where
N
=£ (Lik~Ljk) (y-y*) exp [-
jt-i
lls-sj2
N
-Ix-Xj £ Uik~Ljk)
2 (y-yk) (x-xk) ls-st
jt-i
N
exp[--
Nn
}pk (B. 2)
, x-v , 2 (X~Xk) 2 r ¡S-SjuJ2 ,
-(y-yi>£ (1_)exP[-
k-1
N
B=(x-Xi) Y; (Lik-Ljk) (1-
2 (y-yk)2, r Ils-Sjtl2
k=l
N
) exp [-
IP*
"£ (Lik~Lj*) U-x*)exp(-
s-s,
k-1
]pk
(B. 3)
N
+ (y-yi)£ ^ik-Ljk)
k-l
2(x-xk) (y-y.),)exp[_,.-^",]p^
lls-sJI2
153


Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy
EFFICIENT COMMUNICATIONS FOR NONSYMMETRIC INFORMATION
SOURCES WITH APPLICATION TO PICTURE TRANSMISSION
By
Shahriar Emami
August 1993
Chairperson: Dr. Scott L. Miller
Major Department: Electrical Engineering
This dissertation is concerned with issues related to
nonsymmetric information sources. Signal design, alphabet size
selection and decoding of information from these sources are
among the topics covered in this dissertation. Although the
techniques presented here are applicable to any nonsymmetric
source, the emphasis is placed on video sources. Initially a
model for the statistics of DPCM (Differential Pulse Code
Modulation) of video signals is derived and it is shown that
DPCM of video signals results in a nonsymmetric source.
The problem of signal selection for nonsymmetric sources
in two dimensions is considered. Iterative methods for finding
the minimum error signal (and minimum cost) constellation
subject to an average (or a peak) power constraint are
presented.
vi


17
Let us assume the channel noise is uncorrelated and the
difference signal samples are statistically independent, then
variXj =E[Oi2] +2E[QiNi] +E[Ni2]'£fk2 (1.31)
k
The second term is called the mutual error and can be shown to
be approximately zero if the quantizer is near optimum. The
error power reduces to
varx'i) =E[Q¡] +ElNi2}'£fk2. (1.32)
k
The sum can be evaluated using an identity
ijt2=l+a2+a4+a6 +
k
1
1-a2
=Jb.
(1.33)
The expression for the reconstruction error variance then
becomes
var(Xf_) =var(Qi) +b var(N) (1.34)
We now define the following quantities to relate the quantizer
and noise variances to the differential signal variance,
var(Q) =kg var(E-),
var (N) =Kn var(E£) .
(1.35)
The DPCM prediction gain is also given by


CHAPTER FIVE
EFFICIENT DECODING OF CORRELATED SEQUENCES
5.1 Introduction
Differential pulse code modulation (DPCM) of digitized
pictures results in a stream of binary words which represent
the difference between the current pixel and the previous ones
along the same scan line. Communication channels are almost
never error free. A channel error results in a streak whose
termination may extend to the end of the current scan line
because in a DPCM system the channel errors propagate.
Visually this effect is very disturbing. As a result one must
deal with the errors introduced in the channel in the design
of a DPCM system.
Standard error correction techniques have been used to
combat the channel errors in DPCM systems. The use of error
correction codes such as BCH codes requires addition of
redundancy to the data stream. The redundancy turned out to be
twenty five percent for a bit error rate of 0.01 for a
reasonable picture quality [27],[28].
It is possible to utilize the natural redundancies that
are present in the video signals to enhance the received
signals. Lippmann [29] introduced a method for locating
116


36
solving signal design problem under a peak power constraint.
Previous investigations on signal design have focused on
signal selection for equally likely signals [15]-[17]. There
are some applications where the information source is non-
symmetric. A practical instance in which such a model proves
rewarding is in the transmission of video signals. It was
demonstrated in Chapter Two that DPCM of digitized video
signals results in a nonsymmetric source. In this case signals
should be mapped into a two dimensional signal constellation
in an optimum manner. In other words the goal is to determine
the signal constellation that minimizes the probability of
error (or a given cost function) in presence of additive white
Gaussian noise under an average power(or a peak power)
constraint, given N signals with unequal probabilities.
To illustrate the difficulty of direct approach we will
design a ML receiver for a three signal constellation. We will
also describe a numerical method that uses the Lagrange
multipliers method for optimization. These two methods are
appropriate for smaller signal sets.
Then a number of iterative algorithms are developed.
First a normalized conjugate gradient search algorithm and a
gradient search algorithm will be presented that can be
applied to signal sets of any size and with any probability
distribution. The methods presented here are applicable to the
design of both MAP (maximum a posteriori) and ML (maximum
likelihood) receivers. These methods are generalizations and


122
Because of two interesting features the (M,L) is favored
in practice. The outputs are released in a synchronous manner
and a fixed number of branches are visited at each depth. The
algorithm in one form or another has been used in speech and
image coding previously [34],[35].
Figure 5.1. A tree with branching factor of three is shown. At
each depth only three branches are saved.
5.3 A Modified MAP (MMAP) Receiver
A MAP receiver chooses <£=<3* if
p(ri\vi=ak) p(>p(ri|vi=aJ)p(v^a^) (5.9)
for j=l,2,...,Q and j?k.


64
(-^_1)E iLik-Ljk> (y-y*) exp [ -
OJCi k-l
Z,
|g-g*l
)pk
+ (x-xi)'^2 (Lik~Ljk) (~g~~~2 (y~y*) it1) exP t~ B
*-l oxi -iVo "o
= (-^-)E {Lik~Ljk) (x-xk)exp[--S ^k¡ ]pk
oxi k-l -w~
]pj,
Jr-1 axi "o
|g-g*|2 i
]p*
(3.83)
where
(3.84)
3V ^ | o_o |2
(-0)E ^Lik~hk) (y*-yjt> exp [ ----- ] p
+ (x-xi) E (-2 (y-y*) ^) exp [ Bs^Sjc!1 ] Pk
= (-^~1)E (Lik~Ljk) (x-xk)exp[-^-£A.]pk
(y-Vi) E (< 3r "2 (*~*Jc> > exp [- ,S*k>I ] p*
jc-i yj "o
(3.85)
where
z2-U-xt)({|)+(y-y¡[)(^l)
It can be shown that (refer to Appendix B)
(3.86)


99
the roots of a polynomial [25].
The next step is to find the error values. Berlekamp [23]
has derived a formula for the magnitude of errors if the
location of errors are known.
Eventually to correct the errors we will subtract the
error polynomial from the received polynomial.
4.6 Results
To encode the pictures, a DPCM encoder was utilized to
generate a stream of symbols. The Huffman optimal procedure
[26] was applied to the encoded pictures to produce ternary,
4-ary, 5-ary and 7-ary streams of symbols and the average
message length per symbol were calculated. Then ternary, 4-
ary, 5-ary and 7-ary BCH codes were used to channel encode the
image data. The code rates were selected to be as close as
possible to the quantity hb/hq. The q-ary BCH codes and their
rates for all the pictures are given in Table 4.8.
The symbol probabilities after channel encoding were
calculated. Table 4.9 shows the symbol probabilities before
and after encoding. It turns out that the probabilities still
follow the structure given by (4.13) but the output source is
not quite as asymmetric as the input source. After encoding,
the larger probabilities get slightly smaller and the smaller
probabilities get slightly bigger.
Figures 4.4-4.6 illustrate how the systems under


12
coding is not apparent yet.
Since the code corrects all single errors within a block
of 15 bits, the message error rate for the coded system will
be
Pc.m=E(^5) Pc,/(l-Pc.jb)15-Jc=1.94xl0-6 (1.27)
It is seen by comparing the error rates that the message error
rate has improved by a factor of 58 through the use of a block
code.
Most of the research on block codes has been concentrated
on a subclass of linear codes known as cyclic codes. A cyclic
code word, after any number of cyclic shifts, has the property
of remaining a valid code word from the original set of code
words. Cyclic codes are attractive because they can be easily
implemented with feedback shift registers. The decoding
methods are simple and efficient.
Examples of cyclic and related codes are BCH, Reed-
Solomon, Hamming, Reed-Muller, Golay, quadratic residue and
Goppa codes. The classes form overlapping sets so that a
particular code may be a BCH code and also a quadratic residue
code. Recent applications of these codes to digital
communication include a (31,15) Reed-Solomon code for joint
tactical information distribution system(JTIDS) and a
(127,112) BCH code for INTELSAT V system [4].


72
design relative to a equally likely signal selection, a first
order Gauss-Markov source with a correlation coefficient of
0.9 was synthesized. The output was encoded using a first
order DPCM encoder. A four signal constellation was designed
with the source statistics shown in Table 3.2. The performance
of the non-symmetric constellation was compared with that of
the well known equally likely constellation (for four equally
likely signals, the best constellation is formed by the
vertices of a square [16]). Figure 3.9 demonstrates a
comparison between the systems in terms of output signal-to-
noise ratio (output SNR). For large values of channel signal-
to-noise ratio (channel SNR) the two design procedures result
in an identical performance. As the channel SNR decreases the
curves representing the performance of non-symmetric and
equally likely signal design separate and the difference
between the two systems gets larger. The non-symmetric signal
design is 3 dB (in terms of output SNR) superior to equally
likely signal design for the noisiest channel considered (a
channel SNR of 4 dB) It is seen that significant improvement
in performance can be obtained for noisy channels by utilizing
non-symmetric signal design. The amount of improvement is a
function of source statistics.
A comparison between the equally likely signals and
unequal signal probabilities shows that if the signal
statistics are not very different, the shape of constellation
is not appreciably different from the equally likely case. But


84
(a)
(b)
Figure 4.2. The transmitters for systems employing (a) binary
(b) nonbinary alphabets.
In variable length coding (such as Huffman coding) a
symbol decoding error propagates in the block. Therefore, it
is appropriate to use word error probability as the figure of
merit. For the binary system Pw is given by
Pw=l-(l-pe)n (4.6)
where pc is the crossover probability. For a BPSK system pc is
known to be equal to
pe=OW2e) (4.7)
where e is the signal-to-noise ratio.
II. When the bandwidth constraint is not very strict and
can be somewhat relaxed. Let us allow some bandwidth expansion
for the binary system (Figure 4.2). As a result the nonbinary


4
B_ (i+r)D
(1.10)
where D=R/1 and r is the rolloff factor. Since BT =2B, the
transmission bandwidth of QAM is
d (1+r)R
nT j
(i.ii)
and the spectral efficiency with raised cosine filtering is
given by
log^AT bj ts/sec (1.12)
1 (1+r) Hz
The spectral efficiency increases with the number of the
points in the constellation. However, one can not increase the
spectral efficiency by increasing the number of points in the
signal constellation too much, because as you place more
signals in the constellation the error rate increases.
For reliable communications, the information rate must be kept
below the channel capacity. Therefore, the spectral efficiency
is upper bounded by
ti (1-13)
N
where S/N is the signal-to-noise power ratio.
Two dimensional formats are well suited for high speed
data transmission because of their efficient use of bandwidth.
However, they require coherent detection that implies the need
for synchronization circuits.


18
VJ2£Ht>.1*a.-2ap1, (1.36)
p variEj 1
where p, is equal to R(0) Putting all this together yields
tra r / Y t \
var(X,i)=(k+bkn)' 1 (1-37)
gp
The second term above is the dominant term because the effect
of channel noise
is
much more
destructive
to
the
reconstruction of
the
image than
the effect
due
to
quantization noise,
var(X'j) =K vax(Xi) 1 2a Pi a (1.38)
1-a2
To minimize the variance of the reconstruction error we will
set the derivative of this expression with respect to a to
zero. The optimum value of a turns out to be
a=
^-Pi2
Pi
(1.39)
1.9 Research Objectives
This dissertation is concerned with issues related to
nonsymmetric information sources. To motivate the work on
nonsymmetric sources, it is shown that DPCM of digitized video
signals results in a nonsymmetric information source. One of
the main goals is to address the problem of signal design in


41
P =1-
Pi
tzM
I
Pi-
p1tana]
J exp [--
Pa PitanPi
(Pi-Aii)
Jp|
Nn
) dp2dp1
P2
nN0

/
Pl-0
PjCanaj
J exp[-
PiPitanPj
(Pi-Aii)2
] dp2dpx
P3
nN0
M
/
Pl-0
Pjtanaj
J exp[-
P2--PitanPj
(Pi-AH)
j^]dp2dPl.
lvo
(3.17)
By making the following changes of variables
e = -7= dp 2=desfiro
W
(3.18)
u=-^; dp1=duv/^
W>
(3.19)
the probability of error in terms of the new variables
becomes


25
Equation (2.2) is readily solved for 1¡
*1.1
j x p (x) dx
, =1,2,...,!. (2.6)
*ii
f p(x)dx
*i
The solution of the equations for the general nonuniform
quantizer is difficult. However, a procedure to obtain a
solution by computer iteration has been introduced by Llyod
and Max [9],[10]. For a specified probability density function
and a fixed value of L, 1, is first selected arbitrarily, with
x,=-oo, we solve eq. (2.6) for x2. Next x2 and 1, are used in eq.
(2.5) to obtain 12. The process is repeated to obtain x3 from
eq. (2.6) and 13 from eq. (2.5). Continued iteration finally
stops when 1L is obtained from eq. (2.5). If 1, has been
correctly guessed, then 1L will satisfy the equation with xL+,
=oo. if it does not, 1, is corrected to a new value and the
process is repeated until 1L satisfies eq. (2.6). This
procedure satisfies conditions eq. (2.5) and eq. (2.6) which
are necessary for optimality.
Max [10] used the above procedure to find the quantum
levels for a zero-mean Gaussian message for quantizers up to
36 levels. Paez and Glisson [11] used the procedure to find
optimum levels for signals having either a gamma density or a
Laplace density for L=2,4,8,16, and 32.


34
(b)
Figure 2.3. The histogram for AERIAL MAP picture is compared
with (a) Laplacian Distribution (b) Distribution given in
equation (13). Histogram is shown with a broken line.


EFFICIENT COMMUNICATIONS FOR NONSYMMETRIC INFORMATION
SOURCES WITH APPLICATION TO PICTURE TRANSMISSION
By
SHAHRIAR EMAMI
A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
UNIVERSITY OF FLORIDA
1993


63
(3.78)
The gradient of the cost function depends not only on the
nearest points on the boundary but also on their derivatives.
Let us find the derivative of the components of Sj¡ with
respect to the components of s¡. Differentiating eq. (3.71) and
(3.72) with respect to x¡ and y¡ we get
(3.79)
where
(3.80)
(3.81)
where
w2= (x-x
(3.82)


11
channel coding, let us pick a (15,11) single error correcting
code. Assume a BPSK modulation, a signal-to-noise ratio of
43,776 ( S/N0=43,776) and a data rate of R=4800 b/s. Let Pub
and Pum represent the bit and message error rate for the
uncoded system and Pcb Pcm represent the bit and message error
rate for the coded system, respectively,
without coding
(1.23)
pu,b=0\ -^)=1.02xl0-5
(1.24)
and
PUia=l-(l-PUtb) ll=l. 12x10-4.
(1.25)
With coding
i?c=4800x (15/11) =6545 b/S
(1.26)
) =1.36xlO-4
The bit error rate for the coded system is inferior to that of
the uncoded system and the performance improvement due to


6
can be adjusted to give a smaller quantization error than
would be possible when quantizing the input directly.
1.4.1 Optimum Prediction
We are interested in linear prediction of the form
N
£(n) ='£, aj x(n-j) (1-14)
3-1
which is the weighted sum of N previous samples. The weights
aj are linear prediction coefficients. The filter is optimized
by finding the weights that minimize prediction error in a
mean squared sense
oe2=E[(x(n)-St(n))2) (1-15)
Since mean squared error is a function of aj and
do 2
-=5-^=0; =1,2 N (1.16)
da
is a necessary condition for minimum MSE (mean-squared error).
Evaluating the derivative gives
E[-2ix(n) -£(n))8j?(ri) ] (1.17)
a
Equating this to zero yields,
E[{x(n) -x(n) )x(n-i) ] =0; i=l, 2, . N. (1.18)
This is called the orthogonality principles which states that
minimum error must be orthogonal to all data used in the


108
(a)
(b)
Figure 4.7. A comparison of systems employing non-binary alphabets
against the binary system when the bandwidth can somewhat be
expanded. The solid curve represents the binary system, (a) ternary
(26,14,3) code and binary (15,11,1) code are used and a block is
390 bits long; (b) 4-ary (15,7,3) and binary (31,26,1) are used and
the block length is 465 bits long. The reults are given for a
medium detail picture.


117
erroneous pixels. In this method a pixel is compared with
pixels on the previous and following lines. If the differences
are above a threshold an erroneous pixel is found. The
codeword assignment is such that errors will be either small
or large. Because of high correlation between adjacent lines,
the pixel in error is replaced by the corresponding pixel in
the previous line.
Walsh-Hadamard transformation have been used by Fenwick,
Steel and Vansanji [30]. In this method DPCM blocks are
transformed and sent over the channel. DPCM blocks are
transformed in a manner that when the receiver recovers the
transform coefficients, the presence of an error is detected
and its magnitude is known. Correlation techniques are used to
find the erroneous sequence.
Periodic PCM updates have been used by some researchers
before. A good example is the work by Ngan and Steele [31].
They used several PCM updates per line. If the difference
between the PCM update word and DPCM output is greater than a
threshold, that potion of the line is in error. The pixel
value in error is replaced by a combination of the adjacent
pixels. The percentage increase in bit rate due to side
information was about 12 percent.
Many of the systems presented take advantage of signal
statistics. Steel, Jayant and Schmidt [32] introduced a
statistical block protection procedure. In this work the
statistical features of an input block such as the average and


I
134
(a)
(b)
Figure 5.5. The variation of picture SNR as a function of
channel error rate. The receiver utilizing first (the dotted
curve), second (the dashed curve) and third (the solid curve)
order Markov modelling of the source, (a) MMAP receiver, (b)
MSNR receiver.


References
[1] Habibi, A., "Compression of Nth-Order DPCM Encoder with
Linear Transformation and Block Quantization
Techniques," IEEE Trans, on Commum. Technol. Vol. Com-
19, pp. 948-956, 1971.
[2] Wintz, P. A.,"Transform Picture Coding," Proc. IEEE.
Vol. 60, No. 7, pp. 809-820, 1972.
[3] Jayant, N. S., and Noll, P., Digital Coding of Waveforms.
Englewood Cliffs, NJ: Prentice-Hall, 1984.
[4] Bhargava, V. K., Haccoun, D., Mattyas, R., and Nuspl, P.,
Digital Communications by Satellite Modulation.
Multiple Access and Coding. Wiley, New York, 1981.
[5] Lin, S., and Costello, D. J., Error Control Coding.
Englewood Cliffs, NJ: Prentice-Hall, 1983.
[6] Chang, K. Y., and Donaldson, R. W., Analysis,
"Optimization and Sensitivity Study of Differential PCM
Operating on Noisy Communication Channels," IEEE Trans.
Commun.. Vol. COM-20, pp. 338-350, 1972.
[7] Netravali, A. N., and Limb, J. O., "Picture Coding: A
Review," Proceedings of the IEEE. Vol. 68, No. 3, pp.366-
406, 1980.
[8] Limb, J. 0., "Source Receiver Encoding of television
Signals," Proc. IEEE. Vol. 55, pp. 364-379, 1967.
[9] Lloyd, S.P., "Least Squares Quantization of PCM," IEEE
Trans. Information Theory. Vol. IT-28, No. 2, pp. 129-
136, 1982.
[10] Max, J., "Quantization for Minimum Distortion," IRE
Trans. Information Theory. Vol. IT-6, No. 1,
pp. 7-12, 1960.
[11] Paez, M. D., and Glisson, T. H., "Minimum Mean-Squared-
Error Quantization in Speech PCM and DPCM Systems,"
IEEE Trans. Commun.. Vol. COM-20, No. 2, pp. 225-230,
1972.
155


5
1.3 Source Coding
The purpose of source coding is to remove as much
redundancy as possible from the message. Efficient coding of
messages provides the opportunity for significantly decreasing
the transmission costs. Two main approaches to picture coding
predictive coding and transform coding will be addressed here.
1.4 DPCM
There is considerable correlation between adjacent
samples of speech or image data, and indeed the correlation is
significant even between samples that are several sampling
intervals apart. The meaning of this high correlation is that,
in an average sense, the signal does not change rapidly from
sample to sample so that the difference between adjacent
samples should have a lower variance than the variance of the
signal itself.
The predicted value is the output of the predictor
system, whose input is a quantized version of the input
signal. The difference signal may also be called the
prediction error signal, since it is the amount by which the
predictor fails to exactly predict the input.
Since the variance of error signal is smaller than the
variance of signal, a quantizer with a given number of levels


132
(a)
O
Figure 5.4. The solid circles indicate the neighboring symbols
used in computing conditional statistics, (a) second order
Markov model and (b) third order Markov model.
TABLE 5.1. CODEWORD ASSIGNMENT FOR VARIOUS QUANTIZER LEVELS.
Level
Probability
codeword
1
. 6602
000
2
.1761
110
3
.1205
101
4
.0274
011
5
.0087
111
6
.0054
100
7
.0013
010


(3.44)
For large values of signal-to-noise ratio the conditional
probability of error will be egual to the upper bound. By
using approximation for Q function when the argument is large
and plugging back in (3.43) we get
Pe=N
N
n-i i*n
exp [-
lSi-SnH
4 N
3
(3.45)
Similarly it can be shown that the symbol error rate for
a MAP receiver is given by
Nr
N
Pe~\\l^Vn Z
' n-i l*
X> £
tei-sj
lsrsar+N0lnlj¡i)
exp [
(ls snll2 +Naln (~))2
Px
4W0Dsi-sn|2
(3.46)
3.4.2 Gradient Search Algorithms for an Average Power
Constraint
In search of the minimum, the gradient of the probability
of error is obtained analytically and an iterative gradient
search algorithm which modifies the constellation at each
iteration is used to find the optimum constellation.
In the gradient search algorithm, the iterative rule is
given by


39
d(p,si)> d(p,sk) i*k. (3.8)
where d(x,y) is the Euclidean distance between x and y.
Point H is the intersection of the decision regions and it is
therefore on the boundaries of the decision regions. For the
boundary points the above inequality changes to an equality,
d(H, A) =d(H, B) d(H, A) =d(H, C) (3.9)
or
AH = BH = CH,
(3.10)
and this results in
ai=Pi'
2 P2'
a3=P3. (3.13)
Define Pw to be the probability of word error. The
probability of symbol error is one minus the probability that
a symbol is correct. The probability of a correct symbol is
obtained by averaging all of the conditional probabilities of
a correct symbol, we have
PW=1~Y^P(C I mjpt. (3.14)
i-i
But P(C|m¡) is the result of point p falling in the decision
(3.11)
(3.12)


31
2.7 Discussion
It was shown that the pdf of input to the quantizer can
be fairly approximated by equation eq. (2.13) (which is the
convolution of a gamma and a uniform pdf) or simply a
Laplacian distribution. The theoretical statistics derived
from the models agreed well with the actual source statistics.
It is seen that DPCM of video signals does in fact produce a
nonsymmetric source.


66
computationally very expensive and long. However, one can use
a modified form of the above procedure to save on the
computational expenses. In the modified procedure, the average
cost is computed based on the received data and the gradient
based update rule is replaced with the directional set
(Powell's) method in multidimensions [20] which does not
require the gradient.
3.5.5 Application to DPCM System
In a DPCM system some errors are more costly than others,
therefore we could benefit from utilizing a Bayes receiver. In
Chapter Five an optimum method of selecting the cost matrix
will be presented.
36 Results
In this section we present some numerical results,
elaborate on the design procedure and discuss the performance
of the methods described in this chapter.
The probability of error, in general, is not a convex
function of the signal set, therefore the algorithm can
converge at local as well as global minima. The multi start
technique can be applied to the global optimization problem
[16]. In this technique one selects an optimization technique
and runs it from a number of different starting points. The


Even though efficient techniques for source coding,
channel coding and signal design exists, it is not known how
the choice of alphabet size affects a communication system.
Image transmission systems with various alphabet sizes are
compared on the basis of equal information rate, bandwidth and
average power. The systems employing various alphabet sizes
are analyzed and computer simulations are performed using
pictures with different amount of details.
An optimum procedure for decoding Markov sequences is
developed and the path metric is derived. A heuristic tree
searching algorithm is employed to obtain a suboptimum
solution.
Two other techniques for decoding Markov sequences, a
symbol-by-symbol modified MAP (MMAP) receiver using higher
order statistics and a maximum signal-to-noise ratio (MSNR)
receiver, are also given. The decoding procedures were applied
to image communication over noisy channels.
In summary, the major contributions of this dissertation
were the development of signal selection methods for
nonsymmetric sources, derivation of procedures for decoding of
correlated sources and application of these procedures to the
picture communication in noisy situations.
vii


73
if the signal probabilities are very different from the
equally likely case the shape of the constellation could be
very different from the equally likely constellation. Figure
3.6(b) and Figure 3.6(c) show constellations for two different
five signal sources. The constellations for the extremely non-
symmetric source is completely different from the other one.
Generally speaking the geometry of the constellations
depends upon the power constraint. For example, the optimum
signal constellations for five equally likely signals subject
to average power and peak power constraints are displayed in
Figure 3.8(a) and Figure 3.8(b). Clearly, the choice of power
constrain affects the geometry.
Simulations show that for large values of signal-to-noise
ratio the average cost signal selection does not result in any
improvement relative to the minimum error signal design. For
other values of signal-to-noise ratio the decision regions
form unusual shapes which vary with the signal-to-noise ratio.
Figure 3.10 shows the decision regions for a four signal
constellation with the statistics listed in Table 3.2. Because
of the complexity of the shape, obtaining an analytical
expression for the gradient was not feasible and as a result
the modified procedure in section 3.4.10 was utilized.
To compare the minimum error with the minimum cost signal
selection, a first order Gauss-Markov source with the
correlation coefficient of 0.9 was generated. The output of
the source was encoded with a first order DPCM system. An


118
the maximum magnitude of the differences is transmitted. At
the receiver end the maximum value is used to detect the most
seriously affected pixel, and the average statistic is used to
correct these samples approximately.
Moore and Gibson [33] have used convolutional code for
DPCM of speech. In their DPCM system some of the bits used for
source coding are used for error correction using self-
orthogonal convolutional codes.
Sayood and Borkenhagen [12] have presented a technique
for error correction of DPCM systems. No explicit channel
coding is performed. The output of the source coder is argued
to contain redundancy since certain codeword combinations are
more likely than the others. The decoding is formulated
similar to the Viterbi algorithm and a path metric that
depends on signal statistics is given. Substantial performance
gain in image coding is reported.
A number of methods for decoding correlated sources have
been proposed in this chapter. A minimum error procedure for
the decoding of correlated sequences will be derived which
requires tree decoding. A modified MAP receiver (MMAP) and a
maximal signal-to-noise ratio (MSNR) will also be presented.
The application of the proposed methods to picture
transmission over noisy channels will discussed in section 6.
The results will compared to the performance of a recently
proposed technique.