
Citation 
 Permanent Link:
 http://ufdc.ufl.edu/AA00020973/00001
Material Information
 Title:
 Code timing estimation in directsequence codedivision multipleaccess communication systems
 Creator:
 Smith, Ronald F., 1963
 Publication Date:
 1999
 Language:
 English
 Physical Description:
 xi, 127 leaves : ; 29 cm.
Subjects
 Subjects / Keywords:
 Adaptive filters ( jstor )
Communication systems ( jstor ) Estimators ( jstor ) Mathematical vectors ( jstor ) Matrices ( jstor ) Propagation delay ( jstor ) Random variables ( jstor ) Receivers ( jstor ) Signals ( jstor ) Simulations ( jstor ) Code division multiple access ( lcsh ) Dissertations, Academic  Electrical and Computer Engineering  UF ( lcsh ) Electrical and Computer Engineering thesis, Ph.D ( lcsh )
 Genre:
 bibliography ( marcgt )
nonfiction ( marcgt )
Notes
 Thesis:
 Thesis (Ph.D.)University of Florida, 1999.
 Bibliography:
 Includes bibliographical references (leaves 121126).
 General Note:
 Typescript.
 General Note:
 Vita.
 Statement of Responsibility:
 by Ronald F. Smith.
Record Information
 Source Institution:
 University of Florida
 Holding Location:
 University of Florida
 Rights Management:
 Copyright [name of dissertation author]. Permission granted to the University of Florida to digitize, archive and distribute this item for nonprofit 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:
 030367671 ( ALEPH )
41936483 ( OCLC )

Downloads 
This item has the following downloads:

Full Text 
CODE TIMING ESTIMATION IN DIRECTSEQUENCE
CODEDIVISION MULTIPLEACCESS
COMMUNICATION SYSTEMS
By
RONALD F. SMITH
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
1999
I have been blessed with a wonderful family, and this work is dedicated to them.
To Amy, for all of her support, encouragement and love. To Sarah, for so readily
sharing her silly little giggles and big hugs and kisses.
ACKNOWLEDGMENTS
I would like to express my gratitude to my adviser, Dr. Scott L. Miller, for his
guidance and feedback throughout the course of my research. I also thank the mem
bers of my Ph.D. committee, Drs. Leon Couch II, Jose Principe, Robert Fox, and
Richard Newman, for their interest and willingness to be a part of this effort.
While I have dedicated this work to my wife, Amy, she also deserves special
recognition. She has worked extremely hard supporting our family while I pursued
my graduate studies, and without her efforts this work wouldn't have been possible.
I thank my parents, Ernest and Joyce Smith, and my inlaws, Thomas and Peggy
Meneskie, for their continuous encouragement and support throughout this effort.
I also acknowledge my friends and colleagues, John Miller, Brad Rainbolt and Ali
Almutairi, who have made my time at the University of Florida an interesting expe
rience.
TABLE OF CONTENTS
page
ACKNOW LEDGMENTS ............................. iii
LIST OF FIGURES ................................ vi
KEY TO ABBREVIATIONS ............................ viii
A BSTR A CT . . . . x
CHAPTERS
1 INTRODUCTION ............................ 1
1.1 Introduction to SpreadSpectrum Techniques . 1
1.2 Mathematical Notation . . 6
1.3 Outline of the Dissertation . . 7
2 SYSTEM MODEL FOR THE AWGN CHANNEL . 9
2.1 General Asynchronous DSCDMA System Model for the
AW GN Channel . . . 9
2.2 Simplified Asynchronous DSCDMA System Model for the
AW GN Channel . . ... .. 12
3 REVIEW OF THE LITERATURE . ... .. 14
4 THE MMSE RECEIVER AND SINGLEUSER CODE
ACQUISITION . . . ... .. 21
4.1 The MMSE Receiver . . ... .. 21
4.2 Advantages of Synchronizing the MMSE Receiver ..... 25
4.3 Adaptive Filters and the Estimation Algorithm ....... 29
4.4 Sum m ary . . . 35
5 CHARACTERIZING THE PERFORMANCE OF THE TIMING
ESTIM ATOR ............................. 37
5.1 Transient Statistics of the LMS Filter Weights ....... 37
5.2 Chip Selection Error Probability .................. 45
5.3 Approximation for the Conditional Variance of the Timing
Estim ate . . . 50
5.4 Coarse Acquisition Performance . ... 54
5.5 Sum m ary . . . 59
6 SIMULATING FADING CHANNELS ................... 62
7 CODE ACQUISITION IN NONSTATIONARY ENVIRONMENTS 71
7.1 Frequency Synchronization Errors ................ 75
7.2 Performance in FlatFading Channels .............. 78
7.3 Sum m ary . . . 84
8 INCREASED WINDOW OF OBSERVATION AND TIMING
ACQUISITION ............................ 86
8.1 Updated System Model ..................... 86
8.2 Updated Timing Estimation Algorithm ............. 88
8.3 Sum m ary . . . 93
9 SUMMARY AND FUTURE WORK ................... 94
9.1 Sum m ary . . . 94
9.2 Contributions ................. ... .. .... 96
9.3 Future W ork ........................... 96
APPENDICES
A PROBABILITY OF BIT ERROR USING THE GAUSSIAN
APPROXIMATION ......................... 99
B DERIVATION OF TRANSIENT EQUATIONS FOR THE LMS
FILT ER . . . . 102
B.1 Transient Weight Autocorrelation Matrix Using the Gaussian
Approximation ........................ 102
B.2 Transient Weight Autocorrelation Matrix Using the Known
Statistics of the Receiver's Input Vector .......... 106
C REVIEW OF QUADRATIC FORMS OF GAUSSIAN RANDOM
VARIABLES ............................. 112
D TYPOGRAPHICAL NOTES ........................ 119
REFERENCES ....................... ........ ... 121
BIOGRAPHICAL SKETCH ............................ 127
LIST OF FIGURES
Figure page
1.1 Power spectral densities of the unspread data sequence and the direct
sequence spreadspectrum signals . . .. 3
4.1 Code acquisition with an adaptive receiver . .... .. 22
4.2 Minimum meansquared error as a function of the propagation delay
of user # 1 . . . . .28
4.3 Plot of the cost function f(p, 6) for a singleuser system ........ .. 33
4.4 Plot of the cost function f(p, 6) for a fifteenuser system ........ .. 34
5.1 Performance of equation (5.2) for a singleuser system, a) The analyti
cal mean filter weight vector norm, IIE[w(m)] II (*), and the simulated
mean filter weight vector norm IIw(m) I (solid line) plotted as a func
tion of the training length; b) Comparing the error between the two
mean weight vectors, IE[w(m)] wA(m)I . ... .. 43
5.2 Performance of equation (5.2) for a threeuser system, a) The analyti
cal mean filter weight vector norm, IIE[w(m)]II (*), and the simulated
mean filter weight vector norm IIw(m)II (solid line) plotted as a func
tion of the training length; b) Comparing the error between the two
mean weight vectors, E[w(m)] wA(m)J . 44
5.3 Performance of equations (5.4) and (5.7) for a twouser system. The
Frobenius norm of the LMS filter's transient weight autocovariance
matrix; equation (5.4) (*), equation (5.7) (x) and simulation
results (+ ) . . . . .45
5.4 Comparing incorrect chipselection probability to the Unionbound for
a threeuser system. Simulation results (+), the Chernoff bound
using equations (5.2) and (5.7) (solid line), and the Chernoff bound
using simulated weight vector statistics (*) . .... .. 51
5.5 Comparing the simulated conditional variance of the 61 estimate to the
analytically derived approximation in absence of nearfar MAI. 53
5.6 Comparing the simulated conditional variance of the 61 estimate to the
analytically derived approximation in presence of nearfar MAI. 54
5.7 Average training bits required for correct acquisition as a function of
the number of system users in AWGN channel . ... .. 57
5.8 Average training bits for correct acquisition as a function of the level
of multiaccess interference in AWGN channel . ... .. 58
6.1 Flatfading channel simulator . . ... .. 68
6.2 Theoretical and simulated probability density function for the enve
lope of the fading channel simulator output for two different Rician
parameter values . . . .... ..69
6.3 Magnitude and phase of a slow flatfading Rayleigh process generated
with a carrier frequency of 1.8 GHz, a vehicle speed of 3 MPH, and
a data rate of 9600 BPS . . ... ..69
7.1 Modified adaptive receiver with channel compensation ... .. 74
7.2 Acquisition performance of the adaptive receiver in a fiveuser environ
ment in presence of frequency offset errors . ... .. 77
7.3 Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a thirdorder linear predictor for the channel estima
tion algorithm . . . .. ..80
7.4 Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a tenthorder linear predictor for the channel estima
tion algorithm . . . .. ..81
7.5 Comparing the average training bits for correct acquisition as a func
tion of the level of multiaccess interference in a flatfading channel. 82
7.6 Performance of the timing acquisition algorithm using the hybrid and
modified MMSE receivers . . ... .. 83
8.1 Plot of the cost function f(p, 6) for a fifteenuser system when the
receiver observes two bitintervals . ... .. 91
8.2 Comparing the performance of the timing estimator when the receiver
observes 2bit intervals (*) versus observing 1bit interval (solid line)
as a function of the MAI . . ... .. 92
A.1 The MMSE receiver for the Gaussian approximation ... .. 99
KEY TO ABBREVIATIONS
AWGN: additive white Gaussian noise
BER: biterrorrate
BPS: bitspersecond
BPSK: binary phase shift keying
CRB: CramerRao bound
CTAN: Comprehensive TEJX Archive Network
DSCDMA: directsequence codedivision multipleaccess
DSSS: directsequence spreadspectrum
FHSS: frequencyhopped spreadspectrum
IEEE: Institute of Electrical and Electronics Engineers, Inc.
ISI: intersymbol interference
LFSR: linearfeedback shiftregister
LMS: least mean squared
LOS: lineofsight
LPD: low probability of detection
LPI: low probability of intercept
LSML: largesample maximum likelihood
MAI: multiaccess interference
MASE: multipleantenna sensorsbased estimator
MMSE: minimum meansquared error
MOE: meanoutputenergy
MPH: milesperhour
MSE: meansquared error
MUSIC: multiple signal classification
PCS: personal communication system
PSD: power spectral density
RASE: rapid acquisition by sequential estimation
RLS: recursive leastsquares
RMSEE: root meansquared estimation error
SNR: signaltonoise ratio
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
CODE TIMING ESTIMATION IN DIRECTSEQUENCE
CODEDIVISION MULTIPLEACCESS
COMMUNICATION SYSTEMS
By
Ronald F. Smith
May 1999
Chairman: Dr. Scott L. Miller
Major Department: Electrical and Computer Engineering
This dissertation considers code timing estimation for asynchronous directsequence
codedivision multipleaccess communication systems operating over additive white
Gaussian noise and flatfading channels. Unfortunately, this type of parameter esti
mation is difficult in the presence of multiple system users, and traditional methods
are known to be sensitive to the nearfar problem. The nearfar problem is a con
dition which occurs when the received signal amplitudes of the the multiple system
users are very dissimilar.
A wellknown receiver, referred to in the literature as the minimum meansquared
error (MMSE) receiver, is used to form an estimate of the code timing for a single
user. The MMSE receiver has several desirable properties. The MMSE receiver
is the optimal receiver for a singleuser operating in an additive white Gaussian
noise channel, but is suboptimal in a multiuser scenario. However, the IMMSE
receiver has been observed to be resistant to the nearfar problem. Traditionally,
the MMSE receiver is implemented using an adaptive filter and can, therefore, learn
and adapt to the ambient channel conditions. It will be shown, that with very little
side information, it is feasible to form an estimate of a singleuser's timing offset by
processing the weights of the adaptive filter. The performance of the proposed timing
estimation algorithm has been studied under several different scenarios, and has been
observed to be resistant to the nearfar problem. The complexity of the proposed
timing algorithm is similar to conventional singleuser estimation methods, and is
shown to be of lower complexity than other proposed timing estimation algorithms.
Others have studied the performance of the MMSE receiver in a flatfading chan
nel. It has been found that the MMSE receiver must be modified such that the
received signal is compensated to offset the dynamic phase changes induced by the
channel. In certain environments, this compensation algorithm is shown to degrade
the performance of the timing estimation algorithm. A modification to the operation
of the receiver is proposed to offset some of this performance degradation.
CHAPTER 1
INTRODUCTION
1.1 Introduction to SpreadSpectrum Techniques
In 1941, actress Hedy Lamarr and composer George Antheil filed for a patent
on a "Secret Communication System" [1]. Their device operated on the idea of fre
quently changing the carrier frequency of a radio signal. If the carrier frequency was
updated fast enough, and in an apparently random fashion, an adversary would have
difficulty maintaining contact with the transmitting radio. This means that an ad
versary would only be able to receive a small portion of a message, whether it was
sent coded or uncoded. And, therefore, it is much more difficult for an adversary
to discern the intent or contents of the transmitted message. Of course, this means
that the intended receiver has to dynamically synchronize its carrier frequency with
that of the transmitter. Lamarr and Antheil used paper rolls, similar to those found
in player pianos, as their synchronization method. Hole placement on the roll would
determine the carrier frequency used, and an exact duplicate of the roll used by the
transmitter was given to the intended receiver prior the "secret" communication.
Lamarr and Antheil gave their invention, and the rights to the patent, to the United
States government. Since this development occurred during World War II, this tech
nique initially received a lot of attention. However, the U.S. government deemed the
mechanics of the synchronization method too complex for widescale military use.
The technique developed by Lamarr and Antheil, which is now known as frequency
hopped spreadspectrum (FHSS), is explained in the following few sentences. Let B
represent the bandwidth required to transmit the message signal, using a traditional
radio with a fixed carrier frequency. If we now let the carrier frequency hop through
N discrete carrier frequencies, each separated by B Hz, then the bandwidth of the
transmitted signal becomes NB. Therefore, the bandwidth of the transmitted sig
nal using frequencyhopping has been spread to a wider bandwidth than what is
required to transmit the message signal. However, until the advent of the integrated
circuit, the problem of implementing a synchronization method limited the use of
such systems.
The dual of FHSS is known as directsequence spreadspectrum (DSSS). In DS
SS, a digital signal is modulated by a digital spreadingsequence prior to transmission.
Consider a binary sequence of 1 data where each data bit has duration Tb. If this
data sequence is used to amplitude or phase modulate a carrier signal, the power
spectral density (PSD) of the transmitted signal will have a nulltonull bandwidth of
2/Tb Hz. A spreading sequence is comprised of a pseudorandom sequence of 1 chips
where each chip has duration Tc, and it is assumed that Tb = NTc. By multiplying
the data sequence by the spreading sequence prior to transmission, the PSD of the
transmitted signal now has a nulltonull bandwidth of 2/T, which is N times greater
than the corresponding bandwidth of the original unspread data sequence. The total
transmitted power in the unspread and the DSSS signals is assumed to be equal,
since multiplication by 1 doesn't affect the transmitted power. This means that the
PSD for the DSSS signal will have a much lower amplitude than the PSD for the
unspread signal as shown in Figure 1.1.
In the previous paragraph, we referred to the spreading sequence as being pseudo
random. Ideally the chips of the spreading sequence would be randomly generated,
with +1 occurring with the same probability as 1. However, just as the intended re
ceiver in a FHSS system requires knowledge (and synchronization) of the frequency
hopping algorithm, the intended receiver in a DSSS system must have knowledge
(and synchronization) of the spreading sequence. Therefore, some deterministic
method must be used to generate a spreading sequence. A pseudorandom sequence
can be generated with a linearfeedback shiftregister (LFSR), which is just a bank of
1 PSD of Unspread
Signal
PSD of Spread
Frequency
Figure 1.1: Power spectral densities of the unspread data sequence and the direct
sequence spreadspectrum signals.
flipflops interconnected with the appropriate feedback. Using an LFSR to generate
the spreading sequence, the transmitter and receiver must have common knowledge
of the length of the LFSR, the feedback connections, and the initial state of the LFSR
prior to transmission. With the appropriate feedback, 1 will occur with a proba
bility slightly higher than +1. In addition the sequence will be periodic. However, if
the period is sufficiently long, over a short observation interval the occurrence of +1
and 1 will appear random.
The goals of a covert communication system are to communicate privately with
an asset and to communicate reliably under adverse conditions, possibly caused by
an adversary. Over the last several decades, spreadspectrum systems have received
considerable attention for their use as covert communication systems. The attention
is due to three properties of spreadspectrum systems some of which are hinted at
in Figure 1.1. The first property of interest is the resistance to intentional jamming.
One way for an adversary to disrupt a communication system is to broadcast at a
high power level in the same frequency band. Due to the increased bandwidth utilized
in a spreadspectrum system, a jammer then only achieves partial band jamming or
has to spread the jamming signal over a wider bandwidth. In either case, the effect of
the jamming signal is reduced compared to the effect observed by the same jammer
in the unspread communication system. The second property of interest is the low
probability of detection (LPD). Even if an adversary knows you are transmitting,
without knowledge of the spreading sequence (or frequency hop sequence), it is very
difficult for that adversary to perform a good detection of the message signal. The
third property, which is especially important for clandestine communication, it called
low probability of intercept (LPI). Considering the reduced amplitude of the DSSS
signal's PSD as shown in Figure 1.1, we see that with sufficient spreading the level of
the PSD can be made to approach the level of the background noise floor. This means
that an adversary with a spectrum analyzer that is actively looking for the spread
spectrum transmitter will have difficulty detecting that the transmitter is active. In
other words, in order to locate the transmitter the adversary will have to be physically
located very near the transmitter when a signal is being transmitted. Under ideal
conditions the transmitter will be active for short time intervals, or the asset is able
to detect the presence of the adversary in which case he/she turns the transmitter off.
These conditions greatly decrease the probability of intercept and, therefore, increase
the "life expectancy" of the asset.
As the remainder of this work deals with what is essentially a multiuser direct
sequence spreadspectrum system, we list several references which describe DSSS
systems in more detail. Scholtz [2] and Pickholtz et al. [3] provide good tutorials
on the properties and use of spreadspectrum systems. Ziemer and Peterson [4] give
an extensive examination of these systems. Many references on the generation and
properties of spreading sequences are available, as this subject continues to be a
current research interest [57].
In directsequence codedivision multipleaccess (DSCDMA) systems, each user
has a unique spreading sequence and all users occupy the same frequency band trans
mitting independently of all other users. Consider a K user DSCDMA system in
which each user transmits 1 times its spreading code. We will communicate with
one of the users, called the desired user, using a receiver that is optimized for a
singleuser operating in the presence of additive white Gaussian noise (AWGN). For
this example, the optimum receiver is a filter matched to the desired user's spreading
code. Therefore, the output of the matched filter will depend on the correlation of
the desired user's code with all of the other K 1 users' codes. If all of the users'
spreading codes are orthogonal for any amount of time shift between the codes, the
output of the matched filter will not depend on the presence of the other system
users. A more realistic condition is that the set of spreading codes is chosen so that
the crosscorrelation between any two codes is wellbehaved. This means the output
of the matched filter will be dominated by the desired user, but some component of
the output will be due to contributions of interfering users.
It is possible to receive each of the transmitted signals at a different power level.
For example, this condition could occur if any of the K 1 interfering users is located
physically closer to the receiver than the desired user. In this case the output of
the matched filter may become dominated by the interference. This type of multi
access interference (MAI) is referred to as the nearfar problem. The challenge for
communication systems engineers is to develop receiver structures that are insensitive
to the nearfar problem.
In this dissertation, we are concerned with the problem of achieving synchroniza
tion with the spreading sequence of the desired user. Moon et al. [8] studied the
effects of MAI on traditional techniques (energy detector) that are used in single
user spreadspectrum communication systems. It was found that the average time
to achieve acquisition quickly increases when a nearfar scenario exists with only two
system users. Moon et al. [9] presented a multiuser synchronization method, called
the rapid acquisition by sequential estimation (RASE) algorithm, which essentially
tries to estimate the contents of the LFSR used to generate the spreading sequence for
each user, based on observations of the received signal. They assumed that all users
transmit only their spreading code and that the receiver knows the received ampli
tude of each user. For a total of K users, this algorithm has a complexity of O(K2K)
which is prohibitive for K > 10. A similar singleuser approach was presented by
Barghouthi and Stiiber [10, 11], under the assumption that all users are received at
the same power level. Madhow and Pursley [12], considered the effects of achieving
code synchronization in the presence of MAI on the system capacity. They found
that the system capacity for achieving code synchronization with a matched filter
was much less than the system capacity measured with respect to the biterrorrate
(BER). These results form the motivation for researching code acquisition techniques
for DSCDMA systems.
1.2 Mathematical Notation
Matrices and vectors are typeset in a bold face, for example v or V. In general,
a lowercase letter will be used to denote a vector, while an uppercase letter denotes
a matrix. Unless otherwise noted, all vectors are assumed to be column vectors. In
order to denote the nth element of a vector, a subscript will be used such as vn. The
(i, j)th element of a matrix V is denoted by Vij. A subscript containing the variable
k denotes an element belonging to the kth user in the DSCDMA system.
The transpose, the conjugate (or Hermitian) transpose, and the 2norm of a
columnvector v are denoted by vT, vH and lv = /vHv, respectively. The conju
gate of a scalar quantity will be denoted as a*. For a matrix V we will use iiViIF to
denote the Frobeniusnorm of the matrix defined as
iiv i12 =i2 (Vl)2.
IlVll; ~ ZI ,jl2 (1.1)
i 3
The Kronecker delta function is defined as
1 if m = 0,
6K(m) = (1.2)
0 otherwise.
The expectation of a random variable, /3, either a scalar or nonscalar quantity, will
be denoted as E[/3] or /3. We use the notation O(N) to denote that an algorithm has
numerical complexity on the order of N. When a is a parameter to be estimated, we
denote the estimate as &.
1.3 Outline of the Dissertation
In the next chapter, a system model that is widely used to simulate a DSCDMA
communication system will be presented. The system model is expressed in a conve
nient vectormatrix notation which facilitates simulating a DSCDMA communication
system on a computer. The third chapter provides a brief review of the literature
concerning DSCDMA receivers and timing acquisition techniques. While it is impos
sible to provide a complete review of these subjects, an effort is made to provide the
reader with enough of a review to understand several timing acquisition techniques
and their associated complexities. In the fourth chapter, a lowcomplexity singleuser
timing estimation algorithm is presented. The algorithm is based on processing the
weights of an adaptive filter, which is commonly referred to as the MMSE receiver.
The MMSE receiver is a singlebit singleuser detector, which is a suboptimal receiver
for a multiuser DSCDMA communication system, and is receiving considerable at
tention in the literature. As several papers in the literature ignore the effects of
operating the MMSE receiver asynchronous to the intended user, several examples
of why synchronization is important are presented. In the fifth chapter, the per
formance of the timing estimator is characterized. Several analytical techniques are
used to characterize the performance of the timing estimator, based on a developed
statistical model of the transient statistics of the MMSE receiver weights. Simulation
results are used to study the performance of the timing acquisition algorithm for
a general asynchronous DSCDMA system when two commonly used adaptive algo
rithms are used in the MMSE receiver. It will be shown that for one of these adaptive
algorithms, the performance of the timing acquisition algorithm is nearfar resistant.
In the sixth chapter, the effects of fading on a communication system are discussed
and a commonly used implementation of a flatfading channel simulator is presented.
The performance of the timing acquisition algorithm when the receiver is operating
in two nonstationary environments (frequencyoffset errors and flatfading channels)
is studied in the seventh chapter. Barbosa and Miller [13] have presented a modified
version of the MMSE receiver that can be used in flatfading channels. The perfor
mance of the timing acquisition algorithm for this receiver will be studied. Based on
simulation results, it will be shown that a slight improvement in the performance of
the timing estimator is feasible, through a slight modification in the training cycle of
the modified MMSE receiver. In earlier chapters, the system model and the timing
estimation algorithm were developed such that only one of the DSCDMA system
users can be in the mode of acquiring timing acquisition. In the eighth chapter, a
modification to the system model and the timing estimation algorithm is presented
such that more than one system user can be in the timing acquisition mode. In the
final chapter, a summary of the work presented in this dissertation, and the contri
butions to the area of DSCDMA research, as well as several areas of future research
are provided.
CHAPTER 2
SYSTEM MODEL FOR THE AWGN CHANNEL
2.1 General Asynchronous DSCDMA System Model for the AWGN Channel
In this section, a general system model for the case when the receiver operates
in an AWGN channel and observes only one symbol interval of information at a
time will be presented. After this general model is developed, a simplified version,
which will eventually be shown to lead to a reduction in the numerical complexity
of evaluating the timing estimator, will be presented. The only difference between
these two models, is how we interpret or apply the effect of carrier phase offset to the
input of the receiver for the desired user.
In this work, a binary phase shift keying (BPSK) communication system is used.
There are a total of K system users, each operating at the same carrier frequency fc,
but asynchronous to and independent of the other users. The kth user is assumed to
transmit a signal whose complex envelope [14], Sk(t), is a polar data sequence, dk(l) E
{+1, 1}, which has a unique signature or spreading waveform, Ck(t), superimposed
upon it such that
00
Sk(t) = E dk(l)ck(t lTb) k e {1,2,..., K}. (2.1)
1=00
Therefore, the signal actually transmitted by the kth user is given by
Sk(t) = Re[Sk(t)exp{j27frt}] = Re[Sk(t)exp{jwct}] k E {1,2,...,K} (2.2)
where Re[] returns the real part of its argument. Without loss of generality, it is
assumed that the first user is the desired user and all other users act as interference.
The data bits have a duration Tb while the chips of the spreading sequence have
duration Tc. Each spreading sequence has a period of N = Tb/Tc chips. That is, one
period of a spreading sequence is equal in duration to one bit interval. While this is
not a requirement for DSCDMA systems in general, it is required for the receiver
structure studied in this work. Let the vector Ck = (Ck,O, Ck,1,... Ck,N1)T represent
one period of the kth user's spreading sequence so that the spreading waveform can
be written as
N1
Ck (t) = Ck,n17 (t nT,) (2.3)
n=O
where pT (t) is the chip pulse shape which is taken to be 1 over the interval [0, Tc)
and zero otherwise.
We assume that the receiver is asynchronous in time and phase to the transmitted
signal of the desired user, as well as being asynchronous to all of the interfering users.
Due to the K system users, the received signal is of the form
K
r(t) = v"Pk Re [Sk(t Tk) exp {jwct + jOk}] + n(t). (2.4)
k=1
In the above expression, Tk and Pk are the propagation delay and the average received
signal power, respectively, for the kth user, 0k is the phaseoffset for the kth user,
while n(t) is the additive white Gaussian noise which is assumed to have a onesided
power spectral density of A.M.
In order to process the received signal, the receiver converts r(t) from a bandpass
signal to a baseband signal, R(t), which is then passed through a filter matched to
the chip pulse shape and the output of this matched filter is sampled at the chip rate.
It is assumed that any doublefrequency terms created in the conversion to baseband
are not passed through the chip matched filter. During the receiver's mth bit interval
the nth chip sampled output can be expressed as
mTb+(n+1l)Tc
rm,n = R(t)dt e C. (2.5)
2PP1 llc j
mTb+nTc
During each bit interval, the N chip samples are stored in a received vector,
r(m) = (rm,0, rm,1, ... rm,N1)T and it is this vector quantity that will be processed
by the receiver. Using the above definitions for how each user contributes to r(t) we
can express how each user contributes to r(m) using a notation similar to that used
by Miller [15]. It can be shown that
K Pk
r(m) = P Jk(m)exp{jOk} + Nm E CNx (2.6)
k=l P
where Jk(m) is the contribution to the received vector from the kth user during the
mth receiver bit interval given by
Jk(m) = [z2k1(m)a2k (pk, bk) + Z2k((m)a2k(pk,6k)] (2.7)
where
Z2ki(m)=[dk(m)+dk(m 1)]/2 {1,0,1}
Z2k(m) = [dk(m) dk(m 1)]/2 e {1,0,1} (2.8)
a2kl(pk,(k)= (1 k)Ck) +kCk+)
a2k(pk,bk) =(1 .k) + 6kCk (2.9)
Cn (Ck,Nn,Ck,Nn+l, ... Ck,Nl, Ck,O, Ck,l, ... Ck,Nn1)T (2.10)
k = (Ck,Nn, Ck,Nn+l, ... Ck,Nl, Ck,O, Ck,li ... Ck,Nnl)T (2.11)
E {+1,1} if m E {0,,...,N 1},
Ck,m = (2.12)
0 otherwise.
and Tk =PkTc + kTc with pk E {0,1,..., N 1} and 0 _< k < 1.
Since the receiver is asynchronous to any of the K system users and only observes
one bit interval of information, it is possible to have a data bit transition occur during
the receiver's bit interval. This fact is accounted for in the previous equations. The
quantities Z2ki(m) and z2k(m) are used to indicate if adjacent data bits for the kth
user are similar. If adjacent data bits are the same, then Z2kl(m) will be nonzero,
otherwise z2k(m) is nonzero. The vector quantities ck7 and 'n) denote the nth right
cyclic and the nth modified right cyclic shifts of the kth user's spreading sequence,
respectively. As shown in the previous equations, Jk(m) is just a function of two
slightly offset cyclic shifts of the spreading sequence for the kth user.
In the remainder of this work, we will refer to the propagation delay in terms
of the integer part, pk, and the fractional part, 6k, only. The noise vector, Nm, in
equation (2.6) consists of independent complex Gaussian random variables whose real
and imaginary parts are also independent having zeromean and equal variances of
a2 = N/(2EIb/MAo), where Eb is the average received energy per bit of the desired
user. In the remainder of this work, we define the signaltonoise ratio (SNR) as
SNR = Eb/lAo.
2.2 Simplified Asynchronous DSCDMA System Model for the
AWGN Channel
In the previous section, the receiver was assumed to be phase asynchronous to any
of the K systems users. This is a reasonable assumption, since it is highly unlikely
that all of the oscillators for the users will be phase synchronous as viewed by the
receiver. Certainly, it could be possible for all of the transmitted signals to be phase
synchronous, a situation that would arise at a transmitting base station in a cellular
type communication system where one oscillator can be used to transmit all of the
K user's signals. At any of the mobile receivers, then one could say that all users
experience the same propagation delay and carrier phase shift in the channel from
the base station to the mobile receiver. However, in the communication link from the
mobile user to the base station, different propagation delays and carrier phase shifts
will exist due to the different physical locations of the mobile users.
However, we can model the system as being equivalently phasesynchronous as
follows. We can express equation (2.6) as
K Pk
rm=E (cos(0k)+Ijsin(0k))Jk(m) + Nm eCNx. (2.13)
k=1
We then can create a modified receiver input by only processing the real part of r(m).
That is,
K I
i(m) = Re [r(m)] = PkJ cos()Jk(m) + Nm e RNxl (2.14)
k=l 1
where the noise vector, Nm, in equation (2.14) consists of realvalued Gaussian
random variables that are independent having zeromean and equal variances of
a2 = N/(2Eb/Aro).
Now the cos(Ok) term can be absorbed into the ratio of Pk/P1. Therefore, by
adjusting the level of the SNR appropriately one can equivalently model the effects of
the carrier phase offset for the desired user. To account for the effects of the carrier
phase offset for the other users, the value of Pk/PL for each interfering user is adjusted
as appropriate.
As a result of using this procedure to form an equivalently phasesynchronous
system, we will eventually show that the numerical evaluation of the timing estimator
for the propagation delay for the desired user is simplified. In fact, we will see that
for this model the estimate of 1, the fractional part of the desired user's propagation
delay can be expressed in closed form, which is not the case when the model of
Section 2.1 is used.
CHAPTER 3
REVIEW OF THE LITERATURE
Since the area of DSCDMA has received so much attention in the literature over
the last several years, it is impossible to review all contributions made in the recent
past. However, in this chapter, we will try to summarize any major contributions
as well as those that are directly related to code acquisition in DSCDMA systems.
We will discuss the complexity of the timing estimation algorithms, so that we may
later compare them to the complexity of the timing estimator presented in the next
chapter.
Verdi [16,17] provided a major breakthrough in DSCDMA research by proving
the existence of an optimum multiuser receiver that is nearfar resistant. The receiver
is comprised of a bank of matched filters, one matched filter for each user, followed
by a Viterbi decoder. The receiver requires a lot of sideinformation, in order to
create the matched filters and form a decision metric for the Viterbi algorithm. The
receiver requires knowledge of each users' spreading code, and must be synchronized
with each user. In addition the receiver must know the received signal energy for each
user. The complexity of the receiver is 0(2K) for each binary decision made. The
receiver is considered impractical due to this complexity, which provides motivation
for investigating suboptimal receiver structures.
Lupas and Verdi [18,19] have proposed a suboptimum multiuser receiver which
is known as the linear decorrelating receiver. The receiver is similar to the optimum
receiver [16,17] in that it is comprised of a bank of matched filters, but does not
require knowledge of the received signal energy for each user. The outputs of the bank
of matched filters are processed by a linear transformation (called the decorrelating
filter) that removes the multiuser interference from the output of each filter. The
decorrelating filter cancels the MAI by accounting for the known correlation between
the users' spreading codes. The complexity of the receiver is dependent on taking the
inverse of a matrix, which has complexity 0(N3), to form the decorrelating filter. If
this operation can be done once, the complexity of the receiver is then linear in the
number of users, which is a great improvement over the exponential complexity of
the Viterbi based decoder.
Str6m et al. [20] studied the impact of timing errors on the decorrelating receiver.
It was found that the decorrelating receiver loses its nearfar resistance with the pres
ence of timing errors. That is, the BER of the receiver reaches a nondiminishing
floor even as the SNR increases, when errors are present in the estimates of the
timing offsets. In addition, they observed that the required variance of a timing
estimate is inversely proportional to the level of the MAI. This means as the MAI in
creases, making timing estimation more difficult, a better timing estimate is required
in order for the receiver to maintain its nearfar resistance. Zheng and Barton [21]
perform a similar analysis on the decorrelating receiver. In addition to considering
time synchronization errors, they studied the effects of phase synchronization errors
and frequency synchronization errors. They found that if these quantities are limited
to small, and possibly unrealistic, values the decorrelating receiver still offers a sig
nificant advantage over the conventional matched filter. These results motivate the
need for the development of timing estimators that produce unbiased estimates with
low variances, even in the presence of a severe nearfar environment.
Varanasi and Aazhang [22] have proposed a suboptimal multiuser detector whose
complexity is also linear in the number of users. The receiver requires knowledge of
the received signal strengths of each user, the code sequence of each user, and time and
phase synchronization with each user. The receiver is comprised of a bank of matched
filters (one per user) followed by a multistage interference rejection algorithm. The
interference rejection algorithm is summarized in the following sentences. Consider
the output of one of the matched filters. If a decision is made on the filter output, it
may be corrupted by the interference caused by the other users. If the receiver has
perfect knowledge of the other users' received signal energies and transmitted data
sequences, the receiver could directly calculate the interference at the output of the
matched filter and subtract this quantity from the filter's output. Then the receiver
could make a decision on an interferencefree quantity, and the decision made for
each user would be optimal. However, the receiver does not have perfect knowledge
of such quantities and, therefore, can only form imperfect estimates of the MAI term
at each filter output. The receiver reconstructs the estimate of the MAI at each
filter output during the mth interval, based on the decisions formed in the (m 1)th
interval. Since the reconstruction of the MAI is based on these decisions, perfect
cancellation of the MAI term from the filter output will not occur (and hence the
receiver is suboptimal).
The previous paragraphs give a brief overview of what has been done in terms of
multiuser detectors for DSCDMA. As mentioned the complexity of the these tech
niques, and the amount of required sideinformation, makes them infeasible. However,
all of the receiver structures mentioned require some form of synchronization with the
user spreading sequences. In the remainder of this chapter, we will focus our attention
on techniques that have been proposed to address the issue of code synchronization.
One area of parameter estimation that is receiving considerable attention in the
literature are estimation techniques based on subspace methods [2326]. While Strdm
et al. [23] consider the more general case of sampling the received signal more than
once a chipinterval, we restrict our attention to the system model given in Section 2.1.
We modify the system model slightly, by assuming that all users can transmit random
data sequences. Using this system model, the receiver is operating in a stationary
environment. The autocorrelation matrix of the received signal, which is defined as
R = E[r(m)rH(m)], can be written as
Kp
kR = P [a2k (pk, 6k)a2ki(Pk,6k) + a2k(pk, k)ak(pk, 6k)] + 2u2I (3.1)
k=1
where I is an N x N Identity matrix. Note that R is symmetric, and is also positive
definite as long as a2 > 0. When 2K < N, the eigendecomposition of R can be
expressed as
R= E8 En E, E, H. (3.2)
0 An
As is a diagonal matrix of the 2K largest eigenvalues of R and E, is an N x 2K
matrix of the corresponding eigenvectors. Likewise, An is a diagonal matrix of the
(N2K) smallest eigenvalues (all equal to ao2) of R and En is an N x (N2K) matrix
of the corresponding eigenvectors. The signal subspace is defined to be the subspace
spanned by the set of a2kl(Pk, 6k) and a2k(pk,6k) vectors Vk G {1, 2,..., K}. The
noise subspace is defined to be the orthogonal complement to the signal subspace. The
columns of A, form an orthonormal basis for the signal subspace, and the columns
of An form an orthonormal basis for the noise subspace.
The basis of the multiple signal classification (MUSIC) algorithm, is that since
the set of {al, a2,..., a2g} vectors are in the signal subspace, they are orthogonal
to the noise subspace. Therefore, given perfect knowledge of the noise subspace, one
could find Trk as the solution to En a2k1 (pk, 6k) = 0. Typically, the receiver does not
have perfect knowledge of the noise subspace or R. However, it can form an estimate
of R using the sample correlation matrix defined as
1 M
M r(m)r'(m) (33)
m=1l
Given RM, the receiver can form an estimate of the noise subspace, En, an operation
that has complexity 0(N3). Strdm et al. [23] use this idea to form a timing based on
the MUSIC algorithm. They find an estimate of the kth user's propagation delay as
arg minm a2k ll2 IHI1H 2 (3.4)
Tre[O,Tb) la2kl(T)jI + jla2k(rT)12
The independent work presented by Bensley and Aazhang [24] is similar, but also
addresses the problem of estimating the channel gain and phase in a timeinvariant
multipath channel. As a natural extension of their earlier work, Str6m et al. [25]
consider using the MUSIC based estimator in a timevarying channel.
The MUSICbased timing estimator requires very little side information in order
to form its estimate. The number of system users, as well as the spreading code (to be
used in equation (3.4)) is required. While the complexity of the algorithm is 0(N3),
it requires no training period, and can be used to estimate each user's propagation
delay. However, the algorithm does not work for 2K > N, as the noise subspace
of R has zero rank. It has been noted that for low signaltonoise ratios and little
nearfar effect, that the traditional correlator performs better than the MUSICbased
algorithm. However, as the MAI interference increases the MUSICbased algorithm
outperforms the correlator. It has also been noted that the MUSICbased algorithm
is resistant to the nearfar problem [23,24].
Zheng et al. [27] have proposed a singleuser propagation delay estimator, using
the system model of Section 2.1, that is known as the largesample maximum likeli
hood (LSML) method. This method models the received signal as a known training
sequence (the desired signal) and all other signals including the interfering signal
and thermal noise as unknown colored Gaussian noise that is uncorrelated with the
desired signal. The resulting timing estimate is found by rooting a second order poly
nomial. The coefficients of the polynomial are dependent on a matrix inverse, Q',
where Q is related to the RM matrix given in equation (3.3). In order to evaluate
the matrix inverse, Q must be full rank. This means that the LSML timing estimate
is unavailable until the receiver has observed at least M = N samples of r(m).
As an extension to the LSML timing estimator, Liu et al. [28] have proposed a
similar timing estimation scheme known as the multipleantenna sensorsbased esti
mator (MASE). That is, the estimator is derived using the known training signal and
modeling the MAI and the additive noise as an unknown colored Gaussian random
process. In the MASE algorithm, the receiver observes the outputs of an arbitrary
antenna array of L sensors. In fact, when L = 1 the MASE algorithm reduces to the
LSML algorithm. Like the LSML, the MASE algorithm requires a training sequence,
and the timing estimate is formed by rooting a second order polynomial. Once again,
the coefficients of the polynomial are dependent on the inverse of a matrix. In the
LSML algorithm, the receiver has to acquire at least M > N samples of r(m) in
order for the Q matrix to have full rank. By using L antenna sensors, the MASE
algorithm is able to average over the outputs of the sensors and hence only requires
LM > N samples of r(m) in order for the Q matrix to have full rank. Since the
MASE algorithm requires fewer training symbols than the LSML algorithm, under
conditions where the channel is timevarying, the MASE has the advantage over the
LSML algorithm [28].
In the next chapter, a timing estimation algorithm is presented that is based on
processing the weights of the MMSE receiver. In this dissertation, it is assumed
that the MMSE receiver uses an initial training sequence to adapt the weights to
minimize the MSE between the filter output and the data sequence for a single user.
This means that a known data sequence is used to train the MMSE receiver, prior
to performing data detection. Honig et al. [29] have developed a singleuser detector
based on minimizing the meanoutputenergy (MOE) of an adaptive filter. Note
that the term MOE in actuality is used to denote the variance of the filter output.
The proposed receiver [29] requires knowledge of the timing and spreading sequence
of the desired user. The weights of the filter are expressed in a canonical form as
w(m) = C1 + x(m), where ci is the spreading code of the desired user and x(m) is
orthogonal to cl. The vector x(m) is updated to minimize the variance of the filter
output. This adaptive algorithm does not require knowledge of the desired user's
data sequence, and, therefore, is known as a blind adaptive algorithm.
Madhow [30] uses the blind adaptive receiver in an adhoc fashion to form an
adaptive receiver which only requires knowledge of the desired user's spreading se
quence. Madhow uses a system model similar to that presented in Section 2.2, but
observes two bitintervals (2Tb) of the received signal, and, therefore, has 2N sam
pled outputs of the chipmatched filter comprise the received vector. The prop
agation delay for the desired user is assumed to be one of 2N hypotheses, T1 E
{0, 0.5Tc, 1Tc, 1.5Tc,..., (N 0.5)To}. The received signal is processed by a bank of
2N parallel blind adaptive filters (one for each 71 hypothesis) during a blind acquisi
tion cycle. Let the ith filter be the adaptive filter that has the lowest MOE at the end
of the acquisition cycle. At the end of the acquisition cycle the receiver chooses two
filters, the ith filter and either the (i 1)th or (i + 1)th filter, to form the adaptive
receiver. The output of the ith filter is combined with the other filter in an adhoc
fashion in an attempt to combine the timing hypotheses that are closest to the true
propagation delay for the desired user. Using this method, Madhow is attempting
to create a receiver structure that performs joint acquisition and detection of the de
sired user's signal. However, this method is very complex and doesn't really achieve
synchronization with the desired user. In the next chapter, we demonstrate why syn
chronization is important by considering the effects of being a halfchip asynchronous
to the desired user.
CHAPTER 4
THE MMSE RECEIVER AND SINGLEUSER CODE ACQUISITION
As an alternative to using the complex estimation techniques presented in the
previous chapter, we present a singleuser estimation algorithm which is based on
processing the weights of an adaptive receiver. One benefit of this receiver structure
is that the only side information the receiver requires in order to form its estimate is
the spreading code of the desired user. In addition, the same structure may be used
for data detection. This receiver structure, which has commonly been referred to as
the MMSE receiver, has received significant attention in recent literature.
4.1 The MMSE Receiver
In this section, using certain system assumptions, we will discuss how the weights
of the MMSE receiver, shown in Figure 4.1, may be used to estimate the propagation
delay of the desired user. The weights of the receiver are chosen to minimize the mean
squared error E [e(m) 12], where e(m) is the difference between a desired response
and the filter's output. The vector which minimizes the meansquared error (MSE)
is wellknown to be given by the WienerHopf [31] equation
w(m) =Rl(m)p(m) (4.1)
where R(m) = E [r(m)rH(m)] and p(m) = E[d(m)r(m)] are the autocorrelation
matrix and the steering vector, respectively.
The received vector r(m) corresponds to chipsamples of one period of the spread
ing sequence. Since the receiver is initially asynchronous to the desired user, it is
possible for data bit transitions to occur anywhere within the length of the received
vector. In order to avoid these bit transitions for the desired user, we assume that the
Cl I
10Estimation i
w(m)* algorithm
Convert to H
r(Ft) baseband r(m) Adaptive r (a)w(n) ptdi(m)
tr i a+chip a enr p of code sgn[Re(.)] A ch
l tch.ied filter, w(m) S 2.e1
matched
filter + ( +
_ A daptive 1 e(m )
algorithm I I (M)
Figure 4.1: Code acquisition with an adaptive receiver.
desired user's data bit is a constant di(m) = 1. In other words, the desired user will
transmit an all ones data sequence for purposes of code acquisition. A sidechannel
is used to control the addition of new users into the system, so this restriction is not
unrealistic. Based on the system model presented in Section 2.1, we see that the filter
is operating in a stationary environment. It is assumed that the users' data sequences
and carrier phases are independent of each other and also independent of the additive
white Gaussian noise. The autocorrelation matrix and the steering vector are given
by the following equations.
R= E [r(m)rH(m)] = a,(pi, 61)aT(pi, 61) + 2a21
K (4.2)
+ P [a2k 1(pk, 6k)a2k _i(pk, 4k) + a2k(pk, 6k)ak(pk, 6k)]
k=2P
p = a (pi, 61) exp(j01) (4.3)
where I is an N x N Identity matrix. The MMSE occurs when the filter weights are
optimal, and is given by
Jmin = 1 pHRlp.
(4.4)
In order to solve the WienerHopf equation for the optimal weight vector, Wopt,
we have to invert an N x N matrix, an operation that has complexity of O(N3).
Since R is a symmetric matrix, it can be expressed as
R =VAVT, (4.5)
where V is a matrix whose columns contain the normalized eigenvectors of R and
A is a diagonal matrix that contains the corresponding eigenvalues. This notation is
referred to as the eigendecomposition of the R matrix. Using the above notation,
we can then express the inverse of the autocorrelation matrix as
R1 = VA1VT. (4.6)
Instead of calculating R1 directly, we note that it can be shown that an eigenvector
of P = R 2oa21 is also and eigenvector of R. For each eigenvector of P, adding
2ca2 to the corresponding eigenvalue produces the corresponding eigenvalue for the
matrix R. Therefore, if we find the eigendecomposition of P we immediately have
the eigendecomposition of R.
In equation (4.2), we note that each asynchronous user contributes at most two
outer product terms in the expression for P. If 2K < N, then the P matrix will
not be fullrank and will have at most 2K nonzero eigenvalues. Miller [32,33] used
this fact to simplify the calculation of the eigendecomposition of P, by working
on a 2K x 2K matrix instead of on a N x N matrix. For large ratios of N/(2K)
this technique produces a significant reduction in the complexity of calculating R1.
However, in that work the desired user was assumed to be synchronous (ri = 0) with
the receiver, so one must slightly modify Miller's notation in order to process the
desired user in the more general asynchronous case. Of course when 2K > N, this
technique is not valid and one must find the eigendecomposition of an N x N matrix
(either R or P).
In order to illustrate the dependence of the filter weight vector on the amount
of multiaccess interference, we will observe the optimal filter weight vector for two
cases. For the first case let there be only one chipsynchronous system user, which
implies that r(m) = c) exp (jOi) +Nm. Using the WienerHopf equation, it is found
that
c. exp(ji) (47
Wopt = ICi2 +22
Assuming cl has good autocorrelation properties, this result shows that we could
conceivably form an estimate of Pi, the propagation delay of the desired user, by
finding the peak of the magnitude of the crosscorrelation between cl and Wopt.
For the second case, a second synchronous user is added to the system such that
r(m) = c1 exp (j01) + d2(m/) /P2/P1c2 exp (j02) + Nm. For this second example,
the optimal filter weight vector is
(c1) ^YC2)exp (j01)
Wopt = C 112 2 T (4.8a)
S+20 ^2 2 C1
cT (Pi)
2 1221 (4.8b)
Ic2H +2 P2
Note that 7y is bounded in magnitude by the value of the correlation coefficient of
the two spreading codes, and is therefore expected to be much less than one. In
fact, in the limit P2/P1 + oo, we see that Wopt is the projection of c(P' onto a
(pl)~~~ ~ ont orhgnat 2,tercie
vector orthogonal to c2. In the ideal case that c(P1' is orthogonal to C2, the receiver
completely rejects the second user even though it is received at a much higher power
level. In the general case, for codes with good correlation properties (c2 c1 is small),
the filter should be able to suppress the second user. In the general case, the addition
of the second user causes a small perturbation of the filter weight vector about the
solution given by equation (4.7). We should still be able to form an estimate of pi as
described above.
Equation (4.8a) is used as a basic example of why a timing estimator based on
processing the filter weight vector should even be considered. As more users are in
cluded in the DSCDMA system, additional terms due to the spreading codes are
added to the expression for Wopt. In order to illustrate the dependence of wopt on the
desired user's spreading code in the above example, and to simplify the resulting ex
pressions, we assumed that the interfering user was chipsynchronous to the receiver.
In the general case, each asynchronous interfering user contributes two similar terms
to the expression for Wopt. However, if the set of spreading codes has sufficiently good
crosscorrelation properties one would expect that Wopt is closely related to cl such
that a reasonable estimate of P, is still feasible.
4.2 Advantages of Synchronizing the MMSE Receiver
Consider using the MMSE receiver in a singleuser system. One might ask why
synchronization of the MMSE receiver is even an issue that must be considered. If
the MMSE receiver is used for detection, are the weights optimized to give us the best
performance? The answer is maybe. The weights are optimized for the given value of
T1, the propagation delay of the desired user relative to the receiver. It is conceivable
that Jmin at T1 = 0 is less than the resulting Jmin when Tr1 i 0. Therefore, we should
prefer to operate the receiver synchronous to the desired user's received signal. In
order to expand on this idea, we present three brief cases that demonstrate some of
the costs associated with using the MMSE receiver asynchronous to the desired user.
For the first case, consider using the MMSE receiver to demodulate data in a
single user system where
r(m) ai(pi, 61) + Nm (4.9)
a,(pi, 61) (4.10)
Wopt a,(p, 6)112 + 2210
Consider the signaltonoise ratio of the output statistic z = rH(m)Wopt, which is
defined as SNRo, = E[z]2/Var(z). Note that E[z] = af(pi,6i)wopt, and the expres
sion for Var[z] is evaluated using Var[z] = E[wOtr(m)rH(m)wopt] E[z]2. Direct
substitution of the appropriate terms into this expression produces
Vr2 22 Ial(pi, 61)112
Var~z] 2
(Ilal(pi, 61)112 + 2a2) (4.11)
2o2E[z]
Il ai(pi, 61)112 + 2U2
The result of these expressions is that SNRo = Ilal(pi, 61)112/(2o2). If we assume
that the spreading code has good autocorrelation properties such that the terms
involving c"c1 are negligible when compared to other terms in Iai(pi, 1)12, we can
approximate the output signaltonoise ratio as
SNRo [(1 61)2 ]1 (4.12)
2o2
where o72 = N/(2Eb/.fAo) as previously defined. When 61 is nonzero, there is an
effective loss in the output SNR of the output statistic z. If 61 is uniformly distributed
over the interval [0,1), then the average loss in the output SNR is 2/3 (1.8 dB). This
loss will adversely affect the capacity of the communication system which we typically
want to maximize. Note that this effect can be offset by sampling the received signal
more often than once a chip interval. However, increasing the sampling rate directly
increases the length of the received vector, which leads to an increase in the complexity
of the receiver.
Madhow [34] presents a method where an adaptive receiver is used to perform
joint detection and acquisition by increasing the length of the received vector and the
filter to 2N chip samples. By increasing the length of these vectors to 2N elements,
the desired user does not have to transmit an all ones training sequence. In addition
it may be possible for more than one user to be in the training mode at the same
time. The receiver structure performs joint detection and acquisition without actually
achieving chip synchronism with the desired signal [34]. It is important to ask how
the performance of the MMSE receiver is affected by the lack of chip synchronism.
To address this issue, we calculated the Jmin of the MMSE filter as a function of the
propagation delay of the desired user when the SNR was 10 dB and the length of
the received vector and filter was set to 2N chip samples. This result is shown in
Figure 4.2 for propagation delays in the range of [0, 5) chip intervals. Similar results
are obtained for propagation delays in the range of [5,31) chip intervals but were
omitted for clarity of the plot. There are two things to notice in this plot. If one
were to consider a system that is nearly chip synchronous then it appears that the
performance of the filter, based observing the MMSE, is unaffected by the propagation
delay of the desired user. This result could lead one to the incorrect conclusion that
achieving code synchronization is not important for the MMSE receiver. However,
when the filter is not chip synchronous with the desired signal, the value of 61 falls
in the region (0,1), the MMSE of the filter is sensitive to the propagation delay of
the desired user. Therefore, some method is required for code acquisition and code
tracking even when 2N chip samples are used in the MMSE receiver structure.
As a final example, we will consider the effect of asynchronous operation (with
respect to the desired user) on the resulting probability of bit error. This example
will show that code synchronization is required for the MMSE receiver if one is to
maximize the system capacity. The following briefly explains the outline of the analy
sis used to derive the probability of bit error. Since we are interested in bit errors, we
assume that the filter is at steadystate, and that the desired user is transmitting a
random data sequence. We then model the output of the MMSE receiver, wotr(m),
as a complex Gaussian random variable, and derive an expression for the probability
of bit error. The details of this derivation are given in Appendix A. Modeling the
output of the MMSE receiver as a Gaussian random variable has been recently jus
tified by Poor and Verdi [35], who have compared the bit error rate of the MMSE
0.1 
0.09
0.08
I0.07 \
0.06
0.05
0.04  I
0 1 2 3 4 5
Propagation Delay of User #1 in Chips
Figure 4.2: Minimum meansquared error as a function of the propagation delay of
user #1.
receiver versus the analytical bit error rate of the MMSE receiver when the receiver
output is modeled as a Gaussian random variable. It was found that the Gaussian
approximation was quite good for various number of users and various levels of multi
access interference. In addition, several others have found that this approximation
provides reasonable results [13,36].
Once the DSCDMA environment is configured, # users, delays, phases and power
levels, we can numerically solve for Jmin. We then use equations (A.9) and (A.10)
to find the probability of bit error. By varying the DSCDMA environment, we may
observe the effects of asynchronous operation on the system capacity. For purposes
of this example, we set the SNR to 7 dB and let all interfering users be received at
a power level 10 dB above the desired user. A set of spreading codes for 33 users
were selected at random from a set of Gold codes [5] using N = 31 chips/bit. A set
of 33 propagation delays were chosen from a uniform distribution over the interval
[0, 31). A set of 33 carrier phaseshifts were selected from a uniform distribution over
the interval [0, 27r). If we wanted to analyze a Kuser system, the first K elements
of each of the spreading code, delay, and phaseshift sets would be used to create the
DSCDMA environment.
For the case when the receiver was onehalf chip out of synch with the desired
user, using the Gaussian approximation we found that 4 system users would produce
a 2% biterror rate at the receiver. A total of 9 system users would produce a 5%
errorrate. If the receiver was synchronous with the desired user, a total of 14 users
would produce a 2% biterror rate. Likewise, a total of 24 users would produce a 5%
biterror rate. This shows that a significant improvement in the system capacity can
be achieved if the receiver is synchronous with the desired user. While these results
are only valid for case we examined, they are sufficient to demonstrate the costs of
operating the receiver asynchronous to the desired user.
4.3 Adaptive Filters and the Estimation Algorithm
The complexity of finding the optimal weights, given by the WienerHopf equation,
is 0(N3). However the receiver requires knowledge of the autocorrelation matrix R
and the steering vector p. Typically, these parameters are estimated by the receiver
based on observations of the received signal. In practice, the weights of the filter
are chosen adaptively according to some implementation which usually takes the
form of either the least mean square (LMS) or the recursive leastsquares (RLS)
algorithm. The LMS algorithm has low complexity, 0(2N), does not require any
matrix inversion operations, and is easy to implement. The RLS algorithm is more
complex, O(N2), and performs an iterative matrix inversion by using the Matrix
Inversion Lemma [31]. Since the RLS algorithm is more complex, it is not surprising
that it typically converges faster than the LMS algorithm. While these algorithms
will not produce the same weight vector given by the WienerHopf equation, as the
filter approaches steadystate, they can provide reasonable approximations to the
optimal weight vector.
Our basic idea is to derive a code timing estimation algorithm for a singleuser
system, and then study the performance of the estimator in a multiuser environ
ment. Since the receiver is initially asynchronous to the desired user, r(m) =
a (pi, 61) exp (j1) + Nm. In order to derive a code estimation algorithm, we would
like to have an exact representation of the statistics of the filter weight vector as
the filter adapts. This appears to be intractable. Instead, we model the filter coef
ficients as independent jointly complex Gaussian random variables with a mean of
/3a (pi,5 61) exp (ji) and equal variances, where /3 is some unknown constant. We
choose this expression for the mean since for this scenario the optimal weight vector
is proportional to a (pi, 61) exp (j01). Clearly, this is not an exact description of the
statistics of w(m), and our resulting timing estimator will not produce an optimal es
timate of the desired user's propagation delay. The statistics of the weight vector have
been observed (through simulations) as a function of time when an LMS algorithm
was used, and the LMS stepsize was chosen such that the filter was convergent in the
meansquare. We observed that the components of the filter weight vector become
highly uncorrelated as the filter approaches steadystate. Therefore this assumption
on the statistics of the filter weights seems to be a reasonable approximation and
hence can be used to derive a meaningful timing estimator.
Based on our assumed statistical model for the filter weight vector, we find that
we should choose as our estimates the set {8, /3, p, 6} which minimize
g(, 0, p, S) = Iw(m) / ai (p, 6) exp(jo)112 (4.13)
with p E {0,1,...,N 1} and 0 < S < 1. Note that even though our assumed
mean is based on optimal values, we will use this function to make estimates after
every receiver bit interval. This approach is taken because it can be shown that
when the LMS algorithm is used, the multiuser transient mean filter weight vector
E[w(m)] is highly correlated with the single user optimal weight vector when a small
adaptation stepsize is used. Due to the constraint on the value of p, g(/13, 0, p, 6) is not
differentiable with respect to p. To minimize g(3(0, 0,p, 6), we must find the solutions
to 9g(3, 0,p, 6)/9x = 0 (for x =/3, 0,6), for all possible values of p. Therefore, each
value of p produces a corresponding set of candidates for the subset {0, /3, 6}. Our
estimation algorithm iterates through all possible values of p and produces a list of
all of the corresponding candidates. By choosing the set {0, /3, p, 6} out of this list
which minimizes g(/3, 0,p, 6), we find the global minimum of g(/3, 0,p, 6).
The solutions to Og(/3, 0,p, 6)/9x = 0 for x = 0,/3 are given by
0i (p, 6) = arg (aT(p, 6)w(m)) (4.14a)
/(p~, 6) ',a(p, 6)w(m)I
0lai(p, 6)112 (4.14b)
Substitution of these expressions into the previous equation produces the following
two expressions.
g(o(p, 6),0(p, 6),p,6) = IIw(m)112 f(p,6) (4.15a)
f(p,6) = (p,6)w(m)12 (4.15b)
Ilal(p, 6)112
These equations show that minimizing g(3(p, 6), 9(p, 6),p, 6) corresponds to finding
the set {p, 6} which maximize the cost function f(p, 6).
Since the estimate of pi must be an integer, the cost function f(p, 6) is a piece
wise continuous function. Maximizing f(p, 6) with respect to 6 produces a quadratic
equation whose coefficients depend on the value of p. Therefore, for each possible
value of p we must find the roots of the quadratic equation D2 (p) t2+ D1 (p) + D0o(p),
where the coefficients are given by the following set of equations.
Do(p) = NA1(p) + 2CAo(p) (4.16a)
D1(p) = 2NA2(p) 4CAo(p) (4.16b)
D2(p) = 2C(A2(p) + Ai(p)) (4.16c)
C = N c c1) (4.16d)
Ao(p) = Iw (m)c P)2 (4.16e)
Ai(p) = 2 Re [(cP))Tw(m)w(m)c]P+1) 2Ao(p) (4.16f)
(P+1) 12
A2(p) = IwH(m)ci 2 Ai(p) Ao(p) (4.16g)
By definition, the value of 6 must satisfy the condition 0 < 6 < 1, so we must ignore
any solutions from the previous equation which fall outside of this region. In order
to clarify how we form our estimate of the propagation delay for the desired user, we
present the following algorithm.
Let 7T represent the set of costs corresponding to the set of candidate estimates
of (p, 6), denoted by U. We can find the maximum of f(p, 6) as follows:
Step 1. Let T = {f(0,0),f(1,0),f(2,0),..., f(N 1,0)}.
Step 2. Let U = {0,1,2,..., N 1}.
Step 3. For p = 0,1,... N 1, do the following:
a) Compute the coefficients D2(p), Dj(p), Do(p).
b) Solve for the two roots of the quadratic equation, pi and A2.
c) If 0 < pi < 1 for i {1,2}, add the cost f(p, p) to the set T, and add
(p + pi) to the set U.
Step 4. Let Uk denote the kth element of the set U. Then f = Uk, where k = maxTj.
3
Step 5. The estimates of p, and 61 are given by Pil(m) = [fj and Si(m) = Lfj,
where [xj rounds x to the nearest integer towards zero.
We plot the cost function, f(p, 6), for two different scenarios in Figures 4.3 and 4.4,
under the assumption that the receiver is operating using the optimal filter weights.
Figure 4.3 shows the cost function for a singleuser system when r1 = 15.25T, and
Eb/.Afo was 7 dB. This plot shows that the timing estimate will be f, = 15.25Tc, which
is a perfect estimate. Since we have developed the timing estimator for a singleuser
system this result is not surprising. Of course, when the filter weights are from the
0 1 I I I I I I I I I I I I I I I I I I
0.08
0.06
too
^ 0.04
0.02 
0 ./. ,,y
0 5 10 15 20 25 30
Tested Propagation Delay
Figure 4.3: Plot of the cost function f(p, 6) for a singleuser system.
LMS or RLS algorithms we expect some noise in this estimate due to the noise in the
filter weights. Also, the rate of convergence of the adaptive filter to its steadystate
value will determine how long the filter has to adapt before a reliable estimate of T1
can be made. This issue will be addressed in the next chapter.
Figure 4.4 shows the cost function, f(p, 6), for a more severe environment. In this
figure 14 interfering users have been added to the system, each received at a power
level 10 dB above the desired user. Since the maximum value of f(p, 6) does not
occur near 15.25Tc, the timing estimate will be wrong. In this example f = 7.43Tc.
We have intentionally heavilyloaded the DSCDMA system, to demonstrate what
will happen to the cost function. In general, we can say that at some level of MAI,
the estimation algorithm completely breaks down. We will study the effects of the
level of the MAI on the performance of the timing estimator in the next chapter.
When the simplified system model presented in Section 2.2 is used, the timing
estimator takes on a slightly different form. Recall that the simplified model assumes
that all users are received phasesynchronous, and the weights of the adaptive filter
0 .0 3 1 , i
0.025
0.02 
0.015
0.01
0.005
0
0 5 10 15 20 25 30
Tested Propagation Delay
Figure 4.4: Plot of the cost function f(p, 6) for a fifteenuser system.
are purely real. Using this assumption the autocorrelation matrix and the steering
vector for the adaptive filter are written as
R = E [r(m)rH(m)] = a(P, (l)a (P, i) +
K^ (4.^17)
+ K Pk [a2k 1(pk, 6k)ak (Pk, 6k) + a2k(pk, 6k)ak(pk, 6k)] (4.17)
k=2
p = a(pi, 651) (4.18)
where I is an N x N Identity matrix.
Having the weights of the filter real leads to a small reduction in the complexity
of forming a timing estimate for the desired user. The derivation of the algorithm is
similar to that previously presented. Maximizing the cost function, f(p, 6), produces
a quadratic equation whose roots may be written in closed form. However, one of the
roots produces a trivial result and therefore can be ignored. The other root is given
by the following equation.
NwT(m)c(P+l) T (1) T M)C(P)
S(mc1 1 (4.19) m)
S(N "I"TC(1))(wT(m)c(P+1)'+ wT(m)c~p))
S(N 1 1rcKw 1? + W I)^
Based on the above equation, the timing estimation algorithm is then modified as
follows:
Step 1. Perform steps 1 and 2 of the previous algorithm.
Step 2. For p = 0,1,..., N 1, do the following:
a) Use equation (4.19) to compute the coefficient p1.
b) If 0
set U.
Step 3. Perform steps 4 and 5 of the previous algorithm.
4.4 Summary
The main contribution of this chapter is the presentation of a timing estimator for
a singleuser, based on processing the weights of an adaptive filter. We motivate why
these weights may be used to form a timing estimate by studying how the optimal
weights depend on the DSCDMA environment for two simple cases. The MMSE
receiver has received considerable attention in the literature. One fact that seems to
be ignored by several authors is that there are performance penalties to be paid for
operating the receiver asynchronous to the desired user [34]. Therefore, one may come
to the wrong conclusion that no advantage is offered by synchronizing the receiver
with the desired user. We present three examples as to why operating the MMSE
receiver synchronous to the desired user is important. Certainly, when the system
capacity is considered we must synchronize the receiver with the desired user, if we
want to support as many users as possible in the DSCDMA system.
One may argue that the form of the timing estimator is complex since it depends
on, N, the number of chips/bit. However for large N, even using the traditional
correlator for timing estimation becomes complex. In fact, the correlator uses the
36
exact same estimation algorithm that we have presented in this chapter. The differ
ence between our technique and the correlator, is that we form an estimate based on
observing the filter weights while the correlator forms an estimate by observing sam
ples of the received signal. The additional complexity added to the receiver by our
technique is that of running the adaptive filter (LMS O(2N), RLS 0(N2)). Since the
same adaptivefilter receiver structure can be used for detection, and we have demon
strated that synchronization is desired, including the timing estimator in the receiver
should be an acceptable increase in the overhead of implementing the receiver.
CHAPTER 5
CHARACTERIZING THE PERFORMANCE OF THE
TIMING ESTIMATOR
In this chapter, several methods will be presented in an attempt to characterize the
performance of the adaptive filterbased timing estimator in an AWGN channel. We
are interested in developing analytical tools that can be used to obtain performance
measures on the quality of the timing estimate in hopes of avoiding long Monte
Carlo simulations to get such performance metrics. We first consider a bound on the
probability of correct acquisition when we are only interested in correctly estimating
the chip level timing. Next using traditional techniques, an approximation to the
lower bound on the conditional variance on the timing estimator will be derived. This
expression will be developed by deriving the Cram&rRao bound (CRB) for a fictitious
timing estimator, once several assumptions are made with respect to the statistical
properties of the LMS filter weights. In order to test the quality of these analytical
expressions, they will be compared to simulation results under several different multi
user scenarios. Finally, the coarse acquisition performance of the timing estimator
will be observed through MonteCarlo simulations. The acquisition performance of
the timing estimator will be observed for both the LMS and RLS based adaptive filter
architectures. The effects of multiaccess interference on the timing estimator will be
observed for several different scenarios.
5.1 Transient Statistics of the LMS Filter Weights
In order to derive analytical performance metrics on the timing estimator, we
first require knowledge of the statistics of the filter weights as a function of time.
When we developed the timing estimator, we assumed that the filter weights were
independent and jointly Gaussian random variables. Using these assumptions, one
could then completely define the assumed statistics of the filter weights by defining
the filter weight mean vector and the covariance matrix for the filter weights. Since
the filter is adaptive, these quantities will be dependent on the initial conditions of the
filter as well as the length of time that the filter has adapted (and of course somehow
depend on the statistics of the input signal). The next few paragraphs describe how
the statistics of the the LMS filter, using the system model presented in Section 2.2,
may be derived using several assumptions about the filter and its input signal.
One area that has received a fair amount of attention in the literature is the
tracking performance of the LMS filter [31,3742]. The main subject that has been
addressed in these works, is what bounds must be placed on the LMS stepsize, A, such
that the filter has desirable convergence properties. One of the properties of interest,
is the convergence of the filter weights about the WienerHopf solution. The filter
is said to be convergent in the mean if lim _. E[w(n)] = wopt, the solution to the
WienerHopf equation. As an extension of this idea, even if the filter is convergent in
the mean, it would be very desirable to have a finite variance of the filter weights about
Wopt. The variance of the filter weights about the WienerHopf solution determines
the level of the meansquared error at the filter output. As one would expect, the
bound on p required to minimize the meansquared error of the filter output is less
than the value of p required for convergence in the mean. An intrinsic part of the
analysis on the limits of p for such convergence properties, and what we are really
interested in, is a set of equations that describe the characteristics of the filter weights
as a function of adaptation time.
Senne [37] presented a set of iterative equations to describe the timedependent
filter weight mean vector and the filter weight correlation matrix for an LMS filter
when the filter input was a vector of zeromean Gaussian random variables. However,
in the system model used in Section 2.2, due to the desired user's contribution to the
receiver input, the filter input signal is not zeromean. We can apply the model used
by Senne to derive a similar set of equations to describe the transient statistics of
the filter weight vector. In order to begin the derivation, we make several assump
tion regarding the relationships between the filter weight vector and the input signal.
These assumptions are referred to as the fundamental assumption [37] or indepen
dence assumption on the filter statistics and are included here for convenience. The
fundamental assumption consists of three parts which are:
1) Each receiver input vector r(m) is statistically independent of all previous input
vectors and all previous desired filter outputs di (m).
2) Each desired output dj(m) is dependent on the corresponding input vector r(m)
but is statistically independent of all previous desired outputs and received vectors.
3) All desired quantities and received vectors are mutually Gaussiandistributed ran
dom variables.
In many applications of the LMS filter, the filter input vector is sampled at the
same rate as the output of the filter. Therefore, the first part of the fundamental
assumption is clearly not valid in this situation since the elements of r(m 1) are
shifted by one position in creating r(m). This weakness in the fundamental assump
tion under these conditions is what has driven the study of the convergence properties
of the LMS filter. Of noticeable interest is the work by Douglas [42], in which a nu
merical algorithm is presented to observe the exact transient statistics of the MSE
convergence of the filter. However, this algorithm is very complex in that a set of
linear equations must be dynamically created and solved based on the number of taps
in the LMS filter. As an example of the complexity of this algorithm, for a fivetap
filter, a set of threethousand equations were required to solve for the transient MSE
of the filter. Based on this example, applying to this method to a filter with more
taps does not seem reasonable.
For the adaptive receiver, we have assumed that the input to the filter is sampled
at a rate N times faster than the filter output. This means that r(m) could truly be
statistically independent from all previous receiver inputs. Certainly the contribution
to r(m) from the AWGN is independent from all previous receiver inputs. However,
two symbol intervals of each asynchronous user contribute to r(m). So the first part of
the fundamental assumption, while not being true, will still be asserted and we use the
fundamental assumption to derive the desired equations. We begin the derivation of
the filter transient weight mean vector by manipulating the general LMS filter update
equation given below
w(m + 1) = w(m) PU(m)uT(m)w(m) + pIu(m)d(m) (5.1)
where u(m) is the filter input signal. Using the system model defined in Section 2.2,
u(m) = i(m) defined in equation (2.14), with the caret dropped for notational sim
plicity in future equations.
We note that the desired filter output is not timedependent, and will therefore
drop the dependency on m for this quantity. Taking the expectation of both sides of
equation (5.1) will produce the desired filter transient weight mean vector
E[w(m + 1)] = w(m + 1)
(5.2)
= {I LRrr} w(m) + pfrd
where I is an N x N identity matrix, and Rr = E[r(m)rT(m)] is used for notational
convenience. Once we define the initial conditions on the filter wV(0), realize that
r = a (pi, 61) and let d = 1, we can use the previous equation recursively to describe
the filter transient weight mean vector.
In order to develop an iterative equation for the filter transient weight covariance
matrix, equation (5.1) will be used to find an iterative equation for the filter transient
weight autocorrelation matrix defined as Rw,(m) = E[w(m)wT(m)]. This derivation
is described in Appendix B, but the final result is listed below.
Rww(m + 1) = Rww(m) + Pw(m)iTd lR.(m)R,,
+ trwT (m)d Rrr..ww(m) + 2rrd2
2p2o2 {w1 (m)rI + T (m) + w(m)T I} d
2A2 r*T(m)rT d + 22a02Rww(m)rrT (5.3)
+ 22o2a4R1,(m) + P2a4 trace(Rl,,(m))I
+ l a22TpRu(m)I + 4A2a2 trace(Rl,,(m))rrT
2p2a2rrTR.w(m) + i2rrTRw,(m)rrT
Given the above equation, we can easily define the transient filter weight covariance
matrix using the following equation once we define w(0), R1,(O0), r and Rrr.
Cww(m) = R (m) *(m)wT(m) (5.4)
As an alternative to equation (5.3), we also consider using the transient weight
error covariance matrix which describes the deviation of the filter weights about the
WienerHopf values as a function of time [31]. This is the same equation used by
Miller [32,33] to study the transient MSE response of the LMS filter. The transient
weight error covariance matrix is defined as
K(m) = E [(w(m) Wop0t)(w(m) woPt)T] (5.5)
and can be evaluated at the mth interval using the following iterative equation.
K(m + 1) = K(m) + p[RrrK(m) + K(m)R.,] + '2Rrr trace(RrrK(m))
(5.6)
+ p2 RrK(m)Rrr + A2JminRrr
In order to use equation (5.6) we note that K(0) = WptWoptw, when w(0) = 0, and
Jmin = 1 Woptali(Pi, 61). The transient weight covariance matrix can be found as
shown in equation (5.7).
C.(m) = E [(w(m) *(m))(w(m) w(m))T]
L J (5.7)
= K(m) w(m)*T(m) WoptWoTpt + (m)W Woptw T(m)
Several different DSCDMA environments were simulated, and the transient mean
weight vector and transient autocorrelation matrix of the LMS filter weights were
observed at the mth interval of adaption using the following equations.
1 M
w(m) = ^w(m) (5.8)
i=i
1 M
Rw(M) = M w(m)wT(m) (5.9)
i=1
These quantities were then used to form an estimate of the transient autocovariance
matrix of the LMS filter weights as
Cww(m) = ,(n) W(m)W T(m). (5.10)
Of course, for each trial of a given test, the number of users, the spreading sequences
and propagation delays are fixed. The data sequences for the interfering users and
the AWGN were random during each trial, and independent from trial to trial. The
LMS stepsize, [, was chosen such that the filter was convergent in the meansquare.
By comparing the results produced by the above equations to the corresponding
analytically derived quantities, we can in some sense measure the quality of the
analytical expressions.
For the first comparison, a singleuser environment was simulated using 31 chips/bit
and a SNR of 10 dB, and 71 = 15.5Tc. A set of 3,000 independent trials was used
in computing equations (5.8) through (5.10). Figure 5.1 compares the IIE[w(m)]
produced by equation (5.2) to the observed simulation results. In order to compare
these two quantities directly, the error is plotted as IIE[w(m)] w(m)I in the second
part of the figure.
0.3
0.25
0.2
0.15
0.1
a) Magnitude of Mean vectors
r  I  
IAf' 2 Of i
IU ZU JU V iU ou
b) Magnitude of Difference
Ar
IU 1ou U iU
0 10 20 30 40 50 60 70 80 90 1
# of Training Bits
Figure 5.1:
Performance of equation (5.2) for a singleuser system, a) The analytical
mean filter weight vector norm, E[w(m)] (*), and the simulated mean
filter weight vector norm Iw(m)I (solid line) plotted as a function of the
training length; b) Comparing the error between the two mean weight
vectors, IIE[w(m)] wA(m)11.
As a second test, a threeuser system was simulated. As before the desired user
was received at a SNR of 10 dB and the two interfering users were received at a
power level 10 dB above the desired user. The interfering users had propagation
delays of Tr2 = 6.257Tc and T3 = 28.0125Tc. This environment was simulated using
3,000 independent trials. Figure 5.2 compares equation (5.2) to the simulation results.
Figures 5.1 and 5.2 show that the analytical expression for the transient mean weight
vector of the LMS filter given by equation (5.2) closely tracks the corresponding
simulation results. Several other multiuser scenarios with various levels of SNR and
MAI were simulated, and the performance of equation (5.2) was similar to those
shown in Figures 5.1 and 5.2. Therefore, we use equation (5.2) in the sequel when we
require use of an expression for the LMS filter's transient weight mean vector.
1
" 1(4
4
3.5
0 20 40 60 80 100 120 140 160 180 200
# of Training Bits
Figure 5.2: Performance of equation (5.2) for a threeuser system, a) The analytical
mean filter weight vector norm, IIE[w(m)] (*), and the simulated mean
filter weight vector norm IIw (m) (solid line) plotted as a function of the
training length; b) Comparing the error between the two mean weight
vectors, IIE[w(m)] wA(m)11.
A similar test was used to characterize the performance of equations (5.4) and (5.7),
which are the analytical expressions for the LMS filter's transient weight autocovari
ance matrix. We observed that for a singleuser environment, both equations gave
similar results. However when multipleusers were present, equation (5.4) initially
tracks the simulation results for C1w(m), but at some point diverges from the simu
lation results. However, equation (5.7) was observed to produce similar results, when
compared to Cw,(mn), regardless of the number of users and the level of the MAI. A
twouser system with the desired user received at a SNR of 10 dB and the interfering
user received at a power level 10 dB above the desired user was simulated for 3,000
trials. The results of comparing the Frobenius norm of equations (5.4) and (5.7) to
the Frobenius norm of the simulated Cww,(m) matrix are shown in Figure 5.3. Note
that similar results were observed for other multiuser scenarios. We have observed
x 104
4
3K 3K *
3.5 yt I '
3 W )K Method #1
Sxx x x X Method #2
3 '. "x x + Simulation Results
2.5 x
2
1.5
I
0.5 ''
0 20 40 60 80 100
# of Training Bits
Figure 5.3: Performance of equations (5.4) and (5.7) for a twouser system. The
Frobenius norm of the LMS filter's transient weight autocovariance ma
trix; equation (5.4) (*), equation (5.7) (x) and simulation results (+).
that the Frobenius norm of the difference between equation (5.7) and the simulated
Cw(m) is on the order of 1020% over several different multiuser scenarios.
We have derived a set of analytical expressions, equations (5.2) and (5.7), that
we will now use to describe the transient model of the LMS filter weights under the
assumption that the weights are jointly Gaussian. Since we are using the weights of
the filter to form a timing estimate of the desired user's propagation delay, we are
interested in how the transient response of the filter weights determines the quality of
the resulting timing estimate. In the next two sections, we use our analytical model
for the transient weight statistics, in an attempt to characterize the relationship
between the filter weights and the quality of the timing estimate.
5.2 Chip Selection Error Probability
We are interested in developing an analytical expression for the chip selection
error probability when the fractional part of the propagation delay is ignored. That
is, we want to find an expression for Pr(P1i pi) when the value of 61 is ignored. In
this case, the propagation delay estimate at the ith iteration of the adaptive filter is
formed as shown below.
Pi,=argmaxF(p) p E (0,1,2,...,N1) (5.11)
p
P
F(p) = IwT(i)c1P)I2 (5.12)
Therefore, the chip selection error probability can be expressed using a union bound
[43] as
N1
Pr(fi 5 Pi) = E Pr(F(pi) F(k) < 0). (5.13)
k=O
ktpi
Note that each term in the previous equation can be expressed as
F(pi) F(k) = w(i) (cic (ck) w(i) (5.14)
(5.14)
= wT(i)Akw(i).
Using our assumption that the filter weights are jointly Gaussian, the expression in
equation (5.14) is just a quadratic form of Gaussian random variables. This type
of expression is common in radar applications, weapon control systems and is also
intrinsic to any discussion of variances of random variables [44]. So we now must
concern ourselves with how to evaluate Pr (wT(i)Akw(i) < 0). This type of evalua
tion has received considerable attention by many researchers [4550]. The next few
paragraphs will summarize various methods that may be used to handle this problem.
As shown in Appendix C, the distribution of wT(i)AkW(i) can be shown to have
the same distribution as
n
Y = Z i(Wi bi)2 (5.15)
i=
where the Wi are independent Gaussian random variables of zero mean and unit
variance and the terms Ai and bi depend on the transformation matrix Ak as well the
47
covariance matrix of the filter weights. Deriving the distribution of equation (5.15)
directly does not appear to be mathematically tractable. However, deriving the
corresponding characteristic function is relatively simple. For a random variable that
has a probability density function of fx(x), the characteristic function is given by
+00
x(w) = E[exp (jiwx)] =f fx(x) exp(jwx)dx. (5.16)
00
Likewise, if one is given the characteristic function Dx(w) then the corresponding
probability density function is found using the inversion formula stated below.
+00
fX (x) x (w) exp (jwx) dw (5.17)
00
The characteristic function for equation (5.15) is found to be
( 1 n 2) 1i n b2 n 1 (5 18
Y(w) = exp 2 b exp 2 1 1(5.18)
2 i=1 \2 i=1 2jwA/=1 v1 2jwAi
Johnson and Kotz [45] provide an extensive summary of techniques that can be
used to find fy(y) or the Pr(y < t), where t is some given threshold, when the transfor
mation matrix Ak is positive definite. For specific restrictions on the bi and A, in the
characteristic function, several expansions of Dyu(w) into power series representations
are presented. Then, the inversion formula in equation (5.17) is applied to each term
in the series to produce fy(y). A matrix A is positive definite when xTAx > 0 for
any nonzero vector x. Therefore, if the transformation matrix Ak is positive definite
then a chipselection error can never occur and, therefore, Pr(Y < 0) = 0. We must
determine if the transformation matrix is positive definite. A sufficient condition for
positive definiteness, is that all of the eigenvalues of the matrix Ak are positive [51].
Without actually computing these eigenvalues, we can test for the possible existence
of negative eigenvalues. It is known that the sum of the eigenvalues is equal to the
trace of the matrix. Since all of the diagonal elements of Ak are zero, there must
exist negative as well as positive eigenvalues and therefore the transformation matrix
is not positive definite.
Rice [46] provides several numerical techniques that can be used to evaluate
Pr(Y < t) through applications of equations (5.18) and (5.17) for a general trans
formation matrix. However, it is noted that these techniques can exhibit slow con
vergence due to the oscillatory nature of the exponent in equation (5.18). In order
to improve the convergence rate of the numerical integration techniques, a change of
variables that produces a tilting of the integration path is presented. However, the
change of variables is inversely dependent on the desired threshold t, and is therefore
not valid when t = 0, which is the case at hand.
A quadratic form of independent Gaussian random variables, such as shown in
equation (5.15), can also be shown to have the same distribution as a linear sum of
independent noncentral X2 random variables [44]. That is, equation (5.15) can also
be represented by
r
Y = jXj+ aX0, (5.19)
j=1
where Xj are independent random variables, having noncentral X2 distribution with
nj degrees of freedom and noncentrality parameter 6? for j = 1,..., r and X0 having
a zeromean unitvariance Gaussian distribution. Davies [50], based on inverting
the corresponding characteristic function [49], presents a numerical technique that
can be used to find the distribution of a linear sum of independent noncentral X2
random variables, when the original quadratic form has a general (not necessarily
positive definite) transformation matrix. The technique involves evaluating a series
of exponential terms, where the truncation error is dependent on the number of
terms included in the series. The number of terms required for the integration is
determined approximately by the total number of degrees of freedom and the sum
of the noncentrality parameters as well at the value of t, at which the distribution
function is to be evaluated. Davies presents a table that list the number of terms
required for integration as a function of these parameters. This table shows that as t
approaches zero, the number of terms required for integration quickly increases from
several hundred to tens of thousands.
Therefore, the techniques of Rice or Davies appear to be applicable to a situation
where the Pr(Y < 0) has to be evaluated very infrequently. Recall that at the
receiver's ith interval, we have to evaluate Pr(Y < 0) a total of (Nl) times to form a
bound on the chipselection error probability. If we want to evaluate the chipselection
probability as a function of the LMS filter's adaptation time, these techniques become
numerically prohibitive. That is, we found that MonteCarlo simulations of a specific
DSCDMA environment produce results for the chipselection error probability in less
time than is required to evaluate the same quantity using one of Rice's techniques for
the same environment.
As a means of evaluating Pr(F(pi) F(k) < 0), we instead turn to the moment
generating function. Instead of trying to evaluate the desired expression explicitly,
we use a Chernoff bound to find an upper bound on Pr(F(pi) F(k) < 0) for each
term in equation (5.13) (see Appendix C for details). We then sum these (N 1)
Chernoff bounds to find an upper bound on Pr(pi 54 pl). An example of the resulting
bound using this method is shown in Figure 5.4. A threeuser DSCDMA environment
was simulated with a SNR of 10 dB, and both interfering users were received at a
power level 10 dB above the desired user. The propagation delays for the users were
71 = 15T7, T2 = 6T, and T3 = 28Tc. The LMS stepsize, M, was set to 0.1/(the total
input power) such that the filter was convergent in the meansquare.
There are three curves present in Figure 5.4. The first curve is the simulated
results for the incorrect chipselection probability, which is plotted using a plus sign
(+). The other two curves are the resulting upper bound for this probability, for
two different applications of the Chernoff/Union bound technique. The solid line,
50
represents the upper bound when equations (5.2) and (5.7) are used to describe
the transient mean vector and autocovariance matrix of the LMS filter weights as a
function of the training interval. Note that the resulting upper bound is not very
tight. It seems appropriate to wonder if the looseness of the bound is due to the
Chernoff/Union bound technique, or due to errors present in the model given by
equations (5.2) and (5.7). As a control for this test, the third curve (*) represents the
upper bound found by using the transient mean vector and autocovariance matrix
of the LMS filter weights obtained directly from the simulation. This is slightly
closer to the simulation results, but the upper bound is still very loose. Therefore,
is seems that the application of the Chernoff/Union bounds are the main reason
that the derived upper bound is very loose when compared to the actual simulation
results. Note that this behavior was observed for several other multiuser DSCDMA
scenarios. Unfortunately, based on this result this technique is not very useful in
terms of characterizing the performance of the timing estimator.
5.3 Approximation for the Conditional Variance of the Timing Estimate
A traditional technique that is used to characterize the performance of an esti
mator is to compare the variance of the estimate to the CramerRao bound (CRB).
The CRB is a lower bound on the variance of any unbiased estimator for a given log
likelihood function [52,53]. In this section, we will use the traditional CRB technique
along with several assumptions, in hopes of forming an approximation to the perfor
mance of our timing estimator. Since we are interested in characterizing the transient
performance of our timing estimator, we base our technique on the timedependent
statistical model for the LMS filter weights given in equations (5.2) and (5.7).
Since we have assumed that the filter weights are jointly Gaussian, equations (5.2)
and (5.7) are sufficient to completely define the timedependent statistical model for
the LMS filter weight vector. We do not intend to use this model to derive a new
timing estimation algorithm. However, we use this model to derive a CRB, hoping
+++++
J ulf + + + + + +
103
20 25 30 35 40 45 50 55 60
# of Training Bits
Figure 5.4: Comparing incorrect chipselection probability to the Unionbound for a
threeuser system. Simulation results (+), the Chernoff bound using equa
tions (5.2) and (5.7) (solid line), and the Chernoff bound using simulated
weight vector statistics (*).
that the result can be used to characterize the conditional variance of our estimate
for 61. Using equations (5.2) and (5.7), the timedependent loglikelihood function at
the mth iteration (ignoring constants) is
9(w161, m) = 1 [(w(m) *(m))TC (m)(w(m) w(m))]. (5.20)
The CRB for the above loglikelihood function is then given by the following equation
[52].
V{r "02 (wl( lm) i1 (.1
Var( & _1) > t E g 3 ^ (5.21)
In order to evaluate the above expression, we require knowledge of how Cww(m)
depends on 6i, in order to evaluate the required partial derivatives. This does not
appear to be mathematically tractable since, as seen in equation (5.7), we need to
evaluate E[Ow(m)/c61wT(m)]. In order to continue with this approach, we assume
that OC,(,(m)/O6i = 0. Using this assumption, after straightforward algebraic ma
nipulations, our approximation to the variance of the 61 estimate is given below.
E[( 61)2I, = P] > (1) (1 Cw(m) 0 1 (5.22)
Several auxiliary equations and initial conditions are required in order to evaluate
equation (5.22). The required auxiliary equations are
w(m + 1) rrr o (m) 9
w6) (m) + (I ARrr) 1 + A (5.23)
Oar cp ) + c+1) (5.24)
0+c1
9Rrr ar _T a QT
__r + r (5.25)
961 9651 r 9S 1)
We hasten to stress that we have used the mechanics of the CRB technique to derive
equation (5.22). However, due to the various approximations used along the way our
result is not a lower bound, and is only intended to provide a rough approximation
to the performance of our estimator.
Several different DSCDMA environments were simulated in order to verify the
usefulness of equation (5.22) as an approximation to the conditional variance of the
61 estimate. For the following results, the users' spreading codes were selected from
a set of Gold codes [5] with 31 chips/bit and the desired user was received at a SNR
of 7 dB. The propagation delay for each user was chosen from a uniform distribution
over the interval [0, Tb). Once the spreading codes and delays were selected, they
were considered fixed and the DSCDMA environment was simulated for 2,000 trials.
Therefore, the variables that were considered random between the individual simu
lation trials were the AWGN and the data sequences for the interfering users. Note
that the LMS stepsize, p, was always chosen to be 0.1/(the total input power) such
that the filter was convergent in the meansquare.
10o0
a: 1 User
ib: 5 Users
\ c: 2 Users
S i\,, d: 10 Users
0
1
C
r a
b
1021
0 20 40 60 80 100 120
# of Training Bits
Figure 5.5: Comparing the simulated conditional variance of the 61 estimate to the
analytically derived approximation in absence of nearfar MAI.
Figure 5.5 shows the comparison between equation (5.22) (dashed line) and the
standard deviation of the 61 about 61, conditioned on P, = Pi (solid line) for several
different scenarios. In this figure, all interfering users are received at the same power
level as the desired user. The results shown in this figure look very promising. The
largest error between the simulation results and equation (5.22), which occurs for
10 users, is 6%. Unfortunately, the usefulness of equation (5.22) quickly disappears
as more users are added to the system, or the level of Pk/P1 is increased (or some
combination of these two conditions). As an example of this statement, Figure 5.6
shows how the approximation fails for several multiuser scenarios where all interfering
users are received at some level above the desired user. Upon review of the derivation
of the approximation, the weakness of the approximation appears to stem from the
assumptions that were made to facilitate the derivation, and not from equations (5.2)
and (5.7) as these have been observed to agree very well with simulation results.
10
a: 3 Users,N/F= 15dB
b: 5 Users, N/F = 20 dB
c: 7 Users, N/F = 15dB
U
10"
Cl) C
a b . 
102 I I I I
0 100 200 300 400 500
# of Training Bits
Figure 5.6: Comparing the simulated conditional variance of the 61 estimate to the
analytically derived approximation in presence of nearfar MAI.
5.4 Coarse Acquisition Performance
In this section, we present numerical results for the acquisition performance of the
timing estimator. We compare the performance of our timing estimator with a well
known conventional technique. One approach to timing estimation in a singleuser
spread spectrum system is the correlator. The received signal is correlated with time
delayed versions of the known spreading code, and the timing estimate is given by the
amount of time delay that maximizes the correlation. The receiver forms its estimate
as shown below:
la(I 6) 11I2 1 M
(,,)=argmax I' ra(p )r12M (5.26)
(p,b) Iai(p,6)112 J M =
where M is the number of r(m) samples observed before computing the timing es
timate. As shown by the above equation, this technique also requires an all ones
training sequence. It is also known to be optimal for a singleuser in the presence of
AWGN only, but can be suboptimal in the presence of multiaccess interference.
Our timing estimator will be computed by processing the weights of either an
LMS filter or RLS filter. For the LMS adaptation results, the LMS stepsize P was
chosen to be 0.1/trace(R), where trace(R) is the total filter input power. With
this stepsize, the LMS filter was convergent in the mean and also convergent in the
meansquare. As the metric for measuring the performance of the timing estimator,
we observed the average acquisition time of the timing estimator. The acquisition
time is defined as the number of training bits required so that the probability that the
timing estimate is within one halfchip of the true propagation delay of the desired
user is greater the 90 percent. That is, we recorded the smallest value of m such that
Pr (1i Ti < T,/2) > 90%. (5.27)
In addition, based on the discussion about the advantages of synchronization, we
were interested in observing the root meansquared estimation error (RMSEE) given
correct acquisition.
RMSEE = VE [(f, r\)21 (1f rlTj
The value of the RMSEE gives a measure of how well the receiver could be synchro
nized if the timing estimate was used to update the receiver's timing relative to the
desired user.
Each DSCDMA environment was simulated for 500 independent trials. For each
trial, the user spreading codes were selected at random from a set of Gold codes using
31 chips/bit. For each trial, the propagation delay for each user was chosen from a
uniform distribution over the interval [0, 31), independent from the other users. The
users' phaseshifts were chosen from a uniform distribution over the interval [0, 27r),
independent from the other users. The desired user was received at a SNR of 7 dB.
Also at the start of each trial, the received power level Pk for each interfering user
was selected from a lognormal distribution that had a mean and standard deviation
of 10 dB. That is, Pk/Pi = 100/1 where is a Gaussian random variable with a mean
of 10 and a variance of 100. The lognormal power distribution was used to simulate
environments where different power levels exist due to shadowing or system power
control error. By choosing the simulation parameters at the start of each trial, and
running many independent trials for each environment, the output of this test is an
estimate of the number of training bits required to achieve coarse acquisition.
Figure 5.7 shows the coarse acquisition performance using the correlator, LMS
filter, and the RLS filter in forming a timing estimate as a function of the number
of users. As shown in this figure, for a few number of users the LMS based timing
estimator offers no advantages over the conventional correlator estimator. However
as the number of users increases, the LMS based timing estimator does perform much
better than the correlator. In addition, the maximum RMSEE for all of the points
on the LMS curve was 0.11To, which shows that the timing estimator does produce
a reasonable estimate of the desired users propagation delay. But with 15 system
users, on average 240 training bits are required to achieve coarse acquisition. It is well
known that the convergence rate of an LMS filter, which will affect the performance
of the timing estimator, is dependent on the ratio of the largest eigenvalue of R
to the smallest eigenvalue of R. As the level of nearfar interference increases this
eigenvalue ratio quickly increases. This plot shows a weakness in using the LMS
adaptation algorithm, not a weakness in the timing estimator.
The results shown in Figure 5.7 for the RLS filter support the previous statement.
For the same DSCDMA environment, the RLS filter based timing estimate performs
much better than the correlator or the LMS filter. The RLS algorithm requires an
initial positive definite estimate of the data autocorrelation matrix. For this work, the
softconstrained initialization RLS algorithm was used, in which the initial estimate
of the data autocorrelation matrix is just a positive constant times an identity matrix.
The maximum observed RMSEE was 0.17Tc, which is only slightly worse than the
10 3
+
+
So102 +
10
1 0  """"""""""""'
?. 10 Correlator
I + MMSELMS
MMSE RLS
1 0 0 I
2 4 6 8 10 12 14 16
# Users
Figure 5.7: Average training bits required for correct acquisition as a function of the
number of system users in AWGN channel.
results for the LMS based timing estimator. With 15 users (50% system capacity)
the timing estimator can achieve coarse acquisition with 45 training bits, which is
not an unreasonably long training period. For 10 system users, the RLSbased tim
ing estimator achieves acquisition about 6 times faster than the LMS based timing
estimator. Clearly, the timing estimator based on the RLS filter performs better, in
terms of acquisition time, then the timing estimator based on the LMS filter.
While it is interesting to observe the performance of the timing estimator as a
function of the number of system users, it is equally important to consider the near
far resistance of the timing estimator. In other words, for a fixed number of system
users, we would like to see how the number of training bits required to achieve correct
acquisition depends on the level of the multiaccess interference. For this test, the user
spreading sequences, phases and propagation delays were chosen using the methods
previously described. The desired user was received at a SNR of 7 dB. However,
the received power level Pk for each interfering user was selected from a lognormal
S 3 MM It iL +
a 10 ~''S
2. +
15 Users + +
101 2 2 Users
10 '
0 2 4 6 8 10 12 14
XindB
Figure 5.8: Average training bits for correct acquisition as a function of the level of
multiaccess interference in AWGN channel.
distribution that had a mean and standard deviation of X, where X was allowed
to vary over the range of (0,13) dB. For a given value of K, the number of system
users, and a specific value of X, 500 independent trials were used to find the coarse
acquisition performance of the timing estimator. This scenario was used to simulate
an environment in which there is very loose power control, and the desired user is
received at a power level which on average is one standard deviation below the mean
of the power levels of all the interfering users.
Figure 5.8 shows the coarse acquisition performance of the 3 timing estimation
algorithms for 2 system users and 15 system users as a function of the the value of
X. From this figure it is seen that none of the timing estimators are truly nearfar
resistant. Clearly, the correlator is very sensitive to the amount of nearfar interfer
ence. As was observed in the previous plot, the LMSbased timing estimator performs
better than the correlator, and the RLSbased timing estimator performs better than
the LMSbased timing estimator. For 2 users, the RLS based timing estimation algo
rithm could be considered as nearfar resistant. When the value of X was set to 0 dB,
only 5 training bits were required to achieve correct acquisition. When the value of
X was set to 13 dB, a much more severe nearfar environment, only 7 training bits
were required. With 15 system users, the corresponding number of required training
bits changed from 10 to 75, which we consider as being not enough of an increase in
the training period to label the estimation algorithm as not being nearfar resistant.
However, it is clear from this plot that the length of the training period required to
achieve correct acquisition in a system with a severe nearfar problem when either
the correlator or the LMS based timing estimator is used is unacceptable.
5.5 Summary
In this chapter, we have attempted to characterize the performance of the timing
estimator. An equation for the transient weight mean vector of the LMS filter weights
was derived and tested under several DSCDMA environments. Simulation results
have shown that this equation is in close agreement with the true transient mean of
the LMS filter weights. A second equation was derived and tested for the transient
weight autocovariance matrix of the LMS filter weights. This equation works well
for a singleuser environment, but does not follow the observed response of the filter
weights for a multiuser situation. Therefore, the weight error covariance matrix was
used to derive a second equation for the transient weight autocovariance matrix. This
equation has been observed to agree within about a 1020% error of what is observed
through direct simulation. We then used these two iterative equations to derive two
analytical techniques to characterize the performance of the timing estimator.
The goal of the first technique was to provide a means of observing how quickly
the timing estimator can correctly estimate the propagation delay of the desired user,
in a chipsynchronous environment. The idea was to use the analytical expression as
a means to observe the timing estimator, without running long MonteCarlo simula
tions, or at least to get a sense for how to set up such simulations. Unfortunately,
evaluating the required probability was a difficult task at best, so we used several
methods to form an upper bound on the desired probability. We observed that the
upper bounds were very loose, and so this technique while interesting is of very limited
use.
The second technique was more along the lines of the traditional CramerRao
bound. We desired an analytical expression that could be used as an approximation
to the variance of the 61 estimate (about the true value of 61) conditioned on getting
the integer part of the timing estimate correct. We observed that the resulting ap
proximation was useful for few users, with perfect power control, but was not useful
for more general cases.
Lastly, we characterized the performance of the timing estimator using Monte
Carlo simulation methods. We compared the performance of the adaptive filterbased
timing estimator to a conventional correlatorbased timing estimator. The correlator
based timing estimator is well known to be optimal for a singleuser operating in an
AWGN channel, but can be suboptimal for a multiuser DSCDMA environment.
The metric used to compare the timing estimator, was the average number of training
bits required such that fi was within onehalf a chip of the true value of T1 > 90%
of the time. For few users, with low levels of MAI, we observed that the adaptive
filterbased timing estimator offered a slight improvement over the correlatorbased
timing estimator. However, as the number of system users increases, both the LMS
based and RLSbased implementations of the timing estimate perform better than
the correlator, with the advantage going to the RLSbased algorithm. The nearfar
resistance of the algorithm was tested, by simulating an environment that could occur
in an open loop system, or a system with very loose power control. Neither the LMS
based or the RLSbased estimation algorithms are truly nearfar resistant. However,
the change in the length of the required training interval for the RLSbased algorithm
was so minor that we consider it to be nearfar resistant.
While it is difficult to directly compare our timing estimator with other techniques,
as some of the other techniques estimate more than a singleuser's parameters, one
observation can be made. The complexity of the estimation algorithm is dependent
on N, the number of chips in one period of a spreading sequence. The complex
ity of the estimation algorithm is 0(N2) above the complexity of the traditional
correlatorbased system when the RLS algorithm is used. However, note that for
some environments reliable estimates of rT1 are available with training sequences that
are shorter than N symbols long. In a previous chapter, other estimation algorithms
were presented. Most, if not all, had higher complexities and some required training
sequences that were at least N symbols long before a timing estimate could be made.
Therefore, our timing estimator has the advantage over these other estimators when
short training sequences are required, under the condition that our timing estimator
provides reliable estimates.
CHAPTER 6
SIMULATING FADING CHANNELS
Previously, the communication channel was modeled as an AWGN channel. A
more realistic channel model must be used to account for the dispersive effects ob
served in urban environments. In urban environments, it is possible for multiple
versions of a transmitted signal to arrive at a receiver due to reflections off of ad
jacent buildings, cars or other obstacles. If a total of L versions of the transmitted
signal are received, then the channel is said to have L paths. The signal received from
each path can experience its own time delay, phase shift and attenuation. Therefore,
at the receiver these L signals may add constructively or destructively, a condition
which is known as fading. If relative motion exists between the transmitter and the
receiver (or the reflecting obstacle is in motion), then the received signal is also ob
served to have a shift in its carrier frequency. Also, due to relative motion between
the receiver and the transmitter, it is easy to conceive that the characteristics of the
channel change as a function of time and, therefore, a more realistic channel model
will account for the time varying effects observed on the received signal.
Rappaport [54] provides a summary of the various types of fading, as well as pre
senting several techniques that can be used to simulate fading channels. Proakis [55]
also presents a characterization of fading channels and discusses the effect of fad
ing on the performance of communication systems. The type of fading induced by
a communication channel is dependent on the time/frequency characteristics of the
channel with respect to to the time/frequency characteristics of the transmitted sig
nal. Sklar [56] presents several block diagrams that summarize how these relationships
between the channel and the transmitted signal determine the type of fading induced
by the channel. In addition, Sklar presents a summary of traditional techniques that
have been used to mitigate the effects of fading [57]. In the remainder of this chapter,
we will briefly review the various types of fading and discuss how a fading channel
can be simulated.
Consider a multipath communication channel where there are L distinct paths
with unique propagation delays T1 > T2 > .. TL1 > TL. The quantity TL T1 is
defined as the excess delay spread. As expected, this multipath delay spread causes
time dispersion in the received signal. In fact, if the excess delay spread is significant
when compared to the symbol interval, the channel can be viewed as inducing severe
intersymbol interference (ISI). That is, one transmitted symbol will contribute to
several symbol intervals at the receiver. Since the number of paths, and hence their
corresponding propagation delays, are assumed to change as a function of time, the
excess delay spread is a random variable. The statistical characteristics of the excess
delay spread relative to the symbol interval along with the frequency characteristics
of the channel and transmitted signal determine if the fading is flat or frequency
selective.
If a communication channel has a constant gain and linear phase response over
a bandwidth that is greater than the bandwidth of the transmitted signal, then all
frequency components of the transmitted signal's spectrum will be equally attenuated
by the channel. In this condition, the channel is said to be flat. However, due to the
time varying nature of the multipath channel, the channel's impulse response and
hence its frequency response are time variant. In order to characterize the frequency
response of the channel, the coherence bandwidth is defined. The coherence band
width of a channel denotes a bandwidth in which different frequency components
experience the same amplitude attenuation. The coherence bandwidth of a multi
path channel is related to the excess delay spread in some manner. While no exact
relationship between these two quantities exists, there are several well known rules
of thumb [54]. Let the transmitted signal have a bandwidth, Bs, a symbol duration
Ts, while the channel has a coherence bandwidth Bc and rootmeansquared excess
delay spread of at. The following statements regarding classification of the fading
channel with respect to the transmitted signal will be described by Ts, in order to
describe a general communication system. With Ts used to denote the duration of the
transmitted symbol, we must remember that for a DSCDMA system Ts is equal to
To, the duration of onechip interval, when applying the following statements. If the
bandwidth of the transmitted signal is much less than the coherence bandwidth of the
channel, Bs < Be, and the symbol duration is much greater than the RMS excess
delay spread, Ts at, then the channel induces flat fading. If the bandwidth of the
transmitted signal is greater than the coherence bandwidth of the channel, Bs > Be,
and the symbol duration is less than the RMS excess delay spread, Ts < at, then the
multipath channel induces frequencyselective fading. Note that flat fading channel
models are more prevalent since a frequencyselective fading channel can be viewed as
the composite sum of a multipath channel where each path induces flat fading, but
the propagation delays between the individual paths are set such that the composite
channel induces frequencyselective fading.
Similar to the time dispersion induced by the multipath delays, relative motion
between a transmitter and its receiver induces dispersion in the frequency domain.
Consider a singlepath channel over which a constant amplitude and frequency sinu
soidal signal is transmitted. If there is relative motion between the transmitter and
receiver then there is a frequency shift observed in the received signal. The maximum
observable frequency shift, fD, is known as the Doppler frequency or Doppler spread
and is given by
f V = (6.1)
fD = _f (6.1)
where v is the relative speed between the two devices in m s1, C is the speed of light
in free space and fc is the carrier frequency in Hz. When a timevarying multipath
channel is considered, the observed frequency shift becomes random and timevarying.
The relative speed between a transmitter and receiver determine the rate at which
the impulse response of the channel changes. In order to characterize this rate of
change the coherence time of the channel, Tc, is defined. The coherence time of the
channel is used to characterize the correlation of samples of the channel output. That
is, if two samples of the channel are separated in time by less than the coherence time
of the channel, they will be highly correlated (similar). As before, there are several
known rules of thumb that are used to relate the coherence time of a channel to
the Doppler spread. If the symbol duration is greater than the coherence time of
the channel, Ts > Tc, and the bandwidth of the transmitted signal is less than the
Doppler spread, Bs < fD, then the channel impulse response changes many times
during a symbol interval and the channel induces fast fading. If the symbol interval
is much less than the coherence time of the channel, Ts < Tc, and the bandwidth
of the transmitted signal is much greater than the Doppler spread, Bs > fD, then
the channel is essentially timeinvariant when viewed over several symbol intervals.
In this case, the channel is said to induce slow fading.
The Rayleigh distribution is commonly used to characterize the timevarying
statistics of the envelope of the received signal in a flatfading environment. It is
well known that the random variable z = Vx2 + y2 when x and y are independent
zeromean Gaussian random variables with equal variances a2 follows a Rayleigh
distribution. The probability density function for a Rayleigh distribution is given by
r ex 2 0 < <
fR(r) = 0e h a (otrwi
0 otherwise.
If however there is a dominant path in the communication channel, such as the case
when a lineofsight (LOS) path exists, the envelope of the received signal is modeled
by a Rician distribution and has a probability density function of
r exp r 2+A2) (AO 0,
f(r) ) 2a2 (6.3)
0 otherwise.
where Io() is the zeroorder modified Bessel function of the first kind. Since the
Rayleigh distribution is just a special case of the Rician distribution, it is easy to
create a fading channel simulator that generates both distributions for the envelope
of the received signal.
The fading channel simulator used in this work is based on Clarke's [58] channel
model. In that work, the fading was due to the scattering of electromagnetic waves in
such a way that all paths had the same propagation delay. This means that all paths
form one composite path, and since there is no delay spread the channel induces flat
fading. The received envelope was shown to be a random process whose amplitude
follows a Rayleigh distribution and has an autocorrelation function given by
R(r) = Jo(27rfDT) (6.4)
where Jo() is the zeroorder Bessel function of the first kind. Therefore to simulate
a channel based on this model, we must create a random process whose amplitude is
Rayleigh distributed, in such a way that the autocorrelation of the random process
is similar to equation (6.4).
A thirdorder filter has an transfer function of
3
H(s) = (65Woo
(s2 + 2CwoS + w2) (s + Wo) (6.5)
and a corresponding impulse response of
h(t) = (exp(at) [A sin(3t) B cos(/3t)] + C exp(wot)) u(t) (6.6)
where
A Wo B= C WO
2v/1 (2 2(1
a =wo = o w/1 C2.
By setting wo and C based on the value of the Doppler frequency as shown below,
Wo= 1rfD (6.7)
1.2
C = 0.175 (6.8)
the convolution of h(r) with h(rT) closely follows equation (6.4) over the first major
lobe. This means that if a complex white Gaussian noise process is used as the input
to the thirdorder filter, the autocorrelation of the output process closely follows
equation (6.4) for values of r where samples of the process are highly correlated.
And hence, we use this thirdorder filter to simulate a flatfading channel based on
Clarke's model.
Figure 6.1 shows a block diagram of a fading channel simulator based on the
thirdorder filter that can be used on a digital computer. The continuoustime filter
is converted to a digital filter where the sampling rate is specified, typically equal
to one bit interval. The input to the filter are samples of complex Gaussian noise,
whose real and imaginary parts are independent, have zeromean and equal vari
ances of 1/2. The block diagram accounts for the presence of an LOS, or specu
lar component, such that the envelope of the simulator output can follow both the
Rayleigh and Rician distributions. Therefore the output of the channel simulator is
Samples of 3rd Order Sampled
Complex er (p)+ Flat Fading
Gaussian Noise Process
A(p)exp(j14) LOSor
LOS or
Specular Component
Figure 6.1: Flatfading channel simulator.
a sampled complex Gaussian process having a mean of A exp(jq) and a variance of
a2 = E [Ak exp(jOk) Aexp(j)12].
In Figure 6.1, two parameters are shown to be a function of a variable called p.
The ratio p = A/a is referred to as the Rician parameter and the pair (A, a) are
normalized so that the fading process has unit power.
1
a(p) p> 0 (6.9)
1+ p2
A(p) P p p>0 (6.10)
V/l+p2
A plot comparing the theoretical probability density functions of the envelope of
the fading process given in equations (6.2) and (6.3), to the those observed through
simulations for two values of p is displayed in Figure 6.2. As shown in this figure,
when p = 0 the envelope of the fading process is Rayleigh distributed, while for
other values the envelope of the fading process follows a Rician distribution. As a
final demonstration of the fading channel simulator, Figure 6.3 displays a typical
realization of a flat Rayleigh fading process. The magnitude of the envelope and
the phase of the fading process are shown when the carrier frequency was 1.8 MHz,
the data rate was 9600 bitspersecond (BPS), and the relative speed between the
transmitter and the receiver was 3 milesperhour (MPH). The third order digital
filter, as well as the output of the fading channel simulator where sampled at the
data rate.
0 11 1,,_ I I I
0 0.5 1 1.5 2 2.5 3 3.5
r
Figure 6.2: Theoretical and simulated probability density function for the envelope
of the fading channel simulator output for two different Rician parameter
values.
2000 4000 6000 8000
2000 4000 6000 8000
Sample Number
10000
10000
Magnitude and phase of a slow flatfading Rayleigh process generated
with a carrier frequency of 1.8 GHz, a vehicle speed of 3 MPH, and a
data rate of 9600 BPS.
20 L
0
200
200'
0
Figure 6.3:
Consider simulating a singleuser DSCDMA system operating in a fading channel
environment. For a given Doppler frequency, the fading channel simulator is created
and assume that both the thirdorder filter as well as the simulator output are sam
pled at the symbol rate. In a nonfading channel the contribution to the receiver's
input during the mth symbol interval due to the user is given by J1(m) as shown
in equation (2.7). Multiplication of Ji(m) by the mth sampled output of the fading
channel simulator, A(m) exp(j0(m)), at each symbol interval will produce an input
to the receiver whose envelope experiences fading. In order to simulate a multiuser
fading channel DSCDMA system, each user should have its own fading process that
is independent from all other users. We could implement K independent fading chan
nel simulators, and then multiply Jk(m) by the output of the corresponding fading
channel simulator. However, a simpler approach was used in this work. A single
fading channel simulator was created for a specified Doppler frequency. Then a very
large sequence of samples of the fading process was created. Consider two samples
of this sequence, A(n) exp(jO(n)) and A(n + m) exp(jO(n + m)). We know that if m
is chosen appropriately then the correlation between the samples becomes negligible,
and therefore we can treat the samples as being independent. Using this idea, the
original sequence was partitioned into K sequences such that each sequence is con
sidered to be independent of each other. That is, the sequence for the first user was
created using samples (1, 2,..., m), while the sequence of the second user was created
by using samples (m + 1, m + 2,..., 2m), and so on for each user.
CHAPTER 7
CODE ACQUISITION IN NONSTATIONARY ENVIRONMENTS
It is wellknown that training an adaptive filter in a nonstationary environment
can be difficult [31]. Since our timing estimator is based on processing the weights
of the adaptive filter, a study of how the estimator is affected by such environments
is required. In this chapter, we will consider timing acquisition in two such environ
ments. The first environment occurs due to nonideal downconversion of the received
signal due to frequency synchronization errors. When this condition occurs, the input
to the adaptive filter has a timevarying phase induced on it. The second condition
to be studied is when the received signal has experienced fading, in which case both
the amplitude and phase of the received signal are timevarying.
Barbosa and Miller [13] studied the adaptive detection of DSCDMA signals in
fading channels. In particular, they studied the performance of the adaptive receiver
shown in Figure 4.1 operating in frequency nonselective fading channels. They found
that this receiver structure does not work in a Rayleighfading channel. When the
desired user's signal experiences a deep fade, the receiver frequently loses lock on
the desired signal. When the desired user's signal emerges from the deep fade, the
receiver may emerge into any of 3 possible states:
1. correctly locked on phase to the desired user's signal;
2. locked 180 out of phase to the desired user's signal; or
3. locked either in phase or 180 out of phase to any of the interfering user signals.
In a singleuser environment, only the first two of these conditions are possible. The
first condition is desired. The second condition is not due to the filter's inability to
track the rapid phase changes during a deep fade. It is due to the fact that since the
decisions are unreliable in a deep fade (many errors occur in the feedback loop) the
filter is essentially running "blind" during the deep fade. Once the filter locks onto the
desired user's signal at 180 out of phase, errors will continue to occur. This problem
can be handled by differential encoding and decoding of the data sequence. The third
condition is catastrophic, and is what renders this receiver structure useless in this
type of environment. Barbosa and Miller proposed a modified receiver structure to
alleviate these problems. The next few paragraphs describe this receiver structure
and its derivation.
The main problem with the adaptive receiver structure shown in Figure 4.1 seems
to be loss of phase lock during deep signal fades. Therefore, it stands to reason
that if reliable estimates of the fading process can be made, these phase variations
could be removed from the input to the adaptive filter, Then, maybe the adaptive
receiver could perform adequately, even in a Rayleighfading channel. In a nearfar
environment, the signaltointerferenceplusnoise ratio for the desired user's signal
would probably be too small to allow for phase estimation without a complicated
estimation procedure. However, the filter suppresses the multiaccess interference
such that the signaltointerferenceplusnoise ratio at the filter output is higher than
the same quantity at the filter input. Therefore, the filter weights are used in the
channel estimation procedure.
By taking the real part of the received signal prior to entering the adaptive filter,
the weights of the filter are real and, therefore, the filter makes no attempt to track
the phase of the received signal. Noisy estimates of the amplitude and phase of the
fading process during the mth bit interval for the desired user may be formed as
7y(m) = di(m)wT(m)r(m). The weights of the filter are updated at the bit rate,
which means that y(m) can be updated at the bit rate. Therefore, we must predict
the channel conditions for the current bit interval based on previous values of "y(m).
An Lthorder linear predictor is used to form the current estimate of the channel
conditions, based on past estimates as shown in equation (7.1),
L
=(m) = (m i) (7.1)
ii
and the estimate of the phase during the mth bit interval is found by 0i,m = Z/(m).
The coefficients of the Lthorder linear predictor are chosen to minimize the mean
squared error E[laIm exp(j'i,m) (m)12], where a,1,m and 01,m are the amplitude and
phase induced on the desired user by the channel during the mth bit interval. The
solution of these coefficients depends on the weights of the MMSE filter, which depend
on the multiaccess interference. In order to remove this dependence, the coefficients
were calculated for a singleuser system, and then are used for all other environments.
Not surprisingly, the coefficients of the Lthorder linear predictor, a, are found by
a = C1v (7.2)
where
{v}i = 1 (i,7rfDTb)2 i {1,2,..., L} (7.3)
{B},,j = 1 ((i j)rfDTb)2 ij E {1,2,..., L} (7.4)
C = B + (Eb/A/o)I1 (7.5)
and I is an L x L Identity matrix and fD is the maximum Doppler frequency induced
by the channel. It is important to remember that these equations were derived for
a singleuser and, therefore, in the presence of multiaccess interference the channel
estimator may be very far from optimal.
The modified form of the adaptive receiver, using the Lthorder linear predictor
designed by Barbosa and Miller (equations (7.2) through (7.5)) is shown in Figure 7.1.
Now that we have presented the background of this receiver, we summarize several
C1
w(m) Estimation Ti
wm) algorithm
Convert to
r(t) baseband Adaptive nnn ci
i r+ chip x Rebos filter, w(m) tepo
matched (eco
filter j ?+ 6
__Z Adaptive e(m)
algorithm
Channel phase 44 di (M)
estimator 4 w(m)
Figure 7.1: Modified adaptive receiver with channel compensation.
important results presented by Barbosa and Miller [13]. They found that the perfor
mance of the receiver (detection) is insensitive to L, the order of the linear predictor.
Also, unlike the unmodified form, the performance of the detector is independent of
the rate of the fading process as the detector performed equally well in slowfading as
well as fastfading. However, the most important result is the effect of the modified
receiver structure on the system capacity. For a given level of probability of bit error
rate, it was found that the structure in Figure 7.1 could support more users than the
conventional receiver (which is a correlator). In fact, it was shown that the system
capacity when using the modified receiver structure can be made to approach 100%
(N users), while the conventional receiver typically restricts the system capacity to
on the order of 1020%.
Several comments about the modified receiver structure as it relates to timing
acquisition are required at this point. Barbosa and Miller were interested in detection,
and the weights of the adaptive filter were initialized close to their steadystate values.
The result is that the Lthorder linear predictor can make reliable estimates of the
channel conditions at the start of their simulations. In this work we are interested in
timing acquisition, and therefore the weights of the adaptive filter are initialized to
a vector of all zeros. This means that most likely the initial estimates of the channel
phase will be incorrect. As seen in Figure 7.1, the phase compensation operation,
Re [r(m) exp(j1,m)j, causes a delay in the build of energy in the filter since when
k1,m 01(m) is significant, the phase compensation operation rejects some of the
desired user's input signal from the input to the adaptive filter. The time to achieve
correct acquisition will be dependent in some manner on the transient response of
the channel estimation algorithm.
7.1 Frequency Synchronization Errors
Consider the situation that occurs when the frequency of the kth user's transmit
ting oscillator is not equal to the frequency of the receiver's oscillator. In this case
after downconversion, the input to the adaptive filter for the kth user will have a
linearly timevarying phase. In order to model this effect, we assume that the fre
quency offset for the kth user, Awk, can be expressed as Awk = 27r/(NTcMk) where
Mk is a positive integer and NTc is the bit interval.
Using this notation, we can model the input to the adaptive filter using equa
tion (2.6) once we redefine Jk(m) as shown below.
Jk(m) .NMk exp 27rm\ ]
Mkr (7.6a)
[Z2k1 (m)v2k1 (Pk,tk) + Z2k(m)v2k(pk, 6k)
where the nth components of v2k1(pk, 6k) and V2k(pk, 60k) are given by:
/ c \ j27rn N
V2k l,n(Pk, 6k) = exp N X
\NMk /
{ exp 0Z2r d 1c(Pk+1)
{ NMk ) k,n (7.6b)
+ [exp ( 27_ exp (J2"']c(Pk)
NMk exP {NMk ] c k,,
t j21rn
V2k,n(Pk, 60k) = exp N X
ex 1] NM&) k (7.6c)
NMJk 1 k )
[ (j27r (J275k] e(Pk
+[exp (KNMk) exp ~NMkJ k,n
As a sanity check of these equations, note that in the limit Mk + 00 the above
equations reduce to the previous definition of Jk(m).
The results in Figure 7.2, show the acquisition performance of the estimation
algorithm in the presence of frequency synchronization errors. The desired user was
received at a SNR of 10 dB, and 4 interfering users were each received at a power level
10 dB above the desired user. The desired user was assumed to have a frequency offset
of 1% of the data rate (Mi = 100), while all other users had a frequency offset of 2%
of the data rate (M25 = 50). The timing offsets and initial phaseoffsets for each user
were chosen at random, for each of the 250 trials. For each trial, 3 independent RLS
filters were run in parallel. The first filter was uncompensated, while the second filter
used the modified receiver structure with a 10thorder linear predictor to estimate
the channel phase. As a control, the third filter was compensated with the known
phase induced on the desired user at each bit interval. By doing this, the effects of
the transient response of the channel estimation algorithm on the timing estimator
can be avoided.
As shown in Figure 7.2, the uncompensated filter was initially able to form a good
estimate of propagation delay for the desired user. However, the filter was unable
to track the phase of the input signal and therefore the timing estimation algorithm
eventually breaks down. When the modified receiver structure is used, and the input
signal is compensated using perfect knowledge of the phase induced on the desired
user, the timing estimation algorithm works very well. Once again, this is due to the
fact that the transient response of the channel estimator and the phase compensation
operation are not causing a delay in the buildup of the desired user's signal in the
1
S0.8
VI
0.6
0.4
0.2
0
0
Figure 7.2: Acquisition performance of the adaptive receiver
ment in presence of frequency offset errors.
in a fiveuser environ
weights of the adaptive filter. The 10thorder linear prediction compensated filter,
unlike the uncompensated filter, is able to achieve correct acquisition at the cost of
an increase in the length of the training interval.
Considering the results of Figure 7.2, it seems reasonable that a hybrid of the
uncompensated and compensated filters may be useful in forming a timing esti
mate. That is, it may be possible to start acquisition using an uncompensated filter
(with complex weights) and at some point switch to the compensated filter (with real
weights) such that we get better acquisition performance. We will address this idea
in the next section where we study the performance of the acquisition algorithm in a
flatfading channel.
50 100 150
Time (Symbols)
SCompensated, using perfect knowledge
of phase induced on desired user.
/J\ Compensated,
S10th order linear
Prediction
4 Uncompensated
200
250
300
7.2 Performance in FlatFading Channels
We model the input to the adaptive filter using equation (2.6), by defining the
contribution of the kth user during the mth bit interval as
Jk(m) = ak(m) exp(jOk(m)) [z2k1(m)a2k1(pk, 6k) + Z2k(m)a2k(pk,S k)] (7.7)
where ek (m) and Ok (m) are the amplitude and phase induced on the kth user's signal
by the fading process during the mth bit interval. The fading process for each user is
assumed to be independent of the fading process induced on each of the other users.
The fading parameters, ak(m) and 0k(m), are taken from samples of the flatfading
channel simulator described in the previous chapter.
As a first test of the acquisition algorithm, a tenuser DSCDMA environment
was created with a SNR of 10 dB and all interfering users were received at the same
power level as the desired user (in the absence of fading). A slowly flatfading channel
was created for each user, using a carrier frequency of 1.8 GHz, a data rate of 9600
BPS, and a speed of 3 MPH. These parameters are commonly used to simulate a
personal communication system (PCS) environment in which the users are mobile at
walking speeds. The resulting Doppler frequency, fD, is 8 Hz which is a normalized
rate (normalized to the data rate) of 8.4 x 104.
To test the acquisition algorithm in this environment, a bank of four parallel RLS
filters was used. One filter, was uncompensated as shown in Figure 4.1, and processed
the received signal using complex weights. Two of the filters, used the architecture
shown in Figure 7.1 with a Lthorder linear predictor for the channel estimator. In
the following, we will refer to each of these filters as the modified MMSE receiver.
One of the modified MMSE receivers compensated the received signal using only
the phase estimate of 7(m) (as shown in Figure 7.1). The second modified MMSE
receiver compensated the received signal using both the amplitude and phase of 7(m).
The fourth filter, referred to as the hybrid receiver, was an adhoc mixture of the
two receiver structures. The idea of the hybrid receiver is to jumpstart the channel
estimation algorithm by preloading the adaptive filter's weight vector, w(m), and the
linear predictor vector, 7, hopefully bypassing some of the learning curve required to
form reasonable channel estimates otherwise. The hybrid receiver uses the structure
of Figure 4.1 for 20 bit intervals without making timing estimates. After the 20th bit
interval, the receiver forms an estimate of r1 and then makes an estimate of 01 using
equation (4.14a). This phase estimate is loaded into the Lelements of 7, and the
imaginary parts of the filter's weight vector and the inverse of the correlation matrix
are removed. The receiver then switches to the modified receiver structure shown in
Figure 7.1, and starts the timing and channel estimation algorithms using the (now
real) weights of the adaptive filter.
The results of this test when a thirdorder linear predictor was used for the channel
estimation algorithm are shown in Figure 7.3. As seen in this figure, the acquisition
algorithm based on processing the weights of the uncompensated filter performs bet
ter than any of the other techniques. In general, this situation will not occur as the
level of the MAI changes and/or the Doppler frequency changes, and results similar
to that shown in Figure 7.2 can be expected. Barbosa and Miller [13] observed the
detection performance of the modified MMSE receiver was insensitive to L, the order
of the linear predictor used in channel estimation. In addition, they compensated the
received signal by only using the phase of the channel estimate, '(m), as no advan
tage was observed by also using the amplitude of 7(m) to compensate the received
signal. Since they were interested in detection, they did not observe the startup
transient of the channel estimation algorithm as the weights of the adaptive filter
were initialized to their steadystate values. In Figure 7.3, we observe the effects of
the startup transient of the channel estimation algorithm on the performance of the
timing estimation algorithm for the modified MMSE receiver. Unlike the observations
of Barbosa and Miller, it is observed that including the channel amplitude estimate
0.8 X ^
.0.8 / T Uncompensated
S0.6 Hybrid
V1
t Modified MMSE
k 0.4 (phase)
0.2 Modified MMSE
(phase and amplitude)
0 100 200 300 400 500
Time (Symbols)
Figure 7.3: Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a thirdorder linear predictor for the channel estimation
algorithm.
in the compensation operation imposes a severe penalty in the performance of the
timing estimate when compared to the performance of the timing estimate when only
the channel phase estimate is used. To study the effects of the value of L on the
timing estimation algorithm, the same DSCDMA environment was observed when
a tenthorder linear predictor was used in the channel estimation algorithm. These
results are shown in Figure 7.4. Once again, unlike the observations of Barbosa and
Miller, the performance of the timing estimation algorithm is sensitive to the value of
L. When a tenthorder linear predictor is used in the channel estimation algorithm,
the timing estimation algorithm performs better than when a thirdorder linear pre
dictor is used as noted by the difference between the rates of convergence of the two
probability curves in Figures 7.3 and 7.4. For a variety of DSCDMA environments
(Doppler frequency, MAI, and number of users), it has been observed that no signif
icant improvement in the performance of the timing estimator is achieved by using
a higher order linear predictor in the channel estimation algorithm. Therefore, in
1 1
0.8
,0.6
vi J Modified MMSE
/ J (phase)
S0.4
kf ~Modified MMSE
0.2 (phase and amplitude)
I Hybrid
OI I II
0 100 200 300 400 500
Time (Symbols)
Figure 7.4: Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a tenthorder linear predictor for the channel estimation
algorithm.
the sequel when we refer to the modified MMSE receiver we are using a tenthorder
linear predictor and only the phase of the channel estimate is used to compensate
the received signal. As seen in Figures 7.3 and 7.4, eventhough the hybrid receiver
has been formulated in an adhoc fashion the resulting performance of the timing
estimator clearly shows that the idea has its merits.
In order to further test the hybrid receiver, a study of the effects of the level of
MAI on the performance of the timing estimation algorithm similar to that shown
in Figure 5.8 was performed. The carrier frequency was 900 MHz, the data rate was
9600 BPS, and the relative speed between the receiver and all of the system users was
assumed to be 65 MPH. This scenario is used to depict a DSCDMA system in which
the users are traveling at highway speeds. The Doppler frequency induced by the
flatfading channel was 87 Hz, which is a normalized Doppler rate of 9.1 x 103. Each
DSCDMA environment was simulated for 500 independent trials. For each trial,
the user spreading codes were selected at random from a set of Gold codes using
.200 ".
200 10 Users
^ ^^ /^.Hybrid
1
=,100 srMdfe MME
5 3 Users, Hybrid
0 I I I I
0 2 4 6 8 10 12 14
XindB
Figure 7.5: Comparing the average training bits for correct acquisition as a function
of the level of multiaccess interference in a flatfading channel.
31 chips/bit. For each trial, the propagation delay for each user was chosen from a
uniform distribution over the interval [0, 31), independent from the other users. The
desired user was received at a SNR of 10 dB. Also at the start of each trial, the received
power level Pk for each interfering user was selected from a lognormal distribution
that had a mean and standard deviation of X dB. That is, Pk/P1 = 100/1 where
is a Gaussian random variable with a mean and standard deviation of X dB. By
varying the number of system users and the level of the MAI, it is possible to study
the performance of the hybrid receiver in conditions where the uncompensated filter
shown in Figure 4.1 doesn't perform well, but the hybrid initialization technique may
improve the acquisition performance over that offered by just the modified MMSE
receiver alone.
The results of this test are shown in Figure 7.5 for a threeuser and tenuser DS
CDMA system where the level of the lognormally distributed interference (X) ranged
from 0 dB to 13 dB. For both the threeuser and tenuser systems, the hybrid receiver
performs better than the modified MMSE receiver in terms of the average training
0.8
^0.6
Vi 0 % Uncompensated
U 0.4
0.2
0 IL
0 200 400 600 800 1000
Time (Symbols)
Figure 7.6: Performance of the timing acquisition algorithm using the hybrid and
modified MMSE receivers.
bits required in order to achieve acquisition to within a halfchip of Ti. When the
hybrid receiver was used in the threeuser environment, the performance of the timing
acquisition algorithm appears to be independent of the level of the MAI. This result
is due to the length of the training cycle used in the hybrid receiver before timing
estimates are actually made (20 bits). For a threeuser system, the uncompensated
filter is able to make reliable timing estimates prior to 20 training bits and, therefore,
it may be possible to use a shorter training interval for the hybrid receiver initial
ization cycle without degrading the performance of the timing acquisition algorithm.
The length of the training cycle used to initialize the hybrid receiver is sensitive to
the number of system users and the level of the MAI. However, over the range of users
and lognormally distributed MAI depicted in Figure 7.5, 20 bits of training provide
the hybrid receiver with a definite advantage over the modified MMSE receiver in
terms of timing acquisition performance.
A situation in which the hybrid receiver offers no advantage over the modified
MMSE receiver is shown in Figure 7.6. This result is for the same DSCDMA envi
ronment tested in Figure 7.5, however there are fifteen systemusers and X = 1 dB for
all of the lognormally distributed interfering users. This plot readily demonstrates
how the performance of the timing acquisition algorithm using the hybrid receiver
is "modulated" by the performance of the uncompensated filter and the length of
its training cycle before switching to the modified MMSE receiver structure. While
decreasing the length of the training cycle below 20 bits might not affect the per
formance of the timing acquisition algorithm, certainly increasing the length of the
training cycle to 100 bits would be disastrous.
7.3 Summary
Barbosa and Miller [13] studied the performance of the MMSE receiver in a flat
fading channel, and discovered that the MMSE receiver doesn't work in a fiatfading
Rayleigh channel. They proposed a modified receiver structure which compensates
the received signal such that the adaptive filter doesn't attempt to track the phase
of the desired user's signal. In this chapter, the modified MMSE receiver structure
was used to study the performance of the timing acquisition algorithm in two en
vironments in which the input to the adaptive filter was nonstationary. While the
timing acquisition algorithm is applicable to use with the modified MMSE receiver, it
has been observed that the startup transient of the channel estimation algorithm de
grades the performance of the timing estimation algorithm by increasing the length
of the training interval required to achieve correct acquisition. A hybrid receiver
structure has been presented to circumvent some of the startup transient response of
the channel estimation algorithm. The hybrid receiver uses a short training interval
to form an estimate of the phase induced on the desired user, 01, and then uses the
modified MMSE receiver with the channel estimator preloaded with 91 to continue
the timing acquisition cycle. Through simulation results of a DSCDMA system, it
85
has been observed that the timing acquisition algorithm based on the hybrid receiver
structure can offer an improvement over that of the modified MMSE receiver. How
ever, the initialization cycle for the hybrid receiver is based on an uncompensated
MMSE receiver and therefore with an inappropriate choice of the length of the initial
training cycle, it is possible for the performance of the timing acquisition algorithm
for the hybrid receiver to be worse than that offered by the modified MMSE receiver
alone.
CHAPTER 8
INCREASED WINDOW OF OBSERVATION AND TIMING ACQUISITION
In previous chapters, the adaptive receiver was used to form a timing estimate of
the propagation delay for the desired user. Since the receiver only observed one bit
interval of information at a time, the desired user is forced to use an all ones training
pattern during the code acquisition cycle. However, it is possible to use a known
sequence of ones and zeros to train the adaptive receiver, if the receiver observes more
than one bit interval of information at a time. By letting the observation window of
the receiver be two bit intervals, a full data bit (and possibly two bit transitions) are
guaranteed to occur within the observation window. This increase in the observation
window also allows for more than one user to be in the code acquisition mode at the
same time.
8.1 Updated System Model
In this section, the system model for an observation window of two bit intervals will
be presented. This model is just a special extension of that presented in Section 2.1,
however a notation similar to that presented by Madhow [59] will be used. As with
the previous system model, the receiver converts r(t) to a baseband signal and passes
R(t) through a filter matched to the chip pulse shape and the output of the filter is
sampled at the chip rate. As before, we let the quantity rm,n represent the nth chip
sampled output during the receiver's mth bit interval.
mTb+(n+1l)Tc
rmn : R R(t)dt E C. (8.1)
r2 mTb+nTc
mTb nT,
During each bit interval, a total of N chip samples are gathered and grouped with
the N chip samples from the previous bit interval such that
r(m) = (rmi,o, rml,, I .. r1,N1, rm,o0,rm,1, ..., tim,Ni) (8.2)
and it is this vector quantity that will be used as the input to the adaptive receiver.
As with the previous system model,
K rfkC2,
r(m) = E Jk(m)exp(JOk) +Nm E C2NX1 (8.3)
k=1l P
where Jk(m) is the contribution to the received vector from the kth user during the
mth receiver bit interval. Let the dk,m represent the data bit for the kth user that falls
completely in the two bit observation interval during the receiver's mth bit interval.
Unlike before, we now define the propagation delay for the kth user to be Tk E [0, Tb]
such that Trk = pkTc + kTc with Pk E {0,1,..., N 1} and 0 < 6k < 1, or Tk = NTc.
Let ak represent a vector of length 2N consisting of the N elements of the spreading
sequence for the kth user followed by N zeros.
ak = (Ck,o, ck,1,. .., Ck, 1, 0,..., O)T (8.4)
Let T(L) represent the nth left acyclic shift of a vector, while T() represents the nth
right acyclic shift of a vector. Both functions operate on and produce vectors with
2N elements. Using a general vector x = (X0o, X1,. .. X2N1)T these functions are
defined below.
T()L (x) = T() (x) = x (8.5)
(X)= (xl, X2,..., x2 1,0) (8.6)
TL(1)) (X2NI,..., O) (8.7)
Tl(x) = (0,xo,x,... x22T (8.8)
T(R2N1)(X) = (0,...,,) (8.9)
T(L(X) T(x) (0,...,0)T V n > 2N (8.10)
Using the above definitions we can now define the kth user's contribution to the
received vector as
Jk(m) = dk,mlVk1 + dk,mV + dk,m+lV (8.11)
where
Vk1 (1 6k)TW (ak) + kT.  (ak]) (8.12)
V = (1 6k)TR (ak) + kR +(ak) (8.13)
vk = (1 4k)Tw Pk)(ak) + ikTR+Pkl(ak) (8.14)
Using this notation, the autocorrelation matrix of the adaptive filter's input and the
steering vector are evaluated as shown below.
R E [r(m)r H(m)]
1 ( + V0(V)T + V(V) ) (8.15)
k=l1
+ 2or2I
p = E[r(m)dl,m] = v exp(j0,) (8.16)
8.2 Updated Timing Estimation Algorithm
We are interested in developing a new timing estimator for the propagation delay
of the desired user, now that we have increased the length of the filter. We will use
the same procedure as before. That is, the weights of an MMSE filter for a single
user will be used to form the timing estimator. Therefore, we need to calculate the
weights of the optimal filter for a singleuser operating in an AWGN channel. In
order to study the weights of the WienerHopf filter, we consider the case where we
apply the equivalent phasesynchronous model for the equations listed above. As in
the other system model, this means the filter weights become real and the complex
exponential term is removed from the steering vector expression. In this case, the
optimal weight vector is found using the WienerHopf equation with Rwopt = v,
where we let Wopt = avj1 + bv + cv1.
The coefficients of the optimal weight vector, a, b and c, are given by the solution
to the following equation.
[ + iv1]2 (vIV1)Tv (v)Tv a 0
(viv0 + IIV112 (v?) 1 b = 1 (8.17)
(via) v(0v ) v2
(v1)Tvi1 (vl)TV v 02 + 1v11J10
Using the above equation, we see that the a and c coefficients can only be zero when
both vi1 and v1 are orthogonal to v. This condition occurs when 61 = 0. However,
in the general case when 6j # 0, the a and c coefficients will be much smaller in
magnitude than the b coefficient, since v1 and v1 are only nonzero in one chip
position where v? is nonzero also.
Based on this result, we will ignore the contributions of the vi1 and v, vectors on
the WienerHopf weights when we develop the timing estimator. Using the weights of
the adaptive filter, we estimate the propagation delay for the desired user by finding
the set (/3,O,p,6) that minimize w(m) /3v exp(j0) l2. As in the case where the
receiver only observes one bit of information, the estimate of 1 is given by the set
{p, 6} which maximize the cost function f(p, 6) given below.
f(p,) = v0)Tw(m) (8.1)2
= ilV 112 (8,18)
Maximizing f(p, 6) with respect to 6 produces a quadratic equation whose coeffi
cients depend on the value of p. Therefore for each possible value ofp E {0, 1,..., N
1} we must find the roots of the quadratic equation D2 (p)2 + D, (p)p + D0o(p), where

Full Text 
CODE TIMING ESTIMATION IN DIRECTSEQUENCE
CODEDIVISION MULTIPLEACCESS
COMMUNICATION SYSTEMS
By
RONALD F. SMITH
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
1999
I have been blessed with a wonderful family, and this work is dedicated to them.
To Amy, for all of her support, encouragement and love. To Sarah, for so readily
sharing her silly little giggles and big hugs and kisses.
ACKNOWLEDGMENTS
I would like to express my gratitude to my adviser, Dr. Scott L. Miller, for his
guidance and feedback throughout the course of my research. I also thank the memÂ¬
bers of my Ph.D. committee, Drs. Leon Couch II, Jose Principe, Robert Fox, and
Richard Newman, for their interest and willingness to be a part of this effort.
While I have dedicated this work to my wife, Amy, she also deserves special
recognition. She has worked extremely hard supporting our family while I pursued
my graduate studies, and without her efforts this work wouldnâ€™t have been possible.
I thank my parents, Ernest and Joyce Smith, and my inlaws, Thomas and Peggy
Meneskie, for their continuous encouragement and support throughout this effort.
I also acknowledge my friends and colleagues, John Miller, Brad Rainbolt and Ali
Almutairi, who have made my time at the University of Florida an interesting expeÂ¬
rience.
iii
TABLE OF CONTENTS
page
ACKNOWLEDGMENTS iii
LIST OF FIGURES vi
KEY TO ABBREVIATIONS viii
ABSTRACT x
CHAPTERS
1 INTRODUCTION 1
1.1 Introduction to SpreadSpectrum Techniques 1
1.2 Mathematical Notation 6
1.3 Outline of the Dissertation 7
2 SYSTEM MODEL FOR THE AWGN CHANNEL 9
2.1 General Asynchronous DSCDMA System Model for the
AWGN Channel 9
2.2 Simplified Asynchronous DSCDMA System Model for the
AWGN Channel 12
3 REVIEW OF THE LITERATURE 14
4 THE MMSE RECEIVER AND SINGLEUSER CODE
ACQUISITION 21
4.1 The MMSE Receiver 21
4.2 Advantages of Synchronizing the MMSE Receiver 25
4.3 Adaptive Filters and the Estimation Algorithm 29
4.4 Summary 35
5 CHARACTERIZING THE PERFORMANCE OF THE TIMING
ESTIMATOR 37
5.1 Transient Statistics of the LMS Filter Weights 37
5.2 Chip Selection Error Probability 45
5.3 Approximation for the Conditional Variance of the Timing
Estimate 50
5.4 Coarse Acquisition Performance 54
IV
5.5 Summary 59
6 SIMULATING FADING CHANNELS 62
7 CODE ACQUISITION IN NONSTATION ARY ENVIRONMENTS 71
7.1 Frequency Synchronization Errors 75
7.2 Performance in FlatFading Channels 78
7.3 Summary 84
8 INCREASED WINDOW OF OBSERVATION AND TIMING
ACQUISITION 86
8.1 Updated System Model 86
8.2 Updated Timing Estimation Algorithm 88
8.3 Summary 93
9 SUMMARY AND FUTURE WORK 94
9.1 Summary 94
9.2 Contributions 96
9.3 Future Work 96
APPENDICES
A PROBABILITY OF BIT ERROR USING THE GAUSSIAN
APPROXIMATION 99
B DERIVATION OF TRANSIENT EQUATIONS FOR THE LMS
FILTER 102
B.l Transient Weight Autocorrelation Matrix Using the Gaussian
Approximation 102
B.2 Transient Weight Autocorrelation Matrix Using the Known
Statistics of the Receiverâ€™s Input Vector 106
C REVIEW OF QUADRATIC FORMS OF GAUSSIAN RANDOM
VARIABLES 112
D TYPOGRAPHICAL NOTES 119
REFERENCES 121
BIOGRAPHICAL SKETCH 127
v
LIST OF FIGURES
Figure page
1.1 Power spectral densities of the unspread data sequence and the direct
sequence spreadspectrum signals 3
4.1 Code acquisition with an adaptive receiver 22
4.2 Minimum meansquared error as a function of the propagation delay
of user #1 28
4.3 Plot of the cost function f(p,6) for a singleuser system 33
4.4 Plot of the cost function /(p, Ã“) for a fifteenuser system 34
5.1 Performance of equation (5.2) for a singleuser system, a) The analytiÂ¬
cal mean filter weight vector norm, E[w(m)] (*), and the simulated
mean filter weight vector norm w(m) (solid line) plotted as a funcÂ¬
tion of the training length; b) Comparing the error between the two
mean weight vectors, E[w(m)] â€” w(m) 43
5.2 Performance of equation (5.2) for a threeuser system, a) The analytiÂ¬
cal mean filter weight vector norm, E[w(m)] (*), and the simulated
mean filter weight vector norm w(m) (solid line) plotted as a funcÂ¬
tion of the training length; b) Comparing the error between the two
mean weight vectors, E[w(m)] â€” W(m) 44
5.3 Performance of equations (5.4) and (5.7) for a twouser system. The
Frobenius norm of the LMS filterâ€™s transient weight autocovariance
matrix; equation (5.4) (*), equation (5.7) (x) and simulation
results (+) 45
5.4 Comparing incorrect chipselection probability to the Unionbound for
a threeuser system. Simulation results (+), the Chernoff bound
using equations (5.2) and (5.7) (solid line), and the Chernoff bound
using simulated weight vector statistics (*) 51
5.5 Comparing the simulated conditional variance of the Â¿i estimate to the
analytically derived approximation in absence of nearfar MAI. ... 53
5.6 Comparing the simulated conditional variance of the 8\ estimate to the
analytically derived approximation in presence of nearfar MAI. . . 54
vi
5.7 Average training bits required for correct acquisition as a function of
the number of system users in AWGN channel 57
5.8 Average training bits for correct acquisition as a function of the level
of multiaccess interference in AWGN channel 58
6.1 Flatfading channel simulator 68
6.2 Theoretical and simulated probability density function for the enveÂ¬
lope of the fading channel simulator output for two different Rician
parameter values 69
6.3 Magnitude and phase of a slow flatfading Rayleigh process generated
with a carrier frequency of 1.8 GHz, a vehicle speed of 3 MPH, and
a data rate of 9600 BPS 69
7.1 Modified adaptive receiver with channel compensation 74
7.2 Acquisition performance of the adaptive receiver in a fiveuser environÂ¬
ment in presence of frequency offset errors 77
7.3 Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a thirdorder linear predictor for the channel estimaÂ¬
tion algorithm 80
7.4 Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a tenthorder linear predictor for the channel estimaÂ¬
tion algorithm 81
7.5 Comparing the average training bits for correct acquisition as a funcÂ¬
tion of the level of multiaccess interference in a flatfading channel. 82
7.6 Performance of the timing acquisition algorithm using the hybrid and
modified MMSE receivers 83
8.1 Plot of the cost function f(p,6) for a fifteenuser system when the
receiver observes two bitintervals 91
8.2 Comparing the performance of the timing estimator when the receiver
observes 2bit intervals (*) versus observing 1bit interval (solid line)
as a function of the MAI 92
A.l The MMSE receiver for the Gaussian approximation 99
Vll
KEY TO ABBREVIATIONS
AWGN: additive white Gaussian noise
BER: biterrorrate
BPS: bitspersecond
BPSK: binary phase shift keying
CRB: CramÃ©rRao bound
CTAN: Comprehensive TgX Archive Network
DSCDMA: directsequence codedivision multipleaccess
DSSS: directsequence spreadspectrum
FHSS: frequencyhopped spreadspectrum
IEEE: Institute of Electrical and Electronics Engineers, Inc.
ISI: intersymbol interference
LFSR: linearfeedback shiftregister
LMS: least mean squared
LOS: lineofsight
LPD: low probability of detection
LPI: low probability of intercept
LSML: largesample maximum likelihood
MAI: multiaccess interference
MASE: multipleantenna sensorsbased estimator
MMSE: minimum meansquared error
MOE: meanoutputenergy
MPH: milesperhour
MSE: meansquared error
viii
MUSIC: multiple signal classification
PCS: personal communication system
PSD: power spectral density
RASE: rapid acquisition by sequential estimation
RLS: recursive leastsquares
RMSEE: root meansquared estimation error
SNR: signaltonoise ratio
IX
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
CODE TIMING ESTIMATION IN DIRECTSEQUENCE
CODEDIVISION MULTIPLEACCESS
COMMUNICATION SYSTEMS
By
Ronald F. Smith
May 1999
Chairman: Dr. Scott L. Miller
Major Department: Electrical and Computer Engineering
This dissertation considers code timing estimation for asynchronous directsequence
codedivision multipleaccess communication systems operating over additive white
Gaussian noise and flatfading channels. Unfortunately, this type of parameter estiÂ¬
mation is difficult in the presence of multiple system users, and traditional methods
are known to be sensitive to the nearfar problem. The nearfar problem is a conÂ¬
dition which occurs when the received signal amplitudes of the the multiple system
users are very dissimilar.
A wellknown receiver, referred to in the literature as the minimum meansquared
error (MMSE) receiver, is used to form an estimate of the code timing for a single
user. The MMSE receiver has several desirable properties. The MMSE receiver
is the optimal receiver for a singleuser operating in an additive white Gaussian
noise channel, but is suboptimal in a multiuser scenario. However, the MMSE
receiver has been observed to be resistant to the nearfar problem. Traditionally,
the MMSE receiver is implemented using an adaptive filter and can, therefore, learn
and adapt to the ambient channel conditions. It will be shown, that with very little
side information, it is feasible to form an estimate of a singleuserâ€™s timing offset by
processing the weights of the adaptive filter. The performance of the proposed timing
estimation algorithm has been studied under several different scenarios, and has been
observed to be resistant to the nearfar problem. The complexity of the proposed
timing algorithm is similar to conventional singleuser estimation methods, and is
shown to be of lower complexity than other proposed timing estimation algorithms.
Others have studied the performance of the MMSE receiver in a flatfading chanÂ¬
nel. It has been found that the MMSE receiver must be modified such that the
received signal is compensated to offset the dynamic phase changes induced by the
channel. In certain environments, this compensation algorithm is shown to degrade
the performance of the timing estimation algorithm. A modification to the operation
of the receiver is proposed to offset some of this performance degradation.
xi
CHAPTER 1
INTRODUCTION
1.1 Introduction to SpreadSpectrum Techniques
In 1941, actress Hedy Lamarr and composer George Antheil filed for a patent
on a â€œSecret Communication Systemâ€ [1]. Their device operated on the idea of freÂ¬
quently changing the carrier frequency of a radio signal. If the carrier frequency was
updated fast enough, and in an apparently random fashion, an adversary would have
difficulty maintaining contact with the transmitting radio. This means that an adÂ¬
versary would only be able to receive a small portion of a message, whether it was
sent coded or uncoded. And, therefore, it is much more difficult for an adversary
to discern the intent or contents of the transmitted message. Of course, this means
that the intended receiver has to dynamically synchronize its carrier frequency with
that of the transmitter. Lamarr and Antheil used paper rolls, similar to those found
in player pianos, as their synchronization method. Hole placement on the roll would
determine the carrier frequency used, and an exact duplicate of the roll used by the
transmitter was given to the intended receiver prior the â€œsecretâ€ communication.
Lamarr and Antheil gave their invention, and the rights to the patent, to the United
States government. Since this development occurred during World War II, this techÂ¬
nique initially received a lot of attention. However, the U.S. government deemed the
mechanics of the synchronization method too complex for widescale military use.
The technique developed by Lamarr and Antheil, which is now known as frequency
hopped spreadspectrum (FHSS), is explained in the following few sentences. Let B
represent the bandwidth required to transmit the message signal, using a traditional
radio with a fixed carrier frequency. If we now let the carrier frequency hop through
N discrete carrier frequencies, each separated by B Hz, then the bandwidth of the
1
2
transmitted signal becomes NB. Therefore, the bandwidth of the transmitted sigÂ¬
nal using frequencyhopping has been spread to a wider bandwidth than what is
required to transmit the message signal. However, until the advent of the integrated
circuit, the problem of implementing a synchronization method limited the use of
such systems.
The dual of FHSS is known as directsequence spreadspectrum (DSSS). In DS
SS, a digital signal is modulated by a digital spreadingsequence prior to transmission.
Consider a binary sequence of Â±1 data where each data bit has duration T),. If this
data sequence is used to amplitude or phase modulate a carrier signal, the power
spectral density (PSD) of the transmitted signal will have a nulltonull bandwidth of
2/7), Hz. A spreading sequence is comprised of a pseudorandom sequence of Â±1 chips
where each chip has duration Tc, and it is assumed that TÂ¡Â¡ = NTC. By multiplying
the data sequence by the spreading sequence prior to transmission, the PSD of the
transmitted signal now has a nulltonull bandwidth of 2/Tc which is N times greater
than the corresponding bandwidth of the original unspread data sequence. The total
transmitted power in the unspread and the DSSS signals is assumed to be equal,
since multiplication by Â±1 doesnâ€™t affect the transmitted power. This means that the
PSD for the DSSS signal will have a much lower amplitude than the PSD for the
unspread signal as shown in Figure 1.1.
In the previous paragraph, we referred to the spreading sequence as being pseudoÂ¬
random. Ideally the chips of the spreading sequence would be randomly generated,
with +1 occurring with the same probability as 1. However, just as the intended reÂ¬
ceiver in a FHSS system requires knowledge (and synchronization) of the frequency
hopping algorithm, the intended receiver in a DSSS system must have knowledge
(and synchronization) of the spreading sequence. Therefore, some deterministic
method must be used to generate a spreading sequence. A pseudorandom sequence
can be generated with a linearfeedback shiftregister (LFSR), which is just a bank of
3
Figure 1.1: Power spectral densities of the unspread data sequence and the direct
sequence spreadspectrum signals.
flipflops interconnected with the appropriate feedback. Using an LFSR to generate
the spreading sequence, the transmitter and receiver must have common knowledge
of the length of the LFSR, the feedback connections, and the initial state of the LFSR
prior to transmission. With the appropriate feedback, â€”1 will occur with a probaÂ¬
bility slightly higher than +1. In addition the sequence will be periodic. However, if
the period is sufficiently long, over a short observation interval the occurrence of +1
and â€”1 will appear random.
The goals of a covert communication system are to communicate privately with
an asset and to communicate reliably under adverse conditions, possibly caused by
an adversary. Over the last several decades, spreadspectrum systems have received
considerable attention for their use as covert communication systems. The attention
is due to three properties of spreadspectrum systems some of which are hinted at
in Figure 1.1. The first property of interest is the resistance to intentional jamming.
One way for an adversary to disrupt a communication system is to broadcast at a
high power level in the same frequency band. Due to the increased bandwidth utilized
4
in a spreadspectrum system, a jammer then only achieves partial band jamming or
has to spread the jamming signal over a wider bandwidth. In either case, the effect of
the jamming signal is reduced compared to the effect observed by the same jammer
in the unspread communication system. The second property of interest is the low
probability of detection (LPD). Even if an adversary knows you are transmitting,
without knowledge of the spreading sequence (or frequency hop sequence), it is very
difficult for that adversary to perform a good detection of the message signal. The
third property, which is especially important for clandestine communication, it called
low probability of intercept (LPI). Considering the reduced amplitude of the DSSS
signalâ€™s PSD as shown in Figure 1.1, we see that with sufficient spreading the level of
the PSD can be made to approach the level of the background noise floor. This means
that an adversary with a spectrum analyzer that is actively looking for the spread
spectrum transmitter will have difficulty detecting that the transmitter is active. In
other words, in order to locate the transmitter the adversary will have to be physically
located very near the transmitter when a signal is being transmitted. Under ideal
conditions the transmitter will be active for short time intervals, or the asset is able
to detect the presence of the adversary in which case he/she turns the transmitter off.
These conditions greatly decrease the probability of intercept and, therefore, increase
the â€œlife expectancyâ€ of the asset.
As the remainder of this work deals with what is essentially a multiuser direct
sequence spreadspectrum system, we list several references which describe DSSS
systems in more detail. Scholtz [2] and Pickholtz et al. [3] provide good tutorials
on the properties and use of spreadspectrum systems. Ziemer and Peterson [4] give
an extensive examination of these systems. Many references on the generation and
properties of spreading sequences are available, as this subject continues to be a
current research interest [57].
5
In directsequence codedivision multipleaccess (DSCDMA) systems, each user
has a unique spreading sequence and all users occupy the same frequency band transÂ¬
mitting independently of all other users. Consider a K user DSCDMA system in
which each user transmits Â±1 times its spreading code. We will communicate with
one of the users, called the desired user, using a receiver that is optimized for a
singleuser operating in the presence of additive white Gaussian noise (AWGN). For
this example, the optimum receiver is a filter matched to the desired userâ€™s spreading
code. Therefore, the output of the matched filter will depend on the correlation of
the desired userâ€™s code with all of the other K â€” 1 usersâ€™ codes. If all of the usersâ€™
spreading codes are orthogonal for any amount of time shift between the codes, the
output of the matched filter will not depend on the presence of the other system
users. A more realistic condition is that the set of spreading codes is chosen so that
the crosscorrelation between any two codes is wellbehaved. This means the output
of the matched filter will be dominated by the desired user, but some component of
the output will be due to contributions of interfering users.
It is possible to receive each of the transmitted signals at a different power level.
For example, this condition could occur if any of the K â€” 1 interfering users is located
physically closer to the receiver than the desired user. In this case the output of
the matched filter may become dominated by the interference. This type of multiÂ¬
access interference (MAI) is referred to as the nearfar problem. The challenge for
communication systems engineers is to develop receiver structures that are insensitive
to the nearfar problem.
In this dissertation, we are concerned with the problem of achieving synchronizaÂ¬
tion with the spreading sequence of the desired user. Moon et al. [8] studied the
effects of MAI on traditional techniques (energy detector) that are used in single
user spreadspectrum communication systems. It was found that the average time
to achieve acquisition quickly increases when a nearfar scenario exists with only two
6
system users. Moon et al. [9] presented a multiuser synchronization method, called
the rapid acquisition by sequential estimation (RASE) algorithm, which essentially
tries to estimate the contents of the LFSR used to generate the spreading sequence for
each user, based on observations of the received signal. They assumed that all users
transmit only their spreading code and that the receiver knows the received ampliÂ¬
tude of each user. For a total of K users, this algorithm has a complexity of 0(K2k)
which is prohibitive for K > 10. A similar singleuser approach was presented by
Barghouthi and Stiiber [10,11], under the assumption that all users are received at
the same power level. Madhow and Pursley [12], considered the effects of achieving
code synchronization in the presence of MAI on the system capacity. They found
that the system capacity for achieving code synchronization with a matched filter
was much less than the system capacity measured with respect to the biterrorrate
(BER). These results form the motivation for researching code acquisition techniques
for DSCDMA systems.
1.2 Mathematical Notation
Matrices and vectors are typeset in a bold face, for example v or V. In general,
a lowercase letter will be used to denote a vector, while an uppercase letter denotes
a matrix. Unless otherwise noted, all vectors are assumed to be column vectors. In
order to denote the nth element of a vector, a subscript will be used such as vn. The
(i,j)th element of a matrix V is denoted by VÂ¿Â¿. A subscript containing the variable
k denotes an element belonging to the kth user in the DSCDMA system.
The transpose, the conjugate (or Hermitian) transpose, and the 2norm of a
columnvector v are denoted by vr, \H and v = VvHv, respectively. The conjuÂ¬
gate of a scalar quantity will be denoted as a*. For a matrix V we will use VF to
7
denote the Frobeniusnorm of the matrix defined as
(1.1)
i j
The Kronecker delta function is defined as
1 if m = 0,
(1.2)
0 otherwise.
The expectation of a random variable, /?, either a scalar or nonscalar quantity, will
be denoted as E[f3] or Â¡3. We use the notation O(N) to denote that an algorithm has
numerical complexity on the order of N. When a is a parameter to be estimated, we
denote the estimate as a.
1.3 Outline of the Dissertation
In the next chapter, a system model that is widely used to simulate a DSCDMA
communication system will be presented. The system model is expressed in a conveÂ¬
nient vectormatrix notation which facilitates simulating a DSCDMA communication
system on a computer. The third chapter provides a brief review of the literature
concerning DSCDMA receivers and timing acquisition techniques. While it is imposÂ¬
sible to provide a complete review of these subjects, an effort is made to provide the
reader with enough of a review to understand several timing acquisition techniques
and their associated complexities. In the fourth chapter, a lowcomplexity singleuser
timing estimation algorithm is presented. The algorithm is based on processing the
weights of an adaptive filter, which is commonly referred to as the MMSE receiver.
The MMSE receiver is a singlebit singleuser detector, which is a suboptimal receiver
for a multiuser DSCDMA communication system, and is receiving considerable atÂ¬
tention in the literature. As several papers in the literature ignore the effects of
operating the MMSE receiver asynchronous to the intended user, several examples
8
of why synchronization is important are presented. In the fifth chapter, the perÂ¬
formance of the timing estimator is characterized. Several analytical techniques are
used to characterize the performance of the timing estimator, based on a developed
statistical model of the transient statistics of the MMSE receiver weights. Simulation
results are used to study the performance of the timing acquisition algorithm for
a general asynchronous DSCDMA system when two commonly used adaptive algoÂ¬
rithms are used in the MMSE receiver. It will be shown that for one of these adaptive
algorithms, the performance of the timing acquisition algorithm is nearfar resistant.
In the sixth chapter, the effects of fading on a communication system are discussed
and a commonly used implementation of a flatfading channel simulator is presented.
The performance of the timing acquisition algorithm when the receiver is operating
in two nonstationary environments (frequencyoffset errors and flatfading channels)
is studied in the seventh chapter. Barbosa and Miller [13] have presented a modified
version of the MMSE receiver that can be used in flatfading channels. The perforÂ¬
mance of the timing acquisition algorithm for this receiver will be studied. Based on
simulation results, it will be shown that a slight improvement in the performance of
the timing estimator is feasible, through a slight modification in the training cycle of
the modified MMSE receiver. In earlier chapters, the system model and the timing
estimation algorithm were developed such that only one of the DSCDMA system
users can be in the mode of acquiring timing acquisition. In the eighth chapter, a
modification to the system model and the timing estimation algorithm is presented
such that more than one system user can be in the timing acquisition mode. In the
final chapter, a summary of the work presented in this dissertation, and the contriÂ¬
butions to the area of DSCDMA research, as well as several areas of future research
are provided.
CHAPTER 2
SYSTEM MODEL FOR THE AWGN CHANNEL
2.1 General Asynchronous DSCDMA System Model for the AWGN Channel
In this section, a general system model for the case when the receiver operates
in an AWGN channel and observes only one symbol interval of information at a
time will be presented. After this general model is developed, a simplified version,
which will eventually be shown to lead to a reduction in the numerical complexity
of evaluating the timing estimator, will be presented. The only difference between
these two models, is how we interpret or apply the effect of carrier phase offset to the
input of the receiver for the desired user.
In this work, a binary phase shift keying (BPSK) communication system is used.
There are a total of K system users, each operating at the same carrier frequency /c,
but asynchronous to and independent of the other users. The kth user is assumed to
transmit a signal whose complex envelope [14], Sk(t), is a polar data sequence, dk(l) 6
{+1,1}, which has a unique signature or spreading waveform, ck(t), superimposed
upon it such that
OO
Sk(t) = Y, MM  lTb) ke{ 1,2,..., K}. (2.1)
l=â€”oo
Therefore, the signal actually transmitted by the kth user is given by
sk(t) = Re [Sk(t) exp{j2nfct}] = Re [Sk(t) exp{jujct}] k â‚¬ {1, 2,..., K} (2.2)
where Re[ ] returns the real part of its argument. Without loss of generality, it is
assumed that the first user is the desired user and all other users act as interference.
The data bits have a duration Tb while the chips of the spreading sequence have
duration Tc. Each spreading sequence has a period of N = Tb/Tc chips. That is, one
9
10
period of a spreading sequence is equal in duration to one bit interval. While this is
not a requirement for DSCDMA systems in general, it is required for the receiver
structure studied in this work. Let the vector c*, = (ck,o,Ck,i, â– â– â– ,CktN~i)r represent
one period of the kth userâ€™s spreading sequence so that the spreading waveform can
be written as
JVl
Cfc(i) = ^2 ck,nPTc{t  nTc) (2.3)
n=0
where Prc(0 is the chip pulse shape which is taken to be 1 over the interval [0,TC)
and zero otherwise.
We assume that the receiver is asynchronous in time and phase to the transmitted
signal of the desired user, as well as being asynchronous to all of the interfering users.
Due to the K system users, the received signal is of the form
k
r(t) = ^2 yfifie Re [Sk(t ~ rk) exp {jujct + jOk}\ + n(t). (2.4)
k=1
In the above expression, rfc and Pk are the propagation delay and the average received
signal power, respectively, for the fcth user, Ok is the phaseoffset for the kth user,
while n(t) is the additive white Gaussian noise which is assumed to have a onesided
power spectral density of Wo.
In order to process the received signal, the receiver converts r(t) from a bandpass
signal to a baseband signal, R(t), which is then passed through a filter matched to
the chip pulse shape and the output of this matched filter is sampled at the chip rate.
It is assumed that any doublefrequency terms created in the conversion to baseband
are not passed through the chip matched filter. During the receiverâ€™s mth bit interval
the nth chip sampled output can be expressed as
mTj+(n+l)rc
râ„¢=vmrc / mdt 6C
mTb+nTc
(2.5)
11
During each bit interval, the N chip samples are stored in a received vector,
r(m) = (rmio,rmii,..., rm,ni)t and it is this vector quantity that will be processed
by the receiver. Using the above definitions for how each user contributes to r(t) we
can express how each user contributes to r(m) using a notation similar to that used
by Miller [15]. It can be shown that
k I p
r(m) =V W^Jfc(m)exp{;6Â»fc} + Nm eCNxl (2.6)
LJ V n
k=1 T 1
where Jk(m) is the contribution to the received vector from the kth user during the
mth receiver bit interval given by
Jfc(m) = [22fci(m)a2fc_i(pfc,4) + z2k(m)a2k(pk,5k)]
where
(2.7)
z2ki{m) = [dk(m) + dk(m  l)]/2 G {1, 0,1}
Z2k(m) = [dk(m)  dk(m  l)]/2 G {1, 0,1}
a2jfci(Pfc,Ã¡fc) = (1  h)c{kk) +6kc^k+1)
a2fe(pfc,4) = (l4)c?fc) + 4c^+1)
Ck,m â€”
G {+1, â€”1} if m G {0,1,..., N  1},
otherwise.
(2.8)
(2.9)
nk (ck}Nâ€”ni Ck,Nâ€”Ti+l) â€¢ â€¢ â€¢ > Ck,Nâ€”1> Ck,0i â€¢ â€¢ â€¢ i Ck,Nâ€”nâ€” l) (2.10)
Ã±fc ( Ck,Nâ€”m Ck,Nâ€”n+1, â€¢ â– â€¢ i Ck,Nâ€”1) Cfc,0? Ck,\i â€¢ â€¢ â€¢ > Ck,Nâ€”nâ€”l) (2.11)
(2.12)
and rk = pkTc + 6kTc with pk G {0,1,..., N â€” 1} and 0 < Sk < 1.
Since the receiver is asynchronous to any of the K system users and only observes
one bit interval of information, it is possible to have a data bit transition occur during
the receiverâ€™s bit interval. This fact is accounted for in the previous equations. The
quantities z2fc_i(m) and z2k(m) are used to indicate if adjacent data bits for the &th
12
user are similar. If adjacent data bits are the same, then Z2fci(m) will be nonzero,
otherwise z2k('m) is nonzero. The vector quantities cÂ¡Â¿n) and cj;n) denote the nth right
cyclic and the nth modified right cyclic shifts of the kth userâ€™s spreading sequence,
respectively. As shown in the previous equations, JÂ¡t(m) is just a function of two
slightly offset cyclic shifts of the spreading sequence for the kth user.
In the remainder of this work, we will refer to the propagation delay in terms
of the integer part, pk, and the fractional part, SÂ¡t, only. The noise vector, Nm, in
equation (2.6) consists of independent complex Gaussian random variables whose real
and imaginary parts are also independent having zeromean and equal variances of
cr2 = N/(2Eb/J\f0), where Eb is the average received energy per bit of the desired
user. In the remainder of this work, we define the signaltonoise ratio (SNR) as
SNR = Eb/Mo.
2.2 Simplified Asynchronous DSCDMA System Model for the
AWGN Channel
In the previous section, the receiver was assumed to be phase asynchronous to any
of the K systems users. This is a reasonable assumption, since it is highly unlikely
that all of the oscillators for the users will be phase synchronous as viewed by the
receiver. Certainly, it could be possible for all of the transmitted signals to be phase
synchronous, a situation that would arise at a transmitting base station in a cellular
type communication system where one oscillator can be used to transmit all of the
K userâ€™s signals. At any of the mobile receivers, then one could say that all users
experience the same propagation delay and carrier phase shift in the channel from
the base station to the mobile receiver. However, in the communication link from the
mobile user to the base station, different propagation delays and carrier phase shifts
will exist due to the different physical locations of the mobile users.
13
However, we can model the system as being equivalently phasesynchronous as
follows. We can express equation (2.6) as
r(m)=y',/^(cos(Â«t)+JSin(Â«t))Ji(m)+Nâ€ž â‚¬ CKxl. (2.13)
iriv
We then can create a modified receiver input by only processing the real part of r(m).
That is,
K
r(m)  Re [r(m)] = ^
k=1
where the noise vector, Nm, in equation (2.14) consists of realvalued Gaussian
random variables that are independent having zeromean and equal variances of
a2 = N/(2Eb/Af0).
Now the cos(6k) term can be absorbed into the ratio of Pk/P\â– Therefore, by
adjusting the level of the SNR appropriately one can equivalently model the effects of
the carrier phase offset for the desired user. To account for the effects of the carrier
phase offset for the other users, the value of Pk/P\ for each interfering user is adjusted
as appropriate.
As a result of using this procedure to form an equivalently phasesynchronous
system, we will eventually show that the numerical evaluation of the timing estimator
for the propagation delay for the desired user is simplified. In fact, we will see that
for this model the estimate of <5i, the fractional part of the desired userâ€™s propagation
delay can be expressed in closed form, which is not the case when the model of
Section 2.1 is used.
cos(0fc)J k{m) + Ã‘r
r>Nxl
(2.14)
CHAPTER 3
REVIEW OF THE LITERATURE
Since the area of DSCDMA has received so much attention in the literature over
the last several years, it is impossible to review all contributions made in the recent
past. However, in this chapter, we will try to summarize any major contributions
as well as those that are directly related to code acquisition in DSCDMA systems.
We will discuss the complexity of the timing estimation algorithms, so that we may
later compare them to the complexity of the timing estimator presented in the next
chapter.
VerdÃº [16,17] provided a major breakthrough in DSCDMA research by proving
the existence of an optimum multiuser receiver that is nearfar resistant. The receiver
is comprised of a bank of matched filters, one matched filter for each user, followed
by a Viterbi decoder. The receiver requires a lot of sideinformation, in order to
create the matched filters and form a decision metric for the Viterbi algorithm. The
receiver requires knowledge of each usersâ€™ spreading code, and must be synchronized
with each user. In addition the receiver must know the received signal energy for each
user. The complexity of the receiver is 0(2A) for each binary decision made. The
receiver is considered impractical due to this complexity, which provides motivation
for investigating suboptimal receiver structures.
Lupas and VerdÃº [18,19] have proposed a suboptimum multiuser receiver which
is known as the linear decorrelating receiver. The receiver is similar to the optimum
receiver [16,17] in that it is comprised of a bank of matched filters, but does not
require knowledge of the received signal energy for each user. The outputs of the bank
of matched filters are processed by a linear transformation (called the decorrelating
filter) that removes the multiuser interference from the output of each filter. The
14
15
decorrelating filter cancels the MAI by accounting for the known correlation between
the usersâ€™ spreading codes. The complexity of the receiver is dependent on taking the
inverse of a matrix, which has complexity 0(N3), to form the decorrelating filter. If
this operation can be done once, the complexity of the receiver is then linear in the
number of users, which is a great improvement over the exponential complexity of
the Viterbi based decoder.
Strom et al. [20] studied the impact of timing errors on the decorrelating receiver.
It was found that the decorrelating receiver loses its nearfar resistance with the presÂ¬
ence of timing errors. That is, the BER of the receiver reaches a nondiminishing
floor even as the SNR increases, when errors are present in the estimates of the
timing offsets. In addition, they observed that the required variance of a timing
estimate is inversely proportional to the level of the MAI. This means as the MAI inÂ¬
creases, making timing estimation more difficult, a better timing estimate is required
in order for the receiver to maintain its nearfar resistance. Zheng and Barton [21]
perform a similar analysis on the decorrelating receiver. In addition to considering
time synchronization errors, they studied the effects of phase synchronization errors
and frequency synchronization errors. They found that if these quantities are limited
to small, and possibly unrealistic, values the decorrelating receiver still offers a sigÂ¬
nificant advantage over the conventional matched filter. These results motivate the
need for the development of timing estimators that produce unbiased estimates with
low variances, even in the presence of a severe nearfar environment.
Varanasi and Aazhang [22] have proposed a suboptimal multiuser detector whose
complexity is also linear in the number of users. The receiver requires knowledge of
the received signal strengths of each user, the code sequence of each user, and time and
phase synchronization with each user. The receiver is comprised of a bank of matched
filters (one per user) followed by a multistage interference rejection algorithm. The
interference rejection algorithm is summarized in the following sentences. Consider
16
the output of one of the matched filters. If a decision is made on the filter output, it
may be corrupted by the interference caused by the other users. If the receiver has
perfect knowledge of the other usersâ€™ received signal energies and transmitted data
sequences, the receiver could directly calculate the interference at the output of the
matched filter and subtract this quantity from the filterâ€™s output. Then the receiver
could make a decision on an interferencefree quantity, and the decision made for
each user would be optimal. However, the receiver does not have perfect knowledge
of such quantities and, therefore, can only form imperfect estimates of the MAI term
at each filter output. The receiver reconstructs the estimate of the MAI at each
filter output during the mth interval, based on the decisions formed in the (m â€” l)th
interval. Since the reconstruction of the MAI is based on these decisions, perfect
cancellation of the MAI term from the filter output will not occur (and hence the
receiver is suboptimal).
The previous paragraphs give a brief overview of what has been done in terms of
multiuser detectors for DSCDMA. As mentioned the complexity of the these techÂ¬
niques, and the amount of required sideinformation, makes them infeasible. However,
all of the receiver structures mentioned require some form of synchronization with the
user spreading sequences. In the remainder of this chapter, we will focus our attention
on techniques that have been proposed to address the issue of code synchronization.
One area of parameter estimation that is receiving considerable attention in the
literature are estimation techniques based on subspace methods [2326]. While Strom
et al. [23] consider the more general case of sampling the received signal more than
once a chipinterval, we restrict our attention to the system model given in Section 2.1.
We modify the system model slightly, by assuming that all users can transmit random
data sequences. Using this system model, the receiver is operating in a stationary
environment. The autocorrelation matrix of the received signal, which is defined as
17
R = E[r(ra)rH(ra)], can be written as
^' p
r = [a2fci(Pfc,4)a2fc_i(Pifc,4) + ai2k(Pk,h)^k{pk,Sk)] + 2
i i 1
where I is an N x N Identity matrix. Note that R is symmetric, and is also positive
definite as long as a2 > 0. When 2K < N, the eigendecomposition of R can be
expressed as
R =
As 0
0 An
H
Es En
(3.2)
As is a diagonal matrix of the 2K largest eigenvalues of R and Es is an A x 2K
matrix of the corresponding eigenvectors. Likewise, Aâ€ž is a diagonal matrix of the
(N â€” 2K) smallest eigenvalues (all equal to a2) of R and En is an N x (Nâ€”2K) matrix
of the corresponding eigenvectors. The signal subspace is defined to be the subspace
spanned by the set of &2ki(Pk,tik) and a2k{Pk,h) vectors VA: â‚¬ {1,2,..., K}. The
noise subspace is defined to be the orthogonal complement to the signal subspace. The
columns of As form an orthonormal basis for the signal subspace, and the columns
of An form an orthonormal basis for the noise subspace.
The basis of the multiple signal classification (MUSIC) algorithm, is that since
the set of {al5 a2,..., a2ic} vectors are in the signal subspace, they are orthogonal
to the noise subspace. Therefore, given perfect knowledge of the noise subspace, one
could find Tfc as the solution to E,f &2ki{Pk, h) â€” 0 Typically, the receiver does not
have perfect knowledge of the noise subspace or R. However, it can form an estimate
of R using the sample correlation matrix defined as
1 A
= m (33)
m=1
Given Rw, the receiver can form an estimate of the noise subspace, Ã‰n, an operation
that has complexity 0(N3). Strom et al. [23] use this idea to form a timing based on
18
the MUSIC algorithm. They find an estimate of the kth userâ€™s propagation delay as
MU
Tk
 arg mm
re[0,T6)
lÃ‰"a2k,(T)
2 + Ã‰Â«a2l(r)2
a2ki(r) +a2/b(r)
(3.4)
The independent work presented by Bensley and Aazhang [24] is similar, but also
addresses the problem of estimating the channel gain and phase in a timeinvariant
multipath channel. As a natural extension of their earlier work, Strom et al. [25]
consider using the MUSIC based estimator in a timevarying channel.
The MUSICbased timing estimator requires very little side information in order
to form its estimate. The number of system users, as well as the spreading code (to be
used in equation (3.4)) is required. While the complexity of the algorithm is 0(N3),
it requires no training period, and can be used to estimate each userâ€™s propagation
delay. However, the algorithm does not work for 2K > N, as the noise subspace
of R has zero rank. It has been noted that for low signaltonoise ratios and little
nearfar effect, that the traditional correlator performs better than the MUSICbased
algorithm. However, as the MAI interference increases the MUSICbased algorithm
outperforms the correlator. It has also been noted that the MUSICbased algorithm
is resistant to the nearfar problem [23,24],
Zheng et al. [27] have proposed a singleuser propagation delay estimator, using
the system model of Section 2.1, that is known as the largesample maximum likeliÂ¬
hood (LSML) method. This method models the received signal as a known training
sequence (the desired signal) and all other signals including the interfering signal
and thermal noise as unknown colored Gaussian noise that is uncorrelated with the
desired signal. The resulting timing estimate is found by rooting a second order polyÂ¬
nomial. The coefficients of the polynomial are dependent on a matrix inverse, Q1,
where Q is related to the RM matrix given in equation (3.3). In order to evaluate
the matrix inverse, Q must be full rank. This means that the LSML timing estimate
is unavailable until the receiver has observed at least M â€” N samples of r(m).
19
As an extension to the LSML timing estimator, Liu et al. [28] have proposed a
similar timing estimation scheme known as the multipleantenna sensorsbased estiÂ¬
mator (MASE). That is, the estimator is derived using the known training signal and
modeling the MAI and the additive noise as an unknown colored Gaussian random
process. In the MASE algorithm, the receiver observes the outputs of an arbitrary
antenna array of L sensors. In fact, when L = 1 the MASE algorithm reduces to the
LSML algorithm. Like the LSML, the MASE algorithm requires a training sequence,
and the timing estimate is formed by rooting a second order polynomial. Once again,
the coefficients of the polynomial are dependent on the inverse of a matrix. In the
LSML algorithm, the receiver has to acquire at least M > N samples of r(m) in
order for the Q matrix to have full rank. By using L antenna sensors, the MASE
algorithm is able to average over the outputs of the sensors and hence only requires
LM > N samples of r(m) in order for the Q matrix to have full rank. Since the
MASE algorithm requires fewer training symbols than the LSML algorithm, under
conditions where the channel is timevarying, the MASE has the advantage over the
LSML algorithm [28].
In the next chapter, a timing estimation algorithm is presented that is based on
processing the weights of the MMSE receiver. In this dissertation, it is assumed
that the MMSE receiver uses an initial training sequence to adapt the weights to
minimize the MSE between the filter output and the data sequence for a single user.
This means that a known data sequence is used to train the MMSE receiver, prior
to performing data detection. Honig et al. [29] have developed a singleuser detector
based on minimizing the meanoutputenergy (MOE) of an adaptive filter. Note
that the term MOE in actuality is used to denote the variance of the filter output.
The proposed receiver [29] requires knowledge of the timing and spreading sequence
of the desired user. The weights of the filter are expressed in a canonical form as
w(m) = Ci + x(m), where Ci is the spreading code of the desired user and x(m) is
20
orthogonal to ci. The vector x(m) is updated to minimize the variance of the filter
output. This adaptive algorithm does not require knowledge of the desired userâ€™s
data sequence, and, therefore, is known as a blind adaptive algorithm.
Madhow [30] uses the blind adaptive receiver in an adhoc fashion to form an
adaptive receiver which only requires knowledge of the desired userâ€™s spreading seÂ¬
quence. Madhow uses a system model similar to that presented in Section 2.2, but
observes two bitintervals (27),) of the received signal, and, therefore, has 2N samÂ¬
pled outputs of the chipmatched filter comprise the received vector. The propÂ¬
agation delay for the desired user is assumed to be one of 2N hypotheses, T\ â‚¬
{0,0.5TC, 1TC, 1.5TC,..., (N â€” 0.5)TC}. The received signal is processed by a bank of
2N parallel blind adaptive filters (one for each n hypothesis) during a blind acquisiÂ¬
tion cycle. Let the fth filter be the adaptive filter that has the lowest MOE at the end
of the acquisition cycle. At the end of the acquisition cycle the receiver chooses two
filters, the zth filter and either the (i â€” l)th or (i + l)th filter, to form the adaptive
receiver. The output of the Â¿th filter is combined with the other filter in an adhoc
fashion in an attempt to combine the timing hypotheses that are closest to the true
propagation delay for the desired user. Using this method, Madhow is attempting
to create a receiver structure that performs joint acquisition and detection of the deÂ¬
sired userâ€™s signal. However, this method is very complex and doesnâ€™t really achieve
synchronization with the desired user. In the next chapter, we demonstrate why synÂ¬
chronization is important by considering the effects of being a halfchip asynchronous
to the desired user.
CHAPTER 4
THE MMSE RECEIVER AND SINGLEUSER CODE ACQUISITION
As an alternative to using the complex estimation techniques presented in the
previous chapter, we present a singleuser estimation algorithm which is based on
processing the weights of an adaptive receiver. One benefit of this receiver structure
is that the only side information the receiver requires in order to form its estimate is
the spreading code of the desired user. In addition, the same structure may be used
for data detection. This receiver structure, which has commonly been referred to as
the MMSE receiver, has received significant attention in recent literature.
4.1 The MMSE Receiver
In this section, using certain system assumptions, we will discuss how the weights
of the MMSE receiver, shown in Figure 4.1, may be used to estimate the propagation
delay of the desired user. The weights of the receiver are chosen to minimize the mean
squared error E [e(m)2], where e(m) is the difference between a desired response
and the filterâ€™s output. The vector which minimizes the meansquared error (MSE)
is wellknown to be given by the WienerHopf [31] equation
w(m) = Râ€x(m)p(ra) (4.1)
where R(m) = E [r(m)rH(m)] and p(m) = E[dJ(m)r(m)] are the autocorrelation
matrix and the steering vector, respectively.
The received vector r(m) corresponds to chipsamples of one period of the spreadÂ¬
ing sequence. Since the receiver is initially asynchronous to the desired user, it is
possible for data bit transitions to occur anywhere within the length of the received
vector. In order to avoid these bit transitions for the desired user, we assume that the
21
22
* Estimation t i
w(m) Â», algorithm
r(t)
Convert to
baseband
+ chip
matched
filter
r(m)
.Z.
^ Adaptive
filter, w(m)
H
r (m)w(m)
sgn[Re()]
d\{m)
+
r
Adaptive
e(m) vJJ* g
algorithm
^ i L
d, (m)
Figure 4.1: Code acquisition with an adaptive receiver.
desired userâ€™s data bit is a constant d\{m) = 1. In other words, the desired user will
transmit an all ones data sequence for purposes of code acquisition. A sidechannel
is used to control the addition of new users into the system, so this restriction is not
unrealistic. Based on the system model presented in Section 2.1, we see that the filter
is operating in a stationary environment. It is assumed that the usersâ€™ data sequences
and carrier phases are independent of each other and also independent of the additive
white Gaussian noise. The autocorrelation matrix and the steering vector are given
by the following equations.
R = E [r(m)rH(m)] = ai(pi, (pi, Â¿i) + 2
^' TV
+ 5Z 2P~ [a2*!(P*â€™ 6k)aÃkÃPk, h) + a2k(Pk, 8k)s%k(pk, 5*)]
k=2 Z 1
p = ax (pi, Â¿i) exp^'fli)
(4.2)
(4.3)
where I is an N x N Identity matrix. The MMSE occurs when the filter weights are
optimal, and is given by
Jmm = 1  p"R *P
(4.4)
23
In order to solve the WienerHopf equation for the optimal weight vector. wopt,
we have to invert an N x TV matrix, an operation that has complexity of 0(N3).
Since R is a symmetric matrix, it can be expressed as
R = VAVr, (4.5)
where V is a matrix whose columns contain the normalized eigenvectors of R and
A is a diagonal matrix that contains the corresponding eigenvalues. This notation is
referred to as the eigendecomposition of the R matrix. Using the above notation,
we can then express the inverse of the autocorrelation matrix as
R1 = VA1Vr. (4.6)
Instead of calculating R1 directly, we note that it can be shown that an eigenvector
of P = R â€” 2<72I is also and eigenvector of R. For each eigenvector of P, adding
2
matrix R. Therefore, if we find the eigendecomposition of P we immediately have
the eigendecomposition of R.
In equation (4.2), we note that each asynchronous user contributes at most two
outer product terms in the expression for P. If 2K < N, then the P matrix will
not be fullrank and will have at most 2K nonzero eigenvalues. Miller [32,33] used
this fact to simplify the calculation of the eigendecomposition of P, by working
on a 2K x 2K matrix instead of on a N x N matrix. For large ratios of N/(2K)
this technique produces a significant reduction in the complexity of calculating R_1.
However, in that work the desired user was assumed to be synchronous (ri = 0) with
the receiver, so one must slightly modify Millerâ€™s notation in order to process the
desired user in the more general asynchronous case. Of course when 2K > N, this
technique is not valid and one must find the eigendecomposition of an N x N matrix
(either R or P).
24
In order to illustrate the dependence of the filter weight vector on the amount
of multiaccess interference, we will observe the optimal filter weight vector for two
cases. For the first case let there be only one chipsynchronous system user, which
implies that r(m) = ciPl) exp (j0i) + Nm. Using the WienerHopf equation, it is found
that
wopt =
cjPl)exp(j6>i)
ci2 + 2(72
(4.7)
Assuming Ci has good autocorrelation properties, this result shows that we could
conceivably form an estimate of pi, the propagation delay of the desired user, by
finding the peak of the magnitude of the crosscorrelation between Ci and wopt.
For the second case, a second synchronous user is added to the system such that
r(m) = CiPl) exp (j0i) + d2{m)y/P2/P\c2 exp (jd2) + Nm. For this second example,
the optimal filter weight vector is
wopt =
7
(c?l) â€” 7C2) exp (jOi)
c12 + 2ct2 7c^ciPl)
,T_(Pi)
C2 C1
(4.8a)
(4.8b)
Ic22 + 2 (ft) (j2
Note that 7 is bounded in magnitude by the value of the correlation coefficient of
the two spreading codes, and is therefore expected to be much less than one. In
fact, in the limit P2/P\ â€”>â– 00, we see that wopt is the projection of c[Pl' onto a
vector orthogonal to c2. In the ideal case that c[Pl) is orthogonal to c2, the receiver
completely rejects the second user even though it is received at a much higher power
level. In the general case, for codes with good correlation properties (c^c[Pl) is small),
the filter should be able to suppress the second user. In the general case, the addition
of the second user causes a small perturbation of the filter weight vector about the
solution given by equation (4.7). We should still be able to form an estimate of px as
described above.
25
Equation (4.8a) is used as a basic example of why a timing estimator based on
processing the filter weight vector should even be considered. As more users are inÂ¬
cluded in the DSCDMA system, additional terms due to the spreading codes are
added to the expression for wopt. In order to illustrate the dependence of wopt on the
desired userâ€™s spreading code in the above example, and to simplify the resulting exÂ¬
pressions, we assumed that the interfering user was chipsynchronous to the receiver.
In the general case, each asynchronous interfering user contributes two similar terms
to the expression for wopt. However, if the set of spreading codes has sufficiently good
crosscorrelation properties one would expect that wopt is closely related to Ci such
that a reasonable estimate of pi is still feasible.
4.2 Advantages of Synchronizing the MMSE Receiver
Consider using the MMSE receiver in a singleuser system. One might ask why
synchronization of the MMSE receiver is even an issue that must be considered. If
the MMSE receiver is used for detection, are the weights optimized to give us the best
performance? The answer is maybe. The weights are optimized for the given value of
tÃ, the propagation delay of the desired user relative to the receiver. It is conceivable
that Jmin at Ti = 0 is less than the resulting Jmin when tx Â± 0. Therefore, we should
prefer to operate the receiver synchronous to the desired userâ€™s received signal. In
order to expand on this idea, we present three brief cases that demonstrate some of
the costs associated with using the MMSE receiver asynchronous to the desired user.
For the first case, consider using the MMSE receiver to demodulate data in a
r(m) =
wopt =
al(ifi! ^l) + Nm
ai(pi,Â¿i)
llai(Pi,Â¿i)2 + 2
single user system where
(4.9)
(4.10)
26
Consider the signaltonoise ratio of the output statistic z â€” rH(ra)wopt, which is
defined as SNR0 = E[z]2 / Var(z). Note that E[z\ = af (pi, Â¿i)wopt, and the expresÂ¬
sion for Var[z] is evaluated using Var[z] = Â£'[w^tr(m)rii(m)wopt] â€” E[z]2. Direct
substitution of the appropriate terms into this expression produces
= 2i,Â¿1)2
(la,(Pi,Â¿i)J + 2<7Â»)Ã (411)
2 a2E[z\
lai(pi,5i)2 + 2 a2
The result of these expressions is that SNR0 = ai(pi,Â¿i)2/(2cr2). If we assume
that the spreading code has good autocorrelation properties such that the terms
involving cfc^1* are negligible when compared to other terms in ax(px, Â¿x)2, we can
approximate the output signaltonoise ratio as
SNR0
[(1  Â¿02 + Sj] N
2 a2
(4.12)
where a2 = N/(2Eb/J\f0) as previously defined. When Â¿x is nonzero, there is an
effective loss in the output SNR of the output statistic z. If Â¿x is uniformly distributed
over the interval [0,1), then the average loss in the output SNR is 2/3 (1.8 dB). This
loss will adversely affect the capacity of the communication system which we typically
want to maximize. Note that this effect can be offset by sampling the received signal
more often than once a chip interval. However, increasing the sampling rate directly
increases the length of the received vector, which leads to an increase in the complexity
of the receiver.
Madhow [34] presents a method where an adaptive receiver is used to perform
joint detection and acquisition by increasing the length of the received vector and the
filter to 2N chip samples. By increasing the length of these vectors to 2N elements,
the desired user does not have to transmit an all ones training sequence. In addition
it may be possible for more than one user to be in the training mode at the same
time. The receiver structure performs joint detection and acquisition without actually
27
achieving chip synchronism with the desired signal [34], It is important to ask how
the performance of the MMSE receiver is affected by the lack of chip synchronism.
To address this issue, we calculated the JmÂ¡n of the MMSE filter as a function of the
propagation delay of the desired user when the SNR was 10 dB and the length of
the received vector and filter was set to 2N chip samples. This result is shown in
Figure 4.2 for propagation delays in the range of [0,5) chip intervals. Similar results
are obtained for propagation delays in the range of [5,31) chip intervals but were
omitted for clarity of the plot. There are two things to notice in this plot. If one
were to consider a system that is nearly chip synchronous then it appears that the
performance of the filter, based observing the MMSE, is unaffected by the propagation
delay of the desired user. This result could lead one to the incorrect conclusion that
achieving code synchronization is not important for the MMSE receiver. However,
when the filter is not chip synchronous with the desired signal, the value of Â¿i falls
in the region (0,1), the MMSE of the filter is sensitive to the propagation delay of
the desired user. Therefore, some method is required for code acquisition and code
tracking even when 2N chip samples are used in the MMSE receiver structure.
As a final example, we will consider the effect of asynchronous operation (with
respect to the desired user) on the resulting probability of bit error. This example
will show that code synchronization is required for the MMSE receiver if one is to
maximize the system capacity. The following briefly explains the outline of the analyÂ¬
sis used to derive the probability of bit error. Since we are interested in bit errors, we
assume that the filter is at steadystate, and that the desired user is transmitting a
random data sequence. We then model the output of the MMSE receiver, w^tr(m),
as a complex Gaussian random variable, and derive an expression for the probability
of bit error. The details of this derivation are given in Appendix A. Modeling the
output of the MMSE receiver as a Gaussian random variable has been recently jusÂ¬
tified by Poor and Verdii [35], who have compared the bit error rate of the MMSE
28
Propagation Delay of User #1 in Chips
Figure 4.2: Minimum meansquared error as a function of the propagation delay of
user #1.
receiver versus the analytical bit error rate of the MMSE receiver when the receiver
output is modeled as a Gaussian random variable. It was found that the Gaussian
approximation was quite good for various number of users and various levels of multiÂ¬
access interference. In addition, several others have found that this approximation
provides reasonable results [13,36].
Once the DSCDMA environment is configured, # users, delays, phases and power
levels, we can numerically solve for Jmin. We then use equations (A.9) and (A.10)
to find the probability of bit error. By varying the DSCDMA environment, we may
observe the effects of asynchronous operation on the system capacity. For purposes
of this example, we set the SNR to 7 dB and let all interfering users be received at
a power level 10 dB above the desired user. A set of spreading codes for 33 users
were selected at random from a set of Gold codes [5] using N â€” 31 chips/bit. A set
of 33 propagation delays were chosen from a uniform distribution over the interval
[0,31). A set of 33 carrier phaseshifts were selected from a uniform distribution over
29
the interval [0, 2n). If we wanted to analyze a Kuser system, the first K elements
of each of the spreading code, delay, and phaseshift sets would be used to create the
DSCDMA environment.
For the case when the receiver was onehalf chip out of synch with the desired
user, using the Gaussian approximation we found that 4 system users would produce
a 2% biterror rate at the receiver. A total of 9 system users would produce a 5%
errorrate. If the receiver was synchronous with the desired user, a total of 14 users
would produce a 2% biterror rate. Likewise, a total of 24 users would produce a 5%
biterror rate. This shows that a significant improvement in the system capacity can
be achieved if the receiver is synchronous with the desired user. While these results
are only valid for case we examined, they are sufficient to demonstrate the costs of
operating the receiver asynchronous to the desired user.
4.3 Adaptive Filters and the Estimation Algorithm
The complexity of finding the optimal weights, given by the WienerHopf equation,
is 0(N3). However the receiver requires knowledge of the autocorrelation matrix R
and the steering vector p. Typically, these parameters are estimated by the receiver
based on observations of the received signal. In practice, the weights of the filter
are chosen adaptively according to some implementation which usually takes the
form of either the least mean square (LMS) or the recursive leastsquares (RLS)
algorithm. The LMS algorithm has low complexity, 0(2N), does not require any
matrix inversion operations, and is easy to implement. The RLS algorithm is more
complex, 0(N2), and performs an iterative matrix inversion by using the Matrix
Inversion Lemma [31]. Since the RLS algorithm is more complex, it is not surprising
that it typically converges faster than the LMS algorithm. While these algorithms
will not produce the same weight vector given by the WienerHopf equation, as the
filter approaches steadystate, they can provide reasonable approximations to the
optimal weight vector.
30
Our basic idea is to derive a code timing estimation algorithm for a singleuser
system, and then study the performance of the estimator in a multiuser environÂ¬
ment. Since the receiver is initially asynchronous to the desired user, r(m) =
ai(Pii Â¿1) exP (j01) + Nm. In order to derive a code estimation algorithm, we would
like to have an exact representation of the statistics of the filter weight vector as
the filter adapts. This appears to be intractable. Instead, we model the filter coefÂ¬
ficients as independent jointly complex Gaussian random variables with a mean of
/?ai(pi,
choose this expression for the mean since for this scenario the optimal weight vector
is proportional to ai(plt Â¿i) exp (j8i). Clearly, this is not an exact description of the
statistics of w(m), and our resulting timing estimator will not produce an optimal esÂ¬
timate of the desired userâ€™s propagation delay. The statistics of the weight vector have
been observed (through simulations) as a function of time when an LMS algorithm
was used, and the LMS stepsize was chosen such that the filter was convergent in the
meansquare. We observed that the components of the filter weight vector become
highly uncorrelated as the filter approaches steadystate. Therefore this assumption
on the statistics of the filter weights seems to be a reasonable approximation and
hence can be used to derive a meaningful timing estimator.
Based on our assumed statistical model for the filter weight vector, we find that
we should choose as our estimates the set {#,/?, p, <5} which minimize
g(P,6,p,Ã³) = w(m) â€” (3&i(p, Ã“) exp (jd)\\2 (4.13)
with p e {0,1,..., N 1} and 0 < Ã³ < 1. Note that even though our assumed
mean is based on optimal values, we will use this function to make estimates after
every receiver bit interval. This approach is taken because it can be shown that
when the LMS algorithm is used, the multiuser transient mean filter weight vector
E[w(m)] is highly correlated with the single user optimal weight vector when a small
31
adaptation stepsize is used. Due to the constraint on the value of p, g((3,9, p, 6) is not
differentiable with respect to p. To minimize g((3,9,p,6), we must find the solutions
to dg((3,9,p,6)/dx = 0 (for x = (3,9,6), for all possible values of p. Therefore, each
value of p produces a corresponding set of candidates for the subset {9,(3, Ã¡}. Our
estimation algorithm iterates through all possible values of p and produces a list of
all of the corresponding candidates. By choosing the set {9,(3,p, <5} out of this list
which minimizes g((3,9,p,S), we find the global minimum of g((3,9,p,6).
The solutions to dg((3,9,p,S)/dx = 0 for x = 9,(3 are given by
0i(p, 6) = arg (af (p, S)w(m))
P(pJ)
af(p,Â¿)w(m)
llai(p,Â¿)2
(4.14a)
(4.14b)
Substitution of these expressions into the previous equation produces the following
two expressions.
g0{p,6),9(p,6),p,6) = w(ra)2  f{p,6)
_ af(p,Â¿)w(m)2
[ llai(p, <5)2
(4.15a)
(4.15b)
These equations show that minimizing g((3(p,6),9(p,6),p,6) corresponds to finding
the set {p, $} which maximize the cost function f(p,6).
Since the estimate of p\ must be an integer, the cost function f(p,6) is a piece
wise continuous function. Maximizing f(p, 6) with respect to 6 produces a quadratic
equation whose coefficients depend on the value of p. Therefore, for each possible
value of p we must find the roots of the quadratic equation D2(p)p2 + Di(p)p +D0(p),
where the coefficients are given by the following set of equations.
D0(p) = NAi(p) + 2CA0(p)
D\(p) = 2NA2(p)  4CA0(p)
D2(p) = 2C(A2(p) + A1(p))
(4.16a)
(4.16b)
(4.16c)
C â€” N  cfc^
,(p)i2
Mp) = wH(m)cjP
Ai(p) = 2Re (ciP))Tw(m)wH(m)clp+1) 2AQ{p)
A2(p) = wH(m)ciP+1)2  Ax(p)  Ao(p)
32
(4.16d)
(4.16e)
(4.16f)
(4.16g)
By definition, the value of Â¿ must satisfy the condition 0 < S < 1, so we must ignore
any solutions from the previous equation which fall outside of this region. In order
to clarify how we form our estimate of the propagation delay for the desired user, we
present the following algorithm.
Let T represent the set of costs corresponding to the set of candidate estimates
of (p, 6), denoted by U. We can find the maximum of f(p,S) as follows:
Step 1. Let T = {/(0,0), /(1,0), /(2,0),..., f(N  1,0)}.
Step 2. Let U = {0,1,2,..., N â€” 1}.
Step 3. For p = 0,1,..., N â€” 1, do the following:
a) Compute the coefficients _D2(p), Â£fi(p), D0(p).
b) Solve for the two roots of the quadratic equation, pi and p2.
c) If 0 < pi < 1 for i Â£ {1,2}, add the cost /(p, pÂ¿) to the set T, and add
(p + pi) to the set U.
Step 4. Let Uk denote the kth element of the set U. Then f = Uk, where k = max 7}.
i
Step 5. The estimates of pi and Â¿i are given by pi(ra) = }fj and 5i(m) = f â€” [Fj,
where rounds x to the nearest integer towards zero.
We plot the cost function, /(p, 5), for two different scenarios in Figures 4.3 and 4.4,
under the assumption that the receiver is operating using the optimal filter weights.
Figure 4.3 shows the cost function for a singleuser system when iq = 15.25TC and
Eb/Aio was 7 dB. This plot shows that the timing estimate will be fi = 15.25TC, which
is a perfect estimate. Since we have developed the timing estimator for a singleuser
system this result is not surprising. Of course, when the filter weights are from the
33
Figure 4.3: Plot of the cost function f(p, 6) for a singleuser system.
LMS or RLS algorithms we expect some noise in this estimate due to the noise in the
filter weights. Also, the rate of convergence of the adaptive filter to its steadystate
value will determine how long the filter has to adapt before a reliable estimate of T\
can be made. This issue will be addressed in the next chapter.
Figure 4.4 shows the cost function, f(p, 6), for a more severe environment. In this
figure 14 interfering users have been added to the system, each received at a power
level 10 dB above the desired user. Since the maximum value of f(p,6) does not
occur near 15.25TC, the timing estimate will be wrong. In this example f\ = 7.43TC.
We have intentionally heavilyloaded the DSCDMA system, to demonstrate what
will happen to the cost function. In general, we can say that at some level of MAI,
the estimation algorithm completely breaks down. We will study the effects of the
level of the MAI on the performance of the timing estimator in the next chapter.
When the simplified system model presented in Section 2.2 is used, the timing
estimator takes on a slightly different form. Recall that the simplified model assumes
that all users are received phasesynchronous, and the weights of the adaptive filter
34
Figure 4.4: Plot of the cost function f(p, 6) for a fifteenuser system.
are purely real. Using this assumption the autocorrelation matrix and the steering
vector for the adaptive filter are written as
R = E [r(m)rH(m)] = ai(pi,ii)af(pi,Â£i) +
K Pk (417)
+ TET [a2ii(Pfc^fc)aLi(Pfc>4) + a2*(pjk,5fc)ajb(pfc,5fc)]
2 Pi
k=2 1
P = (pi.il)
(4.18)
where I is an N x N Identity matrix.
Having the weights of the filter real leads to a small reduction in the complexity
of forming a timing estimate for the desired user. The derivation of the algorithm is
similar to that previously presented. Maximizing the cost function, f(p, S), produces
a quadratic equation whose roots may be written in closed form. However, one of the
roots produces a trivial result and therefore can be ignored. The other root is given
35
by the following equation.
= NwT(m)c(f+1)  cfc(11)wr(m)c(1p) ig
^ (N â€” cfci1))(wT(m)c(1p+1) + wr(m)CiP))
Based on the above equation, the timing estimation algorithm is then modified as
follows:
Step 1. Perform steps 1 and 2 of the previous algorithm.
Step 2. For p = 0,1,..., TV â€” 1, do the following:
a) Use equation (4.19) to compute the coefficient pi.
b) If 0 < pi < 1, add the cost f(p, pi) to the set T, and add (p + p\) to the
set U.
Step 3. Perform steps 4 and 5 of the previous algorithm.
4.4 Summary
The main contribution of this chapter is the presentation of a timing estimator for
a singleuser, based on processing the weights of an adaptive filter. We motivate why
these weights may be used to form a timing estimate by studying how the optimal
weights depend on the DSCDMA environment for two simple cases. The MMSE
receiver has received considerable attention in the literature. One fact that seems to
be ignored by several authors is that there are performance penalties to be paid for
operating the receiver asynchronous to the desired user [34]. Therefore, one may come
to the wrong conclusion that no advantage is offered by synchronizing the receiver
with the desired user. We present three examples as to why operating the MMSE
receiver synchronous to the desired user is important. Certainly, when the system
capacity is considered we must synchronize the receiver with the desired user, if we
want to support as many users as possible in the DSCDMA system.
One may argue that the form of the timing estimator is complex since it depends
on, N, the number of chips/bit. However for large N, even using the traditional
correlator for timing estimation becomes complex. In fact, the correlator uses the
36
exact same estimation algorithm that we have presented in this chapter. The differÂ¬
ence between our technique and the correlator, is that we form an estimate based on
observing the filter weights while the correlator forms an estimate by observing samÂ¬
ples of the received signal. The additional complexity added to the receiver by our
technique is that of running the adaptive filter (LMS 0(2N), RLS 0(N2)). Since the
same adaptivefilter receiver structure can be used for detection, and we have demonÂ¬
strated that synchronization is desired, including the timing estimator in the receiver
should be an acceptable increase in the overhead of implementing the receiver.
CHAPTER 5
CHARACTERIZING THE PERFORMANCE OF THE
TIMING ESTIMATOR
In this chapter, several methods will be presented in an attempt to characterize the
performance of the adaptive filterbased timing estimator in an AWGN channel. We
are interested in developing analytical tools that can be used to obtain performance
measures on the quality of the timing estimate in hopes of avoiding long Monte
Carlo simulations to get such performance metrics. We first consider a bound on the
probability of correct acquisition when we are only interested in correctly estimating
the chip level timing. Next using traditional techniques, an approximation to the
lower bound on the conditional variance on the timing estimator will be derived. This
expression will be developed by deriving the CramÃ©rRao bound (CRB) for a fictitious
timing estimator, once several assumptions are made with respect to the statistical
properties of the LMS filter weights. In order to test the quality of these analytical
expressions, they will be compared to simulation results under several different multiÂ¬
user scenarios. Finally, the coarse acquisition performance of the timing estimator
will be observed through MonteCarlo simulations. The acquisition performance of
the timing estimator will be observed for both the LMS and RLS based adaptive filter
architectures. The effects of multiaccess interference on the timing estimator will be
observed for several different scenarios.
5.1 Transient Statistics of the LMS Filter Weights
In order to derive analytical performance metrics on the timing estimator, we
first require knowledge of the statistics of the filter weights as a function of time.
When we developed the timing estimator, we assumed that the filter weights were
37
38
independent and jointly Gaussian random variables. Using these assumptions, one
could then completely define the assumed statistics of the filter weights by defining
the filter weight mean vector and the covariance matrix for the filter weights. Since
the filter is adaptive, these quantities will be dependent on the initial conditions of the
filter as well as the length of time that the filter has adapted (and of course somehow
depend on the statistics of the input signal). The next few paragraphs describe how
the statistics of the the LMS filter, using the system model presented in Section 2.2,
may be derived using several assumptions about the filter and its input signal.
One area that has received a fair amount of attention in the literature is the
tracking performance of the LMS filter [31,3742], The main subject that has been
addressed in these works, is what bounds must be placed on the LMS stepsize, //, such
that the filter has desirable convergence properties. One of the properties of interest,
is the convergence of the filter weights about the WienerHopf solution. The filter
is said to be convergent in the mean if lim^oo E[w(n)] = wopt, the solution to the
WienerHopf equation. As an extension of this idea, even if the filter is convergent in
the mean, it would be very desirable to have a finite variance of the filter weights about
wopt. The variance of the filter weights about the WienerHopf solution determines
the level of the meansquared error at the filter output. As one would expect, the
bound on Â¡i required to minimize the meansquared error of the filter output is less
than the value of /r required for convergence in the mean. An intrinsic part of the
analysis on the limits of /j for such convergence properties, and what we are really
interested in, is a set of equations that describe the characteristics of the filter weights
as a function of adaptation time.
Senne [37] presented a set of iterative equations to describe the timedependent
filter weight mean vector and the filter weight correlation matrix for an LMS filter
when the filter input was a vector of zeromean Gaussian random variables. However,
in the system model used in Section 2.2, due to the desired userâ€™s contribution to the
39
receiver input, the filter input signal is not zeromean. We can apply the model used
by Senne to derive a similar set of equations to describe the transient statistics of
the filter weight vector. In order to begin the derivation, we make several assumpÂ¬
tion regarding the relationships between the filter weight vector and the input signal.
These assumptions are referred to as the fundamental assumption [37] or indepenÂ¬
dence assumption on the filter statistics and are included here for convenience. The
fundamental assumption consists of three parts which are:
1) Each receiver input vector r(m) is statistically independent of all previous input
vectors and all previous desired filter outputs di(ra).
2) Each desired output d\(m) is dependent on the corresponding input vector r(m)
but is statistically independent of all previous desired outputs and received vectors.
3) All desired quantities and received vectors are mutually Gaussiandistributed ranÂ¬
dom variables.
In many applications of the LMS filter, the filter input vector is sampled at the
same rate as the output of the filter. Therefore, the first part of the fundamental
assumption is clearly not valid in this situation since the elements of r(m â€” 1) are
shifted by one position in creating r(m). This weakness in the fundamental assumpÂ¬
tion under these conditions is what has driven the study of the convergence properties
of the LMS filter. Of noticeable interest is the work by Douglas [42], in which a nuÂ¬
merical algorithm is presented to observe the exact transient statistics of the MSE
convergence of the filter. However, this algorithm is very complex in that a set of
linear equations must be dynamically created and solved based on the number of taps
in the LMS filter. As an example of the complexity of this algorithm, for a fivetap
filter, a set of threethousand equations were required to solve for the transient MSE
of the filter. Based on this example, applying to this method to a filter with more
taps does not seem reasonable.
40
For the adaptive receiver, we have assumed that the input to the filter is sampled
at a rate N times faster than the filter output. This means that r(m) could truly be
statistically independent from all previous receiver inputs. Certainly the contribution
to r(m) from the AWGN is independent from all previous receiver inputs. However,
two symbol intervals of each asynchronous user contribute to r(m). So the first part of
the fundamental assumption, while not being true, will still be asserted and we use the
fundamental assumption to derive the desired equations. We begin the derivation of
the filter transient weight mean vector by manipulating the general LMS filter update
equation given below
w(m + 1) = w(m) â€” /uu(m)ur(m)w(m) + /xu(m)d(m) (5.1)
where u(m) is the filter input signal. Using the system model defined in Section 2.2,
u(m) = r(m) defined in equation (2.14), with the caret dropped for notational simÂ¬
plicity in future equations.
We note that the desired filter output is not timedependent, and will therefore
drop the dependency on m for this quantity. Taking the expectation of both sides of
equation (5.1) will produce the desired filter transient weight mean vector
E[w(m + 1)] = w(m + 1)
= {I â€” //Rrr} w(m) + Â¡lid
(5.2)
where I is an N x N identity matrix, and Rrr = E[r(m)rT(m)] is used for notational
convenience. Once we define the initial conditions on the filter w(0), realize that
r = ai(pi, Â¿i) and let d = 1, we can use the previous equation recursively to describe
the filter transient weight mean vector.
In order to develop an iterative equation for the filter transient weight covariance
matrix, equation (5.1) will be used to find an iterative equation for the filter transient
weight autocorrelation matrix defined as Kww(m) = E[w(m)wr(m)]. This derivation
41
is described in Appendix B, but the final result is listed below.
Rww{m + 1) = Rww{m) + /xw(m)rTd  /j,Rww(m)Rrr
+ /.Â¿fwT(m)d â€” iiRrrRww(m) + fi2Rrrd2
â€” 2fj?a2 {wr(m)rl + rwr(m) + w(m)rr} d
â€” 2/i2fwT(m)ffrd + 2/i2 cr2Rww(m)rrT (53)
+ 2^i2a4Rww(m) + Â¿Ã2
+ /i2cr2r'rRÃÃ,it,(m)fI + fÃo2 trace(Ru,u,(m))rrr
+ 2n2a2frTRww(m) + ^2rfTRJl,â€ž,(m)ffT
Given the above equation, we can easily define the transient filter weight covariance
matrix using the following equation once we define w(0), Ru,tl,(0), r and Rrr.
Cww(m) = Rww(m)  w(m)wT(m) (5.4)
As an alternative to equation (5.3), we also consider using the transient weight
error covariance matrix which describes the deviation of the filter weights about the
WienerHopf values as a function of time [31]. This is the same equation used by
Miller [32,33] to study the transient MSE response of the LMS filter. The transient
weight error covariance matrix is defined as
K(m) = E (w(m)  wopt)(w(m)  vv^
(5.5)
and can be evaluated at the mth interval using the following iterative equation.
K(m + 1) = K(m) + Â¿i[R,.,.K(m) + K(m)R,.r] + ^2Rrr trace(RrrK(m))
+ p2RrrK(m)Rrr + n2JminRrr
(5.6)
In order to use equation (5.6) we note that K(0) = w0ptwjpt, when w(0) = 0, and
Jmin = 1  Woptai(pi,Â¿i). The transient weight covariance matrix can be found as
shown in equation (5.7).
42
Cww(m) = E (w(m)  w(m))(w(m)  w(m))
= K(m)  w(m)wr(m)  woptwJpt + w(m)w^pt + woptwT(m)
(5.7)
Several different DSCDMA environments were simulated, and the transient mean
weight vector and transient autocorrelation matrix of the LMS filter weights were
observed at the mth interval of adaption using the following equations.
1 M
w(m) = ^J^w(m)
i=l
M
= Â¿Â¿w(m)wr(m)
1 t=i
(5.8)
(5.9)
These quantities were then used to form an estimate of the transient autocovariance
matrix of the LMS filter weights as
A T ,
Cww(m)  Rww(m)  w(m)w (m).
(5.10)
Of course, for each trial of a given test, the number of users, the spreading sequences
and propagation delays are fixed. The data sequences for the interfering users and
the AWGN were random during each trial, and independent from trial to trial. The
LMS stepsize, n, was chosen such that the filter was convergent in the meansquare.
By comparing the results produced by the above equations to the corresponding
analytically derived quantities, we can in some sense measure the quality of the
analytical expressions.
For the first comparison, a singleuser environment was simulated using 31 chips/bit
and a SNR of 10 dB, and T\ = 15.5TC. A set of 3,000 independent trials was used
in computing equations (5.8) through (5.10). Figure 5.1 compares the E[w(m)]
produced by equation (5.2) to the observed simulation results. In order to compare
these two quantities directly, the error is plotted as E[w(m)]  W(m) in the second
part of the figure.
43
Figure 5.1: Performance of equation (5.2) for a singleuser system, a) The analytical
mean filter weight vector norm, E[w(m)] (*), and the simulated mean
filter weight vector norm W(m) (solid line) plotted as a function of the
training length; b) Comparing the error between the two mean weight
vectors, E[w(m)] â€” w(m).
As a second test, a threeuser system was simulated. As before the desired user
was received at a SNR of 10 dB and the two interfering users were received at a
power level 10 dB above the desired user. The interfering users had propagation
delays of t? = 6.257Tc and 73 = 28.0125TC. This environment was simulated using
3,000 independent trials. Figure 5.2 compares equation (5.2) to the simulation results.
Figures 5.1 and 5.2 show that the analytical expression for the transient mean weight
vector of the LMS filter given by equation (5.2) closely tracks the corresponding
simulation results. Several other multiuser scenarios with various levels of SNR and
MAI were simulated, and the performance of equation (5.2) was similar to those
shown in Figures 5.1 and 5.2. Therefore, we use equation (5.2) in the sequel when we
require use of an expression for the LMS filterâ€™s transient weight mean vector.
44
Figure 5.2: Performance of equation (5.2) for a threeuser system, a) The analytical
mean filter weight vector norm, E[w(m)] (*), and the simulated mean
filter weight vector norm W(m) (solid line) plotted as a function of the
training length; b) Comparing the error between the two mean weight
vectors, E[w(m)]  w(m).
A similar test was used to characterize the performance of equations (5.4) and (5.7),
which are the analytical expressions for the LMS filterâ€™s transient weight autocovariÂ¬
ance matrix. We observed that for a singleuser environment, both equations gave
similar results. However when multipleusers were present, equation (5.4) initially
tracks the simulation results for Cww(m), but at some point diverges from the simuÂ¬
lation results. However, equation (5.7) was observed to produce similar results, when
compared to Cww(m), regardless of the number of users and the level of the MAI. A
twouser system with the desired user received at a SNR of 10 dB and the interfering
user received at a power level 10 dB above the desired user was simulated for 3,000
trials. The results of comparing the Frobenius norm of equations (5.4) and (5.7) to
the Frobenius norm of the simulated Cww(m) matrix are shown in Figure 5.3. Note
that similar results were observed for other multiuser scenarios. We have observed
45
Figure 5.3: Performance of equations (5.4) and (5.7) for a twouser system. The
Frobenius norm of the LMS filterâ€™s transient weight autocovariance maÂ¬
trix; equation (5.4) (*), equation (5.7) (x) and simulation results (+).
that the Frobenius norm of the difference between equation (5.7) and the simulated
Cww(m) is on the order of 1020% over several different multiuser scenarios.
We have derived a set of analytical expressions, equations (5.2) and (5.7), that
we will now use to describe the transient model of the LMS filter weights under the
assumption that the weights are jointly Gaussian. Since we are using the weights of
the filter to form a timing estimate of the desired userâ€™s propagation delay, we are
interested in how the transient response of the filter weights determines the quality of
the resulting timing estimate. In the next two sections, we use our analytical model
for the transient weight statistics, in an attempt to characterize the relationship
between the filter weights and the quality of the timing estimate.
5.2 Chip Selection Error Probability
We are interested in developing an analytical expression for the chip selection
error probability when the fractional part of the propagation delay is ignored. That
46
is, we want to find an expression for Pr(pi Â± pi) when the value of <5i is ignored. In
this case, the propagation delay estimate at the zth iteration of the adaptive filter is
formed as shown below.
Pi = argmaxF(p) p â‚¬ (0,1,2,..., N  1)
v
F(p) = wI'(Â«)c?'l2
(5.11)
(5.12)
Therefore, the chip selection error probability can be expressed using a union bound
[43] as
Nl
Prtfi i Pi) = V PrlF(pi)  F(k) < 0).
(5.13)
kâ€”0
fc^pi
Note that each term in the previous equation can be expressed as
F(pi) â€” F(k) = w T(i) ^c[Pl^(ciPl)) â€” j w(?)
= WT(t)A,w(!).
(5.14)
Using our assumption that the filter weights are jointly Gaussian, the expression in
equation (5.14) is just a quadratic form of Gaussian random variables. This type
of expression is common in radar applications, weapon control systems and is also
intrinsic to any discussion of variances of random variables [44]. So we now must
concern ourselves with how to evaluate Pr (wr(i)A;,w(z) < 0). This type of evaluaÂ¬
tion has received considerable attention by many researchers [4550]. The next few
paragraphs will summarize various methods that may be used to handle this problem.
As shown in Appendix C, the distribution of wr(z)Ajtw(z) can be shown to have
the same distribution as
U = ^AÂ¿(VUÂ¿6Â¿)2 (5.15)
Â¿=i
where the are independent Gaussian random variables of zero mean and unit
variance and the terms and 6Â¿ depend on the transformation matrix Ak as well the
47
covariance matrix of the filter weights. Deriving the distribution of equation (5.15)
directly does not appear to be mathematically tractable. However, deriving the
corresponding characteristic function is relatively simple. For a random variable that
has a probability density function of fx(x), the characteristic function is given by
+ 00
$x(w) = E[exp (Jux)] = j fx{x) exp (jux) dx. (5.16)
â€” OO
Likewise, if one is given the characteristic function $x(w) then the corresponding
probability density function is found using the inversion formula stated below.
+oo
fx(x) = J $x{uj)exp(jux)duj
â€” OO
The characteristic function for equation (5.15) is found to be
(5.17)
1
(5.18)
v/1  2ju\i
Johnson and Kotz [45] provide an extensive summary of techniques that can be
used to find fy{y) or the Pr(y < t), where t is some given threshold, when the transforÂ¬
mation matrix Ak is positive definite. For specific restrictions on the and in the
characteristic function, several expansions of y(u;) into power series representations
are presented. Then, the inversion formula in equation (5.17) is applied to each term
in the series to produce fy(y). A matrix A is positive definite when xrAx > 0 for
any nonzero vector x. Therefore, if the transformation matrix Ak is positive definite
then a chipselection error can never occur and, therefore, Pr(K < 0) = 0. We must
determine if the transformation matrix is positive definite. A sufficient condition for
positive definiteness, is that all of the eigenvalues of the matrix Ak are positive [51].
Without actually computing these eigenvalues, we can test for the possible existence
of negative eigenvalues. It is known that the sum of the eigenvalues is equal to the
trace of the matrix. Since all of the diagonal elements of Ak are zero, there must
48
exist negative as well as positive eigenvalues and therefore the transformation matrix
is not positive definite.
Rice [46] provides several numerical techniques that can be used to evaluate
Pr(T < t) through applications of equations (5.18) and (5.17) for a general transÂ¬
formation matrix. However, it is noted that these techniques can exhibit slow conÂ¬
vergence due to the oscillatory nature of the exponent in equation (5.18). In order
to improve the convergence rate of the numerical integration techniques, a change of
variables that produces a tilting of the integration path is presented. However, the
change of variables is inversely dependent on the desired threshold t, and is therefore
not valid when t â€” 0, which is the case at hand.
A quadratic form of independent Gaussian random variables, such as shown in
equation (5.15), can also be shown to have the same distribution as a linear sum of
independent noncentral x2 random variables [44]. That is, equation (5.15) can also
be represented by
r
(5.19)
where Xj are independent random variables, having noncentral \2 distribution with
rij degrees of freedom and noncentrality parameter S? for j = 1,..., r and X0 having
a zeromean unitvariance Gaussian distribution. Davies [50], based on inverting
the corresponding characteristic function [49], presents a numerical technique that
can be used to find the distribution of a linear sum of independent noncentral x2
random variables, when the original quadratic form has a general (not necessarily
positive definite) transformation matrix. The technique involves evaluating a series
of exponential terms, where the truncation error is dependent on the number of
terms included in the series. The number of terms required for the integration is
determined approximately by the total number of degrees of freedom and the sum
of the noncentrality parameters as well at the value of t, at which the distribution
49
function is to be evaluated. Davies presents a table that list the number of terms
required for integration as a function of these parameters. This table shows that as t
approaches zero, the number of terms required for integration quickly increases from
several hundred to tens of thousands.
Therefore, the techniques of Rice or Davies appear to be applicable to a situation
where the Pr(Y < 0) has to be evaluated very infrequently. Recall that at the
receiverâ€™s ith interval, we have to evaluate Pr(Y < 0) a total of (Nâ€”l) times to form a
bound on the chipselection error probability. If we want to evaluate the chipselection
probability as a function of the LMS filterâ€™s adaptation time, these techniques become
numerically prohibitive. That is, we found that MonteCarlo simulations of a specific
DSCDMA environment produce results for the chipselection error probability in less
time than is required to evaluate the same quantity using one of Riceâ€™s techniques for
the same environment.
As a means of evaluating Pr(F(pi) â€” F(k) < 0), we instead turn to the moment
generating function. Instead of trying to evaluate the desired expression explicitly,
we use a Chernoff bound to find an upper bound on Pr(F(px) â€” F(k) < 0) for each
term in equation (5.13) (see Appendix C for details). We then sum these (N  1)
Chernoff bounds to find an upper bound on Pr(pi ^ px). An example of the resulting
bound using this method is shown in Figure 5.4. A threeuser DSCDMA environment
was simulated with a SNR of 10 dB, and both interfering users were received at a
power level 10 dB above the desired user. The propagation delays for the users were
T\ â€” 15Tc, t2 = 6TC and r3 = 28Tc. The LMS stepsize, /x, was set to 0.1/(the total
input power) such that the filter was convergent in the meansquare.
There are three curves present in Figure 5.4. The first curve is the simulated
results for the incorrect chipselection probability, which is plotted using a plus sign
(+). The other two curves are the resulting upper bound for this probability, for
two different applications of the Chernoff/Union bound technique. The solid line,
50
represents the upper bound when equations (5.2) and (5.7) are used to describe
the transient mean vector and autocovariance matrix of the LMS filter weights as a
function of the training interval. Note that the resulting upper bound is not very
tight. It seems appropriate to wonder if the looseness of the bound is due to the
Chernoff/Union bound technique, or due to errors present in the model given by
equations (5.2) and (5.7). As a control for this test, the third curve (*) represents the
upper bound found by using the transient mean vector and autocovariance matrix
of the LMS filter weights obtained directly from the simulation. This is slightly
closer to the simulation results, but the upper bound is still very loose. Therefore,
is seems that the application of the Chernoff/Union bounds are the main reason
that the derived upper bound is very loose when compared to the actual simulation
results. Note that this behavior was observed for several other multiuser DSCDMA
scenarios. Unfortunately, based on this result this technique is not very useful in
terms of characterizing the performance of the timing estimator.
5.3 Approximation for the Conditional Variance of the Timing Estimate
A traditional technique that is used to characterize the performance of an estiÂ¬
mator is to compare the variance of the estimate to the CramÃ©rRao bound (CRB).
The CRB is a lower bound on the variance of any unbiased estimator for a given log
likelihood function [52,53]. In this section, we will use the traditional CRB technique
along with several assumptions, in hopes of forming an approximation to the perforÂ¬
mance of our timing estimator. Since we are interested in characterizing the transient
performance of our timing estimator, we base our technique on the timedependent
statistical model for the LMS filter weights given in equations (5.2) and (5.7).
Since we have assumed that the filter weights are jointly Gaussian, equations (5.2)
and (5.7) are sufficient to completely define the timedependent statistical model for
the LMS filter weight vector. We do not intend to use this model to derive a new
timing estimation algorithm. However, we use this model to derive a CRB, hoping
51
Figure 5.4: Comparing incorrect chipselection probability to the Unionbound for a
threeuser system. Simulation results (+), the Chernoff bound using equaÂ¬
tions (5.2) and (5.7) (solid line), and the Chernoff bound using simulated
weight vector statistics (*).
that the result can be used to characterize the conditional variance of our estimate
for Â¿i. Using equations (5.2) and (5.7), the timedependent loglikelihood function at
the mth iteration (ignoring constants) is
1
S(wÂ¿i,ra) =  [(w(m)  w(m))TC^(m)(w(m)  w(m))] .
(5.20)
The CRB for the above loglikelihood function is then given by the following equation
[52].
Var(Ã¡x â€” Â¿i) > < â€” E
dÃ³Â¡
}
i
(5.21)
In order to evaluate the above expression, we require knowledge of how Cww(m)
depends on Â¿i, in order to evaluate the required partial derivatives. This does not
appear to be mathematically tractable since, as seen in equation (5.7), we need to
evaluate E[dw(m)/dÂ¿iwr(m)]. In order to continue with this approach, we assume
52
that dCww(m)/dÃ³x = 0. Using this assumption, after straightforward algebraic maÂ¬
nipulations, our approximation to the variance of the <5i estimate is given below.
i
(5.22)
Several auxiliary equations and initial conditions are required in order to evaluate
equation (5.22). The required auxiliary equations are
(5.23)
OT _ r(pd i r(pi+:)
d5[~~c' +Cl
(5.24)
(5.25)
We hasten to stress that we have used the mechanics of the CRB technique to derive
equation (5.22). However, due to the various approximations used along the way our
result is not a lower bound, and is only intended to provide a rough approximation
to the performance of our estimator.
Several different DSCDMA environments were simulated in order to verify the
usefulness of equation (5.22) as an approximation to the conditional variance of the
Â¿i estimate. For the following results, the usersâ€™ spreading codes were selected from
a set of Gold codes [5] with 31 chips/bit and the desired user was received at a SNR
of 7 dB. The propagation delay for each user was chosen from a uniform distribution
over the interval [0,7),). Once the spreading codes and delays were selected, they
were considered fixed and the DSCDMA environment was simulated for 2,000 trials.
Therefore, the variables that were considered random between the individual simuÂ¬
lation trials were the AWGN and the data sequences for the interfering users. Note
that the LMS stepsize, Â¿Ã, was always chosen to be 0.1/(the total input power) such
that the filter was convergent in the meansquare.
53
Figure 5.5: Comparing the simulated conditional variance of the Â¿1 estimate to the
analytically derived approximation in absence of nearfar MAI.
Figure 5.5 shows the comparison between equation (5.22) (dashed line) and the
standard deviation of the Â¿1 about Â¿1, conditioned on = p\ (solid line) for several
different scenarios. In this figure, all interfering users are received at the same power
level as the desired user. The results shown in this figure look very promising. The
largest error between the simulation results and equation (5.22), which occurs for
10 users, is 6%. Unfortunately, the usefulness of equation (5.22) quickly disappears
as more users are added to the system, or the level of Pk/P\ is increased (or some
combination of these two conditions). As an example of this statement, Figure 5.6
shows how the approximation fails for several multiuser scenarios where all interfering
users are received at some level above the desired user. Upon review of the derivation
of the approximation, the weakness of the approximation appears to stem from the
assumptions that were made to facilitate the derivation, and not from equations (5.2)
and (5.7) as these have been observed to agree very well with simulation results.
54
10
o
a: 3 Users, N/F= 15 dB
b: 5 Users, N/F = 20 dB
c: 7 Users, N/F= 15dB
a
b
10
2
0
100 200 300 400 500
# of Training Bits
Figure 5.6: Comparing the simulated conditional variance of the Â¿i estimate to the
analytically derived approximation in presence of nearfar MAI.
5.4 Coarse Acquisition Performance
In this section, we present numerical results for the acquisition performance of the
timing estimator. We compare the performance of our timing estimator with a well
known conventional technique. One approach to timing estimation in a singleuser
spread spectrum system is the correlator. The received signal is correlated with time
delayed versions of the known spreading code, and the timing estimate is given by the
amount of time delay that maximizes the correlation. The receiver forms its estimate
as shown below:
(5.26)
where M is the number of r(m) samples observed before computing the timing esÂ¬
timate. As shown by the above equation, this technique also requires an all ones
training sequence. It is also known to be optimal for a singleuser in the presence of
AWGN only, but can be suboptimal in the presence of multiaccess interference.
55
Our timing estimator will be computed by processing the weights of either an
LMS filter or RLS filter. For the LMS adaptation results, the LMS stepsize n was
chosen to be 0.1/trace(R), where trace(R) is the total filter input power. With
this stepsize, the LMS filter was convergent in the mean and also convergent in the
meansquare. As the metric for measuring the performance of the timing estimator,
we observed the average acquisition time of the timing estimator. The acquisition
time is defined as the number of training bits required so that the probability that the
timing estimate is within one halfchip of the true propagation delay of the desired
user is greater the 90 percent. That is, we recorded the smallest value of m such that
Pr (fi  n < TJ2) > 90%. (5.27)
In addition, based on the discussion about the advantages of synchronization, we
were interested in observing the root meansquared estimation error (RMSEE) given
correct acquisition.
RMSEE = ^/^[(fxnfldnnl^T^)] (5.28)
The value of the RMSEE gives a measure of how well the receiver could be synchroÂ¬
nized if the timing estimate was used to update the receiverâ€™s timing relative to the
desired user.
Each DSCDMA environment was simulated for 500 independent trials. For each
trial, the user spreading codes were selected at random from a set of Gold codes using
31 chips/bit. For each trial, the propagation delay for each user was chosen from a
uniform distribution over the interval [0,31), independent from the other users. The
usersâ€™ phaseshifts were chosen from a uniform distribution over the interval [ 0, 27t),
independent from the other users. The desired user was received at a SNR of 7 dB.
Also at the start of each trial, the received power level for each interfering user
was selected from a lognormal distribution that had a mean and standard deviation
56
of 10 dB. That is, Pk/P\ â€” 10i/10 where Â£ is a Gaussian random variable with a mean
of 10 and a variance of 100. The lognormal power distribution was used to simulate
environments where different power levels exist due to shadowing or system power
control error. By choosing the simulation parameters at the start of each trial, and
running many independent trials for each environment, the output of this test is an
estimate of the number of training bits required to achieve coarse acquisition.
Figure 5.7 shows the coarse acquisition performance using the correlator, LMS
filter, and the RLS filter in forming a timing estimate as a function of the number
of users. As shown in this figure, for a few number of users the LMS based timing
estimator offers no advantages over the conventional correlator estimator. However
as the number of users increases, the LMS based timing estimator does perform much
better than the correlator. In addition, the maximum RMSEE for all of the points
on the LMS curve was 0.11TC, which shows that the timing estimator does produce
a reasonable estimate of the desired users propagation delay. But with 15 system
users, on average 240 training bits are required to achieve coarse acquisition. It is well
known that the convergence rate of an LMS filter, which will affect the performance
of the timing estimator, is dependent on the ratio of the largest eigenvalue of R
to the smallest eigenvalue of R. As the level of nearfar interference increases this
eigenvalue ratio quickly increases. This plot shows a weakness in using the LMS
adaptation algorithm, not a weakness in the timing estimator.
The results shown in Figure 5.7 for the RLS filter support the previous statement.
For the same DSCDMA environment, the RLS filter based timing estimate performs
much better than the correlator or the LMS filter. The RLS algorithm requires an
initial positive definite estimate of the data autocorrelation matrix. For this work, the
softconstrained initialization RLS algorithm was used, in which the initial estimate
of the data autocorrelation matrix is just a positive constant times an identity matrix.
The maximum observed RMSEE was 0.17TC, which is only slightly worse than the
57
Figure 5.7: Average training bits required for correct acquisition as a function of the
number of system users in AWGN channel.
results for the LMS based timing estimator. With 15 users (50% system capacity)
the timing estimator can achieve coarse acquisition with 45 training bits, which is
not an unreasonably long training period. For 10 system users, the RLSbased timÂ¬
ing estimator achieves acquisition about 6 times faster than the LMS based timing
estimator. Clearly, the timing estimator based on the RLS filter performs better, in
terms of acquisition time, then the timing estimator based on the LMS filter.
While it is interesting to observe the performance of the timing estimator as a
function of the number of system users, it is equally important to consider the near
far resistance of the timing estimator. In other words, for a fixed number of system
users, we would like to see how the number of training bits required to achieve correct
acquisition depends on the level of the multiaccess interference. For this test, the user
spreading sequences, phases and propagation delays were chosen using the methods
previously described. The desired user was received at a SNR of 7 dB. However,
the received power level P* for each interfering user was selected from a lognormal
58
Figure 5.8: Average training bits for correct acquisition as a function of the level of
multiaccess interference in AWGN channel.
distribution that had a mean and standard deviation of X, where X was allowed
to vary over the range of (0,13) dB. For a given value of K, the number of system
users, and a specific value of A, 500 independent trials were used to find the coarse
acquisition performance of the timing estimator. This scenario was used to simulate
an environment in which there is very loose power control, and the desired user is
received at a power level which on average is one standard deviation below the mean
of the power levels of all the interfering users.
Figure 5.8 shows the coarse acquisition performance of the 3 timing estimation
algorithms for 2 system users and 15 system users as a function of the the value of
X. From this figure it is seen that none of the timing estimators are truly nearfar
resistant. Clearly, the correlator is very sensitive to the amount of nearfar interferÂ¬
ence, As was observed in the previous plot, the LMSbased timing estimator performs
better than the correlator, and the RLSbased timing estimator performs better than
59
the LMSbased timing estimator. For 2 users, the RLS based timing estimation algoÂ¬
rithm could be considered as nearfar resistant. When the value of X was set to 0 dB,
only 5 training bits were required to achieve correct acquisition. When the value of
X was set to 13 dB, a much more severe nearfar environment, only 7 training bits
were required. With 15 system users, the corresponding number of required training
bits changed from 10 to 75, which we consider as being not enough of an increase in
the training period to label the estimation algorithm as not being nearfar resistant.
However, it is clear from this plot that the length of the training period required to
achieve correct acquisition in a system with a severe nearfar problem when either
the correlator or the LMS based timing estimator is used is unacceptable.
5.5 Summary
In this chapter, we have attempted to characterize the performance of the timing
estimator. An equation for the transient weight mean vector of the LMS filter weights
was derived and tested under several DSCDMA environments. Simulation results
have shown that this equation is in close agreement with the true transient mean of
the LMS filter weights. A second equation was derived and tested for the transient
weight autocovariance matrix of the LMS filter weights. This equation works well
for a singleuser environment, but does not follow the observed response of the filter
weights for a multiuser situation. Therefore, the weight error covariance matrix was
used to derive a second equation for the transient weight autocovariance matrix. This
equation has been observed to agree within about a 1020% error of what is observed
through direct simulation. We then used these two iterative equations to derive two
analytical techniques to characterize the performance of the timing estimator.
The goal of the first technique was to provide a means of observing how quickly
the timing estimator can correctly estimate the propagation delay of the desired user,
in a chipsynchronous environment. The idea was to use the analytical expression as
60
a means to observe the timing estimator, without running long MonteCarlo simulaÂ¬
tions, or at least to get a sense for how to set up such simulations. Unfortunately,
evaluating the required probability was a difficult task at best, so we used several
methods to form an upper bound on the desired probability. We observed that the
upper bounds were very loose, and so this technique while interesting is of very limited
use.
The second technique was more along the lines of the traditional CramÃ©rRao
bound. We desired an analytical expression that could be used as an approximation
to the variance of the Â¿i estimate (about the true value of Â¿1) conditioned on getting
the integer part of the timing estimate correct. We observed that the resulting apÂ¬
proximation was useful for few users, with perfect power control, but was not useful
for more general cases.
Lastly, we characterized the performance of the timing estimator using Monte
Carlo simulation methods. We compared the performance of the adaptive filterbased
timing estimator to a conventional correlatorbased timing estimator. The correlator
based timing estimator is well known to be optimal for a singleuser operating in an
AWGN channel, but can be suboptimal for a multiuser DSCDMA environment.
The metric used to compare the timing estimator, was the average number of training
bits required such that fx was within onehalf a chip of the true value of rx > 90%
of the time. For few users, with low levels of MAI, we observed that the adaptive
filterbased timing estimator offered a slight improvement over the correlatorbased
timing estimator. However, as the number of system users increases, both the LMS
based and RLSbased implementations of the timing estimate perform better than
the correlator, with the advantage going to the RLSbased algorithm. The nearfar
resistance of the algorithm was tested, by simulating an environment that could occur
in an open loop system, or a system with very loose power control. Neither the LMS
based or the RLSbased estimation algorithms are truly nearfar resistant. However,
61
the change in the length of the required training interval for the RLSbased algorithm
was so minor that we consider it to be nearfar resistant.
While it is difficult to directly compare our timing estimator with other techniques,
as some of the other techniques estimate more than a singleuserâ€™s parameters, one
observation can be made. The complexity of the estimation algorithm is dependent
on N, the number of chips in one period of a spreading sequence. The complexÂ¬
ity of the estimation algorithm is 0(iV2) above the complexity of the traditional
correlatorbased system when the RLS algorithm is used. However, note that for
some environments reliable estimates of T\ are available with training sequences that
are shorter than N symbols long. In a previous chapter, other estimation algorithms
were presented. Most, if not all, had higher complexities and some required training
sequences that were at least N symbols long before a timing estimate could be made.
Therefore, our timing estimator has the advantage over these other estimators when
short training sequences are required, under the condition that our timing estimator
provides reliable estimates.
CHAPTER 6
SIMULATING FADING CHANNELS
Previously, the communication channel was modeled as an AWGN channel. A
more realistic channel model must be used to account for the dispersive effects obÂ¬
served in urban environments. In urban environments, it is possible for multiple
versions of a transmitted signal to arrive at a receiver due to reflections off of adÂ¬
jacent buildings, cars or other obstacles. If a total of L versions of the transmitted
signal are received, then the channel is said to have L paths. The signal received from
each path can experience its own time delay, phase shift and attenuation. Therefore,
at the receiver these L signals may add constructively or destructively, a condition
which is known as fading. If relative motion exists between the transmitter and the
receiver (or the reflecting obstacle is in motion), then the received signal is also obÂ¬
served to have a shift in its carrier frequency. Also, due to relative motion between
the receiver and the transmitter, it is easy to conceive that the characteristics of the
channel change as a function of time and, therefore, a more realistic channel model
will account for the time varying effects observed on the received signal.
Rappaport [54] provides a summary of the various types of fading, as well as preÂ¬
senting several techniques that can be used to simulate fading channels. Proakis [55]
also presents a characterization of fading channels and discusses the effect of fadÂ¬
ing on the performance of communication systems. The type of fading induced by
a communication channel is dependent on the time/frequency characteristics of the
channel with respect to to the time/frequency characteristics of the transmitted sigÂ¬
nal. Sklar [56] presents several block diagrams that summarize how these relationships
between the channel and the transmitted signal determine the type of fading induced
by the channel. In addition, Sklar presents a summary of traditional techniques that
62
63
have been used to mitigate the effects of fading [57]. In the remainder of this chapter,
we will briefly review the various types of fading and discuss how a fading channel
can be simulated.
Consider a multipath communication channel where there are L distinct paths
with unique propagation delays iq > r2 > . ..rÂ¿_ i > rÂ¿. The quantity â€” T\ is
defined as the excess delay spread. As expected, this multipath delay spread causes
time dispersion in the received signal. In fact, if the excess delay spread is significant
when compared to the symbol interval, the channel can be viewed as inducing severe
intersymbol interference (ISI). That is, one transmitted symbol will contribute to
several symbol intervals at the receiver. Since the number of paths, and hence their
corresponding propagation delays, are assumed to change as a function of time, the
excess delay spread is a random variable. The statistical characteristics of the excess
delay spread relative to the symbol interval along with the frequency characteristics
of the channel and transmitted signal determine if the fading is flat or frequency
selective.
If a communication channel has a constant gain and linear phase response over
a bandwidth that is greater than the bandwidth of the transmitted signal, then all
frequency components of the transmitted signalâ€™s spectrum will be equally attenuated
by the channel. In this condition, the channel is said to be flat. However, due to the
time varying nature of the multipath channel, the channelâ€™s impulse response and
hence its frequency response are time variant. In order to characterize the frequency
response of the channel, the coherence bandwidth is defined. The coherence bandÂ¬
width of a channel denotes a bandwidth in which different frequency components
experience the same amplitude attenuation. The coherence bandwidth of a multi
path channel is related to the excess delay spread in some manner. While no exact
relationship between these two quantities exists, there are several well known rules
of thumb [54]. Let the transmitted signal have a bandwidth, Bs, a symbol duration
64
T5, while the channel has a coherence bandwidth Be and rootmeansquared excess
delay spread of at. The following statements regarding classification of the fading
channel with respect to the transmitted signal will be described by T5, in order to
describe a general communication system. With T$ used to denote the duration of the
transmitted symbol, we must remember that for a DSCDMA system Ts is equal to
Tc, the duration of onechip interval, when applying the following statements. If the
bandwidth of the transmitted signal is much less than the coherence bandwidth of the
channel, Bs
delay spread, Ts >> crt, then the channel induces flat fading. If the bandwidth of the
transmitted signal is greater than the coherence bandwidth of the channel, Bs > Be,
and the symbol duration is less than the RMS excess delay spread, Ts < crt, then the
multipath channel induces frequencyselective fading. Note that flat fading channel
models are more prevalent since a frequencyselective fading channel can be viewed as
the composite sum of a multipath channel where each path induces flat fading, but
the propagation delays between the individual paths are set such that the composite
channel induces frequencyselective fading.
Similar to the time dispersion induced by the multipath delays, relative motion
between a transmitter and its receiver induces dispersion in the frequency domain.
Consider a singlepath channel over which a constant amplitude and frequency sinuÂ¬
soidal signal is transmitted. If there is relative motion between the transmitter and
receiver then there is a frequency shift observed in the received signal. The maximum
observable frequency shift, fp, is known as the Doppler frequency or Doppler spread
and is given by
(6.1)
65
where v is the relative speed between the two devices in m s 1, C is the speed of light
in free space and fc is the carrier frequency in Hz. When a timevarying multipath
channel is considered, the observed frequency shift becomes random and timevarying.
The relative speed between a transmitter and receiver determine the rate at which
the impulse response of the channel changes. In order to characterize this rate of
change the coherence time of the channel, Tc, is defined. The coherence time of the
channel is used to characterize the correlation of samples of the channel output. That
is, if two samples of the channel are separated in time by less than the coherence time
of the channel, they will be highly correlated (similar). As before, there are several
known rules of thumb that are used to relate the coherence time of a channel to
the Doppler spread. If the symbol duration is greater than the coherence time of
the channel, Ts > Tc, and the bandwidth of the transmitted signal is less than the
Doppler spread, Bs < /o, then the channel impulse response changes many times
during a symbol interval and the channel induces fast fading. If the symbol interval
is much less than the coherence time of the channel, Ts C Tc, and the bandwidth
of the transmitted signal is much greater than the Doppler spread, Bs Ãd, then
the channel is essentially timeinvariant when viewed over several symbol intervals.
In this case, the channel is said to induce slow fading.
The Rayleigh distribution is commonly used to characterize the timevarying
statistics of the envelope of the received signal in a flatfading environment. It is
well known that the random variable z = y/x2 + y2 when x and y are independent
zeromean Gaussian random variables with equal variÃ¡nces a2 follows a Rayleigh
distribution. The probability density function for a Rayleigh distribution is given by
otherwise.
(6.2)
66
If however there is a dominant path in the communication channel, such as the case
when a lineofsight (LOS) path exists, the envelope of the received signal is modeled
by a Rician distribution and has a probability density function of
(6.3)
otherwise.
where 70() is the zeroorder modified Bessel function of the first kind. Since the
Rayleigh distribution is just a special case of the Rician distribution, it is easy to
create a fading channel simulator that generates both distributions for the envelope
of the received signal.
The fading channel simulator used in this work is based on Clarkeâ€™s [58] channel
model. In that work, the fading was due to the scattering of electromagnetic waves in
such a way that all paths had the same propagation delay. This means that all paths
form one composite path, and since there is no delay spread the channel induces flat
fading. The received envelope was shown to be a random process whose amplitude
follows a Rayleigh distribution and has an autocorrelation function given by
R(t) = M2nfDT)
(6.4)
where J0() is the zeroorder Bessel function of the first kind. Therefore to simulate
a channel based on this model, we must create a random process whose amplitude is
Rayleigh distributed, in such a way that the autocorrelation of the random process
is similar to equation (6.4).
A thirdorder filter has an transfer function of
(6.5)
67
and a corresponding impulse response of
h(t) = (exp(â€”ctf) [Asin(/3t)  B cos((3t)\ + Cexp(a>0Â¿)) u{t) (66)
where
A =
UJq
20^?
a = (u0
B â€” C =
2(10
/3 = W 1C2.
By setting uq and C based on the value of the Doppler frequency as shown below,
UJq
2t\J_d
1.2
c = 0.175
(6.7)
(6.8)
the convolution of h[r) with h(â€”r) closely follows equation (6.4) over the first major
lobe. This means that if a complex white Gaussian noise process is used as the input
to the thirdorder filter, the autocorrelation of the output process closely follows
equation (6.4) for values of r where samples of the process are highly correlated.
And hence, we use this thirdorder filter to simulate a flatfading channel based on
Clarkeâ€™s model.
Figure 6.1 shows a block diagram of a fading channel simulator based on the
thirdorder filter that can be used on a digital computer. The continuoustime filter
is converted to a digital filter where the sampling rate is specified, typically equal
to one bit interval. The input to the filter are samples of complex Gaussian noise,
whose real and imaginary parts are independent, have zeromean and equal variÂ¬
ances of 1/2. The block diagram accounts for the presence of an LOS, or specuÂ¬
lar component, such that the envelope of the simulator output can follow both the
Rayleigh and Rician distributions. Therefore the output of the channel simulator is
68
Samples of
Complex
Gaussian Noise
Figure 6.1: Flatfading channel simulator.
a sampled complex Gaussian process having a mean of A exp(jcfi) and a variance of
a2 = E [\Ak exp(j0k)  Aexp(j(Â¡))\2].
In Figure 6.1, two parameters are shown to be a function of a variable called p.
The ratio p â€” A/a is referred to as the Rician parameter and the pair {A, a) are
normalized so that the fading process has unit power.
<*{p)  R 2 P  0
y/l + p2
(6.9)
A(P)~ nrâ€”T
V1 + p
(6.10)
A plot comparing the theoretical probability density functions of the envelope of
the fading process given in equations (6.2) and (6.3), to the those observed through
simulations for two values of p is displayed in Figure 6.2. As shown in this figure,
when p = 0 the envelope of the fading process is Rayleigh distributed, while for
other values the envelope of the fading process follows a Rician distribution. As a
final demonstration of the fading channel simulator, Figure 6.3 displays a typical
realization of a flat Rayleigh fading process. The magnitude of the envelope and
the phase of the fading process are shown when the carrier frequency was 1.8 MHz,
the data rate was 9600 bitspersecond (BPS), and the relative speed between the
transmitter and the receiver was 3 milesperhour (MPH). The third order digital
filter, as well as the output of the fading channel simulator where sampled at the
data rate.
69
O 0.5 1 1.5 2 2.5 3 3.5
r
Figure 6.2: Theoretical and simulated probability density function for the envelope
of the fading channel simulator output for two different Rician parameter
values.
Sample Number
Figure 6.3: Magnitude and phase of a slow flatfading Rayleigh process generated
with a carrier frequency of 1.8 GHz, a vehicle speed of 3 MPH, and a
data rate of 9600 BPS.
70
Consider simulating a singleuser DSCDMA system operating in a fading channel
environment. For a given Doppler frequency, the fading channel simulator is created
and assume that both the thirdorder filter as well as the simulator output are samÂ¬
pled at the symbol rate. In a nonfading channel the contribution to the receiverâ€™s
input during the mth symbol interval due to the user is given by Ji(m) as shown
in equation (2.7). Multiplication of Ji(m) by the mth sampled output of the fading
channel simulator, A(m) exp(j0(m)), at each symbol interval will produce an input
to the receiver whose envelope experiences fading. In order to simulate a multiuser
fading channel DSCDMA system, each user should have its own fading process that
is independent from all other users. We could implement K independent fading chanÂ¬
nel simulators, and then multiply Jfc(m) by the output of the corresponding fading
channel simulator. However, a simpler approach was used in this work. A single
fading channel simulator was created for a specified Doppler frequency. Then a very
large sequence of samples of the fading process was created. Consider two samples
of this sequence, A(n) exp{j0(n)) and A(n + m) exp(j0(n + m)). We know that if m
is chosen appropriately then the correlation between the samples becomes negligible,
and therefore we can treat the samples as being independent. Using this idea, the
original sequence was partitioned into K sequences such that each sequence is conÂ¬
sidered to be independent of each other. That is, the sequence for the first user was
created using samples (1,2,..., m), while the sequence of the second user was created
by using samples (m + 1, m + 2,..., 2m), and so on for each user.
CHAPTER 7
CODE ACQUISITION IN NONSTATIONARY ENVIRONMENTS
It is wellknown that training an adaptive filter in a nonstationary environment
can be difficult [31]. Since our timing estimator is based on processing the weights
of the adaptive filter, a study of how the estimator is affected by such environments
is required. In this chapter, we will consider timing acquisition in two such environÂ¬
ments. The first environment occurs due to nonideal downconversion of the received
signal due to frequency synchronization errors. When this condition occurs, the input
to the adaptive filter has a timevarying phase induced on it. The second condition
to be studied is when the received signal has experienced fading, in which case both
the amplitude and phase of the received signal are timevarying.
Barbosa and Miller [13] studied the adaptive detection of DSCDMA signals in
fading channels. In particular, they studied the performance of the adaptive receiver
shown in Figure 4.1 operating in frequency nonselective fading channels. They found
that this receiver structure does not work in a Rayleighfading channel. When the
desired userâ€™s signal experiences a deep fade, the receiver frequently loses lock on
the desired signal. When the desired userâ€™s signal emerges from the deep fade, the
receiver may emerge into any of 3 possible states:
1. correctly locked on phase to the desired userâ€™s signal;
2. locked 180Â° out of phase to the desired userâ€™s signal; or
3. locked either in phase or 180Â° out of phase to any of the interfering user signals.
In a singleuser environment, only the first two of these conditions are possible. The
first condition is desired. The second condition is not due to the filterâ€™s inability to
track the rapid phase changes during a deep fade. It is due to the fact that since the
decisions are unreliable in a deep fade (many errors occur in the feedback loop) the
71
72
filter is essentially running â€œblindâ€ during the deep fade. Once the filter locks onto the
desired userâ€™s signal at 180Â° out of phase, errors will continue to occur. This problem
can be handled by differential encoding and decoding of the data sequence. The third
condition is catastrophic, and is what renders this receiver structure useless in this
type of environment. Barbosa and Miller proposed a modified receiver structure to
alleviate these problems. The next few paragraphs describe this receiver structure
and its derivation.
The main problem with the adaptive receiver structure shown in Figure 4.1 seems
to be loss of phase lock during deep signal fades. Therefore, it stands to reason
that if reliable estimates of the fading process can be made, these phase variations
could be removed from the input to the adaptive filter, Then, maybe the adaptive
receiver could perform adequately, even in a Rayleighfading channel. In a nearfar
environment, the signaltointerferenceplusnoise ratio for the desired userâ€™s signal
would probably be too small to allow for phase estimation without a complicated
estimation procedure. However, the filter suppresses the multiaccess interference
such that the signaltointerferenceplusnoise ratio at the filter output is higher than
the same quantity at the filter input. Therefore, the filter weights are used in the
channel estimation procedure.
By taking the real part of the received signal prior to entering the adaptive filter,
the weights of the filter are real and, therefore, the filter makes no attempt to track
the phase of the received signal. Noisy estimates of the amplitude and phase of the
fading process during the mth bit interval for the desired user may be formed as
7{m) = di(m)wr(m)r(m). The weights of the filter are updated at the bit rate,
which means that 7(771) can be updated at the bit rate. Therefore, we must predict
the channel conditions for the current bit interval based on previous values of 7(777).
An Lthorder linear predictor is used to form the current estimate of the channel
73
conditions, based on past estimates as shown in equation (7.1),
L
7 {m) = ^2aa{mi) (7.1)
1=1
and the estimate of the phase during the mth bit interval is found by 6^m = Z7(m).
The coefficients of the Lthorder linear predictor are chosen to minimize the mean
squared error E[ai,m exp(j0im) â€” 7(ra)2], where c*i,m and 0i,m are the amplitude and
phase induced on the desired user by the channel during the mth bit interval. The
solution of these coefficients depends on the weights of the MMSE filter, which depend
on the multiaccess interference. In order to remove this dependence, the coefficients
were calculated for a singleuser system, and then are used for all other environments.
Not surprisingly, the coefficients of the Lthorder linear predictor, a. are found by
a = C1v (7.2)
where
MÂ¿ = 1  (mfDTb)2 *â‚¬{1,2,...,/,} (7.3)
{B}y = 1 â€œ ((* â€œ j)*fDTbf i,ie{l,2,...,L} (7.4)
C = B + (Eb/Ky1! (7.5)
and I is an L x L Identity matrix and fD is the maximum Doppler frequency induced
by the channel. It is important to remember that these equations were derived for
a singleuser and, therefore, in the presence of multiaccess interference the channel
estimator may be very far from optimal.
The modified form of the adaptive receiver, using the Lthorder linear predictor
designed by Barbosa and Miller (equations (7.2) through (7.5)) is shown in Figure 7.1.
Now that we have presented the background of this receiver, we summarize several
74
* Estimation ti
w(m^ algorithm
important results presented by Barbosa and Miller [13]. They found that the perforÂ¬
mance of the receiver (detection) is insensitive to L, the order of the linear predictor.
Also, unlike the unmodified form, the performance of the detector is independent of
the rate of the fading process as the detector performed equally well in slowfading as
well as fastfading. However, the most important result is the effect of the modified
receiver structure on the system capacity. For a given level of probability of bit error
rate, it was found that the structure in Figure 7.1 could support more users than the
conventional receiver (which is a correlator). In fact, it was shown that the system
capacity when using the modified receiver structure can be made to approach 100%
(N users), while the conventional receiver typically restricts the system capacity to
on the order of 1020%.
Several comments about the modified receiver structure as it relates to timing
acquisition are required at this point. Barbosa and Miller were interested in detection,
and the weights of the adaptive filter were initialized close to their steadystate values.
The result is that the Lthorder linear predictor can make reliable estimates of the
channel conditions at the start of their simulations. In this work we are interested in
timing acquisition, and therefore the weights of the adaptive filter are initialized to
75
a vector of all zeros. This means that most likely the initial estimates of the channel
phase will be incorrect. As seen in Figure 7.1, the phase compensation operation,
Re r(m) exp(â€” j9\^m) , causes a delay in the build of energy in the filter since when
0\m â€” Oi(m) is significant, the phase compensation operation rejects some of the
desired userâ€™s input signal from the input to the adaptive filter. The time to achieve
correct acquisition will be dependent in some manner on the transient response of
the channel estimation algorithm.
7.1 Frequency Synchronization Errors
Consider the situation that occurs when the frequency of the A:th userâ€™s transmitÂ¬
ting oscillator is not equal to the frequency of the receiverâ€™s oscillator. In this case
after downconversion, the input to the adaptive filter for the kth user will have a
linearly timevarying phase. In order to model this effect, we assume that the freÂ¬
quency offset for the kth user, Au>k, can be expressed as Au>k = 2n/(NTcMk) where
Mk is a positive integer and NTC is the bit interval.
Using this notation, we can model the input to the adaptive filter using equaÂ¬
tion (2.6) once we redefine Jk(m) as shown below.
(7.6a)
Z2ki(m)v2ki(Pk, 4) + z2k(m)v2k(pk, 4)
where the nth components of v2ki(pk,Ã³k) and v2k(pk,Sk) are given by:
(7.6b)
76
As a sanity check of these equations, note that in the limit Mk â€”Â»â– oo the above
equations reduce to the previous definition of Jfc(m).
The results in Figure 7.2, show the acquisition performance of the estimation
algorithm in the presence of frequency synchronization errors. The desired user was
received at a SNR of 10 dB, and 4 interfering users were each received at a power level
10 dB above the desired user. The desired user was assumed to have a frequency offset
of 1% of the data rate (Mi = 100), while all other users had a frequency offset of 2%
of the data rate (M25 = 50). The timing offsets and initial phaseoffsets for each user
were chosen at random, for each of the 250 trials. For each trial, 3 independent RLS
filters were run in parallel. The first filter was uncompensated, while the second filter
used the modified receiver structure with a 10thorder linear predictor to estimate
the channel phase. As a control, the third filter was compensated with the known
phase induced on the desired user at each bit interval. By doing this, the effects of
the transient response of the channel estimation algorithm on the timing estimator
can be avoided.
As shown in Figure 7.2, the uncompensated filter was initially able to form a good
estimate of propagation delay for the desired user. However, the filter was unable
to track the phase of the input signal and therefore the timing estimation algorithm
eventually breaks down. When the modified receiver structure is used, and the input
signal is compensated using perfect knowledge of the phase induced on the desired
user, the timing estimation algorithm works very well. Once again, this is due to the
fact that the transient response of the channel estimator and the phase compensation
operation are not causing a delay in the buildup of the desired userâ€™s signal in the
77
1
^ 0.8
0.6
i
jÂ±L 0.4
TT
a,
0.2
0
0 50 100 150 200 250 300
Time (Symbols)
Figure 7.2: Acquisition performance of the adaptive receiver in a fiveuser environÂ¬
ment in presence of frequency offset errors.
weights of the adaptive filter. The 10thorder linear prediction compensated filter,
unlike the uncompensated filter, is able to achieve correct acquisition at the cost of
an increase in the length of the training interval.
Considering the results of Figure 7.2, it seems reasonable that a hybrid of the
uncompensated and compensated filters may be useful in forming a timing estiÂ¬
mate. That is, it may be possible to start acquisition using an uncompensated filter
(with complex weights) and at some point switch to the compensated filter (with real
weights) such that we get better acquisition performance. We will address this idea
in the next section where we study the performance of the acquisition algorithm in a
flatfading channel.
78
7.2 Performance in FlatFading Channels
We model the input to the adaptive filter using equation (2.6), by defining the
contribution of the kth user during the mth bit interval as
Jfc(m) = afc(m)exp(j0fe(m)) [z2ki{m)a2ki(pk,Sk) + z2k{'rn)&2k{Pk,&k)\ (77)
where ock{m) and 0fc(m) are the amplitude and phase induced on the kth userâ€™s signal
by the fading process during the mth bit interval. The fading process for each user is
assumed to be independent of the fading process induced on each of the other users.
The fading parameters, ajt(m) and 9k{m), are taken from samples of the flatfading
channel simulator described in the previous chapter.
As a first test of the acquisition algorithm, a tenuser DSCDMA environment
was created with a SNR of 10 dB and all interfering users were received at the same
power level as the desired user (in the absence of fading). A slowly flatfading channel
was created for each user, using a carrier frequency of 1.8 GHz, a data rate of 9600
BPS, and a speed of 3 MPH. These parameters are commonly used to simulate a
personal communication system (PCS) environment in which the users are mobile at
walking speeds. The resulting Doppler frequency, fD, is 8 Hz which is a normalized
rate (normalized to the data rate) of 8.4 x 104.
To test the acquisition algorithm in this environment, a bank of four parallel RLS
filters was used. One filter, was uncompensated as shown in Figure 4.1, and processed
the received signal using complex weights. Two of the filters, used the architecture
shown in Figure 7.1 with a Lthorder linear predictor for the channel estimator. In
the following, we will refer to each of these filters as the modified MMSE receiver.
One of the modified MMSE receivers compensated the received signal using only
the phase estimate of 7(m) (as shown in Figure 7.1). The second modified MMSE
receiver compensated the received signal using both the amplitude and phase of 7(771).
The fourth filter, referred to as the hybrid receiver, was an adhoc mixture of the
79
two receiver structures. The idea of the hybrid receiver is to jumpstart the channel
estimation algorithm by preloading the adaptive filterâ€™s weight vector, w(m), and the
linear predictor vector, 7, hopefully bypassing some of the learning curve required to
form reasonable channel estimates otherwise. The hybrid receiver uses the structure
of Figure 4.1 for 20 bit intervals without making timing estimates. After the 20th bit
interval, the receiver forms an estimate of T\ and then makes an estimate of 9i using
equation (4.14a). This phase estimate is loaded into the Lelements of 7, and the
imaginary parts of the filterâ€™s weight vector and the inverse of the correlation matrix
are removed. The receiver then switches to the modified receiver structure shown in
Figure 7.1, and starts the timing and channel estimation algorithms using the (now
real) weights of the adaptive filter.
The results of this test when a thirdorder linear predictor was used for the channel
estimation algorithm are shown in Figure 7.3. As seen in this figure, the acquisition
algorithm based on processing the weights of the uncompensated filter performs betÂ¬
ter than any of the other techniques. In general, this situation will not occur as the
level of the MAI changes and/or the Doppler frequency changes, and results similar
to that shown in Figure 7.2 can be expected. Barbosa and Miller [13] observed the
detection performance of the modified MMSE receiver was insensitive to L, the order
of the linear predictor used in channel estimation. In addition, they compensated the
received signal by only using the phase of the channel estimate, 7(777), as no advanÂ¬
tage was observed by also using the amplitude of 7(771) to compensate the received
signal. Since they were interested in detection, they did not observe the startup
transient of the channel estimation algorithm as the weights of the adaptive filter
were initialized to their steadystate values. In Figure 7.3, we observe the effects of
the startup transient of the channel estimation algorithm on the performance of the
timing estimation algorithm for the modified MMSE receiver. Unlike the observations
of Barbosa and Miller, it is observed that including the channel amplitude estimate
80
Figure 7.3: Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a thirdorder linear predictor for the channel estimation
algorithm.
in the compensation operation imposes a severe penalty in the performance of the
timing estimate when compared to the performance of the timing estimate when only
the channel phase estimate is used. To study the effects of the value of L on the
timing estimation algorithm, the same DSCDMA environment was observed when
a tenthorder linear predictor was used in the channel estimation algorithm. These
results are shown in Figure 7.4. Once again, unlike the observations of Barbosa and
Miller, the performance of the timing estimation algorithm is sensitive to the value of
L. When a tenthorder linear predictor is used in the channel estimation algorithm,
the timing estimation algorithm performs better than when a thirdorder linear preÂ¬
dictor is used as noted by the difference between the rates of convergence of the two
probability curves in Figures 7.3 and 7.4. For a variety of DSCDMA environments
(Doppler frequency, MAI, and number of users), it has been observed that no signifÂ¬
icant improvement in the performance of the timing estimator is achieved by using
a higher order linear predictor in the channel estimation algorithm. Therefore, in
81
Figure 7.4: Performance of the acquisition algorithm in a slowly flatfading Rayleigh
channel using a tenthorder linear predictor for the channel estimation
algorithm.
the sequel when we refer to the modified MMSE receiver we are using a tenthorder
linear predictor and only the phase of the channel estimate is used to compensate
the received signal. As seen in Figures 7.3 and 7.4, eventhough the hybrid receiver
has been formulated in an adhoc fashion the resulting performance of the timing
estimator clearly shows that the idea has its merits.
In order to further test the hybrid receiver, a study of the effects of the level of
MAI on the performance of the timing estimation algorithm similar to that shown
in Figure 5.8 was performed. The carrier frequency was 900 MHz, the data rate was
9600 BPS, and the relative speed between the receiver and all of the system users was
assumed to be 65 MPH. This scenario is used to depict a DSCDMA system in which
the users are traveling at highway speeds. The Doppler frequency induced by the
flatfading channel was 87 Hz, which is a normalized Doppler rate of 9.1 x 10~3. Each
DSCDMA environment was simulated for 500 independent trials. For each trial,
the user spreading codes were selected at random from a set of Gold codes using
82
Figure 7.5: Comparing the average training bits for correct acquisition as a function
of the level of multiaccess interference in a flatfading channel.
31 chips/bit. For each trial, the propagation delay for each user was chosen from a
uniform distribution over the interval [0,31), independent from the other users. The
desired user was received at a SNR of 10 dB. Also at the start of each trial, the received
power level Pk for each interfering user was selected from a lognormal distribution
that had a mean and standard deviation of X dB. That is, Pk/P\ = 10^10 where
Â£ is a Gaussian random variable with a mean and standard deviation of A dB. By
varying the number of system users and the level of the MAI, it is possible to study
the performance of the hybrid receiver in conditions where the uncompensated filter
shown in Figure 4.1 doesnâ€™t perform well, but the hybrid initialization technique may
improve the acquisition performance over that offered by just the modified MMSE
receiver alone.
The results of this test are shown in Figure 7.5 for a threeuser and tenuser DS
CDMA system where the level of the lognormally distributed interference (X) ranged
from 0 dB to 13 dB. For both the threeuser and tenuser systems, the hybrid receiver
performs better than the modified MMSE receiver in terms of the average training
83
Figure 7.6: Performance of the timing acquisition algorithm using the hybrid and
modified MMSE receivers.
bits required in order to achieve acquisition to within a halfchip of T\. When the
hybrid receiver was used in the threeuser environment, the performance of the timing
acquisition algorithm appears to be independent of the level of the MAI. This result
is due to the length of the training cycle used in the hybrid receiver before timing
estimates are actually made (20 bits). For a threeuser system, the uncompensated
filter is able to make reliable timing estimates prior to 20 training bits and, therefore,
it may be possible to use a shorter training interval for the hybrid receiver initialÂ¬
ization cycle without degrading the performance of the timing acquisition algorithm.
The length of the training cycle used to initialize the hybrid receiver is sensitive to
the number of system users and the level of the MAI. However, over the range of users
and lognormally distributed MAI depicted in Figure 7.5, 20 bits of training provide
the hybrid receiver with a definite advantage over the modified MMSE receiver in
terms of timing acquisition performance.
84
A situation in which the hybrid receiver offers no advantage over the modified
MMSE receiver is shown in Figure 7.6. This result is for the same DSCDMA enviÂ¬
ronment tested in Figure 7.5, however there are fifteen systemusers and X = 1 dB for
all of the lognormally distributed interfering users. This plot readily demonstrates
how the performance of the timing acquisition algorithm using the hybrid receiver
is â€œmodulatedâ€ by the performance of the uncompensated filter and the length of
its training cycle before switching to the modified MMSE receiver structure. While
decreasing the length of the training cycle below 20 bits might not affect the perÂ¬
formance of the timing acquisition algorithm, certainly increasing the length of the
training cycle to 100 bits would be disastrous.
7.3 Summary
Barbosa and Miller [13] studied the performance of the MMSE receiver in a flatÂ¬
fading channel, and discovered that the MMSE receiver doesnâ€™t work in a flatfading
Rayleigh channel. They proposed a modified receiver structure which compensates
the received signal such that the adaptive filter doesnâ€™t attempt to track the phase
of the desired userâ€™s signal. In this chapter, the modified MMSE receiver structure
was used to study the performance of the timing acquisition algorithm in two enÂ¬
vironments in which the input to the adaptive filter was nonstationary. While the
timing acquisition algorithm is applicable to use with the modified MMSE receiver, it
has been observed that the startup transient of the channel estimation algorithm deÂ¬
grades the performance of the timing estimation algorithm by increasing the length
of the training interval required to achieve correct acquisition. A hybrid receiver
structure has been presented to circumvent some of the startup transient response of
the channel estimation algorithm. The hybrid receiver uses a short training interval
to form an estimate of the phase induced on the desired user, Â§i, and then uses the
modified MMSE receiver with the channel estimator preloaded with Â§i to continue
the timing acquisition cycle. Through simulation results of a DSCDMA system, it
85
has been observed that the timing acquisition algorithm based on the hybrid receiver
structure can offer an improvement over that of the modified MMSE receiver. HowÂ¬
ever, the initialization cycle for the hybrid receiver is based on an uncompensated
MMSE receiver and therefore with an inappropriate choice of the length of the initial
training cycle, it is possible for the performance of the timing acquisition algorithm
for the hybrid receiver to be worse than that offered by the modified MMSE receiver
alone.
CHAPTER 8
INCREASED WINDOW OF OBSERVATION AND TIMING ACQUISITION
In previous chapters, the adaptive receiver was used to form a timing estimate of
the propagation delay for the desired user. Since the receiver only observed one bit
interval of information at a time, the desired user is forced to use an all ones training
pattern during the code acquisition cycle. However, it is possible to use a known
sequence of ones and zeros to train the adaptive receiver, if the receiver observes more
than one bit interval of information at a time. By letting the observation window of
the receiver be two bit intervals, a full data bit (and possibly two bit transitions) are
guaranteed to occur within the observation window. This increase in the observation
window also allows for more than one user to be in the code acquisition mode at the
same time.
8.1 Updated System Model
In this section, the system model for an observation window of two bit intervals will
be presented. This model is just a special extension of that presented in Section 2.1,
however a notation similar to that presented by Madhow [59] will be used. As with
the previous system model, the receiver converts r(t) to a baseband signal and passes
R(t) through a filter matched to the chip pulse shape and the output of the filter is
sampled at the chip rate. As before, we let the quantity rm^n represent the nth chip
sampled output during the receiverâ€™s mth bit interval.
TnTi,\(n\l)Tc
Tm'n ~ \/2~P[T f â‚¬ C. (8.1)
mTb+nTc
86
87
During each bit interval, a total of N chip samples are gathered and grouped with
the N chip samples from the previous bit interval such that
r(m) = {Xmâ€”1,0) I'mâ€”1,1) â€¢ â€¢ â€¢ > I'm,0) Tm,\i â€¢ â€¢ â€¢ i ^m,Nâ€” l) (8.2)
and it is this vector quantity that will be used as the input to the adaptive receiver.
As with the previous system model,
r(m) = Â¿
Jt=i
where Jk(m) is the contribution to the received vector from the kth user during the
mth receiver bit interval. Let the (4,m represent the data bit for the kth user that falls
completely in the two bit observation interval during the receiverâ€™s mth bit interval.
Unlike before, we now define the propagation delay for the Arth user to be rk G [0,TÂ¡,]
such that rk â€” pkTc + SkTc with pk G {0,1,..., N  1} and 0 < Sk < 1, or rk = NTC.
Let ak represent a vector of length 2N consisting of the N elements of the spreading
sequence for the kth user followed by N zeros.
C/e, 1) â€¢ â€¢ â€¢ > Ck,Nâ€”1) 0j â€¢ â€¢ â€¢ > 0) (8.4)
Let T^'0 represent the nth left acyclic shift of a vector, while represents the nth
right acyclic shift of a vector. Both functions operate on and produce vectors with
2N elements. Using a general vector x = (rr0,Â£i,... ,x2ni)t these functions are
defined below.
T^(x) = Tj^(x) = x
(8.5)
T i1}(x) = (xi,x2,...,x2Ni,0)T
(8.6)
r1}(x) = (x2wi,0,...,0)t
(8.7)
T2)(x) = (0, X0, X\, .. .,x2N2)t
(8.8)
Jfc(ra) exp(j9k) + Nm G C:
QNxl
(8.3)
88
TfiN_1)(x) = (O,..., O, x0)T (8.9)
tÂ£Â°(x) = T^(x) = (0,...,0)r Vn>2 N (8.10)
Using the above definitions we can now define the kth userâ€™s contribution to the
received vector as
Jfc(^) â€” dk,mâ€”k dk'fnVfc + dfc,m+lV^. (8.11)
where
V,1 = (1  4)Tfw)(a*) +
vÂ» = (1  Ãt)Tg*>(aÂ») + 4TÂ£*+1)(a*)
vi = (1  ÃÂ»)TSf,+M)(aÂ») + itT^,+"+1(al)
(8.12)
(8.13)
(8.14)
Using this notation, the autocorrelation matrix of the adaptive filterâ€™s input and the
steering vector are evaluated as shown below.
R = E [r(m)rH(m)]
= y (v*:lKâ€œl)r + vM)r + vÂ¿(vÂ¿)T)
fc=i v '
(8.15)
+ 2
p = E[r(m)di,m] = v? exp(j6Â»i)
(8.16)
8.2 Updated Timing Estimation Algorithm
We are interested in developing a new timing estimator for the propagation delay
of the desired user, now that we have increased the length of the filter. We will use
the same procedure as before. That is, the weights of an MMSE filter for a single
user will be used to form the timing estimator. Therefore, we need to calculate the
weights of the optimal filter for a singleuser operating in an AWGN channel. In
89
order to study the weights of the WienerHopf filter, we consider the case where we
apply the equivalent phasesynchronous model for the equations listed above. As in
the other system model, this means the filter weights become real and the complex
exponential term is removed from the steering vector expression. In this case, the
optimal weight vector is found using the WienerHopf equation with Rwopt = vÂ°,
where we let wopt = avf1 + + cv}.
The coefficients of the optimal weight vector, a, b and c, are given by the solution
to the following equation.
^dllvr1!!2
(vfTv? G2 + Vl
(vrTv? (vrTvi
0"2 (V0)TV1
,1II2
a
0
b
=
1
c
0
(8.17)
(vr )rvJ (v?)TvJ a2+ Vi,
Using the above equation, we see that the a and c coefficients can only be zero when
both vf1 and v{ are orthogonal to vÂ°. This condition occurs when ^ = 0. However,
in the general case when 8\ ^ 0, the a and c coefficients will be much smaller in
magnitude than the b coefficient, since vi: and v{ are only nonzero in one chip
position where vÂ° is nonzero also.
Based on this result, we will ignore the contributions of the vf1 and vj vectors on
the WienerHopf weights when we develop the timing estimator. Using the weights of
the adaptive filter, we estimate the propagation delay for the desired user by finding
the set (0,Q,p,8) that minimize w(m)  /3vJ exp(j0)2. As in the case where the
receiver only observes one bit of information, the estimate of T\ is given by the set
{p, Â¿} which maximize the cost function f(p,8) given below.
f{pJ) =
l(v?)rw(m)
11trO112
(8.18)
Maximizing / (p, 8) with respect to 8 produces a quadratic equation whose coeffiÂ¬
cients depend on the value ofp. Therefore for each possible value of p G {0,1,.... N â€”
1} we must find the roots of the quadratic equation D2(p)p2 + Di(p)p + D0{p), where
90
the coefficients are given by the following set of equations.
D0(p) = NA1(p) + 2CA0(p)
D\(p) = 2NA2(p)  iCA0(p)
D2 (p) = â€” 2C(A2(p) + Ai(p))
C = N afT^iaO
A0(p) = w//(m)T^(a1)2
Ai(p) = 2Re (T^)(ai))rw(m)w/i(m)T^+1)(a1)
 2A0(p)
A2(p) = w/i(m)T^,+1)(a1)2  Ax(p)  A0{p)
(8.19a)
(8.19b)
(8.19c)
(8.19d)
(8.19e)
(8.19f)
(8.19g)
The propagation delay estimation algorithm for the desired user is summarized
as follows. Let T represent the set of costs corresponding to the set of candidate
estimates of (p,6), denoted by U. We can find the maximum of f(p,6) as follows:
Step 1. Let T = {/(0,0), /(1,0), /(2,0),..., f(N, 0)}.
Step 2. Let U = {0,1,2,..., A^}.
Step 3. For p = 0,1,..., N â€” 1, do the following:
a) Compute the coefficients D2(p), D\(p), D0(p).
b) Solve for the two roots of the quadratic equation, p\ and p2
c) If 0 < pi < 1 for i E {1,2}, add the cost f{p,p,i) to the set T, and add
(p + pi) to the set U.
Step 4. Let Â¿4 denote the kth element of the set U. Then f = Z4, where k = max7j.
j
Step 5. The estimates of px and are given by pi(ra) = [fj and Si(m) â€” f â€” [fj,
where rounds x to the nearest integer towards zero.
The updated system model/timing estimation algorithm are briefly tested to
demonstrate what affects, if any, are observed on the performance of the timing
estimate. In Chapter 4, several plots of the cost function, f(p,S), are shown to
91
Figure 8.1: Plot of the cost function /(p, Â¿) for a fifteenuser system when the receiver
observes two bitintervals.
demonstrate how the timing estimate is formed under two given scenarios when the
WienerHopf weights are used. In Figure 4.4, the cost function is shown for an enviÂ¬
ronment where there are fifteenusers and SNR = 7 dB, T\ â€” 15.25TC and all of the
interfering users are received at a power level 10 dB above the desired user. Using the
timing estimator of Chapter 4, the resulting timing estimate is T\ = 7.43TC. Clearly,
in this environment the system is overloaded to the point where a reliable estimate is
infeasible. Using the same environment, we plot the cost function for the case when
the receiver observes twobit intervals in Figure 8.1. In this case, the timing estimate
is T\ = 15.1523TC which is a reliable estimate. We can justify this improved perforÂ¬
mance, to some extent, by considering how the timing estimate is formed for the two
different system models. When the receiver observes only onebit interval, the filter
weights are processed with a cyclicshift of the spreading code of the desired user.
In the case when the receiver observes twobit intervals, an acyclicshifting operation
is used. The difference between the two operations results in the sidelobes (those
values of r not near Ti) of the cost function being modulated differently. Therefore, it
92
Figure 8.2: Comparing the performance of the timing estimator when the receiver
observes 2bit intervals (*) versus observing 1bit interval (solid line) as
a function of the MAI.
is easy to conceive that it is possible (but not guaranteed) that the amplitude of the
peak sidelobe in Figure 4.4, is reduced in amplitude such that the correct estimate
is formed.
As a further test of the updated system model/timing estimation algorithm we
compare the performance of the timing estimate to those results shown in Figure 5.8.
The nearfar performance of the timing estimate was observed for a multiuser enviÂ¬
ronment with a SNR = 7 dB, and all interfering users received at power levels that
were lognormally distributed. The power of each interfering user relative to the deÂ¬
sired user was Pk/Pi = 10f/10, where f was a Gaussian random variable with mean
and standard deviation of X dB. The performance of the timing estimate when the
filter weights are adapted using the RLS algorithm is shown in Figure 8.2 for both
system models. A seen in this figure, there is only a slight difference between the
performance of the two systems over a wide range of users and level of MAI.
93
8.3 Summary
In this chapter, we have modified the system model so that the receiver observes
two bit intervals of data. If only one bit interval of data is observed, the training
sequence is restricted to being an all ones sequence to avoid bit transitions in the
desired userâ€™s received signal. By letting the receiver observe two bit intervals of
data, the desired user can now send a known data sequence during training. In
addition, it is possible for more than one system user to be in the code acquisition
mode at the same time.
Another possible advantage of using a longer observation window, is due to what
the receiver can do after acquiring code acquisition. Consider what happens when
the receiver only observes one bit interval of information. After code acquisition, the
receiver must update its system timing relative to the received signal, in order to
maximize the system capacity. This means that the filter will have to enter another,
possibly long, training cycle to adjust the weights so that data detection can begin.
In the case when two bit intervals are observed by the receiver, the length of the
second training cycle may be reduced. Once the receiver has an estimate of the
desired userâ€™s propagation delay, the receiver could update its timing and weights
relative to the desired userâ€™s signal at the same time. That is, out of the 2N weights,
the filter could pick out the appropriate bank of consecutive N + 1 weights and then
operate on those weights. This would allow the receiver to decrease its complexity
after code acquisition. Since the weights of the N + 1 length filter are initialized to
those values that are within a chip of being synchronous with the desired user, it is
feasible that the filter can begin data detection quickly and slowly adjust its timing
to get synchronous with the desired user.
CHAPTER 9
SUMMARY AND FUTURE WORK
9.1 Summary
The nearfar problem is an impediment in the development of asynchronous DS
CDMA communication systems. The nearfar problem exists when various system
users are received at very dissimilar power levels. Conventional detectors are sensitive
to the nearfar problem, in that it is possible for the multiaccess interference to comÂ¬
pletely destroy the receiverâ€™s ability to detect a weak user. Several nearfar receivers
exist, but their complexity and the amount of required sideinformation make them
impractical to implement. Those researchers that study the problem of detection in
asynchronous DSCDMA environments usually assume that the receiver operates synÂ¬
chronous to the desired user. Traditionally the biterrorrate is used to determine the
overall system capacity. However, earlier work has shown that the ability to achieve
code synchronization, and not the biterrorrate, will determine the overall system
capacity. This motivates the need for nearfar resistant synchronization methods.
In Chapter 3, a review of several proposed timing estimation algorithms was
presented. It was shown that each of these techniques were of rather high complexity.
The MMSE receiver is wellknown to be nearfar resistant and is receiving considerable
attention in the literature. In Chapter 4, a lowcomplexity timing acquisition method
based on processing the weights of the MMSE receiver was presented. The timing
acquisition algorithm was developed assuming a singleuser system, where the desired
user must transmit an all ones training sequence. In this case, the MMSE receiver
adapts in the mean to a sum of two timeshifted versions of the spreading code of the
desired user. The complexity of the timing estimation algorithm was slightly higher
than that of the conventional correlator (by the complexity of running the adaptive
94
95
filter). An important contribution of Chapter 4 is an overview of why synchronization
of the MMSE receiver is important. While this may seem trivial, enough literature
exists where the costs of operating the MMSE receiver asynchronous to the desired
user are ignored. This omission in the literature may lead researchers to the wrong
conclusion that synchronizing the MMSE receiver is unimportant.
The performance of the acquisition algorithm was characterized in Chapter 5.
For the LMS filter, when the system model of Section 2.2 was used, a set of iterative
equations was developed to describe the statistics of the LMS filter weights as a funcÂ¬
tion of time. This model was used as a basis for characterizing the timing estimator
using two different analytical techniques. Unfortunately, both techniques produced
results that are of limited use. This fact stems from several assumptions used in the
derivation of the analytical methods, as the statistical model for the transient LMS
filter weight statistics has been shown to provide reasonable performance between the
simulated and analytical results. Even though the timing estimation algorithm was
developed for a singleuser system, it was shown to perform well in multiuser enviÂ¬
ronments. When the RLS algorithm was used to adapt the weights of the receiver,
the timing acquisition algorithm was observed to be nearfar resistant.
The performance of the timing estimation algorithm was studied in Chapter 7,
for two nonstationary environments resulting from either frequencysynchronization
errors or flatfading channels. The modified MMSE receiver proposed by Barbosa and
Miller [13] can be used in these scenarios for purposes of timing acquisition. It was
shown that the performance of the timing estimator was dependent on the order of
the linear predictor used in the modified MMSE receiver (up to order 10). However,
the phasecompensation algorithm delays the buildup of energy in the weights of the
adaptive filter. This affects the performance of the timing estimation algorithm by
increasing the length of the training cycle required to achieve acquisition. A hybrid
96
form of the modified MMSE receiver was shown to offset some of this performance
degradation.
Finally, in Chapter 8, a modification to the system model and timing estimation
algorithm was presented such that the required training sequence for the MMSE
receiver doesnâ€™t have to be all ones. The modified timing estimator was compared
to the original timing estimator under a multiuser nearfar environment. There
was no appreciable difference observed between the performance of the two timing
estimators. The advantage of the modified timing estimator is that more than one
systemuser can be in the acquisition mode at the same time.
9.2 Contributions
The main contributions made in this dissertation are:
â€¢ Development of a low complexity singleuser nearfar resistant timing estimator
based on the wellknown MMSE receiver.
â€¢ Clearly demonstrating why synchronizing the MMSE receiver is important.
â€¢ Studying the performance of the estimator in two nonstationary channels and
proposing a hybrid technique based on the modified MMSE receiver.
â€¢ Modifying the timing estimator such that more than one systemuser can be in
the acquisition mode at the same time.
9.3 Future Work
There are several areas of future research related to timing acquisition and synÂ¬
chronization in DSCDMA communication systems. In this dissertation, the weights
of an adaptive filter have been processed to form a coarse estimate of the propagation
delay of the desired user. In Chapter 4, it was shown that the MMSE receiver must
be synchronized with the desired user in order to maximize the system capacity. AfÂ¬
ter the receiver adjusts its timing based on the timing estimate, a separate tracking
97
mode must be used synchronize to the desired userâ€™s signal and maintain this synÂ¬
chronization while the filter is used for data detection. While the timing estimation
algorithm provides coarse acquisition, its usefulness in a tracking mode is possibly
limited. Consider the weights of the adaptive filter when the receiver is in the data
detection mode. A change in the relative timing between the receiver and the desired
user will marginally affect (for small timing errors) the filter output error signal. This
means that the effect of the timing change wonâ€™t appear in the filter weights for many
bit intervals. When the time shift is detected by the timing estimation algorithm,
most likely the timing relationship between the receiver and the desired user has
already changed. To the best of the authorâ€™s knowledge the problem of tracking in
DSCDMA communication systems hasnâ€™t been sufficiently addressed.
Another area that is receiving attention in the literature is that of overlay commuÂ¬
nication systems. In an overlay system, the DSCDMA system shares bandwidth with
a narrowband communication system. While the issue of code acquisition in overÂ¬
lay scenarios has been studied [6062] the emphasis of overlay research has been the
nature of how the narrowband system affects the DSCDMA system performance.
Recent work has studied the effect of the DSCDMA system on the performance of the
narrowband system [63]. It has been shown that the DSCDMA system must place
nulls in its PSD where the narrowband system exists in order for the narrowband
system to coexist with the DSCDMA system. The nulls in PSD of the DSCDMA
system are created by filtering the spreading sequence used by each user. At this time,
it is unclear how this filtering operation will affect the problem of timing acquisition
and tracking.
A final area for future research deals with multipath frequencyselective fading
channels. A recent work by Miller et al. [64] studies the performance of several
different MMSE receiver architectures in frequency selective fading channels. One
proposed architecture is to combine the outputs of a bank of filters (either the modified
98
MMSE [13] or the MOE [29]), one for each received path of the desired userâ€™s signal.
The performance of this architecture is compared to a single MMSE receiver in which
the steering vector is formed with a priori knowledge of the fading parameters and
delays of each path of the desired userâ€™s signal. It is shown, that the single MMSE
receiver offers a significant improvement in BER performance over the multiple filter
architecture in a fast fading environment. Therefore without knowledge of the fading
parameters (for each path), a substantial decrease in the system capacity will occur.
This result motivates the need for nearfar resistant parameter estimation algorithms
for fast fading multipath channels.
APPENDIX A
PROBABILITY OF BIT ERROR USING THE GAUSSIAN APPROXIMATION
In this appendix, we consider the probability of bit error for an MMSE receiver
operating asynchronous to all users. Since we are interested in detection, we now
assume that all users transmit random data sequences. The output of the MMSE
filter, shown in Figure A.l, will be modeled as a complex Gaussian random variable.
Using this model we derive an expression for the probability of bit error.
Now that the desired user is assumed to transmit a Â±1 data sequence, we must
update our system model. The input to the receiver during the mth symbol interval
may now be expressed as
r(m) = <Â¡i(m) [(1  Ã,)T^>(c,) + Ã,tÂ£'+1,(c1)] exp(;#,) + f(m), (A.l)
where f (m) accounts for the intersymbol interference (ISI) from the desired user, the
MAI interference and the complex AWGN as shown below.
f(m) = di(m  1) f(l  +Ã¡,T<"!â€™11)(c1)l exp(j*i)
K
+ '52\ly1 Jfc(m) exP t?0fc) + n â€ž
(A.2)
k=2
lit)
Convert to
baseband
+ chip
matched
filter
r(m)
MMSE
filter, w
H, \
r (m)w
sgn[Re()]
d\{m)
e(m)
dAm)
Figure A.l: The MMSE receiver for the Gaussian approximation.
99
100
In the above equations, the TÂ¿( ) and TÃ±( ) operators are the acyclic left and right
shift operators of an iVelement vector and Jfc(ra) is defined as shown in equation (2.7).
The steering vector for the MMSE receiver is p = E[di(m)r(m)], while the autoÂ¬
correlation matrix is R = E[r(m)r/i(m)]. Using the above equations, these quantities
can be expressed as shown in the following equations.
(1  exp(iÂ«,)] (A.3)
^' p
R = la^i{PkJk)^2ki(PkJk) + aL2k{Pk,Sk)a^k{pk,Sk)] + 2a2l
k=i Â¿n
= ppÃ + E[f(m)fiÃ(m)] (A4)
= ppH + R
The filter output, y(m), at the optimal weights wopt = R xp is given by
y{m) = w"tr(m) = di(m)w"tp + wfptr(m).
(A.5)
We model the term w^tr(m) as a complex Gaussian random variable with a mean
of zero and a variance of a2. We find the variance by evaluating
a2 = E[w*tr(m)rH(m)wopt]
= wJ,tE[f(m)i"(m)]wopt
= w^t [R â€œ PPH] wÂ°Pt
= p^R^p (l  pHR_1p).
(A.6)
Recognizing that Jmin = 1  pHR_1p, we find that
& â€” (1 'AniiO'Ar
(A.7)
and
di(â„¢)wfptp = di(m) (1  Jmin).
(A.8)
101
The filter forms its decision by comparing Re[y(m)] to a threshold of zero. Since
we have modeled the output of the filter as a complex Gaussian random variable, we
see that Re[y(m)] is a Gaussian random variable with a mean of d\(m) (1 â€” Jmin) and
a variance of (1 â€” Jmin)Jmmâ– Assuming that d\(m) is equally likely to be Â±1, the
probability of error is
Pr(error) = Q  ^n)
\ y *^min
where
00
Q(x) = â€”j= f exp(t2/2)dt (A.10)
v27t J
X
and Q(x) is known as the the complementary errorfunction integral [43].
(A9)
APPENDIX B
DERIVATION OF TRANSIENT EQUATIONS FOR THE LMS FILTER
B.l Transient Weight Autocorrelation Matrix Using
the Gaussian Approximation
In this section, a recursive equation for the LMS filter transient weight autocorreÂ¬
lation matrix will be derived. The fundamental assumption, described in Section 5.1,
will be used to make the derivation mathematically tractable. The input to the reÂ¬
ceiver will be assumed to have a known mean, but the elements of the received vector
will be modeled as independent identically distributed Gaussian random variables.
We begin by taking the outer product of equation (5.1).
w(m + l)wT(m + 1) = w(m)wr(m) + nw(m)uT(m)d
â€” //w(m)w7 (m)u(rn)u7 (m) + Â¿m(m)w7 (m)d
â€” /xu(m)ur(m)w(m)w:r(m) + Â¿Â¿2u(ra)ur(ra)d2
(Term #1) â€” /Â¿2u(ra)wr(m)u(m)ur(ra)
â€” /ru(m)u7 (m)w(m)uT(m)d
(Term #2) + //2u(m)ur(m)w(m)wr(m)u(m)ur(m)
(B.l)
In general filtering applications, the statistics of u(ra) are dependent on the value
of m. However, we will assume that these statistics are not timevarying and thereÂ¬
fore drop the dependence on m on this quantity when evaluating the corresponding
expectations. Taking the expectation of the quantities in equation (B.l) is relatively
simple, except for the terms in lines four through six of the equation which require
careful algebraic manipulations. These special terms have been denoted as terms #1
and #2 in equation (B.l) and their corresponding expectations are defined below.
102
103
Ignoring the scalars Â¿u and d, the expectation of term #1 can be written as
N
(E[u(m)wr(m)u(m)ur(m)])^ wk E^u*^]. (B.2)
fc=i
In order to evaluate E[uÂ¿u^Uj] we use the fundamental assumption which asserts
that the inputs to the filter are independent Gaussian distributed random variables.
We define an N element vector z whose elements are independent zeromean Gaussian
random variables each having a variance of a2, where a2 is the variance of the AWGN
at the input to the LMS filter. Using this definition we can express
E[ujUfcUj] = E[(z< + uÂ¿)(zfc + ufc)(z j + Uj)].
(B.3)
After multiplying out all of the terms inside the above expectation, we see that there
are only four terms that contribute to the expectation as shown below.
E[ujU*Uj] = ufc E[zjZj] + uÂ¿ E[zfcZj] + uj E[zÂ¿zfc] + UjUfcUj (B.4)
Due to the assumed statistical independence of the elements of z, each of the expecÂ¬
tation terms in the above equation is easily evaluated as
E[zmzn] = <720K(ra  n)
(B.5)
Substituting equation (B.5) into equation (B.2) and considering all possible comÂ¬
binations of i, k and j we see that term #1, neglecting the scalars n and d, of
equation (B.l) can be written in a convenient form shown below.
Note that due to the symmetry of the above equation we can also immediately state
104
Next, we consider evaluating term #2, which ignoring any scalar terms can be
expressed as
N N
(E[u(m)uT(m)w(m)wr(m)u(m)ur(m)]) = ^ ^ E[wfcwp] E[uÂ¿ufcupUj]. (B.8)
p=1 fc=l
Note that we will use z as before to evaluate E[uÂ¿UfcUpUj]. The result of this substiÂ¬
tution and expansion of terms is listed below.
E[ujUfcUpUj] = E[ziZkzpZj\
+ E[ziZk]Ã¼jÃ¼p
+ E[zÂ¿Zp]Ã¼jÃ¼fc
+ E[z,Zj]Ã¼jfcÃ¼p
+ E[zfcZp]Ã¼Â¿Ã¼j
+ E[zjZjt]Ã¼Â¿Ã¼p
+ E[zjZp]Ã¼jÃ¼fc
+ Ãœj ÃœfcÃœpÃœj
(term #1)
(term #2)
(term #3)
(term #4)
(term #5) (B.9)
(term #6)
(term #7)
(term #8)
Due to the complexity of the above equation, each of the terms and its contribution
to equation (B.8) will be described in full detail.
We begin with the expansion of the first term in equation (B.9). In order to
evaluate this term we must use the Gaussian momentfactoring theorem [43].
E[zjZfcZpZj] = a46K(i  k) Ã³K(j  p) {1  SK(k  p)}
+ a4Ã³K(i  p) SK(j  k) {1  Sx(k  p)}
(B.10)
+ 04Â¿k(Â¿ ~ j) (k â€” p) {1 â€” â€” k)}
+ 3a46K(i  j) Ã³K(i  k) SK(i  p)
105
The contribution of the above equation to equation (B.8) can be expressed as shown
below.
N N
^ Y] E[wAwP] E[zjZfczpZj] = (2cr4Rtt,u,(m) + a4 trace(Rttlu,(m))l). ^ (B.ll)
p=i fc=i
Based on the previous analysis, the contributions of terms #2 and #3 from equaÂ¬
tion (B.9) to equation (B.8) are readily available.
N N
yy yy E[wA.wp] {E^z^Ã¼jÃ¼p + E^ZpJÃ¼jÃ¼fc} = (2cr2Ru,tt,(m)Ã¼Ã¼T). (B.12)
p=i fe=i
We next consider the contribution of term #4 to equation (B.8). Note that this term
only contributes to those elements along the diagonal, that is when i = j.
N N
yy yy E[wA.wp] EfzjZpJÃ¼jÃ¼fc = (a2uTKww(m)ui). . (B.13)
P=1 fc=l
Similarly, term #5 only contributes to those elements when k â€” p, which produces
the following equation.
N N
yy yy E[wfcwp] EfzfcZpJÃ¼jÃ¼., = (a2 trace (Rww(m))mT). . (B.14)
p=1 fc=i
Following the analysis above, the contributions from terms #6 through #8 are readily
listed below.
N N
yy yy E[w*wp] {EfzjZfcjÃ¼iÃ¼p + E[zjzpjÃ¼iÃ¼jt} = (2cr2Ã¼Ã¼TRww(m)). (B.15)
p=i Jt=i
N N
yy yy EfwjtWpJÃ¼jÃ¼fcÃ¼pÃ¼j = (Ã¼Ã¼tru11Ã)(m)Ã¼Ã¼r).. (B.16)
P=1 fe=l
106
We can now combine all of the terms from the appropriate equations to get the
final iterative equation for the transient filter weight autocorrelation matrix.
RuÃ±ara + 1) = Rww(m) + pw(m)uTd  pRww(m)Ruu
+ Â¿iÃ¼wT(m)d  /uRuuRtt,w(m) + fiRuud2
â€” 2fio2 {wr(m)Ã¼l + Ã¼wr(m) + w(m)Ã¼r} d
â€” 2/i2Ã¼w T(m)Ã¼Ã±Td + 2/j,2a2Rww(m)Ã¼Ã±T (B.17)
+ 2^2a4Rww(m) + n2a4 trace(RU)U,(m))I
+ ^Ã2cr2Ã¼TRu,!i)(m)Ã¼I + fio2 trace(Ru,tt,(m))Ã¼Ã¼r
+ 2/i2
Now given equation (B.17), we can easily define the transient filter weight covariance
matrix using the following equation once we define w(0), Râ€ž,ty(0), Ã¼ and Ruu.
Cww(m) = Rww(m)  w(m)wr(m) (B.18)
B.2 Transient Weight Autocorrelation Matrix Using
the Known Statistics of the Receiverâ€™s Input Vector
In the previous section, when we evaluated the terms E[u,u*.Uj] and E[u,UfcUpUj]
we assumed that the individual elements of u were independent. We have essentially
ignored the highly structured nature of the LMS filterâ€™s received vector for our DS
CDMA system model. However, by using this assumption the resulting analysis of the
transient equations for the filter weight vector is extremely simplified. In this section,
we will use the known statistics of r(m) in order to study the transient statistics of
w(m) in more detail. The complexity of the resulting equations will make it painfully
obvious why the independence assumption is routinely used for this problem.
107
Recall that u(m) = r(m), where
r (m) = ai(px, Â¿1) + v(m) + Nm
(B. 19)
k=2
(B.20)
In order to simplify the following expressions, let ax and represent the xth component
of the vector ax(pi, Â¿i). Also let the term vx represent the xth element of the v(m)
vector from the previous equation. Using this notation, and recalling that the data
bits for the interfering users are assumed to be independent and equally likely to be
Â±1, bruteforce algebraic manipulations are used to find the desired quantities.
E[uÂ¿u*;Uj] = aÂ¿a*aÂ¿ + o2a.jÃ³K(i  k) +
+ a2ak6K(i  j) + aÂ¿ E[vfcVj]
+ ak EfvjVj] + aj E[vÂ¿vfc]
(B.21)
We are really interested in substituting the previous equation in to equation (B.2).
In order to simplify the notation for this substitution as well as other following exÂ¬
pressions, let (a2fc_i)a; and (a2A;)x denote the xth component of a2ki(Pk,h) and
a2k{Pk,Sk), respectively.
E[vsvt] = (P)4it
r=lrÂ£ + lr0
(B.22)
(B.23)
where
k=2
(B.24)
(B.25)
108
Using the vectormatrix notation from the previous section, the following equation
results.
E[u(m)wr(m)u(m)ur(m)] = E[u(m)u7 (m)w(m)ur(m)]
= cr2wr(m)a1 (pi, Â¿i)I + a2ai(pi, Â¿i)wr(m)
+ ai(Pi,^i)wr(m)ai(p1,Â¿i)af(pi,Ã¡1) (B.26)
+ <72w(m)a[(pi,Ã¡i) + wr(m)ai(pi,Â¿i)r
+ rw(m)af(p1,Ã¡1) + a1(p1,Ã¡1)wr(m)r
In order to evaluate E[uÂ¿UjtUpUj], we will expand this expression as a sum of three
functions.
E[uÂ¿UfcUpUj] = A(ui, uk, Up, uj) + B{uÃ, ukl up, u,) + C(uit uk, up, Uj) (B.27)
Each of these functions will be expanded in the following equations.
We begin with expanding the j4(uÂ¿, Ujt, up, uÂ¿) term.
A(uj, ufc, Up, Uj) = a2a.pa.jSK(i  k) + apaj E[vÂ¿vfc]
I a2akajÃ³K(i  p) + akaÃ³ E[vÂ¿vp]
+ a2aiajÃ³K(k p) + a,a_, E[vfcvp]
+ aÂ¿a;.apaj
(B.28)
Substituting this term into equation (B.8), we find
N N
^ ^ E[w;.Wp]i4(uÂ¿, uk, Up, Uj) = (2ÃT2Rti)U)(m)a1(p1,Â¿i)a[(p1,Ã¡1)).J
p=1 k1
+ (o2 trace (Rwu;(m))ai(pi, Â¿i)a[(pi,Si)). .
+ (2rRu,ttl(m)a1(pi, Â¿i)a[ (pi, Â¿i)) {j
+ (trace (Rtt)It;(m)r)ai(pi, Â¿i)a[(px, Ã¡1))i .
+ (ai(px, Ã³i)a[ (p i, Ã¡1)RU)U,(m)a1(p1, Â¿i)af (pi, Â¿i)) tJ
(B.29)
109
Next, we consider the B(uÂ¿, u*, up, Uj) term and its contribution to the transient
correlation matrix.
B{ui, uk, Up, Uj) = cr4SK(i  k) 6K(j  p) {1  SK(k  p)}
+ a46K(i  p) Ã³K{j k){ 1  SK(k  p)}
+ a4Ã³K{i  j) Ã³K(k p){ 1  5k(*  *0}
+ 3a4dK(i  j) <5k(*  k) SK(i  p) (B.30)
+ <72afcapÂ¿K(Ã  j) + E[vfcVp]Ã³K(Ã  j)
+ a^AiApSKik  j) +
+ ^2aiajtÂ¿K(p  j) + a2 E[vjVfc]Ã¡K(p  j)
Substitution of this equation into equation (B.8) leads to the following equation.
N N
^ ^2 E[wa.wp]B(uÂ¿, Ufc, Up, Uj) = (2a4Rww(m) + o4 trace (R^m))!)
p=i fc=i
+ (
+ (2
+ (cr2 trace (Ru,u,(m)r)l).
+ (2iT2rR11Jtl)(m)),j.
Finally, the last term is considered.
C(uÂ¿, u*, Up, Uj) = aÂ¿afc EfvpVj] + a2 E[vfcVj]$K(t  p)
+ afcap E[vjVj] + a2 E[viVj]5K(fc  p)
+ a;ap E[vfcVj] + a2 E[vpVj]<5K(i  k)
+ E[vjVfcVpVj]
(B.32)
110
In order to evaluate the above expression, the following auxiliary equation is required.
k 02
E[vsVtVuVx] = ^2 [(a2fcl)s(a2fcl)i(a2fcl)u(a2fcl)i]
â€”9 1
k=2
K
Pi
(B.33)
^2 [(a2fc)s(a2fc)Ã(a2fc)u(a2fc)I]
k=2 Z 1
Substitution of equation (B.32) into equation (B.8) leads to the following equation.
N N
^^E[wfcwp]C'(uÂ¿,ufc,up,uj) = (2cr2Rttni,(ra)r) _
pâ€”1 k=1
+ (ai(pi, Â¿i)af (pi, Â¿i)RUJU,(m)r)
+ (cr2 trace (Hww (m))r)
+ (a[(p1,<5i)Ru,[U(m)ai(pi,Â¿i)r).J
+ x (rÂ£Râ€ž(m)rÂ£ + r0RU)it,(m)ro)jj
(B.34)
Ill
We now combine all of the appropriate terms to get the final expression for the
transient filter weight autocorrelation matrix.
R.wwipi + 1) = + pw(m)aj (pi,6x)d pRww(m)Ruu
+ pax(px, Sx)wT(m)d  pRuuRww{m) + p2Ruud2
â€” 2//2cr2w:r(m)ai(pi, <5i)Id â€” 2p2a2ax(px,6x)'wT(m)d
â€” 2p2ax (pi, Â¿i)wr(m)ai (px, Â¿1 )af (px,Sx)d
â€” 2p2cr2w(ra)af (px, 6x)d â€” 2p2wT (m)ax(px,
â€” 2p2rw(m)ai(pi, 6x)d â€” 2p2ax(px, Ã³x)wT(m)Td
+ 2p2cr2Rlt,u;(m)ai(px, Â¿i)af (px, Â¿i)
+ p2a2 trace (RU)it,(m))ai(pi, Sx)ax (px, Sx)
+ 2p2rRum(m)ai(p1,Ã¡i)a[(pi,Ã¡i)
(B.35)
+ p2 trace (Rww(m)r)ax (px, Ã³x)a[(px, Sx)
+ U2ai(Pi, (pi,Ã³i)Rww(m)ax(px, Ã³x)a[(px,Ã³x)
+ 2p2a4Rww(m) + p2a4 trace (Runâ€ž(m))I
+ p2a2af (pi, Sx)Rww(m)ax(px, Â¿i)I
+ 2pVa1(p1, Ã³x)a[(px, Ã³x)Rww(m)
+ p2o2 trace (RÃ(,tÃ¼(m)r)I + 2p2cr2rRlt,u,(ra)
+ 2/i2(j2RU)jt;(m)r + p2ax (px, Sx)a[(px, Ã³x)Rww(m)r
+ tfo1 trace (Râ€ž,â€ž,(m))r + Â¿f2ai(p!, Ã¡1)Ru,u;(m)af (plt Â¿i)r
2 2
+ ^rÂ£Ru,u,(m)rÂ£; + yr0ROT(m)r0
APPENDIX C
REVIEW OF QUADRATIC FORMS OF GAUSSIAN RANDOM VARIABLES
Let z be an nelement column vector, whose elements are jointly Gaussian random
variables with a mean of z and a symmetric positive definite covariance matrix of V.
Let A be an nbyn symmetric matrix. Using these definitions, the quantity zrAz is
known as a quadratic form of Gaussian random variables. The following paragraphs
explain in detail how the characteristic function for a quadratic form of Gaussian
random variables can be derived.
We are interested in evaluating Pr(zTAz < y), where y is a given threshold. Using
the known joint probability density function of z this probability can be expressed as
Pv(zTAz
where dz is used to denote dz\dz2 ... dzn and the ndimensional integral is integrated
over the region D â€” {R" : Z7AZ < y}. We can simplify the above expression by
defining the change of variables m = z â€” z, in which case the previous expression
becomes
Pr(zrAz
where the integration region is now D = {Rn : (M  z)rA (M z) < y}
Note the covariance matrix V is positive definite and can therefore be expressed
using a Cholesky decomposition. This means that
V = LLr, (C.3)
112
113
where L is a nonsingular lower triangular matrix. The following expressions are
based on simple algebraic manipulations of the previous equation.
V = LLrl = L2 (C.4)
V1 = (Ll)rL1 (C.5)
Let x = L 1m. Using this change of variables the joint probability density function
of x is then
/x(x) =
/m(hi)
IL1!
(C.6)
m=Lx
Using this change of variables, equation (C.2) can be expressed as
Pr(zTAz < â€ž) = /"7(Â¡Ãj^Â¡LH.â€œP (5
(C.7)
where the region of integration is
D = r" : (L(X  L_1z))rAL(X  L1z) < y}
= {i" : (X  L_1z)T(LrAL)(X  L1z) < y} .
We can further simplify equation (C.7) by expressing
LtAL = PAPr
(C.8)
(C.9)
where P is a matrix of orthonormal eigenvalues and A is a diagonal matrix of the
corresponding eigenvalues. Using the above equation we can also state that
A = PrLrALP.
(C.10)
114
Now define a further change of variables, w = Px. Using this change of variables,
equation (C.7) can be expressed as
Pr(zrAz < y) = j â– â– â– J * exp ^l(Prw)TPTw j dw
D ' '
/"/Â¿^exp(4wTw)â€˜ÃW
D '
(C.ll)
where the region of integration is
D = ]Rn : (P(W  PrL_1z))T(LTAL)P(W  P^L^z) < y}
= {ir : (W  PrL_1z)7 A(W  PtL~1z) < y} .
(C.12)
Using equations (C.ll) and (C.12) we see that in order to calculate Pr(zrAz < y),
we could concentrate on the distribution of
n
Y = YJK{Wi hi)2 (C.13)
Â¿=i
where the Wi are independent Gaussian random variables of zero mean and unit
variance, is the Â¿th eigenvalue of LrAL and 6, is the *th component of PrL_1z.
However instead of deriving the distribution of Y, we instead concentrate on its
moment generating function. Let the moment generating function of Y be defined as
H(s) = E[exp($y)]. Then we can express the moment generating function as
H(s) â€” E[exp(sy)] = E
exp ( Xi(Wi  biY
Â¿=i
(C.14)
Due to the statistical independence of the !UÂ¿, the above expression can be simplified
to
if(s) = nE[exp(sAi(W'ji.i)2)]
1=1
(C.15)
115
We now concentrate on evaluating each term in the previous equation. We start with
the following equation.
OO
E [exp (sAÂ¿(WÂ¿ â€” 6Â¿)2)] = Ã â€”j= exp(s\i(w â€” bi)2) exp(â€”w2/2) dw (C.
J y2ir
16)
In order to simplify the above integrand, expand the terms in the exponent and use
the change of variables x = wy/1 â€” 2AÂ¿s. After performing these operations, the
above equation becomes
E [exp (s\i(Wi  bi)2)] = exp(sAÂ¿6t2)
IWneXP{
2sXu x (C.17)
11 : x ) exp(â€”x2/2) dx.
vi  2AiS
In order to simplify the above integrand, complete the square of the terms in the
exponent. The result of this operation is
o 1 / 9
E [exp (*W  *.)â€™)] = p (j^j) exp(SA,6>)
j7^eXP{J2
â€”  hr +
2s\ibÃ
Vl ~ 2AjS
(C.18)
dx.
The integral in the above expression is recognized as the probability density function
of a Gaussian random variable. Therefore the integral evaluates to one and the above
expression can be written as
E [exp (sXm  W*)] = ^=4= exp (Â£g) exp(^)
1,
exp â€” b\ exp
1 52
2 1  2A, s
(C.19)
Vl  2AiS
Direct substitution of equation (C.19) into equation (C.15) produces the moment
generating function of Y.
2A,;S
(C.20)
116
As the name implies, the moment generating function of a random variable may
be used to evaluate the moments of the random variable as shown below.
mn = E[Yn]
dnH{s)
dsn
5=0
(C.21)
Using the two previous equations, one can show that the first two moments of Y
are given by the following equations.
n
+ (C.22)
t=l
m2 = Â¿ Kb] Â¿ AÂ¿(2 + b?)
1=1
n
i=1
n n
+yi +462)+xk
(C.23)
1=1
Â¿=1 fc=l
k^i
These two moments can be used to find the variance of Y since Var(Y) = m2 â€” raf.
Next, we consider how a Chernoff bound can be used to find a limit on Pr(Y < a).
Note that the desired probability can be found by evaluating
a
Pr(Y
(C.24)
where fY(y) is the probability density function of Y. Let U(x) be the unitstep
function defined as
1 if x > 0,
0 otherwise.
We can then express equation (C.24) as
a oo
Pr(K
â€”OO â€” oo
(C.25)
(C.26)
117
Next we can find an upper limit on the desired probability using the fact that
oo oo
J U(a â€” y)fY{y) dy < J exp{s{y  a))fY{y) dy real(s) < 0. (C.27)
â€” OO â€”OO
We recognize the right hand side of the above equation as an exponential times the
previously derived moment generating function. Therefore, the upper limit on the
desired probability is
Pr(y < a) < exp(â€”as)H(s) real(s) < 0. (C.28)
At this point, a couple of comments are in order. We initially stated that the
transformation matrix A was symmetric. However if the matrix is not symmetric, it
can easily be transformed into a symmetric matrix. Consider the general quadratic
form
X = zrAz. (C.29)
Note that the previous equation may also be expressed as
n n
X = y ] zkzj(A)k,jâ€¢ (C.30)
k=1 j=1
If we consider the terms where k ^ j, the contribution to X for each of these terms
is then ZkZj[(A)kj + (A)^*.]. We can therefore transform A into a new symmetric
matrix B where B = [A + Ar]/2 and X = zrBz.
When we consider the general quadratic form X â€” zrAz, we see that if A is
positive definite then by definition X will always be greater than zero. There are
several necessary and sufficient conditions for a real symmetric matrix to be positive
definite. One of these conditions is that all of the eigenvalues of the matrix must be
positive. Without actually computing the eigenvalues of a matrix, we can test for the
possible existence of negative eigenvalues. It is known that the sum of the eigenvalues
equals the trace of the matrix, where the trace is the sum of all of the elements along
118
the main diagonal. Therefore, if the trace of a real symmetric matrix is less than or
equal to zero, then the matrix is not positive definite. Only when the transformation
matrix A is not positive definite is it possible for the Pr(zTAz < 0) to be nonzero.
APPENDIX D
TYPOGRAPHICAL NOTES
This dissertation was typed and edited using GNU Emacs 19.34.1 with the
AUCTeX 9.81 and RefTeX 3.17 programs on a NeXTstationÂ®1 Turbo workstation.
These programs greatly facilitate the creation of errorfree DT^X 2e source code, and
are highly recommended by the author. The dissertation was typeset using DI^X 2e
version <1998/06/01>. The author extends his gratitude to those individuals who
have created these programs and have made them freely available to the masses.
The Graduate School at the University of Florida has several typesetting requireÂ¬
ments that are traditionally considered inappropriate by the general DT]eX 2e commuÂ¬
nity. The specific requirements deal with linespacing, underlining section headings
and general page layout. In order to meet these requirements, the author spent quite
a bit of time creating a DIkX2Â£ class file called ufthesis.CLS that modifies the
standard report class to meet the page layout guidelines and facilitates switching the
linespacing as required. Kopka and Daly [65], Goosens et al. [66], as well as the
classes.dtx and ltsect.dtx files (part of the standard M^X2e distribution) were very
useful in creating the class file. In addition, the DTEX2e packages ulem.sty (version
1997/04/21) by Donald Arsenau and SETSPACE.STY (version 6.4 1998/11/26) by GeÂ¬
offrey Tobin and SECTSTY.STY (version vl.8.4 1998/12/11) by Rowland McDonnell
were used to format section headings and adjust linespacing. These packages are
readily available from your nearest Comprehensive TfÃX Archive Network (CTAN)
site [66].
1 NeXT is a registered trademark of NeXT Software Inc.
119
120
The document class file is currently available via the World Wide Web at http: //
barney. ee. uf 1. edu/~rsmith/LaTeX2e or http: //www. math. uf 1. edu/~nring, where
information regarding the package and its use are available. The package is distributed
with integrated documentation, where the description and the coding are merged into
a single source file called ufthesis.dtx. The DocStrip utility is used to separate the
ufthesis.CLS document class file, while the doc package is used to typeset the docÂ¬
umentation. This packaging aids in the maintenance and revision of UTeX2Â£ source
coding. In addition, by providing documentation, it is hoped that other RTeX2Â£
users will be able to use this document class file and modify it to suit their needs.
For the near future, the author can be reached via email at smith_ron@usa.net and,
time permitting, is willing to address questions regarding the use of this document
class. Please contact the author if you cannot find the document class package at the
above URLs, as the author may have moved this information to a more permanent
website.
REFERENCES
[1] D. R. Hughes and D. Hendricks, â€œSpreadspectrum radio,â€ Scientific American,
vol. 278, no. 4, pp. 9496, Apr. 1998.
[2] R. A. Scholtz, â€œThe spread spectrum concept,â€ IEEE Transactions on CommuÂ¬
nications, vol. 8, no. COM25, pp. 748755, Aug. 1977.
[3] R. L. Pickholtz, D. L. Schilling, and L. B. Milstein, â€œTheory of spreadspectrum
communicationsâ€”a tutorial,â€ IEEE Transactions on Communications, vol.
COM30, no. 5, pp. 855884, May 1982.
[4] R. E. Ziemer and R. L. Peterson, Digital Communications and Spread Spectrum
Systems, Macmillan, 1985.
[5] D. V. Sarwate and M. B. Pursley, â€œCrosscorrelation properties of psuedorandom
and related sequences,â€ Proceedings of the IEEE, vol. 68, no. 5, pp. 593619,
May 1980.
[6] E. H. Dinan and B. Jabbari, â€œSpreading codes for direct sequence CDMA and
wideband CDMA networks,â€ IEEE Communications Magazine, vol. 36, no. 9,
pp. 4854, Sept. 1998.
[7] H. Fukumasa, R. Kohno, and H. Imai, â€œDesign of psuedonoise sequences with
good odd and even correlation properties for DS/CDMA,â€ IEEE Journal on
Selected Areas in Communications, vol. 12, no. 5, pp. 828836, June 1994.
[8] T. K. Moon, R. T. Short, and C. K. Rushforth, â€œAverage acquisition time for
SSMA channels,â€ in Proceedings IEEE Military Communications Conference,
McLean, VA, 1991, vol. 3, pp. 10421046.
[9] T. K. Moon, R. T. Short, and C. K. Rushforth, â€œA RASE approach to acquisition
in SSMA channels,â€ in Proceedings IEEE Military Communications Conference,
McLean, VA, 1991, vol. 1, pp. 10371041.
[10] R. T. Barghouthi and G. L. Stiiber, â€œSequence acquisition techniques for CDMA
systems employing Kasami sequences,â€ in Proceedings IEEE Military CommuÂ¬
nications Conference, San Diego, CA, 1992, vol. 2, pp. 633637.
[11] R. T. Barghouthi and G. L. Stiiber, â€œRapid sequence acquisition for DS/CDMA
systems employing Kasami sequences,â€ IEEE Transactions on Communications,
vol. 42, no. 2/3/4, pp. 19571968, Feb./Mar./Apr. 1994.
121
122
[12] U. Madhow and M. B. Pursley, â€œAcquisition based capacity of directsequence
spreadspectrum communication networks,â€ in Proceedings of the Conference
of Information Sciences and Systems, The John Hopkins University, Baltimore,
MD, Mar. 1991.
[13] A. N. Barbosa and S. L. Miller, â€œAdaptive detection of DS/CDMA signals in
fading channels,â€ IEEE Transactions on Communications, vol. 46, no. 1, pp.
115124, Jan. 1998.
[14] L. W. Couch II, Digital and Analog Communication Systems, PrenticeHall, 5th
edition, 1997.
[15] S. L. Miller, â€œAn adaptive directsequence codedivision multipleaccess receiver
for multiuser interference rejection,â€ IEEE Transactions on Communications,
vol. 43, no. 2/3/4, pp. 17461755, Feb./Mar./Apr. 1995.
[16] S. VerdÃº, â€œMinimum probability of error for asynchronous Gaussian multiple
access channels,â€ IEEE Transactions on Information Theory, vol. 32, no. 1, pp.
8596, Jan. 1986.
[17] S. VerdÃº, â€œOptimum multiuser asymptotic efficiency,â€ IEEE Transactions on
Communications, vol. COM34, no. 9, pp. 890897, Sept. 1986.
[18] R. Lupas and S. VerdÃº, â€œLinear multiuser detectors for synchronous codeÂ¬
division multipleaccess channels,â€ IEEE Transactions on Information Theory,
vol. 35, no. 1, pp. 123136, Jan. 1989.
[19] R. Lupas and S. VerdÃº, â€œNearfar resistance of multiuser detectors in asynÂ¬
chronous channels,â€ IEEE Transactions on Communications, vol. 38, no. 4, pp.
496508, Apr. 1990.
[20] S. Parkvall, E. Strom, and B. Ottersten, â€œThe impact of timing errors on the
performance of linear DSCDMA receivers,â€ IEEE Journal on Selected Areas in
Communications, vol. 14, no. 8, pp. 16601667, Oct. 1996.
[21] F.C. Zheng and S. K. Barton, â€œOn the performance of nearfar resistant CDMA
detectors in the presence of synchronization errors,â€ IEEE Transactions on
Communications, vol. 43, no. 12, pp. 30373045, Dec. 1995.
[22] M. K. Varanasi and B. Aazhang, â€œMultistage detection in asynchronous codeÂ¬
division multipleaccess communications,â€ IEEE Transactions on CommunicaÂ¬
tions, vol. 38, no. 4, pp. 509519, Apr. 1990.
[23] E. G. Strom, S. Parkvall, S. L. Miller, and B. E. Ottersten, â€œPropagation delay
estimation in asynchronous directsequence codedivision multiple access sysÂ¬
tems," IEEE Transactions on Communications, vol. 44, no. 1, pp. 8493, Jan.
1996.
123
[24] S. E. Bensley and B. Aazhang, â€œSubspacebased channel estimation for code
division multiple access communication systems,â€ IEEE Transactions on ComÂ¬
munications, vol. 44, no. 8, pp. 10091020, Aug. 1996.
[25] E. G. StrÃ³m, S. Parkvall, S. L. Miller, and B. E. Ottersten, â€œDSCDMA synÂ¬
chronization in timevarying fading channels,â€ IEEE Journal on Selected Areas
in Communications, vol. 14, no. 8, pp. 16361642, Oct. 1996.
[26] S. E. Bensley and B. Aazhang, â€œMaximumLikelihood synchronization of a single
user for codedivision multipleaccess communication systems,â€ IEEE TransacÂ¬
tions on Communications, vol. 46, no. 3, pp. 392399, Mar. 1998.
[27] D. Zheng, J. Li, S. L. Miller, and E. G. Strom, â€œAn efficient codetiming estimator
for DSCDMA signals,â€ IEEE Transactions on Signal Processing, vol. 45, no. 1,
pp. 8289, Jan. 1997.
[28] Z.S. Liu, J. Li, and S. L. Miller, â€œAn efficient codetiming estimator for receiver
diversity DSCDMA systems,â€ IEEE Transactions on Communications, vol. 46,
no. 6, pp. 826835, June 1998.
[29] M. L. Honig, U. Madhow, and S. VerdÃº, â€œBlind adaptive multiuser detection,â€
IEEE Transactions on Information Theory, vol. 41, no. 4, pp. 944960, July
1995.
[30] U. Madhow, â€œBlind adaptive interference suppression for the nearfar resistant
acquisition and demodulation of directsequence CDMA signals,â€ IEEE TransÂ¬
actions on Signal Processing, vol. 45, no. 1, pp. 124136, Jan. 1997.
[31] S. Haykin, Adaptive Filter Theory, PrenticeHall, 2nd edition, 1991.
[32] S. L. Miller, â€œTransient behavior of the minimum meansquared error receiver
for directsequence codedivision multipleaccess systems,â€ in Proceedings IEEE
Military Communications Conference, Fort Monmouth, NJ, 1994, vol. 1, pp.
5256.
[33] S. L. Miller, â€œTraining analysis of adaptive interference suppression for direct
sequence codedivision multipleaccess systems,â€ IEEE Transactions on ComÂ¬
munications, vol. 44, no. 4, pp. 488495, Apr. 1996.
[34] U. Madhow, â€œAdaptive interference suppression for joint acquisition and deÂ¬
modulation of directsequence CDMA signals,â€ in Proceedings IEEE Military
Communications Conference, San Diego, CA, 1995, vol. 3, pp. 12001204.
[35] H. V. Poor and S. VerdÃº, â€œProbability of error in MMSE multiuser detection,â€
IEEE Transactions on Information Theory, vol. 43, no. 3, pp. 858871, May
1997.
124
[36] E. G. Strom, DirectSequence CodeDivision MultipleAccess Systems: NearFar
Resistant Parameter Estimation and Detection, Ph.D. dissertation, University
of Florida, 1994.
[37] K. D. Senne, Adaptive Linear DiscreteTime Estimation, Ph.D. dissertation,
Stanford University, June 1968.
[38] L. I. Horowitz and K. D. Senne, â€œPerformance advantage of complex LMS
for controlling narrowband adaptive arrays,â€ IEEE Transactions on Acoustics,
Speech and Signal Processing, vol. ASSP29, pp. 722736, June 1981.
[39] B. Fisher and N. J. Bershad, â€œThe complex LMS adaptive algorithmâ€”transient
weight mean and covariance with applications to the ALE,â€ IEEE Transactions
on Acoustics, Speech and Signal Processing, vol. ASSP31, no. 1, pp. 3444, Jan.
1983.
[40] A. Fueur and E. Weinstein, â€œConvergence analysis of LMS filters with uncorÂ¬
related data,â€ IEEE Transactions on Acoustics, Speech and Signal Processing,
vol. ASSP33, pp. 222230, Feb. 1985.
[41] S. C. Douglas and T. H.Y. Meng, â€œExact expectation analysis of the LMS adapÂ¬
tive filter without the independence assumption,â€ in Proceedings IEEE InterÂ¬
national Conference on Acoustics, Speech and Signal Processing, San Francisco,
CA, 1992, vol. 4, pp. 6164.
[42] S. C. Douglas, â€œExact expectation analysis of the LMS adaptive filter for corÂ¬
related Gaussian input data,â€ in Proceedings IEEE International Conference
on Acoustics, Speech and Signal Processing, Minneapolis, MN, 1993, vol. 3, pp.
519522.
[43] C. W. Helstrom, Probability and Stochastic Processes for Engineers, Macmillan
Publishing Company, 2nd edition, 1991.
[44] H. ScheffÃ©, The Analysis of Variance, John Wiley & Sons, Inc., 1959.
[45] N. L. Johnson and S. Kotz, Continuous Univariate Distributions, Houghton
Mifflin, 1970.
[46] S. O. Rice, â€œDistribution of quadratic forms in normal random variablesâ€”
evaluation by numerical integration,â€ SIAM J. SCI. STAT. COMPUT., vol.
1, no. 4, pp. 438448, Dec. 1980.
[47] J. Sheil and I. O'Muircheartaigh, â€œThe distribution of nonnegative quadratic
forms in normal variables,â€ Applied Statistics, vol. 26, no. 1, pp. 9298, 1977.
[48] J. P. Imhof, â€œComputing the distribution of quadratic forms in normal variables,â€
Biometrika, vol. 48, no. 3, pp. 419426, 1961.
125
[49] R. B. Davies, â€œNumerical inversion of a characteristic function,â€ Biometrika,
vol. 60, no. 2, pp. 415417, 1973.
[50] R. B. Davies, â€œThe distribution of a linear combination of random variables,â€
Applied Statistics, vol. 29, no. 3, pp. 323333, 1980.
[51] G. Strang, Linear Algebra and its Applications, Harcourt Brace Jovanovich,
Inc., 3rd edition, 1988.
[52] H. L. van Trees, Detection, Estimation, and Modulation Theory: Part I, John
Wiley & Sons, Inc., 1968.
[53] H. V. Poor, An Introduction to Signal Detection and Estimation, Springer
Verlag, 2nd edition, 1994.
[54] T. S. Rappaport, Wireless Communications: Principles and Practice, Prentice
Hall, 1996.
[55] J. G. Proakis, Digital Communications, McGrawHill, Inc., 2nd edition, 1989.
[56] B. Sklar, â€œRayleigh fading channels in mobile digital communication systems
Part I: Characterization,â€ IEEE Communications Magazine, vol. 35, no. 7, pp.
90100, July 1997.
[57] B. Sklar, â€œRayleigh fading channels in mobile digital communication systems
Part II: Mitigation,â€ IEEE Communications Magazine, vol. 35, no. 7, pp. 102â€”
109, July 1997.
[58] R. H. Clarke, â€œA statistical theory of mobileradio reception,â€ The Bell System
Technical Journal, vol. 47, no. 6, pp. 9571000, JulyAug. 1968.
[59] U. Madhow, â€œMMSE interference suppression for timing acquisition and demodÂ¬
ulation in directsequence CDMA systems,â€ IEEE Transactions on CommuniÂ¬
cations, vol. 46, no. 8, pp. 10651075, Aug. 1998.
[60] S. L. Miller, â€œPerformance of code acquisition for DSCDMA overlay,â€ in
Proceedings IEEE Military Communications Conference, San Diego, CA, 1995,
vol. 2, pp. 626630.
[61] L. D. Gottesman, Interference Rejection Techniques to Aid Coarse Acquisition in
DirectSequence SpreadSpectrum, Ph.D. dissertation, University of California,
San Diego, 1993.
[62] L. D. Gottesman and L. B. Milstein, â€œThe coarse acquisition performance of a
CDMA overlay system,â€ IEEE Journal on Selected Areas in Communications,
vol. 14, no. 8, pp. 16271635, Oct. 1996.
[63] B. J. Rainbolt, DirectSequence CodeDivision MultipleAccess Overlay Systems,
Ph.D. dissertation, University of Florida, Dec. 1998.
126
[64] S. L. Miller, M. L. Honig, M. Shensa, and L. B. Milstein, â€œPerformance analÂ¬
ysis of MMSE receivers for DSCDMA in frequency selective fading channels,â€
submitted to IEEE Transactions on Communications, 1998.
[65] H. Kopka and P. W. Daly, A Guide to $TeX2s Document Preparation for
Beginners and Advanced Users, AddisonWesley, 2nd edition, 1995.
[66] M. Goosens, F. Mittlebach, and A. Samarin, The BTpfX Companion, Addison
Wesley, 2nd edition, 1994.
BIOGRAPHICAL SKETCH
Ronald F. Smith was born in Pittsburgh, Pennsylvania in 1963. He received his
Bachelor of Science degree from the University of Pittsburgh in 1985. From 1985
through 1989, Ronald was employed by the Government Systems Sector of the Harris
Corporation, specializing in the design of highspeed digital design circuits for comÂ¬
munication systems. In the fall of 1989, Ronald entered the graduate program at the
University of Florida with the Department of Electrical and Computer Engineering.
In May 1992, Ronald received the Master of Engineering degree. His interests include
the design and analysis of communication systems for spread spectrum systems.
127
I certify that I have read this study and that in my opinion it conforms to acceptÂ¬
able standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
Sc JrAA .
Scott L. Miller, Chairman
Associate Professor of Electrical and
Computer Engineering
I certify that I have read this study and that in my opinion it conforms to acceptÂ¬
able standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy. â€ž
fa. iU
Leon W. Couch II, Cochair
Professor of Electrical and Computer
Engineering
I certify that I have read this study and that in my opinion it conforms to acceptÂ¬
able standards of scholarly presentation and is fullyjjdeq&ate, in scope and quality,
as a dissertation for the degree of Doctor of ]
and Computer
I certify that I have read this study and that in my opinion it conforms to acceptÂ¬
able standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
Robert M. Fox
Associate Professor of Electrical and
Computer Engineering
I certify that I have read this study and that in my opinion it conforms to acceptÂ¬
able standards of scholarly presentation and is fully adequate, in scope and quality,
as a dissertation for the degree of Doctor of Philosophy.
Richard Newman
Assistant Professor of Computer and
Information Science and Engineering
This dissertation was submitted to the Graduate Faculty of the College of EnÂ¬
gineering and to the Graduate School and was accepted as partial fulfillment of the
requirements for the degree of Doctor of Philosophy.
May 1999
Winfred M. Phillips
Dean, College of Engineering
M. J. Ohanian
Dean, Graduate School
LD
1780
199 a
.s\*sr
UNIVERSITY OF FLORIDA
l m ni mu in" 
3 1262 08555 3484
PAGE 1
&2'( 7,0,1* (67,0$7,21 ,1 ',5(&76(48(1&( &2'(',9,6,21 08/7,3/($&&(66 &20081,&$7,21 6<67(06 %\ 521$/' ) 60,7+ $ ',66(57$7,21 35(6(17(' 72 7+( *5$'8$7( 6&+22/ 2) 7+( 81,9(56,7< 2) )/25,'$ ,1 3$57,$/ )8/),//0(17 2) 7+( 5(48,5(0(176 )25 7+( '(*5(( 2) '2&725 2) 3+,/2623+< 81,9(56,7< 2) )/25,'$
PAGE 2
, KDYH EHHQ EOHVVHG ZLWK D ZRQGHUIXO IDPLO\ DQG WKLV ZRUN LV GHGLFDWHG WR WKHP 7R $P\ IRU DOO RI KHU VXSSRUW HQFRXUDJHPHQW DQG ORYH 7R 6DUDK IRU VR UHDGLO\ VKDULQJ KHU VLOO\ OLWWOH JLJJOHV DQG ELJ KXJV DQG NLVVHV
PAGE 3
$&.12:/('*0(176 ZRXOG OLNH WR H[SUHVV P\ JUDWLWXGH WR P\ DGYLVHU 'U 6FRWW / 0LOOHU IRU KLV JXLGDQFH DQG IHHGEDFN WKURXJKRXW WKH FRXUVH RI P\ UHVHDUFK DOVR WKDQN WKH PHPn EHUV RI P\ 3K' FRPPLWWHH 'UV /HRQ &RXFK ,, RVH 3ULQFLSH 5REHUW )R[ DQG 5LFKDUG 1HZPDQ IRU WKHLU LQWHUHVW DQG ZLOOLQJQHVV WR EH D SDUW RI WKLV HIIRUW :KLOH KDYH GHGLFDWHG WKLV ZRUN WR P\ ZLIH $P\ VKH DOVR GHVHUYHV VSHFLDO UHFRJQLWLRQ 6KH KDV ZRUNHG H[WUHPHO\ KDUG VXSSRUWLQJ RXU IDPLO\ ZKLOH SXUVXHG P\ JUDGXDWH VWXGLHV DQG ZLWKRXW KHU HIIRUWV WKLV ZRUN ZRXOGQfW KDYH EHHQ SRVVLEOH WKDQN P\ SDUHQWV (UQHVW DQG R\FH 6PLWK DQG P\ LQODZV 7KRPDV DQG 3HJJ\ 0HQHVNLH IRU WKHLU FRQWLQXRXV HQFRXUDJHPHQW DQG VXSSRUW WKURXJKRXW WKLV HIIRUW DOVR DFNQRZOHGJH P\ IULHQGV DQG FROOHDJXHV RKQ 0LOOHU %UDG 5DLQEROW DQG $OL $OPXWDLUL ZKR KDYH PDGH P\ WLPH DW WKH 8QLYHUVLW\ RI )ORULGD DQ LQWHUHVWLQJ H[SHn ULHQFH LLL
PAGE 4
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
PAGE 5
6XPPDU\ 6,08/$7,1* )$',1* &+$11(/6 &2'( $&48,6,7,21 ,1 12167$7,21 $5< (19,5210(176 )UHTXHQF\ 6\QFKURQL]DWLRQ (UURUV 3HUIRUPDQFH LQ )ODW)DGLQJ &KDQQHOV 6XPPDU\ ,1&5($6(' :,1'2: 2) 2%6(59$7,21 $1' 7,0,1* $&48,6,7,21 8SGDWHG 6\VWHP 0RGHO 8SGDWHG 7LPLQJ (VWLPDWLRQ $OJRULWKP 6XPPDU\ 6800$5< $1' )8785( :25. 6XPPDU\ &RQWULEXWLRQV )XWXUH :RUN $33(1',&(6 $ 352%$%,/,7< 2) %,7 (5525 86,1* 7+( *$866,$1 $3352;,0$7,21 % '(5,9$7,21 2) 75$16,(17 (48$7,216 )25 7+( /06 ),/7(5 %O 7UDQVLHQW :HLJKW $XWRFRUUHODWLRQ 0DWUL[ 8VLQJ WKH *DXVVLDQ $SSUR[LPDWLRQ % 7UDQVLHQW :HLJKW $XWRFRUUHODWLRQ 0DWUL[ 8VLQJ WKH .QRZQ 6WDWLVWLFV RI WKH 5HFHLYHUfV ,QSXW 9HFWRU & 5(9,(: 2) 48$'5$7,& )2506 2) *$866,$1 5$1'20 9$5,$%/(6 7<32*5$3+,&$/ 127(6 5()(5(1&(6 %,2*5$3+,&$/ 6.(7&+ Y
PAGE 6
/,67 2) ),*85(6 )LJXUH SDJH 3RZHU VSHFWUDO GHQVLWLHV RI WKH XQVSUHDG GDWD VHTXHQFH DQG WKH GLUHFW VHTXHQFH VSUHDGVSHFWUXP VLJQDOV &RGH DFTXLVLWLRQ ZLWK DQ DGDSWLYH UHFHLYHU 0LQLPXP PHDQVTXDUHG HUURU DV D IXQFWLRQ RI WKH SURSDJDWLRQ GHOD\ RI XVHU 3ORW RI WKH FRVW IXQFWLRQ ISf IRU D VLQJOHXVHU V\VWHP 3ORW RI WKH FRVW IXQFWLRQ S Â•f IRU D ILIWHHQXVHU V\VWHP 3HUIRUPDQFH RI HTXDWLRQ f IRU D VLQJOHXVHU V\VWHP Df 7KH DQDO\WLn FDO PHDQ ILOWHU ZHLJKW YHFWRU QRUP __(>ZPf@__ rf DQG WKH VLPXODWHG PHDQ ILOWHU ZHLJKW YHFWRU QRUP __ZPf__ VROLG OLQHf SORWWHG DV D IXQFn WLRQ RI WKH WUDLQLQJ OHQJWK Ef &RPSDULQJ WKH HUURU EHWZHHQ WKH WZR PHDQ ZHLJKW YHFWRUV __(>ZPf@ fÂ§ ZPf__ 3HUIRUPDQFH RI HTXDWLRQ f IRU D WKUHHXVHU V\VWHP Df 7KH DQDO\WLn FDO PHDQ ILOWHU ZHLJKW YHFWRU QRUP __(>ZPf@__ rf DQG WKH VLPXODWHG PHDQ ILOWHU ZHLJKW YHFWRU QRUP __ZPf__ VROLG OLQHf SORWWHG DV D IXQFn WLRQ RI WKH WUDLQLQJ OHQJWK Ef &RPSDULQJ WKH HUURU EHWZHHQ WKH WZR PHDQ ZHLJKW YHFWRUV __(>ZPf@ fÂ§ :Pf__ 3HUIRUPDQFH RI HTXDWLRQV f DQG f IRU D WZRXVHU V\VWHP 7KH )UREHQLXV QRUP RI WKH /06 ILOWHUfV WUDQVLHQW ZHLJKW DXWRFRYDULDQFH PDWUL[ HTXDWLRQ f rf HTXDWLRQ f [f DQG VLPXODWLRQ UHVXOWV f &RPSDULQJ LQFRUUHFW FKLSVHOHFWLRQ SUREDELOLW\ WR WKH 8QLRQERXQG IRU D WKUHHXVHU V\VWHP 6LPXODWLRQ UHVXOWV f WKH &KHUQRII ERXQG XVLQJ HTXDWLRQV f DQG f VROLG OLQHf DQG WKH &KHUQRII ERXQG XVLQJ VLPXODWHG ZHLJKW YHFWRU VWDWLVWLFV rf &RPSDULQJ WKH VLPXODWHG FRQGLWLRQDO YDULDQFH RI WKH ÂL HVWLPDWH WR WKH DQDO\WLFDOO\ GHULYHG DSSUR[LPDWLRQ LQ DEVHQFH RI QHDUIDU 0$, &RPSDULQJ WKH VLPXODWHG FRQGLWLRQDO YDULDQFH RI WKH ÂL HVWLPDWH WR WKH DQDO\WLFDOO\ GHULYHG DSSUR[LPDWLRQ LQ SUHVHQFH RI QHDUIDU 0$, YL
PAGE 7
$YHUDJH WUDLQLQJ ELWV UHTXLUHG IRU FRUUHFW DFTXLVLWLRQ DV D IXQFWLRQ RI WKH QXPEHU RI V\VWHP XVHUV LQ $:*1 FKDQQHO $YHUDJH WUDLQLQJ ELWV IRU FRUUHFW DFTXLVLWLRQ DV D IXQFWLRQ RI WKH OHYHO RI PXOWLDFFHVV LQWHUIHUHQFH LQ $:*1 FKDQQHO )ODWIDGLQJ FKDQQHO VLPXODWRU 7KHRUHWLFDO DQG VLPXODWHG SUREDELOLW\ GHQVLW\ IXQFWLRQ IRU WKH HQYHn ORSH RI WKH IDGLQJ FKDQQHO VLPXODWRU RXWSXW IRU WZR GLIIHUHQW 5LFLDQ SDUDPHWHU YDOXHV 0DJQLWXGH DQG SKDVH RI D VORZ IODWIDGLQJ 5D\OHLJK SURFHVV JHQHUDWHG ZLWK D FDUULHU IUHTXHQF\ RI *+] D YHKLFOH VSHHG RI 03+ DQG D GDWD UDWH RI %36 0RGLILHG DGDSWLYH UHFHLYHU ZLWK FKDQQHO FRPSHQVDWLRQ $FTXLVLWLRQ SHUIRUPDQFH RI WKH DGDSWLYH UHFHLYHU LQ D ILYHXVHU HQYLURQn PHQW LQ SUHVHQFH RI IUHTXHQF\ RIIVHW HUURUV 3HUIRUPDQFH RI WKH DFTXLVLWLRQ DOJRULWKP LQ D VORZO\ IODWIDGLQJ 5D\OHLJK FKDQQHO XVLQJ D WKLUGRUGHU OLQHDU SUHGLFWRU IRU WKH FKDQQHO HVWLPDn WLRQ DOJRULWKP 3HUIRUPDQFH RI WKH DFTXLVLWLRQ DOJRULWKP LQ D VORZO\ IODWIDGLQJ 5D\OHLJK FKDQQHO XVLQJ D WHQWKRUGHU OLQHDU SUHGLFWRU IRU WKH FKDQQHO HVWLPDn WLRQ DOJRULWKP &RPSDULQJ WKH DYHUDJH WUDLQLQJ ELWV IRU FRUUHFW DFTXLVLWLRQ DV D IXQFn WLRQ RI WKH OHYHO RI PXOWLDFFHVV LQWHUIHUHQFH LQ D IODWIDGLQJ FKDQQHO 3HUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP XVLQJ WKH K\EULG DQG PRGLILHG 006( UHFHLYHUV 3ORW RI WKH FRVW IXQFWLRQ ISf IRU D ILIWHHQXVHU V\VWHP ZKHQ WKH UHFHLYHU REVHUYHV WZR ELWLQWHUYDOV &RPSDULQJ WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU ZKHQ WKH UHFHLYHU REVHUYHV ELW LQWHUYDOV rf YHUVXV REVHUYLQJ ELW LQWHUYDO VROLG OLQHf DV D IXQFWLRQ RI WKH 0$, $O 7KH 006( UHFHLYHU IRU WKH *DXVVLDQ DSSUR[LPDWLRQ 9OO
PAGE 8
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
PAGE 9
086,& PXOWLSOH VLJQDO FODVVLILFDWLRQ 3&6 SHUVRQDO FRPPXQLFDWLRQ V\VWHP 36' SRZHU VSHFWUDO GHQVLW\ 5$6( UDSLG DFTXLVLWLRQ E\ VHTXHQWLDO HVWLPDWLRQ 5/6 UHFXUVLYH OHDVWVTXDUHV 506(( URRW PHDQVTXDUHG HVWLPDWLRQ HUURU 615 VLJQDOWRQRLVH UDWLR ,;
PAGE 10
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n PDWLRQ LV GLIILFXOW LQ WKH SUHVHQFH RI PXOWLSOH V\VWHP XVHUV DQG WUDGLWLRQDO PHWKRGV DUH NQRZQ WR EH VHQVLWLYH WR WKH QHDUIDU SUREOHP 7KH QHDUIDU SUREOHP LV D FRQn GLWLRQ ZKLFK RFFXUV ZKHQ WKH UHFHLYHG VLJQDO DPSOLWXGHV RI WKH WKH PXOWLSOH V\VWHP XVHUV DUH YHU\ GLVVLPLODU $ ZHOONQRZQ UHFHLYHU UHIHUUHG WR LQ WKH OLWHUDWXUH DV WKH PLQLPXP PHDQVTXDUHG HUURU 006(f UHFHLYHU LV XVHG WR IRUP DQ HVWLPDWH RI WKH FRGH WLPLQJ IRU D VLQJOH XVHU 7KH 006( UHFHLYHU KDV VHYHUDO GHVLUDEOH SURSHUWLHV 7KH 006( UHFHLYHU LV WKH RSWLPDO UHFHLYHU IRU D VLQJOHXVHU RSHUDWLQJ LQ DQ DGGLWLYH ZKLWH *DXVVLDQ QRLVH FKDQQHO EXW LV VXERSWLPDO LQ D PXOWLXVHU VFHQDULR +RZHYHU WKH 006(
PAGE 11
UHFHLYHU KDV EHHQ REVHUYHG WR EH UHVLVWDQW WR WKH QHDUIDU SUREOHP 7UDGLWLRQDOO\ WKH 006( UHFHLYHU LV LPSOHPHQWHG XVLQJ DQ DGDSWLYH ILOWHU DQG FDQ WKHUHIRUH OHDUQ DQG DGDSW WR WKH DPELHQW FKDQQHO FRQGLWLRQV ,W ZLOO EH VKRZQ WKDW ZLWK YHU\ OLWWOH VLGH LQIRUPDWLRQ LW LV IHDVLEOH WR IRUP DQ HVWLPDWH RI D VLQJOHXVHUf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n QHO ,W KDV EHHQ IRXQG WKDW WKH 006( UHFHLYHU PXVW EH PRGLILHG VXFK WKDW WKH UHFHLYHG VLJQDO LV FRPSHQVDWHG WR RIIVHW WKH G\QDPLF SKDVH FKDQJHV LQGXFHG E\ WKH FKDQQHO ,Q FHUWDLQ HQYLURQPHQWV WKLV FRPSHQVDWLRQ DOJRULWKP LV VKRZQ WR GHJUDGH WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWLRQ DOJRULWKP $ PRGLILFDWLRQ WR WKH RSHUDWLRQ RI WKH UHFHLYHU LV SURSRVHG WR RIIVHW VRPH RI WKLV SHUIRUPDQFH GHJUDGDWLRQ [L
PAGE 12
&+$37(5 ,1752'8&7,21 ,QWURGXFWLRQ WR 6SUHDG6SHFWUXP 7HFKQLTXHV ,Q DFWUHVV +HG\ /DPDUU DQG FRPSRVHU *HRUJH $QWKHLO ILOHG IRU D SDWHQW RQ D f6HFUHW &RPPXQLFDWLRQ 6\VWHPf >@ 7KHLU GHYLFH RSHUDWHG RQ WKH LGHD RI IUHn TXHQWO\ FKDQJLQJ WKH FDUULHU IUHTXHQF\ RI D UDGLR VLJQDO ,I WKH FDUULHU IUHTXHQF\ ZDV XSGDWHG IDVW HQRXJK DQG LQ DQ DSSDUHQWO\ UDQGRP IDVKLRQ DQ DGYHUVDU\ ZRXOG KDYH GLIILFXOW\ PDLQWDLQLQJ FRQWDFW ZLWK WKH WUDQVPLWWLQJ UDGLR 7KLV PHDQV WKDW DQ DGn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fVHFUHWf FRPPXQLFDWLRQ /DPDUU DQG $QWKHLO JDYH WKHLU LQYHQWLRQ DQG WKH ULJKWV WR WKH SDWHQW WR WKH 8QLWHG 6WDWHV JRYHUQPHQW 6LQFH WKLV GHYHORSPHQW RFFXUUHG GXULQJ :RUOG :DU ,, WKLV WHFKn QLTXH LQLWLDOO\ UHFHLYHG D ORW RI DWWHQWLRQ +RZHYHU WKH 86 JRYHUQPHQW GHHPHG WKH PHFKDQLFV RI WKH V\QFKURQL]DWLRQ PHWKRG WRR FRPSOH[ IRU ZLGHVFDOH PLOLWDU\ XVH 7KH WHFKQLTXH GHYHORSHG E\ /DPDUU DQG $QWKHLO ZKLFK LV QRZ NQRZQ DV IUHTXHQF\ KRSSHG VSUHDGVSHFWUXP )+66f LV H[SODLQHG LQ WKH IROORZLQJ IHZ VHQWHQFHV /HW % UHSUHVHQW WKH EDQGZLGWK UHTXLUHG WR WUDQVPLW WKH PHVVDJH VLJQDO XVLQJ D WUDGLWLRQDO UDGLR ZLWK D IL[HG FDUULHU IUHTXHQF\ ,I ZH QRZ OHW WKH FDUULHU IUHTXHQF\ KRS WKURXJK 1 GLVFUHWH FDUULHU IUHTXHQFLHV HDFK VHSDUDWHG E\ % +] WKHQ WKH EDQGZLGWK RI WKH
PAGE 13
WUDQVPLWWHG VLJQDO EHFRPHV 1% 7KHUHIRUH WKH EDQGZLGWK RI WKH WUDQVPLWWHG VLJn QDO XVLQJ IUHTXHQF\KRSSLQJ KDV EHHQ VSUHDG WR D ZLGHU EDQGZLGWK WKDQ ZKDW LV UHTXLUHG WR WUDQVPLW WKH PHVVDJH VLJQDO +RZHYHU XQWLO WKH DGYHQW RI WKH LQWHJUDWHG FLUFXLW WKH SUREOHP RI LPSOHPHQWLQJ D V\QFKURQL]DWLRQ PHWKRG OLPLWHG WKH XVH RI VXFK V\VWHPV 7KH GXDO RI )+66 LV NQRZQ DV GLUHFWVHTXHQFH VSUHDGVSHFWUXP '666f ,Q '6 66 D GLJLWDO VLJQDO LV PRGXODWHG E\ D GLJLWDO VSUHDGLQJVHTXHQFH SULRU WR WUDQVPLVVLRQ &RQVLGHU D ELQDU\ VHTXHQFH RI s GDWD ZKHUH HDFK GDWD ELW KDV GXUDWLRQ 7f ,I WKLV GDWD VHTXHQFH LV XVHG WR DPSOLWXGH RU SKDVH PRGXODWH D FDUULHU VLJQDO WKH SRZHU VSHFWUDO GHQVLW\ 36'f RI WKH WUDQVPLWWHG VLJQDO ZLOO KDYH D QXOOWRQXOO EDQGZLGWK RI f +] $ VSUHDGLQJ VHTXHQFH LV FRPSULVHG RI D SVHXGRUDQGRP VHTXHQFH RI s FKLSV ZKHUH HDFK FKLS KDV GXUDWLRQ 7F DQG LW LV DVVXPHG WKDW 7cc 17& %\ PXOWLSO\LQJ WKH GDWD VHTXHQFH E\ WKH VSUHDGLQJ VHTXHQFH SULRU WR WUDQVPLVVLRQ WKH 36' RI WKH WUDQVPLWWHG VLJQDO QRZ KDV D QXOOWRQXOO EDQGZLGWK RI 7F ZKLFK LV 1 WLPHV JUHDWHU WKDQ WKH FRUUHVSRQGLQJ EDQGZLGWK RI WKH RULJLQDO XQVSUHDG GDWD VHTXHQFH 7KH WRWDO WUDQVPLWWHG SRZHU LQ WKH XQVSUHDG DQG WKH '666 VLJQDOV LV DVVXPHG WR EH HTXDO VLQFH PXOWLSOLFDWLRQ E\ s GRHVQfW DIIHFW WKH WUDQVPLWWHG SRZHU 7KLV PHDQV WKDW WKH 36' IRU WKH '666 VLJQDO ZLOO KDYH D PXFK ORZHU DPSOLWXGH WKDQ WKH 36' IRU WKH XQVSUHDG VLJQDO DV VKRZQ LQ )LJXUH ,Q WKH SUHYLRXV SDUDJUDSK ZH UHIHUUHG WR WKH VSUHDGLQJ VHTXHQFH DV EHLQJ SVHXGRn UDQGRP ,GHDOO\ WKH FKLSV RI WKH VSUHDGLQJ VHTXHQFH ZRXOG EH UDQGRPO\ JHQHUDWHG ZLWK RFFXUULQJ ZLWK WKH VDPH SUREDELOLW\ DV +RZHYHU MXVW DV WKH LQWHQGHG UHn FHLYHU LQ D )+66 V\VWHP UHTXLUHV NQRZOHGJH DQG V\QFKURQL]DWLRQf RI WKH IUHTXHQF\ KRSSLQJ DOJRULWKP WKH LQWHQGHG UHFHLYHU LQ D '666 V\VWHP PXVW KDYH NQRZOHGJH DQG V\QFKURQL]DWLRQf RI WKH VSUHDGLQJ VHTXHQFH 7KHUHIRUH VRPH GHWHUPLQLVWLF PHWKRG PXVW EH XVHG WR JHQHUDWH D VSUHDGLQJ VHTXHQFH $ SVHXGRUDQGRP VHTXHQFH FDQ EH JHQHUDWHG ZLWK D OLQHDUIHHGEDFN VKLIWUHJLVWHU /)65f ZKLFK LV MXVW D EDQN RI
PAGE 14
)LJXUH 3RZHU VSHFWUDO GHQVLWLHV RI WKH XQVSUHDG GDWD VHTXHQFH DQG WKH GLUHFW VHTXHQFH VSUHDGVSHFWUXP VLJQDOV IOLSIORSV LQWHUFRQQHFWHG ZLWK WKH DSSURSULDWH IHHGEDFN 8VLQJ DQ /)65 WR JHQHUDWH WKH VSUHDGLQJ VHTXHQFH WKH WUDQVPLWWHU DQG UHFHLYHU PXVW KDYH FRPPRQ NQRZOHGJH RI WKH OHQJWK RI WKH /)65 WKH IHHGEDFN FRQQHFWLRQV DQG WKH LQLWLDO VWDWH RI WKH /)65 SULRU WR WUDQVPLVVLRQ :LWK WKH DSSURSULDWH IHHGEDFN fÂ§ ZLOO RFFXU ZLWK D SUREDn ELOLW\ VOLJKWO\ KLJKHU WKDQ ,Q DGGLWLRQ WKH VHTXHQFH ZLOO EH SHULRGLF +RZHYHU LI WKH SHULRG LV VXIILFLHQWO\ ORQJ RYHU D VKRUW REVHUYDWLRQ LQWHUYDO WKH RFFXUUHQFH RI DQG fÂ§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
PAGE 15
LQ D VSUHDGVSHFWUXP V\VWHP D MDPPHU WKHQ RQO\ DFKLHYHV SDUWLDO EDQG MDPPLQJ RU KDV WR VSUHDG WKH MDPPLQJ VLJQDO RYHU D ZLGHU EDQGZLGWK ,Q HLWKHU FDVH WKH HIIHFW RI WKH MDPPLQJ VLJQDO LV UHGXFHG FRPSDUHG WR WKH HIIHFW REVHUYHG E\ WKH VDPH MDPPHU LQ WKH XQVSUHDG FRPPXQLFDWLRQ V\VWHP 7KH VHFRQG SURSHUW\ RI LQWHUHVW LV WKH ORZ SUREDELOLW\ RI GHWHFWLRQ /3'f (YHQ LI DQ DGYHUVDU\ NQRZV \RX DUH WUDQVPLWWLQJ ZLWKRXW NQRZOHGJH RI WKH VSUHDGLQJ VHTXHQFH RU IUHTXHQF\ KRS VHTXHQFHf LW LV YHU\ GLIILFXOW IRU WKDW DGYHUVDU\ WR SHUIRUP D JRRG GHWHFWLRQ RI WKH PHVVDJH VLJQDO 7KH WKLUG SURSHUW\ ZKLFK LV HVSHFLDOO\ LPSRUWDQW IRU FODQGHVWLQH FRPPXQLFDWLRQ LW FDOOHG ORZ SUREDELOLW\ RI LQWHUFHSW /3,f &RQVLGHULQJ WKH UHGXFHG DPSOLWXGH RI WKH '666 VLJQDOf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fOLIH H[SHFWDQF\f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
PAGE 16
,Q GLUHFWVHTXHQFH FRGHGLYLVLRQ PXOWLSOHDFFHVV '6&'0$f V\VWHPV HDFK XVHU KDV D XQLTXH VSUHDGLQJ VHTXHQFH DQG DOO XVHUV RFFXS\ WKH VDPH IUHTXHQF\ EDQG WUDQVn PLWWLQJ LQGHSHQGHQWO\ RI DOO RWKHU XVHUV &RQVLGHU D XVHU '6&'0$ V\VWHP LQ ZKLFK HDFK XVHU WUDQVPLWV s WLPHV LWV VSUHDGLQJ FRGH :H ZLOO FRPPXQLFDWH ZLWK RQH RI WKH XVHUV FDOOHG WKH GHVLUHG XVHU XVLQJ D UHFHLYHU WKDW LV RSWLPL]HG IRU D VLQJOHXVHU RSHUDWLQJ LQ WKH SUHVHQFH RI DGGLWLYH ZKLWH *DXVVLDQ QRLVH $:*1f )RU WKLV H[DPSOH WKH RSWLPXP UHFHLYHU LV D ILOWHU PDWFKHG WR WKH GHVLUHG XVHUfV VSUHDGLQJ FRGH 7KHUHIRUH WKH RXWSXW RI WKH PDWFKHG ILOWHU ZLOO GHSHQG RQ WKH FRUUHODWLRQ RI WKH GHVLUHG XVHUfV FRGH ZLWK DOO RI WKH RWKHU fÂ§ XVHUVf FRGHV ,I DOO RI WKH XVHUVf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fÂ§ LQWHUIHULQJ XVHUV LV ORFDWHG SK\VLFDOO\ FORVHU WR WKH UHFHLYHU WKDQ WKH GHVLUHG XVHU ,Q WKLV FDVH WKH RXWSXW RI WKH PDWFKHG ILOWHU PD\ EHFRPH GRPLQDWHG E\ WKH LQWHUIHUHQFH 7KLV W\SH RI PXOWLn DFFHVV LQWHUIHUHQFH 0$,f LV UHIHUUHG WR DV WKH QHDUIDU SUREOHP 7KH FKDOOHQJH IRU FRPPXQLFDWLRQ V\VWHPV HQJLQHHUV LV WR GHYHORS UHFHLYHU VWUXFWXUHV WKDW DUH LQVHQVLWLYH WR WKH QHDUIDU SUREOHP ,Q WKLV GLVVHUWDWLRQ ZH DUH FRQFHUQHG ZLWK WKH SUREOHP RI DFKLHYLQJ V\QFKURQL]Dn WLRQ ZLWK WKH VSUHDGLQJ VHTXHQFH RI WKH GHVLUHG XVHU 0RRQ HW DO >@ VWXGLHG WKH HIIHFWV RI 0$, RQ WUDGLWLRQDO WHFKQLTXHV HQHUJ\ GHWHFWRUf WKDW DUH XVHG LQ VLQJOH XVHU VSUHDGVSHFWUXP FRPPXQLFDWLRQ V\VWHPV ,W ZDV IRXQG WKDW WKH DYHUDJH WLPH WR DFKLHYH DFTXLVLWLRQ TXLFNO\ LQFUHDVHV ZKHQ D QHDUIDU VFHQDULR H[LVWV ZLWK RQO\ WZR
PAGE 17
V\VWHP XVHUV 0RRQ HW DO >@ SUHVHQWHG D PXOWLXVHU V\QFKURQL]DWLRQ PHWKRG FDOOHG WKH UDSLG DFTXLVLWLRQ E\ VHTXHQWLDO HVWLPDWLRQ 5$6(f DOJRULWKP ZKLFK HVVHQWLDOO\ WULHV WR HVWLPDWH WKH FRQWHQWV RI WKH /)65 XVHG WR JHQHUDWH WKH VSUHDGLQJ VHTXHQFH IRU HDFK XVHU EDVHG RQ REVHUYDWLRQV RI WKH UHFHLYHG VLJQDO 7KH\ DVVXPHG WKDW DOO XVHUV WUDQVPLW RQO\ WKHLU VSUHDGLQJ FRGH DQG WKDW WKH UHFHLYHU NQRZV WKH UHFHLYHG DPSOLn WXGH RI HDFK XVHU )RU D WRWDO RI XVHUV WKLV DOJRULWKP KDV D FRPSOH[LW\ RI .Nf ZKLFK LV SURKLELWLYH IRU $ VLPLODU VLQJOHXVHU DSSURDFK ZDV SUHVHQWHG E\ %DUJKRXWKL DQG 6WLLEHU >@ XQGHU WKH DVVXPSWLRQ WKDW DOO XVHUV DUH UHFHLYHG DW WKH VDPH SRZHU OHYHO 0DGKRZ DQG 3XUVOH\ >@ FRQVLGHUHG WKH HIIHFWV RI DFKLHYLQJ FRGH V\QFKURQL]DWLRQ LQ WKH SUHVHQFH RI 0$, RQ WKH V\VWHP FDSDFLW\ 7KH\ IRXQG WKDW WKH V\VWHP FDSDFLW\ IRU DFKLHYLQJ FRGH V\QFKURQL]DWLRQ ZLWK D PDWFKHG ILOWHU ZDV PXFK OHVV WKDQ WKH V\VWHP FDSDFLW\ PHDVXUHG ZLWK UHVSHFW WR WKH ELWHUURUUDWH %(5f 7KHVH UHVXOWV IRUP WKH PRWLYDWLRQ IRU UHVHDUFKLQJ FRGH DFTXLVLWLRQ WHFKQLTXHV IRU '6&'0$ V\VWHPV 0DWKHPDWLFDO 1RWDWLRQ 0DWULFHV DQG YHFWRUV DUH W\SHVHW LQ D EROG IDFH IRU H[DPSOH Y RU 9 ,Q JHQHUDO D ORZHUFDVH OHWWHU ZLOO EH XVHG WR GHQRWH D YHFWRU ZKLOH DQ XSSHUFDVH OHWWHU GHQRWHV D PDWUL[ 8QOHVV RWKHUZLVH QRWHG DOO YHFWRUV DUH DVVXPHG WR EH FROXPQ YHFWRUV ,Q RUGHU WR GHQRWH WKH QWK HOHPHQW RI D YHFWRU D VXEVFULSW ZLOO EH XVHG VXFK DV YQ 7KH LMfWK HOHPHQW RI D PDWUL[ 9 LV GHQRWHG E\ 9ÂÂ $ VXEVFULSW FRQWDLQLQJ WKH YDULDEOH N GHQRWHV DQ HOHPHQW EHORQJLQJ WR WKH NWK XVHU LQ WKH '6&'0$ V\VWHP 7KH WUDQVSRVH WKH FRQMXJDWH RU +HUPLWLDQf WUDQVSRVH DQG WKH QRUP RI D FROXPQYHFWRU Y DUH GHQRWHG E\ Y7 ?+ DQG __Y__ 9Y+Y UHVSHFWLYHO\ 7KH FRQMXn JDWH RI D VFDODU TXDQWLW\ ZLOO EH GHQRWHG DV Dr )RU D PDWUL[ 9 ZH ZLOO XVH __9__) WR
PAGE 18
GHQRWH WKH )UREHQLXVQRUP RI WKH PDWUL[ GHILQHG DV f L M 7KH .URQHFNHU GHOWD IXQFWLRQ LV GHILQHG DV LI P f 6.Pf RWKHUZLVH 7KH H[SHFWDWLRQ RI D UDQGRP YDULDEOH HLWKHU D VFDODU RU QRQVFDODU TXDQWLW\ ZLOO EH GHQRWHG DV (>I@ RU c :H XVH WKH QRWDWLRQ 21f WR GHQRWH WKDW DQ DOJRULWKP KDV QXPHULFDO FRPSOH[LW\ RQ WKH RUGHU RI 1 :KHQ D LV D SDUDPHWHU WR EH HVWLPDWHG ZH GHQRWH WKH HVWLPDWH DV D 2XWOLQH RI WKH 'LVVHUWDWLRQ ,Q WKH QH[W FKDSWHU D V\VWHP PRGHO WKDW LV ZLGHO\ XVHG WR VLPXODWH D '6&'0$ FRPPXQLFDWLRQ V\VWHP ZLOO EH SUHVHQWHG 7KH V\VWHP PRGHO LV H[SUHVVHG LQ D FRQYHn QLHQW YHFWRUPDWUL[ QRWDWLRQ ZKLFK IDFLOLWDWHV VLPXODWLQJ D '6&'0$ FRPPXQLFDWLRQ V\VWHP RQ D FRPSXWHU 7KH WKLUG FKDSWHU SURYLGHV D EULHI UHYLHZ RI WKH OLWHUDWXUH FRQFHUQLQJ '6&'0$ UHFHLYHUV DQG WLPLQJ DFTXLVLWLRQ WHFKQLTXHV :KLOH LW LV LPSRVn VLEOH WR SURYLGH D FRPSOHWH UHYLHZ RI WKHVH VXEMHFWV DQ HIIRUW LV PDGH WR SURYLGH WKH UHDGHU ZLWK HQRXJK RI D UHYLHZ WR XQGHUVWDQG VHYHUDO WLPLQJ DFTXLVLWLRQ WHFKQLTXHV DQG WKHLU DVVRFLDWHG FRPSOH[LWLHV ,Q WKH IRXUWK FKDSWHU D ORZFRPSOH[LW\ VLQJOHXVHU WLPLQJ HVWLPDWLRQ DOJRULWKP LV SUHVHQWHG 7KH DOJRULWKP LV EDVHG RQ SURFHVVLQJ WKH ZHLJKWV RI DQ DGDSWLYH ILOWHU ZKLFK LV FRPPRQO\ UHIHUUHG WR DV WKH 006( UHFHLYHU 7KH 006( UHFHLYHU LV D VLQJOHELW VLQJOHXVHU GHWHFWRU ZKLFK LV D VXERSWLPDO UHFHLYHU IRU D PXOWLXVHU '6&'0$ FRPPXQLFDWLRQ V\VWHP DQG LV UHFHLYLQJ FRQVLGHUDEOH DWn WHQWLRQ LQ WKH OLWHUDWXUH $V VHYHUDO SDSHUV LQ WKH OLWHUDWXUH LJQRUH WKH HIIHFWV RI RSHUDWLQJ WKH 006( UHFHLYHU DV\QFKURQRXV WR WKH LQWHQGHG XVHU VHYHUDO H[DPSOHV
PAGE 19
RI ZK\ V\QFKURQL]DWLRQ LV LPSRUWDQW DUH SUHVHQWHG ,Q WKH ILIWK FKDSWHU WKH SHUn IRUPDQFH RI WKH WLPLQJ HVWLPDWRU LV FKDUDFWHUL]HG 6HYHUDO DQDO\WLFDO WHFKQLTXHV DUH XVHG WR FKDUDFWHUL]H WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU EDVHG RQ D GHYHORSHG VWDWLVWLFDO PRGHO RI WKH WUDQVLHQW VWDWLVWLFV RI WKH 006( UHFHLYHU ZHLJKWV 6LPXODWLRQ UHVXOWV DUH XVHG WR VWXG\ WKH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP IRU D JHQHUDO DV\QFKURQRXV '6&'0$ V\VWHP ZKHQ WZR FRPPRQO\ XVHG DGDSWLYH DOJRn ULWKPV DUH XVHG LQ WKH 006( UHFHLYHU ,W ZLOO EH VKRZQ WKDW IRU RQH RI WKHVH DGDSWLYH DOJRULWKPV WKH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP LV QHDUIDU UHVLVWDQW ,Q WKH VL[WK FKDSWHU WKH HIIHFWV RI IDGLQJ RQ D FRPPXQLFDWLRQ V\VWHP DUH GLVFXVVHG DQG D FRPPRQO\ XVHG LPSOHPHQWDWLRQ RI D IODWIDGLQJ FKDQQHO VLPXODWRU LV SUHVHQWHG 7KH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP ZKHQ WKH UHFHLYHU LV RSHUDWLQJ LQ WZR QRQVWDWLRQDU\ HQYLURQPHQWV IUHTXHQF\RIIVHW HUURUV DQG IODWIDGLQJ FKDQQHOVf LV VWXGLHG LQ WKH VHYHQWK FKDSWHU %DUERVD DQG 0LOOHU >@ KDYH SUHVHQWHG D PRGLILHG YHUVLRQ RI WKH 006( UHFHLYHU WKDW FDQ EH XVHG LQ IODWIDGLQJ FKDQQHOV 7KH SHUIRUn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n EXWLRQV WR WKH DUHD RI '6&'0$ UHVHDUFK DV ZHOO DV VHYHUDO DUHDV RI IXWXUH UHVHDUFK DUH SURYLGHG
PAGE 20
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f FRPPXQLFDWLRQ V\VWHP LV XVHG 7KHUH DUH D WRWDO RI V\VWHP XVHUV HDFK RSHUDWLQJ DW WKH VDPH FDUULHU IUHTXHQF\ F EXW DV\QFKURQRXV WR DQG LQGHSHQGHQW RI WKH RWKHU XVHUV 7KH $WK XVHU LV DVVXPHG WR WUDQVPLW D VLJQDO ZKRVH FRPSOH[ HQYHORSH >@ 6NWf LV D SRODU GDWD VHTXHQFH GNOf ^` ZKLFK KDV D XQLTXH VLJQDWXUH RU VSUHDGLQJ ZDYHIRUP FNWf VXSHULPSRVHG XSRQ LW VXFK WKDW 22 6NWf < 00 O7Ef NH^ .` f O fÂ§22 7KHUHIRUH WKH VLJQDO DFWXDOO\ WUDQVPLWWHG E\ WKH NWK XVHU LV JLYHQ E\ VNWf 5H >6NWf H[S^MQIFW`@ 5H >6NWf H[S^MXMFW`@ N H ^ .` f ZKHUH 5H> @ UHWXUQV WKH UHDO SDUW RI LWV DUJXPHQW :LWKRXW ORVV RI JHQHUDOLW\ LW LV DVVXPHG WKDW WKH ILUVW XVHU LV WKH GHVLUHG XVHU DQG DOO RWKHU XVHUV DFW DV LQWHUIHUHQFH 7KH GDWD ELWV KDYH D GXUDWLRQ 7E ZKLOH WKH FKLSV RI WKH VSUHDGLQJ VHTXHQFH KDYH GXUDWLRQ 7F (DFK VSUHDGLQJ VHTXHQFH KDV D SHULRG RI 1 7E7F FKLSV 7KDW LV RQH
PAGE 21
SHULRG RI D VSUHDGLQJ VHTXHQFH LV HTXDO LQ GXUDWLRQ WR RQH ELW LQWHUYDO :KLOH WKLV LV QRW D UHTXLUHPHQW IRU '6&'0$ V\VWHPV LQ JHQHUDO LW LV UHTXLUHG IRU WKH UHFHLYHU VWUXFWXUH VWXGLHG LQ WKLV ZRUN /HW WKH YHFWRU Fr FAR FALFN1Lf7 UHSUHVHQW RQH SHULRG RI WKH NWK XVHUfV VSUHDGLQJ VHTXHQFH VR WKDW WKH VSUHDGLQJ ZDYHIRUP FDQ EH ZULWWHQ DV 1 &IFLf A FNQ37F^W Q7Ff f Q ZKHUH 3UF LV WKH FKLS SXOVH VKDSH ZKLFK LV WDNHQ WR EH RYHU WKH LQWHUYDO >7&f DQG ]HUR RWKHUZLVH :H DVVXPH WKDW WKH UHFHLYHU LV DV\QFKURQRXV LQ WLPH DQG SKDVH WR WKH WUDQVPLWWHG VLJQDO RI WKH GHVLUHG XVHU DV ZHOO DV EHLQJ DV\QFKURQRXV WR DOO RI WKH LQWHUIHULQJ XVHUV 'XH WR WKH V\VWHP XVHUV WKH UHFHLYHG VLJQDO LV RI WKH IRUP N UWf A ILW 5H >6NW 7IFf H[S ^MXHW M2N`? QWf f N ,Q WKH DERYH H[SUHVVLRQ UIF DQG 3N DUH WKH SURSDJDWLRQ GHOD\ DQG WKH DYHUDJH UHFHLYHG VLJQDO SRZHU UHVSHFWLYHO\ IRU WKH IFWK XVHU N LV WKH SKDVHRIIVHW IRU WKH NWK XVHU ZKLOH QWf LV WKH DGGLWLYH ZKLWH *DXVVLDQ QRLVH ZKLFK LV DVVXPHG WR KDYH D RQHVLGHG SRZHU VSHFWUDO GHQVLW\ RI :R ,Q RUGHU WR SURFHVV WKH UHFHLYHG VLJQDO WKH UHFHLYHU FRQYHUWV UWf IURP D EDQGSDVV VLJQDO WR D EDVHEDQG VLJQDO 5Wf ZKLFK LV WKHQ SDVVHG WKURXJK D ILOWHU PDWFKHG WR WKH FKLS SXOVH VKDSH DQG WKH RXWSXW RI WKLV PDWFKHG ILOWHU LV VDPSOHG DW WKH FKLS UDWH ,W LV DVVXPHG WKDW DQ\ GRXEOHIUHTXHQF\ WHUPV FUHDWHG LQ WKH FRQYHUVLRQ WR EDVHEDQG DUH QRW SDVVHG WKURXJK WKH FKLS PDWFKHG ILOWHU 'XULQJ WKH UHFHLYHUfV PWK ELW LQWHUYDO WKH QWK FKLS VDPSOHG RXWSXW FDQ EH H[SUHVVHG DV P7EQOf7F U PUF PGW & P7EQ7F f
PAGE 22
'XULQJ HDFK ELW LQWHUYDO WKH 1 FKLS VDPSOHV DUH VWRUHG LQ D UHFHLYHG YHFWRU UPf UPLRUPLL UPMYBLfW DQG LW LV WKLV YHFWRU TXDQWLW\ WKDW ZLOO EH SURFHVVHG E\ WKH UHFHLYHU 8VLQJ WKH DERYH GHILQLWLRQV IRU KRZ HDFK XVHU FRQWULEXWHV WR UWf ZH FDQ H[SUHVV KRZ HDFK XVHU FRQWULEXWHV WR UPf XVLQJ D QRWDWLRQ VLPLODU WR WKDW XVHG E\ 0LOOHU >@ ,W FDQ EH VKRZQ WKDW N S UPf ? IFUDf H[S^M2N` 1P H&1[O f /9 0 N 7 ZKHUH NPf LV WKH FRQWULEXWLRQ WR WKH UHFHLYHG YHFWRU IURP WKH NWK XVHU GXULQJ WKH PWK UHFHLYHU ELW LQWHUYDO JLYHQ E\ IFPf >IFLPfDIFBLSIFf ]NPfDNSN f@ ZKHUH f ]NL^Pf >GNPf GNP Of@ ^ ` =N^Pf >GNPf GNP Of@ ^` DrL3IHIFf 6Nf3Wf N3nWf tN3N6Nf ONfFrNf NFrNf Qf L V7 &N ?&N1fÂ§QL &N1fÂ§Q" f f f &N1fÂ§! nNILL &IFO f f f &NWO?fÂ§QfÂ§Of aQf L A AN 9 &N1fÂ§P AN1fÂ§QOL f Â‘ f L AN1fÂ§ M &IF! f Â‘ f &IF9fÂ§QfÂ§ M H ^` LI P ^ ,9 ` RWKHUZLVH f f f f f DQG UN SN7F N7F ZLWK SN ^ 1 ` DQG 6N 6LQFH WKH UHFHLYHU LV DV\QFKURQRXV WR DQ\ RI WKH V\VWHP XVHUV DQG RQO\ REVHUYHV RQH ELW LQWHUYDO RI LQIRUPDWLRQ LW LV SRVVLEOH WR KDYH D GDWD ELW WUDQVLWLRQ RFFXU GXULQJ WKH UHFHLYHUfV ELW LQWHUYDO 7KLV IDFW LV DFFRXQWHG IRU LQ WKH SUHYLRXV HTXDWLRQV 7KH TXDQWLWLHV ]NLPf DQG ]NPf DUH XVHG WR LQGLFDWH LI DGMDFHQW GDWD ELWV IRU WKH NWK
PAGE 23
XVHU DUH VLPLODU ,I DGMDFHQW GDWD ELWV DUH WKH VDPH WKHQ =IFLPf ZLOO EH QRQ]HUR RWKHUZLVH AIHPf LV QRQ]HUR 7KH YHFWRU TXDQWLWLHV DQG FMQf GHQRWH WKH QWK ULJKW F\FOLF DQG WKH QWK PRGLILHG ULJKW F\FOLF VKLIWV RI WKH NWK XVHUfV VSUHDGLQJ VHTXHQFH UHVSHFWLYHO\ $V VKRZQ LQ WKH SUHYLRXV HTXDWLRQV cWPf LV MXVW D IXQFWLRQ RI WZR VOLJKWO\ RIIVHW F\FOLF VKLIWV RI WKH VSUHDGLQJ VHTXHQFH IRU WKH NWK XVHU ,Q WKH UHPDLQGHU RI WKLV ZRUN ZH ZLOO UHIHU WR WKH SURSDJDWLRQ GHOD\ LQ WHUPV RI WKH LQWHJHU SDUW Sr DQG WKH IUDFWLRQDO SDUW 6N RQO\ 7KH QRLVH YHFWRU 1P LQ HTXDWLRQ f FRQVLVWV RI LQGHSHQGHQW FRPSOH[ *DXVVLDQ UDQGRP YDULDEOHV ZKRVH UHDO DQG LPDJLQDU\ SDUWV DUH DOVR LQGHSHQGHQW KDYLQJ ]HURPHDQ DQG HTXDO YDULDQFHV RI FU 1(E?IRf ZKHUH (E LV WKH DYHUDJH UHFHLYHG HQHUJ\ SHU ELW RI WKH GHVLUHG XVHU ,Q WKH UHPDLQGHU RI WKLV ZRUN ZH GHILQH WKH VLJQDOWRQRLVH UDWLR 615f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fV VLJQDOV $W DQ\ RI WKH PRELOH UHFHLYHUV WKHQ RQH FRXOG VD\ WKDW DOO XVHUV H[SHULHQFH WKH VDPH SURSDJDWLRQ GHOD\ DQG FDUULHU SKDVH VKLIW LQ WKH FKDQQHO IURP WKH EDVH VWDWLRQ WR WKH PRELOH UHFHLYHU +RZHYHU LQ WKH FRPPXQLFDWLRQ OLQN IURP WKH PRELOH XVHU WR WKH EDVH VWDWLRQ GLIIHUHQW SURSDJDWLRQ GHOD\V DQG FDUULHU SKDVH VKLIWV ZLOO H[LVW GXH WR WKH GLIIHUHQW SK\VLFDO ORFDWLRQV RI WKH PRELOH XVHUV
PAGE 24
+RZHYHU ZH FDQ PRGHO WKH V\VWHP DV EHLQJ HTXLYDOHQWO\ SKDVHV\QFKURQRXV DV IROORZV :H FDQ H[SUHVV HTXDWLRQ f DV A L S UPf fMFRVWfMVPHWffNPf :P f &.[O f IFfÂ§ f :H WKHQ FDQ FUHDWH D PRGLILHG UHFHLYHU LQSXW E\ RQO\ SURFHVVLQJ WKH UHDO SDUW RI UPf 7KDW LV UPf 5H >UPf@ A N ZKHUH WKH QRLVH YHFWRU 1P LQ HTXDWLRQ f FRQVLVWV RI UHDOYDOXHG *DXVVLDQ UDQGRP YDULDEOHV WKDW DUH LQGHSHQGHQW KDYLQJ ]HURPHDQ DQG HTXDO YDULDQFHV RI D 1(E$If 1RZ WKH FRV4Nf WHUP FDQ EH DEVRUEHG LQWR WKH UDWLR RI 3N3LÂ‘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fV SURSDJDWLRQ GHOD\ FDQ EH H[SUHVVHG LQ FORVHG IRUP ZKLFK LV QRW WKH FDVH ZKHQ WKH PRGHO RI 6HFWLRQ LV XVHG FRVIHfN^Pf Â“. f!9[ f
PAGE 25
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f VSUHDGLQJ FRGH DQG PXVW EH V\QFKURQL]HG ZLWK HDFK XVHU ,Q DGGLWLRQ WKH UHFHLYHU PXVW NQRZ WKH UHFHLYHG VLJQDO HQHUJ\ IRU HDFK XVHU 7KH FRPSOH[LW\ RI WKH UHFHLYHU LV $f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f WKDW UHPRYHV WKH PXOWLXVHU LQWHUIHUHQFH IURP WKH RXWSXW RI HDFK ILOWHU 7KH
PAGE 26
GHFRUUHODWLQJ ILOWHU FDQFHOV WKH 0$, E\ DFFRXQWLQJ IRU WKH NQRZQ FRUUHODWLRQ EHWZHHQ WKH XVHUVf VSUHDGLQJ FRGHV 7KH FRPSOH[LW\ RI WKH UHFHLYHU LV GHSHQGHQW RQ WDNLQJ WKH LQYHUVH RI D PDWUL[ ZKLFK KDV FRPSOH[LW\ 1f WR IRUP WKH GHFRUUHODWLQJ ILOWHU ,I WKLV RSHUDWLRQ FDQ EH GRQH RQFH WKH FRPSOH[LW\ RI WKH UHFHLYHU LV WKHQ OLQHDU LQ WKH QXPEHU RI XVHUV ZKLFK LV D JUHDW LPSURYHPHQW RYHU WKH H[SRQHQWLDO FRPSOH[LW\ RI WKH 9LWHUEL EDVHG GHFRGHU 6WURP HW DO >@ VWXGLHG WKH LPSDFW RI WLPLQJ HUURUV RQ WKH GHFRUUHODWLQJ UHFHLYHU ,W ZDV IRXQG WKDW WKH GHFRUUHODWLQJ UHFHLYHU ORVHV LWV QHDUIDU UHVLVWDQFH ZLWK WKH SUHVn HQFH RI WLPLQJ HUURUV 7KDW LV WKH %(5 RI WKH UHFHLYHU UHDFKHV D QRQGLPLQLVKLQJ IORRU HYHQ DV WKH 615 LQFUHDVHV ZKHQ HUURUV DUH SUHVHQW LQ WKH HVWLPDWHV RI WKH WLPLQJ RIIVHWV ,Q DGGLWLRQ WKH\ REVHUYHG WKDW WKH UHTXLUHG YDULDQFH RI D WLPLQJ HVWLPDWH LV LQYHUVHO\ SURSRUWLRQDO WR WKH OHYHO RI WKH 0$, 7KLV PHDQV DV WKH 0$, LQn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n QLILFDQW DGYDQWDJH RYHU WKH FRQYHQWLRQDO PDWFKHG ILOWHU 7KHVH UHVXOWV PRWLYDWH WKH QHHG IRU WKH GHYHORSPHQW RI WLPLQJ HVWLPDWRUV WKDW SURGXFH XQELDVHG HVWLPDWHV ZLWK ORZ YDULDQFHV HYHQ LQ WKH SUHVHQFH RI D VHYHUH QHDUIDU HQYLURQPHQW 9DUDQDVL DQG $D]KDQJ >@ KDYH SURSRVHG D VXERSWLPDO PXOWLXVHU GHWHFWRU ZKRVH FRPSOH[LW\ LV DOVR OLQHDU LQ WKH QXPEHU RI XVHUV 7KH UHFHLYHU UHTXLUHV NQRZOHGJH RI WKH UHFHLYHG VLJQDO VWUHQJWKV RI HDFK XVHU WKH FRGH VHTXHQFH RI HDFK XVHU DQG WLPH DQG SKDVH V\QFKURQL]DWLRQ ZLWK HDFK XVHU 7KH UHFHLYHU LV FRPSULVHG RI D EDQN RI PDWFKHG ILOWHUV RQH SHU XVHUf IROORZHG E\ D PXOWLVWDJH LQWHUIHUHQFH UHMHFWLRQ DOJRULWKP 7KH LQWHUIHUHQFH UHMHFWLRQ DOJRULWKP LV VXPPDUL]HG LQ WKH IROORZLQJ VHQWHQFHV &RQVLGHU
PAGE 27
WKH RXWSXW RI RQH RI WKH PDWFKHG ILOWHUV ,I D GHFLVLRQ LV PDGH RQ WKH ILOWHU RXWSXW LW PD\ EH FRUUXSWHG E\ WKH LQWHUIHUHQFH FDXVHG E\ WKH RWKHU XVHUV ,I WKH UHFHLYHU KDV SHUIHFW NQRZOHGJH RI WKH RWKHU XVHUVf UHFHLYHG VLJQDO HQHUJLHV DQG WUDQVPLWWHG GDWD VHTXHQFHV WKH UHFHLYHU FRXOG GLUHFWO\ FDOFXODWH WKH LQWHUIHUHQFH DW WKH RXWSXW RI WKH PDWFKHG ILOWHU DQG VXEWUDFW WKLV TXDQWLW\ IURP WKH ILOWHUfV RXWSXW 7KHQ WKH UHFHLYHU FRXOG PDNH D GHFLVLRQ RQ DQ LQWHUIHUHQFHIUHH TXDQWLW\ DQG WKH GHFLVLRQ PDGH IRU HDFK XVHU ZRXOG EH RSWLPDO +RZHYHU WKH UHFHLYHU GRHV QRW KDYH SHUIHFW NQRZOHGJH RI VXFK TXDQWLWLHV DQG WKHUHIRUH FDQ RQO\ IRUP LPSHUIHFW HVWLPDWHV RI WKH 0$, WHUP DW HDFK ILOWHU RXWSXW 7KH UHFHLYHU UHFRQVWUXFWV WKH HVWLPDWH RI WKH 0$, DW HDFK ILOWHU RXWSXW GXULQJ WKH PWK LQWHUYDO EDVHG RQ WKH GHFLVLRQV IRUPHG LQ WKH P fÂ§ OfWK LQWHUYDO 6LQFH WKH UHFRQVWUXFWLRQ RI WKH 0$, LV EDVHG RQ WKHVH GHFLVLRQV SHUIHFW FDQFHOODWLRQ RI WKH 0$, WHUP IURP WKH ILOWHU RXWSXW ZLOO QRW RFFXU DQG KHQFH WKH UHFHLYHU LV VXERSWLPDOf 7KH SUHYLRXV SDUDJUDSKV JLYH D EULHI RYHUYLHZ RI ZKDW KDV EHHQ GRQH LQ WHUPV RI PXOWLXVHU GHWHFWRUV IRU '6&'0$ $V PHQWLRQHG WKH FRPSOH[LW\ RI WKH WKHVH WHFKn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
PAGE 28
5 (>UUDfU+UDf@ FDQ EH ZULWWHQ DV An S 5 RSn >DrL^3NILNfD=NL^3NNf DIFSIFIFfDESIFMEf@ FU, f L L ZKHUH LV DQ 1 [ 1 ,GHQWLW\ PDWUL[ 1RWH WKDW 5 LV V\PPHWULF DQG LV DOVR SRVLWLYH GHILQLWH DV ORQJ DV D :KHQ 1 WKH HLJHQGHFRPSRVLWLRQ RI 5 FDQ EH H[SUHVVHG DV 5 $V $Q + (V (Q f $V LV D GLDJRQDO PDWUL[ RI WKH ODUJHVW HLJHQYDOXHV RI 5 DQG (V LV DQ $ [ PDWUL[ RI WKH FRUUHVSRQGLQJ HLJHQYHFWRUV /LNHZLVH $Q LV D GLDJRQDO PDWUL[ RI WKH 1 fÂ§ .f VPDOOHVW HLJHQYDOXHV DOO HTXDO WR Df RI 5 DQG (Q LV DQ 1 [ 1fÂ§.f PDWUL[ RI WKH FRUUHVSRQGLQJ HLJHQYHFWRUV 7KH VLJQDO VXEVSDFH LV GHILQHG WR EH WKH VXEVSDFH VSDQQHG E\ WKH VHW RI 6/NL3NKf DQG DN3NANf YHFWRUV 9$ f ^ .` 7KH QRLVH VXEVSDFH LV GHILQHG WR EH WKH RUWKRJRQDO FRPSOHPHQW WR WKH VLJQDO VXEVSDFH 7KH FROXPQV RI $V IRUP DQ RUWKRQRUPDO EDVLV IRU WKH VLJQDO VXEVSDFH DQG WKH FROXPQV RI $Q IRUP DQ RUWKRQRUPDO EDVLV IRU WKH QRLVH VXEVSDFH 7KH EDVLV RI WKH PXOWLSOH VLJQDO FODVVLILFDWLRQ 086,&f DOJRULWKP LV WKDW VLQFH WKH VHW RI ^DO D DLF` YHFWRUV DUH LQ WKH VLJQDO VXEVSDFH WKH\ DUH RUWKRJRQDO WR WKH QRLVH VXEVSDFH 7KHUHIRUH JLYHQ SHUIHFW NQRZOHGJH RI WKH QRLVH VXEVSDFH RQH FRXOG ILQG UN DV WKH VROXWLRQ WR (ADNL^3NKf fÂ§ 7\SLFDOO\ WKH UHFHLYHU GRHV QRW KDYH SHUIHFW NQRZOHGJH RI WKH QRLVH VXEVSDFH RU 5 +RZHYHU LW FDQ IRUP DQ HVWLPDWH RI 5 XVLQJ WKH VDPSOH FRUUHODWLRQ PDWUL[ GHILQHG DV $ 5P P f P *LYHQ 5Z WKH UHFHLYHU FDQ IRUP DQ HVWLPDWH RI WKH QRLVH VXEVSDFH Â‹Q DQ RSHUDWLRQ WKDW KDV FRPSOH[LW\ 1f 6WURP HW DO >@ XVH WKLV LGHD WR IRUP D WLPLQJ EDVHG RQ
PAGE 29
WKH 086,& DOJRULWKP 7KH\ ILQG DQ HVWLPDWH RI WKH NWK XVHUfV SURSDJDWLRQ GHOD\ DV _Â‹DWL7f+ >_Â‹mDDW7f__ n08 DUJ PP UH>RUf __DMIHLUf__ __DIF7f f 7KH LQGHSHQGHQW ZRUN SUHVHQWHG E\ %HQVOH\ DQG $D]KDQJ >@ LV VLPLODU EXW DOVR DGGUHVVHV WKH SUREOHP RI HVWLPDWLQJ WKH FKDQQHO JDLQ DQG SKDVH LQ D WLPHLQYDULDQW PXOWLSDWK FKDQQHO $V D QDWXUDO H[WHQVLRQ RI WKHLU HDUOLHU ZRUN 6WURP HW DO >@ FRQVLGHU XVLQJ WKH 086,& EDVHG HVWLPDWRU LQ D WLPHYDU\LQJ FKDQQHO 7KH 086,&EDVHG WLPLQJ HVWLPDWRU UHTXLUHV YHU\ OLWWOH VLGH LQIRUPDWLRQ LQ RUGHU WR IRUP LWV HVWLPDWH 7KH QXPEHU RI V\VWHP XVHUV DV ZHOO DV WKH VSUHDGLQJ FRGH WR EH XVHG LQ HTXDWLRQ ff LV UHTXLUHG :KLOH WKH FRPSOH[LW\ RI WKH DOJRULWKP LV 1f LW UHTXLUHV QR WUDLQLQJ SHULRG DQG FDQ EH XVHG WR HVWLPDWH HDFK XVHUf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n KRRG /60/f PHWKRG 7KLV PHWKRG PRGHOV WKH UHFHLYHG VLJQDO DV D NQRZQ WUDLQLQJ VHTXHQFH WKH GHVLUHG VLJQDOf DQG DOO RWKHU VLJQDOV LQFOXGLQJ WKH LQWHUIHULQJ VLJQDO DQG WKHUPDO QRLVH DV XQNQRZQ FRORUHG *DXVVLDQ QRLVH WKDW LV XQFRUUHODWHG ZLWK WKH GHVLUHG VLJQDO 7KH UHVXOWLQJ WLPLQJ HVWLPDWH LV IRXQG E\ URRWLQJ D VHFRQG RUGHU SRO\n QRPLDO 7KH FRHIILFLHQWV RI WKH SRO\QRPLDO DUH GHSHQGHQW RQ D PDWUL[ LQYHUVH 4 ZKHUH 4 LV UHODWHG WR WKH 50 PDWUL[ JLYHQ LQ HTXDWLRQ f ,Q RUGHU WR HYDOXDWH WKH PDWUL[ LQYHUVH 4 PXVW EH IXOO UDQN 7KLV PHDQV WKDW WKH /60/ WLPLQJ HVWLPDWH LV XQDYDLODEOH XQWLO WKH UHFHLYHU KDV REVHUYHG DW OHDVW 0 fÂ§ 1 VDPSOHV RI UPf
PAGE 30
$V DQ H[WHQVLRQ WR WKH /60/ WLPLQJ HVWLPDWRU /LX HW DO >@ KDYH SURSRVHG D VLPLODU WLPLQJ HVWLPDWLRQ VFKHPH NQRZQ DV WKH PXOWLSOHDQWHQQD VHQVRUVEDVHG HVWLn PDWRU 0$6(f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f LQ RUGHU IRU WKH 4 PDWUL[ WR KDYH IXOO UDQN %\ XVLQJ / DQWHQQD VHQVRUV WKH 0$6( DOJRULWKP LV DEOH WR DYHUDJH RYHU WKH RXWSXWV RI WKH VHQVRUV DQG KHQFH RQO\ UHTXLUHV /0 1 VDPSOHV RI UPf LQ RUGHU IRU WKH 4 PDWUL[ WR KDYH IXOO UDQN 6LQFH WKH 0$6( DOJRULWKP UHTXLUHV IHZHU WUDLQLQJ V\PEROV WKDQ WKH /60/ DOJRULWKP XQGHU FRQGLWLRQV ZKHUH WKH FKDQQHO LV WLPHYDU\LQJ WKH 0$6( KDV WKH DGYDQWDJH RYHU WKH /60/ DOJRULWKP >@ ,Q WKH QH[W FKDSWHU D WLPLQJ HVWLPDWLRQ DOJRULWKP LV SUHVHQWHG WKDW LV EDVHG RQ SURFHVVLQJ WKH ZHLJKWV RI WKH 006( UHFHLYHU ,Q WKLV GLVVHUWDWLRQ LW LV DVVXPHG WKDW WKH 006( UHFHLYHU XVHV DQ LQLWLDO WUDLQLQJ VHTXHQFH WR DGDSW WKH ZHLJKWV WR PLQLPL]H WKH 06( EHWZHHQ WKH ILOWHU RXWSXW DQG WKH GDWD VHTXHQFH IRU D VLQJOH XVHU 7KLV PHDQV WKDW D NQRZQ GDWD VHTXHQFH LV XVHG WR WUDLQ WKH 006( UHFHLYHU SULRU WR SHUIRUPLQJ GDWD GHWHFWLRQ +RQLJ HW DO >@ KDYH GHYHORSHG D VLQJOHXVHU GHWHFWRU EDVHG RQ PLQLPL]LQJ WKH PHDQRXWSXWHQHUJ\ 02(f RI DQ DGDSWLYH ILOWHU 1RWH WKDW WKH WHUP 02( LQ DFWXDOLW\ LV XVHG WR GHQRWH WKH YDULDQFH RI WKH ILOWHU RXWSXW 7KH SURSRVHG UHFHLYHU >@ UHTXLUHV NQRZOHGJH RI WKH WLPLQJ DQG VSUHDGLQJ VHTXHQFH RI WKH GHVLUHG XVHU 7KH ZHLJKWV RI WKH ILOWHU DUH H[SUHVVHG LQ D FDQRQLFDO IRUP DV ZPf &L [Pf ZKHUH &L LV WKH VSUHDGLQJ FRGH RI WKH GHVLUHG XVHU DQG [Pf LV
PAGE 31
RUWKRJRQDO WR &L 7KH YHFWRU [Pf LV XSGDWHG WR PLQLPL]H WKH YDULDQFH RI WKH ILOWHU RXWSXW 7KLV DGDSWLYH DOJRULWKP GRHV QRW UHTXLUH NQRZOHGJH RI WKH GHVLUHG XVHUfV GDWD VHTXHQFH DQG WKHUHIRUH LV NQRZQ DV D EOLQG DGDSWLYH DOJRULWKP 0DGKRZ >@ XVHV WKH EOLQG DGDSWLYH UHFHLYHU LQ DQ DGKRF IDVKLRQ WR IRUP DQ DGDSWLYH UHFHLYHU ZKLFK RQO\ UHTXLUHV NQRZOHGJH RI WKH GHVLUHG XVHUfV VSUHDGLQJ VHn TXHQFH 0DGKRZ XVHV D V\VWHP PRGHO VLPLODU WR WKDW SUHVHQWHG LQ 6HFWLRQ EXW REVHUYHV WZR ELWLQWHUYDOV ff RI WKH UHFHLYHG VLJQDO DQG WKHUHIRUH KDV 1 VDPn SOHG RXWSXWV RI WKH FKLSPDWFKHG ILOWHU FRPSULVH WKH UHFHLYHG YHFWRU 7KH SURSn DJDWLRQ GHOD\ IRU WKH GHVLUHG XVHU LV DVVXPHG WR EH RQH RI 1 K\SRWKHVHV 7? f ^7& 7& 7& 1 fÂ§ f7&` 7KH UHFHLYHG VLJQDO LV SURFHVVHG E\ D EDQN RI 1 SDUDOOHO EOLQG DGDSWLYH ILOWHUV RQH IRU HDFK Q K\SRWKHVLVf GXULQJ D EOLQG DFTXLVLn WLRQ F\FOH /HW WKH IWK ILOWHU EH WKH DGDSWLYH ILOWHU WKDW KDV WKH ORZHVW 02( DW WKH HQG RI WKH DFTXLVLWLRQ F\FOH $W WKH HQG RI WKH DFTXLVLWLRQ F\FOH WKH UHFHLYHU FKRRVHV WZR ILOWHUV WKH mWK ILOWHU DQG HLWKHU WKH L fÂ§ OfWK RU L OfWK ILOWHU WR IRUP WKH DGDSWLYH UHFHLYHU 7KH RXWSXW RI WKH ÂWK ILOWHU LV FRPELQHG ZLWK WKH RWKHU ILOWHU LQ DQ DGKRF IDVKLRQ LQ DQ DWWHPSW WR FRPELQH WKH WLPLQJ K\SRWKHVHV WKDW DUH FORVHVW WR WKH WUXH SURSDJDWLRQ GHOD\ IRU WKH GHVLUHG XVHU 8VLQJ WKLV PHWKRG 0DGKRZ LV DWWHPSWLQJ WR FUHDWH D UHFHLYHU VWUXFWXUH WKDW SHUIRUPV MRLQW DFTXLVLWLRQ DQG GHWHFWLRQ RI WKH GHn VLUHG XVHUfV VLJQDO +RZHYHU WKLV PHWKRG LV YHU\ FRPSOH[ DQG GRHVQfW UHDOO\ DFKLHYH V\QFKURQL]DWLRQ ZLWK WKH GHVLUHG XVHU ,Q WKH QH[W FKDSWHU ZH GHPRQVWUDWH ZK\ V\Qn FKURQL]DWLRQ LV LPSRUWDQW E\ FRQVLGHULQJ WKH HIIHFWV RI EHLQJ D KDOIFKLS DV\QFKURQRXV WR WKH GHVLUHG XVHU
PAGE 32
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f_@ ZKHUH HPf LV WKH GLIIHUHQFH EHWZHHQ D GHVLUHG UHVSRQVH DQG WKH ILOWHUfV RXWSXW 7KH YHFWRU ZKLFK PLQLPL]HV WKH PHDQVTXDUHG HUURU 06(f LV ZHOONQRZQ WR EH JLYHQ E\ WKH :LHQHU+RSI >@ HTXDWLRQ ZPf 5f[PfSUDf f ZKHUH 5Pf ( >UPfU+UDf@ DQG SPf (>GPfUPf@ DUH WKH DXWRFRUUHODWLRQ PDWUL[ DQG WKH VWHHULQJ YHFWRU UHVSHFWLYHO\ 7KH UHFHLYHG YHFWRU UPf FRUUHVSRQGV WR FKLSVDPSOHV RI RQH SHULRG RI WKH VSUHDGn LQJ VHTXHQFH 6LQFH WKH UHFHLYHU LV LQLWLDOO\ DV\QFKURQRXV WR WKH GHVLUHG XVHU LW LV SRVVLEOH IRU GDWD ELW WUDQVLWLRQV WR RFFXU DQ\ZKHUH ZLWKLQ WKH OHQJWK RI WKH UHFHLYHG YHFWRU ,Q RUGHU WR DYRLG WKHVH ELW WUDQVLWLRQV IRU WKH GHVLUHG XVHU ZH DVVXPH WKDW WKH
PAGE 33
r (VWLPDWLRQ W L ZPf } DOJRULWKP UWf &RQYHUW WR EDVHEDQG FKLS PDWFKHG ILOWHU UPf A $GDSWLYH ILOWHU ZPf + U PfZPf VJQ>5Hf@ G?^Pf U $GDSWLYH HPf J DOJRULWKP A L / G Pf )LJXUH &RGH DFTXLVLWLRQ ZLWK DQ DGDSWLYH UHFHLYHU GHVLUHG XVHUfV GDWD ELW LV D FRQVWDQW G?^Pf ,Q RWKHU ZRUGV WKH GHVLUHG XVHU ZLOO WUDQVPLW DQ DOO RQHV GDWD VHTXHQFH IRU SXUSRVHV RI FRGH DFTXLVLWLRQ $ VLGHFKDQQHO LV XVHG WR FRQWURO WKH DGGLWLRQ RI QHZ XVHUV LQWR WKH V\VWHP VR WKLV UHVWULFWLRQ LV QRW XQUHDOLVWLF %DVHG RQ WKH V\VWHP PRGHO SUHVHQWHG LQ 6HFWLRQ ZH VHH WKDW WKH ILOWHU LV RSHUDWLQJ LQ D VWDWLRQDU\ HQYLURQPHQW ,W LV DVVXPHG WKDW WKH XVHUVf GDWD VHTXHQFHV DQG FDUULHU SKDVHV DUH LQGHSHQGHQW RI HDFK RWKHU DQG DOVR LQGHSHQGHQW RI WKH DGGLWLYH ZKLWH *DXVVLDQ QRLVH 7KH DXWRFRUUHODWLRQ PDWUL[ DQG WKH VWHHULQJ YHFWRU DUH JLYHQ E\ WKH IROORZLQJ HTXDWLRQV 5 ( >UPfU+UDf@ DLSL fDI SL f W, An S = 3a >Dr3rf 6NftONL3N f DN3N fDeNSN f@ N = S DLSLfH[SMf f f ZKHUH LV DQ 1 [ 1 ,GHQWLW\ PDWUL[ 7KH 006( RFFXUV ZKHQ WKH ILOWHU ZHLJKWV DUH RSWLPDO DQG LV JLYHQ E\ PP S5 r3 f
PAGE 34
,Q RUGHU WR VROYH WKH :LHQHU+RSI HTXDWLRQ IRU WKH RSWLPDO ZHLJKW YHFWRU ZRSW ZH KDYH WR LQYHUW DQ 1 [ 79 PDWUL[ DQ RSHUDWLRQ WKDW KDV FRPSOH[LW\ RI 1f 6LQFH 5 LV D V\PPHWULF PDWUL[ LW FDQ EH H[SUHVVHG DV 5 9$9U f ZKHUH 9 LV D PDWUL[ ZKRVH FROXPQV FRQWDLQ WKH QRUPDOL]HG HLJHQYHFWRUV RI 5 DQG $ LV D GLDJRQDO PDWUL[ WKDW FRQWDLQV WKH FRUUHVSRQGLQJ HLJHQYDOXHV 7KLV QRWDWLRQ LV UHIHUUHG WR DV WKH HLJHQGHFRPSRVLWLRQ RI WKH 5 PDWUL[ 8VLQJ WKH DERYH QRWDWLRQ ZH FDQ WKHQ H[SUHVV WKH LQYHUVH RI WKH DXWRFRUUHODWLRQ PDWUL[ DV 5 9$9U f ,QVWHDG RI FDOFXODWLQJ 5 GLUHFWO\ ZH QRWH WKDW LW FDQ EH VKRZQ WKDW DQ HLJHQYHFWRU RI 3 5 fÂ§ LV DOVR DQG HLJHQYHFWRU RI 5 )RU HDFK HLJHQYHFWRU RI 3 DGGLQJ D WR WKH FRUUHVSRQGLQJ HLJHQYDOXH SURGXFHV WKH FRUUHVSRQGLQJ HLJHQYDOXH IRU WKH PDWUL[ 5 7KHUHIRUH LI ZH ILQG WKH HLJHQGHFRPSRVLWLRQ RI 3 ZH LPPHGLDWHO\ KDYH WKH HLJHQGHFRPSRVLWLRQ RI 5 ,Q HTXDWLRQ f ZH QRWH WKDW HDFK DV\QFKURQRXV XVHU FRQWULEXWHV DW PRVW WZR RXWHU SURGXFW WHUPV LQ WKH H[SUHVVLRQ IRU 3 ,I 1 WKHQ WKH 3 PDWUL[ ZLOO QRW EH IXOOUDQN DQG ZLOO KDYH DW PRVW QRQ]HUR HLJHQYDOXHV 0LOOHU >@ XVHG WKLV IDFW WR VLPSOLI\ WKH FDOFXODWLRQ RI WKH HLJHQGHFRPSRVLWLRQ RI 3 E\ ZRUNLQJ RQ D [ PDWUL[ LQVWHDG RI RQ D 1 [ 1 PDWUL[ )RU ODUJH UDWLRV RI 1.f WKLV WHFKQLTXH SURGXFHV D VLJQLILFDQW UHGXFWLRQ LQ WKH FRPSOH[LW\ RI FDOFXODWLQJ 5B +RZHYHU LQ WKDW ZRUN WKH GHVLUHG XVHU ZDV DVVXPHG WR EH V\QFKURQRXV 7? f ZLWK WKH UHFHLYHU VR RQH PXVW VOLJKWO\ PRGLI\ 0LOOHUfV QRWDWLRQ LQ RUGHU WR SURFHVV WKH GHVLUHG XVHU LQ WKH PRUH JHQHUDO DV\QFKURQRXV FDVH 2I FRXUVH ZKHQ 1 WKLV WHFKQLTXH LV QRW YDOLG DQG RQH PXVW ILQG WKH HLJHQGHFRPSRVLWLRQ RI DQ 1 [ 1 PDWUL[ HLWKHU 5 RU 3f
PAGE 35
,Q RUGHU WR LOOXVWUDWH WKH GHSHQGHQFH RI WKH ILOWHU ZHLJKW YHFWRU RQ WKH DPRXQW RI PXOWLDFFHVV LQWHUIHUHQFH ZH ZLOO REVHUYH WKH RSWLPDO ILOWHU ZHLJKW YHFWRU IRU WZR FDVHV )RU WKH ILUVW FDVH OHW WKHUH EH RQO\ RQH FKLSV\QFKURQRXV V\VWHP XVHU ZKLFK LPSOLHV WKDW UPf FIfOf H[S MLf 1P 8VLQJ WKH :LHQHU+RSI HTXDWLRQ LW LV IRXQG WKDW ZRSW FM3OfH[SM!Lf __FL__ f $VVXPLQJ &L KDV JRRG DXWRFRUUHODWLRQ SURSHUWLHV WKLV UHVXOW VKRZV WKDW ZH FRXOG FRQFHLYDEO\ IRUP DQ HVWLPDWH RI SL WKH SURSDJDWLRQ GHOD\ RI WKH GHVLUHG XVHU E\ ILQGLQJ WKH SHDN RI WKH PDJQLWXGH RI WKH FURVVFRUUHODWLRQ EHWZHHQ &L DQG ZRSW )RU WKH VHFRQG FDVH D VHFRQG V\QFKURQRXV XVHU LV DGGHG WR WKH V\VWHP VXFK WKDW UPf FA3Of H[S MLf GPf\33LF H[S MGf 1P )RU WKLV VHFRQG H[DPSOH WKH RSWLPDO ILOWHU ZHLJKW YHFWRU LV ZRSW F"Of &f H[S M2Lf ??FL?? D &AFL3Of WS Lf & & Df Ef ,1, Df D 1RWH WKDW LV ERXQGHG LQ PDJQLWXGH E\ WKH YDOXH RI WKH FRUUHODWLRQ FRHIILFLHQW RI WKH WZR VSUHDGLQJ FRGHV DQG LV WKHUHIRUH H[SHFWHG WR EH PXFK OHVV WKDQ RQH ,Q IDFW LQ WKH OLPLW 33? fÂ§g ZH VHH WKDW ZRSW LV WKH SURMHFWLRQ RI RQWR D YHFWRU RUWKRJRQDO WR F ,Q WKH LGHDO FDVH WKDW F>3Of LV RUWKRJRQDO WR F WKH UHFHLYHU FRPSOHWHO\ UHMHFWV WKH VHFRQG XVHU HYHQ WKRXJK LW LV UHFHLYHG DW D PXFK KLJKHU SRZHU OHYHO ,Q WKH JHQHUDO FDVH IRU FRGHV ZLWK JRRG FRUUHODWLRQ SURSHUWLHV FAF>3Of LV VPDOOf WKH ILOWHU VKRXOG EH DEOH WR VXSSUHVV WKH VHFRQG XVHU ,Q WKH JHQHUDO FDVH WKH DGGLWLRQ RI WKH VHFRQG XVHU FDXVHV D VPDOO SHUWXUEDWLRQ RI WKH ILOWHU ZHLJKW YHFWRU DERXW WKH VROXWLRQ JLYHQ E\ HTXDWLRQ f :H VKRXOG VWLOO EH DEOH WR IRUP DQ HVWLPDWH RI S[ DV GHVFULEHG DERYH
PAGE 36
(TXDWLRQ Df LV XVHG DV D EDVLF H[DPSOH RI ZK\ D WLPLQJ HVWLPDWRU EDVHG RQ SURFHVVLQJ WKH ILOWHU ZHLJKW YHFWRU VKRXOG HYHQ EH FRQVLGHUHG $V PRUH XVHUV DUH LQn FOXGHG LQ WKH '6&'0$ V\VWHP DGGLWLRQDO WHUPV GXH WR WKH VSUHDGLQJ FRGHV DUH DGGHG WR WKH H[SUHVVLRQ IRU ZRSW ,Q RUGHU WR LOOXVWUDWH WKH GHSHQGHQFH RI ZRSW RQ WKH GHVLUHG XVHUfV VSUHDGLQJ FRGH LQ WKH DERYH H[DPSOH DQG WR VLPSOLI\ WKH UHVXOWLQJ H[n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s 7KHUHIRUH ZH VKRXOG SUHIHU WR RSHUDWH WKH UHFHLYHU V\QFKURQRXV WR WKH GHVLUHG XVHUfV UHFHLYHG VLJQDO ,Q RUGHU WR H[SDQG RQ WKLV LGHD ZH SUHVHQW WKUHH EULHI FDVHV WKDW GHPRQVWUDWH VRPH RI WKH FRVWV DVVRFLDWHG ZLWK XVLQJ WKH 006( UHFHLYHU DV\QFKURQRXV WR WKH GHVLUHG XVHU )RU WKH ILUVW FDVH FRQVLGHU XVLQJ WKH 006( UHFHLYHU WR GHPRGXODWH GDWD LQ D UPf ZRSW DO3OL AOf 1P DL3LÂLf OODL3LÂLf__ FU VLQJOH XVHU V\VWHP ZKHUH f f
PAGE 37
&RQVLGHU WKH VLJQDOWRQRLVH UDWLR RI WKH RXWSXW VWDWLVWLF ] fÂ§ U+UDfZRSW ZKLFK LV GHILQHG DV 615 (>]` 9DU]f 1RWH WKDW (>]f DI SL ÂLfZRSW DQG WKH H[SUHVn VLRQ IRU 9DU>]@ LV HYDOXDWHG XVLQJ 9DU>]@ en>ZAWUPfULLPfZRSW@ fÂ§ (>]@ 'LUHFW VXEVWLWXWLRQ RI WKH DSSURSULDWH WHUPV LQWR WKLV H[SUHVVLRQ SURGXFHV 7__DLM!LÂf__ OODOSÂOf__ f Qf D(>]? O_DLSLLf__ HU 7KH UHVXOW RI WKHVH H[SUHVVLRQV LV WKDW 615 __DLSL ÂLf__FUf ,I ZH DVVXPH WKDW WKH VSUHDGLQJ FRGH KDV JRRG DXWRFRUUHODWLRQ SURSHUWLHV VXFK WKDW WKH WHUPV LQYROYLQJ FIFAr DUH QHJOLJLEOH ZKHQ FRPSDUHG WR RWKHU WHUPV LQ __DLSL ÂLf__ ZH FDQ DSSUR[LPDWH WKH RXWSXW VLJQDOWRQRLVH UDWLR DV 615 > IWf 6M@ 1 D f ZKHUH D 1(E?If DV SUHYLRXVO\ GHILQHG :KHQ ÂL LV QRQ]HUR WKHUH LV DQ HIIHFWLYH ORVV LQ WKH RXWSXW 615 RI WKH RXWSXW VWDWLVWLF ] ,I ÂL LV XQLIRUPO\ GLVWULEXWHG RYHU WKH LQWHUYDO >f WKHQ WKH DYHUDJH ORVV LQ WKH RXWSXW 615 LV G%f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
PAGE 38
DFKLHYLQJ FKLS V\QFKURQLVP ZLWK WKH GHVLUHG VLJQDO >@ ,W LV LPSRUWDQW WR DVN KRZ WKH SHUIRUPDQFH RI WKH 006( UHFHLYHU LV DIIHFWHG E\ WKH ODFN RI FKLS V\QFKURQLVP 7R DGGUHVV WKLV LVVXH ZH FDOFXODWHG WKH PcQ RI WKH 006( ILOWHU DV D IXQFWLRQ RI WKH SURSDJDWLRQ GHOD\ RI WKH GHVLUHG XVHU ZKHQ WKH 615 ZDV G% DQG WKH OHQJWK RI WKH UHFHLYHG YHFWRU DQG ILOWHU ZDV VHW WR 1 FKLS VDPSOHV 7KLV UHVXOW LV VKRZQ LQ )LJXUH IRU SURSDJDWLRQ GHOD\V LQ WKH UDQJH RI >f FKLS LQWHUYDOV 6LPLODU UHVXOWV DUH REWDLQHG IRU SURSDJDWLRQ GHOD\V LQ WKH UDQJH RI >f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ÂL IDOOV LQ WKH UHJLRQ f WKH 006( RI WKH ILOWHU LV VHQVLWLYH WR WKH SURSDJDWLRQ GHOD\ RI WKH GHVLUHG XVHU 7KHUHIRUH VRPH PHWKRG LV UHTXLUHG IRU FRGH DFTXLVLWLRQ DQG FRGH WUDFNLQJ HYHQ ZKHQ 1 FKLS VDPSOHV DUH XVHG LQ WKH 006( UHFHLYHU VWUXFWXUH $V D ILQDO H[DPSOH ZH ZLOO FRQVLGHU WKH HIIHFW RI DV\QFKURQRXV RSHUDWLRQ ZLWK UHVSHFW WR WKH GHVLUHG XVHUf RQ WKH UHVXOWLQJ SUREDELOLW\ RI ELW HUURU 7KLV H[DPSOH ZLOO VKRZ WKDW FRGH V\QFKURQL]DWLRQ LV UHTXLUHG IRU WKH 006( UHFHLYHU LI RQH LV WR PD[LPL]H WKH V\VWHP FDSDFLW\ 7KH IROORZLQJ EULHIO\ H[SODLQV WKH RXWOLQH RI WKH DQDO\n VLV XVHG WR GHULYH WKH SUREDELOLW\ RI ELW HUURU 6LQFH ZH DUH LQWHUHVWHG LQ ELW HUURUV ZH DVVXPH WKDW WKH ILOWHU LV DW VWHDG\VWDWH DQG WKDW WKH GHVLUHG XVHU LV WUDQVPLWWLQJ D UDQGRP GDWD VHTXHQFH :H WKHQ PRGHO WKH RXWSXW RI WKH 006( UHFHLYHU ZAWUPf DV D FRPSOH[ *DXVVLDQ UDQGRP YDULDEOH DQG GHULYH DQ H[SUHVVLRQ IRU WKH SUREDELOLW\ RI ELW HUURU 7KH GHWDLOV RI WKLV GHULYDWLRQ DUH JLYHQ LQ $SSHQGL[ $ 0RGHOLQJ WKH RXWSXW RI WKH 006( UHFHLYHU DV D *DXVVLDQ UDQGRP YDULDEOH KDV EHHQ UHFHQWO\ MXVn WLILHG E\ 3RRU DQG 9HUGLL >@ ZKR KDYH FRPSDUHG WKH ELW HUURU UDWH RI WKH 006(
PAGE 39
3URSDJDWLRQ 'HOD\ RI 8VHU LQ &KLSV )LJXUH 0LQLPXP PHDQVTXDUHG HUURU DV D IXQFWLRQ RI WKH SURSDJDWLRQ GHOD\ RI XVHU UHFHLYHU YHUVXV WKH DQDO\WLFDO ELW HUURU UDWH RI WKH 006( UHFHLYHU ZKHQ WKH UHFHLYHU RXWSXW LV PRGHOHG DV D *DXVVLDQ UDQGRP YDULDEOH ,W ZDV IRXQG WKDW WKH *DXVVLDQ DSSUR[LPDWLRQ ZDV TXLWH JRRG IRU YDULRXV QXPEHU RI XVHUV DQG YDULRXV OHYHOV RI PXOWLn DFFHVV LQWHUIHUHQFH ,Q DGGLWLRQ VHYHUDO RWKHUV KDYH IRXQG WKDW WKLV DSSUR[LPDWLRQ SURYLGHV UHDVRQDEOH UHVXOWV >@ 2QFH WKH '6&'0$ HQYLURQPHQW LV FRQILJXUHG XVHUV GHOD\V SKDVHV DQG SRZHU OHYHOV ZH FDQ QXPHULFDOO\ VROYH IRU PLQ :H WKHQ XVH HTXDWLRQV $f DQG $ f WR ILQG WKH SUREDELOLW\ RI ELW HUURU %\ YDU\LQJ WKH '6&'0$ HQYLURQPHQW ZH PD\ REVHUYH WKH HIIHFWV RI DV\QFKURQRXV RSHUDWLRQ RQ WKH V\VWHP FDSDFLW\ )RU SXUSRVHV RI WKLV H[DPSOH ZH VHW WKH 615 WR G% DQG OHW DOO LQWHUIHULQJ XVHUV EH UHFHLYHG DW D SRZHU OHYHO G% DERYH WKH GHVLUHG XVHU $ VHW RI VSUHDGLQJ FRGHV IRU XVHUV ZHUH VHOHFWHG DW UDQGRP IURP D VHW RI *ROG FRGHV >@ XVLQJ 1 fÂ§ FKLSVELW $ VHW RI SURSDJDWLRQ GHOD\V ZHUH FKRVHQ IURP D XQLIRUP GLVWULEXWLRQ RYHU WKH LQWHUYDO >f $ VHW RI FDUULHU SKDVHVKLIWV ZHUH VHOHFWHG IURP D XQLIRUP GLVWULEXWLRQ RYHU
PAGE 40
WKH LQWHUYDO > Qf ,I ZH ZDQWHG WR DQDO\]H D .XVHU V\VWHP WKH ILUVW HOHPHQWV RI HDFK RI WKH VSUHDGLQJ FRGH GHOD\ DQG SKDVHVKLIW VHWV ZRXOG EH XVHG WR FUHDWH WKH '6&'0$ HQYLURQPHQW )RU WKH FDVH ZKHQ WKH UHFHLYHU ZDV RQHKDOI FKLS RXW RI V\QFK ZLWK WKH GHVLUHG XVHU XVLQJ WKH *DXVVLDQ DSSUR[LPDWLRQ ZH IRXQG WKDW V\VWHP XVHUV ZRXOG SURGXFH D b ELWHUURU UDWH DW WKH UHFHLYHU $ WRWDO RI V\VWHP XVHUV ZRXOG SURGXFH D b HUURUUDWH ,I WKH UHFHLYHU ZDV V\QFKURQRXV ZLWK WKH GHVLUHG XVHU D WRWDO RI XVHUV ZRXOG SURGXFH D b ELWHUURU UDWH /LNHZLVH D WRWDO RI XVHUV ZRXOG SURGXFH D b ELWHUURU UDWH 7KLV VKRZV WKDW D VLJQLILFDQW LPSURYHPHQW LQ WKH V\VWHP FDSDFLW\ FDQ EH DFKLHYHG LI WKH UHFHLYHU LV V\QFKURQRXV ZLWK WKH GHVLUHG XVHU :KLOH WKHVH UHVXOWV DUH RQO\ YDOLG IRU FDVH ZH H[DPLQHG WKH\ DUH VXIILFLHQW WR GHPRQVWUDWH WKH FRVWV RI RSHUDWLQJ WKH UHFHLYHU DV\QFKURQRXV WR WKH GHVLUHG XVHU $GDSWLYH )LOWHUV DQG WKH (VWLPDWLRQ $OJRULWKP 7KH FRPSOH[LW\ RI ILQGLQJ WKH RSWLPDO ZHLJKWV JLYHQ E\ WKH :LHQHU+RSI HTXDWLRQ LV 1f +RZHYHU WKH UHFHLYHU UHTXLUHV NQRZOHGJH RI WKH DXWRFRUUHODWLRQ PDWUL[ 5 DQG WKH VWHHULQJ YHFWRU S 7\SLFDOO\ WKHVH SDUDPHWHUV DUH HVWLPDWHG E\ WKH UHFHLYHU EDVHG RQ REVHUYDWLRQV RI WKH UHFHLYHG VLJQDO ,Q SUDFWLFH WKH ZHLJKWV RI WKH ILOWHU DUH FKRVHQ DGDSWLYHO\ DFFRUGLQJ WR VRPH LPSOHPHQWDWLRQ ZKLFK XVXDOO\ WDNHV WKH IRUP RI HLWKHU WKH OHDVW PHDQ VTXDUH /06f RU WKH UHFXUVLYH OHDVWVTXDUHV 5/6f DOJRULWKP 7KH /06 DOJRULWKP KDV ORZ FRPSOH[LW\ 1f GRHV QRW UHTXLUH DQ\ PDWUL[ LQYHUVLRQ RSHUDWLRQV DQG LV HDV\ WR LPSOHPHQW 7KH 5/6 DOJRULWKP LV PRUH FRPSOH[ 1f DQG SHUIRUPV DQ LWHUDWLYH PDWUL[ LQYHUVLRQ E\ XVLQJ WKH 0DWUL[ ,QYHUVLRQ /HPPD >@ 6LQFH WKH 5/6 DOJRULWKP LV PRUH FRPSOH[ LW LV QRW VXUSULVLQJ WKDW LW W\SLFDOO\ FRQYHUJHV IDVWHU WKDQ WKH /06 DOJRULWKP :KLOH WKHVH DOJRULWKPV ZLOO QRW SURGXFH WKH VDPH ZHLJKW YHFWRU JLYHQ E\ WKH :LHQHU+RSI HTXDWLRQ DV WKH ILOWHU DSSURDFKHV VWHDG\VWDWH WKH\ FDQ SURYLGH UHDVRQDEOH DSSUR[LPDWLRQV WR WKH RSWLPDO ZHLJKW YHFWRU
PAGE 41
2XU EDVLF LGHD LV WR GHULYH D FRGH WLPLQJ HVWLPDWLRQ DOJRULWKP IRU D VLQJOHXVHU V\VWHP DQG WKHQ VWXG\ WKH SHUIRUPDQFH RI WKH HVWLPDWRU LQ D PXOWLXVHU HQYLURQn PHQW 6LQFH WKH UHFHLYHU LV LQLWLDOO\ DV\QFKURQRXV WR WKH GHVLUHG XVHU UPf DL3LL Âf H[3 Mf 1P ,Q RUGHU WR GHULYH D FRGH HVWLPDWLRQ DOJRULWKP ZH ZRXOG OLNH WR KDYH DQ H[DFW UHSUHVHQWDWLRQ RI WKH VWDWLVWLFV RI WKH ILOWHU ZHLJKW YHFWRU DV WKH ILOWHU DGDSWV 7KLV DSSHDUV WR EH LQWUDFWDEOH ,QVWHDG ZH PRGHO WKH ILOWHU FRHIn ILFLHQWV DV LQGHSHQGHQW MRLQWO\ FRPSOH[ *DXVVLDQ UDQGRP YDULDEOHV ZLWK D PHDQ RI "DLSL Ef H[S MLf DQG HTXDO YDULDQFHV ZKHUH c LV VRPH XQNQRZQ FRQVWDQW :H FKRRVH WKLV H[SUHVVLRQ IRU WKH PHDQ VLQFH IRU WKLV VFHQDULR WKH RSWLPDO ZHLJKW YHFWRU LV SURSRUWLRQDO WR DLSL ÂLf H[S MLf &OHDUO\ WKLV LV QRW DQ H[DFW GHVFULSWLRQ RI WKH VWDWLVWLFV RI ZPf DQG RXU UHVXOWLQJ WLPLQJ HVWLPDWRU ZLOO QRW SURGXFH DQ RSWLPDO HVn WLPDWH RI WKH GHVLUHG XVHUfV SURSDJDWLRQ GHOD\ 7KH VWDWLVWLFV RI WKH ZHLJKW YHFWRU KDYH EHHQ REVHUYHG WKURXJK VLPXODWLRQVf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` ZKLFK PLQLPL]H J3Sf __ZPf fÂ§ DLS Â•f H[S Mf?? f ZLWK S H ^ 1 ` DQG 1RWH WKDW HYHQ WKRXJK RXU DVVXPHG PHDQ LV EDVHG RQ RSWLPDO YDOXHV ZH ZLOO XVH WKLV IXQFWLRQ WR PDNH HVWLPDWHV DIWHU HYHU\ UHFHLYHU ELW LQWHUYDO 7KLV DSSURDFK LV WDNHQ EHFDXVH LW FDQ EH VKRZQ WKDW ZKHQ WKH /06 DOJRULWKP LV XVHG WKH PXOWLXVHU WUDQVLHQW PHDQ ILOWHU ZHLJKW YHFWRU (>ZPf@ LV KLJKO\ FRUUHODWHG ZLWK WKH VLQJOH XVHU RSWLPDO ZHLJKW YHFWRU ZKHQ D VPDOO
PAGE 42
DGDSWDWLRQ VWHSVL]H LV XVHG 'XH WR WKH FRQVWUDLQW RQ WKH YDOXH RI S J S f LV QRW GLIIHUHQWLDEOH ZLWK UHVSHFW WR S 7R PLQLPL]H JSf ZH PXVW ILQG WKH VROXWLRQV WR GJSfG[ IRU [ f IRU DOO SRVVLEOH YDOXHV RI S 7KHUHIRUH HDFK YDOXH RI S SURGXFHV D FRUUHVSRQGLQJ VHW RI FDQGLGDWHV IRU WKH VXEVHW ^ Â£` 2XU HVWLPDWLRQ DOJRULWKP LWHUDWHV WKURXJK DOO SRVVLEOH YDOXHV RI S DQG SURGXFHV D OLVW RI DOO RI WKH FRUUHVSRQGLQJ FDQGLGDWHV %\ FKRRVLQJ WKH VHW ^S ` RXW RI WKLV OLVW ZKLFK PLQLPL]HV J^S6f ZH ILQG WKH JOREDO PLQLPXP RI JSf 7KH VROXWLRQV WR GJS6fG[ IRU [ DUH JLYHQ E\ LSf DUJ DIS6fZPff 33f DISÂfZPf_ OODLSÂf__ Df Ef 6XEVWLWXWLRQ RI WKHVH H[SUHVVLRQV LQWR WKH SUHYLRXV HTXDWLRQ SURGXFHV WKH IROORZLQJ WZR H[SUHVVLRQV J^SfSfSf __ZUDf__ I^Sf B _DISÂfZPf_ > OODU S Âf __ Df Ef 7KHVH HTXDWLRQV VKRZ WKDW PLQLPL]LQJ JSfSfSf FRUUHVSRQGV WR ILQGLQJ WKH VHW ^S ` ZKLFK PD[LPL]H WKH FRVW IXQFWLRQ ISf 6LQFH WKH HVWLPDWH RI S? PXVW EH DQ LQWHJHU WKH FRVW IXQFWLRQ ISf LV D SLHFH ZLVH FRQWLQXRXV IXQFWLRQ 0D[LPL]LQJ IS f ZLWK UHVSHFW WR SURGXFHV D TXDGUDWLF HTXDWLRQ ZKRVH FRHIILFLHQWV GHSHQG RQ WKH YDOXH RI S 7KHUHIRUH IRU HDFK SRVVLEOH YDOXH RI S ZH PXVW ILQG WKH URRWV RI WKH TXDGUDWLF HTXDWLRQ 'SfS 'LSfS 'Sf ZKHUH WKH FRHIILFLHQWV DUH JLYHQ E\ WKH IROORZLQJ VHW RI HTXDWLRQV 'Sf 1$LSf &$Sf 'L^Sf 1$Sf &$Sf 'Sf &$Sf $Sff Df Ef Ff
PAGE 43
& fÂ§ 1 FIFA SfL 03f _Z+PfFL3 $LSf 5H FL3ff7ZPfZ+PfF_OSf $Sf $Sf _Z+PfFL3f_ $[Sf $Sf Gf Hf If Jf %\ GHILQLWLRQ WKH YDOXH RI Â PXVW VDWLVI\ WKH FRQGLWLRQ 6 VR ZH PXVW LJQRUH DQ\ VROXWLRQV IURP WKH SUHYLRXV HTXDWLRQ ZKLFK IDOO RXWVLGH RI WKLV UHJLRQ ,Q RUGHU WR FODULI\ KRZ ZH IRUP RXU HVWLPDWH RI WKH SURSDJDWLRQ GHOD\ IRU WKH GHVLUHG XVHU ZH SUHVHQW WKH IROORZLQJ DOJRULWKP /HW 7 UHSUHVHQW WKH VHW RI FRVWV FRUUHVSRQGLQJ WR WKH VHW RI FDQGLGDWH HVWLPDWHV RI S f GHQRWHG E\ 8 :H FDQ ILQG WKH PD[LPXP RI IS6f DV IROORZV 6WHS /HW 7 ^f f f I1 f` 6WHS /HW 8 ^ 1 ` 6WHS )RU S 1 fÂ§ GR WKH IROORZLQJ Df &RPSXWH WKH FRHIILFLHQWV 'Sf 'LSf 'Sf Ef 6ROYH IRU WKH WZR URRWV RI WKH TXDGUDWLF HTXDWLRQ SL DQG S Ff ,I SL IRU L e ^` DGG WKH FRVW S SÂf WR WKH VHW 7 DQG DGG S SLf WR WKH VHW 8 6WHS /HW 8N GHQRWH WKH NWK HOHPHQW RI WKH VHW 8 7KHQ I 8N ZKHUH N PD[ ` M 6WHS 7KH HVWLPDWHV RI S? DQG ÂL DUH JLYHQ E\ SLUDf `IM DQG LPf fÂ§ I fÂ§ >IM ZKHUH URXQGV [ WR WKH QHDUHVW LQWHJHU WRZDUGV ]HUR :H SORW WKH FRVW IXQFWLRQ S Â£f IRU WZR GLIIHUHQW VFHQDULRV LQ )LJXUHV DQG XQGHU WKH DVVXPSWLRQ WKDW WKH UHFHLYHU LV RSHUDWLQJ XVLQJ WKH RSWLPDO ILOWHU ZHLJKWV )LJXUH VKRZV WKH FRVW IXQFWLRQ IRU D VLQJOHXVHU V\VWHP ZKHQ LT 7& DQG (E$LR ZDV G% 7KLV SORW VKRZV WKDW WKH WLPLQJ HVWLPDWH ZLOO EH IL 7& ZKLFK LV D SHUIHFW HVWLPDWH 6LQFH ZH KDYH GHYHORSHG WKH WLPLQJ HVWLPDWRU IRU D VLQJOHXVHU V\VWHP WKLV UHVXOW LV QRW VXUSULVLQJ 2I FRXUVH ZKHQ WKH ILOWHU ZHLJKWV DUH IURP WKH
PAGE 44
)LJXUH 3ORW RI WKH FRVW IXQFWLRQ IS f IRU D VLQJOHXVHU V\VWHP /06 RU 5/6 DOJRULWKPV ZH H[SHFW VRPH QRLVH LQ WKLV HVWLPDWH GXH WR WKH QRLVH LQ WKH ILOWHU ZHLJKWV $OVR WKH UDWH RI FRQYHUJHQFH RI WKH DGDSWLYH ILOWHU WR LWV VWHDG\VWDWH YDOXH ZLOO GHWHUPLQH KRZ ORQJ WKH ILOWHU KDV WR DGDSW EHIRUH D UHOLDEOH HVWLPDWH RI 7? FDQ EH PDGH 7KLV LVVXH ZLOO EH DGGUHVVHG LQ WKH QH[W FKDSWHU )LJXUH VKRZV WKH FRVW IXQFWLRQ IS f IRU D PRUH VHYHUH HQYLURQPHQW ,Q WKLV ILJXUH LQWHUIHULQJ XVHUV KDYH EHHQ DGGHG WR WKH V\VWHP HDFK UHFHLYHG DW D SRZHU OHYHO G% DERYH WKH GHVLUHG XVHU 6LQFH WKH PD[LPXP YDOXH RI ISf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
PAGE 45
)LJXUH 3ORW RI WKH FRVW IXQFWLRQ S f IRU D ILIWHHQXVHU V\VWHP DUH SXUHO\ UHDO 8VLQJ WKLV DVVXPSWLRQ WKH DXWRFRUUHODWLRQ PDWUL[ DQG WKH VWHHULQJ YHFWRU IRU WKH DGDSWLYH ILOWHU DUH ZULWWHQ DV 5 ( >UPfU+Pf@ DLSLLLfDISLLf U, 3N f 7(7 >DLL3IFAIFfD/L3IF!f DrSMNIFfDESIFOIFf@ 3L IF 3 SLLOf f ZKHUH LV DQ 79 [ 1 ,GHQWLW\ PDWUL[ +DYLQJ WKH ZHLJKWV RI WKH ILOWHU UHDO OHDGV WR D VPDOO UHGXFWLRQ LQ WKH FRPSOH[LW\ RI IRUPLQJ D WLPLQJ HVWLPDWH IRU WKH GHVLUHG XVHU 7KH GHULYDWLRQ RI WKH DOJRULWKP LV VLPLODU WR WKDW SUHYLRXVO\ SUHVHQWHG 0D[LPL]LQJ WKH FRVW IXQFWLRQ IS 6f SURGXFHV D TXDGUDWLF HTXDWLRQ ZKRVH URRWV PD\ EH ZULWWHQ LQ FORVHG IRUP +RZHYHU RQH RI WKH URRWV SURGXFHV D WULYLDO UHVXOW DQG WKHUHIRUH FDQ EH LJQRUHG 7KH RWKHU URRW LV JLYHQ
PAGE 46
E\ WKH IROORZLQJ HTXDWLRQ 1Z7PfFIf FIFfZUPfFSf LJ A 1 fÂ§ FIFLffZ7PfFSf ZUPf&L3ff %DVHG RQ WKH DERYH HTXDWLRQ WKH WLPLQJ HVWLPDWLRQ DOJRULWKP LV WKHQ PRGLILHG DV IROORZV 6WHS 3HUIRUP VWHSV DQG RI WKH SUHYLRXV DOJRULWKP 6WHS )RU S 79 fÂ§ GR WKH IROORZLQJ Df 8VH HTXDWLRQ f WR FRPSXWH WKH FRHIILFLHQW SL Ef ,I SL O DGG WKH FRVW IS SLf WR WKH VHW 7 DQG DGG S S?f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
PAGE 47
H[DFW VDPH HVWLPDWLRQ DOJRULWKP WKDW ZH KDYH SUHVHQWHG LQ WKLV FKDSWHU 7KH GLIIHUn HQFH EHWZHHQ RXU WHFKQLTXH DQG WKH FRUUHODWRU LV WKDW ZH IRUP DQ HVWLPDWH EDVHG RQ REVHUYLQJ WKH ILOWHU ZHLJKWV ZKLOH WKH FRUUHODWRU IRUPV DQ HVWLPDWH E\ REVHUYLQJ VDPn SOHV RI WKH UHFHLYHG VLJQDO 7KH DGGLWLRQDO FRPSOH[LW\ DGGHG WR WKH UHFHLYHU E\ RXU WHFKQLTXH LV WKDW RI UXQQLQJ WKH DGDSWLYH ILOWHU /06 $Uf 5/6 1ff 6LQFH WKH VDPH DGDSWLYHILOWHU UHFHLYHU VWUXFWXUH FDQ EH XVHG IRU GHWHFWLRQ DQG ZH KDYH GHPRQn VWUDWHG WKDW V\QFKURQL]DWLRQ LV GHVLUHG LQFOXGLQJ WKH WLPLQJ HVWLPDWRU LQ WKH UHFHLYHU VKRXOG EH DQ DFFHSWDEOH LQFUHDVH LQ WKH RYHUKHDG RI LPSOHPHQWLQJ WKH UHFHLYHU
PAGE 48
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f IRU D ILFWLWLRXV WLPLQJ HVWLPDWRU RQFH VHYHUDO DVVXPSWLRQV DUH PDGH ZLWK UHVSHFW WR WKH VWDWLVWLFDO SURSHUWLHV RI WKH /06 ILOWHU ZHLJKWV ,Q RUGHU WR WHVW WKH TXDOLW\ RI WKHVH DQDO\WLFDO H[SUHVVLRQV WKH\ ZLOO EH FRPSDUHG WR VLPXODWLRQ UHVXOWV XQGHU VHYHUDO GLIIHUHQW PXOWLn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
PAGE 49
LQGHSHQGHQW DQG MRLQWO\ *DXVVLDQ UDQGRP YDULDEOHV 8VLQJ WKHVH DVVXPSWLRQV RQH FRXOG WKHQ FRPSOHWHO\ GHILQH WKH DVVXPHG VWDWLVWLFV RI WKH ILOWHU ZHLJKWV E\ GHILQLQJ WKH ILOWHU ZHLJKW PHDQ YHFWRU DQG WKH FRYDULDQFH PDWUL[ IRU WKH ILOWHU ZHLJKWV 6LQFH WKH ILOWHU LV DGDSWLYH WKHVH TXDQWLWLHV ZLOO EH GHSHQGHQW RQ WKH LQLWLDO FRQGLWLRQV RI WKH ILOWHU DV ZHOO DV WKH OHQJWK RI WLPH WKDW WKH ILOWHU KDV DGDSWHG DQG RI FRXUVH VRPHKRZ GHSHQG RQ WKH VWDWLVWLFV RI WKH LQSXW VLJQDOf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f@ ZRSW WKH VROXWLRQ WR WKH :LHQHU+RSI HTXDWLRQ $V DQ H[WHQVLRQ RI WKLV LGHD HYHQ LI WKH ILOWHU LV FRQYHUJHQW LQ WKH PHDQ LW ZRXOG EH YHU\ GHVLUDEOH WR KDYH D ILQLWH YDULDQFH RI WKH ILOWHU ZHLJKWV DERXW ZRSW 7KH YDULDQFH RI WKH ILOWHU ZHLJKWV DERXW WKH :LHQHU+RSI VROXWLRQ GHWHUPLQHV WKH OHYHO RI WKH PHDQVTXDUHG HUURU DW WKH ILOWHU RXWSXW $V RQH ZRXOG H[SHFW WKH ERXQG RQ cL UHTXLUHG WR PLQLPL]H WKH PHDQVTXDUHG HUURU RI WKH ILOWHU RXWSXW LV OHVV WKDQ WKH YDOXH RI c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fV FRQWULEXWLRQ WR WKH
PAGE 50
UHFHLYHU LQSXW WKH ILOWHU LQSXW VLJQDO LV QRW ]HURPHDQ :H FDQ DSSO\ WKH PRGHO XVHG E\ 6HQQH WR GHULYH D VLPLODU VHW RI HTXDWLRQV WR GHVFULEH WKH WUDQVLHQW VWDWLVWLFV RI WKH ILOWHU ZHLJKW YHFWRU ,Q RUGHU WR EHJLQ WKH GHULYDWLRQ ZH PDNH VHYHUDO DVVXPSn WLRQ UHJDUGLQJ WKH UHODWLRQVKLSV EHWZHHQ WKH ILOWHU ZHLJKW YHFWRU DQG WKH LQSXW VLJQDO 7KHVH DVVXPSWLRQV DUH UHIHUUHG WR DV WKH IXQGDPHQWDO DVVXPSWLRQ >@ RU LQGHSHQn GHQFH DVVXPSWLRQ RQ WKH ILOWHU VWDWLVWLFV DQG DUH LQFOXGHG KHUH IRU FRQYHQLHQFH 7KH IXQGDPHQWDO DVVXPSWLRQ FRQVLVWV RI WKUHH SDUWV ZKLFK DUH f (DFK UHFHLYHU LQSXW YHFWRU UPf LV VWDWLVWLFDOO\ LQGHSHQGHQW RI DOO SUHYLRXV LQSXW YHFWRUV DQG DOO SUHYLRXV GHVLUHG ILOWHU RXWSXWV G?Pf f (DFK GHVLUHG RXWSXW G?Pf LV GHSHQGHQW RQ WKH FRUUHVSRQGLQJ LQSXW YHFWRU UPf EXW LV VWDWLVWLFDOO\ LQGHSHQGHQW RI DOO SUHYLRXV GHVLUHG RXWSXWV DQG UHFHLYHG YHFWRUV f $OO GHVLUHG TXDQWLWLHV DQG UHFHLYHG YHFWRUV DUH PXWXDOO\ *DXVVLDQGLVWULEXWHG UDQn GRP YDULDEOHV ,Q PDQ\ DSSOLFDWLRQV RI WKH /06 ILOWHU WKH ILOWHU LQSXW YHFWRU LV VDPSOHG DW WKH VDPH UDWH DV WKH RXWSXW RI WKH ILOWHU 7KHUHIRUH WKH ILUVW SDUW RI WKH IXQGDPHQWDO DVVXPSWLRQ LV FOHDUO\ QRW YDOLG LQ WKLV VLWXDWLRQ VLQFH WKH HOHPHQWV RI UP fÂ§ f DUH VKLIWHG E\ RQH SRVLWLRQ LQ FUHDWLQJ UPf 7KLV ZHDNQHVV LQ WKH IXQGDPHQWDO DVVXPSn WLRQ XQGHU WKHVH FRQGLWLRQV LV ZKDW KDV GULYHQ WKH VWXG\ RI WKH FRQYHUJHQFH SURSHUWLHV RI WKH /06 ILOWHU 2I QRWLFHDEOH LQWHUHVW LV WKH ZRUN E\ 'RXJODV >@ LQ ZKLFK D QXn PHULFDO DOJRULWKP LV SUHVHQWHG WR REVHUYH WKH H[DFW WUDQVLHQW VWDWLVWLFV RI WKH 06( FRQYHUJHQFH RI WKH ILOWHU +RZHYHU WKLV DOJRULWKP LV YHU\ FRPSOH[ LQ WKDW D VHW RI OLQHDU HTXDWLRQV PXVW EH G\QDPLFDOO\ FUHDWHG DQG VROYHG EDVHG RQ WKH QXPEHU RI WDSV LQ WKH /06 ILOWHU $V DQ H[DPSOH RI WKH FRPSOH[LW\ RI WKLV DOJRULWKP IRU D ILYHWDS ILOWHU D VHW RI WKUHHWKRXVDQG HTXDWLRQV ZHUH UHTXLUHG WR VROYH IRU WKH WUDQVLHQW 06( RI WKH ILOWHU %DVHG RQ WKLV H[DPSOH DSSO\LQJ WR WKLV PHWKRG WR D ILOWHU ZLWK PRUH WDSV GRHV QRW VHHP UHDVRQDEOH
PAGE 51
)RU WKH DGDSWLYH UHFHLYHU ZH KDYH DVVXPHG WKDW WKH LQSXW WR WKH ILOWHU LV VDPSOHG DW D UDWH 1 WLPHV IDVWHU WKDQ WKH ILOWHU RXWSXW 7KLV PHDQV WKDW UPf FRXOG WUXO\ EH VWDWLVWLFDOO\ LQGHSHQGHQW IURP DOO SUHYLRXV UHFHLYHU LQSXWV &HUWDLQO\ WKH FRQWULEXWLRQ WR UPf IURP WKH $:*1 LV LQGHSHQGHQW IURP DOO SUHYLRXV UHFHLYHU LQSXWV +RZHYHU WZR V\PERO LQWHUYDOV RI HDFK DV\QFKURQRXV XVHU FRQWULEXWH WR UPf 6R WKH ILUVW SDUW RI WKH IXQGDPHQWDO DVVXPSWLRQ ZKLOH QRW EHLQJ WUXH ZLOO VWLOO EH DVVHUWHG DQG ZH XVH WKH IXQGDPHQWDO DVVXPSWLRQ WR GHULYH WKH GHVLUHG HTXDWLRQV :H EHJLQ WKH GHULYDWLRQ RI WKH ILOWHU WUDQVLHQW ZHLJKW PHDQ YHFWRU E\ PDQLSXODWLQJ WKH JHQHUDO /06 ILOWHU XSGDWH HTXDWLRQ JLYHQ EHORZ ZP f ZPf fÂ§ XXPfXUPfZPf [XPfGPf f ZKHUH XPf LV WKH ILOWHU LQSXW VLJQDO 8VLQJ WKH V\VWHP PRGHO GHILQHG LQ 6HFWLRQ XPf IPf GHILQHG LQ HTXDWLRQ f ZLWK WKH FDUHW GURSSHG IRU QRWDWLRQDO VLPn SOLFLW\ LQ IXWXUH HTXDWLRQV :H QRWH WKDW WKH GHVLUHG ILOWHU RXWSXW LV QRW WLPHGHSHQGHQW DQG ZLOO WKHUHIRUH GURS WKH GHSHQGHQF\ RQ P IRU WKLV TXDQWLW\ 7DNLQJ WKH H[SHFWDWLRQ RI ERWK VLGHV RI HTXDWLRQ f ZLOO SURGXFH WKH GHVLUHG ILOWHU WUDQVLHQW ZHLJKW PHDQ YHFWRU (>ZP f@ ZP f ^, fÂ§ 5UU` ZPf cOLG f ZKHUH LV DQ 1 [ M9 LGHQWLW\ PDWUL[ DQG 5UU (>UPfU7Pf@ LV XVHG IRU QRWDWLRQDO FRQYHQLHQFH 2QFH ZH GHILQH WKH LQLWLDO FRQGLWLRQV RQ WKH ILOWHU Zf UHDOL]H WKDW U DLSL ÂLf DQG OHW G ZH FDQ XVH WKH SUHYLRXV HTXDWLRQ UHFXUVLYHO\ WR GHVFULEH WKH ILOWHU WUDQVLHQW ZHLJKW PHDQ YHFWRU ,Q RUGHU WR GHYHORS DQ LWHUDWLYH HTXDWLRQ IRU WKH ILOWHU WUDQVLHQW ZHLJKW FRYDULDQFH PDWUL[ HTXDWLRQ f ZLOO EH XVHG WR ILQG DQ LWHUDWLYH HTXDWLRQ IRU WKH ILOWHU WUDQVLHQW ZHLJKW DXWRFRUUHODWLRQ PDWUL[ GHILQHG DV 5LLf8fPf (>ZPfZUPf@ 7KLV GHULYDWLRQ
PAGE 52
LV GHVFULEHG LQ $SSHQGL[ % EXW WKH ILQDO UHVXOW LV OLVWHG EHORZ 5:Z^UQ f 5ZZ^Pf LZUDfUUG [58f8Pf5UU WUZ7UDfG fÂ§ L5UU5Xf8fPf L5UUG fÂ§ SFU ^ZUPfUO UZUPf ZPfUU` G fÂ§ SIZUPfIIUG LFU5ZXfPfUIU f LLD7OZZPf ÂÂFU WUDFH58f8Pff, QDI7n5ZZPfI, QR WUDFH5XXPffUUU QDIU75ZZPf AII75OfPfUI7 *LYHQ WKH DERYH HTXDWLRQ ZH FDQ HDVLO\ GHILQH WKH WUDQVLHQW ILOWHU ZHLJKW FRYDULDQFH PDWUL[ XVLQJ WKH IROORZLQJ HTXDWLRQ RQFH ZH GHILQH Zf 5XWOf U DQG 5UU &ZZPf 5ZZPf ZPfZ7Pf f $V DQ DOWHUQDWLYH WR HTXDWLRQ f ZH DOVR FRQVLGHU XVLQJ WKH WUDQVLHQW ZHLJKW HUURU FRYDULDQFH PDWUL[ ZKLFK GHVFULEHV WKH GHYLDWLRQ RI WKH ILOWHU ZHLJKWV DERXW WKH :LHQHU+RSI YDOXHV DV D IXQFWLRQ RI WLPH >@ 7KLV LV WKH VDPH HTXDWLRQ XVHG E\ 0LOOHU >@ WR VWXG\ WKH WUDQVLHQW 06( UHVSRQVH RI WKH /06 ILOWHU 7KH WUDQVLHQW ZHLJKW HUURU FRYDULDQFH PDWUL[ LV GHILQHG DV .Pf ( ZPf ZRSWfZPf :RSWf f DQG FDQ EH HYDOXDWHG DW WKH PWK LQWHUYDO XVLQJ WKH IROORZLQJ LWHUDWLYH HTXDWLRQ .P f .Pf ÂL>5U.Pf .Pf5@ L5UU WUDFH5UU.Pff U5UU.Pf5UU QP cQ5UU f ,Q RUGHU WR XVH HTXDWLRQ f ZH QRWH WKDW .f ZRSWZASW ZKHQ Zf DQG GPLQ :RSWDLSLÂLf 7KH WUDQVLHQW ZHLJKW FRYDULDQFH PDWUL[ FDQ EH IRXQG DV VKRZQ LQ HTXDWLRQ f
PAGE 53
&ZZPf ( ZPf ZPffZPf ZPff .Pf ZPfZUPf ZRSWZSW ZPfZASW ZRSWZ7Pf f 6HYHUDO GLIIHUHQW '6&'0$ HQYLURQPHQWV ZHUH VLPXODWHG DQG WKH WUDQVLHQW PHDQ ZHLJKW YHFWRU DQG WUDQVLHQW DXWRFRUUHODWLRQ PDWUL[ RI WKH /06 ILOWHU ZHLJKWV ZHUH REVHUYHG DW WKH PWK LQWHUYDO RI DGDSWLRQ XVLQJ WKH IROORZLQJ HTXDWLRQV 0 ZPf L O 0 W L f f 7KHVH TXDQWLWLHV ZHUH WKHQ XVHG WR IRUP DQ HVWLPDWH RI WKH WUDQVLHQW DXWRFRYDULDQFH PDWUL[ RI WKH /06 ILOWHU ZHLJKWV DV 7 &ZZPf 5ZXPf ZPfZ Pf f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f WKURXJK f )LJXUH FRPSDUHV WKH __(>ZPf@__ SURGXFHG E\ HTXDWLRQ f WR WKH REVHUYHG VLPXODWLRQ UHVXOWV ,Q RUGHU WR FRPSDUH WKHVH WZR TXDQWLWLHV GLUHFWO\ WKH HUURU LV SORWWHG DV __(>ZPf@ :Pf__ LQ WKH VHFRQG SDUW RI WKH ILJXUH
PAGE 54
)LJXUH 3HUIRUPDQFH RI HTXDWLRQ f IRU D VLQJOHXVHU V\VWHP Df 7KH DQDO\WLFDO PHDQ ILOWHU ZHLJKW YHFWRU QRUP __(>ZPf@__ rf DQG WKH VLPXODWHG PHDQ ILOWHU ZHLJKW YHFWRU QRUP __:Pf__ VROLG OLQHf SORWWHG DV D IXQFWLRQ RI WKH WUDLQLQJ OHQJWK Ef &RPSDULQJ WKH HUURU EHWZHHQ WKH WZR PHDQ ZHLJKW YHFWRUV __(>ZPf@ fÂ§ ZPf__ $V D VHFRQG WHVW D WKUHHXVHU V\VWHP ZDV VLPXODWHG $V EHIRUH WKH GHVLUHG XVHU ZDV UHFHLYHG DW D 615 RI G% DQG WKH WZR LQWHUIHULQJ XVHUV ZHUH UHFHLYHG DW D SRZHU OHYHO G% DERYH WKH GHVLUHG XVHU 7KH LQWHUIHULQJ XVHUV KDG SURSDJDWLRQ GHOD\V RI U 7& DQG 7& 7KLV HQYLURQPHQW ZDV VLPXODWHG XVLQJ LQGHSHQGHQW WULDOV )LJXUH FRPSDUHV HTXDWLRQ f WR WKH VLPXODWLRQ UHVXOWV )LJXUHV DQG VKRZ WKDW WKH DQDO\WLFDO H[SUHVVLRQ IRU WKH WUDQVLHQW PHDQ ZHLJKW YHFWRU RI WKH /06 ILOWHU JLYHQ E\ HTXDWLRQ f FORVHO\ WUDFNV WKH FRUUHVSRQGLQJ VLPXODWLRQ UHVXOWV 6HYHUDO RWKHU PXOWLXVHU VFHQDULRV ZLWK YDULRXV OHYHOV RI 615 DQG 0$, ZHUH VLPXODWHG DQG WKH SHUIRUPDQFH RI HTXDWLRQ f ZDV VLPLODU WR WKRVH VKRZQ LQ )LJXUHV DQG 7KHUHIRUH ZH XVH HTXDWLRQ f LQ WKH VHTXHO ZKHQ ZH UHTXLUH XVH RI DQ H[SUHVVLRQ IRU WKH /06 ILOWHUfV WUDQVLHQW ZHLJKW PHDQ YHFWRU
PAGE 55
)LJXUH 3HUIRUPDQFH RI HTXDWLRQ f IRU D WKUHHXVHU V\VWHP Df 7KH DQDO\WLFDO PHDQ ILOWHU ZHLJKW YHFWRU QRUP __(>ZPf@__ rf DQG WKH VLPXODWHG PHDQ ILOWHU ZHLJKW YHFWRU QRUP __:Pf__ VROLG OLQHf SORWWHG DV D IXQFWLRQ RI WKH WUDLQLQJ OHQJWK Ef &RPSDULQJ WKH HUURU EHWZHHQ WKH WZR PHDQ ZHLJKW YHFWRUV __(>ZPf@ ZPf__ $ VLPLODU WHVW ZDV XVHG WR FKDUDFWHUL]H WKH SHUIRUPDQFH RI HTXDWLRQV f DQG f ZKLFK DUH WKH DQDO\WLFDO H[SUHVVLRQV IRU WKH /06 ILOWHUfV WUDQVLHQW ZHLJKW DXWRFRYDULn DQFH PDWUL[ :H REVHUYHG WKDW IRU D VLQJOHXVHU HQYLURQPHQW ERWK HTXDWLRQV JDYH VLPLODU UHVXOWV +RZHYHU ZKHQ PXOWLSOHXVHUV ZHUH SUHVHQW HTXDWLRQ f LQLWLDOO\ WUDFNV WKH VLPXODWLRQ UHVXOWV IRU &ZZPf EXW DW VRPH SRLQW GLYHUJHV IURP WKH VLPXn ODWLRQ UHVXOWV +RZHYHU HTXDWLRQ f ZDV REVHUYHG WR SURGXFH VLPLODU UHVXOWV ZKHQ FRPSDUHG WR &ZZPf UHJDUGOHVV RI WKH QXPEHU RI XVHUV DQG WKH OHYHO RI WKH 0$, $ WZRXVHU V\VWHP ZLWK WKH GHVLUHG XVHU UHFHLYHG DW D 615 RI G% DQG WKH LQWHUIHULQJ XVHU UHFHLYHG DW D SRZHU OHYHO G% DERYH WKH GHVLUHG XVHU ZDV VLPXODWHG IRU WULDOV 7KH UHVXOWV RI FRPSDULQJ WKH )UREHQLXV QRUP RI HTXDWLRQV f DQG f WR WKH )UREHQLXV QRUP RI WKH VLPXODWHG &ZZPf PDWUL[ DUH VKRZQ LQ )LJXUH 1RWH WKDW VLPLODU UHVXOWV ZHUH REVHUYHG IRU RWKHU PXOWLXVHU VFHQDULRV :H KDYH REVHUYHG
PAGE 56
)LJXUH 3HUIRUPDQFH RI HTXDWLRQV f DQG f IRU D WZRXVHU V\VWHP 7KH )UREHQLXV QRUP RI WKH /06 ILOWHUfV WUDQVLHQW ZHLJKW DXWRFRYDULDQFH PDn WUL[ HTXDWLRQ f rf HTXDWLRQ f [f DQG VLPXODWLRQ UHVXOWV f WKDW WKH )UREHQLXV QRUP RI WKH GLIIHUHQFH EHWZHHQ HTXDWLRQ f DQG WKH VLPXODWHG &ZZPf LV RQ WKH RUGHU RI b RYHU VHYHUDO GLIIHUHQW PXOWLXVHU VFHQDULRV :H KDYH GHULYHG D VHW RI DQDO\WLFDO H[SUHVVLRQV HTXDWLRQV f DQG f WKDW ZH ZLOO QRZ XVH WR GHVFULEH WKH WUDQVLHQW PRGHO RI WKH /06 ILOWHU ZHLJKWV XQGHU WKH DVVXPSWLRQ WKDW WKH ZHLJKWV DUH MRLQWO\ *DXVVLDQ 6LQFH ZH DUH XVLQJ WKH ZHLJKWV RI WKH ILOWHU WR IRUP D WLPLQJ HVWLPDWH RI WKH GHVLUHG XVHUf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
PAGE 57
LV ZH ZDQW WR ILQG DQ H[SUHVVLRQ IRU 3USL s SLf ZKHQ WKH YDOXH RI ÂL LV LJQRUHG ,Q WKLV FDVH WKH SURSDJDWLRQ GHOD\ HVWLPDWH DW WKH ]WK LWHUDWLRQ RI WKH DGDSWLYH ILOWHU LV IRUPHG DV VKRZQ EHORZ 3L DUJPD[)Sf S f 1 f Y )Sf _Z,nmfF"nO_ f f 7KHUHIRUH WKH FKLS VHOHFWLRQ HUURU SUREDELOLW\ FDQ EH H[SUHVVHG XVLQJ D XQLRQ ERXQG >@ DV 1O 3UWIL ) 3Lf 9 3U)SLf )LNf f f NfÂ§ IFASL 1RWH WKDW HDFK WHUP LQ WKH SUHYLRXV HTXDWLRQ FDQ EH H[SUHVVHG DV )SLf fÂ§ )Nf Z7Lf AF>3OAFL3Off fÂ§ &L3Nf&L3Af M Z"f :7Wf$_Zf f 8VLQJ RXU DVVXPSWLRQ WKDW WKH ILOWHU ZHLJKWV DUH MRLQWO\ *DXVVLDQ WKH H[SUHVVLRQ LQ HTXDWLRQ f LV MXVW D TXDGUDWLF IRUP RI *DXVVLDQ UDQGRP YDULDEOHV 7KLV W\SH RI H[SUHVVLRQ LV FRPPRQ LQ UDGDU DSSOLFDWLRQV ZHDSRQ FRQWURO V\VWHPV DQG LV DOVR LQWULQVLF WR DQ\ GLVFXVVLRQ RI YDULDQFHV RI UDQGRP YDULDEOHV >@ 6R ZH QRZ PXVW FRQFHUQ RXUVHOYHV ZLWK KRZ WR HYDOXDWH 3U ZULf$Z]f f 7KLV W\SH RI HYDOXDn WLRQ KDV UHFHLYHG FRQVLGHUDEOH DWWHQWLRQ E\ PDQ\ UHVHDUFKHUV >@ 7KH QH[W IHZ SDUDJUDSKV ZLOO VXPPDUL]H YDULRXV PHWKRGV WKDW PD\ EH XVHG WR KDQGOH WKLV SUREOHP $V VKRZQ LQ $SSHQGL[ & WKH GLVWULEXWLRQ RI ZU]f$ÂZ]f FDQ EH VKRZQ WR KDYH WKH VDPH GLVWULEXWLRQ DV < AA;LL:LEWf f Â L ZKHUH WKH DUH LQGHSHQGHQW *DXVVLDQ UDQGRP YDULDEOHV RI ]HUR PHDQ DQG XQLW YDULDQFH DQG WKH WHUPV DQG ? GHSHQG RQ WKH WUDQVIRUPDWLRQ PDWUL[ $N DV ZHOO WKH
PAGE 58
FRYDULDQFH PDWUL[ RI WKH ILOWHU ZHLJKWV 'HULYLQJ WKH GLVWULEXWLRQ RI HTXDWLRQ f GLUHFWO\ GRHV QRW DSSHDU WR EH PDWKHPDWLFDOO\ WUDFWDEOH +RZHYHU GHULYLQJ WKH FRUUHVSRQGLQJ FKDUDFWHULVWLF IXQFWLRQ LV UHODWLYHO\ VLPSOH )RU D UDQGRP YDULDEOH WKDW KDV D SUREDELOLW\ GHQVLW\ IXQFWLRQ RI I[[f WKH FKDUDFWHULVWLF IXQFWLRQ LV JLYHQ E\ rZf (>H[S MZ[f@ I[[f H[S MXL[f G[ f fÂ§ 22 /LNHZLVH LI RQH LV JLYHQ WKH FKDUDFWHULVWLF IXQFWLRQ [Zf WKHQ WKH FRUUHVSRQGLQJ SUREDELOLW\ GHQVLW\ IXQFWLRQ LV IRXQG XVLQJ WKH LQYHUVLRQ IRUPXOD VWDWHG EHORZ RR rrf A AZfH[3aDFfGD fÂ§22 7KH FKDUDFWHULVWLF IXQFWLRQ IRU HTXDWLRQ f LV IRXQG WR EH f f Y MX?L RKQVRQ DQG .RW] >@ SURYLGH DQ H[WHQVLYH VXPPDU\ RI WHFKQLTXHV WKDW FDQ EH XVHG WR ILQG I\^\f RU WKH 3U\ Wf ZKHUH W LV VRPH JLYHQ WKUHVKROG ZKHQ WKH WUDQVIRUn PDWLRQ PDWUL[ $N LV SRVLWLYH GHILQLWH )RU VSHFLILF UHVWULFWLRQV RQ WKH DQG LQ WKH FKDUDFWHULVWLF IXQFWLRQ VHYHUDO H[SDQVLRQV RI \Xf LQWR SRZHU VHULHV UHSUHVHQWDWLRQV DUH SUHVHQWHG 7KHQ WKH LQYHUVLRQ IRUPXOD LQ HTXDWLRQ f LV DSSOLHG WR HDFK WHUP LQ WKH VHULHV WR SURGXFH I\\f $ PDWUL[ $ LV SRVLWLYH GHILQLWH ZKHQ [U$[ IRU DQ\ QRQ]HUR YHFWRU [ 7KHUHIRUH LI WKH WUDQVIRUPDWLRQ PDWUL[ $N LV SRVLWLYH GHILQLWH WKHQ D FKLSVHOHFWLRQ HUURU FDQ QHYHU RFFXU DQG WKHUHIRUH 3U. f :H PXVW GHWHUPLQH LI WKH WUDQVIRUPDWLRQ PDWUL[ LV SRVLWLYH GHILQLWH $ VXIILFLHQW FRQGLWLRQ IRU SRVLWLYH GHILQLWHQHVV LV WKDW DOO RI WKH HLJHQYDOXHV RI WKH PDWUL[ $N DUH SRVLWLYH >@ :LWKRXW DFWXDOO\ FRPSXWLQJ WKHVH HLJHQYDOXHV ZH FDQ WHVW IRU WKH SRVVLEOH H[LVWHQFH RI QHJDWLYH HLJHQYDOXHV ,W LV NQRZQ WKDW WKH VXP RI WKH HLJHQYDOXHV LV HTXDO WR WKH WUDFH RI WKH PDWUL[ 6LQFH DOO RI WKH GLDJRQDO HOHPHQWV RI $N DUH ]HUR WKHUH PXVW
PAGE 59
H[LVW QHJDWLYH DV ZHOO DV SRVLWLYH HLJHQYDOXHV DQG WKHUHIRUH WKH WUDQVIRUPDWLRQ PDWUL[ LV QRW SRVLWLYH GHILQLWH 5LFH >@ SURYLGHV VHYHUDO QXPHULFDO WHFKQLTXHV WKDW FDQ EH XVHG WR HYDOXDWH 3U7 Wf WKURXJK DSSOLFDWLRQV RI HTXDWLRQV f DQG f IRU D JHQHUDO WUDQVn IRUPDWLRQ PDWUL[ +RZHYHU LW LV QRWHG WKDW WKHVH WHFKQLTXHV FDQ H[KLELW VORZ FRQn YHUJHQFH GXH WR WKH RVFLOODWRU\ QDWXUH RI WKH H[SRQHQW LQ HTXDWLRQ f ,Q RUGHU WR LPSURYH WKH FRQYHUJHQFH UDWH RI WKH QXPHULFDO LQWHJUDWLRQ WHFKQLTXHV D FKDQJH RI YDULDEOHV WKDW SURGXFHV D WLOWLQJ RI WKH LQWHJUDWLRQ SDWK LV SUHVHQWHG +RZHYHU WKH FKDQJH RI YDULDEOHV LV LQYHUVHO\ GHSHQGHQW RQ WKH GHVLUHG WKUHVKROG W DQG LV WKHUHIRUH QRW YDOLG ZKHQ W fÂ§ ZKLFK LV WKH FDVH DW KDQG $ TXDGUDWLF IRUP RI LQGHSHQGHQW *DXVVLDQ UDQGRP YDULDEOHV VXFK DV VKRZQ LQ HTXDWLRQ f FDQ DOVR EH VKRZQ WR KDYH WKH VDPH GLVWULEXWLRQ DV D OLQHDU VXP RI LQGHSHQGHQW QRQFHQWUDO [ UDQGRP YDULDEOHV >@ 7KDW LV HTXDWLRQ f FDQ DOVR EH UHSUHVHQWHG E\ U f ZKHUH ;M DUH LQGHSHQGHQW UDQGRP YDULDEOHV KDYLQJ QRQFHQWUDO ? GLVWULEXWLRQ ZLWK ULM GHJUHHV RI IUHHGRP DQG QRQFHQWUDOLW\ SDUDPHWHU 6M IRU M U DQG ; KDYLQJ D ]HURPHDQ XQLWYDULDQFH *DXVVLDQ GLVWULEXWLRQ 'DYLHV >@ EDVHG RQ LQYHUWLQJ WKH FRUUHVSRQGLQJ FKDUDFWHULVWLF IXQFWLRQ >@ SUHVHQWV D QXPHULFDO WHFKQLTXH WKDW FDQ EH XVHG WR ILQG WKH GLVWULEXWLRQ RI D OLQHDU VXP RI LQGHSHQGHQW QRQFHQWUDO [ UDQGRP YDULDEOHV ZKHQ WKH RULJLQDO TXDGUDWLF IRUP KDV D JHQHUDO QRW QHFHVVDULO\ SRVLWLYH GHILQLWHf WUDQVIRUPDWLRQ PDWUL[ 7KH WHFKQLTXH LQYROYHV HYDOXDWLQJ D VHULHV RI H[SRQHQWLDO WHUPV ZKHUH WKH WUXQFDWLRQ HUURU LV GHSHQGHQW RQ WKH QXPEHU RI WHUPV LQFOXGHG LQ WKH VHULHV 7KH QXPEHU RI WHUPV UHTXLUHG IRU WKH LQWHJUDWLRQ LV GHWHUPLQHG DSSUR[LPDWHO\ E\ WKH WRWDO QXPEHU RI GHJUHHV RI IUHHGRP DQG WKH VXP RI WKH QRQFHQWUDOLW\ SDUDPHWHUV DV ZHOO DW WKH YDOXH RI W DW ZKLFK WKH GLVWULEXWLRQ
PAGE 60
IXQFWLRQ LV WR EH HYDOXDWHG 'DYLHV SUHVHQWV D WDEOH WKDW OLVW WKH QXPEHU RI WHUPV UHTXLUHG IRU LQWHJUDWLRQ DV D IXQFWLRQ RI WKHVH SDUDPHWHUV 7KLV WDEOH VKRZV WKDW DV W DSSURDFKHV ]HUR WKH QXPEHU RI WHUPV UHTXLUHG IRU LQWHJUDWLRQ TXLFNO\ LQFUHDVHV IURP VHYHUDO KXQGUHG WR WHQV RI WKRXVDQGV 7KHUHIRUH WKH WHFKQLTXHV RI 5LFH RU 'DYLHV DSSHDU WR EH DSSOLFDEOH WR D VLWXDWLRQ ZKHUH WKH 3U< f KDV WR EH HYDOXDWHG YHU\ LQIUHTXHQWO\ 5HFDOO WKDW DW WKH UHFHLYHUfV LWK LQWHUYDO ZH KDYH WR HYDOXDWH 3U< f D WRWDO RI 1fÂ§Of WLPHV WR IRUP D ERXQG RQ WKH FKLSVHOHFWLRQ HUURU SUREDELOLW\ ,I ZH ZDQW WR HYDOXDWH WKH FKLSVHOHFWLRQ SUREDELOLW\ DV D IXQFWLRQ RI WKH /06 ILOWHUfV DGDSWDWLRQ WLPH WKHVH WHFKQLTXHV EHFRPH QXPHULFDOO\ SURKLELWLYH 7KDW LV ZH IRXQG WKDW 0RQWH&DUOR VLPXODWLRQV RI D VSHFLILF '6&'0$ HQYLURQPHQW SURGXFH UHVXOWV IRU WKH FKLSVHOHFWLRQ HUURU SUREDELOLW\ LQ OHVV WLPH WKDQ LV UHTXLUHG WR HYDOXDWH WKH VDPH TXDQWLW\ XVLQJ RQH RI 5LFHfV WHFKQLTXHV IRU WKH VDPH HQYLURQPHQW $V D PHDQV RI HYDOXDWLQJ 3U)SLf fÂ§ )Nf f ZH LQVWHDG WXUQ WR WKH PRPHQW JHQHUDWLQJ IXQFWLRQ ,QVWHDG RI WU\LQJ WR HYDOXDWH WKH GHVLUHG H[SUHVVLRQ H[SOLFLWO\ ZH XVH D &KHUQRII ERXQG WR ILQG DQ XSSHU ERXQG RQ 3U)SLf fÂ§ )Nf f IRU HDFK WHUP LQ HTXDWLRQ f VHH $SSHQGL[ & IRU GHWDLOVf :H WKHQ VXP WKHVH 1 f &KHUQRII ERXQGV WR ILQG DQ XSSHU ERXQG RQ 3USL A S[f $Q H[DPSOH RI WKH UHVXOWLQJ ERXQG XVLQJ WKLV PHWKRG LV VKRZQ LQ )LJXUH $ WKUHHXVHU '6&'0$ HQYLURQPHQW ZDV VLPXODWHG ZLWK D 615 RI G% DQG ERWK LQWHUIHULQJ XVHUV ZHUH UHFHLYHG DW D SRZHU OHYHO G% DERYH WKH GHVLUHG XVHU 7KH SURSDJDWLRQ GHOD\V IRU WKH XVHUV ZHUH 7? fÂ§ 7F U 7& DQG U 7F 7KH /06 VWHSVL]H [ ZDV VHW WR WKH WRWDO LQSXW SRZHUf VXFK WKDW WKH ILOWHU ZDV FRQYHUJHQW LQ WKH PHDQVTXDUH 7KHUH DUH WKUHH FXUYHV SUHVHQW LQ )LJXUH 7KH ILUVW FXUYH LV WKH VLPXODWHG UHVXOWV IRU WKH LQFRUUHFW FKLSVHOHFWLRQ SUREDELOLW\ ZKLFK LV SORWWHG XVLQJ D SOXV VLJQ f 7KH RWKHU WZR FXUYHV DUH WKH UHVXOWLQJ XSSHU ERXQG IRU WKLV SUREDELOLW\ IRU WZR GLIIHUHQW DSSOLFDWLRQV RI WKH &KHUQRII8QLRQ ERXQG WHFKQLTXH 7KH VROLG OLQH
PAGE 61
UHSUHVHQWV WKH XSSHU ERXQG ZKHQ HTXDWLRQV f DQG f DUH XVHG WR GHVFULEH WKH WUDQVLHQW PHDQ YHFWRU DQG DXWRFRYDULDQFH PDWUL[ RI WKH /06 ILOWHU ZHLJKWV DV D IXQFWLRQ RI WKH WUDLQLQJ LQWHUYDO 1RWH WKDW WKH UHVXOWLQJ XSSHU ERXQG LV QRW YHU\ WLJKW ,W VHHPV DSSURSULDWH WR ZRQGHU LI WKH ORRVHQHVV RI WKH ERXQG LV GXH WR WKH &KHUQRII8QLRQ ERXQG WHFKQLTXH RU GXH WR HUURUV SUHVHQW LQ WKH PRGHO JLYHQ E\ HTXDWLRQV f DQG f $V D FRQWURO IRU WKLV WHVW WKH WKLUG FXUYH rf UHSUHVHQWV WKH XSSHU ERXQG IRXQG E\ XVLQJ WKH WUDQVLHQW PHDQ YHFWRU DQG DXWRFRYDULDQFH PDWUL[ RI WKH /06 ILOWHU ZHLJKWV REWDLQHG GLUHFWO\ IURP WKH VLPXODWLRQ 7KLV LV VOLJKWO\ FORVHU WR WKH VLPXODWLRQ UHVXOWV EXW WKH XSSHU ERXQG LV VWLOO YHU\ ORRVH 7KHUHIRUH LV VHHPV WKDW WKH DSSOLFDWLRQ RI WKH &KHUQRII8QLRQ ERXQGV DUH WKH PDLQ UHDVRQ WKDW WKH GHULYHG XSSHU ERXQG LV YHU\ ORRVH ZKHQ FRPSDUHG WR WKH DFWXDO VLPXODWLRQ UHVXOWV 1RWH WKDW WKLV EHKDYLRU ZDV REVHUYHG IRU VHYHUDO RWKHU PXOWLXVHU '6&'0$ VFHQDULRV 8QIRUWXQDWHO\ EDVHG RQ WKLV UHVXOW WKLV WHFKQLTXH LV QRW YHU\ XVHIXO LQ WHUPV RI FKDUDFWHUL]LQJ WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU $SSUR[LPDWLRQ IRU WKH &RQGLWLRQDO 9DULDQFH RI WKH 7LPLQJ (VWLPDWH $ WUDGLWLRQDO WHFKQLTXH WKDW LV XVHG WR FKDUDFWHUL]H WKH SHUIRUPDQFH RI DQ HVWLn PDWRU LV WR FRPSDUH WKH YDULDQFH RI WKH HVWLPDWH WR WKH &UDPU5DR ERXQG &5%f 7KH &5% LV D ORZHU ERXQG RQ WKH YDULDQFH RI DQ\ XQELDVHG HVWLPDWRU IRU D JLYHQ ORJ OLNHOLKRRG IXQFWLRQ >@ ,Q WKLV VHFWLRQ ZH ZLOO XVH WKH WUDGLWLRQDO &5% WHFKQLTXH DORQJ ZLWK VHYHUDO DVVXPSWLRQV LQ KRSHV RI IRUPLQJ DQ DSSUR[LPDWLRQ WR WKH SHUIRUn PDQFH RI RXU WLPLQJ HVWLPDWRU 6LQFH ZH DUH LQWHUHVWHG LQ FKDUDFWHUL]LQJ WKH WUDQVLHQW SHUIRUPDQFH RI RXU WLPLQJ HVWLPDWRU ZH EDVH RXU WHFKQLTXH RQ WKH WLPHGHSHQGHQW VWDWLVWLFDO PRGHO IRU WKH /06 ILOWHU ZHLJKWV JLYHQ LQ HTXDWLRQV f DQG f 6LQFH ZH KDYH DVVXPHG WKDW WKH ILOWHU ZHLJKWV DUH MRLQWO\ *DXVVLDQ HTXDWLRQV f DQG f DUH VXIILFLHQW WR FRPSOHWHO\ GHILQH WKH WLPHGHSHQGHQW VWDWLVWLFDO PRGHO IRU WKH /06 ILOWHU ZHLJKW YHFWRU :H GR QRW LQWHQG WR XVH WKLV PRGHO WR GHULYH D QHZ WLPLQJ HVWLPDWLRQ DOJRULWKP +RZHYHU ZH XVH WKLV PRGHO WR GHULYH D &5% KRSLQJ
PAGE 62
)LJXUH &RPSDULQJ LQFRUUHFW FKLSVHOHFWLRQ SUREDELOLW\ WR WKH 8QLRQERXQG IRU D WKUHHXVHU V\VWHP 6LPXODWLRQ UHVXOWV f WKH &KHUQRII ERXQG XVLQJ HTXDn WLRQV f DQG f VROLG OLQHf DQG WKH &KHUQRII ERXQG XVLQJ VLPXODWHG ZHLJKW YHFWRU VWDWLVWLFV rf WKDW WKH UHVXOW FDQ EH XVHG WR FKDUDFWHUL]H WKH FRQGLWLRQDO YDULDQFH RI RXU HVWLPDWH IRU L 8VLQJ HTXDWLRQV f DQG f WKH WLPHGHSHQGHQW ORJOLNHOLKRRG IXQFWLRQ DW WKH PWK LWHUDWLRQ LJQRULQJ FRQVWDQWVf LV JZcÂ• Pf >ZPf ZPffU&fePfZUDf ZPff@ f 7KH &5% IRU WKH DERYH ORJOLNHOLKRRG IXQFWLRQ LV WKHQ JLYHQ E\ WKH IROORZLQJ HTXDWLRQ >@ 9DUÂ rf! GJZ?OPf G6c f ,Q RUGHU WR HYDOXDWH WKH DERYH H[SUHVVLRQ ZH UHTXLUH NQRZOHGJH RI KRZ &ZZPf GHSHQGV RQ ÂL LQ RUGHU WR HYDOXDWH WKH UHTXLUHG SDUWLDO GHULYDWLYHV 7KLV GRHV QRW DSSHDU WR EH PDWKHPDWLFDOO\ WUDFWDEOH VLQFH DV VHHQ LQ HTXDWLRQ f ZH QHHG WR HYDOXDWH (>ZUDfF0[ZUPf@ ,Q RUGHU WR FRQWLQXH ZLWK WKLV DSSURDFK ZH DVVXPH
PAGE 63
WKDW G&ZZPfG6L 8VLQJ WKLV DVVXPSWLRQ DIWHU VWUDLJKWIRUZDUG DOJHEUDLF PDn QLSXODWLRQV RXU DSSUR[LPDWLRQ WR WKH YDULDQFH RI WKH HVWLPDWH LV JLYHQ EHORZ L f 6HYHUDO DX[LOLDU\ HTXDWLRQV DQG LQLWLDO FRQGLWLRQV DUH UHTXLUHG LQ RUGHU WR HYDOXDWH HTXDWLRQ f 7KH UHTXLUHG DX[LOLDU\ HTXDWLRQV DUH f 27 B USG L USLf G>aaFn &O f f :H KDVWHQ WR VWUHVV WKDW ZH KDYH XVHG WKH PHFKDQLFV RI WKH &5% WHFKQLTXH WR GHULYH HTXDWLRQ f +RZHYHU GXH WR WKH YDULRXV DSSUR[LPDWLRQV XVHG DORQJ WKH ZD\ RXU UHVXOW LV QRW D ORZHU ERXQG DQG LV RQO\ LQWHQGHG WR SURYLGH D URXJK DSSUR[LPDWLRQ WR WKH SHUIRUPDQFH RI RXU HVWLPDWRU 6HYHUDO GLIIHUHQW '6&'0$ HQYLURQPHQWV ZHUH VLPXODWHG LQ RUGHU WR YHULI\ WKH XVHIXOQHVV RI HTXDWLRQ f DV DQ DSSUR[LPDWLRQ WR WKH FRQGLWLRQDO YDULDQFH RI WKH ÂL HVWLPDWH )RU WKH IROORZLQJ UHVXOWV WKH XVHUVf VSUHDGLQJ FRGHV ZHUH VHOHFWHG IURP D VHW RI *ROG FRGHV >@ ZLWK FKLSVELW DQG WKH GHVLUHG XVHU ZDV UHFHLYHG DW D 615 RI G% 7KH SURSDJDWLRQ GHOD\ IRU HDFK XVHU ZDV FKRVHQ IURP D XQLIRUP GLVWULEXWLRQ RYHU WKH LQWHUYDO >ff 2QFH WKH VSUHDGLQJ FRGHV DQG GHOD\V ZHUH VHOHFWHG WKH\ ZHUH FRQVLGHUHG IL[HG DQG WKH '6&'0$ HQYLURQPHQW ZDV VLPXODWHG IRU WULDOV 7KHUHIRUH WKH YDULDEOHV WKDW ZHUH FRQVLGHUHG UDQGRP EHWZHHQ WKH LQGLYLGXDO VLPXn ODWLRQ WULDOV ZHUH WKH $:*1 DQG WKH GDWD VHTXHQFHV IRU WKH LQWHUIHULQJ XVHUV 1RWH WKDW WKH /06 VWHSVL]H Â ZDV DOZD\V FKRVHQ WR EH WKH WRWDO LQSXW SRZHUf VXFK WKDW WKH ILOWHU ZDV FRQYHUJHQW LQ WKH PHDQVTXDUH
PAGE 64
)LJXUH &RPSDULQJ WKH VLPXODWHG FRQGLWLRQDO YDULDQFH RI WKH Â HVWLPDWH WR WKH DQDO\WLFDOO\ GHULYHG DSSUR[LPDWLRQ LQ DEVHQFH RI QHDUIDU 0$, )LJXUH VKRZV WKH FRPSDULVRQ EHWZHHQ HTXDWLRQ f GDVKHG OLQHf DQG WKH VWDQGDUG GHYLDWLRQ RI WKH Â DERXW Â FRQGLWLRQHG RQ S? VROLG OLQHf IRU VHYHUDO GLIIHUHQW VFHQDULRV ,Q WKLV ILJXUH DOO LQWHUIHULQJ XVHUV DUH UHFHLYHG DW WKH VDPH SRZHU OHYHO DV WKH GHVLUHG XVHU 7KH UHVXOWV VKRZQ LQ WKLV ILJXUH ORRN YHU\ SURPLVLQJ 7KH ODUJHVW HUURU EHWZHHQ WKH VLPXODWLRQ UHVXOWV DQG HTXDWLRQ f ZKLFK RFFXUV IRU XVHUV LV b 8QIRUWXQDWHO\ WKH XVHIXOQHVV RI HTXDWLRQ f TXLFNO\ GLVDSSHDUV DV PRUH XVHUV DUH DGGHG WR WKH V\VWHP RU WKH OHYHO RI 3N3L LV LQFUHDVHG RU VRPH FRPELQDWLRQ RI WKHVH WZR FRQGLWLRQVf $V DQ H[DPSOH RI WKLV VWDWHPHQW )LJXUH VKRZV KRZ WKH DSSUR[LPDWLRQ IDLOV IRU VHYHUDO PXOWLXVHU VFHQDULRV ZKHUH DOO LQWHUIHULQJ XVHUV DUH UHFHLYHG DW VRPH OHYHO DERYH WKH GHVLUHG XVHU 8SRQ UHYLHZ RI WKH GHULYDWLRQ RI WKH DSSUR[LPDWLRQ WKH ZHDNQHVV RI WKH DSSUR[LPDWLRQ DSSHDUV WR VWHP IURP WKH DVVXPSWLRQV WKDW ZHUH PDGH WR IDFLOLWDWH WKH GHULYDWLRQ DQG QRW IURP HTXDWLRQV f DQG f DV WKHVH KDYH EHHQ REVHUYHG WR DJUHH YHU\ ZHOO ZLWK VLPXODWLRQ UHVXOWV
PAGE 65
R D 8VHUV 1) G% E 8VHUV 1) G% F 8VHUV 1) G% D E RI 7UDLQLQJ %LWV )LJXUH &RPSDULQJ WKH VLPXODWHG FRQGLWLRQDO YDULDQFH RI WKH Â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f ZKHUH 0 LV WKH QXPEHU RI UPf VDPSOHV REVHUYHG EHIRUH FRPSXWLQJ WKH WLPLQJ HVn WLPDWH $V VKRZQ E\ WKH DERYH HTXDWLRQ WKLV WHFKQLTXH DOVR UHTXLUHV DQ DOO RQHV WUDLQLQJ VHTXHQFH ,W LV DOVR NQRZQ WR EH RSWLPDO IRU D VLQJOHXVHU LQ WKH SUHVHQFH RI $:*1 RQO\ EXW FDQ EH VXERSWLPDO LQ WKH SUHVHQFH RI PXOWLDFFHVV LQWHUIHUHQFH
PAGE 66
2XU WLPLQJ HVWLPDWRU ZLOO EH FRPSXWHG E\ SURFHVVLQJ WKH ZHLJKWV RI HLWKHU DQ /06 ILOWHU RU 5/6 ILOWHU )RU WKH /06 DGDSWDWLRQ UHVXOWV WKH /06 VWHSVL]H Q ZDV FKRVHQ WR EH WUDFH5f ZKHUH WUDFH5f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f b f ,Q DGGLWLRQ EDVHG RQ WKH GLVFXVVLRQ DERXW WKH DGYDQWDJHV RI V\QFKURQL]DWLRQ ZH ZHUH LQWHUHVWHG LQ REVHUYLQJ WKH URRW PHDQVTXDUHG HVWLPDWLRQ HUURU 506((f JLYHQ FRUUHFW DFTXLVLWLRQ 506(( AA>I[QIOGQQOA7Af@ f 7KH YDOXH RI WKH 506(( JLYHV D PHDVXUH RI KRZ ZHOO WKH UHFHLYHU FRXOG EH V\QFKURn QL]HG LI WKH WLPLQJ HVWLPDWH ZDV XVHG WR XSGDWH WKH UHFHLYHUfV WLPLQJ UHODWLYH WR WKH GHVLUHG XVHU (DFK '6&'0$ HQYLURQPHQW ZDV VLPXODWHG IRU LQGHSHQGHQW WULDOV )RU HDFK WULDO WKH XVHU VSUHDGLQJ FRGHV ZHUH VHOHFWHG DW UDQGRP IURP D VHW RI *ROG FRGHV XVLQJ FKLSVELW )RU HDFK WULDO WKH SURSDJDWLRQ GHOD\ IRU HDFK XVHU ZDV FKRVHQ IURP D XQLIRUP GLVWULEXWLRQ RYHU WKH LQWHUYDO >f LQGHSHQGHQW IURP WKH RWKHU XVHUV 7KH XVHUVf SKDVHVKLIWV ZHUH FKRVHQ IURP D XQLIRUP GLVWULEXWLRQ RYHU WKH LQWHUYDO > Wf LQGHSHQGHQW IURP WKH RWKHU XVHUV 7KH GHVLUHG XVHU ZDV UHFHLYHG DW D 615 RI G% $OVR DW WKH VWDUW RI HDFK WULDO WKH UHFHLYHG SRZHU OHYHO IRU HDFK LQWHUIHULQJ XVHU ZDV VHOHFWHG IURP D ORJQRUPDO GLVWULEXWLRQ WKDW KDG D PHDQ DQG VWDQGDUG GHYLDWLRQ
PAGE 67
RI G% 7KDW LV 3N3? fÂ§ L ZKHUH e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
PAGE 68
)LJXUH $YHUDJH WUDLQLQJ ELWV UHTXLUHG IRU FRUUHFW DFTXLVLWLRQ DV D IXQFWLRQ RI WKH QXPEHU RI V\VWHP XVHUV LQ $:*1 FKDQQHO UHVXOWV IRU WKH /06 EDVHG WLPLQJ HVWLPDWRU :LWK XVHUV b V\VWHP FDSDFLW\f WKH WLPLQJ HVWLPDWRU FDQ DFKLHYH FRDUVH DFTXLVLWLRQ ZLWK WUDLQLQJ ELWV ZKLFK LV QRW DQ XQUHDVRQDEO\ ORQJ WUDLQLQJ SHULRG )RU V\VWHP XVHUV WKH 5/6EDVHG WLPn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r IRU HDFK LQWHUIHULQJ XVHU ZDV VHOHFWHG IURP D ORJQRUPDO
PAGE 69
)LJXUH $YHUDJH WUDLQLQJ ELWV IRU FRUUHFW DFTXLVLWLRQ DV D IXQFWLRQ RI WKH OHYHO RI PXOWLDFFHVV LQWHUIHUHQFH LQ $:*1 FKDQQHO GLVWULEXWLRQ WKDW KDG D PHDQ DQG VWDQGDUG GHYLDWLRQ RI ; ZKHUH ; ZDV DOORZHG WR YDU\ RYHU WKH UDQJH RI f G% )RU D JLYHQ YDOXH RI WKH QXPEHU RI V\VWHP XVHUV DQG D VSHFLILF YDOXH RI $ LQGHSHQGHQW WULDOV ZHUH XVHG WR ILQG WKH FRDUVH DFTXLVLWLRQ SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU 7KLV VFHQDULR ZDV XVHG WR VLPXODWH DQ HQYLURQPHQW LQ ZKLFK WKHUH LV YHU\ ORRVH SRZHU FRQWURO DQG WKH GHVLUHG XVHU LV UHFHLYHG DW D SRZHU OHYHO ZKLFK RQ DYHUDJH LV RQH VWDQGDUG GHYLDWLRQ EHORZ WKH PHDQ RI WKH SRZHU OHYHOV RI DOO WKH LQWHUIHULQJ XVHUV )LJXUH VKRZV WKH FRDUVH DFTXLVLWLRQ SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWLRQ DOJRULWKPV IRU V\VWHP XVHUV DQG V\VWHP XVHUV DV D IXQFWLRQ RI WKH WKH YDOXH RI ; )URP WKLV ILJXUH LW LV VHHQ WKDW QRQH RI WKH WLPLQJ HVWLPDWRUV DUH WUXO\ QHDUIDU UHVLVWDQW &OHDUO\ WKH FRUUHODWRU LV YHU\ VHQVLWLYH WR WKH DPRXQW RI QHDUIDU LQWHUIHUn HQFH $V ZDV REVHUYHG LQ WKH SUHYLRXV SORW WKH /06EDVHG WLPLQJ HVWLPDWRU SHUIRUPV EHWWHU WKDQ WKH FRUUHODWRU DQG WKH 5/6EDVHG WLPLQJ HVWLPDWRU SHUIRUPV EHWWHU WKDQ
PAGE 70
WKH /06EDVHG WLPLQJ HVWLPDWRU )RU XVHUV WKH 5/6 EDVHG WLPLQJ HVWLPDWLRQ DOJRn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b HUURU RI ZKDW LV REVHUYHG WKURXJK GLUHFW VLPXODWLRQ :H WKHQ XVHG WKHVH WZR LWHUDWLYH HTXDWLRQV WR GHULYH WZR DQDO\WLFDO WHFKQLTXHV WR FKDUDFWHUL]H WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU 7KH JRDO RI WKH ILUVW WHFKQLTXH ZDV WR SURYLGH D PHDQV RI REVHUYLQJ KRZ TXLFNO\ WKH WLPLQJ HVWLPDWRU FDQ FRUUHFWO\ HVWLPDWH WKH SURSDJDWLRQ GHOD\ RI WKH GHVLUHG XVHU LQ D FKLSV\QFKURQRXV HQYLURQPHQW 7KH LGHD ZDV WR XVH WKH DQDO\WLFDO H[SUHVVLRQ DV
PAGE 71
D PHDQV WR REVHUYH WKH WLPLQJ HVWLPDWRU ZLWKRXW UXQQLQJ ORQJ 0RQWH&DUOR VLPXODn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Âf FRQGLWLRQHG RQ JHWWLQJ WKH LQWHJHU SDUW RI WKH WLPLQJ HVWLPDWH FRUUHFW :H REVHUYHG WKDW WKH UHVXOWLQJ DSn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b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
PAGE 72
WKH FKDQJH LQ WKH OHQJWK RI WKH UHTXLUHG WUDLQLQJ LQWHUYDO IRU WKH 5/6EDVHG DOJRULWKP ZDV VR PLQRU WKDW ZH FRQVLGHU LW WR EH QHDUIDU UHVLVWDQW :KLOH LW LV GLIILFXOW WR GLUHFWO\ FRPSDUH RXU WLPLQJ HVWLPDWRU ZLWK RWKHU WHFKQLTXHV DV VRPH RI WKH RWKHU WHFKQLTXHV HVWLPDWH PRUH WKDQ D VLQJOHXVHUfV SDUDPHWHUV RQH REVHUYDWLRQ FDQ EH PDGH 7KH FRPSOH[LW\ RI WKH HVWLPDWLRQ DOJRULWKP LV GHSHQGHQW RQ 1 WKH QXPEHU RI FKLSV LQ RQH SHULRG RI D VSUHDGLQJ VHTXHQFH 7KH FRPSOH[n LW\ RI WKH HVWLPDWLRQ DOJRULWKP LV L9f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
PAGE 73
&+$37(5 6,08/$7,1* )$',1* &+$11(/6 3UHYLRXVO\ WKH FRPPXQLFDWLRQ FKDQQHO ZDV PRGHOHG DV DQ $:*1 FKDQQHO $ PRUH UHDOLVWLF FKDQQHO PRGHO PXVW EH XVHG WR DFFRXQW IRU WKH GLVSHUVLYH HIIHFWV REn VHUYHG LQ XUEDQ HQYLURQPHQWV ,Q XUEDQ HQYLURQPHQWV LW LV SRVVLEOH IRU PXOWLSOH YHUVLRQV RI D WUDQVPLWWHG VLJQDO WR DUULYH DW D UHFHLYHU GXH WR UHIOHFWLRQV RII RI DGn MDFHQW EXLOGLQJV FDUV RU RWKHU REVWDFOHV ,I D WRWDO RI / YHUVLRQV RI WKH WUDQVPLWWHG VLJQDO DUH UHFHLYHG WKHQ WKH FKDQQHO LV VDLG WR KDYH / SDWKV 7KH VLJQDO UHFHLYHG IURP HDFK SDWK FDQ H[SHULHQFH LWV RZQ WLPH GHOD\ SKDVH VKLIW DQG DWWHQXDWLRQ 7KHUHIRUH DW WKH UHFHLYHU WKHVH / VLJQDOV PD\ DGG FRQVWUXFWLYHO\ RU GHVWUXFWLYHO\ D FRQGLWLRQ ZKLFK LV NQRZQ DV IDGLQJ ,I UHODWLYH PRWLRQ H[LVWV EHWZHHQ WKH WUDQVPLWWHU DQG WKH UHFHLYHU RU WKH UHIOHFWLQJ REVWDFOH LV LQ PRWLRQf WKHQ WKH UHFHLYHG VLJQDO LV DOVR REn VHUYHG WR KDYH D VKLIW LQ LWV FDUULHU IUHTXHQF\ $OVR GXH WR UHODWLYH PRWLRQ EHWZHHQ WKH UHFHLYHU DQG WKH WUDQVPLWWHU LW LV HDV\ WR FRQFHLYH WKDW WKH FKDUDFWHULVWLFV RI WKH FKDQQHO FKDQJH DV D IXQFWLRQ RI WLPH DQG WKHUHIRUH D PRUH UHDOLVWLF FKDQQHO PRGHO ZLOO DFFRXQW IRU WKH WLPH YDU\LQJ HIIHFWV REVHUYHG RQ WKH UHFHLYHG VLJQDO 5DSSDSRUW >@ SURYLGHV D VXPPDU\ RI WKH YDULRXV W\SHV RI IDGLQJ DV ZHOO DV SUHn VHQWLQJ VHYHUDO WHFKQLTXHV WKDW FDQ EH XVHG WR VLPXODWH IDGLQJ FKDQQHOV 3URDNLV >@ DOVR SUHVHQWV D FKDUDFWHUL]DWLRQ RI IDGLQJ FKDQQHOV DQG GLVFXVVHV WKH HIIHFW RI IDGn LQJ RQ WKH SHUIRUPDQFH RI FRPPXQLFDWLRQ V\VWHPV 7KH W\SH RI IDGLQJ LQGXFHG E\ D FRPPXQLFDWLRQ FKDQQHO LV GHSHQGHQW RQ WKH WLPHIUHTXHQF\ FKDUDFWHULVWLFV RI WKH FKDQQHO ZLWK UHVSHFW WR WR WKH WLPHIUHTXHQF\ FKDUDFWHULVWLFV RI WKH WUDQVPLWWHG VLJn QDO 6NODU >@ SUHVHQWV VHYHUDO EORFN GLDJUDPV WKDW VXPPDUL]H KRZ WKHVH UHODWLRQVKLSV EHWZHHQ WKH FKDQQHO DQG WKH WUDQVPLWWHG VLJQDO GHWHUPLQH WKH W\SH RI IDGLQJ LQGXFHG E\ WKH FKDQQHO ,Q DGGLWLRQ 6NODU SUHVHQWV D VXPPDU\ RI WUDGLWLRQDO WHFKQLTXHV WKDW
PAGE 74
KDYH EHHQ XVHG WR PLWLJDWH WKH HIIHFWV RI IDGLQJ >@ ,Q WKH UHPDLQGHU RI WKLV FKDSWHU ZH ZLOO EULHIO\ UHYLHZ WKH YDULRXV W\SHV RI IDGLQJ DQG GLVFXVV KRZ D IDGLQJ FKDQQHO FDQ EH VLPXODWHG &RQVLGHU D PXOWLSDWK FRPPXQLFDWLRQ FKDQQHO ZKHUH WKHUH DUH / GLVWLQFW SDWKV ZLWK XQLTXH SURSDJDWLRQ GHOD\V 7? U 7OL 7Of 7KH TXDQWLW\ fÂ§ 7? LV GHILQHG DV WKH H[FHVV GHOD\ VSUHDG $V H[SHFWHG WKLV PXOWLSDWK GHOD\ VSUHDG FDXVHV WLPH GLVSHUVLRQ LQ WKH UHFHLYHG VLJQDO ,Q IDFW LI WKH H[FHVV GHOD\ VSUHDG LV VLJQLILFDQW ZKHQ FRPSDUHG WR WKH V\PERO LQWHUYDO WKH FKDQQHO FDQ EH YLHZHG DV LQGXFLQJ VHYHUH LQWHUV\PERO LQWHUIHUHQFH ,6,f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fV VSHFWUXP ZLOO EH HTXDOO\ DWWHQXDWHG E\ WKH FKDQQHO ,Q WKLV FRQGLWLRQ WKH FKDQQHO LV VDLG WR EH IODW +RZHYHU GXH WR WKH WLPH YDU\LQJ QDWXUH RI WKH PXOWLSDWK FKDQQHO WKH FKDQQHOfV LPSXOVH UHVSRQVH DQG KHQFH LWV IUHTXHQF\ UHVSRQVH DUH WLPH YDULDQW ,Q RUGHU WR FKDUDFWHUL]H WKH IUHTXHQF\ UHVSRQVH RI WKH FKDQQHO WKH FRKHUHQFH EDQGZLGWK LV GHILQHG 7KH FRKHUHQFH EDQGn ZLGWK RI D FKDQQHO GHQRWHV D EDQGZLGWK LQ ZKLFK GLIIHUHQW IUHTXHQF\ FRPSRQHQWV H[SHULHQFH WKH VDPH DPSOLWXGH DWWHQXDWLRQ 7KH FRKHUHQFH EDQGZLGWK RI D PXOWL SDWK FKDQQHO LV UHODWHG WR WKH H[FHVV GHOD\ VSUHDG LQ VRPH PDQQHU :KLOH QR H[DFW UHODWLRQVKLS EHWZHHQ WKHVH WZR TXDQWLWLHV H[LVWV WKHUH DUH VHYHUDO ZHOO NQRZQ UXOHV RI WKXPE >@ /HW WKH WUDQVPLWWHG VLJQDO KDYH D EDQGZLGWK %V D V\PERO GXUDWLRQ
PAGE 75
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n VRLGDO VLJQDO LV WUDQVPLWWHG ,I WKHUH LV UHODWLYH PRWLRQ EHWZHHQ WKH WUDQVPLWWHU DQG UHFHLYHU WKHQ WKHUH LV D IUHTXHQF\ VKLIW REVHUYHG LQ WKH UHFHLYHG VLJQDO 7KH PD[LPXP REVHUYDEOH IUHTXHQF\ VKLIW IS LV NQRZQ DV WKH 'RSSOHU IUHTXHQF\ RU 'RSSOHU VSUHDG DQG LV JLYHQ E\ f
PAGE 76
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f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Â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f
PAGE 77
,I KRZHYHU WKHUH LV D GRPLQDQW SDWK LQ WKH FRPPXQLFDWLRQ FKDQQHO VXFK DV WKH FDVH ZKHQ D OLQHRIVLJKW /26f SDWK H[LVWV WKH HQYHORSH RI WKH UHFHLYHG VLJQDO LV PRGHOHG E\ D 5LFLDQ GLVWULEXWLRQ DQG KDV D SUREDELOLW\ GHQVLW\ IXQFWLRQ RI f RWKHUZLVH ZKHUH f LV WKH ]HURRUGHU PRGLILHG %HVVHO IXQFWLRQ RI WKH ILUVW NLQG 6LQFH WKH 5D\OHLJK GLVWULEXWLRQ LV MXVW D VSHFLDO FDVH RI WKH 5LFLDQ GLVWULEXWLRQ LW LV HDV\ WR FUHDWH D IDGLQJ FKDQQHO VLPXODWRU WKDW JHQHUDWHV ERWK GLVWULEXWLRQV IRU WKH HQYHORSH RI WKH UHFHLYHG VLJQDO 7KH IDGLQJ FKDQQHO VLPXODWRU XVHG LQ WKLV ZRUN LV EDVHG RQ &ODUNHfV >@ FKDQQHO PRGHO ,Q WKDW ZRUN WKH IDGLQJ ZDV GXH WR WKH VFDWWHULQJ RI HOHFWURPDJQHWLF ZDYHV LQ VXFK D ZD\ WKDW DOO SDWKV KDG WKH VDPH SURSDJDWLRQ GHOD\ 7KLV PHDQV WKDW DOO SDWKV IRUP RQH FRPSRVLWH SDWK DQG VLQFH WKHUH LV QR GHOD\ VSUHDG WKH FKDQQHO LQGXFHV IODW IDGLQJ 7KH UHFHLYHG HQYHORSH ZDV VKRZQ WR EH D UDQGRP SURFHVV ZKRVH DPSOLWXGH IROORZV D 5D\OHLJK GLVWULEXWLRQ DQG KDV DQ DXWRFRUUHODWLRQ IXQFWLRQ JLYHQ E\ 5Wf 0QI'7f f ZKHUH f LV WKH ]HURRUGHU %HVVHO IXQFWLRQ RI WKH ILUVW NLQG 7KHUHIRUH WR VLPXODWH D FKDQQHO EDVHG RQ WKLV PRGHO ZH PXVW FUHDWH D UDQGRP SURFHVV ZKRVH DPSOLWXGH LV 5D\OHLJK GLVWULEXWHG LQ VXFK D ZD\ WKDW WKH DXWRFRUUHODWLRQ RI WKH UDQGRP SURFHVV LV VLPLODU WR HTXDWLRQ f $ WKLUGRUGHU ILOWHU KDV DQ WUDQVIHU IXQFWLRQ RI f
PAGE 78
DQG D FRUUHVSRQGLQJ LPSXOVH UHVSRQVH RI KWf H[SfÂ§FWIf >$ VLQWf % FRVWf? & H[SaXMWff XWf f ZKHUH $ 8T A" D &AR % fÂ§ & : & %\ VHWWLQJ XT DQG & EDVHG RQ WKH YDOXH RI WKH 'RSSOHU IUHTXHQF\ DV VKRZQ EHORZ 8T W?BG & f f WKH FRQYROXWLRQ RI K>Uf ZLWK LfÂ§Uf FORVHO\ IROORZV HTXDWLRQ f RYHU WKH ILUVW PDMRU OREH 7KLV PHDQV WKDW LI D FRPSOH[ ZKLWH *DXVVLDQ QRLVH SURFHVV LV XVHG DV WKH LQSXW WR WKH WKLUGRUGHU ILOWHU WKH DXWRFRUUHODWLRQ RI WKH RXWSXW SURFHVV FORVHO\ IROORZV HTXDWLRQ f IRU YDOXHV RI U ZKHUH VDPSOHV RI WKH SURFHVV DUH KLJKO\ FRUUHODWHG $QG KHQFH ZH XVH WKLV WKLUGRUGHU ILOWHU WR VLPXODWH D IODWIDGLQJ FKDQQHO EDVHG RQ &ODUNHfV PRGHO )LJXUH VKRZV D EORFN GLDJUDP RI D IDGLQJ FKDQQHO VLPXODWRU EDVHG RQ WKH WKLUGRUGHU ILOWHU WKDW FDQ EH XVHG RQ D GLJLWDO FRPSXWHU 7KH FRQWLQXRXVWLPH ILOWHU LV FRQYHUWHG WR D GLJLWDO ILOWHU ZKHUH WKH VDPSOLQJ UDWH LV VSHFLILHG W\SLFDOO\ HTXDO WR RQH ELW LQWHUYDO 7KH LQSXW WR WKH ILOWHU DUH VDPSOHV RI FRPSOH[ *DXVVLDQ QRLVH ZKRVH UHDO DQG LPDJLQDU\ SDUWV DUH LQGHSHQGHQW KDYH ]HURPHDQ DQG HTXDO YDULn DQFHV RI 7KH EORFN GLDJUDP DFFRXQWV IRU WKH SUHVHQFH RI DQ /26 RU VSHFXn ODU FRPSRQHQW VXFK WKDW WKH HQYHORSH RI WKH VLPXODWRU RXWSXW FDQ IROORZ ERWK WKH 5D\OHLJK DQG 5LFLDQ GLVWULEXWLRQV 7KHUHIRUH WKH RXWSXW RI WKH FKDQQHO VLPXODWRU LV
PAGE 79
6DPSOHV RI &RPSOH[ *DXVVLDQ 1RLVH )LJXUH )ODWIDGLQJ FKDQQHO VLPXODWRU D VDPSOHG FRPSOH[ *DXVVLDQ SURFHVV KDYLQJ D PHDQ RI $ H[SMFSf DQG D YDULDQFH RI D ( >?$N H[SMNf $H[SMcff?@ ,Q )LJXUH WZR SDUDPHWHUV DUH VKRZQ WR EH D IXQFWLRQ RI D YDULDEOH FDOOHG S 7KH UDWLR S $D LV UHIHUUHG WR DV WKH 5LFLDQ SDUDPHWHU DQG WKH SDLU $ Df DUH QRUPDOL]HG VR WKDW WKH IDGLQJ SURFHVV KDV XQLW SRZHU r^3f 5 3 a 9 S f $Sfa SUU 3Ar 9 3 f $ SORW FRPSDULQJ WKH WKHRUHWLFDO SUREDELOLW\ GHQVLW\ IXQFWLRQV RI WKH HQYHORSH RI WKH IDGLQJ SURFHVV JLYHQ LQ HTXDWLRQV f DQG f WR WKH WKRVH REVHUYHG WKURXJK VLPXODWLRQV IRU WZR YDOXHV RI S LV GLVSOD\HG LQ )LJXUH $V VKRZQ LQ WKLV ILJXUH ZKHQ S WKH HQYHORSH RI WKH IDGLQJ SURFHVV LV 5D\OHLJK GLVWULEXWHG ZKLOH IRU RWKHU YDOXHV WKH HQYHORSH RI WKH IDGLQJ SURFHVV IROORZV D 5LFLDQ GLVWULEXWLRQ $V D ILQDO GHPRQVWUDWLRQ RI WKH IDGLQJ FKDQQHO VLPXODWRU )LJXUH GLVSOD\V D W\SLFDO UHDOL]DWLRQ RI D IODW 5D\OHLJK IDGLQJ SURFHVV 7KH PDJQLWXGH RI WKH HQYHORSH DQG WKH SKDVH RI WKH IDGLQJ SURFHVV DUH VKRZQ ZKHQ WKH FDUULHU IUHTXHQF\ ZDV 0+] WKH GDWD UDWH ZDV ELWVSHUVHFRQG %36f DQG WKH UHODWLYH VSHHG EHWZHHQ WKH WUDQVPLWWHU DQG WKH UHFHLYHU ZDV PLOHVSHUKRXU 03+f 7KH WKLUG RUGHU GLJLWDO ILOWHU DV ZHOO DV WKH RXWSXW RI WKH IDGLQJ FKDQQHO VLPXODWRU ZKHUH VDPSOHG DW WKH GDWD UDWH
PAGE 80
2 U )LJXUH 7KHRUHWLFDO DQG VLPXODWHG SUREDELOLW\ GHQVLW\ IXQFWLRQ IRU WKH HQYHORSH RI WKH IDGLQJ FKDQQHO VLPXODWRU RXWSXW IRU WZR GLIIHUHQW 5LFLDQ SDUDPHWHU YDOXHV 6DPSOH 1XPEHU )LJXUH 0DJQLWXGH DQG SKDVH RI D VORZ IODWIDGLQJ 5D\OHLJK SURFHVV JHQHUDWHG ZLWK D FDUULHU IUHTXHQF\ RI *+] D YHKLFOH VSHHG RI 03+ DQG D GDWD UDWH RI %36
PAGE 81
&RQVLGHU VLPXODWLQJ D VLQJOHXVHU '6&'0$ V\VWHP RSHUDWLQJ LQ D IDGLQJ FKDQQHO HQYLURQPHQW )RU D JLYHQ 'RSSOHU IUHTXHQF\ WKH IDGLQJ FKDQQHO VLPXODWRU LV FUHDWHG DQG DVVXPH WKDW ERWK WKH WKLUGRUGHU ILOWHU DV ZHOO DV WKH VLPXODWRU RXWSXW DUH VDPn SOHG DW WKH V\PERO UDWH ,Q D QRQIDGLQJ FKDQQHO WKH FRQWULEXWLRQ WR WKH UHFHLYHUfV LQSXW GXULQJ WKH PWK V\PERO LQWHUYDO GXH WR WKH XVHU LV JLYHQ E\ LPf DV VKRZQ LQ HTXDWLRQ f 0XOWLSOLFDWLRQ RI LPf E\ WKH PWK VDPSOHG RXWSXW RI WKH IDGLQJ FKDQQHO VLPXODWRU $Pf H[SMPff DW HDFK V\PERO LQWHUYDO ZLOO SURGXFH DQ LQSXW WR WKH UHFHLYHU ZKRVH HQYHORSH H[SHULHQFHV IDGLQJ ,Q RUGHU WR VLPXODWH D PXOWLXVHU IDGLQJ FKDQQHO '6&'0$ V\VWHP HDFK XVHU VKRXOG KDYH LWV RZQ IDGLQJ SURFHVV WKDW LV LQGHSHQGHQW IURP DOO RWKHU XVHUV :H FRXOG LPSOHPHQW LQGHSHQGHQW IDGLQJ FKDQn QHO VLPXODWRUV DQG WKHQ PXOWLSO\ rPf E\ WKH RXWSXW RI WKH FRUUHVSRQGLQJ IDGLQJ FKDQQHO VLPXODWRU +RZHYHU D VLPSOHU DSSURDFK ZDV XVHG LQ WKLV ZRUN $ VLQJOH IDGLQJ FKDQQHO VLPXODWRU ZDV FUHDWHG IRU D VSHFLILHG 'RSSOHU IUHTXHQF\ 7KHQ D YHU\ ODUJH VHTXHQFH RI VDPSOHV RI WKH IDGLQJ SURFHVV ZDV FUHDWHG &RQVLGHU WZR VDPSOHV RI WKLV VHTXHQFH $Qf H[S^MGQff DQG $Q Pf H[SMGQ Pff :H NQRZ WKDW LI P LV FKRVHQ DSSURSULDWHO\ WKHQ WKH FRUUHODWLRQ EHWZHHQ WKH VDPSOHV EHFRPHV QHJOLJLEOH DQG WKHUHIRUH ZH FDQ WUHDW WKH VDPSOHV DV EHLQJ LQGHSHQGHQW 8VLQJ WKLV LGHD WKH RULJLQDO VHTXHQFH ZDV SDUWLWLRQHG LQWR VHTXHQFHV VXFK WKDW HDFK VHTXHQFH LV FRQn VLGHUHG WR EH LQGHSHQGHQW RI HDFK RWKHU 7KDW LV WKH VHTXHQFH IRU WKH ILUVW XVHU ZDV FUHDWHG XVLQJ VDPSOHV Pf ZKLOH WKH VHTXHQFH RI WKH VHFRQG XVHU ZDV FUHDWHG E\ XVLQJ VDPSOHV P P Pf DQG VR RQ IRU HDFK XVHU
PAGE 82
&+$37(5 &2'( $&48,6,7,21 ,1 12167$7,21$5< (19,5210(176 ,W LV ZHOONQRZQ WKDW WUDLQLQJ DQ DGDSWLYH ILOWHU LQ D QRQVWDWLRQDU\ HQYLURQPHQW FDQ EH GLIILFXOW >@ 6LQFH RXU WLPLQJ HVWLPDWRU LV EDVHG RQ SURFHVVLQJ WKH ZHLJKWV RI WKH DGDSWLYH ILOWHU D VWXG\ RI KRZ WKH HVWLPDWRU LV DIIHFWHG E\ VXFK HQYLURQPHQWV LV UHTXLUHG ,Q WKLV FKDSWHU ZH ZLOO FRQVLGHU WLPLQJ DFTXLVLWLRQ LQ WZR VXFK HQYLURQn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fV VLJQDO H[SHULHQFHV D GHHS IDGH WKH UHFHLYHU IUHTXHQWO\ ORVHV ORFN RQ WKH GHVLUHG VLJQDO :KHQ WKH GHVLUHG XVHUfV VLJQDO HPHUJHV IURP WKH GHHS IDGH WKH UHFHLYHU PD\ HPHUJH LQWR DQ\ RI SRVVLEOH VWDWHV FRUUHFWO\ ORFNHG RQ SKDVH WR WKH GHVLUHG XVHUfV VLJQDO ORFNHG r RXW RI SKDVH WR WKH GHVLUHG XVHUfV VLJQDO RU ORFNHG HLWKHU LQ SKDVH RU r RXW RI SKDVH WR DQ\ RI WKH LQWHUIHULQJ XVHU VLJQDOV ,Q D VLQJOHXVHU HQYLURQPHQW RQO\ WKH ILUVW WZR RI WKHVH FRQGLWLRQV DUH SRVVLEOH 7KH ILUVW FRQGLWLRQ LV GHVLUHG 7KH VHFRQG FRQGLWLRQ LV QRW GXH WR WKH ILOWHUfV LQDELOLW\ WR WUDFN WKH UDSLG SKDVH FKDQJHV GXULQJ D GHHS IDGH ,W LV GXH WR WKH IDFW WKDW VLQFH WKH GHFLVLRQV DUH XQUHOLDEOH LQ D GHHS IDGH PDQ\ HUURUV RFFXU LQ WKH IHHGEDFN ORRSf WKH
PAGE 83
ILOWHU LV HVVHQWLDOO\ UXQQLQJ fEOLQGf GXULQJ WKH GHHS IDGH 2QFH WKH ILOWHU ORFNV RQWR WKH GHVLUHG XVHUfV VLJQDO DW r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f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f GLPfZUPfUPf 7KH ZHLJKWV RI WKH ILOWHU DUH XSGDWHG DW WKH ELW UDWH ZKLFK PHDQV WKDW Pf FDQ EH XSGDWHG DW WKH ELW UDWH 7KHUHIRUH ZH PXVW SUHGLFW WKH FKDQQHO FRQGLWLRQV IRU WKH FXUUHQW ELW LQWHUYDO EDVHG RQ SUHYLRXV YDOXHV RI Pf $Q /WKRUGHU OLQHDU SUHGLFWRU LV XVHG WR IRUP WKH FXUUHQW HVWLPDWH RI WKH FKDQQHO
PAGE 84
FRQGLWLRQV EDVHG RQ SDVW HVWLPDWHV DV VKRZQ LQ HTXDWLRQ f / 0 rLr1 f Of fÂ§ DQG WKH HVWLPDWH RI WKH SKDVH GXULQJ WKH PWK ELW LQWHUYDO LV IRXQG E\ AP =Pf 7KH FRHIILFLHQWV RI WKH /WKRUGHU OLQHDU SUHGLFWRU DUH FKRVHQ WR PLQLPL]H WKH PHDQ VTXDUHG HUURU (>_DLP H[SMI"LPf fÂ§ UDf_@ ZKHUH FrLP DQG LP DUH WKH DPSOLWXGH DQG SKDVH LQGXFHG RQ WKH GHVLUHG XVHU E\ WKH FKDQQHO GXULQJ WKH PWK ELW LQWHUYDO 7KH VROXWLRQ RI WKHVH FRHIILFLHQWV GHSHQGV RQ WKH ZHLJKWV RI WKH 006( ILOWHU ZKLFK GHSHQG RQ WKH PXOWLDFFHVV LQWHUIHUHQFH ,Q RUGHU WR UHPRYH WKLV GHSHQGHQFH WKH FRHIILFLHQWV ZHUH FDOFXODWHG IRU D VLQJOHXVHU V\VWHP DQG WKHQ DUH XVHG IRU DOO RWKHU HQYLURQPHQWV 1RW VXUSULVLQJO\ WKH FRHIILFLHQWV RI WKH /WKRUGHU OLQHDU SUHGLFWRU D DUH IRXQG E\ D &Y f ZKHUH 0Â PI'7Ef rf^` f ^%`LOLMfQI'7Ef LM ^` f & % (E.f f DQG LV DQ / [ / ,GHQWLW\ PDWUL[ DQG I' LV WKH PD[LPXP 'RSSOHU IUHTXHQF\ LQGXFHG E\ WKH FKDQQHO ,W LV LPSRUWDQW WR UHPHPEHU WKDW WKHVH HTXDWLRQV ZHUH GHULYHG IRU D VLQJOHXVHU DQG WKHUHIRUH LQ WKH SUHVHQFH RI PXOWLDFFHVV LQWHUIHUHQFH WKH FKDQQHO HVWLPDWRU PD\ EH YHU\ IDU IURP RSWLPDO 7KH PRGLILHG IRUP RI WKH DGDSWLYH UHFHLYHU XVLQJ WKH /WKRUGHU OLQHDU SUHGLFWRU GHVLJQHG E\ %DUERVD DQG 0LOOHU HTXDWLRQV f WKURXJK ff LV VKRZQ LQ )LJXUH 1RZ WKDW ZH KDYH SUHVHQWHG WKH EDFNJURXQG RI WKLV UHFHLYHU ZH VXPPDUL]H VHYHUDO
PAGE 85
r (VWLPDWLRQ 7L ZPf IU DOJRULWKP LPSRUWDQW UHVXOWV SUHVHQWHG E\ %DUERVD DQG 0LOOHU >@ 7KH\ IRXQG WKDW WKH SHUIRUn PDQFH RI WKH UHFHLYHU GHWHFWLRQf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f ,Q IDFW LW ZDV VKRZQ WKDW WKH V\VWHP FDSDFLW\ ZKHQ XVLQJ WKH PRGLILHG UHFHLYHU VWUXFWXUH FDQ EH PDGH WR DSSURDFK b 1 XVHUVf ZKLOH WKH FRQYHQWLRQDO UHFHLYHU W\SLFDOO\ UHVWULFWV WKH V\VWHP FDSDFLW\ WR RQ WKH RUGHU RI b 6HYHUDO FRPPHQWV DERXW WKH PRGLILHG UHFHLYHU VWUXFWXUH DV LW UHODWHV WR WLPLQJ DFTXLVLWLRQ DUH UHTXLUHG DW WKLV SRLQW %DUERVD DQG 0LOOHU ZHUH LQWHUHVWHG LQ GHWHFWLRQ DQG WKH ZHLJKWV RI WKH DGDSWLYH ILOWHU ZHUH LQLWLDOL]HG FORVH WR WKHLU VWHDG\VWDWH YDOXHV 7KH UHVXOW LV WKDW WKH /WKRUGHU OLQHDU SUHGLFWRU FDQ PDNH UHOLDEOH HVWLPDWHV RI WKH FKDQQHO FRQGLWLRQV DW WKH VWDUW RI WKHLU VLPXODWLRQV ,Q WKLV ZRUN ZH DUH LQWHUHVWHG LQ WLPLQJ DFTXLVLWLRQ DQG WKHUHIRUH WKH ZHLJKWV RI WKH DGDSWLYH ILOWHU DUH LQLWLDOL]HG WR
PAGE 86
D YHFWRU RI DOO ]HURV 7KLV PHDQV WKDW PRVW OLNHO\ WKH LQLWLDO HVWLPDWHV RI WKH FKDQQHO SKDVH ZLOO EH LQFRUUHFW $V VHHQ LQ )LJXUH WKH SKDVH FRPSHQVDWLRQ RSHUDWLRQ 5H UPf H[SfÂ§ M?APf FDXVHV D GHOD\ LQ WKH EXLOG RI HQHUJ\ LQ WKH ILOWHU VLQFH ZKHQ ?P fÂ§ LPf LV VLJQLILFDQW WKH SKDVH FRPSHQVDWLRQ RSHUDWLRQ UHMHFWV VRPH RI WKH GHVLUHG XVHUfV LQSXW VLJQDO IURP WKH LQSXW WR WKH DGDSWLYH ILOWHU 7KH WLPH WR DFKLHYH FRUUHFW DFTXLVLWLRQ ZLOO EH GHSHQGHQW LQ VRPH PDQQHU RQ WKH WUDQVLHQW UHVSRQVH RI WKH FKDQQHO HVWLPDWLRQ DOJRULWKP )UHTXHQF\ 6\QFKURQL]DWLRQ (UURUV &RQVLGHU WKH VLWXDWLRQ WKDW RFFXUV ZKHQ WKH IUHTXHQF\ RI WKH $WK XVHUfV WUDQVPLWn WLQJ RVFLOODWRU LV QRW HTXDO WR WKH IUHTXHQF\ RI WKH UHFHLYHUfV RVFLOODWRU ,Q WKLV FDVH DIWHU GRZQFRQYHUVLRQ WKH LQSXW WR WKH DGDSWLYH ILOWHU IRU WKH NWK XVHU ZLOO KDYH D OLQHDUO\ WLPHYDU\LQJ SKDVH ,Q RUGHU WR PRGHO WKLV HIIHFW ZH DVVXPH WKDW WKH IUHn TXHQF\ RIIVHW IRU WKH NWK XVHU $X!N FDQ EH H[SUHVVHG DV $X!N Q17F0Nf ZKHUH 0N LV D SRVLWLYH LQWHJHU DQG 17& LV WKH ELW LQWHUYDO 8VLQJ WKLV QRWDWLRQ ZH FDQ PRGHO WKH LQSXW WR WKH DGDSWLYH ILOWHU XVLQJ HTXDn WLRQ f RQFH ZH UHGHILQH NPf DV VKRZQ EHORZ Df =NLPfYNL3N tNf ]NPfYNSN f ZKHUH WKH QWK FRPSRQHQWV RI YNBLSN6Nf DQG YNSNNf DUH JLYHQ E\ Ef
PAGE 87
$V D VDQLW\ FKHFN RI WKHVH HTXDWLRQV QRWH WKDW LQ WKH OLPLW 0N fÂ§}Â‘ RR WKH DERYH HTXDWLRQV UHGXFH WR WKH SUHYLRXV GHILQLWLRQ RI IFPf 7KH UHVXOWV LQ )LJXUH VKRZ WKH DFTXLVLWLRQ SHUIRUPDQFH RI WKH HVWLPDWLRQ DOJRULWKP LQ WKH SUHVHQFH RI IUHTXHQF\ V\QFKURQL]DWLRQ HUURUV 7KH GHVLUHG XVHU ZDV UHFHLYHG DW D 615 RI G% DQG LQWHUIHULQJ XVHUV ZHUH HDFK UHFHLYHG DW D SRZHU OHYHO G% DERYH WKH GHVLUHG XVHU 7KH GHVLUHG XVHU ZDV DVVXPHG WR KDYH D IUHTXHQF\ RIIVHW RI b RI WKH GDWD UDWH 0L f ZKLOH DOO RWKHU XVHUV KDG D IUHTXHQF\ RIIVHW RI b RI WKH GDWD UDWH 0 f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fV VLJQDO LQ WKH
PAGE 88
A Ms/ 77 D 7LPH 6\PEROVf )LJXUH $FTXLVLWLRQ SHUIRUPDQFH RI WKH DGDSWLYH UHFHLYHU LQ D ILYHXVHU HQYLURQn PHQW LQ SUHVHQFH RI IUHTXHQF\ RIIVHW HUURUV ZHLJKWV RI WKH DGDSWLYH ILOWHU 7KH WKRUGHU OLQHDU SUHGLFWLRQ FRPSHQVDWHG ILOWHU XQOLNH WKH XQFRPSHQVDWHG ILOWHU LV DEOH WR DFKLHYH FRUUHFW DFTXLVLWLRQ DW WKH FRVW RI DQ LQFUHDVH LQ WKH OHQJWK RI WKH WUDLQLQJ LQWHUYDO &RQVLGHULQJ WKH UHVXOWV RI )LJXUH LW VHHPV UHDVRQDEOH WKDW D K\EULG RI WKH XQFRPSHQVDWHG DQG FRPSHQVDWHG ILOWHUV PD\ EH XVHIXO LQ IRUPLQJ D WLPLQJ HVWLn PDWH 7KDW LV LW PD\ EH SRVVLEOH WR VWDUW DFTXLVLWLRQ XVLQJ DQ XQFRPSHQVDWHG ILOWHU ZLWK FRPSOH[ ZHLJKWVf DQG DW VRPH SRLQW VZLWFK WR WKH FRPSHQVDWHG ILOWHU ZLWK UHDO ZHLJKWVf VXFK WKDW ZH JHW EHWWHU DFTXLVLWLRQ SHUIRUPDQFH :H ZLOO DGGUHVV WKLV LGHD LQ WKH QH[W VHFWLRQ ZKHUH ZH VWXG\ WKH SHUIRUPDQFH RI WKH DFTXLVLWLRQ DOJRULWKP LQ D IODWIDGLQJ FKDQQHO
PAGE 89
3HUIRUPDQFH LQ )ODW)DGLQJ &KDQQHOV :H PRGHO WKH LQSXW WR WKH DGDSWLYH ILOWHU XVLQJ HTXDWLRQ f E\ GHILQLQJ WKH FRQWULEXWLRQ RI WKH NWK XVHU GXULQJ WKH PWK ELW LQWHUYDO DV MWPf DNPf H[SM2NPff >]NLPfDNLSN6Nf ]N^PftN^SN Nf@ f ZKHUH DNPf DQG NPf DUH WKH DPSOLWXGH DQG SKDVH LQGXFHG RQ WKH NWK XVHUfV VLJQDO E\ WKH IDGLQJ SURFHVV GXULQJ WKH PWK ELW LQWHUYDO 7KH IDGLQJ SURFHVV IRU HDFK XVHU LV DVVXPHG WR EH LQGHSHQGHQW RI WKH IDGLQJ SURFHVV LQGXFHG RQ HDFK RI WKH RWKHU XVHUV 7KH IDGLQJ SDUDPHWHUV DNPf DQG NPf DUH WDNHQ IURP VDPSOHV RI WKH IODWIDGLQJ FKDQQHO VLPXODWRU GHVFULEHG LQ WKH SUHYLRXV FKDSWHU $V D ILUVW WHVW RI WKH DFTXLVLWLRQ DOJRULWKP D WHQXVHU '6&'0$ HQYLURQPHQW ZDV FUHDWHG ZLWK D 615 RI G% DQG DOO LQWHUIHULQJ XVHUV ZHUH UHFHLYHG DW WKH VDPH SRZHU OHYHO DV WKH GHVLUHG XVHU LQ WKH DEVHQFH RI IDGLQJf $ VORZO\ IODWIDGLQJ FKDQQHO ZDV FUHDWHG IRU HDFK XVHU XVLQJ D FDUULHU IUHTXHQF\ RI *+] D GDWD UDWH RI %36 DQG D VSHHG RI 03+ 7KHVH SDUDPHWHUV DUH FRPPRQO\ XVHG WR VLPXODWH D SHUVRQDO FRPPXQLFDWLRQ V\VWHP 3&6f HQYLURQPHQW LQ ZKLFK WKH XVHUV DUH PRELOH DW ZDONLQJ VSHHGV 7KH UHVXOWLQJ 'RSSOHU IUHTXHQF\ I' LV +] ZKLFK LV D QRUPDOL]HG UDWH QRUPDOL]HG WR WKH GDWD UDWHf RI [ 7R WHVW WKH DFTXLVLWLRQ DOJRULWKP LQ WKLV HQYLURQPHQW D EDQN RI IRXU SDUDOOHO 5/6 ILOWHUV ZDV XVHG 2QH ILOWHU ZDV XQFRPSHQVDWHG DV VKRZQ LQ )LJXUH DQG SURFHVVHG WKH UHFHLYHG VLJQDO XVLQJ FRPSOH[ ZHLJKWV 7ZR RI WKH ILOWHUV XVHG WKH DUFKLWHFWXUH VKRZQ LQ )LJXUH ZLWK D /WKRUGHU OLQHDU SUHGLFWRU IRU WKH FKDQQHO HVWLPDWRU ,Q WKH IROORZLQJ ZH ZLOO UHIHU WR HDFK RI WKHVH ILOWHUV DV WKH PRGLILHG 006( UHFHLYHU 2QH RI WKH PRGLILHG 006( UHFHLYHUV FRPSHQVDWHG WKH UHFHLYHG VLJQDO XVLQJ RQO\ WKH SKDVH HVWLPDWH RI Pf DV VKRZQ LQ )LJXUH f 7KH VHFRQG PRGLILHG 006( UHFHLYHU FRPSHQVDWHG WKH UHFHLYHG VLJQDO XVLQJ ERWK WKH DPSOLWXGH DQG SKDVH RI f 7KH IRXUWK ILOWHU UHIHUUHG WR DV WKH K\EULG UHFHLYHU ZDV DQ DGKRF PL[WXUH RI WKH
PAGE 90
WZR UHFHLYHU VWUXFWXUHV 7KH LGHD RI WKH K\EULG UHFHLYHU LV WR MXPSVWDUW WKH FKDQQHO HVWLPDWLRQ DOJRULWKP E\ SUHORDGLQJ WKH DGDSWLYH ILOWHUfV ZHLJKW YHFWRU ZPf DQG WKH OLQHDU SUHGLFWRU YHFWRU KRSHIXOO\ E\SDVVLQJ VRPH RI WKH OHDUQLQJ FXUYH UHTXLUHG WR IRUP UHDVRQDEOH FKDQQHO HVWLPDWHV RWKHUZLVH 7KH K\EULG UHFHLYHU XVHV WKH VWUXFWXUH RI )LJXUH IRU ELW LQWHUYDOV ZLWKRXW PDNLQJ WLPLQJ HVWLPDWHV $IWHU WKH WK ELW LQWHUYDO WKH UHFHLYHU IRUPV DQ HVWLPDWH RI 7? DQG WKHQ PDNHV DQ HVWLPDWH RI ? XVLQJ HTXDWLRQ Df 7KLV SKDVH HVWLPDWH LV ORDGHG LQWR WKH /HOHPHQWV RI DQG WKH LPDJLQDU\ SDUWV RI WKH ILOWHUfV ZHLJKW YHFWRU DQG WKH LQYHUVH RI WKH FRUUHODWLRQ PDWUL[ DUH UHPRYHG 7KH UHFHLYHU WKHQ VZLWFKHV WR WKH PRGLILHG UHFHLYHU VWUXFWXUH VKRZQ LQ )LJXUH DQG VWDUWV WKH WLPLQJ DQG FKDQQHO HVWLPDWLRQ DOJRULWKPV XVLQJ WKH QRZ UHDOf ZHLJKWV RI WKH DGDSWLYH ILOWHU 7KH UHVXOWV RI WKLV WHVW ZKHQ D WKLUGRUGHU OLQHDU SUHGLFWRU ZDV XVHG IRU WKH FKDQQHO HVWLPDWLRQ DOJRULWKP DUH VKRZQ LQ )LJXUH $V VHHQ LQ WKLV ILJXUH WKH DFTXLVLWLRQ DOJRULWKP EDVHG RQ SURFHVVLQJ WKH ZHLJKWV RI WKH XQFRPSHQVDWHG ILOWHU SHUIRUPV EHWn WHU WKDQ DQ\ RI WKH RWKHU WHFKQLTXHV ,Q JHQHUDO WKLV VLWXDWLRQ ZLOO QRW RFFXU DV WKH OHYHO RI WKH 0$, FKDQJHV DQGRU WKH 'RSSOHU IUHTXHQF\ FKDQJHV DQG UHVXOWV VLPLODU WR WKDW VKRZQ LQ )LJXUH FDQ EH H[SHFWHG %DUERVD DQG 0LOOHU >@ REVHUYHG WKH GHWHFWLRQ SHUIRUPDQFH RI WKH PRGLILHG 006( UHFHLYHU ZDV LQVHQVLWLYH WR / WKH RUGHU RI WKH OLQHDU SUHGLFWRU XVHG LQ FKDQQHO HVWLPDWLRQ ,Q DGGLWLRQ WKH\ FRPSHQVDWHG WKH UHFHLYHG VLJQDO E\ RQO\ XVLQJ WKH SKDVH RI WKH FKDQQHO HVWLPDWH f DV QR DGYDQn WDJH ZDV REVHUYHG E\ DOVR XVLQJ WKH DPSOLWXGH RI f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
PAGE 91
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n GLFWRU LV XVHG DV QRWHG E\ WKH GLIIHUHQFH EHWZHHQ WKH UDWHV RI FRQYHUJHQFH RI WKH WZR SUREDELOLW\ FXUYHV LQ )LJXUHV DQG )RU D YDULHW\ RI '6&'0$ HQYLURQPHQWV 'RSSOHU IUHTXHQF\ 0$, DQG QXPEHU RI XVHUVf LW KDV EHHQ REVHUYHG WKDW QR VLJQLIn LFDQW LPSURYHPHQW LQ WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU LV DFKLHYHG E\ XVLQJ D KLJKHU RUGHU OLQHDU SUHGLFWRU LQ WKH FKDQQHO HVWLPDWLRQ DOJRULWKP 7KHUHIRUH LQ
PAGE 92
)LJXUH 3HUIRUPDQFH RI WKH DFTXLVLWLRQ DOJRULWKP LQ D VORZO\ IODWIDGLQJ 5D\OHLJK FKDQQHO XVLQJ D WHQWKRUGHU OLQHDU SUHGLFWRU IRU WKH FKDQQHO HVWLPDWLRQ DOJRULWKP WKH VHTXHO ZKHQ ZH UHIHU WR WKH PRGLILHG 006( UHFHLYHU ZH DUH XVLQJ D WHQWKRUGHU OLQHDU SUHGLFWRU DQG RQO\ WKH SKDVH RI WKH FKDQQHO HVWLPDWH LV XVHG WR FRPSHQVDWH WKH UHFHLYHG VLJQDO $V VHHQ LQ )LJXUHV DQG HYHQWKRXJK WKH K\EULG UHFHLYHU KDV EHHQ IRUPXODWHG LQ DQ DGKRF IDVKLRQ WKH UHVXOWLQJ SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU FOHDUO\ VKRZV WKDW WKH LGHD KDV LWV PHULWV ,Q RUGHU WR IXUWKHU WHVW WKH K\EULG UHFHLYHU D VWXG\ RI WKH HIIHFWV RI WKH OHYHO RI 0$, RQ WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWLRQ DOJRULWKP VLPLODU WR WKDW VKRZQ LQ )LJXUH ZDV SHUIRUPHG 7KH FDUULHU IUHTXHQF\ ZDV 0+] WKH GDWD UDWH ZDV %36 DQG WKH UHODWLYH VSHHG EHWZHHQ WKH UHFHLYHU DQG DOO RI WKH V\VWHP XVHUV ZDV DVVXPHG WR EH 03+ 7KLV VFHQDULR LV XVHG WR GHSLFW D '6&'0$ V\VWHP LQ ZKLFK WKH XVHUV DUH WUDYHOLQJ DW KLJKZD\ VSHHGV 7KH 'RSSOHU IUHTXHQF\ LQGXFHG E\ WKH IODWIDGLQJ FKDQQHO ZDV +] ZKLFK LV D QRUPDOL]HG 'RSSOHU UDWH RI [ a (DFK '6&'0$ HQYLURQPHQW ZDV VLPXODWHG IRU LQGHSHQGHQW WULDOV )RU HDFK WULDO WKH XVHU VSUHDGLQJ FRGHV ZHUH VHOHFWHG DW UDQGRP IURP D VHW RI *ROG FRGHV XVLQJ
PAGE 93
)LJXUH &RPSDULQJ WKH DYHUDJH WUDLQLQJ ELWV IRU FRUUHFW DFTXLVLWLRQ DV D IXQFWLRQ RI WKH OHYHO RI PXOWLDFFHVV LQWHUIHUHQFH LQ D IODWIDGLQJ FKDQQHO FKLSVELW )RU HDFK WULDO WKH SURSDJDWLRQ GHOD\ IRU HDFK XVHU ZDV FKRVHQ IURP D XQLIRUP GLVWULEXWLRQ RYHU WKH LQWHUYDO >f LQGHSHQGHQW IURP WKH RWKHU XVHUV 7KH GHVLUHG XVHU ZDV UHFHLYHG DW D 615 RI G% $OVR DW WKH VWDUW RI HDFK WULDO WKH UHFHLYHG SRZHU OHYHO 3N IRU HDFK LQWHUIHULQJ XVHU ZDV VHOHFWHG IURP D ORJQRUPDO GLVWULEXWLRQ WKDW KDG D PHDQ DQG VWDQGDUG GHYLDWLRQ RI ; G% 7KDW LV 3N3? A ZKHUH e LV D *DXVVLDQ UDQGRP YDULDEOH ZLWK D PHDQ DQG VWDQGDUG GHYLDWLRQ RI $ G% %\ YDU\LQJ WKH QXPEHU RI V\VWHP XVHUV DQG WKH OHYHO RI WKH 0$, LW LV SRVVLEOH WR VWXG\ WKH SHUIRUPDQFH RI WKH K\EULG UHFHLYHU LQ FRQGLWLRQV ZKHUH WKH XQFRPSHQVDWHG ILOWHU VKRZQ LQ )LJXUH GRHVQfW SHUIRUP ZHOO EXW WKH K\EULG LQLWLDOL]DWLRQ WHFKQLTXH PD\ LPSURYH WKH DFTXLVLWLRQ SHUIRUPDQFH RYHU WKDW RIIHUHG E\ MXVW WKH PRGLILHG 006( UHFHLYHU DORQH 7KH UHVXOWV RI WKLV WHVW DUH VKRZQ LQ )LJXUH IRU D WKUHHXVHU DQG WHQXVHU '6 &'0$ V\VWHP ZKHUH WKH OHYHO RI WKH ORJQRUPDOO\ GLVWULEXWHG LQWHUIHUHQFH ;f UDQJHG IURP G% WR G% )RU ERWK WKH WKUHHXVHU DQG WHQXVHU V\VWHPV WKH K\EULG UHFHLYHU SHUIRUPV EHWWHU WKDQ WKH PRGLILHG 006( UHFHLYHU LQ WHUPV RI WKH DYHUDJH WUDLQLQJ
PAGE 94
)LJXUH 3HUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP XVLQJ WKH K\EULG DQG PRGLILHG 006( UHFHLYHUV ELWV UHTXLUHG LQ RUGHU WR DFKLHYH DFTXLVLWLRQ WR ZLWKLQ D KDOIFKLS RI 7? :KHQ WKH K\EULG UHFHLYHU ZDV XVHG LQ WKH WKUHHXVHU HQYLURQPHQW WKH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP DSSHDUV WR EH LQGHSHQGHQW RI WKH OHYHO RI WKH 0$, 7KLV UHVXOW LV GXH WR WKH OHQJWK RI WKH WUDLQLQJ F\FOH XVHG LQ WKH K\EULG UHFHLYHU EHIRUH WLPLQJ HVWLPDWHV DUH DFWXDOO\ PDGH ELWVf )RU D WKUHHXVHU V\VWHP WKH XQFRPSHQVDWHG ILOWHU LV DEOH WR PDNH UHOLDEOH WLPLQJ HVWLPDWHV SULRU WR WUDLQLQJ ELWV DQG WKHUHIRUH LW PD\ EH SRVVLEOH WR XVH D VKRUWHU WUDLQLQJ LQWHUYDO IRU WKH K\EULG UHFHLYHU LQLWLDOn L]DWLRQ F\FOH ZLWKRXW GHJUDGLQJ WKH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP 7KH OHQJWK RI WKH WUDLQLQJ F\FOH XVHG WR LQLWLDOL]H WKH K\EULG UHFHLYHU LV VHQVLWLYH WR WKH QXPEHU RI V\VWHP XVHUV DQG WKH OHYHO RI WKH 0$, +RZHYHU RYHU WKH UDQJH RI XVHUV DQG ORJQRUPDOO\ GLVWULEXWHG 0$, GHSLFWHG LQ )LJXUH ELWV RI WUDLQLQJ SURYLGH WKH K\EULG UHFHLYHU ZLWK D GHILQLWH DGYDQWDJH RYHU WKH PRGLILHG 006( UHFHLYHU LQ WHUPV RI WLPLQJ DFTXLVLWLRQ SHUIRUPDQFH
PAGE 95
$ VLWXDWLRQ LQ ZKLFK WKH K\EULG UHFHLYHU RIIHUV QR DGYDQWDJH RYHU WKH PRGLILHG 006( UHFHLYHU LV VKRZQ LQ )LJXUH 7KLV UHVXOW LV IRU WKH VDPH '6&'0$ HQYLn URQPHQW WHVWHG LQ )LJXUH KRZHYHU WKHUH DUH ILIWHHQ V\VWHPXVHUV DQG ; G% IRU DOO RI WKH ORJQRUPDOO\ GLVWULEXWHG LQWHUIHULQJ XVHUV 7KLV SORW UHDGLO\ GHPRQVWUDWHV KRZ WKH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP XVLQJ WKH K\EULG UHFHLYHU LV fPRGXODWHGf E\ WKH SHUIRUPDQFH RI WKH XQFRPSHQVDWHG ILOWHU DQG WKH OHQJWK RI LWV WUDLQLQJ F\FOH EHIRUH VZLWFKLQJ WR WKH PRGLILHG 006( UHFHLYHU VWUXFWXUH :KLOH GHFUHDVLQJ WKH OHQJWK RI WKH WUDLQLQJ F\FOH EHORZ ELWV PLJKW QRW DIIHFW WKH SHUn IRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP FHUWDLQO\ LQFUHDVLQJ WKH OHQJWK RI WKH WUDLQLQJ F\FOH WR ELWV ZRXOG EH GLVDVWURXV 6XPPDU\ %DUERVD DQG 0LOOHU >@ VWXGLHG WKH SHUIRUPDQFH RI WKH 006( UHFHLYHU LQ D IODWn IDGLQJ FKDQQHO DQG GLVFRYHUHG WKDW WKH 006( UHFHLYHU GRHVQfW ZRUN LQ D IODWIDGLQJ 5D\OHLJK FKDQQHO 7KH\ SURSRVHG D PRGLILHG UHFHLYHU VWUXFWXUH ZKLFK FRPSHQVDWHV WKH UHFHLYHG VLJQDO VXFK WKDW WKH DGDSWLYH ILOWHU GRHVQfW DWWHPSW WR WUDFN WKH SKDVH RI WKH GHVLUHG XVHUfV VLJQDO ,Q WKLV FKDSWHU WKH PRGLILHG 006( UHFHLYHU VWUXFWXUH ZDV XVHG WR VWXG\ WKH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP LQ WZR HQn YLURQPHQWV LQ ZKLFK WKH LQSXW WR WKH DGDSWLYH ILOWHU ZDV QRQVWDWLRQDU\ :KLOH WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP LV DSSOLFDEOH WR XVH ZLWK WKH PRGLILHG 006( UHFHLYHU LW KDV EHHQ REVHUYHG WKDW WKH VWDUWXS WUDQVLHQW RI WKH FKDQQHO HVWLPDWLRQ DOJRULWKP GHn JUDGHV WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWLRQ DOJRULWKP E\ LQFUHDVLQJ WKH OHQJWK RI WKH WUDLQLQJ LQWHUYDO UHTXLUHG WR DFKLHYH FRUUHFW DFTXLVLWLRQ $ K\EULG UHFHLYHU VWUXFWXUH KDV EHHQ SUHVHQWHG WR FLUFXPYHQW VRPH RI WKH VWDUWXS WUDQVLHQW UHVSRQVH RI WKH FKDQQHO HVWLPDWLRQ DOJRULWKP 7KH K\EULG UHFHLYHU XVHV D VKRUW WUDLQLQJ LQWHUYDO WR IRUP DQ HVWLPDWH RI WKH SKDVH LQGXFHG RQ WKH GHVLUHG XVHU iL DQG WKHQ XVHV WKH PRGLILHG 006( UHFHLYHU ZLWK WKH FKDQQHO HVWLPDWRU SUHORDGHG ZLWK iL WR FRQWLQXH WKH WLPLQJ DFTXLVLWLRQ F\FOH 7KURXJK VLPXODWLRQ UHVXOWV RI D '6&'0$ V\VWHP LW
PAGE 96
KDV EHHQ REVHUYHG WKDW WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP EDVHG RQ WKH K\EULG UHFHLYHU VWUXFWXUH FDQ RIIHU DQ LPSURYHPHQW RYHU WKDW RI WKH PRGLILHG 006( UHFHLYHU +RZn HYHU WKH LQLWLDOL]DWLRQ F\FOH IRU WKH K\EULG UHFHLYHU LV EDVHG RQ DQ XQFRPSHQVDWHG 006( UHFHLYHU DQG WKHUHIRUH ZLWK DQ LQDSSURSULDWH FKRLFH RI WKH OHQJWK RI WKH LQLWLDO WUDLQLQJ F\FOH LW LV SRVVLEOH IRU WKH SHUIRUPDQFH RI WKH WLPLQJ DFTXLVLWLRQ DOJRULWKP IRU WKH K\EULG UHFHLYHU WR EH ZRUVH WKDQ WKDW RIIHUHG E\ WKH PRGLILHG 006( UHFHLYHU DORQH
PAGE 97
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f DUH JXDUDQWHHG WR RFFXU ZLWKLQ WKH REVHUYDWLRQ ZLQGRZ 7KLV LQFUHDVH LQ WKH REVHUYDWLRQ ZLQGRZ DOVR DOORZV IRU PRUH WKDQ RQH XVHU WR EH LQ WKH FRGH DFTXLVLWLRQ PRGH DW WKH VDPH WLPH 8SGDWHG 6\VWHP 0RGHO ,Q WKLV VHFWLRQ WKH V\VWHP PRGHO IRU DQ REVHUYDWLRQ ZLQGRZ RI WZR ELW LQWHUYDOV ZLOO EH SUHVHQWHG 7KLV PRGHO LV MXVW D VSHFLDO H[WHQVLRQ RI WKDW SUHVHQWHG LQ 6HFWLRQ KRZHYHU D QRWDWLRQ VLPLODU WR WKDW SUHVHQWHG E\ 0DGKRZ >@ ZLOO EH XVHG $V ZLWK WKH SUHYLRXV V\VWHP PRGHO WKH UHFHLYHU FRQYHUWV UWf WR D EDVHEDQG VLJQDO DQG SDVVHV 5Wf WKURXJK D ILOWHU PDWFKHG WR WKH FKLS SXOVH VKDSH DQG WKH RXWSXW RI WKH ILOWHU LV VDPSOHG DW WKH FKLS UDWH $V EHIRUH ZH OHW WKH TXDQWLW\ UPAQ UHSUHVHQW WKH QWK FKLS VDPSOHG RXWSXW GXULQJ WKH UHFHLYHUfV PWK ELW LQWHUYDO P7E?Q?Of7F 7UQnQ a ?a3>7 I 5^Wf GW f & f P7EQ7F
PAGE 98
'XULQJ HDFK ELW LQWHUYDO D WRWDO RI 1 FKLS VDPSOHV DUH JDWKHUHG DQG JURXSHG ZLWK WKH 1 FKLS VDPSOHV IURP WKH SUHYLRXV ELW LQWHUYDO VXFK WKDW UPf ^;PfÂ§f ,nPfÂ§f f f f 8QfÂ§9fÂ§OL 8Q! 7P?L f f f L AP1fÂ§ Of f DQG LW LV WKLV YHFWRU TXDQWLW\ WKDW ZLOO EH XVHG DV WKH LQSXW WR WKH DGDSWLYH UHFHLYHU $V ZLWK WKH SUHYLRXV V\VWHP PRGHO UPf Â W L ZKHUH NPf LV WKH FRQWULEXWLRQ WR WKH UHFHLYHG YHFWRU IURP WKH NWK XVHU GXULQJ WKH PWK UHFHLYHU ELW LQWHUYDO /HW WKH GNAP UHSUHVHQW WKH GDWD ELW IRU WKH NWK XVHU WKDW IDOOV FRPSOHWHO\ LQ WKH WZR ELW REVHUYDWLRQ LQWHUYDO GXULQJ WKH UHFHLYHUfV PWK ELW LQWHUYDO 8QOLNH EHIRUH ZH QRZ GHILQH WKH SURSDJDWLRQ GHOD\ IRU WKH $UWK XVHU WR EH UN >@@ VXFK WKDW UN fÂ§ SN7F 6N7F ZLWK SN ^ 1 fÂ§ ` DQG N RU UN 17& /HW DN UHSUHVHQW D YHFWRU RI OHQJWK 1 FRQVLVWLQJ RI WKH 1 HOHPHQWV RI WKH VSUHDGLQJ VHTXHQFH IRU WKH NWK XVHU IROORZHG E\ 1 ]HURV AIFf &H f f f f &N1fÂ§f M f f f f f /HW 7An UHSUHVHQW WKH QWK OHIW DF\FOLF VKLIW RI D YHFWRU ZKLOH UHSUHVHQWV WKH QWK ULJKW DF\FOLF VKLIW RI D YHFWRU %RWK IXQFWLRQV RSHUDWH RQ DQG SURGXFH YHFWRUV ZLWK 1 HOHPHQWV 8VLQJ D JHQHUDO YHFWRU [ UUeL [QLfW WKHVH IXQFWLRQV DUH GHILQHG EHORZ 7A[f 7A[f [ f 7 O?[f [L[[1Lf7 f U`[f ;1 f Â‘ f f7 f 7`[f R ; ;L ;1f7 f IFUDf H[SMNf 1P & L:[O f
PAGE 99
7IL1Bf[f 2 2 [f7 f 7Â `[f 7A[f 2 f7 9 Q 1 f 8VLQJ WKH DERYH GHILQLWLRQV ZH FDQ QRZ GHILQH WKH NWK XVHUfV FRQWULEXWLRQ WR WKH UHFHLYHG YHFWRU DV IFAf fÂ§ GNPfÂ§N GNPUPfUPf@ \ YrO.fOfU Yr.fU YÂYÂf7f IF L )O Y n f W, S (>UPfGLP@ Y" H[SM}Lf f 8SGDWHG 7LPLQJ (VWLPDWLRQ $OJRULWKP :H DUH LQWHUHVWHG LQ GHYHORSLQJ D QHZ WLPLQJ HVWLPDWRU IRU WKH SURSDJDWLRQ GHOD\ RI WKH GHVLUHG XVHU QRZ WKDW ZH KDYH LQFUHDVHG WKH OHQJWK RI WKH ILOWHU :H ZLOO XVH WKH VDPH SURFHGXUH DV EHIRUH 7KDW LV WKH ZHLJKWV RI DQ 006( ILOWHU IRU D VLQJOH XVHU ZLOO EH XVHG WR IRUP WKH WLPLQJ HVWLPDWRU 7KHUHIRUH ZH QHHG WR FDOFXODWH WKH ZHLJKWV RI WKH RSWLPDO ILOWHU IRU D VLQJOHXVHU RSHUDWLQJ LQ DQ $:*1 FKDQQHO ,Q
PAGE 100
RUGHU WR VWXG\ WKH ZHLJKWV RI WKH :LHQHU+RSI ILOWHU ZH FRQVLGHU WKH FDVH ZKHUH ZH DSSO\ WKH HTXLYDOHQW SKDVHV\QFKURQRXV PRGHO IRU WKH HTXDWLRQV OLVWHG DERYH $V LQ WKH RWKHU V\VWHP PRGHO WKLV PHDQV WKH ILOWHU ZHLJKWV EHFRPH UHDO DQG WKH FRPSOH[ H[SRQHQWLDO WHUP LV UHPRYHG IURP WKH VWHHULQJ YHFWRU H[SUHVVLRQ ,Q WKLV FDVH WKH RSWLPDO ZHLJKW YHFWRU LV IRXQG XVLQJ WKH :LHQHU+RSI HTXDWLRQ ZLWK 5ZRSW Yr ZKHUH ZH OHW ZRSW DYI FY^ 7KH FRHIILFLHQWV RI WKH RSWLPDO ZHLJKW YHFWRU D E DQG F DUH JLYHQ E\ WKH VROXWLRQ WR WKH IROORZLQJ HTXDWLRQ A O_YUQ YI7Y" __9[ YI7Y" YU7Y` Yf7YO ,, D E F f YUUYM Y"f7Z LLYLL 8VLQJ WKH DERYH HTXDWLRQ ZH VHH WKDW WKH D DQG F FRHIILFLHQWV FDQ RQO\ EH ]HUR ZKHQ ERWK YI DQG Y^ DUH RUWKRJRQDO WR Yr 7KLV FRQGLWLRQ RFFXUV ZKHQ ÂL +RZHYHU LQ WKH JHQHUDO FDVH ZKHQ A WKH D DQG F FRHIILFLHQWV ZLOO EH PXFK VPDOOHU LQ PDJQLWXGH WKDQ WKH E FRHIILFLHQW VLQFH Y/M DQG YM DUH RQO\ QRQ]HUR LQ RQH FKLS SRVLWLRQ ZKHUH Yr LV QRQ]HUR DOVR %DVHG RQ WKLV UHVXOW ZH ZLOO LJQRUH WKH FRQWULEXWLRQV RI WKH YA DQG Y` YHFWRUV RQ WKH :LHQHU+RSI ZHLJKWV ZKHQ ZH GHYHORS WKH WLPLQJ HVWLPDWRU 8VLQJ WKH ZHLJKWV RI WKH DGDSWLYH ILOWHU ZH HVWLPDWH WKH SURSDJDWLRQ GHOD\ IRU WKH GHVLUHG XVHU E\ ILQGLQJ WKH VHW 4Sf WKDW PLQLPL]H __ZPf YH[SMf__ $V LQ WKH FDVH ZKHUH WKH UHFHLYHU RQO\ REVHUYHV RQH ELW RI LQIRUPDWLRQ WKH HVWLPDWH RI 7? LV JLYHQ E\ WKH VHW ^S Â` ZKLFK PD[LPL]H WKH FRVW IXQFWLRQ ISf JLYHQ EHORZ ISf ,.fUZPf ,OLL20 f 0D[LPL]LQJ S f ZLWK UHVSHFW WR SURGXFHV D TXDGUDWLF HTXDWLRQ ZKRVH FRHIILn FLHQWV GHSHQG RQ WKH YDOXH RIS 7KHUHIRUH IRU HDFK SRVVLEOH YDOXH RI S ^ 1fÂ§ ` ZH PXVW ILQG WKH URRWV RI WKH TXDGUDWLF HTXDWLRQ 'SfS 'LSfS '^Sf ZKHUH
PAGE 101
WKH FRHIILFLHQWV DUH JLYHQ E\ WKH IROORZLQJ VHW RI HTXDWLRQV 'Sf 1$Sf &$Sf '?^Sf 1$Sf L&$Sf 'Sf fÂ§ &$Sf $LSff & 1 DI7AD2 $Sf _ZPf7ADf_ $?Sf 5H 7AfDLffUZPfZLPf7AfDf $Sf $Sf _ZIURf7MfDf_ $[Sf $^Sf Df Ef Ff Gf Hf If Jf 7KH SURSDJDWLRQ GHOD\ HVWLPDWLRQ DOJRULWKP IRU WKH GHVLUHG XVHU LV VXPPDUL]HG DV IROORZV /HW 7 UHSUHVHQW WKH VHW RI FRVWV FRUUHVSRQGLQJ WR WKH VHW RI FDQGLGDWH HVWLPDWHV RI Sf GHQRWHG E\ 8 :H FDQ ILQG WKH PD[LPXP RI ISf DV IROORZV 6WHS /HW 7 ^f f f I1 f` 6WHS /HW 8 ^ L9` 6WHS )RU S 1 fÂ§ GR WKH IROORZLQJ Df &RPSXWH WKH FRHIILFLHQWV 'Sf '?Sf 'Sf Ef 6ROYH IRU WKH WZR URRWV RI WKH TXDGUDWLF HTXDWLRQ S? DQG S Ff ,I SL IRU L ( ^` DGG WKH FRVW I^SSLf WR WKH VHW 7 DQG DGG S SLf WR WKH VHW 8 6WHS /HW Â GHQRWH WKH NWK HOHPHQW RI WKH VHW 8 7KHQ I = ZKHUH N PD[M M 6WHS 7KH HVWLPDWHV RI S[ DQG A DUH JLYHQ E\ SLUDf >n&O DQG 0Pf fÂ§ I a /Af ZKHUH URXQGV [ WR WKH QHDUHVW LQWHJHU WRZDUGV ]HUR 7KH XSGDWHG V\VWHP PRGHOWLPLQJ HVWLPDWLRQ DOJRULWKP DUH EULHIO\ WHVWHG WR GHPRQVWUDWH ZKDW DIIHFWV LI DQ\ DUH REVHUYHG RQ WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWH ,Q &KDSWHU VHYHUDO SORWV RI WKH FRVW IXQFWLRQ IS6f DUH VKRZQ WR
PAGE 102
)LJXUH 3ORW RI WKH FRVW IXQFWLRQ S Âf IRU D ILIWHHQXVHU V\VWHP ZKHQ WKH UHFHLYHU REVHUYHV WZR ELWLQWHUYDOV GHPRQVWUDWH KRZ WKH WLPLQJ HVWLPDWH LV IRUPHG XQGHU WZR JLYHQ VFHQDULRV ZKHQ WKH :LHQHU+RSI ZHLJKWV DUH XVHG ,Q )LJXUH WKH FRVW IXQFWLRQ LV VKRZQ IRU DQ HQYLn URQPHQW ZKHUH WKHUH DUH ILIWHHQXVHUV DQG 615 G% 7? fÂ§ 7& DQG DOO RI WKH LQWHUIHULQJ XVHUV DUH UHFHLYHG DW D SRZHU OHYHO G% DERYH WKH GHVLUHG XVHU 8VLQJ WKH WLPLQJ HVWLPDWRU RI &KDSWHU WKH UHVXOWLQJ WLPLQJ HVWLPDWH LV 7? 7& &OHDUO\ LQ WKLV HQYLURQPHQW WKH V\VWHP LV RYHUORDGHG WR WKH SRLQW ZKHUH D UHOLDEOH HVWLPDWH LV LQIHDVLEOH 8VLQJ WKH VDPH HQYLURQPHQW ZH SORW WKH FRVW IXQFWLRQ IRU WKH FDVH ZKHQ WKH UHFHLYHU REVHUYHV WZRELW LQWHUYDOV LQ )LJXUH ,Q WKLV FDVH WKH WLPLQJ HVWLPDWH LV IL 7& ZKLFK LV D UHOLDEOH HVWLPDWH :H FDQ MXVWLI\ WKLV LPSURYHG SHUIRUn PDQFH WR VRPH H[WHQW E\ FRQVLGHULQJ KRZ WKH WLPLQJ HVWLPDWH LV IRUPHG IRU WKH WZR GLIIHUHQW V\VWHP PRGHOV :KHQ WKH UHFHLYHU REVHUYHV RQO\ RQHELW LQWHUYDO WKH ILOWHU ZHLJKWV DUH SURFHVVHG ZLWK D F\FOLFVKLIW RI WKH VSUHDGLQJ FRGH RI WKH GHVLUHG XVHU ,Q WKH FDVH ZKHQ WKH UHFHLYHU REVHUYHV WZRELW LQWHUYDOV DQ DF\FOLFVKLIWLQJ RSHUDWLRQ LV XVHG 7KH GLIIHUHQFH EHWZHHQ WKH WZR RSHUDWLRQV UHVXOWV LQ WKH VLGHOREHV WKRVH YDOXHV RI U QRW QHDU 7Lf RI WKH FRVW IXQFWLRQ EHLQJ PRGXODWHG GLIIHUHQWO\ 7KHUHIRUH LW
PAGE 103
)LJXUH &RPSDULQJ WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWRU ZKHQ WKH UHFHLYHU REVHUYHV ELW LQWHUYDOV rf YHUVXV REVHUYLQJ ELW LQWHUYDO VROLG OLQHf DV D IXQFWLRQ RI WKH 0$, LV HDV\ WR FRQFHLYH WKDW LW LV SRVVLEOH EXW QRW JXDUDQWHHGf WKDW WKH DPSOLWXGH RI WKH SHDN VLGHOREH LQ )LJXUH LV UHGXFHG LQ DPSOLWXGH VXFK WKDW WKH FRUUHFW HVWLPDWH LV IRUPHG $V D IXUWKHU WHVW RI WKH XSGDWHG V\VWHP PRGHOWLPLQJ HVWLPDWLRQ DOJRULWKP ZH FRPSDUH WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWH WR WKRVH UHVXOWV VKRZQ LQ )LJXUH 7KH QHDUIDU SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWH ZDV REVHUYHG IRU D PXOWLXVHU HQYLn URQPHQW ZLWK D 615 G% DQG DOO LQWHUIHULQJ XVHUV UHFHLYHG DW SRZHU OHYHOV WKDW ZHUH ORJQRUPDOO\ GLVWULEXWHG 7KH SRZHU RI HDFK LQWHUIHULQJ XVHU UHODWLYH WR WKH GHn VLUHG XVHU ZDV 3N3L A ZKHUH I ZDV D *DXVVLDQ UDQGRP YDULDEOH ZLWK PHDQ DQG VWDQGDUG GHYLDWLRQ RI $ G% 7KH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWH ZKHQ WKH ILOWHU ZHLJKWV DUH DGDSWHG XVLQJ WKH 5/6 DOJRULWKP LV VKRZQ LQ )LJXUH IRU ERWK V\VWHP PRGHOV $ VHHQ LQ WKLV ILJXUH WKHUH LV RQO\ D VOLJKW GLIIHUHQFH EHWZHHQ WKH SHUIRUPDQFH RI WKH WZR V\VWHPV RYHU D ZLGH UDQJH RI XVHUV DQG OHYHO RI 0$,
PAGE 104
6XPPDU\ ,Q WKLV FKDSWHU ZH KDYH PRGLILHG WKH V\VWHP PRGHO VR WKDW WKH UHFHLYHU REVHUYHV WZR ELW LQWHUYDOV RI GDWD ,I RQO\ RQH ELW LQWHUYDO RI GDWD LV REVHUYHG WKH WUDLQLQJ VHTXHQFH LV UHVWULFWHG WR EHLQJ DQ DOO RQHV VHTXHQFH WR DYRLG ELW WUDQVLWLRQV LQ WKH GHVLUHG XVHUfV UHFHLYHG VLJQDO %\ OHWWLQJ WKH UHFHLYHU REVHUYH WZR ELW LQWHUYDOV RI GDWD WKH GHVLUHG XVHU FDQ QRZ VHQG D NQRZQ GDWD VHTXHQFH GXULQJ WUDLQLQJ ,Q DGGLWLRQ LW LV SRVVLEOH IRU PRUH WKDQ RQH V\VWHP XVHU WR EH LQ WKH FRGH DFTXLVLWLRQ PRGH DW WKH VDPH WLPH $QRWKHU SRVVLEOH DGYDQWDJH RI XVLQJ D ORQJHU REVHUYDWLRQ ZLQGRZ LV GXH WR ZKDW WKH UHFHLYHU FDQ GR DIWHU DFTXLULQJ FRGH DFTXLVLWLRQ &RQVLGHU ZKDW KDSSHQV ZKHQ WKH UHFHLYHU RQO\ REVHUYHV RQH ELW LQWHUYDO RI LQIRUPDWLRQ $IWHU FRGH DFTXLVLWLRQ WKH UHFHLYHU PXVW XSGDWH LWV V\VWHP WLPLQJ UHODWLYH WR WKH UHFHLYHG VLJQDO LQ RUGHU WR PD[LPL]H WKH V\VWHP FDSDFLW\ 7KLV PHDQV WKDW WKH ILOWHU ZLOO KDYH WR HQWHU DQRWKHU SRVVLEO\ ORQJ WUDLQLQJ F\FOH WR DGMXVW WKH ZHLJKWV VR WKDW GDWD GHWHFWLRQ FDQ EHJLQ ,Q WKH FDVH ZKHQ WZR ELW LQWHUYDOV DUH REVHUYHG E\ WKH UHFHLYHU WKH OHQJWK RI WKH VHFRQG WUDLQLQJ F\FOH PD\ EH UHGXFHG 2QFH WKH UHFHLYHU KDV DQ HVWLPDWH RI WKH GHVLUHG XVHUfV SURSDJDWLRQ GHOD\ WKH UHFHLYHU FRXOG XSGDWH LWV WLPLQJ DQG ZHLJKWV UHODWLYH WR WKH GHVLUHG XVHUf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
PAGE 105
&+$37(5 6800$5< $1' )8785( :25. 6XPPDU\ 7KH QHDUIDU SUREOHP LV DQ LPSHGLPHQW LQ WKH GHYHORSPHQW RI DV\QFKURQRXV '6 &'0$ FRPPXQLFDWLRQ V\VWHPV 7KH QHDUIDU SUREOHP H[LVWV ZKHQ YDULRXV V\VWHP XVHUV DUH UHFHLYHG DW YHU\ GLVVLPLODU SRZHU OHYHOV &RQYHQWLRQDO GHWHFWRUV DUH VHQVLWLYH WR WKH QHDUIDU SUREOHP LQ WKDW LW LV SRVVLEOH IRU WKH PXOWLDFFHVV LQWHUIHUHQFH WR FRPn SOHWHO\ GHVWUR\ WKH UHFHLYHUfV DELOLW\ WR GHWHFW D ZHDN XVHU 6HYHUDO QHDUIDU UHFHLYHUV H[LVW EXW WKHLU FRPSOH[LW\ DQG WKH DPRXQW RI UHTXLUHG VLGHLQIRUPDWLRQ PDNH WKHP LPSUDFWLFDO WR LPSOHPHQW 7KRVH UHVHDUFKHUV WKDW VWXG\ WKH SUREOHP RI GHWHFWLRQ LQ DV\QFKURQRXV '6&'0$ HQYLURQPHQWV XVXDOO\ DVVXPH WKDW WKH UHFHLYHU RSHUDWHV V\Qn FKURQRXV WR WKH GHVLUHG XVHU 7UDGLWLRQDOO\ WKH ELWHUURUUDWH LV XVHG WR GHWHUPLQH WKH RYHUDOO V\VWHP FDSDFLW\ +RZHYHU HDUOLHU ZRUN KDV VKRZQ WKDW WKH DELOLW\ WR DFKLHYH FRGH V\QFKURQL]DWLRQ DQG QRW WKH ELWHUURUUDWH ZLOO GHWHUPLQH WKH RYHUDOO V\VWHP FDSDFLW\ 7KLV PRWLYDWHV WKH QHHG IRU QHDUIDU UHVLVWDQW V\QFKURQL]DWLRQ PHWKRGV ,Q &KDSWHU D UHYLHZ RI VHYHUDO SURSRVHG WLPLQJ HVWLPDWLRQ DOJRULWKPV ZDV SUHVHQWHG ,W ZDV VKRZQ WKDW HDFK RI WKHVH WHFKQLTXHV ZHUH RI UDWKHU KLJK FRPSOH[LW\ 7KH 006( UHFHLYHU LV ZHOONQRZQ WR EH QHDUIDU UHVLVWDQW DQG LV UHFHLYLQJ FRQVLGHUDEOH DWWHQWLRQ LQ WKH OLWHUDWXUH ,Q &KDSWHU D ORZFRPSOH[LW\ WLPLQJ DFTXLVLWLRQ PHWKRG EDVHG RQ SURFHVVLQJ WKH ZHLJKWV RI WKH 006( UHFHLYHU ZDV SUHVHQWHG 7KH WLPLQJ DFTXLVLWLRQ DOJRULWKP ZDV GHYHORSHG DVVXPLQJ D VLQJOHXVHU V\VWHP ZKHUH WKH GHVLUHG XVHU PXVW WUDQVPLW DQ DOO RQHV WUDLQLQJ VHTXHQFH ,Q WKLV FDVH WKH 006( UHFHLYHU DGDSWV LQ WKH PHDQ WR D VXP RI WZR WLPHVKLIWHG YHUVLRQV RI WKH VSUHDGLQJ FRGH RI WKH GHVLUHG XVHU 7KH FRPSOH[LW\ RI WKH WLPLQJ HVWLPDWLRQ DOJRULWKP ZDV VOLJKWO\ KLJKHU WKDQ WKDW RI WKH FRQYHQWLRQDO FRUUHODWRU E\ WKH FRPSOH[LW\ RI UXQQLQJ WKH DGDSWLYH
PAGE 106
ILOWHUf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n WLRQ RI WLPH 7KLV PRGHO ZDV XVHG DV D EDVLV IRU FKDUDFWHUL]LQJ WKH WLPLQJ HVWLPDWRU XVLQJ WZR GLIIHUHQW DQDO\WLFDO WHFKQLTXHV 8QIRUWXQDWHO\ ERWK WHFKQLTXHV SURGXFHG UHVXOWV WKDW DUH RI OLPLWHG XVH 7KLV IDFW VWHPV IURP VHYHUDO DVVXPSWLRQV XVHG LQ WKH GHULYDWLRQ RI WKH DQDO\WLFDO PHWKRGV DV WKH VWDWLVWLFDO PRGHO IRU WKH WUDQVLHQW /06 ILOWHU ZHLJKW VWDWLVWLFV KDV EHHQ VKRZQ WR SURYLGH UHDVRQDEOH SHUIRUPDQFH EHWZHHQ WKH VLPXODWHG DQG DQDO\WLFDO UHVXOWV (YHQ WKRXJK WKH WLPLQJ HVWLPDWLRQ DOJRULWKP ZDV GHYHORSHG IRU D VLQJOHXVHU V\VWHP LW ZDV VKRZQ WR SHUIRUP ZHOO LQ PXOWLXVHU HQYLn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f +RZHYHU WKH SKDVHFRPSHQVDWLRQ DOJRULWKP GHOD\V WKH EXLOGXS RI HQHUJ\ LQ WKH ZHLJKWV RI WKH DGDSWLYH ILOWHU 7KLV DIIHFWV WKH SHUIRUPDQFH RI WKH WLPLQJ HVWLPDWLRQ DOJRULWKP E\ LQFUHDVLQJ WKH OHQJWK RI WKH WUDLQLQJ F\FOH UHTXLUHG WR DFKLHYH DFTXLVLWLRQ $ K\EULG
PAGE 107
IRUP RI WKH PRGLILHG 006( UHFHLYHU ZDV VKRZQ WR RIIVHW VRPH RI WKLV SHUIRUPDQFH GHJUDGDWLRQ )LQDOO\ LQ &KDSWHU D PRGLILFDWLRQ WR WKH V\VWHP PRGHO DQG WLPLQJ HVWLPDWLRQ DOJRULWKP ZDV SUHVHQWHG VXFK WKDW WKH UHTXLUHG WUDLQLQJ VHTXHQFH IRU WKH 006( UHFHLYHU GRHVQfW KDYH WR EH DOO RQHV 7KH PRGLILHG WLPLQJ HVWLPDWRU ZDV FRPSDUHG WR WKH RULJLQDO WLPLQJ HVWLPDWRU XQGHU D PXOWLXVHU QHDUIDU HQYLURQPHQW 7KHUH ZDV QR DSSUHFLDEOH GLIIHUHQFH REVHUYHG EHWZHHQ WKH SHUIRUPDQFH RI WKH WZR WLPLQJ HVWLPDWRUV 7KH DGYDQWDJH RI WKH PRGLILHG WLPLQJ HVWLPDWRU LV WKDW PRUH WKDQ RQH V\VWHPXVHU FDQ EH LQ WKH DFTXLVLWLRQ PRGH DW WKH VDPH WLPH &RQWULEXWLRQV 7KH PDLQ FRQWULEXWLRQV PDGH LQ WKLV GLVVHUWDWLRQ DUH f 'HYHORSPHQW RI D ORZ FRPSOH[LW\ VLQJOHXVHU QHDUIDU UHVLVWDQW WLPLQJ HVWLPDWRU EDVHG RQ WKH ZHOONQRZQ 006( UHFHLYHU f &OHDUO\ GHPRQVWUDWLQJ ZK\ V\QFKURQL]LQJ WKH 006( UHFHLYHU LV LPSRUWDQW f 6WXG\LQJ WKH SHUIRUPDQFH RI WKH HVWLPDWRU LQ WZR QRQVWDWLRQDU\ FKDQQHOV DQG SURSRVLQJ D K\EULG WHFKQLTXH EDVHG RQ WKH PRGLILHG 006( UHFHLYHU f 0RGLI\LQJ WKH WLPLQJ HVWLPDWRU VXFK WKDW PRUH WKDQ RQH V\VWHPXVHU FDQ EH LQ WKH DFTXLVLWLRQ PRGH DW WKH VDPH WLPH )XWXUH :RUN 7KHUH DUH VHYHUDO DUHDV RI IXWXUH UHVHDUFK UHODWHG WR WLPLQJ DFTXLVLWLRQ DQG V\Qn FKURQL]DWLRQ LQ '6&'0$ FRPPXQLFDWLRQ V\VWHPV ,Q WKLV GLVVHUWDWLRQ WKH ZHLJKWV RI DQ DGDSWLYH ILOWHU KDYH EHHQ SURFHVVHG WR IRUP D FRDUVH HVWLPDWH RI WKH SURSDJDWLRQ GHOD\ RI WKH GHVLUHG XVHU ,Q &KDSWHU LW ZDV VKRZQ WKDW WKH 006( UHFHLYHU PXVW EH V\QFKURQL]HG ZLWK WKH GHVLUHG XVHU LQ RUGHU WR PD[LPL]H WKH V\VWHP FDSDFLW\ $In WHU WKH UHFHLYHU DGMXVWV LWV WLPLQJ EDVHG RQ WKH WLPLQJ HVWLPDWH D VHSDUDWH WUDFNLQJ
PAGE 108
PRGH PXVW EH XVHG V\QFKURQL]H WR WKH GHVLUHG XVHUfV VLJQDO DQG PDLQWDLQ WKLV V\Qn FKURQL]DWLRQ ZKLOH WKH ILOWHU LV XVHG IRU GDWD GHWHFWLRQ :KLOH WKH WLPLQJ HVWLPDWLRQ DOJRULWKP SURYLGHV FRDUVH DFTXLVLWLRQ LWV XVHIXOQHVV LQ D WUDFNLQJ PRGH LV SRVVLEO\ OLPLWHG &RQVLGHU WKH ZHLJKWV RI WKH DGDSWLYH ILOWHU ZKHQ WKH UHFHLYHU LV LQ WKH GDWD GHWHFWLRQ PRGH $ FKDQJH LQ WKH UHODWLYH WLPLQJ EHWZHHQ WKH UHFHLYHU DQG WKH GHVLUHG XVHU ZLOO PDUJLQDOO\ DIIHFW IRU VPDOO WLPLQJ HUURUVf WKH ILOWHU RXWSXW HUURU VLJQDO 7KLV PHDQV WKDW WKH HIIHFW RI WKH WLPLQJ FKDQJH ZRQfW DSSHDU LQ WKH ILOWHU ZHLJKWV IRU PDQ\ ELW LQWHUYDOV :KHQ WKH WLPH VKLIW LV GHWHFWHG E\ WKH WLPLQJ HVWLPDWLRQ DOJRULWKP PRVW OLNHO\ WKH WLPLQJ UHODWLRQVKLS EHWZHHQ WKH UHFHLYHU DQG WKH GHVLUHG XVHU KDV DOUHDG\ FKDQJHG 7R WKH EHVW RI WKH DXWKRUfV NQRZOHGJH WKH SUREOHP RI WUDFNLQJ LQ '6&'0$ FRPPXQLFDWLRQ V\VWHPV KDVQfW EHHQ VXIILFLHQWO\ DGGUHVVHG $QRWKHU DUHD WKDW LV UHFHLYLQJ DWWHQWLRQ LQ WKH OLWHUDWXUH LV WKDW RI RYHUOD\ FRPPXn QLFDWLRQ V\VWHPV ,Q DQ RYHUOD\ V\VWHP WKH '6&'0$ V\VWHP VKDUHV EDQGZLGWK ZLWK D QDUURZEDQG FRPPXQLFDWLRQ V\VWHP :KLOH WKH LVVXH RI FRGH DFTXLVLWLRQ LQ RYHUn OD\ VFHQDULRV KDV EHHQ VWXGLHG >@ WKH HPSKDVLV RI RYHUOD\ UHVHDUFK KDV EHHQ WKH QDWXUH RI KRZ WKH QDUURZEDQG V\VWHP DIIHFWV WKH '6&'0$ V\VWHP SHUIRUPDQFH 5HFHQW ZRUN KDV VWXGLHG WKH HIIHFW RI WKH '6&'0$ V\VWHP RQ WKH SHUIRUPDQFH RI WKH QDUURZEDQG V\VWHP >@ ,W KDV EHHQ VKRZQ WKDW WKH '6&'0$ V\VWHP PXVW SODFH QXOOV LQ LWV 36' ZKHUH WKH QDUURZEDQG V\VWHP H[LVWV LQ RUGHU IRU WKH QDUURZEDQG V\VWHP WR FRH[LVW ZLWK WKH '6&'0$ V\VWHP 7KH QXOOV LQ 36' RI WKH '6&'0$ V\VWHP DUH FUHDWHG E\ ILOWHULQJ WKH VSUHDGLQJ VHTXHQFH XVHG E\ HDFK XVHU $W WKLV WLPH LW LV XQFOHDU KRZ WKLV ILOWHULQJ RSHUDWLRQ ZLOO DIIHFW WKH SUREOHP RI WLPLQJ DFTXLVLWLRQ DQG WUDFNLQJ $ ILQDO DUHD IRU IXWXUH UHVHDUFK GHDOV ZLWK PXOWLSDWK IUHTXHQF\VHOHFWLYH IDGLQJ FKDQQHOV $ UHFHQW ZRUN E\ 0LOOHU HW DO >@ VWXGLHV WKH SHUIRUPDQFH RI VHYHUDO GLIIHUHQW 006( UHFHLYHU DUFKLWHFWXUHV LQ IUHTXHQF\ VHOHFWLYH IDGLQJ FKDQQHOV 2QH SURSRVHG DUFKLWHFWXUH LV WR FRPELQH WKH RXWSXWV RI D EDQN RI ILOWHUV HLWKHU WKH PRGLILHG
PAGE 109
006( >@ RU WKH 02( >@f RQH IRU HDFK UHFHLYHG SDWK RI WKH GHVLUHG XVHUfV VLJQDO 7KH SHUIRUPDQFH RI WKLV DUFKLWHFWXUH LV FRPSDUHG WR D VLQJOH 006( UHFHLYHU LQ ZKLFK WKH VWHHULQJ YHFWRU LV IRUPHG ZLWK D SULRUL NQRZOHGJH RI WKH IDGLQJ SDUDPHWHUV DQG GHOD\V RI HDFK SDWK RI WKH GHVLUHG XVHUfV VLJQDO ,W LV VKRZQ WKDW WKH VLQJOH 006( UHFHLYHU RIIHUV D VLJQLILFDQW LPSURYHPHQW LQ %(5 SHUIRUPDQFH RYHU WKH PXOWLSOH ILOWHU DUFKLWHFWXUH LQ D IDVW IDGLQJ HQYLURQPHQW 7KHUHIRUH ZLWKRXW NQRZOHGJH RI WKH IDGLQJ SDUDPHWHUV IRU HDFK SDWKf D VXEVWDQWLDO GHFUHDVH LQ WKH V\VWHP FDSDFLW\ ZLOO RFFXU 7KLV UHVXOW PRWLYDWHV WKH QHHG IRU QHDUIDU UHVLVWDQW SDUDPHWHU HVWLPDWLRQ DOJRULWKPV IRU IDVW IDGLQJ PXOWLSDWK FKDQQHOV
PAGE 110
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s GDWD VHTXHQFH ZH PXVW XSGDWH RXU V\VWHP PRGHO 7KH LQSXW WR WKH UHFHLYHU GXULQJ WKH PWK V\PERO LQWHUYDO PD\ QRZ EH H[SUHVVHG DV UPf cLPf > Âf7A!Ff LWeFf@ H[SMcf IPf $Of ZKHUH I Pf DFFRXQWV IRU WKH LQWHUV\PERO LQWHUIHUHQFH ,6,f IURP WKH GHVLUHG XVHU WKH 0$, LQWHUIHUHQFH DQG WKH FRPSOH[ $:*1 DV VKRZQ EHORZ IPf GLP f IO Lf7ISL!FLf LfOInnfFOfO H[S MmLf n?O\ IFPf H[3 8IDf 1U $f N OLWf &RQYHUW WR EDVHEDQG FKLS PDWFKHG ILOWHU UPf 006( ILOWHU Z + U PfZ VJQ>5Hf@ G?Pf HPf G$Pf )LJXUH $O 7KH 006( UHFHLYHU IRU WKH *DXVVLDQ DSSUR[LPDWLRQ
PAGE 111
,Q WKH DERYH HTXDWLRQV WKH 7Â f DQG 7 f RSHUDWRUV DUH WKH DF\FOLF OHIW DQG ULJKW VKLIW RSHUDWRUV RI DQ L9HOHPHQW YHFWRU DQG IFUDf LV GHILQHG DV VKRZQ LQ HTXDWLRQ f 7KH VWHHULQJ YHFWRU IRU WKH 006( UHFHLYHU LV S (>GLPfUPf@ ZKLOH WKH DXWRn FRUUHODWLRQ PDWUL[ LV 5 (>UPfU+Pf@ 8VLQJ WKH DERYH HTXDWLRQV WKHVH TXDQWLWLHV FDQ EH H[SUHVVHG DV VKRZQ LQ WKH IROORZLQJ HTXDWLRQV Lf7JALGf L7JFfH[S2f@ $f A A S 5 78 ODNL3NNfANL3NNf DrSMIF!LIFfDESIFIFf@ FU, N L ÂQ fÂ§ SSZ (>UUDfU+Pf@ $f SS+ 5 Â‘mSLLf 7KH ILOWHU RXWSXW \Pf DW WKH RSWLPDO ZHLJKWV ZRSW 5 [S LV JLYHQ E\ \^Pf ZWUPf GPfZISWS ZISWUPf $f :H PRGHO WKH WHUP ZAWUPf DV D FRPSOH[ *DXVVLDQ UDQGRP YDULDEOH ZLWK D PHDQ RI ]HUR DQG D YDULDQFH RI D :H ILQG WKH YDULDQFH E\ HYDOXDWLQJ D (>ZISWLPfL+PfZRSW@ ZW(>IPfLPf@ZRSW ZAW >5 33+@ ZRSW S+US 3ASf $f 5HFRJQL]LQJ WKDW PLQ S+5BS ZH ILQG WKDW t fÂ§ PLQfQ $f DQG GLr1fZISWS GLPf PLQf $f
PAGE 112
7KH ILOWHU IRUPV LWV GHFLVLRQ E\ FRPSDULQJ 5H>\Pf@ WR D WKUHVKROG RI ]HUR 6LQFH ZH KDYH PRGHOHG WKH RXWSXW RI WKH ILOWHU DV D FRPSOH[ *DXVVLDQ UDQGRP YDULDEOH ZH VHH WKDW 5H>\Pf@ LV D *DXVVLDQ UDQGRP YDULDEOH ZLWK D PHDQ RI G?^Pf fÂ§ PLQf DQG D YDULDQFH RI fÂ§ PLQfPPÂ‘ $VVXPLQJ WKDW G?Pf LV HTXDOO\ OLNHO\ WR EH s WKH SUREDELOLW\ RI HUURU LV 3UHUURUf 4 > ? \ rAPLQ ZKHUH 4[f fÂ§ I H[SWfGW $f YW ; DQG 4[f LV NQRZQ DV WKH WKH FRPSOHPHQWDU\ HUURUIXQFWLRQ LQWHJUDO >@ $f
PAGE 113
$33(1',; % '(5,9$7,21 2) 75$16,(17 (48$7,216 )25 7+( /06 ),/7(5 %O 7UDQVLHQW :HLJKW $XWRFRUUHODWLRQ 0DWUL[ 8VLQJ WKH *DXVVLDQ $SSUR[LPDWLRQ ,Q WKLV VHFWLRQ D UHFXUVLYH HTXDWLRQ IRU WKH /06 ILOWHU WUDQVLHQW ZHLJKW DXWRFRUUHn ODWLRQ PDWUL[ ZLOO EH GHULYHG 7KH IXQGDPHQWDO DVVXPSWLRQ GHVFULEHG LQ 6HFWLRQ ZLOO EH XVHG WR PDNH WKH GHULYDWLRQ PDWKHPDWLFDOO\ WUDFWDEOH 7KH LQSXW WR WKH UHn FHLYHU ZLOO EH DVVXPHG WR KDYH D NQRZQ PHDQ EXW WKH HOHPHQWV RI WKH UHFHLYHG YHFWRU ZLOO EH PRGHOHG DV LQGHSHQGHQW LGHQWLFDOO\ GLVWULEXWHG *DXVVLDQ UDQGRP YDULDEOHV :H EHJLQ E\ WDNLQJ WKH RXWHU SURGXFW RI HTXDWLRQ f ZP OfZ7P f ZPfZUPf QZPfX7PfG fÂ§ ZPfZ PfXUQfX Pf ÂPPfZ PfG fÂ§ [XPfXUPfZPfZUPf ÂÂXUDfXUUDfG 7HUP f fÂ§ ÂXUDfZUPfXPfXUUDfL fÂ§ UXPfX PfZPfXUPfG 7HUP f XPfXUPfZPfZUPfXPfX7Pf %Of ,Q JHQHUDO ILOWHULQJ DSSOLFDWLRQV WKH VWDWLVWLFV RI XUDf DUH GHSHQGHQW RQ WKH YDOXH RI P +RZHYHU ZH ZLOO DVVXPH WKDW WKHVH VWDWLVWLFV DUH QRW WLPHYDU\LQJ DQG WKHUHn IRUH GURS WKH GHSHQGHQFH RQ P RQ WKLV TXDQWLW\ ZKHQ HYDOXDWLQJ WKH FRUUHVSRQGLQJ H[SHFWDWLRQV 7DNLQJ WKH H[SHFWDWLRQ RI WKH TXDQWLWLHV LQ HTXDWLRQ %Of LV UHODWLYHO\ VLPSOH H[FHSW IRU WKH WHUPV LQ OLQHV IRXU WKURXJK VL[ RI WKH HTXDWLRQ ZKLFK UHTXLUH FDUHIXO DOJHEUDLF PDQLSXODWLRQV 7KHVH VSHFLDO WHUPV KDYH EHHQ GHQRWHG DV WHUPV DQG LQ HTXDWLRQ %Of DQG WKHLU FRUUHVSRQGLQJ H[SHFWDWLRQV DUH GHILQHG EHORZ
PAGE 114
,JQRULQJ WKH VFDODUV ÂX DQG G WKH H[SHFWDWLRQ RI WHUP FDQ EH ZULWWHQ DV 1 (>XPfZUPfXPfXUPf@fA ZN (>XÂXIF8M@ %f IF L ,Q RUGHU WR HYDOXDWH (>XÂ8MW8M@ ZH XVH WKH IXQGDPHQWDO DVVXPSWLRQ ZKLFK DVVHUWV WKDW WKH LQSXWV WR WKH ILOWHU DUH LQGHSHQGHQW *DXVVLDQ GLVWULEXWHG UDQGRP YDULDEOHV :H GHILQH DQ 1 HOHPHQW YHFWRU ] ZKRVH HOHPHQWV DUH LQGHSHQGHQW ]HURPHDQ *DXVVLDQ UDQGRP YDULDEOHV HDFK KDYLQJ D YDULDQFH RI D ZKHUH R LV WKH YDULDQFH RI WKH $:*1 DW WKH LQSXW WR WKH /06 ILOWHU 8VLQJ WKLV GHILQLWLRQ ZH FDQ H[SUHVV (IXM8IF8M@ (>]Â Lf]IF IFf]M Mf@ %f $IWHU PXOWLSO\LQJ RXW DOO RI WKH WHUPV LQVLGH WKH DERYH H[SHFWDWLRQ ZH VHH WKDW WKHUH DUH RQO\ IRXU WHUPV WKDW FRQWULEXWH WR WKH H[SHFWDWLRQ DV VKRZQ EHORZ (>XÂ8MIF8M@ XIF (>]Â=M@ XL (>]N=M@ XA (>]Â]IF@ XÂXIF8M %f 'XH WR WKH DVVXPHG VWDWLVWLFDO LQGHSHQGHQFH RI WKH HOHPHQWV RI ] HDFK RI WKH H[SHFn WDWLRQ WHUPV LQ WKH DERYH HTXDWLRQ LV HDVLO\ HYDOXDWHG DV (>]P]Q@ D.P Qf %f 6XEVWLWXWLQJ HTXDWLRQ %f LQWR HTXDWLRQ %f DQG FRQVLGHULQJ DOO SRVVLEOH FRPn ELQDWLRQV RI L N DQG M ZH VHH WKDW WHUP QHJOHFWLQJ WKH VFDODUV Q DQG G RI HTXDWLRQ %Of FDQ EH ZULWWHQ LQ D FRQYHQLHQW IRUP VKRZQ EHORZ (>XPfZUPfXPfXfPf@ D ^ZUPfXO XZUPf ZPfU` %f XZUPfXX7 1RWH WKDW GXH WR WKH V\PPHWU\ RI WKH DERYH HTXDWLRQ ZH FDQ DOVR LPPHGLDWHO\ VWDWH (>XPfXUPfZPfX Pf@ D ^Z7Pf, ZUPf ZPf7` ZUPfU %f
PAGE 115
1H[W ZH FRQVLGHU HYDOXDWLQJ WHUP ZKLFK LJQRULQJ DQ\ VFDODU WHUPV FDQ EH H[SUHVVHG DV 1 1 (>XPfX7PfZPfZUPfXPfXUPf@f A A (>ZIHZS@ (>XÂXIFXS8M@ %f S N ? 1RWH WKDW ZH ZLOO XVH ] DV EHIRUH WR HYDOXDWH (>X8IF8S8M@ 7KH UHVXOW RI WKLV VXEVWLn WXWLRQ DQG H[SDQVLRQ RI WHUPV LV OLVWHG EHORZ (WXM8IF8S8M@ (>]L=N]S=M? (>]L=IF@ÂS (I]L=SMIF (>]Â=M@IFS (W]IF=SMLM (>]M=MW@ÂS (>]M=S@MIF M IFSM WHUP f WHUP f WHUP f WHUP f WHUP f %f WHUP f WHUP f WHUP f 'XH WR WKH FRPSOH[LW\ RI WKH DERYH HTXDWLRQ HDFK RI WKH WHUPV DQG LWV FRQWULEXWLRQ WR HTXDWLRQ %f ZLOO EH GHVFULEHG LQ IXOO GHWDLO :H EHJLQ ZLWK WKH H[SDQVLRQ RI WKH ILUVW WHUP LQ HTXDWLRQ %f ,Q RUGHU WR HYDOXDWH WKLV WHUP ZH PXVW XVH WKH *DXVVLDQ PRPHQWIDFWRULQJ WKHRUHP >@ (>]M=IF=S=M@ D.L Nf .M Sf ^ 6.N Sf` D.r Sf 6.M Nf ^ 6[N Sf` %f ÂNÂ a Mf N fÂ§ Sf ^ fÂ§ fÂ§ Nf` D.L Mf .L Nf 6.L Sf
PAGE 116
7KH FRQWULEXWLRQ RI WKH DERYH HTXDWLRQ WR HTXDWLRQ %f FDQ EH H[SUHVVHG DV VKRZQ EHORZ 1 1 A <@ (>Z$Z3@ (>]]IF]S=M@ 5XXPf D WUDFH5WWOXPffOf A %OOf S L IF L %DVHG RQ WKH SUHYLRXV DQDO\VLV WKH FRQWULEXWLRQV RI WHUPV DQG IURP HTXDn WLRQ %f WR HTXDWLRQ %f DUH UHDGLO\ DYDLODEOH 1 1 \\ \\ (>Z$ZS@ ^(A]AMS (A=SMMIF` 5XWWPf7f %f S L IH L :H QH[W FRQVLGHU WKH FRQWULEXWLRQ RI WHUP WR HTXDWLRQ %f 1RWH WKDW WKLV WHUP RQO\ FRQWULEXWHV WR WKRVH HOHPHQWV DORQJ WKH GLDJRQDO WKDW LV ZKHQ L M 1 1 \\ \\ (>Z$ZS@ (I]M=SMIF D7%ZZPff %f 3 IF O 6LPLODUO\ WHUP RQO\ FRQWULEXWHV WR WKRVH HOHPHQWV ZKHQ N fÂ§ S ZKLFK SURGXFHV WKH IROORZLQJ HTXDWLRQ 1 1 \\ \\ (>ZIFZS@ (A=SA2 D WUDFH 5ZZPffP7f %f S IF L )ROORZLQJ WKH DQDO\VLV DERYH WKH FRQWULEXWLRQV IURP WHUPV WKURXJK DUH UHDGLO\ OLVWHG EHORZ 1 1 \\\\(>Z:S@ ^(I]M=AMS (I]M=SMr` SAA5APff %f 3 IF O 1 1 \\ \\ (IZIF:SMMIHS U5XOOOUQfUf" %f S L IF L
PAGE 117
:H FDQ QRZ FRPELQH DOO RI WKH WHUPV IURP WKH DSSURSULDWH HTXDWLRQV WR JHW WKH ILQDO LWHUDWLYH HTXDWLRQ IRU WKH WUDQVLHQW ILOWHU ZHLJKW DXWRFRUUHODWLRQ PDWUL[ 5Z:P f 5ZZPf SZPfX7G S5ZZPf5XX ÂPZ7PfG cM5XX5ZZPf IL5XXG fÂ§ ILR ^ZUPfO ZUPf ZPfU` G fÂ§ ÂZ 7Pf7G ILR5ZZPf7 %f AD5ZZPf QD WUDFH58f8Pff, AD75XLfPf, ILR WUDFH5XWWPffU L7758ffPf [U5WXXPf 1RZ JLYHQ HTXDWLRQ %f ZH FDQ HDVLO\ GHILQH WKH WUDQVLHQW ILOWHU ZHLJKW FRYDULDQFH PDWUL[ XVLQJ WKH IROORZLQJ HTXDWLRQ RQFH ZH GHILQH Zf 5ffW\f DQG 5XX &ZZPf 5ZZPf ZPfZUPf %f % 7UDQVLHQW :HLJKW $XWRFRUUHODWLRQ 0DWUL[ 8VLQJ WKH .QRZQ 6WDWLVWLFV RI WKH 5HFHLYHUfV ,QSXW 9HFWRU ,Q WKH SUHYLRXV VHFWLRQ ZKHQ ZH HYDOXDWHG WKH WHUPV (>XXr8M@ DQG (>X8IF8S8M@ ZH DVVXPHG WKDW WKH LQGLYLGXDO HOHPHQWV RI X ZHUH LQGHSHQGHQW :H KDYH HVVHQWLDOO\ LJQRUHG WKH KLJKO\ VWUXFWXUHG QDWXUH RI WKH /06 ILOWHUfV UHFHLYHG YHFWRU IRU RXU '6 &'0$ V\VWHP PRGHO +RZHYHU E\ XVLQJ WKLV DVVXPSWLRQ WKH UHVXOWLQJ DQDO\VLV RI WKH WUDQVLHQW HTXDWLRQV IRU WKH ILOWHU ZHLJKW YHFWRU LV H[WUHPHO\ VLPSOLILHG ,Q WKLV VHFWLRQ ZH ZLOO XVH WKH NQRZQ VWDWLVWLFV RI UPf LQ RUGHU WR VWXG\ WKH WUDQVLHQW VWDWLVWLFV RI ZPf LQ PRUH GHWDLO 7KH FRPSOH[LW\ RI WKH UHVXOWLQJ HTXDWLRQV ZLOO PDNH LW SDLQIXOO\ REYLRXV ZK\ WKH LQGHSHQGHQFH DVVXPSWLRQ LV URXWLQHO\ XVHG IRU WKLV SUREOHP
PAGE 118
5HFDOO WKDW XPf UPf ZKHUH U Pf DLSL ÂLf YPf 1P % f N %f ,Q RUGHU WR VLPSOLI\ WKH IROORZLQJ H[SUHVVLRQV OHW D[ DQG UHSUHVHQW WKH [WK FRPSRQHQW RI WKH YHFWRU DASLAf $OVR OHW WKH WHUP Y[ UHSUHVHQW WKH [WK HOHPHQW RI WKH YPf YHFWRU IURP WKH SUHYLRXV HTXDWLRQ 8VLQJ WKLV QRWDWLRQ DQG UHFDOOLQJ WKDW WKH GDWD ELWV IRU WKH LQWHUIHULQJ XVHUV DUH DVVXPHG WR EH LQGHSHQGHQW DQG HTXDOO\ OLNHO\ WR EH s EUXWHIRUFH DOJHEUDLF PDQLSXODWLRQV DUH XVHG WR ILQG WKH GHVLUHG TXDQWLWLHV (IXM8IF8M@ DADADM DDM.L Nf DDL.N Mf DDN.L Mf DÂ (>YIF9M@ DN (IYM9M@ DM (>YÂYIF@ %f :H DUH UHDOO\ LQWHUHVWHG LQ VXEVWLWXWLQJ WKH SUHYLRXV HTXDWLRQ LQ WR HTXDWLRQ %f ,Q RUGHU WR VLPSOLI\ WKH QRWDWLRQ IRU WKLV VXEVWLWXWLRQ DV ZHOO DV RWKHU IROORZLQJ H[n SUHVVLRQV OHW DIFBLfD DQG DIFf[ GHQRWH WKH [WK FRPSRQHQW RI DNLSNKf DQG DN^3N6Nf UHVSHFWLYHO\ (.YL@ 3fW U OUe OU %f %f ZKHUH %f %f
PAGE 119
8VLQJ WKH YHFWRUPDWUL[ QRWDWLRQ IURP WKH SUHYLRXV VHFWLRQ WKH IROORZLQJ HTXDWLRQ UHVXOWV (>XPfZUPfXPfXUPf@ (>XPfX PfZPfXUPf@ UZUPfD SL ÂLf, DDLSL ÂLfZUPf DL3LALfZUPfDLSÂLfDISLÂ£f %f UZPfD>SLÂ£Lf ZUPfDLSLÂLfU UZPfDISÂ£f DSÂ£fZUPfU ,Q RUGHU WR HYDOXDWH (>XÂXAXS8M@ ZH ZLOO H[SDQG WKLV H[SUHVVLRQ DV D VXP RI WKUHH IXQFWLRQV (>XÂ8MIF8S8M@ $X XN 8S XMf %X XNO XS Xf &XÂ XIF XS 8Mf %f (DFK RI WKHVH IXQFWLRQV ZLOO EH H[SDQGHG LQ WKH IROORZLQJ HTXDWLRQV :H EHJLQ ZLWK H[SDQGLQJ WKH XÂ XA XS 8Mf WHUP $XÂ XIF 8S 8Mf DDSDM.L Nf DSDM (>YÂYIF@ DDNDM.L Sf DIFD (>YYS@ DDLDM6.N Sf (>YIFYS@ DÂDADSDM %f 6XEVWLWXWLQJ WKLV WHUP LQWR HTXDWLRQ %f ZH ILQG 1 1 A A (>ZIF:S@LXM XN 8S 8Mf 75WLf8fPfDSÂLfD>SÂ£LffS L IF L D WUDFH 5:,WfPffDLSL ÂLfD>SLLff U5XOWPfDSL ÂLfDI SLÂLffL Â WUDFH 5WWf,WPfUfDLSL ÂLfD>S[ ÂLffLÂ DLS[ Â[fDI SL Â£f58f8PfDS Â£LfDI SL ÂLff 4 %f
PAGE 120
1H[W ZH FRQVLGHU WKH 5XÂ Xr XS 8Mf WHUP DQG LWV FRQWULEXWLRQ WR WKH WUDQVLHQW FRUUHODWLRQ PDWUL[ %XL XIF XS XMf D6.L Nf 6.M Sf ^ 6.N Sf` Sf 6.^M Nf^ 6.N Sf` D6.L Mf .N Sf ^ .L Nf` D6.L Mf .L Nf 6.L Sf %f DDNDS.L Mf D (>YIFYS@Â. Mf DÂDSÂ£.IF Mf R (>YÂ9S@.IF Mf DÂDIFÂ£.S Mf D (>YÂYIF@Â£.S Mf 6XEVWLWXWLRQ RI WKLV HTXDWLRQ LQWR HTXDWLRQ %f OHDGV WR WKH IROORZLQJ HTXDWLRQ 1 1 A H>ZrZS@XÂ! 8IF 8S XMf D.ZZPf D WUDFH 5APfff S L IF L WIDISL$f5 ZZ PfDSLÂ£fOfFUDLS ÂLfDI SL ÂLM5APff M WUDFH 5XWXPfUfOfL DU5ZZPff )LQDOO\ WKH ODVW WHUP LV FRQVLGHUHG &8L XIF 8S XMf DÂDIF (>YS9M@ D (>YIFYÂ@Â.r Sf DIFDS (>YM9B@ D (>YÂ9M@Â.IF Sf DÂDS (>YIF9M@ D (>YS9M@.L Nf (>YYIF9S9M@ %f
PAGE 121
,Q RUGHU WR HYDOXDWH WKH DERYH H[SUHVVLRQ WKH IROORZLQJ DX[LOLDU\ HTXDWLRQ LV UHTXLUHG N (>Y9W9X9@ AA >DIFOfVDIFOfLDIFOfXDLWOfL@ fÂ§ N 3L %f A >DIFfVDcWfWDIFfXDIFf,@ N = 6XEVWLWXWLRQ RI HTXDWLRQ %f LQWR HTXDWLRQ %f OHDGV WR WKH IROORZLQJ HTXDWLRQ 1 1 AO(>ZIF:S@&XLXMIHfXOXf D5WWf8PfUfS O N DLSL ÂLfDI SL ÂLf588PfUf FU WUDFH +ZZ 0fUf D>SLf5X>8PfDLSLÂLfUf[ 7M5fURfU% UR5A0UR\ %f
PAGE 122
,OO :H QRZ FRPELQH DOO RI WKH DSSURSULDWH WHUPV WR JHW WKH ILQDO H[SUHVVLRQ IRU WKH WUDQVLHQW ILOWHU ZHLJKW DXWRFRUUHODWLRQ PDWUL[ 5ZZLSL f 5ZZ^Pf SZPfD[ ^SL[fG S5ZZPf5XX 6LfZ7PfG Q5XX5ZZPf S5XXG fÂ§ SDY7PfD[S[ [f,G fÂ§ SDD[S[[fnZ7PfG fÂ§ SD[SL ÂLfZUPfDLS[ Â fDI S[6[fG fÂ§ SFUZUDfDI S[ [fG fÂ§ SZ7PfD[S[[f7G fÂ§ S7ZPfD[S[ [fG fÂ§ SD[S[ [fZ7Pf7G SD5ZZPfD[ S[ ÂLfDI S[ ;f SD WUDFH 58fLWPffDLSL [fD[ S[ 6[f SU5ZZPfD[S[ [fD>S[ [f %f S WUDFH 5ZZPf7fD[ S[ [fDI S[ Â£[f ADLSL ÂLfD> SL f5XXPfDLSL Â[fDI S[[f SFU5ZZPf S9 WUDFH 5ZZPffO S9DI SL 6[f5ZZPfD[S[ ÂLf, S9DS Â[fD> S[ S9 WUDFH 5WPfUf, SFUU5OW8fUDf SD5ZZPf7 SD[SL Â[fDI SOW ÂLf5APfU 3R WUDFH 5AfPffU SD[S[ Â£f5XXPfDI S[ Â[fU AU e5ZXPfU H \U527PfU
PAGE 123
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f ZKHUH \ LV D JLYHQ WKUHVKROG 8VLQJ WKH NQRZQ MRLQW SUREDELOLW\ GHQVLW\ IXQFWLRQ RI ] WKLV SUREDELOLW\ FDQ EH H[SUHVVHG DV 3Y]7$]\f Â‘Â‘Â‘ Wff_Y_L Hr3 Vf79f] ]ff G] &Of ZKHUH G] LV XVHG WR GHQRWH G]?G] G]Q DQG WKH QGLPHQVLRQDO LQWHJUDO LV LQWHJUDWHG RYHU WKH UHJLRQ fÂ§ ^5 =$= \` :H FDQ VLPSOLI\ WKH DERYH H[SUHVVLRQ E\ GHILQLQJ WKH FKDQJH RI YDULDEOHV P ] fÂ§ ] LQ ZKLFK FDVH WKH SUHYLRXV H[SUHVVLRQ EHFRPHV 3U]U$] \f A A_\_D H[S MPf9fPf GP &f ZKHUH WKH LQWHJUDWLRQ UHJLRQ LV QRZ ^5Q 0 ]fU$ 0 rf \` 1RWH WKH FRYDULDQFH PDWUL[ 9 LV SRVLWLYH GHILQLWH DQG FDQ WKHUHIRUH EH H[SUHVVHG XVLQJ D &KROHVN\ GHFRPSRVLWLRQ 7KLV PHDQV WKDW 9 //U &f
PAGE 124
ZKHUH / LV D QRQVLQJXODU ORZHU WULDQJXODU PDWUL[ 7KH IROORZLQJ H[SUHVVLRQV DUH EDVHG RQ VLPSOH DOJHEUDLF PDQLSXODWLRQV RI WKH SUHYLRXV HTXDWLRQ _9_ _/__/UO _/_ &f 9 /OfU/ &f /HW [ / P 8VLQJ WKLV FKDQJH RI YDULDEOHV WKH MRLQW SUREDELOLW\ GHQVLW\ IXQFWLRQ RI [ LV WKHQ [[f 0Pf ,/ P /[ &f 8VLQJ WKLV FKDQJH RI YDULDEOHV HTXDWLRQ &f FDQ EH H[SUHVVHG DV 3U]U$] \ ff H;3 r Y f ZKHUH WKH UHJLRQ RI LQWHJUDWLRQ LV _U /; /]ff7$/; /B]f \` ^L ; /]f7/U$/f; /B]f \` :H FDQ IXUWKHU VLPSOLI\ HTXDWLRQ &f E\ H[SUHVVLQJ /W$/ 3$3U &f &f ZKHUH 3 LV D PDWUL[ RI RUWKRQRUPDO HLJHQYDOXHV DQG $ LV D GLDJRQDO PDWUL[ RI WKH FRUUHVSRQGLQJ HLJHQYDOXHV 8VLQJ WKH DERYH HTXDWLRQ ZH FDQ DOVR VWDWH WKDW $ 3U/U$/3 & f
PAGE 125
1RZ GHILQH D IXUWKHU FKDQJH RI YDULDEOHV Z 3[ 8VLQJ WKLV FKDQJH RI YDULDEOHV HTXDWLRQ &f FDQ EH H[SUHVVHG DV 3U]U$] \f M Â‘ Â‘ Â‘ r H[S L3UZf737Zf GZ n n ÂAH[SZ7ZfLZ n &OOf ZKHUH WKH UHJLRQ RI LQWHJUDWLRQ LV _@5Q 3: 3U/B]ff7/7$/f3: 3A]f \` ^LU : 3U/B]f $: 3Af \` &f 8VLQJ HTXDWLRQV &OOf DQG &f ZH VHH WKDW LQ RUGHU WR FDOFXODWH 3U]U$] \f ZH FRXOG FRQFHQWUDWH RQ WKH GLVWULEXWLRQ RI Q < H[S\f@ 7KHQ ZH FDQ H[SUHVV WKH PRPHQW JHQHUDWLQJ IXQFWLRQ DV +Vf fÂ§ (>H[SV)f@ ( H[S r !:0D &f ? Â 'XH WR WKH VWDWLVWLFDO LQGHSHQGHQFH RI WKH :c WKH DERYH H[SUHVVLRQ FDQ EH VLPSOLILHG WR LIVf Q(>H[SV$L:nMLLff@ &f
PAGE 126
:H QRZ FRQFHQWUDWH RQ HYDOXDWLQJ HDFK WHUP LQ WKH SUHYLRXV HTXDWLRQ :H VWDUW ZLWK WKH IROORZLQJ HTXDWLRQ 22 ( >H[S V$Â:Â fÂ§ tÂff@ fÂ§M H[SV?LZ fÂ§ Eff H[SfÂ§Zf GZ & Y LU f ,Q RUGHU WR VLPSOLI\ WKH DERYH LQWHJUDQG H[SDQG WKH WHUPV LQ WKH H[SRQHQW DQG XVH WKH FKDQJH RI YDULDEOHV [ Z?fÂ§ ?V $IWHU SHUIRUPLQJ WKHVH RSHUDWLRQV WKH DERYH HTXDWLRQ EHFRPHV ( >H[S Âff@ ,:QH;3^ V;X [ &f [ f H[SfÂ§[f G[ 9, $A ,Q RUGHU WR VLPSOLI\ WKH DERYH LQWHJUDQG FRPSOHWH WKH VTXDUH RI WKH WHUPV LQ WKH H[SRQHQW 7KH UHVXOW RI WKLV RSHUDWLRQ LV R "$K ? ( >H[S 6$: rf!f@ H[S MAf m3Â$rf MAH;3^fÂ§ KU V?MEL 9O a $M6 &f G[ 7KH LQWHJUDO LQ WKH DERYH H[SUHVVLRQ LV UHFRJQL]HG DV WKH SUREDELOLW\ GHQVLW\ IXQFWLRQ RI D *DXVVLDQ UDQGRP YDULDEOH 7KHUHIRUH WKH LQWHJUDO HYDOXDWHV WR RQH DQG WKH DERYH H[SUHVVLRQ FDQ EH ZULWWHQ DV ( >H[S $: mrf@ A H[S eJf H[S$"f H;3 fRtL H;3 86 $ V &f 9O $L6 'LUHFW VXEVWLWXWLRQ RI HTXDWLRQ &f LQWR HTXDWLRQ &f SURGXFHV WKH PRPHQW JHQHUDWLQJ IXQFWLRQ RI < $6 &f
PAGE 127
$V WKH QDPH LPSOLHV WKH PRPHQW JHQHUDWLQJ IXQFWLRQ RI D UDQGRP YDULDEOH PD\ EH XVHG WR HYDOXDWH WKH PRPHQWV RI WKH UDQGRP YDULDEOH DV VKRZQ EHORZ PQ (>
PAGE 128
1H[W ZH FDQ ILQG DQ XSSHU OLPLW RQ WKH GHVLUHG SUREDELOLW\ XVLQJ WKH IDFW WKDW RR RR 8D fÂ§ \fI<^\f G\ H[S^V^\ DffI<^\f G\ UHDOVf &f fÂ§ 22 fÂ§22 :H UHFRJQL]H WKH ULJKW KDQG VLGH RI WKH DERYH HTXDWLRQ DV DQ H[SRQHQWLDO WLPHV WKH SUHYLRXVO\ GHULYHG PRPHQW JHQHUDWLQJ IXQFWLRQ 7KHUHIRUH WKH XSSHU OLPLW RQ WKH GHVLUHG SUREDELOLW\ LV 3U\ Df H[SfÂ§DVf+Vf UHDOVf &f $W WKLV SRLQW D FRXSOH RI FRPPHQWV DUH LQ RUGHU :H LQLWLDOO\ VWDWHG WKDW WKH WUDQVIRUPDWLRQ PDWUL[ $ ZDV V\PPHWULF +RZHYHU LI WKH PDWUL[ LV QRW V\PPHWULF LW FDQ HDVLO\ EH WUDQVIRUPHG LQWR D V\PPHWULF PDWUL[ &RQVLGHU WKH JHQHUDO TXDGUDWLF IRUP ; ]U$] &f 1RWH WKDW WKH SUHYLRXV HTXDWLRQ PD\ DOVR EH H[SUHVVHG DV Q Q ; fÂ§ \L ]N]M$fNM &f N M O ,I ZH FRQVLGHU WKH WHUPV ZKHUH N A M WKH FRQWULEXWLRQ WR ; IRU HDFK RI WKHVH WHUPV LV WKHQ =N=M>$f&M $fA@ :H FDQ WKHUHIRUH WUDQVIRUP $ LQWR D QHZ V\PPHWULF PDWUL[ % ZKHUH % >$ $U@ DQG ; ]7%] :KHQ ZH FRQVLGHU WKH JHQHUDO TXDGUDWLF IRUP ; fÂ§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
PAGE 129
WKH PDLQ GLDJRQDO 7KHUHIRUH LI WKH WUDFH RI D UHDO V\PPHWULF PDWUL[ LV OHVV WKDQ RU HTXDO WR ]HUR WKHQ WKH PDWUL[ LV QRW SRVLWLYH GHILQLWH 2QO\ ZKHQ WKH WUDQVIRUPDWLRQ PDWUL[ $ LV QRW SRVLWLYH GHILQLWH LV LW SRVVLEOH IRU WKH 3U]U$] f WR EH QRQ]HUR
PAGE 130
$33(1',; 7<32*5$3+,&$/ 127(6 7KLV GLVVHUWDWLRQ ZDV W\SHG DQG HGLWHG XVLQJ *18 (PDFV ZLWK WKH $8&7H; DQG 5HI7H; SURJUDPV RQ D 1H;7VWDWLRQp 7XUER ZRUNVWDWLRQ 7KHVH SURJUDPV JUHDWO\ IDFLOLWDWH WKH FUHDWLRQ RI HUURUIUHH '7A; H VRXUFH FRGH DQG DUH KLJKO\ UHFRPPHQGHG E\ WKH DXWKRU 7KH GLVVHUWDWLRQ ZDV W\SHVHW XVLQJ ',A; H YHUVLRQ 7KH DXWKRU H[WHQGV KLV JUDWLWXGH WR WKRVH LQGLYLGXDOV ZKR KDYH FUHDWHG WKHVH SURJUDPV DQG KDYH PDGH WKHP IUHHO\ DYDLODEOH WR WKH PDVVHV 7KH *UDGXDWH 6FKRRO DW WKH 8QLYHUVLW\ RI )ORULGD KDV VHYHUDO W\SHVHWWLQJ UHTXLUHn PHQWV WKDW DUH WUDGLWLRQDOO\ FRQVLGHUHG LQDSSURSULDWH E\ WKH JHQHUDO '7@H; H FRPPXn QLW\ 7KH VSHFLILF UHTXLUHPHQWV GHDO ZLWK OLQHVSDFLQJ XQGHUOLQLQJ VHFWLRQ KHDGLQJV DQG JHQHUDO SDJH OD\RXW ,Q RUGHU WR PHHW WKHVH UHTXLUHPHQWV WKH DXWKRU VSHQW TXLWH D ELW RI WLPH FUHDWLQJ D $,A;e FODVV ILOH FDOOHG XIWKHVLV&/6 WKDW PRGLILHV WKH VWDQGDUG UHSRUW FODVV WR PHHW WKH SDJH OD\RXW JXLGHOLQHV DQG IDFLOLWDWHV VZLWFKLQJ WKH OLQHVSDFLQJ DV UHTXLUHG .RSND DQG 'DO\ >@ *RRVHQV HW DO >@ DV ZHOO DV WKH FODVVHVGW[ DQG OWVHFWGW[ ILOHV SDUW RI WKH VWDQGDUG ,$,A;e GLVWULEXWLRQf ZHUH YHU\ XVHIXO LQ FUHDWLQJ WKH FODVV ILOH ,Q DGGLWLRQ WKH '7(;H SDFNDJHV XOHPVW\ YHUVLRQ f E\ 'RQDOG $UVHQDX DQG 6(763$&(67< YHUVLRQ f E\ *Hn RIIUH\ 7RELQ DQG 6(&767<67< YHUVLRQ YO f E\ 5RZODQG 0F'RQQHOO ZHUH XVHG WR IRUPDW VHFWLRQ KHDGLQJV DQG DGMXVW OLQHVSDFLQJ 7KHVH SDFNDJHV DUH UHDGLO\ DYDLODEOH IURP \RXU QHDUHVW &RPSUHKHQVLYH 7I; $UFKLYH 1HWZRUN &7$1f VLWH >@ 1H;7 LV D UHJLVWHUHG WUDGHPDUN RI 1H;7 6RIWZDUH ,QF
PAGE 131
7KH GRFXPHQW FODVV ILOH LV FXUUHQWO\ DYDLODEOH YLD WKH :RUOG :LGH :HE DW KWWS EDUQH\ HH XI HGXaUVPLWK/D7H;H RU KWWS ZZZ PDWK XI HGXaQULQJ ZKHUH LQIRUPDWLRQ UHJDUGLQJ WKH SDFNDJH DQG LWV XVH DUH DYDLODEOH 7KH SDFNDJH LV GLVWULEXWHG ZLWK LQWHJUDWHG GRFXPHQWDWLRQ ZKHUH WKH GHVFULSWLRQ DQG WKH FRGLQJ DUH PHUJHG LQWR D VLQJOH VRXUFH ILOH FDOOHG XIWKHVLVGW[ 7KH 'RF6WULS XWLOLW\ LV XVHG WR VHSDUDWH WKH XIWKHVLV&/6 GRFXPHQW FODVV ILOH ZKLOH WKH GRF SDFNDJH LV XVHG WR W\SHVHW WKH GRFn XPHQWDWLRQ 7KLV SDFNDJLQJ DLGV LQ WKH PDLQWHQDQFH DQG UHYLVLRQ RI 87H;e VRXUFH FRGLQJ ,Q DGGLWLRQ E\ SURYLGLQJ GRFXPHQWDWLRQ LW LV KRSHG WKDW RWKHU 57H;e XVHUV ZLOO EH DEOH WR XVH WKLV GRFXPHQW FODVV ILOH DQG PRGLI\ LW WR VXLW WKHLU QHHGV )RU WKH QHDU IXWXUH WKH DXWKRU FDQ EH UHDFKHG YLD HPDLO DW VPLWKBURQ#XVDQHW DQG WLPH SHUPLWWLQJ LV ZLOOLQJ WR DGGUHVV TXHVWLRQV UHJDUGLQJ WKH XVH RI WKLV GRFXPHQW FODVV 3OHDVH FRQWDFW WKH DXWKRU LI \RX FDQQRW ILQG WKH GRFXPHQW FODVV SDFNDJH DW WKH DERYH 85/V DV WKH DXWKRU PD\ KDYH PRYHG WKLV LQIRUPDWLRQ WR D PRUH SHUPDQHQW ZHEVLWH
PAGE 132
5()(5(1&(6 >@ 5 +XJKHV DQG +HQGULFNV f6SUHDGVSHFWUXP UDGLRf 6FLHQWLILF $PHULFDQ YRO QR SS $SU >@ 5 $ 6FKROW] f7KH VSUHDG VSHFWUXP FRQFHSWf ,((( 7UDQVDFWLRQV RQ &RPPXn QLFDWLRQV YRO QR &20 SS $XJ >@ 5 / 3LFNKROW] / 6FKLOOLQJ DQG / % 0LOVWHLQ f7KHRU\ RI VSUHDGVSHFWUXP FRPPXQLFDWLRQVfÂ§D WXWRULDOf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO &20 QR SS 0D\ >@ 5 ( =LHPHU DQG 5 / 3HWHUVRQ 'LJLWDO &RPPXQLFDWLRQV DQG 6SUHDG 6SHFWUXP 6\VWHPV 0DFPLOODQ >@ 9 6DUZDWH DQG 0 % 3XUVOH\ f&URVVFRUUHODWLRQ SURSHUWLHV RI SVXHGRUDQGRP DQG UHODWHG VHTXHQFHVf 3URFHHGLQJV RI WKH ,((( YRO QR SS 0D\ >@ ( + 'LQDQ DQG % DEEDUL f6SUHDGLQJ FRGHV IRU GLUHFW VHTXHQFH &'0$ DQG ZLGHEDQG &'0$ QHWZRUNVf ,((( &RPPXQLFDWLRQV 0DJD]LQH YRO QR SS 6HSW >@ + )XNXPDVD 5 .RKQR DQG + ,PDL f'HVLJQ RI SVXHGRQRLVH VHTXHQFHV ZLWK JRRG RGG DQG HYHQ FRUUHODWLRQ SURSHUWLHV IRU '6&'0$f ,((( RXUQDO RQ 6HOHFWHG $UHDV LQ &RPPXQLFDWLRQV YRO QR SS XQH >@ 7 0RRQ 5 7 6KRUW DQG & 5XVKIRUWK f$YHUDJH DFTXLVLWLRQ WLPH IRU 660$ FKDQQHOVf LQ 3URFHHGLQJV ,((( 0LOLWDU\ &RPPXQLFDWLRQV &RQIHUHQFH 0F/HDQ 9$ YRO SS >@ 7 0RRQ 5 7 6KRUW DQG & 5XVKIRUWK f$ 5$6( DSSURDFK WR DFTXLVLWLRQ LQ 660$ FKDQQHOVf LQ 3URFHHGLQJV ,((( 0LOLWDU\ &RPPXQLFDWLRQV &RQIHUHQFH 0F/HDQ 9$ YRO SS >@ 5 7 %DUJKRXWKL DQG / 6WLLEHU f6HTXHQFH DFTXLVLWLRQ WHFKQLTXHV IRU &'0$ V\VWHPV HPSOR\LQJ .DVDPL VHTXHQFHVf LQ 3URFHHGLQJV ,((( 0LOLWDU\ &RPPXn QLFDWLRQV &RQIHUHQFH 6DQ 'LHJR &$ YRO SS >@ 5 7 %DUJKRXWKL DQG / 6WLLEHU f5DSLG VHTXHQFH DFTXLVLWLRQ IRU '6&'0$ V\VWHPV HPSOR\LQJ .DVDPL VHTXHQFHVf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO QR SS )HE0DU$SU
PAGE 133
>@ 8 0DGKRZ DQG 0 % 3XUVOH\ f$FTXLVLWLRQ EDVHG FDSDFLW\ RI GLUHFWVHTXHQFH VSUHDGVSHFWUXP FRPPXQLFDWLRQ QHWZRUNVf LQ 3URFHHGLQJV RI WKH &RQIHUHQFH RI ,QIRUPDWLRQ 6FLHQFHV DQG 6\VWHPV 7KH RKQ +RSNLQV 8QLYHUVLW\ %DOWLPRUH 0' 0DU >@ $ 1 %DUERVD DQG 6 / 0LOOHU f$GDSWLYH GHWHFWLRQ RI '6&'0$ VLJQDOV LQ IDGLQJ FKDQQHOVf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO QR SS DQ >@ / : &RXFK ,, 'LJLWDO DQG $QDORJ &RPPXQLFDWLRQ 6\VWHPV 3UHQWLFH+DOO WK HGLWLRQ >@ 6 / 0LOOHU f$Q DGDSWLYH GLUHFWVHTXHQFH FRGHGLYLVLRQ PXOWLSOHDFFHVV UHFHLYHU IRU PXOWLXVHU LQWHUIHUHQFH UHMHFWLRQf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO QR SS )HE0DU$SU >@ 6 9HUG f0LQLPXP SUREDELOLW\ RI HUURU IRU DV\QFKURQRXV *DXVVLDQ PXOWLSOH DFFHVV FKDQQHOVf ,((( 7UDQVDFWLRQV RQ ,QIRUPDWLRQ 7KHRU\ YRO QR SS DQ >@ 6 9HUG f2SWLPXP PXOWLXVHU DV\PSWRWLF HIILFLHQF\f ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO &20 QR SS 6HSW >@ 5 /XSDV DQG 6 9HUG f/LQHDU PXOWLXVHU GHWHFWRUV IRU V\QFKURQRXV FRGHn GLYLVLRQ PXOWLSOHDFFHVV FKDQQHOVf ,((( 7UDQVDFWLRQV RQ ,QIRUPDWLRQ 7KHRU\ YRO QR SS DQ >@ 5 /XSDV DQG 6 9HUG f1HDUIDU UHVLVWDQFH RI PXOWLXVHU GHWHFWRUV LQ DV\Qn FKURQRXV FKDQQHOVf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO QR SS $SU >@ 6 3DUNYDOO ( 6WURP DQG % 2WWHUVWHQ f7KH LPSDFW RI WLPLQJ HUURUV RQ WKH SHUIRUPDQFH RI OLQHDU '6&'0$ UHFHLYHUVf ,((( RXUQDO RQ 6HOHFWHG $UHDV LQ &RPPXQLFDWLRQV YRO QR SS 2FW >@ )& =KHQJ DQG 6 %DUWRQ f2Q WKH SHUIRUPDQFH RI QHDUIDU UHVLVWDQW &'0$ GHWHFWRUV LQ WKH SUHVHQFH RI V\QFKURQL]DWLRQ HUURUVf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO QR SS 'HF >@ 0 9DUDQDVL DQG % $D]KDQJ f0XOWLVWDJH GHWHFWLRQ LQ DV\QFKURQRXV FRGHn GLYLVLRQ PXOWLSOHDFFHVV FRPPXQLFDWLRQVf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDn WLRQV YRO QR SS $SU >@ ( 6WURP 6 3DUNYDOO 6 / 0LOOHU DQG % ( 2WWHUVWHQ f3URSDJDWLRQ GHOD\ HVWLPDWLRQ LQ DV\QFKURQRXV GLUHFWVHTXHQFH FRGHGLYLVLRQ PXOWLSOH DFFHVV V\Vn WHPV ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO QR SS DQ
PAGE 134
>@ 6 ( %HQVOH\ DQG % $D]KDQJ f6XEVSDFHEDVHG FKDQQHO HVWLPDWLRQ IRU FRGH GLYLVLRQ PXOWLSOH DFFHVV FRPPXQLFDWLRQ V\VWHPVf ,((( 7UDQVDFWLRQV RQ &RPn PXQLFDWLRQV YRO QR SS $XJ >@ ( 6WUP 6 3DUNYDOO 6 / 0LOOHU DQG % ( 2WWHUVWHQ f'6&'0$ V\Qn FKURQL]DWLRQ LQ WLPHYDU\LQJ IDGLQJ FKDQQHOVf ,((( RXUQDO RQ 6HOHFWHG $UHDV LQ &RPPXQLFDWLRQV YRO QR SS 2FW >@ 6 ( %HQVOH\ DQG % $D]KDQJ f0D[LPXP/LNHOLKRRG V\QFKURQL]DWLRQ RI D VLQJOH XVHU IRU FRGHGLYLVLRQ PXOWLSOHDFFHVV FRPPXQLFDWLRQ V\VWHPVf ,((( 7UDQVDFn WLRQV RQ &RPPXQLFDWLRQV YRO QR SS 0DU >@ =KHQJ /L 6 / 0LOOHU DQG ( 6WURP f$Q HIILFLHQW FRGHWLPLQJ HVWLPDWRU IRU '6&'0$ VLJQDOVf ,((( 7UDQVDFWLRQV RQ 6LJQDO 3URFHVVLQJ YRO QR SS DQ >@ =6 /LX /L DQG 6 / 0LOOHU f$Q HIILFLHQW FRGHWLPLQJ HVWLPDWRU IRU UHFHLYHU GLYHUVLW\ '6&'0$ V\VWHPVf ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV YRO QR SS XQH >@ 0 / +RQLJ 8 0DGKRZ DQG 6 9HUG f%OLQG DGDSWLYH PXOWLXVHU GHWHFWLRQf ,((( 7UDQVDFWLRQV RQ ,QIRUPDWLRQ 7KHRU\ YRO QR SS XO\ >@ 8 0DGKRZ f%OLQG DGDSWLYH LQWHUIHUHQFH VXSSUHVVLRQ IRU WKH QHDUIDU UHVLVWDQW DFTXLVLWLRQ DQG GHPRGXODWLRQ RI GLUHFWVHTXHQFH &'0$ VLJQDOVf ,((( 7UDQVn DFWLRQV RQ 6LJQDO 3URFHVVLQJ YRO QR SS DQ >@ 6 +D\NLQ $GDSWLYH )LOWHU 7KHRU\ 3UHQWLFH+DOO QG HGLWLRQ >@ 6 / 0LOOHU f7UDQVLHQW EHKDYLRU RI WKH PLQLPXP PHDQVTXDUHG HUURU UHFHLYHU IRU GLUHFWVHTXHQFH FRGHGLYLVLRQ PXOWLSOHDFFHVV V\VWHPVf LQ 3URFHHGLQJV ,((( 0LOLWDU\ &RPPXQLFDWLRQV &RQIHUHQFH )RUW 0RQPRXWK 1YRO SS >@ 6 / 0LOOHU f7UDLQLQJ DQDO\VLV RI DGDSWLYH LQWHUIHUHQFH VXSSUHVVLRQ IRU GLUHFW VHTXHQFH FRGHGLYLVLRQ PXOWLSOHDFFHVV V\VWHPVf ,((( 7UDQVDFWLRQV RQ &RPn PXQLFDWLRQV YRO QR SS $SU >@ 8 0DGKRZ f$GDSWLYH LQWHUIHUHQFH VXSSUHVVLRQ IRU MRLQW DFTXLVLWLRQ DQG GHn PRGXODWLRQ RI GLUHFWVHTXHQFH &'0$ VLJQDOVf LQ 3URFHHGLQJV ,((( 0LOLWDU\ &RPPXQLFDWLRQV &RQIHUHQFH 6DQ 'LHJR &$ YRO SS >@ + 9 3RRU DQG 6 9HUG f3UREDELOLW\ RI HUURU LQ 006( PXOWLXVHU GHWHFWLRQf ,((( 7UDQVDFWLRQV RQ ,QIRUPDWLRQ 7KHRU\ YRO QR SS 0D\
PAGE 135
>@ ( 6WURP 'LUHFW6HTXHQFH &RGH'LYLVLRQ 0XOWLSOH$FFHVV 6\VWHPV 1HDU)DU 5HVLVWDQW 3DUDPHWHU (VWLPDWLRQ DQG 'HWHFWLRQ 3K' GLVVHUWDWLRQ 8QLYHUVLW\ RI )ORULGD >@ 6HQQH $GDSWLYH /LQHDU 'LVFUHWH7LPH (VWLPDWLRQ 3K' GLVVHUWDWLRQ 6WDQIRUG 8QLYHUVLW\ XQH >@ / +RURZLW] DQG 6HQQH f3HUIRUPDQFH DGYDQWDJH RI FRPSOH[ /06 IRU FRQWUROOLQJ QDUURZEDQG DGDSWLYH DUUD\Vf ,((( 7UDQVDFWLRQV RQ $FRXVWLFV 6SHHFK DQG 6LJQDO 3URFHVVLQJ YRO $663 SS XQH >@ % )LVKHU DQG 1 %HUVKDG f7KH FRPSOH[ /06 DGDSWLYH DOJRULWKPfÂ§WUDQVLHQW ZHLJKW PHDQ DQG FRYDULDQFH ZLWK DSSOLFDWLRQV WR WKH $/(f ,((( 7UDQVDFWLRQV RQ $FRXVWLFV 6SHHFK DQG 6LJQDO 3URFHVVLQJ YRO $663 QR SS DQ >@ $ )XHXU DQG ( :HLQVWHLQ f&RQYHUJHQFH DQDO\VLV RI /06 ILOWHUV ZLWK XQFRUn UHODWHG GDWDf ,((( 7UDQVDFWLRQV RQ $FRXVWLFV 6SHHFK DQG 6LJQDO 3URFHVVLQJ YRO $663 SS )HE >@ 6 & 'RXJODV DQG 7 +< 0HQJ f([DFW H[SHFWDWLRQ DQDO\VLV RI WKH /06 DGDSn WLYH ILOWHU ZLWKRXW WKH LQGHSHQGHQFH DVVXPSWLRQf LQ 3URFHHGLQJV ,((( ,QWHUn QDWLRQDO &RQIHUHQFH RQ $FRXVWLFV 6SHHFK DQG 6LJQDO 3URFHVVLQJ 6DQ )UDQFLVFR &$ YRO SS >@ 6 & 'RXJODV f([DFW H[SHFWDWLRQ DQDO\VLV RI WKH /06 DGDSWLYH ILOWHU IRU FRUn UHODWHG *DXVVLDQ LQSXW GDWDf LQ 3URFHHGLQJV ,((( ,QWHUQDWLRQDO &RQIHUHQFH RQ $FRXVWLFV 6SHHFK DQG 6LJQDO 3URFHVVLQJ 0LQQHDSROLV 01 YRO SS >@ & : +HOVWURP 3UREDELOLW\ DQG 6WRFKDVWLF 3URFHVVHV IRU (QJLQHHUV 0DFPLOODQ 3XEOLVKLQJ &RPSDQ\ QG HGLWLRQ >@ + 6FKHII 7KH $QDO\VLV RI 9DULDQFH RKQ :LOH\ t 6RQV ,QF >@ 1 / RKQVRQ DQG 6 .RW] &RQWLQXRXV 8QLYDULDWH 'LVWULEXWLRQV +RXJKWRQ 0LIIOLQ >@ 6 2 5LFH f'LVWULEXWLRQ RI TXDGUDWLF IRUPV LQ QRUPDO UDQGRP YDULDEOHVfÂ§ HYDOXDWLRQ E\ QXPHULFDO LQWHJUDWLRQf 6,$0 6&, 67$7 &20387 YRO QR SS 'HF >@ 6KHLO DQG 2n0XLUFKHDUWDLJK f7KH GLVWULEXWLRQ RI QRQQHJDWLYH TXDGUDWLF IRUPV LQ QRUPDO YDULDEOHVf $SSOLHG 6WDWLVWLFV YRO QR SS >@ 3 ,PKRI f&RPSXWLQJ WKH GLVWULEXWLRQ RI TXDGUDWLF IRUPV LQ QRUPDO YDULDEOHVf %LRPHWULND YRO QR SS
PAGE 136
>@ 5 % 'DYLHV f1XPHULFDO LQYHUVLRQ RI D FKDUDFWHULVWLF IXQFWLRQf %LRPHWULND YRO QR SS >@ 5 % 'DYLHV f7KH GLVWULEXWLRQ RI D OLQHDU FRPELQDWLRQ RI [ UDQGRP YDULDEOHVf $SSOLHG 6WDWLVWLFV YRO QR SS >@ 6WUDQJ /LQHDU $OJHEUD DQG LWV $SSOLFDWLRQV +DUFRXUW %UDFH RYDQRYLFK ,QF UG HGLWLRQ >@ + / YDQ 7UHHV 'HWHFWLRQ (VWLPDWLRQ DQG 0RGXODWLRQ 7KHRU\ 3DUW RKQ :LOH\ t 6RQV ,QF >@ + 9 3RRU $Q ,QWURGXFWLRQ WR 6LJQDO 'HWHFWLRQ DQG (VWLPDWLRQ 6SULQJHU 9HUODJ QG HGLWLRQ >@ 7 6 5DSSDSRUW :LUHOHVV &RPPXQLFDWLRQV 3ULQFLSOHV DQG 3UDFWLFH 3UHQWLFH +DOO >@ 3URDNLV 'LJLWDO &RPPXQLFDWLRQV 0F*UDZ+LOO ,QF QG HGLWLRQ >@ % 6NODU f5D\OHLJK IDGLQJ FKDQQHOV LQ PRELOH GLJLWDO FRPPXQLFDWLRQ V\VWHPV 3DUW &KDUDFWHUL]DWLRQf ,((( &RPPXQLFDWLRQV 0DJD]LQH YRO QR SS XO\ >@ % 6NODU f5D\OHLJK IDGLQJ FKDQQHOV LQ PRELOH GLJLWDO FRPPXQLFDWLRQ V\VWHPV 3DUW ,, 0LWLJDWLRQf ,((( &RPPXQLFDWLRQV 0DJD]LQH YRO QR SS fÂ§ XO\ >@ 5 + &ODUNH f$ VWDWLVWLFDO WKHRU\ RI PRELOHUDGLR UHFHSWLRQf 7KH %HOO 6\VWHP 7HFKQLFDO RXUQDO YRO QR SS XO\$XJ >@ 8 0DGKRZ f006( LQWHUIHUHQFH VXSSUHVVLRQ IRU WLPLQJ DFTXLVLWLRQ DQG GHPRGn XODWLRQ LQ GLUHFWVHTXHQFH &'0$ V\VWHPVf ,((( 7UDQVDFWLRQV RQ &RPPXQLn FDWLRQV YRO QR SS $XJ >@ 6 / 0LOOHU f3HUIRUPDQFH RI FRGH DFTXLVLWLRQ IRU '6&'0$ RYHUOD\f LQ 3URFHHGLQJV ,((( 0LOLWDU\ &RPPXQLFDWLRQV &RQIHUHQFH 6DQ 'LHJR &$ YRO SS >@ / *RWWHVPDQ ,QWHUIHUHQFH 5HMHFWLRQ 7HFKQLTXHV WR $LG &RDUVH $FTXLVLWLRQ LQ 'LUHFW6HTXHQFH 6SUHDG6SHFWUXP 3K' GLVVHUWDWLRQ 8QLYHUVLW\ RI &DOLIRUQLD 6DQ 'LHJR >@ / *RWWHVPDQ DQG / % 0LOVWHLQ f7KH FRDUVH DFTXLVLWLRQ SHUIRUPDQFH RI D &'0$ RYHUOD\ V\VWHPf ,((( RXUQDO RQ 6HOHFWHG $UHDV LQ &RPPXQLFDWLRQV YRO QR SS 2FW >@ % 5DLQEROW 'LUHFW6HTXHQFH &RGH'LYLVLRQ 0XOWLSOH$FFHVV 2YHUOD\ 6\VWHPV 3K' GLVVHUWDWLRQ 8QLYHUVLW\ RI )ORULGD 'HF
PAGE 137
>@ 6 / 0LOOHU 0 / +RQLJ 0 6KHQVD DQG / % 0LOVWHLQ f3HUIRUPDQFH DQDOn \VLV RI 006( UHFHLYHUV IRU '6&'0$ LQ IUHTXHQF\ VHOHFWLYH IDGLQJ FKDQQHOVf VXEPLWWHG WR ,((( 7UDQVDFWLRQV RQ &RPPXQLFDWLRQV >@ + .RSND DQG 3 : 'DO\ $ *XLGH WR %7H;V 'RFXPHQW 3UHSDUDWLRQ IRU %HJLQQHUV DQG $GYDQFHG 8VHUV $GGLVRQ:HVOH\ QG HGLWLRQ >@ 0 *RRVHQV ) 0LWWOHEDFK DQG $ 6DPDULQ 7KH %7SI; &RPSDQLRQ $GGLVRQ :HVOH\ QG HGLWLRQ
PAGE 138
%,2*5$3+,&$/ 6.(7&+ 5RQDOG ) 6PLWK ZDV ERUQ LQ 3LWWVEXUJK 3HQQV\OYDQLD LQ +H UHFHLYHG KLV %DFKHORU RI 6FLHQFH GHJUHH IURP WKH 8QLYHUVLW\ RI 3LWWVEXUJK LQ )URP WKURXJK 5RQDOG ZDV HPSOR\HG E\ WKH *RYHUQPHQW 6\VWHPV 6HFWRU RI WKH +DUULV &RUSRUDWLRQ VSHFLDOL]LQJ LQ WKH GHVLJQ RI KLJKVSHHG GLJLWDO GHVLJQ FLUFXLWV IRU FRPn PXQLFDWLRQ V\VWHPV ,Q WKH IDOO RI 5RQDOG HQWHUHG WKH JUDGXDWH SURJUDP DW WKH 8QLYHUVLW\ RI )ORULGD ZLWK WKH 'HSDUWPHQW RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ ,Q 0D\ 5RQDOG UHFHLYHG WKH 0DVWHU RI (QJLQHHULQJ GHJUHH +LV LQWHUHVWV LQFOXGH WKH GHVLJQ DQG DQDO\VLV RI FRPPXQLFDWLRQ V\VWHPV IRU VSUHDG VSHFWUXP V\VWHPV
PAGE 139
, FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWn DEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 6F U$$ 6FRWW / 0LOOHU &KDLUPDQ $VVRFLDWH 3URIHVVRU RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWn DEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ B ID L8 /HRQ : &RXFK ,, &RFKDLU 3URIHVVRU RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWn DEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\MMGHDtDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI @ DQG &RPSXWHU FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWn DEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 5REHUW 0 )R[ $VVRFLDWH 3URIHVVRU RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ FHUWLI\ WKDW KDYH UHDG WKLV VWXG\ DQG WKDW LQ P\ RSLQLRQ LW FRQIRUPV WR DFFHSWn DEOH VWDQGDUGV RI VFKRODUO\ SUHVHQWDWLRQ DQG LV IXOO\ DGHTXDWH LQ VFRSH DQG TXDOLW\ DV D GLVVHUWDWLRQ IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 5LFKDUG 1HZPDQ $VVLVWDQW 3URIHVVRU RI &RPSXWHU DQG ,QIRUPDWLRQ 6FLHQFH DQG (QJLQHHULQJ
PAGE 140
7KLV GLVVHUWDWLRQ ZDV VXEPLWWHG WR WKH *UDGXDWH )DFXOW\ RI WKH &ROOHJH RI (Qn JLQHHULQJ DQG WR WKH *UDGXDWH 6FKRRO DQG ZDV DFFHSWHG DV SDUWLDO IXOILOOPHQW RI WKH UHTXLUHPHQWV IRU WKH GHJUHH RI 'RFWRU RI 3KLORVRSK\ 0D\ :LQIUHG 0 3KLOOLSV 'HDQ &ROOHJH RI (QJLQHHULQJ 0 2KDQLDQ 'HDQ *UDGXDWH 6FKRRO
PAGE 141
/' [LVU 81,9(56,7< 2) )/25,'$ O P QL PX LQ
xml version 1.0 encoding UTF8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchemainstance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID EOEBLVAS1_C21927 INGEST_TIME 20140430T16:04:42Z PACKAGE AA00020973_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES

