partial response signaling with a maximum likelihood ... · are applied to find the optimal channel...
TRANSCRIPT
,
\
PARTIAL RESPONSE SIGNALING WITH A MAXIMUM.t' .
LIKELIHOOD SEQUENCE ESTIMATION RECEIVER
by
N. Chan, B.Eng.
\
A Thesis .
Submitted to the Faculty of Graduate Studies
in Partial Fulfilmen~ of the Requirements
for the Degree
Master of Engineering
'~ ' ..
McMaster University
December 1980
MASTER OF .ENGINEERING {1980)Electrical and ~omputer Engineering
McMASTER UNIVERSITYHamilton~ Ontario
•~ .
TITLE: Partial Response Signaling with a Maximum LikelihoodSequence Estimatio~ Receiver
AUTHOR: N. Chan~ B.Eng. (Concordia University)
SUPERVISOR: Dr. J. B. Anderson
NUMBER OF PAGES: vii, 115
·...
ABSTRACT
This thesis evaluates a bandwidth efficient data transmitting
system woich is modelled as a PRS system. A maximum likelihood,
receiver implementing VA is assumed at,the receiving end. An algor
i thm is developed to compute a fundamental performance parameter of
the system, called the free distance..
99% energy bandwidth and
intersy~bol in~erference (lSI) degradation are used to measure the
performance of the system.\
\
Nonlinear progr9mming and minimax methods
are applied to find the optimal channel codes under different criteria.
Three different set~ of optimal channel codes have been found; first,
the worst~case channel codes in terms of lSI degraQation, secondly, ,~
the minimum 99% energy bandwidth channel codes and finally the mini-
mum 99% energy bandwidth channel with fixed lSI degradation constraint.
Two PRS systems with different pulse shaping filter ar~ considered.
First, an ideal low pass filter with minimum- Nyquist band\"idth is
evaluated for channel lengths up to twelve. Then a spectra1 raised~
cosine filter with roll-off factor equal to one is evaluated for~
channel lengths up to four. The two PRS systems show that a longer
channel can have better performance in- consideration of both band~
width and lSI degradation. the raised cosine filter causes no per-
formance penalty in narrow band channels.
; i i
).
ACKNOWLEDGEMENTS
The author wishes to 'thank Dr. J. B. Anderson for his. .
guidance and inspiration throughout the course of this program.
The author would also like to express ~is appreciation
from a number of useful discussions with Dr. M. R. M. Rizk.
The financial assi.stance provided by the Department of
~ Electrical and Computer Engineering is gratefully acknowledged.
Finally, the author also wishes to thank Ms. linda Hunter
for typing this thesis.
I
iv
, .
• •
· .
,
ITABLE OF CONTENTS
6
p~e
CHAPTER 1 INTRODUCTION
CHAPTER 2 ABANDWIDTH EFFICIENT SIGNALING SCHEME: PARTIAL ?RESPONSE SIGNALING
2.1 Partial Response Signaling System (PRS) 7
2.2 Choice of Pulse ~ha~ing Filter G(w) '10
2.3 Energy. and Bandwidth of PRS System 13
2.4 Finite State Machine 14
2.5 Tree and Trellis
CHAPTER '3 MAXIMUMt1RELIHOOD RECEIVER
3.1 A Nonlinear .Receiver Structure
15
17
17
3.2 Maximum Likelihood SequenGe Est}mation 20for PRS Systems .
\
3.3 Modifi€d Vjterbi Algorithm 22
\
3.4 Performance of the Receiv~r
3.5 Some Discussion
CHAPTER 4 COMPUTATION OF MINIMUM DISTANCE - dfree4.1 Represe~tation gf d?f
2 ree4.2 Properties of dfree
4.3 An'Algorithm for Computation of d~ree
4.4 Complexity and Efficiency of the ModifiedStack Algorithm
v
25
28
31<
31
33
35
38
I ~
page
CHAPTER 5 OPTIMISATIO~ TECHNIQUES
5.1 Problem Statement
43(43
5.2 The Kuhn-Tucker Conditions 45
5.3 Techniques to Solve the Constrained Problem 46, .
5.4 A Computer Package Fl~pt 5 48
~ 5.5 Application to Our Work 50
CHAPTER 6 THE MINIMUM FREE DISTANCES OF PRS SYSTEMS WITH 51CHANNEL LENGTHS UP TO 12
6.1 Normali~ed d2f and Degradation 51
'. ree6.2 Computation of the Worst Possible Degradation 53
.6.3 Analysis of the R€sults 55
CHAPTER 7 MINIMIZATION OF BANDWIDTH WITH AND WITHOUT POWER 63CONSTRAINTS
-\0
7.1 DefinitiDn of 99% Energy Bandwidth 63
7.2 Minimization of Bandwidth 66
7.3 Discussion of the Results 67
7.4 Minimization of Bandwidth with DB Constraint 73
1/
CHAPTER 8
7.5 Discussion of the Optimal Bandwidth' Channelwith DB Constraint
CONSIDERATION OF RAISED-COSINE FUNCTION FORPULSE SHAPING .
75
84
8.1 Bandwidth of a PRS System with Raised-Cosine 84Filter . I
8.2 Minimum Bandwidth of the PRS System with a' 88Raised-Cosine Filter
vi
f
page
8.3 Degradation of the PRS System
8.4 The Worst Possible Degradation
93
98
8.5 Minimization of Bandwidth with Degradation 98Constraint
CHAPTER 9 CONCLUSION 102\.t
tAPPENDIX A 107
APPENDIX B 109
""'-REFERENCES t 113
\
(l>
vii
\ '
CHAPTER 1
INTRODUCTION
1.1 A.Digital Commun~ation System
Increasing demand for faster, bandwidth efficient an~ reliable
data transmission has caused interest in various signaling schemes and
receivers to fulfil these objectives. For instance, on a satellite
communication ~ystem an~a mobile radio system, ?oth the transmission
bandwidth and signal to nciise ratio 1SNR) are of concern. Since the
deman~r communication·is increas4ng rapidly, the efficient use'of
channel bandwidth is important to reduce the channel cost directly
and conserve the radio spectrum. The immense transmission distance,
limited power supply and/or unfavourable channels of these c6mmun
ication systems )increase the.im~ortance of reliable data transmission
with manageable transmission power.
The primary sources of distortion in a high data rate trans
mission.system 'are intersymbol interference (lSI) and noise. lSI is
produced by the tail of a baseband pulse at each symbol period which
interferes with the neighbour pulses. Multipath interference has a
similar effect.
The purpose of this thesis is to design a bandwidth efficient•
transmitting system, a so called partial response signal.ing (PRS)
system or correlative encoder. A maximum likelihood (ML) receiver
employs the Viter?i Algorithm (VA) to decode the received sequence.
1
~'\
2
Fig~ 1.1 shows the digital communication system model that is used in
our work. !he inpJt data is assumed to be a sequence of impulse {O,l},
and the transmitting filter correlates the input sequence for spectral
shaping to achieve better bandwidth performance. A bandlimited channel
corrupteG by white Gausstan noise is. assumed. On the receiver1s side,
a matched filter and a maximum likelihood sequence estimator implementing'"
the V~ are used as an optimum receiver.,
Kabal and Pasupathy [lJ presented a detailed study ojY01fferent
PRS schemes from the viewpoint of spectral properties. PRS schemes are
based Dn the idea of introducing some correlation (or lSI) among the
data for spectral shaping so as to increase bandwidth efficiency.
Since the controlled amount of lSI is introduced to the adjacent
symbols in a finite discrete way, the lSI can be eliminated at the,
receiver by subtracting the previ~us weighted symbols from present
symbol.
Various receiver structures have been proposed to combat
the effect of lSI [2J-[5]~ One simple struct~re used in pract{ce is
called the Jinear equalizer. Its performance has been analysed by\
Lucky [2]. Another kind of receiver that implements a non-linear
technique, called the decision feedba~ ~qua1izer, was anglysed by
Price [3]'. Amore recent nonlinear r.eceiver was proposed by Forney
[5] which consists of a whitened matched filter, a sampler with symbol
rate and a recursive non1 inear p'rocesso'r, called the Viterbi Algorithm
(VA). The VA was devised by Vite:bi [6] as a decoding algorithm for
convolutional codes.' Omura [7J pointed out that the VA can be ~on-
.'
\\
\
\
sidered ilS a dynamic programming splution to ~ decoding problem. 3
Forn~y [3] and Kobayashi [8] . showed . that the A is ,indeed amaximum likelihood decoder performing maximum llkelihood estimation,.(MLSE) in pulse amplitude modulation (PAM) system wi h lSI. The
\performance analysis of a MLSE implementing the VA w S reported
, '
•by Forney [3J, [9J. A simil ar recei ver was proposed b Ungerboeck
[4], in which the VA was modified to deal with correla~ive noise
__directly without a pre-whitening filter. An _adapti.v.e"F-Ae~e for the
'receiving filter is ~lso proposed in the same paper. ~agee [10].- -~ --- - I'
. I
proposed another ~daptive ML receiver WhiCh'com~ines rorn~yls ~eceiver
with a cha~~el estimator so that it can be applied tQ\ an unknown slowly• • #' )
time-varying dispersive channel.
Forney showed that the probability of error of MLSE is domin
ated by the minimum distance at moderate SNR (5]. Magee and Proakis,
[llJ suggested a method to estimate the worst possible performance- .
/ over a chennel with a fixed energy finite duration pulse response.
The results for estimated worst performance were shown for channel
lengths up to ten_ Anderson and Foschini [12J later provided a
procedure for computing the minimum distance for classes of a'few
hundred states. A combined functional analysis and computer search-.
l).
approach was used to find the minimum distance.~
Their results indic-
ated that Magee's results [11] were only true for the channe1 lengths
less than seven.
An ~val'uation on power, bandwidth and complexity in MLSE
was done by Wong [13J. The transmitter was modelled as a partial
4
.r~sponse signaling system, while Forney's receiver was applied at
the receiving end. 'A double dynamic programming technique was
developed and used for camputation of the minimum dis~~nce...
Power,
bandwidth and complexity were evaluated for ~hqgnel lengths up ,to
four by plotting the corresponding contour maps.'
,, .
\1.2
I/
Organization of the Thesis
In_,Chapter 2--of this tflesi-s,~-PRS system.and it's equivalent
finite state ma.chine model is--d:i-s-c~ ...----fflChapter 3, -a ~1'1rrecei ver
proposed by Ungerboeck, [4) is presented and compared with Forney's
receiver [5].
Amodified stack algorithm i~ived in Chapter 4 for com-'. .
putation of the minimum distance.' Efficiency of the algorithm is
d~scussed and compared with the double dynamic programming method.
An optimization technique and the corresponding computer package are
introduced in Chapter 5. The computer package is used to solve the
problem of optimizing d2f . in Chapter 6 and optimizing bandwidth in- ree
Chapter 7 and 8.
Combining the\"algoritQm and technique proposed in Chapter 4
and 5, we have found the worst degradation of the MLSE for channel
lengtbs u~ to 12 in Chapter 6, while in Chapter 7, we optimize the- ----
bandwidth with and without the power constraint for channel lengths
up to 10. Finally, a spectral raised cosine function is introduced
for pulse shaping rather than the rectangu)ar spectral function used
-------iQJu.:evious chapters.. The same optimizatiqn problems in "Chapter
6 and· 7 are done for the modified system,;n Chapter 8.
)
,
.. To' ..... l"-;.
•
5
.v
."
PRe; pilter
ITransmitting I
~Channel
~-Filter I FilterI
--I
-----. -- ------- --1.
II
Input Dal;;;:>o{Xd _,
0,1 I
L
---- --- --- - --I
+ ~Noise
r-1aXimun-likeliliOO'i. Receiver
VA
ML Decexlf>.r
IIIReceiving
Filter 1
I_____ -1
II
t-t--------..
r:~tirratedd
Sequence ~t--
{Xi} I /
~I -
Firy.~.l A diqitnl communication system.0"1
"I
("
)
CHAPTER 2
A BANDWIDTH EFFICIENT SIGNALING SCHEME:PARTIAL RESPONSE SIGNALING
In communication systems, most channels are bandlimited
in some sense, so that efficient use of bandwidth is one of the
objects of a transmission system. In this chapter, partial response
signaling is introduced as a bandwidth efficient signaling scheme.
2.1 Partia'l Response Signaling System (PRS)
Lender [14] first introduced PRS for data transmission. A
PRS system is based on the allowance of a controlled amount of inter-.
symbol interference (ISI)which is used for spectral shaping so as
to redistribute the spectral energy of the signal. Most of the
energy concentrates in low frequency region or some other frequency
region depending on the correlation introduced among the signals.
Since the lSI is known, its effect can be removed or diminished at
the receiver. In comparison, a conventional pulse amplitude mod-
ulation (PAM) system eliminates lSI by creating a large number of
signal levels. A PRS system can achieve high data rate by signaling
at a higher rate with fewer levels. Therefore, PRS system will have
better error rate performance than the conventional PAM system [2J.
A PRS system can be considered as a cascade of a digitalI
7
8
transversal filter F(w) and a pulse shaping filter G(w), as shown in
Fig. 2.1 [1],
The digital transversal filter F(w) is equivalent to a shist
register with Ktaps. It correlates each input datum with L = K- 1
succeeding input data. By controlling the tap coefficients, we can
achieve the desired spectrum at the output of F(w). In our study,
the tap coefficients can be any real numbers.
The transfer function of F(w) is
I:> LF(w) = [
i=O
-jwTf. e, (2.1)
where L is number of del~y units.
f. is a tap coefficient1
T is one tap delay time which is equal to one symbol period.
Its impulse response isL ..
F( D) r f. Oi (2.2)= i=O 1
,where 0 is the Huffman's delay pperator [5J.
F(w) is a periodic function with period liT. The analog
filter G(w) ~onverts the samples YK to an analog waveform and in so~
doing bandlimits the resulting system function, hopefully without too'
mu~h change to the sample values introduced by F(O). The choice of
G(w)' will be discussed in more detail in next section.
The analogy between PRS and convolutional coding was pointed '