Performance analysis of the maximum likelihood sequence estimator for known channel impulse responses by Harold Fred Fisher A thesis submitted to the Graduate Faculty in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE in Electrical Engineering Montana State University © Copyright by Harold Fred Fisher (1973) Abstract: In a pulse amplitude modulation system, intersymbol interference is the primary detriment to reliable high rate digital transmission over narrow bandwidth high signal to noise ratio channels. The maximum likelihood sequence estimator is a receiver structure designed for use under intersymbol interference conditions. The objective of this thesis is to develop and use an analysis technique for evaluating the performance of the maximum likelihood sequence estimator for linear channels with known impulse responses. The performance criterion used is the fractional difference in minimum euclidean weight due to intersymbol interference. Discrete time signal models and flow graph theory are used to develop the analysis technique. PERMISSION TO COPY In p r e s e n tin g t h i s th e s is in p a r t i a l f u l f i l l m e n t o f th e requirem ents f o r an advanced degree a t Montana S ta te U n i v e r s i t y , L i b r a r y s h a ll make i t I agree t h a t the f r e e l y a v a ila b le f o r in s p e c tio n . I f u r t h e r agree t h a t perm ission f o r e x t e n s iv e copying o f t h i s th e s is f o r s c h o l a r l y purposes may be granted by my m a jor p r o f e s s o r , o r , o f L ib ra rie s . It in h is absence, by t h e . D i r e c t o r i s understood t h a t any copying o r p u b l i c a t i o n o f t h i s th e s is f o r f in a n c ia l g a in s h a l l not be a llo w e d my w r i t t e n p e rm is s io n . S ig n a tu r e Date v (V v if. /R / 9 ~7 3 PERFORMANCE ANALYSIS OF THE MAXIMUM LIKELIHOOD SEQUENCE ESTIMATOR FOR KNOWN CHANNEL IMPULSE RESPONSES by HAROLD FRED FISHER A th e s is subm itted to th e Graduate F a c u lt y in p a r t i a l f u l f i l l m e n t o f th e requirem ents f o r the degree of MASTER OF SCIENCE in E le c tric a l E ngineering Approved: Head, M ajor Department MONTANA STATE UNIVERSITY Bozeman, Montana August, 1973 iii ACKNOWLEDGMENT The a u th o r wishes t o thank h is committee C . K . R u s h fo r t h 9 D .A . Rudberg and M.A .F a u lk n e r f o r t h e i r guidance in h e lp in g p re p are t h i s t h e s i s . A ls o , th e a u th o r expresses h is a p p r e c i a t i o n to R.E.Leo h is re se a rch a s s i s t a n t s h i p a d v is o r f o r h is p a tie n c e s and understand ing showed d u rin g th e e n t i r e tim e spent p r e p a r in g th e t h e s i s . iv TABLE OF CONTENTS Page ii V I T A ........................ . iii TABLE OF CONTENTS iv LIST OF TABLES vi ACKNOWLEDGMENT . v ii LIST OF FIGURES ABSTRACT v iii < . . . Chapter I. IL I IN T R O D U C T IO N ............................................« CHANNEL MODELS AND THE OPTIMUM RECEIVER ■ ... DISCRETE TIME A N A L Y S I S ............................ 6 . . . 8 ........................ Tl ’ A RUNNING EXAMPLE ...................................... 14 ' PERFORMANCE ANALYSIS .................................... 21 D-TRANSFORM ................................. .... CHIP D-TRANSFORM III. 6 ERROR EVENTS . . . . . . 21 .......................................... 23 . . . . EUCLIDEAN WEIGHT FINDING:MINIMUM WEIGHT * ............... ... ............................ 24 V Chapter IV . . Page ANALYSIS USING FLOW GRAPHS ERROR STATE . . . . . . . ............................................... 27 ...................................................................... 28 MINIMUM PATH THROUGH FLOW GRAPH ............................ 29 MINIMUM EUCLIDEAN DISTANCE WITH NO INTERSYMBOL INTE RFE RE NCE ................................................................. .... FRACTIONAL DIFFERENCE IN MINIMUM WEIGHT MINIMUM DISTANCE PATH ALGORITHM V. V I. SEVERAL EXAMPLES .......................................... REFERENCES................................. ... 32 ■. 33 ......................................... . 35 . .-. 43 . 52 ....................................... . . . . ....................................... ; ................... .... CONCLUSION AND SUMMARY APPENDIX . . . . . . . ................................................ • • • • ............................................................. .... 54 65 vi LIST OF TABLES T a b le Page 1. Program O utput f o r Examples I through 5 . . . . . . 2. Program Output f o r Examples .6 through 8 ............................ 49 50 v ii LIST OF FIGURES F ig u re I* 2. ■ 3. Page P.A.M. Channel w i t h White Gaussion Noise and th e Maximum L i k e lih o o d Sequence R e c eiv er s tru c tu re . ........................................................ D i s c r e t e Time Model o f P.A.M. System 3 . . . . . . . . P.A.M. Channel w i t h White Gaussian N o is e, F i l t e r Matched t o h ( t ) , and T ra n s v e rs a l f i l t e r w i t h D i s c r e t e Time Response 1 / f (D- ^ ) . . . . . . 11 . 14 4. Channel Response t o ( a ) x Q = I , .(b) x-j = I , ( c ) X3 = I , (d ) X zl = I and ( e ) Xn = I , X1 = I , x 2 = Oi x3 = and x^ = I .................................. 15 5. C onvolutional Encoder E q u iv a le n t t o th e F l a t Top Pulse Channel ............................ ................... ... 16 6. Chip Function f o r i = 0,1 and 17 7. Flow Graph o f F l a t Top Pulse .............................. 8. Closed Loop Paths Adding Zero Weight 9. Flow Graph w i t h Minimum D is ta n c e Paths Only . . . . . . I 31 Flow Graph o f F l a t Top Pulse Example 38 10. " . " ' 2 ■ ■ . . . . . . . . . . . . . . . . . . 30 ................................. 31 . ............................ v ‘ ■ viii ABSTRACT In a pulse a m p litu d e m odulation system, intersym bol i n t e r f e r e n c e is th e p rim a ry d e t r i m e n t to r e l i a b l e high r a t e d i g i t a l tr a n s m is s io n over narrow bandwidth high s ig n a l to noise r a t i o c h a n n els . The maximum l i k e ­ lih o o d sequence e s t im a t o r i s a r e c e i v e r s t r u c t u r e designed f o r use under intersym bol i n t e r f e r e n c e c o n d i t i o n s . The o b j e c t i v e o f t h i s th e s is i s to develop and use an a n a l y s is te c h n iq u e f o r e v a l u a t i n g th e performance o f th e maximum l i k e l i h o o d sequence e s t im a t o r f o r l i n e a r channels w ith known impulse responses. The performance c r i t e r i o n used i s th e f r a c t i o n a l d i f f e r e n c e i n minimum e u c lid e a n w e ig h t due t o intersym bol i n t e r f e r e n c e . D is c r e t e tim e s ig n a l models and f l o w graph th e o r y a r e used to develop th e a n a l y s is te c h n iq u e . I. INTRODUCTION In a p ulse a m p litu d e m odulation system whenever th e in fo r m a tio n r a t e , measured in u n i t s o f pulses per second (b a u d ) , i s l a r g e r than th e bandwidth o f th e tra n s m is s io n c h a n n e l, intersym bol s u lt [ I ] . i n t e f e r e n c e may r e ­ In o t h e r words, th e pulses o v e r la p i n t o a d j a c e n t tim e and may cause an erroneous d e c is io n a t th e r e c e i v e r . i s th e p rim a ry d e t e r r e n t t o r e l i a b l e , s lo ts T h is phenomenon high r a t e d i g i t a l tra n s m is s io n over narrow bandw idth, high s i g n a l - t o - n o i s e r a t i o c h a n n e ls . te le p h o n e channels used f o r d a ta tr a n s m is s io n a r e t y p i c a l V o ic e -g ra d e examples. The o b j e c t i v e o f t h i s th e s is i s to develop and use a te c h n iq u e f o r a n a ly z in g th e performance o f th e maximum l i k e l i h o o d sequence e s t im a to r [ 2] f o r l i n e a r channels whose im pulse response i s lo n g e r than th e source symbol s e p a r a tio n . Many r e c e i v e r s t r u c t u r e s have been discussed and developed [ 3 , 4 , 5 ]. P robably th e most t a l k e d about s t r u c t u r e has been th e optimum r e ­ c e i v e r s t r u c t u r e - - t h e one t h a t makes symbol d e c is io n s based bh the e n t i r e r e c e iv e d sequence [ 5 ] . This s t r u c t u r e was not pursued, however, s in c e maximum l i k e l i h o o d c a l c u l a t i o n s in c re a se d e x p o n e n t ia l I y w ith s e ­ quence le n g th making hardware development too complex and d i f f i c u l t . As an a l t e r n a t i v e to t h i s overwhelming c o m p le x ity problem , s t r u c t u r e s were developed t h a t made sim ple symbol-by-symbol d e c is io n s [ 4 ] . I The best forms o f these r e c e i v e r s t r u c t u r e s wefe d e riv e d using c e r t a i n c r i t e r i o n o f o p t i m a l i t y , such as minimum p r o b a b i l i t y o f e r r o r , minimum p r o b a b l i t y o f e r r o r w i t h intersym bol i n t e r f e r e n c e fo r c e d to z e r o , and minimum mean-square 2 erro r. filte r In e v e ry case these r e c e i v e r s t r u c t u r e s tu rn e d o u t to be a matched in cascade w it h a tapped d e la y l i n e (tran sversal f i l t e r s ) N o n lin e a r r e c e i v e r s have a ls o been looked a t [ 6] . [3 ]. S everal optimum n o n l i n e a r s t r u c t u r e s have been d e v e lo p e d , but these tu rn e d o u t t o be v e r y complex. T h is caused th e e x p e r ts t o lo o k a t suboptimaI n o n lin e a r r e c e i ­ v ers such as d e c is io n fe edback. These were f a r too complex and d i f f i c u l t to a n a ly z e t o j u s t i f y t h e i r use. 1 . ^ i The most r e c e n t r e c e i v e r s t r u c t u r e devised f o r channels in tr o d u c in g gaussian n oise and intersym bol i n t e r f e r e n c e i s a maximum l i k e l i h o o d s e ­ quence e s t im a t o r o f th e e n t i r e r e c e iv e d sequence [ I ] . U n lik e p a s t form u­ l a t i o n s o f l i k e l i h o o d sequence e s t im a to r s i t s c o m p le x ity does not in c re a s e w i t h sequence le n g th ,' but i s p r o p o r tio n a l to where M i s th e s iz e o f th e i n p u t a lp h a b e t and L i s th e le n g th o f channel impulse response in u n i t s o f symbol s e p a r a t i o n . In f a c t t h i s r e c e i v e r can e a s i l y be im p le ­ mented and a n a ly z e d . F ig u re I i s a b lock diagram o f a P .A.M . channel w i t h a d d i t i v e w h ite gaussian n oise cascaded w i t h th e maximum l i k e l i h o o d sequence r e c e i v e r . The r e c e i v e r c o n s is ts o f a l i n e a r f i l t e r c a l l e d th e whitened matched f i l ­ t e r , a sampler ta k in g samples once e v e r y T seconds (symbol s e p a ra tio n ), and a n o n l i n e a r r e c u r s i v e a lg o r it h m c a l l e d the V i t e r b i A lg o r ith m . F ig u re I From , i=N s ( t ) =; I 1=0 6 xi h (t-iT ) (I) ; 3 n ( t ) White Gaussian 9 Noise x ,x O Sample Once Every T Seconds Channel V ite rb i s (t) i=N x (t) = I A lg o rith m Whitened Matched F ilte r x . 6( t - i T ) i =0 1 F ig u re I . r(t z0,z l P.A.M. channel w it h w h ite gaussian noise and th e maximum l i k e l i h o o d sequence r e c e i v e r s t r u c t u r e . where s ( t ) = o u tp u t o f channel whose impulse response i s h ( t ) x ( t ) = in p u t sequence I = in p u t symbol s e p a r a tio n Xi = ( i + l ) th symbol o f th e i n p u t sequence The im plem entation o f the whitened matched f i l t e r In s te a d o f having a bank o f matched f i l t e r s , s e n t , o n ly one f i l t e r s t a n t i s necessary. one f o r e v e r y in p u t symbol [ w ( - t ) ] , w ith samples taken once e v e r y symbol This lea v es the o u tp u t samples w it h a l l i n fo r m a tio n making them a s e t o f s u f f i c i e n t s t a t i s t i c s o f th e t r a n s m it t e d sequence. " ' ’ zN + L - I i s q u i t e sim ple. the necessary f o r th e e s t im a tio n In a d d i t i o n , the o u tp u t sequence, Zq , a sequence o f s t a t i s t i c a l l y d i s t r i b u t e d gaussian random v a r i a b l e s . in ­ z ^, in dependent, i d e n t i c a l l y This i s necessary to insure the simple and r e c u r s i v e p r o p e r ty o f th e V i t e r b i a lg o r it h m . The V i t e r b i a lg o r it h m was o r i g i n a l l y used to decode c o n v o lu tio n a l 4 codes. Because o f th e s i m i l a r i t y between a d a ta tra n s m is s io n channel causing intersym bol i n t e r f e r e n c e and a c o n v o lu t io n a l encoder the V i t e r b i a lg o r it h m can be used f o r maximum l i k e l i h o o d sequence e s t im a t i o n [ I ] . mentioned e a r l i e r , th e a lg o r it h m g iv es l i k e l i h o o d c a l c u l a t i o n w ith com­ p l e x i t y p r o p o r tio n a l isons and As t o M*- . To implement th e a lg o r it h m o n ly a d d i t i o n s per r e c e iv e d symbol a r e r e q u i r e d . L -I i s t e r s f o r remembering th e s u r v i v i n g path and M compar­ A ls o , reg- r e g i s t e r s f o r remem­ be rin g , th e a p p r o p r i a t e m e tr ic s a re needed [ 7 ] . Much o f th e l i t e r a t u r e d e a lin g w i t h d e t e c t io n th e o r y c l a s s i f i e s a r e c e i v e r optimum when th e p r o b a b i l i t y o f e r r o r i s m inim ized [ 2 , 3 , 8 , 9 , 10]. For th e maximum l i k e l i h o o d sequence e s t i m a t o r , a bound on the minimum p r o b a b i l i t y o f e r r o r , can be determ ined by knowing th e minimum e u c lid e a n d is ta n c e o r w eig h t o f th e o u tp u t s ig n a l space. Minimum eucludean d i s ­ tance i s th e d is ta n c e between th e two c l o s e s t s ig n a ls o r sequences o f the o u tp u t s ig n a l space. I f th e channel causes intersym bol th e minimum d is ta n c e may become s m a l l e r . in te rfe re n c e , Some channels cause a l a r g e r r e d u c t io n in d is ta n c e then o t h e r s , hence, causing a l a r g e r increase, in p ro b a b ility o f e rro r. th e system. If it This decrease d i r e c t l y reduces th e r e l i a b i l i t y o f i s known which channels y i e l d th e s m a lle s t d i f f e r e n c e s (p e r c e n t r e d u c t io n ) then communication systems can be de vis e d t h a t p ro v id e maximum r e l i a b i l i t y in th e presence o f intersym bol i n t e r f e r e n c e and a d d i t i v e w h ite gaussi an n o is e . The performance c r i t e r i a used i s th e f r a c t i o n a l d i f f e r e n c e in. minimum 5 e u c lid e a n w e ig h t. To de te rm in e t h i s f o r each c h a n n e l, minimum e u c lid e a n w e ig h t w i t h and w it h o u t inte rsy m b o l i n t e r f e r e n c e must be found. A s sociated w it h each channel impulse response th e r e i s a s e t o f e r r o r s t a t e s t h a t form nodes o f a flo w graph. These nodes a re connected i n such a way t h a t a t r a n s i t i o n from one node to a n o th e r determ ines a s in g le r e c e i v e r o u tp u t e r r o r e v e n t. F u r t h e r , e u c lid e a n w e ig h t f o r each t r a n s i t i o n can be shown t o be th e square o f a s in g le o u tp u t e r r o r e v e n t. T h is m ig h t suggest using the f lo w graph to s o lv e f o r minimum e u c lid e a n w e ig h t by f i n d i n g the s h o r t e s t path from some i n i t i a l node t o a f i n a l node. Indeed t h i s can be done and i s developed and used here in th e form o f an . XDS Extended F o r tr a n IV program as our performance a n a l y s is te c h n iq u e . As a f i r s t step toward th e flo w gra p h , a more com plete a n a ly s is o f th e w h ite n e d , matched f i l t e r w i l l be g iv e n . A ls o , an e q u i v a l e n t d i s c r e t e ­ tim e model o f th e P.A.M. channel and maximum l i k e l i h o o d sequence r e c e i v e r w ill be p r e s e n te d . T h is w i l l le a d t o d e f i n i t i o n s o f in p u t and outp u t e r r o r e v e n ts , s t a t e e r r o r e v e n ts , e u c lid e a n w e ig h t , and then f i n a l l y to ■ minimum e u c lid e a n w e ig h t. 1. II. CHANNEL MODELS AND THE OPTIMUM RECEIVER Reference to F ig u re T shows th e o u tp u t r ( t ) o f th e n o is y channel to be a sum o f two s i g n a l s , w h ite gaussian noise n ( t ) and s i g n a l ; s ( t ) given by ( I ) . It i s w e ll known t h a t d e t e c t i o n o f s ig n a ls t h a t a r e l i n e a r com- b in a t io n s o f some s e t o f b a s is s ig n a ls i s accomplished using a bank o f filte rs each matched to a ba sis s ig n a l a r e tim e t r a n s l a t i o n s o f h ( t ) [8] , (th e channel Since th e s ig n a ls o f ( I ) impulse r e s p o n s e ), o n ly one f i l t e r matched to th e channel w it h samples taken once e v e ry T seconds i s n e cessary. its L e t t i n g a ( t ) r e p r e s e n t th e o u tp u t o f h ( - t ) samples form a sequence o f numbers a ( 0) , a ( l ) , ..., (matched f i l t e r ) , a (k ), .... a(N + L - I ) . .+ O O a (k ) = k = 0, I , r(t)h (t-k T )d t —oo i+OO i=N i Now, l e t each i n t e g r a l h (t-iT )h (t-k T )d t + —oo ..., N + L - I I+Oo n (t)h (t-k T )d t —OO (2) term o f ( 2 ) be re p re s e n te d as »+oo R k -i h (t-k T )h (t-iT )d t —00 (3 ) and .+ o o n (t)h (t-k T )d t \ (4 ) Equation ( 2 ) now has a d i s c r e t e r e p r e s e n t a t i o n given by i=N 3 (k ) ^ l0 (5 ) Xi Rk - i ' + nk 7 What can be s a id about the c o r r e l a t i o n o f the n o is e samples given by (4 )? To begin w i t h , th e i n p u t n oise n ( t ) c o r r e l a t i o n fu n c t i o n o <s ( t ) where a u n i t bandwidth. s e p a ra tio n , i . e . „ i s w h ite gaussian w it h a u t o - i s th e v a lu e o f th e power over a I f th e d u r a t i o n o f h ( t ) i s l a r g e r than th e in p u t symbol i f L > I where L i s th e s m a lle s t i n t e g e r such t h a t : h ( t ) - 0 f o r t > L I , then a t l e a s t a d ja c e n t samples a r e c o r r e l a t e d . To . g e t an e x a c t measure o f th e c o r r e l a t i o n o f the n oise samples, t h e i r a u to ­ c o r r e l a t i o n c o e f f i c i e n t must be d e te rm in e d . Taking th e expected v a lu e o f th e p ro d u c t nj^nl, 0 <_ i and k <. N + L - I , gives EK ni) ■ "2rR-I R e f e r r in g to e q u a tio n ( 3 ) , because o f th e o v e r l a p . it is e v id e n t th a t A ls o , Rk_t <6> i s nonzero f o r k - l < L i s s y m m e tr ic a l, i . e . , R ^ j =. R ._ k I t was mentioned e a r l i e r t h a t th e w h ite n e d , matched f i l t e r p r o p e r t y t h a t i t s o u tp u t samples a r e s t a t i s t i c a l l y independent, has th e This i s necessary t o in s u r e th e sim ple and r e c u r s i v e n a tu re o f th e V i t e r b i a l g o r ­ B u t, e q u a tio n s ( 6 ) and ( 3 ) show t h a t th e noise sampled nk a r e c o r r ­ ith m . e l a t e d whenever L > I *, t h e r e f o r e , th e sampled outpu ts o f h ( - t ) a re n o t s t a t i s t i c a l l y independent. formed. Some w h ite n in g t r a n s fo r m a t io n must be p e r ­ From random v a r i a b l e th e o r y ] t i s known t h a t a l i n e a r t r a n s f o r - m ation on a sequence o f gaussian random v a r i a b l e s i s a ls o gaussian. th e r, if F u r­ th e sequence i s s t a t i s t i c a l l y dependent, then th e r e i s a l i n e a r tr a n s f o r m a t io n t h a t w i l l produce a s t a t i s t i c a l l y independent sequence. 8 T h is i s th e f i n a l process performed by th e w h ite n e d , matched f i l t e r and i s accomplished, by a tr a n s v e r s a l o f th e tr a n s v e r s a l filte r filte r (tapped d e la y l i n e ) . i s postponed u n t i l Discussion th e d i s c r e t e tim e model i s in tr o d u c e d . R e c all from F ig u re I t h a t th e in p u t to th e channel i s a sequence o f numbers, each separated from i t s a d ja c e n t neighbors by T seconds. As­ s o c ia te d w it h t h i s sequence ( o r any tim e sequence o f numbers) i s a formal power s e r i e s i n D. T h is s e r i e s i s c a l l e d th e D -tra n s fo rm o f x ( t ) and i s g iven by i=N x (D ) = I . x .D 1 i =0 1 (7 ) whehe x . i s a symbol chosen a t th e t r a n s m i t t e r ?rpm som§ predeterm ined I ^ v.- A a lp h a b e t ; e . g . , in th e b in a r y c a s e , Xi e q u a lSr IroK O ". - D can be thought o f as a d e la y o p e r a t o r r e p r e s e n t in g I T " u n i t s o f d e la y . Since th e o u tp u t samples o f h ( - t ) a ra tio n T, i t sample a ( k ) R .. I too has a D -tr a n s fo r m . form a sequence w it h symbol sep­ E quation ( 5 ) i n d i c a t e d t h a t each i s a f u n c t i o n o f n|J, and th e f i r s t k + I v a lu e s f o r Xi and T h is looks as i f th e D -tra n s fo rm o f th e sequence a ( 0 ) , a ( l ) , could be r e p re s e n te d in terms o f the D -tra n s fo rm s p f Xq , x-j, ..., ..., x^ ; Hq , n ^ ........... n^, and th e sequence whose c o e f f i c i e n t s a re given by ( 3 ) . Now e q u a tio n ( 5 ) has terms r e p r e s e n tin g the c o n v o lu tio n o f th e c o e f f i ­ c i e n t s Xi and Ri . Whenever two p o lynom ials a r e m u l t i p l i e d , each term o f th e product i s th e c o n v o lu tio n o f terms from each. Hence, th e D -tra n s fo rm 9 o f th e samples a ( k ) can be expressed as a (D ) = X (D )R (D ) + n ' ( D ) (8 ) where x (D ) i s th e i n p u t sequence t r a n s fo r m , n ' (D) is th e c o r r e l a t e d n o is e sample tr a n s fo r m and R(D) i s the tr a n s fo r m o f th e sequence whose c o e f f i ­ c i e n t s a r e g iven by ( 3 ) . N o t ic e , e q u a tio n ( 3 ) i s a f u n c t i o n o f th e d i f f e r e n c e k - L For e x ­ am ple, R 2 •+CO h (t-9 T )h (t-7 T )d t 9 -7 -CO i+OO = R 3-1 ~ A ls o , R ^ equals R a bout th e k - i o r i g i n . ^ h (t-3 T )h (t-T )d t making the c o rres ponding sequence symmetrical The number o f c o e f f i c i e n t s in th e sequence depends on th e amount o f o v e r l a p , o r th e v a lu e o f L . A ll t o t a l e d , t h e r e a re 2 ( L - 1 ) + I nonzero terms having a D -t r a n s fo r m . i= L -l I R(D) = i= l-L The D -tra n s fo rm R(D) w i l l (9 ) 1 be c a l l e d th e a u t o c o r r e l a t i o n fu n c t io n o r th e pulse a u t o c o r r e l a t i o n f u n c t i o n . te r m s , i t _• R1D1 has 2 L -2 complex r o o t s . Since i t c o n ta in s 2L-1 nonzero Since R(D) i s sym m etrical R(D™ ) ] an in v e r s e o f a r o o t i s a ls o a root,.. T h is makes i t [R(D) = p o s s ib le to ■' I ‘ f a c t o r R(D) i n t o a p roduc t o f p o ly n o m ia ls , f ( D ) and f ( D ). We a r e now in a p o s i t i o n t o t a l k about th e tr a n s v e r s a l filte r, filte r. T h is remember, must tr a n s fo r m th e c o r r e l a t e d noise sequence n ' ( D ) i n t o 10 But th e c o e f f i c i e n t s o f ( 6 ) a r e j u s t p those o f ( 3 ) tim es th e n o is e power pe r u n i t bandwidth (o ) ; t h e r e f o r e , a w h ite gaussian noise sequence. th e p ulse a u t o c o r r e l a t i o n f u n c t i o n o f n ' ( D ) is RJ1(D ) = O2 R(D) = o 2f ( D ) f ( D - 1 ) (1 0 ) The w h ite n o is e sequence must have a p u ls e a u t o c o r r e l a t i o n f u n c t i o n t h a t i s a c o n s ta n t. Let n ' (D) = n(D )f(D ™ 1 ) (1 1 ) where n(D ) i s our d e s ir e d w h ite n o is e sequence. E quation ( 8 ) now can be w r i t t e n as a (D ) = x ( D ) f ( D ) f (D- 1 ) + n(D )f(D ™ 1 ) (12) Passing th e sequence a (D ) through a f i l t e r w i t h response f " 1 (D™1 ) ( t r a n s ­ versal f i l t e r ) giv es z (D ) = a ( D ) / f ( D " ^ ) = x ( D ) f (D) + n(D) (1 3 ) w it h p ulse a u t o c o r r e l a t i o n fu n c t i o n f o r n(D ) equal to EWD,n(D-l)] , ^'(D)l P1(Dl )I , 2 m , ,2 (14) f ( D ) f (D ' ) where o2 i s a c o n s ta n t. Equation (1 3 ) t e l l s us t h a t th e P.A.M. channel and w h ite n e d , matched f i l t e r can be r e p re s e n te d as a d i s c r e t e time f i l t e r w it h impulse response f (D) plus an a d d i t i v e w h ite gaussian noise sequence (see F ig u r e 2 ) . n n(D ) o White Noise Sequence si In p u t Sequence V o— f (D) ■> x(D ) e O - ^ y (D ) = x ( D ) f (D) z(D ) Output Sequence F ig u re 2. D is c r e t e tim e model o f P.A.M. system. Nothing a t t h i s p o i n t has been s aid about the o v e r a l l w (t). What kind o f waveform i s i t ? w (t-k T ), k = 0, I , ...» the samples z k , k = 0 , f o r e s t im a t i n g x (D )? Does indeed the s e t o f fu n c tio n s N + L - I ,fo r m an orthonormal I, .... tim e response, N + L - l b a s i s , in s u r in g are a set o f s u f f i c i e n t s t a t i s t i c s These q ue stions can be answered by lo o k in g a t a n­ o th e r d i s c r e t e tim e t r a n s f o r m a t i o n , th e c h ip D -tra n s fo rm . Think o f the response h ( t ) as a sum o f L m u tu a lly e x c l u s iv e chips H1- ( t ) , 0 < i <_ L - I , where h ^ ( t ) = h ( t + i T ) f o r 0 £ t <_ T and zero el s e w here, i= L -l I h (t) = h .(t-iT ) i =0 Now ta k e th e D -tr a n s fo r m o f the c h ip sequence, i = 0 , I , i = L -I h (D ,t) = I 1=0 The fu n c t i o n h ( D , t ) (15) 1 ..., • h .(t)D 1 1 i s c a l l e d the c h ip D -tra n s fo rm o f h ( t ) . L - I , (16) 12 The f u n c t i o n f (D~^) i s a polynom ial d iv is io n I / f(D tim e sequence. ) in D*^ o f degree L - I . i s c a r r i e d o u t , what r e s u l t s i s p o s s ib ly an i n f i n i t e Remember, (F ig u r e I ) th e in p u t sequence Xq , i=N continuous tim e r e p r e s e n t a t i o n x ( t ) = Y x . 6 ( t - i T ) . i t 1 v e rs a ! f i l t e r I f th e x -j , had L ik e w is e , th e t r a n s ­ sequences can be g e n e r a l l y r e p re s e n te d in terms o f a w eighted sqm o f d i r a c d e l t a fu n c t i o n tim e t r a n s l a t i o n s . I =oo g ( t ) = I g 1S ( M T ) i =0 where g ( t ) is i t s Since g ( t ) Each c h ip g . ( t ) . (1 7 ) impulse response. i s a tim e f u n c t i o n , i t has a c h ip D -t r a n s fo r m . i s a w eighted d e l t a f u n c t i o n , g . 6 ( t ) . I I g (D ,t) Therefore • * = g , 6 ( I ) D i = g ( D ) 6( t ) 1 (1 8 ) . and a ls o . x (D ,t) = x ( D ) 6( t ) . Now th e w h ite n e d , matched f i l t e r filte r h (-t) i s re p re s e n te d in terms o f a matched [ c h i p D -tra n s fo rm h(D \ t ) ] t e r g(D ) = f™^(D- i ) . (1 9 ) in cascade w it h a tr a n s v e r s a l What i s th e c h ip D -tra n s fo rm o f w ( - t ) ? fil Note, t h a t h ( D , t ) and g ( D , t ) a r e fu n c tio n s o f both D and T, t h e r e f o r e , W(D- 1 S t) = h(D- 1 , t ) * g ( D , t ) = h(D- 1 , t ) * 6 ( t ) g ( D ) = h(D- 1 , t ) g ( D ) = h(D- 1 , t ) / f ( D - 1 ) ■ ( 20) 13 A s i m i l a r development shows t h a t f o r th e o u tp u t s ig n a l s ( t ) o f the chan­ n e l, s (D ,t) = h (D ,t)x (D ) (2 1 ) E quation ( 9 ) g iv e s th e pulse a u t o c o r r e l a t i o n fu n c t io n o f h ( t ) where the c o e f f i c i e n t s a re given by ( 3 ) . R(D) has a n o th e r r e p r e s e n t a t i o n in terms o f th e c h ip D -tr a n s fo r m h ( D , t ) : ■+00 R(D) = h (D ,t)h (p ™ ( 22) ,t)d t How about th e o r t h o n o r m a l i t y o f w ( t ) o r w ( D , t ) ? k = 0,- . . . , N +. L - I , r e p r e s e n t an orthonormal p ulse a u t o c o r r e l a t i o n f u n c t i o n o f w ( t ) . Does th e s e t w ( t - k T ) , basis? From (2 2 ) and ( 2 0 ) , , -I ,” h ( D , t ) h ( D ~ 1 , t ) d t w (D ,t)w (D ~ , t ) d t = J O f ( D ) f ( D “V ) jO Rw (D ) = L e t ' s look a t th e = I (2 3 ) (2 3 ) f(D )f(D " ') Rw(D) = I i n d i c a t i n g th e s e t w ( t - k T ) , k = 0, ..., N + L - I , i s o r th o n o r ­ m a l, t h e r e f o r e t h a t th e samples z k , k. = 0 , I ............ N + L - I , a re a s e t o f s u f f i c i e n t s t a t i s t i c s f o r th e e s t im a t io n o f x ( D ) . The w h ite n e d , matched f i l t e r now c o n s is ts o f a matched f i l t e r , h ( - t ) , ;■ I H t' ! ■ ' in cascade w i t h a t r a n s v e r s a l f i l t e r , f " (D“ ) ( s e e 'F i g u r e 3 ) w ith a c h ip D -tr a n s fo r m w ( D , t ) g iv en by ( 2 0 ) . 14 n ( t ) White Gaussian Noise s (t) r ( t ) Channel F ig u re 3. Matched F ilte r a (0 ),a (l) T ra n s v e rs a l F ilte r P.A.M. channel w i t h w h ite gaussian n o is e , f i l t e r matched to h ( t ) , and t r a n s v e r s a l f i l t e r w it h d i s c r e t e tim e response l / f ( D " 1 ). To g a in deeper i n s i g h t i n t o the communication system i l l u s t r a t e d by F ig u re 3 , l e t ' s work through a running example. m ission channel has an impulse response g iv en by th e f l a t pulse o f F ig u re 4 (a ). L e t ' s assume the t r a n s ­ A ls o , l e t ' s c o n s id e r the case where the symbols to be tr a n s m it te d are b in a ry . S p e c i f i c a l l y , c o n s id e r an i n p u t sequence X2 = 0 , X3 = I , and X4 = I . xq = 1 » xI = 1 • T h e r e f o r e , from (1 5 ) s (t) = h (t) + h ( t - l ) + h (t-3 ) + h (t-4 ) The channel can be thought o f as a t h r e e - s t a g e s h i f t r e g i s t e r and an adder ( F ig u r e 5 ) . 15 h (t) I O ro __ I____ I____ 1 4 5 6 7 8 6 7 8 (a) h (t-l) 2 f 0 I 2 3 1 5 h(t-3 ) 2 41 - O I 4 (c) ^ (t-4 ) s (t) 21 -■ (e) Fig u re 4 . Channel response to ( a ) x n = I , (b ) x-, = 1 , ( c ) X3 = I , (d ) X, = I and ( e ) x n = I ; X 1 = I , x , = 0 , x . = I and rI . 16 1 1 0 11 1 2 2 2 2 2 1 F ig u re 5. C o n v o lu tio n a l encoder e q u i v a l e n t to the f l a t top pulse c h an n el. White gaussian noise i s added to the s i g n a l , r ( t ) want t o e s tim a te th e tr a n s m it t e d sequence. a (k ) = = s (t) + n ( t ) . From e q u a tio n s ( 2 ) and ( 4 ) + nj. where i= 4 'k \L h (t-iT )h (t-k T )d t i =0 i= 4 h (t)h (t)d t = 3 + 2 = 5 5O % i= 4 h(t)h (t-l)d t = 2 + 3 + I = 6 si M o X1 i= 4 h ( t ) h ( t - 2)d t = 1 + 2 + 2+1 '2 M o * 1 i= 4 h (t)h (t-3 )d t = 1 + 3 S3 We now + 2 = 6 =6 17 I x. s (t)h (t-4 )d t = 2 + 3 = 5 I = O — 0° Now, ta k e th e D -tra n s fo rm acc o rd in g to (7 ), s ' (D) = 5 + 6D + 6D2 + 6D3 + 5D4 But from ( 8 ) s ' ( D ) = X (D )R (D ) L e t ' s t r y f i n d i n g R(D) by using th e above e xpression x ( D ) R ( D ) . sequence i s I I O l I, so i t s The i n p u t D -tr a n s fo r m i s x (D ) = I + D + D3 + D4 We can solve f o r R(D) using one o f two methods, e i t h e r by (2 2 ) or ( 9 ) . L e t ' s t r y both. h (D ,t). To use (2 2 ) we must s o lv e f o r th e c h ip D -tra n s fo rm From (1 5 ) h( t) = hg(t) + Ii1U - I ) + h2( t - 2 ) where h ^ ( t ) i s given by F ig u re 6 . h,(t)' I -- O O F ig u re 6 . 2 Chip f u n c t i o n f o r i = 0 , I , 3 and 2. t Now, a p p ly in g th e d e f i n i t i o n o f th e c h ip D -tra n s fo rm (1 6 ) h ( D , t ) = hQ( t ) + h -|( t)D + h2 ( t ) D = h .(t) I (I 2 + D + D^) L e t ' s solve f o r R(D) using both methods. I. From (2 2 ) h (D ,t)h (D " \t)d t R(D) 0 (I + D + D2 ) ( I + D" 1 + D '2 ) h f ( t ) d t R(D) = D" 2 + 2D™ 1 + 3 + 2D + D+2 (I 2. + D1 + O f ) ( I I -I + D™' + D"^) = f ( D ) f ( D ™ ' ) From ( 3 ) and ( 9 ) fh 2 ( t ) d t Jn R1 ■ R_i Ro = R o = '2 hence '- 2 h (t)h (t+ l)d t = 2 0 rl ■ h ( t ) h ( t + 2) d t = I Jo . R(D) = I D ™ 2 + 2D™1 + 3D + 2D + I D2 Now s' (D) i s 19 S1(D) = X (D )R (D ) = D“ 2 + 3D" 1 + 5 + 6D - 6D2 + GD3 + GD9 + 3D5 + D6 N o tic e th e e x t r a terms in s'(D). M u l t i p l i c a t i o n o f R(D) and x (D ) im p lie s c o n v o lu tio n o f th e tim e sequences. T h is assumes sampling s t a r t e d a t minus i n f i n i t y and te r m in a te d a t plus i n f i n i t y . B u t, th e in p u t sequence S1(D) s t a t e d a t tim e k = 0 and ended a t tim e k = 4 , so, o n ly those terms w here, k = 0 , ...» Now, l e t ' s 4 o f s'(D) need to be remembered. lo o k a t th e waveform w (D ,t) h .(t)(l h (D ,t) + D + D2 ) hi ( t ) I + D + D2 f (D) T h is im p lie s t h a t w ( t ) = h ^ ( t ) . N e x t, l e t ' s f i n d th e samples fo r k = 0, .... 6. /1+00 r(T )w (T -k )d t -CO hoo [s (t) + n (t)]w (t-k )d t -OO The s ig n a l p o r tio n s o f z ^ , k = 0 , ..., 6 , a re s (t)h .. ( t - k ) d t I, Z^ = 2 , z'0 = 2, z ' = 2 , z l = 2, z ' = 2 , zl Now, r e f e r t o th e d i s c r e t e tim e model o f F ig u re 2 where = I 20 y (D ) = x (D )f(D ) = (I =1 + D + D3 + D4 ) ( l + D + D2 ) + 2D + 2D2 + 2D3 + 2D4 + 2D5 + I D 6 N o t ic e , t h i s i s th e D -tra n s fo rm o f th e s ig n a l p o r tio n s o f z ^ , k = 0 , The V i t e r b i a lg o r it h m r e q u ir e s th e samples z ^ , k = 0 , I , be s t a t i s t i c a l l y independent gaussian random v a r i a b l e s . noise p o r tio n s o f th e samples a r e random, l e t ' s I+OO » + 0 0 n (t)n (s )w (t-k )w (s -j)d td s +00I +00, & (t-s )w (t-k )w (s " j)d td s J 6, Since o n ly the look a t t h e i r a u t o c o r r e l a tio n c o e ffic ie n t. E (n kn , ) ...» E (nkn . ) = 0; k M 0O O h j ( t - j ) d t = 'a ; -k = j Hence, th e n oise sequence is independent ( w h i t e ) . III. PERFORMANCE ANALYSIS Comparison o f th e o u tp u t sequence y ( D ) and in p u t sequence x (D ) sug­ gests t h a t th e f l a t - t o p in te rfe re n c e . pulse w it h L = 3T c r e a t e s severe intersym bol How severe i s th e intersym bol in te rfe re n c e . What f r a c t i o n ­ a l r e d u c t io n in th e minimum e u c lid e a n w e ig h t,d o e s th e f l a t - t o p pulse p r o ­ duce? This q u e s tio n lea d s d i r e c t l y to th e d e f i n i t i o n o f an e r r o r e vent E and i t s e u c lid e a n w eig h t d ( E ) . L e t ' s assume t h a t th e P.A.M. t r a n s m i t t e r sends th e sequence Xg = I , X1 = 1 , X2 = 0 , Xg = I , X4 = I b u t , th e r e c e i v e r e s tim a te s th e t r a n s m it t e d sequence as xO = I, X1 ^ 0, X2 = !, x3 = I, X4 = I These sequences have D -tran s fo rm s x (D ) = I + D + D3 + D4 and x ( D ) = I + D2 + D3 + D4 9 . re s p e c tiv e ly . Now, t a k in g th e d i f f e r e n c e between th e a c t u a l tr a n s m it t e d sequence and th e e s tim a te d sequence g iv e s th e i n p u t e r r o r sequence xO " x O = °» x i " x i = + T » x2 ~ x2 = 1 x 3 " x3 = 0 ’ x4 ™ x4 = 0 w it h D -tra n s fo rm Ex (D) = ( x 0 - x 0 )D ° + (X 1 - X1 ID 1 + .(x 2 - x 2 ) D2 + ( x 3 - x 3 )D 3 .+' (x 4 - %4 )D4 = 0 + D - D2 + OD3 + OD4 = D(1 - D ) = x (D ) - x (D ) 22 I f th e r e c e i v e r made a wrong e s t im a te o f x ( D ) , what about i t s y ( D ) C y (D )]? e s tim a te o f From F ig u re 2 y ( D ) = f (D )x (D ) (2 5 ) y ( D ) = f (D )x (D ) (2 6 ) hence The d i f f e r e n c e between y ( D ) and y (D ) Ey (D) = y (D ) - J (D ) is . . - A = f ( D ) [ x ( D ) - x ( D ) ] = f (D)E (D) . (2 7 ) ■ L e t ' s lo o k a t E^(D) f o r th e f l a t to p pulse [ f (D) = I + D + D^] and in p u t e r r o r sequence D - D 2 : Ey (D) = ( I + D + D2 H l - D)D = D - D^ = y (D ) - y (D ) Consider th e two sequences whose D -tran s fo rm s a re y (D ) and y ( D ) . i s a . p o s s i b l e ouput v e c t o r c onta ined in th e o u tp u t v e c t o r space. tance between th e two v e c to r s i s j u s t th e e u c lid e a n d is ta n c e The d i s ­ (th e square r o o t o f th e sum o f th e squares o f th e d i f f e r e n c e in each component). e u c lid e a n w e ig h t i s j u s t th e square o f t h i s . . Each The B u t, Ew(D) i s a polynomial y ... in D whose c o e f f i c i e n t s a r e th e d i f f e r e n c e s o f the components o f y (D ) and .• A •• - . y ( D ) , t h e r e f o r e , th e e u c lid e a n w e ig h t i s j u s t th e sum o f th e squares o f 23 th e components o f E y ( D ) , and g iven as d 2 (E ) = • j (2 8 ) Equation (2 8 ) can be o b ta in e d by a n o th e r method. Consider the p o l y ­ nomial Ey (D)Ey ( D - I ) = Ex ( D ) f ( D ) f ( D ' 1 )Ex (D‘ 1 ) = E (D )R (D )E (D"1 ) X * The D^ c o e f f i c i e n t o f t h i s polynomial (2 9 ) i s given by ( 2 8 ) , hence d 2 (E ) = [E y (D )E y (D' 1) ] 0 (3 0 ) L e t ' s de te rm in e th e e u c lid e a n w e i g h t . o f our running example. A pply­ in g ( 2 8 ) , we have d 2 (E) = 1 + 1 = 2 I s t h i s th e s m a lle s t e u c lid e a n w e ig h t p o s s ib le w ith th e f l a t top pulse? L e t ' s lo o k a t th e e u c lid e a n w e ig h t o f th e f l a t top p u ls e f o r several o t h e r i n p u t e r r o r sequences and see. L e t ' s f i r s t t r y s in g le e r r o r s I . A s in g le e r r o r : ., Ex (D) = ±Dk , 0 < k < N Ex (D) = +Dk ( l + D + D2 ) ' . • 24 d 2 (E ) = [ ( I + D + D2 ) ( + D " k ) ( + p “ k ) ( l . + D' 1 + D" 2) ] 0 = [R (D )]q = 3 N o t ic e , sin ce (+ p k )(jjD™k ) = + 1 , Id2 (E ) i s independent o f o v e r a l l d e la y and s ig n . From example ( I ) e u c lid e a n w e ig h ts . 2. above, we know s i n g l e e r r o r s do n o t g iv e s m a lle r So, l e t ' s t r y some more double e r r o r p a t t e r n s . Ex (D) = I - D E (D) = ( I y - D ) ( I + D + D2 ) = I -D 3 d 2 (E ) = I + I = 2 3. Ex (D) = D2 - D 3 Ey (D) = D2 ( I d 2 (E ) = [ ( I - D3 ) - D3 )D 2 D" 2 ( I - D“ 3) ] 0 = 2 2 As f o r th e s in g le e r r o r s , f o r double e r r o r s d ( E ) , i s independent o f o v e r ­ a ll d e la y . 4. Ex (D) = I + Dk Ew(D) = ( I y + Dk ) ( I I <_ k <_ N + D + D2 ) = I + D + D2 + Dk + Dk+1 + Dk+2 . 25 k = I d? (E ) = 1 + 22 + 2 2 + I = 1 0 k = 2 d2 (E ) = 1 + I + 2 2+ 1 + 1 = 8 k > 3 d2 (E) 5, 1 + 1+1 Ex (D) = I - Dr Ex (D) = ( I + 1 + 1 + 1=6 2 < k < N - Dk ) ( l + D + D2 ) = I + D + D2 . - Dk - Dk+1 - Dk+2 k = 2 d2 (E ) = 1 k> +1 +1 +1 =4 3 d2 (E ) = 1 + 1 + 1 + 1 + 1 + 1 = 6 L e t ' s t r y some t r i p l e e r r o r s . 6. Ev (D) = I - D + D2 A Ey (D) = I + D2 + D4 d2 (E ) = I + I + I = 3 26 7- Ev (D) = I + D + D2 . 8. 2 d (E ) = 1 + Ex (D) = 1 . . d 2 (E ) = 7 4 + 9 + 4 + 1=19 - D - D 2 IV . ANALYSIS USING FLOW GRAPHS P From th e examples g iv e n , we can guess t h a t maybe d (E) - 2 , where Ex (D) = Dk ( l - D ), is th e minimum e u c lid e a n w e ig h t f o r th e f l a t top p u ls e . B u t, we need a more r e l i a b l e method t h a t w i l l always g iv e th e minimum e u c lid e a n w e ig h t f o r an a r b i t r a r y h ( t ) d r f ( D ) . One such method i s to f i n d the s h o r t e s t path through a f l o w graph r e p r e s e n tin g th e e u c lid e a n w e ig h t d i s t r i b u t i o n o f th e o u tp u t e r r o r sequences. Ey (D) = Ex ( D ) f (D) L e t ' s ta k e a lo o k a t the i t h c o e f f i c i e n t o f Ev ( D ) , E , i = 0 , I , x y^ ..., N + L - I. Each such c o e f f i c i e n t o f th e product o f Ex (D) i s given k=m I < "y i . k=0 ~x i - k (3 1 ) k where L - I i f L - I. < i if L - I > i What we need i s a flo w graph such t h a t a p a r t i c u l a r path through th e graph w i l l g iv e us th e c o e f f i c i e n t s o f a p a r t i c u l a r o u tp u t e r r o r se­ quence. From ( 3 1 ) , th e s t a t e e r r o r , E si can be d e f i n e d . (3 2 ) i-L Knowing two successive s t a t e s Ec and Ec i th e o u tp u t c o e f f i c i e n t E , . yi , we can f i n d i+1 I f we l e t each node o f th e graph be a 28 r p o s s ib le s t a t e node E 0 , then passing through i t am ounts.to s t a r t i n g a t and p ro g re ss in g s e q u e n t i a l l y through to node E . SN+L-1 Each O l i n k between th e s t a t e E and Ec has assigned to i t si si+ l summing th e E 's f o r each path w i l l ^i E yi , th e re fo re , g iv e th e e u c lid e a n w e ig h t o f t h a t p a th . For a b in a r y i n p u t a lp h a b e t , which i s assumed thro u g h o u t th e e n t i r e p a p e r, each e lem ent o f a s t a t e ta k e s one o f t h r e e p o s s ib le values E x T h is means th e r e a r e 3 ^ a ll (3 3 ) k d i f f e r e n t , s t a t e s o f th e graph. To in s u re t h a t nonzero c o e f f i c i e n t s o f E (D) a r e d e te r m in e d , th e f i n a l th e a l l z e r o s t a t e . T h is makes a graph o f 3 and f i n a l nodes being th e a l l L- I node must be ' + I nodes, th e i n i t i a l zero s t a t e . S in c e , t r a n s i t i o n through th e graph is done s e q u e n t i a l l y , l i n k s from nodes w it h s t a t e E s B u t, s t a t e (3 2 ). ( 3 2 ) must go o n ly t o nodes w ith s t a t e (3 4 ) 1+1 M -L + l ( 3 4 ) c o n ta in s L -2 elements t h a t a r e a ls o c o n ta in e d in s t a t e The o n ly e lem ent o f (3 4 ) t h a t i s not c onta ined in »• - . Since i t '-I ■''VV - • >• (3 2 ) • 1• i s Ex . ' ^ ta k e s on o n ly t h r e e v a l u e s , t h e r e a r e th r e e l i n k s l e a v i n g each node and t h r e e l i n k s going t o each node. Now since we a r e i n t e r e s t e d 29 2 o n ly in nonzero d (E) v a l u e s , th e r e is no l i n k from the i n i t i a l th e f i n a l node to node; hence, t h e r e a r e o n ly two l i n k s le a v in g the i n i t i a l node and two l i n k s going to th e f i n a l node. To see what t h i s flo w graph looks l i k e , l e t ' s c o n s t r u c t i t f o r the f l a t top pulse w it h d i s c r e t e impulse response f (D) = I + D + D2 . L = 3 , th e r e a r e 10 nodes composing th e graph (see F ig u re 7 ) . s p e c tio n we see t h a t two s h o r t e s t paths through the graph a re d min = I + 0 + 0 + I In p u t e r r o r 0 0 , - I I 0 0 sequence and ( o j ) — ^— (T o )— > d 2min = 1 — + 0 + 0 + 1 (<h = 2 In p u t e r r o r 0 0 1 , - 1 0 0 sequence )— Since By i n ­ 30 F ig u re 7 . Flow graph o f f l a t top p ulse example. 31 Look a t th e two loops shown below in F ig u re 8 . Loop I F ig u re 8 . Closed loop paths adding zero w e ig h t. We can t r a v e l A ll Loop 2 around these paths any number o f times and no t add w e ig h t. o paths c o n ta in in g these loops have d (E ) = 2 , corresponding to m i n i ­ mum d is ta n c e paths. The flo w graph i s redrawn w it h a l l d e le te d e xc e p t minimum d is ta n c e routes (F ig u r e 9 ) . F ig u re 9. Flow graph w ith minimum d is ta n c e paths o n ly . 32 Using a f l o w graph, we a r e a b le t o f i n d th e minimum d is ta n c e f o r a g iven d i s c r e t e impulse response f ( D ) . how severe th e intersym bol th e r e i s no intersym bol B u t, t h i s s t i l l in te rfe re n c e is * in te rfe re n c e . is tra n s m itte d . We need t o f i n d d Then th e f r a c t i o n a l minimum e u c lid e a n w e ig h t can be found . s in g le symbol does not t e l l us min when d i f f e r e n c e in Consider th e case.w here o n ly a The w h ite n e d , matched f i l t e r would have to process o n ly one p u ls e , so t h e r e would be no intersym bol in te rfe re n c e . Since we a r e assuming a b in a r y a lp h a b e t , t h e r e a r e o n ly two in p u t sequences p o s s ib le X1 (D) = Dk 0 < k < N I H O O ■ I I rxT and I f th e t r a n s m i t t e r sends x-| ( D ) , but th e r e c e i v e r e s tim a te s x (D ) to be X(D) = X2 (D) i n s t e a d , then Ex (D) = x (D ) - x (D ) = X1 (D) 7 X1 (D) = Dk th e re fo re , Ey (D) = Dkf (D) and d2 (E ) = [ f ( D ) D kD"kf (D- 1 J l 0 = [ f ( D ) f ( D - 1 ) ] 0 i= L -l o (3 5 ) 33 Equation (3 5 ) would n o t change i f in s te a d th e t r a n s m i t t e r s e n t a 0 and th e r e c e i v e r made a wrong d e c i s i o n . nonzero Ex ( D ) , i t Since (3 5 ) i s t r u e f o r a l l o f th e i s th e minimum e u c lid e a n w e ig h t w it h no intersym bol in te rfe re n c e . p L e t ' s f i n d d min w ith no intersym bol top p u l s e , f (D) = I + D + D2 . ' i n t e r f e r e n c e f o r the f l a t Using (3 5 ) d2min = [ ( I = [I + D + D2 ) ( I + D™1 + D"2 ) ] 0 + 2D"1 + 3 + 2D + 1 ] Q = 3 T h is i s th e same r e s u l t o b ta in e d w i t h a s in g le in p u t e r r o r sequence w ith inte rsy m b o l in te rfe re n c e (Example I ) . What i s the f r a c t i o n a l d i f f e r e n c e i n d min due to intersym bol t e r f e r e n c e w i t h th e f l a t to p pulse? Using (3 5 ) and d2min 1 = 2 in ­ (in te r- symbol i n t e r f e r e n c e ) , (d 2min " d2min I ) / d 2min = \ • = .333 ... , ' T h is says t h a t th e f l a t to p pulse reduces the minimum e u c lid e a n w eig h t by o n e - t h i r d . i• • G e n e r a liz in g th e above f o r an a r b i t r a r y f ( D ) g iv e s th e f r a c t i o n a l d iffe re n c e 34 f d = (d 2min - d2min I ) / d 2min (36) where d^min =. minimum e u c lid e a n w e ig h t w i t h o u t intersym bol in te rfe re n c e .' d m i n , I = minimum e u c lid e a n w eig h t w it h in te rs y m b o l i n t e r f e r e n c e . Expression (3 6 ) i s th e c r i t e r i o n t h a t w i l l be used t o compare the p e r ­ formance o f our maximum l i k e l i h o o d sequence r e c e i v e r f o r s e v e ra l channel im pulse responses. Its performance can a ls o be e v a lu a te d by comparing th e upper bounds on th e minimum p r o b a b i l i t y o f e r r o r . The i n t e r e s t e d re a d e r i s r e f e r r e d to r e fe r e n c e [ 2 ] . The impulse response given as a running example was f o r the case ‘i L = 2. What about when L i s g r e a t e r than 3? W i l T a minimum d is ta n c e path through the; f l o w graph be as easy to f i n d ? I f L = 4 , t h a t means I ■ th e number o f nodes in th e graph i s t w e n t y - e i g h t . For. L = 5 , th e r e a re . 82 nodes. 3 L -I +1, • . • As L ^becomes l a r g e r , th e graph grows e x p o n e n t i a l l y as making; th e d e te r m in a tio n o f a minimum d is ta n c e path i n c r e a s i n g ­ ly d iffic u lt. T h is makes i t necessary to use computer programming to f i n d th e minimum d is ta n c e path . I t was mentioned a t th e b eginnin g o f t h i s paper t h a t i t s o b j e c t i v e 'I ‘ : i s to develop a te c h n iq u e f o r a n a ly z in g th e performance o f th e maximum . ' ' l i k e l i h o o d sequence e s t im a t o r f o r known c h a n n e ls . ■ i \ ■ : f . - ■: •; ■ ■ . Thd f l o w graph a n a l y s is .u' . , • i • s____________________ =____________ • r I 35 was in tro d u c e d as a p a r t o f t h i s o b j e c t i v e . y s i s us efu l To make th e f l o w graph a n a l - f o r a r b i t r a r y value s o f L, an XDS Extended F o r tr a n IV computer * program has been w r i t t e n t h a t s e ts up th e graph and f i n d s a minimum d i s ­ ta n c e path through i t f o r a r b i t r a r y L and f ( D ) p a i r s (channel impulse re s p o n s e s ). The minimum d is ta n c e path f i n d e r p o r t i o n o f th e program uses an a l ­ g o rith m in tro d u c e d by W. E. D i j k s t r a f o r d e te rm in in g s h o r t e s t paths through a graph [ 1 1 , 1 2 ] . L e t ' s ta k e a lo o k a t t h i s a lg o r it h m and use i t t o f i n d th e minimum d is ta n c e f o r our ru n n in g example. This p a r t i c u l a r s h o r t e s t path a lg o r it h m f i n d s th e s h o r t e s t r o u te from a given s t a r t i n g node, Ng . to some o t h e r given node in the graph* I t accomplishes t h i s by f i n d i n g th e s h o r t e s t path from t h i s node to a l l o t h e r nodes N. o f th e graph. The reason f o r doing t h i s i s t h a t any node may be an i n t e r m e d i a t e node on th e s h o r t e s t path from Ng to Nf . t h e r e i s more than one s h o r t e s t path from N^ to N^, i d e le te s a l l but one* i s a t r e e w it h N - I L (i,j) p a th ). (L (i,j) The r e s u l t i n g gra p h , a f t e r a l l f If s a lg o r it h m d e le tio n s (N i s the number o f nodes in th e f l o w graph) l i n k s i s th e path from N^ to Nj. w i t h no o t h e r nodes on t h a t Each l i n k L ( i , j ) i s th e s h o r t e s t l i n k between N^ and Nj.. As mentioned above, th e a lg o r it h m b u ild s a t r e e which c o n ta in s s h o r t e s t paths from th e s t a r t i n g node"N to a l l g in n in g , t h e r e a r e no l i n k s o f th e t r e e so a l l o th e r nodes. At the be­ l i n k s a r e no n tre e l i n k s , The a lg o r it h m then a tte m p ts to in c re a s e th e number o f t r e e l i n k s u n t i l • 36 th e re are N - I in number. To b e g in , l e t be th e a c tu a l s h o r t e s t d is ta n c e from Ng to and l e t LgJc be th e s h o r t e s t d is ta n c e from Ng to N^ using t r e e l i n k s and at, l e a s t one no n tre e l i n k . no n tre e l i n k then Lg^ = I f th e path from Ng to N^ needs more than one Now, i f L (i,k ) Njc where N^ i s c o n ta in e d i n th e t r e e , If i s the l i n k between N^ and then i t th e r e i s no l i n k between N. and N^ then d ^ has a d is ta n c e 0 I d ^ = <_ A ls o , i f node N^ i s a n e ig h b o rin g node o f th e e x i s t i n g t r e e , whose nodes a r e designated N^, then 0 <. d^^ <_ The a lg o r it h m then examines a l l n e ig h b o rin g nodes o f th e t r e e , and f o r each d e term ines th e minimum Lg^ f o r a l l N .:(3 7 ) Lsk d m| n ( Ls i + dik> I t then de te rm in e s th e minimum Lg^ f o r a l l Lsk ■ 7 N^, ' (3 8 ) ' Lsk; and makes t h i s node N^ a t r e e node w i t h d is ta n c e L (39) sp L in k L ( i , p ) now. becomes a n e w ,tr e e l i n k w it h d is ta n c e d ^ . Each tim e e q u atio n s added t o th e t r e e . (3 7 ) through (3 9 ) a r e e x e c u te d , a new l i n k i s T h is means t h a t we must R e c a lc u la te Lg^ f o r a l l n e ig h b o rin g nodes o f th e new t r e e . To do t h i s , we lo o k a t a l l n e ig h b o rin g N^ and compare Lg^ determ ined e a r l i e r w i t h i s l a r g e r th a n Lgp + dpj , th e new Lgk becomes Lgp + nodes + dp... I f dp i . I f Lgk i s Lg^ 37 s m a lle r no change i s made. This o p e r a t io n is i n d i c a t e d by LSk = " " ' " f - s k - l SP + dPkj (4 0 ) The above a lg o r it h m i s summarized using th e f o l l o w i n g s te p s . I. To s t a r t , Ng i s th e o n ly node o f th e t r e e so Lg^ = dg^ and Lgg = 0. II. The minimum o f a l l Lgk i s fo u n d , Lgp = min L 1g k , and L ( i , p ) is made a new t r e e l i n k , Lcn = Lcn. sp ■ sp III. The number o f l i n k s o f th e t r e e i s compared w i t h N - I ... If it becomes equal t o t h i s v a lu e th e a lg o r it h m i s te r m in a t e d . IV . Re-examine each Lgk found. L e t Lgk = min ( L g k , Lgp + cW V. Return to Step I I . To c a r r y o u t th e above a l g o r i t h m , each node, N i s given a la b e l (L ,i). L i s e i t h e r Lgk o r Lgk depending on whether Nk i s a t r e e node o r a no n tre e node. The l a s t node on th e s h o r t e s t path from Ng l t o Nk is d e s ig n a te d as i . p L e t ' s use th e above a lg o r it h m to f i n d d min w ith th e flo w graph o f th e f l a t top p u ls e . I t s graph i s redrawn here as F ig u re 10. has been given a d e s ig n a tio n N1- , i = 1 , 2 , . . . , • = Each node 10. " ' To b e g in , th e t r e e c o n ta in s o n ly one node N^ and no t r e e l i n k s . s i n g l e node N-j has two neighbors N^ and Ny. The We g iv e them temporary l a b e l s / I 38 F ig u r e 10. Flow graph o f f l a t top p u ls e example. 39 ( L j 4 J ) = ( d ^ . l ) = ( I 9I ) and ( L j 79I ) = ( 1 , 1 ) . Now9 min ( I 9I ) = I, but both have d is t a n c e one; t h e r e f o r e , a r b i t r a r i l y chose e i t h e r , say N7 . The t r e e now c o n s is ts o f th e l i n k L ( 1 , 7 ) o n ly . The n e ig h b o rin g nodes o f the t r e e a r e N39 N59 Ng 9 Ng9 hence; L j3 = min ( L j 39 L17 + Ci73) = min (“ , I + L j4 = min ( L j 49 L17 + Ci74) = min ( I , I + ») = I L j6 = min ( L j g 9L17 + d ^ , ) = min (» , I + 0) = I L jg = min ( L j g9 L17 + d7 g ) = min (<», I + 4) = 5 I) = 2 Since L j 4 and L j 6 a r e e q u a l , a r b i t r a r i l y chose one, say L j 4 . now has th e permanent l a b e l L ( 1 , 4 ) and L ( 1 » 7 ) . (1 ,1 ). Node N4 The t r e e c o n s is ts o f th e two l i n k s C o n t in u in g 9 th e n e ig h b o rin g nodes a r e Ng 9 N59 Ng 9 N39 Ng 9 and Ng9 t h e r e f o r e ; L j2 = min ( L j g 9L^4 + dg4 ) = min (°°, I + I ) =2 L j3 = min ( L j 39 L14 + d3 4 ) = min (2 , I + =°) = 2 L j5 = min ( L j 59 L14 + d4 5 ) = min (°°9 I + 4) = 5 L j 6 = min ( L j g 9 L14 + d4 g ) = min ( I 9 I + 00) . = ! Hs =m 1r^ (Hs' Lu + ^s 5 =m 1n, ("■ Ji+ *;’ . - ' '''' (.'.y, . L j 9 = min ( L j g9 L 14 + d4 g ) = min ( 5 , 1 + » ) = 5 - ' - ( Since L j g i s one o f th e s m a l l e s t i t L ( 4 , 8 ) becomes a t r e e l i n k . becomes a permanent l a b e l ( 1 , 4 ) and P ro ce e d in g , th e n e ig h b o rin g nodes a re now / 40 N2 , N3 , Ng, Ng and Ng w i t h d is ta n c e s L j2 = min ( L j 2 , L j3 = min ( L j 3 , L18 + dg3 ) = min (2 , I + 0) = I L j5 = min ( L j 5 , L18 + d8 5 ) = min (5 , I + ~) = 5 L j6 = min ( L j 6 , L18 + d8 6 ) = min (I, I + I) = 1 . L jg = min ( L j g , L18 + d8 g ) = min (5 , I + I) = 2 + dg2 ) Now choose L j 6 as a permenant l a b e l =min ( 2 , I (1 ,7 ). + «=) L in k L ( 6 , 7 ) l i n k and N31 N2 , N5 and Ng a r e th e n e ig h b o rin g nodes. = 2 ; becomes a t r e e Pushing on, L j 2 = min ( L j 2 , L16 + dg2 ) = min ( 2 , I + 0) = I L j 3 = min ( L j 3 , L16 + dg3 ) = min ( 2 , I +<*>) = 2 L j5 = min ( L j 5 , L16 + dg5 ) = min (5 , I + I ) = 2 L j g = min ( L j g , L16 + d6 g) = min (2, I + ~) = 2 The s m a lle s t i s L j 2 and i t becomes a permanent la b e l f i v e l i n k s o f th e t r e e w i t h fo u r l e f t t o go. (1 ,6 ). There a re C o n tin u in g , L j 3 = min ( L j 3 , L12 + d2 3 ) = min (2 , I + °°) = 2 L j 5 = min ( L j 5 , L 12 + d2 5 ) = min (2 , I + °°) = 2 L j g «? min ( L j g , L12 + d 2g) = min (2 , I + °?) = 2 L j L 10 = min ( L j 9 l 0 , Lj2 + d 2 1 0 ) = min ( » , I + I ) = 2 41 Now, p ic k a t random one o f th e f o u r as a permanent l a b e l , say L j 39 ( 2 , 8 ) . There a r e th r e e nodes l e f t , N3 , Ng and N ^ , L j 5 = min ( L j 6 , L ^3 + d3 5 ) L j g = min ( L j g , L13 hence; = min ( 2 , 2 + » + d3 g ) = min ( 2 , 2 + ~ ) =2 =2 Ljjlo = mi" (Lj.io. L13 + (I3 j l 0 ) = mf" (2. 2 + I ) = 2 Once a g a i n , randomly choose one, say L j g . th e re a re seven t r e e l i n k s w ith two t o go. ' L j 5 = min ( L j 5 , Li g I t has a l a b e l (2 ,8 ). So f a r Proceeding as b e fo re + dg 5 ) = min ( 2 , 2 + 4 ) =2 H .IO = m,n <Ll ', 1 0 \ L19 + d9,lo ) * m' n (2i 2 + ” ) = 2 L e t ' s p ic k node N15 t h i s tim e . I t s permanent la b e l have e i g h t t r e e l i n k s so we must proceed on u n t i l is (2 ,2 ). We now t h e r e a r e n in e . L1 5 = m1n 1M s 1 Ll , 1 0 + dI O 1S 1 = min (2> 2 I " ” ) = 2 Node N5 becomes a node o f th e t r e e w i t h l a b e l nine l i n k s (N - I ) ( 2, 6) . B u t, t h e r e a r e now o f th e t r e e , so th e a lg o r it h m t e r m i n a t e s . Remember, we wanted t o f i n d th e minimum d is ta n c e from th e s t a r t i n g node N1 to th e f i n a l node Ni q . Node 10 has permanent l a b e l (2 ,2 ), th e re ­ fo re min I Ll , 1 0 Using th e a lg o r it h m j u s t i l l u s t r a t e d w it h th e f l a t top pulse example we now have a te c h n iq u e f o r f i n d i n g th e minimum e u c lid e a n w e ig h t f o r an a r b i t r a r y L and f ( D ) p a i r . T h is te c h n iq u e along w i t h e q u a tio n s (3 5 ) and 42 ( 3 6 ) a r e implemented by an Extended F o r tr a n IV .program and used to a n a ly z e th e performance o f th e maximum l i k e l i h o o d sequence r e c e i v e r f o r s e v e ra l channel impulse response examples. r V. SEVERAL EXAMPLES To use the program in i t s e n t i r e t y , i t w ith v a lu e s f o r L and f ( D ) . it i s necessary o n ly to supply For exam ple, the f l a t top pulse f o r L = 3 has a d i s c r e t e tim e response f (D) = I + D + D^. needs as d a ta i s L = Al I th e program 3 and th e c o e f f i c i e n t s o f f ( D ) , f 0 = I , f , = I and f2 = I . Having developed a program t h a t f i n d s th e minimum d is ta n c e through th e e r r o r s t a t e f lo w g ra p h , l e t ' s use i t t o a n a ly z e the performance o f o u r r e c e i v e r s t r u c t u r e f o r s ev e ral examples o f channel Example Two: Q uarter-W ave Cosine Impulse Response cos h (t) h (t) = { Tit T 0 L = 3 T = I I. Pulse A u t o c o r r e l a t io n Function S o lv in g f o r C o e f f i c i e n t s : R0 = R1 - R-1 * impulse response. hi ( t ) d t = 0 cos ( ^ ) d t = 1 .5 0 ' c o s ( ^ ) cos ^ ^ d t = .867 0 < T < 3 elsewhere ** I R2 = R- 2 = / tTtT cos^ 0 COS(^ Z J l) d t = .2 5 0 R(D) = .2 5 D "2 +. .8 6 7 D "1 + 1 .5 + 867D1 + ..25D"2 2. D i s c r e t e Impulse Response L e t t i n g f (D) = a + bD + cD2 Then: R(D), = acD " 2 + (ab + be) D" 1 + ( a 2 + b2 + c 2 ) + (ab + be) D+1 + acD2 Equating C o e f f i c i e n t s : . . Rri = a 2 + b2 + c 2 + 1 .5 0 R-| = R_-j = ab + be = .867 R2 = R-2 = ac = '2 5 0 I . a 8 - a 6 + .12 6 7a 4 - A p o s itiv e real .0 6 2 5 a 2 + .00 3 9 = 0 r o o t is a = .9 6 5 0 Hence : f (D) = .9 6 5 0 + .7083D1 + . 2 5 9 1 D2 45 Example T h re e: S t r a i g h t Line Impulse Response -T /3 + I 0 <_ t <_ 3 h (t)= { h (t) elsewhere L = 3 T = I S o lv in g as in Example One: 1. Pulse A u t o c o r r e l a t i o n Function R(D) = 1 .333D " 2 + 4 .6 7 D " 1 + 8 . 9 9 9 + 4.67D + 1 .3 3 3 D 2 2. D i s c r e t e Impulse Response f (D) = 2 .5 2 8 + 1.528D + .528D2 Example Four: Decaying E xponential Function !.S e ' 1 h (t) h(t) = { 0 < t < 3 46 I. Pulse A u t o c o r r e l a t i o n Function hQ( t ) = h(t) 0 I t I = 0 elsew here h (D ,t) = (I R(D) (I (I 1 + B- 1 D + e " 2D2 )h Q( t ) + e - 1 D + e ” 2 D2 ) ( I + t f V 1 + e ' 2D ' 2 )h 2 ( t ) d t + e ' V 1 + e " 1D"2 ) ( l + C- 1 D + e " 2D2 ) f ( - D ) X f (D) 2. D is c r e t e Impulse Response f (D) = I . + Example F i v e : e " 1 D + e " 2 D2 Suppressed C a r r i e r A.M. M odula tion 0 I t I 3 elsewhere I. Pulse A u t o c o r r e l a t io n Function R(D) = .02 7 D "2 - .0 8 2 D '1 + .4 7 6 - .082D1 + .027D2 47 2. D is c r e t e Impulse Response f (D) = .6 8 0 - Example S i x : 1. .1 1 4 0 + .040D2 Voice-G rade Telephone Channel Pulse A u t o c o r r e l a t io n Function R(D) = -.194D"2 + .157D "12 + 1 .5 1 8 + .1570 - 2. [ 1 3 , 14] .194D 2 D is c r e te Impulse Response f (D) = 1 .2 1 2 5 + .1492D1 Examples Seven and E ig h t: .1599D2 E x p o n e n tia ls f o r I > 3 48 3 /2 . e I. h(t) -3 /4 t. = { O < t < 4 elsewhere O L = 4 I. R(D) 1 .1 6 4 (1 .+ e ' 3 / 4 D + e ' 3/ 2D2 + e " 9/ 4D3 ) ( l e ' 3/ 2D" 2 + e " 9/ V 2. + e ^ V 1..+ 3) f (D) = 1 .0 7 9 + 509D1 + .242D2 + .114D3 3 /2 e “ 3 t/5 II. h ( t ) ={ 0 0 < t < 5 elsewhere L = 5 1. R(D) = 1 .3 1 3 (1 + e ” 2/ 5D1 + e " 6/ 5D2 + e " 9/ 5D3 + e “ 12/ 5D4 ) ( l . e ' 2/ 5D' 1 + e ' 6/ 5D' 2 + e " 9/ 5 D™ 3 + e " 12/f5D ' 4 ) 2. f (D) = 1 .1 4 6 + .6 2 8 0 + .345D 2 + .189D3 + .104D 4 + I MPULSE KLSPONSE -V- -V- CLRAlI UN L Y Y WI TH IMERSYMfiCL Y Y interference Y interference Y Y F(D) Y V V- 3 3 vF O= YF I = YF2 = V Y Y Y Y Y * V 3 Y Y Y Y 3 B «9558 I -1537 . Y Y I • 266C Y Y Y 8.6665 Y Y Y Y »4 7 7 0 Y Dy y 2 Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y Y EX(D) V Y »3332 Y .1560 .0370 • .CCCC Y '4770 Y Y Y Y Y Y C 1-1 Y C Y Y I .1537 V V YK O= * 6 8 COY v F I = " • 11 A C Y Y f 2 = »C4C0v 2 .CCCC Y Y YFO= 1 » 0 0 CCY Y Fl = »3 6 7 9 v Y F 2 = »I 3 5 3 Y Y Y Y ;v Y YFO= 2 » b 2 7 5 v Y F l =1 * 5 2 7 5 Y v F 2 «= . 5 2 7 5 V Y Y Y Y Y Y V Y V V V V V V Y V V V V Y V V Y V V- V V V V V V Y V V V Y V V V1 V V Y V V V V Y 1*5000 Y SEQUENCE Dy v 2 * D I y y 2 Y • OCGC Y Y C 1-1 C Y Y Y Y Y Y Y Y Y Y Y Y CM C Y Y O <-« Y 3 * OCOO ERROR Y H X-* 3 C I YY2 Y Y »5650Y • 7 O8 3 Y • 2591v Y difference O 2* * V K O = Y Y Y V Y Y Dw 2 Y Y I * OOOOv v F l =1 • OOCOv YFS =1 ♦ OCOOv Y Y S. RESPONSE NO I M F R SY MBOL Y I NPUT Y FRACTI ONAL Y Y v -V -V -V- V >;■ V -Y- v- V- V V -Y V V- V V- V V V V V -V- V V V V Y Y W Y 4« Y Y Y 2• I MPULSE Y •V- I • Y Y Y - V - ELCi ICF.AN WEI GHT MI NI MUM discrete -V* O E X Y A * r V F- Y i Y * Program Output fo r Examples I through 5 O # *-* Table I . • V -V -V V 4 V -V 8 . •V V -V V V 5 I NTERSYMBOL V interference V interference V with > V D v v 2. V V V V V V V V Y Y Y V V V V V V V V V V V V V V 1*5180 V V V v- V e .OCCC 1.4581 .OCCO V V V V V V V V V V V V V I «8679 V V V V V I » 7 C2 1 ERROR V V V V V V V V V V V V V V V V V V V -V V V V V V V V SEQUENCE V Dv v S - C I v v E V V V EX(D) V D v v2 V I . Sl SC V 1*4981 FRACTI ONAL DI FFERENCE V V V VFO= I • I 4 5 0 v v-F I = » 6 2 0 0 v vF2 = * 3 4 5 0 v vF3 = . 1 8 7 0 v vF 4 = * I 0 5 0v D I vv2 ^ V V V-FO= I . 0 8 0 0 v v F I = .SlOOv v F 2 = .2420V v F3 * « 1140v * V V V V V V CTl O V V V V V V C l C C V O V V 3 V NC I NTERSYMBOL V vFO* 1 . 2 1 2 5 v v FI = . 1492v vF2 * “ • 1599v V V V • 0 8 8 8 V V V V V V V O 6• I NPUT weight O V- euclidean V A v RESPONSE v I MPULSE V Viv V P 'Y- DURATI ON V RESPONSE v V V L V -v F(D) v E •V I v v // * -V- V- V -V- -V- -V -V - -Y- V V V -V- -V- V V V V V V V V V V V -Y- 7 MI NI MUM discrete V K b* V I MPULSE O L X * Program Output fo r Examples 6 through 8 »-> 8 M- Table 2. 51 Tables I and 2 d i s p l a y th e program r e s u l t s f o r the e i g h t channel impulse response examples. The f i r s t s i x examples a r e f o r impulse responses w i t h L = 3. these s i x th e l a s t t h r e e g iv e ze ro f r a c t i o n a l e u c lid e a n w e ig h t. opes. N o tic e , a l l Of d i f f e r e n c e in minimum th r e e have e x p o n e n t ia l I y decaying e n v e l ­ Of th e s i x examples we would e x p e c t these l a s t t h r e e to g ive the b e s t performance in terms o f s m a lle s t in c re a s e in e r r o r r a t e . Since those impulse responses having decaying e x p o n e n tia l envelopes gave the best r e s u l t s , where L = 3 , v a lu e s f o r L. th e y were e v a lu a te d using l a r g e r For L = 4 (example seven) the decaying e x p o n e n tia l giv es ze ro f r a c t i o n a l d i f f e r e n c e , but f o r L = 5 th e minimum e u c lid e a n w eig h t i s reduced by a lm ost 9 p e r c e n t due t o intersym bol in te rfe re n c e . V I. CONCLUSION AND SUMMARY , A performance c r i t e r i o n and a s s o c ia te d a n a l y s is te c h n iq u es have been in tro d u c e d f o r e v a l u a t i n g th e performance o f the maximum l i k e l i h o o d se­ quence e s t im a t o r o v er l i n e a r channels w it h known channel se. impulse respon.-. Since th e minimum e u c lid e a n w eig h t o r d is ta n c e o f th e o u tp u t s ig n a l space g iv e s th e r e c e i v e r s t r u c t u r e s margin a g a in s t e r r o r when th e r e is a d d i t i v e w h ite gaussian n oise on th e c h a n n e l, the f r a c t i o n a l d iffe re n c e : in th e minimum w e ig h t g iv e s us a way o f comparing th e performance o f th e s tru c tu re fo r d iffe re n t.c h a n n e ls . When a p a r t i c u l a r channel giv es the s m a lle s t f r a c t i o n a l d i f f e r e n c e we know t h a t i t g iv es th e s m a lle s t i n c r ­ ease in e r r o r r a t e . For th e l i n e a r channels a n a ly z e d , t h i s channel would p ro b a b ly produce th e most r e l i a b l e communication, system. In a P .A.M . system, when d a ta i s t r a n s m it t e d over narrow bandwidth high s ig n a l t o n oise r a t i o channel w i t h an i n fo r m a tio n r a t e l a r g e r than th e channel bandw idth, inte rsy m bol (bauds) i n t e r f e r e n c e may r e s u l t . The maximum l i k e l i h o o d sequence e s t im a t o r was designed t o help a l l e v i a t e th is in te rfe re n c e . The o b j e c t i v e o f t h i s t h e s is i s to develop a c r i t e r i o n and a s s o c ia te d a n a l y s is te c hniques t o e v a l u a t e th e performance o f th e maximum l i k e l i h o o d sequence e s t im a t o r f o r l i n e a r channels w it h known impulse responses. . T h is r e c e i v e r s t r u c t u r e i s optimum in th e sense t h a t i t m inim izes the p ro b a b ility o f e r r o r . Since bounds on th e p r o b a b i l i t y o f e r r o r can be found by knowing th e minimum e u c lid e a n w e ig h t o f th e o u tp u t s ig n a l space. 53 th e performance c r i t e r i o n used i s th e f r a c t i o n a l d i f f e r e n c e in minimum e u c lid e a n w e ig h t o f th e o u tp u t s ig n a l space. D i s c r e t e t i m e , D -tr a n s fo r m and c h ip D -tra n s fo rm models o f th e r e c e i v e r s t r u c t u r e a re developed. From these d i s c r e t e tim e m odels, d e f i n i t i o n s o f i n p u t , o u tp u t and s t a t e e r r o r eve nts a r e g iv e n . These a r e used to c o n s t r u c t a flo w graph r e p r e s e n tin g th e e u c lid e a n w eig h t d i s t r i b u t i o n o f th e o u tp u t e r r o r e v e n ts . T h is f l o w graph along w it h a s h o r te s t path I f i n d i n g a lg o r it h m i s implemented by an XDS Extended F o r tr a n IV program and used t o f i n d th e minimum e u c lid e a n w e ig h t in th e presence o f intersym bol in te rfe re n c e . The program uses t h i s minimum e u c lid e a n w e ig h t along w i t h th e minimum w e ig h t w ith no intersym bol c a l c u l a t e th e f r a c t i o n a l d iffe re n c e . analy ze d using th e program. i n t e r f e r e n c e to S everal channel examples are APPENDIX 1' ?• 2• 4« 5. 5. 7. E. 5. IC . 11 • 12 • 12 • 14. IS* 16 * 17 * IS ­ IS* SC • SI * 22* 25 * 2H* 25* 26 * 27 * 2S* 22 * 3C * 31 * C C C C C C C C C C C C C C C C C C C C C C f * * * * * * * * * * * * * * * ¥ tf * * * * T H I S PROGRAM CALCULATES THE f I M r U M E u C L l C C A\ wE I C h T Cr THE OUTPUT Si OLENCE CF THE v A X l r U r u I K E L l H C C C SEQUENCE ESTI MATOR EOF CHANNEL I H P l LSE RESPONSES CAUSI NG I M E c SYr SOL I NT E RF E RE NCE * I T ACCOMPLI SHES T H ' S SY F I N D I N G THE SHCRTEST PATH THROUGH A FLCw CRARH REPRESENTI NG THE EUCLI DEAN WEI GHT D I S T R I B U T I O N CF JhE OUTFLT ERROR EVENTS* THE PROGRAM ALSL CALCULATES Tr-E FRACTI ONAL DI FFERENCE I N Ml N l ML H EuCLl CEAN U t I G H T DUE TC I N T E r SY^BOL I NTERFERENCE* I n p u t s tu t h e p r o g r a m are l the input s y m b o l SEPARATI ON AND T h E C O E F F I C I E N T S OF F ( C ) / THE DI SCRETE TI ME MODEL I MPULSE RESPONSE• SEVERAL L a n d F ( D ) PAI RS CA n be i n p u t t e d a t one T I ME * EACH p a i r T a k e s T wo CARDS U l T h T h E FOLLOWI NG FORMAT: I : l CARD 2 : F(J) card > format( 12) in , FORMAT(LF6*4) DI ME N S I O N F ( CU I O ) / E X ( O : 41> > L 1 ( 8 5 ) / N O D E ( 8 5 ) DI ME N S I O N L l N K T ( 8 5 ) , L I N K P ( 8 5 I , O I S T I 8 5 , 3 ) DI ME NS I ON X ( I O ) , I ASTNODE I 8 5 ) , P T R N l 8 5 ) I NTEGER E X , X , u , w , P E R M , V , T E v P , M, PT R N , U l / Wl REAL L I , L l R L2 = 0 CNT = O C Jf 2? C * * * RFAD L * * * 25 24 35 SCO I 36 27 38 35 4C 41 860 C C * 2 OOO 42 42 44 84C 48 45 SC 51 52 55 54 SE 56 57 L MAKES NC SENSE RFADF * * IN T HI S PROGRAM, t ! * READ I 1 1'5> 2 ) L / ( F I J * 1 ) / J = I > L ) FORMAT I N F G * 4) in it ia l iz e variables * * * NUN * 3 9 9 V ♦ I I F I L ‘ FO • I I GOTO 8 5 0 LE=3*9(L-2) ’ NASKl =C N A SK 3 3 CO 3 I = I z V - I N A S K= I SC I NASKz 2 J NASKl = I CR(MASKzNASKl ) 3 55 6C 61 62 62 CONTI NUE " ASK=ISC(MASKz-29( V - I ) ) P ERN=I TEMP=N=O CO 5 I =I z NUMl PTRNt I ) = L I N K T t I I = L I N K P t D = C LltI 1=1000000 64 66 ? * 9 9 * 56 65 R E A D t l C S z I z En D = I I O ) FORMAT( 1 2 ) I F ( L - I ) S40z S6C# 8 6 0 OUTPUT 1LCOK B U O z L K l GOTO H O V= L - I 6 5 CU 6 J = I Z 3 D I S T t I z J ) = 1 OOOOOO CONTI NUE CONTI NUE in Ct * * '67 65 65 7C 71 7c 73 L I ( I ) - L A S T N d O E l l >*0 IP (L •EG •L2 ) GOTO 503 2 CO h O D E (J)=G CONTINUE o o o o o n 4 DC 2 0 0 A = 1 , V X(A)=O CO 4 J = I f N U r f CONSTRUCTION CF FLOW GRAPH building the nodes * * v * CXj Il I sc. s i. Sc . SG* Si • Sb . 86 • S7 • Sg . 85 • 9C • SI • Sr • SG • 94 . 95. sc. 37 . SE • SS . * * $ 15 11 7 8 5 IC 12 13 5 C3 C I=I I F ( X( I )- I I 7 ,8,9 X(I)=I g o t o 10 X < I >=2 COTOl C X(I)=O I F l x ( I ) . NE * O ) GOTO 12 1 = 1+1 I F ( I . L E . V ) GOTOl l DO 1 3 I = 1 > V X ( I I = I S C I X I I I f E jM I - I ) ) NODE( J) = I O R I X I I J f N C D E ( J ) ) Xl I J = I S C I X I I ) f - 2 * ( I - I ) ) CONTI NUE J=J+l _ I F ( j • L E • NUM ) GOTO 15 CONTI NUE Ul "4 I OC ICl I OE I 03 104 I CS C C * 2C NODE I = NODE( I ) XODEI = I AND ( NODE I ^ nI ASKI ) NODEK = NODEt K I NODEK=I AND( I S C t N C D E K , 2 ) , M A S K l ) I F ( \ C D E K • n E • NODE I ) K = K t l J G G T C E C PTRNt I I =K CO 16 J = C , V - l MASK=I S C ( M A S K , 2 * J ) E X t J ) = I ANDt MASKi NODE( I ) ) MASK=I SC I v A S K , - E f J I E X ( J ) = IS C ( E X t J ) , =2*J) IF tE X (J ).£0.2) EX(J)=-I CONTI NUE MASK= I SCt MAS K, 2 * ( V - l ) ) E X t V I = I A N O t v ASK, NOCE( K) ) MASK=I SC( MASK, - E v ( V - I ) ) E X ( V l = I S C t E X ( V ) , - 2 v ( V - l )) I F t E X ( V ) . E C . 2) E X ( V ) = - I J =V EYJ = O DC I S H = O , J EYJ = E Y J f E X I I I I v F I J - I I I CONTI NUE F = ( KfLE-PTRNl I ) ) / L E DISTI!,PI=EYJvEYJ I F ( I • EQ » I ) NNN = EJ GOTO 19 ICg ICS 17 111 112 1 12 IG 121 122 1 22 124 1 25 126 127 I2S I2 S 13 I 3C 131 132 132 ¥ I=I K= 2 ICG 114 HS HG 117 HS HS I2C THE NCtES AND SOL v In G F:s 21 107 lie linking N N N = 3 IS I F ( P . L I * NNN) K = Kt L E J GOTO 17 I =It l I F ( I * L E • N U M - I ) GOTO 21 (TI 03 134 136 I 36 137 135 135 I4 C C C C C •C C V V V v I 4I 142 143 22 MI Sl MUM Ft-Tu F I NDE R INITIAL STEP CF ALGORITHM, v v = P T RN ( I ) TEMP- M • F=I LI ( M ) = D I S T ( I i P ) 1 4 4 L aSTnCDC(M)=I 1 45 M=MVLE P = ( Mv l E - P T R nM I ) I / L E I F I P i L E i ? ) L I N K T t M - L E ) * M< GGTC2 2 LINKT(M-LE)=O FLAC=O GOTO 65 146 147 145 145 I SC 151 152 155 154 155 156 157 152 155 I6C 161 162 163 164 165 166 167 162 ALGCRI ThM Ul v ALL NEI GHBORI NG NODES ARE LABELED (LtI) v - = n T R N ( PERM) P= (MtLE-PTRNtPERMe) )/LE IE <P •G T • 3) GOTO 45 C ^=TEMP IF(AiEGiC) IF(w.EG.M) 55 36 GOTO 30 G=PERM IF(QiEGiO) IFI-.EQiQ) 3 G0 T 0 3 5 COT 0 3 7 W=LINKT( U ) C=LINKP(Q) 37 GOTO 36 M=MvLE GOTO 40 LI NKT(M)-TEMPiTEMF = MiLASTNCDE(M)=PERMj GOTO 37 GOTO 37 VD 165.I 7C* 171. 172. * 172. 174. 175. 1 76 . 177. 178. 1/5. I SC • ISl . 162 . 185 • 1 84 . 185. 186 • 187. 188. 185. I 5 C• 191 . 195. 193. 1 94 . 195 . 1 96 . 1 97 . 196. 139 . 20C • C C C 45 .SC 51 52 6C 55 * * * =TEMP IF(WEO-O) Ll =W LSK = L S K ' FLAC = I i 71 L=LINKTIUl) I F ( L . F G . O ) GOTO 75 I F I L l ( U ) - L I (w)) 7 1 ,7 2 ,7 2 W= U Al =Ll cn o = MI N I GOTO 85 Wl = W 7C < + DPK) Is= TEPP I F ( k . E Q . C ) 3 0 TO 65 P = Wt L E - P T P N I PERN I CT = LE F=P/ DT I F ( P * E G • I * I GOTO 51 I F l p . E G . 2 . ) GOTO 51 I F I P . EG • 3 . ) GOTO Si CT= 1COOOOC• GOTO 52 CT = C I S T I P E R t f , P ) L l R = L I I PERM) +DT I F l L l I w) - L 1 R ) 5 5 # 5 5 , 6 0 Ll ( W) =LlR LASTNCOE( W) =PERM W= L I N K T I w ) GOTO SC C C C 65 LIK « • ■ » P r IM N (LLI IKK ' , L S P LSK' ¥ 72 75 85 u l= L l\K l(U l ) GOTO 70 I F ( W l . E G . W) T E r P = L I N K T ( h ) L l N K T ( Wl J = L I N K T U ' ) L I n k T( W) = O L l N K f I W) =PERM FERr =W N= N + 1 I F l N . L T . N U M - I ) GCTC 80 CONTI NUE C C C * W = L a S T n CDE * I * calculation cf Ex(D) * * * NUM) O Il ' ) 201 . 202. 202 . 2 04 . 205 . 206 • 207 . 208 . 202. 2 1 C• 211 • 212. 212 • 214. 215. 216 . 217 . 218 • 212 . 220 . 221 . 222. 222 • 224 . 225 . 226. 227 . 228 . 222 . 230 . 2 31 • 232 . 620 600 610 "ASK=I S C ( MASK, 2 » ( V - I ) ) NOOE I = NGDE ( W) NODEI = I SC ( I A N D t N O D E U MA S K ) i - 2 * ( V - D ) IF IN G D E l-I1610,610,600 NuCEI=-I E X ( J J ) =NODEI W=L a S T n ODC(W) IF (W .NE.CI C C C * 850 800 JJ = J J t l ; GOTO 6 2 0 * * CALCULATI ON CF FRACTI ONAL DI FFERENCE D M I n =C DO RCC I H = O , V DMl N = D M l N t F ( I H U F ( I H ) I F ( L - E O - I ) L I ( NUM) = DMI N D I F F = ( D M I N - L l ( NUM) ) / D M l N CALL WRI TESUS I F H < L I , N L M , DM I N , D I F F / J J , CNT / E X ) 223 • 224 . 225 . 2 36 . 237 . 2 3%. lit I COc • L2 = L GOTO 5 0 0 WRITE!1C8.1000) FORMAT ( ' I « ) STOP END SUBRCLT I NE . ' RI TESUB ( F , L > L I > NLiB , DM I N # D I FF # L J# C NT # E X ) DI MENSI ON L l i 4 C 5 ) > F ( C : i C ) / E X ( C : 5 C ) i nteger ex real LI DATA L I N E / ' * ' / I F I C N T • E G * ! I CCTD 4 0 0 .RITE 1108/323) 323 FORMAT I ' I ' ) CLTPLT • ' / • «> • i / ' I, • •>• I , I •, I • .R ITE (108,203) 30C FORM AT I 1 4 X , ' E ' , 3 X , ' I M P U L S E ' , 4 X , ' D I S C R E T E ' , 1 2 X , ' MI N I p l m e u c l i d e a n W CEI Gh T ' , 1 3 X , ' I N P U T ' ) 301 . R l TE I 1 0 8 , 3 0 2 ) 302 FORMAT I 1 4 X , ' X ' , 1 X / ' * ' , 1 C X , ' * ' , 1 C X , ' * ' , 4 4 X , ' * ' ) * . R I TE I 1 0 2 , 3 0 3 ) m 303 FORMAT I 1 4 X , ' A S l X , ' * ' , I X , ' RE S P ON S E ' , I X , ' * ' , I X , ' I M P U L S E ' , 2 X , ' * ' , 7 ^ C, ' N O ' , I 3 X , ' W I T H ' , 7 X , ' F R A C T I O N A L ' , I X , ' * ' , 2 X , ' ERROR • ) .HITE I 108,304) 304 FORMAT I l 4 X , 'Ml S I X , ' * ' , 1 C X , ' * ' , 1 0 X , * * ' , 2 X , • I n TERS y M B 0 L ' , 2 x , » * S 2 X C , ' I NTERSYMBOL S 2 X , ' * S I X , ' D I F F E R E N C E ' , I X , ' * • ) WRI TE 11 0 8 , 3 0 5 ) 305 FORMAT I I 4 X , ' P S I X , « * S I X, «CUR A T I CN S I X , ' * S I X , ' RESPONSE S I X , • * S EX c , ' i nterference • , i x , • * s ? x , ' i nterference •, i x , •* •, i s x , • * s i x , ' secuen CCE' ) WRITE!1 0 8 , 3 0 6 ) 306 307 4 CI 4 0* FORMAT! 1 4 X , ' L S 0*2',IX ,'*') WRI TE I 1 0 8 , 3 0 7 ) FORMAT! 1 4 X , ' E S C, ' 0 * * 2 S 5X, ' * S CI WRI TE! 1 0 8 , 4 0 4 ) FORMAT 11 4 X , ' F S C ,'*') I X , ' * S 1 0 X , ' * S l C X , ' * S 1 5 X , ' * S i 5 x , ' * ' , I X , 'C**2-CI* I X , ' * S 4 X , ' L S 5 X , ' * ' , 3 X , 'FID ) S 3 X , S S 6 X 5 X , ' DI * * 2 S 5 X , ' * S I X, S I X , ' * S 2 X , • FX I C ) ' I X , ' * S I C X , ' * S I C X , ' * S 1 5 X, ' * S 1 5 X , ' * S 4 X , ' C * * 2 S 4 X 35 3c 37 3S 35 . • . • • 2 C5 4C• 41 • 4 CC 402 42« 42 . 4 C3 44 . 45 . 46 • 47 . 42 . 45. bC • 4C5 4 Cb LP- SC WHI TE I ! O S , 3 0 9 ) I P , ( L I N E F u R y A T ( I 3 X , NAl ) CNT-I NI = 6 wRITE( 1 0 2 , 4 0 2 I FOKyAT I I 6 X, ' * ' , 1 CX, • * ' , w R I T E ( 1 0 S , 4 0 3 ) M , L , F ( O) FCRy ATl I P X , I P , ' . ' , I X , ' * C, ' * ' , 4 X , F 6 * 4 , 5 X , ' * ' , PX, CG 4 0 5 J - l , L - I LRI TE ( l f ; £ , 406 ) J , F ( J ) F O R y A T d O X , ' * ' , I OX , ' C F ' NI * M + 1 RETURN END >LP ) • ) - J ) ; J=C,JJI 4X,F6,4,5X ',IPX,'*') CTl 4^ /■» * - .HtiTtiaa. --it • 65 REFERENCES [1 ] Mischa S c h w a rtz, In fo rm a tio n T ra n s m is s io n , M o d u la tio n and N o is e . New Y ork: M c G ra w -H ill, 1970* Ch. 2-3. [2 ] G. David Forn ey, "M axim um -Likelihood Sequence E s tim a tio n o f D i g i t a l Sequences in th e Presences o f In tersym b o l In t e r f e r e n c e ," IEEE T ra n s. In fo rm . T h e o ry, Vol I T - 1 8 , pp. 3 6 3 -3 7 8 , May 1972. [3 ] R. W. Lucky, J . S a lz , and E. J . Weldon, O r . , P r in c ip le s o f Data Communication. New Y ork: M c G ra w -H ill, 1 9 6 8 , Ch 5 . [4 ] Donald W. T u f t s , " N y q u is t's Problem — The J o in t O p tim iz a tio n o f Trans m i t t e r and R e c e iv e r in Pulse A m plitude M o d u la tio n ," P ro c . IEEE, V o l. 5 3 , pp. 2 4 8 -2 5 9 , Mar. 1965. [5 ] C. W. H e !s tro m , S t a t i s t i c a l Pergamon, 1960, S e c t. I V . 5. [ 6] R. W. Chang and R. C. Hancock, "On R e c e iv e r S tru c tu re s f o r Channels Having Memroy," IEEE T ra n s . In fo rm . T h e o ry, V o l. I T - 1 2 , pp. 4 6 3 4 6 8 , O ct. 1966. [7 ] Andrew 0 . V i t e r b i , "C o n v o lu tio n a l Codes and t h e j r Performance in Communication System s," IEEE T ra n s . Commun. T e ch n o !. , V o l. COM-19, O ct. 1 9 7 1 ^ pp. 7 4 1 -7 7 1 . [ 8} 0 . M. W ozG ncraft and I . M. Jacobs, P r in c ip le s o f Communication E n g in e e rin g . New Y ork: W ile y , 1965. [9 ] D. J . S a k fis o n , Communication T h e o ry: Transm ission o f Waveforms and D i g i t a l In fo r m a tio n . New Y o rk: John W ile y and Sons, I n c . , 1 9 6 8 , Ch. 7 . Theory o f S ignal D e te c tio n . New Y ork: [ 1 0 ] Jim K. Omuira, "O ptim al R e c e iv e r Design f o r C o n v o lu tio n a l Codes and Channels w ith Memory V ia C ontrol T h e o re tic a l C oncepts," In fo rm a tio n S c ie n c e s , V o l. 3 , 1 9 7 1 , pp. 2 4 3 -2 6 6 . [1 1 ] T . C. Hu, In te g e r Programming and Network Flow s., Add 1son-W esley P u b lis h in g Company, 1 9 6 9 , Ch. 10. R eading , Ma s s . : [1 2 ] E. W. D i j k s t r a , "A Note on Two Problems in Connection w ith G raphs," Num. M a th ., V o l. I , 1 9 5 9 , pp. 2 6 9 -2 7 1 . [ 1 3 ] J . R. Davey, "Modems," P ro c . o f th e IE E E , V o l. 6 0 , No. 1 1 , Nov. 1 97 2 , pp. 1 2 8 9 -1 2 9 0 . 66 [ M ] . F. P. D u ffy and I . .W. T h a tc h e r, J r . , "Analog T ransm ission P e r fo r ­ mance on th e Sw itched Telecom m unications N e tw o rk ,". B e ll System T e ch n ic a l J o u r n a l, V o i. 5 0 , No. 4 , A p r i l , 1971, pp. .1 3 1 1 -1 3 9 7 , MONTANA STATE UNIVERSITY LIBRARIES 0013707 2 1762 I • - F533 cop.2 • Fisher, Harold F Performance analysis of the maximum likeli­ hood sequence estimator mams and aooaksk ^ A UM M vQrd.-t<.v ^vw , /I ^ 7 / /- _> /-Xvfi. College Floce Iindery