Chapter 5: Applications using S.D.E.’s Channel state-estimation State-space channel estimation using Kalman filtering Channel parameter identificationa Nonlinear filtering Power control for flat fading channels Convex optimization and predictable strategies Channel capacity Optimal encoding and decoding Chapter 5: Linear Channel State-Estimation The various terms of the state-space description are: X I (t ) AI (t ) X (t ) , A(t ) X Q (t ) 0 0 BI (t ) , B(t ) , AQ (t ) BQ (t ) BI (t ) BQ (t ) 0 k (t ) T G (t ,t ) cos c t D cos c t 0 sin c t 0 s (t t ) sin c t f I (t ) F t : specular components f Q (t ) I (t ) 1 0 0 0 Q(t ) 0 0 1 0 X (t ) Note that the parameters depend on the propagation environment represented by t Chapter 5: Channel Simulations First must find model parameters for a given structure Method 1: Approximate the power spectral density (see short-term fading model) Method 2: From explicit equations and data we have X (t ) e X (0) e A(t t ) BW (t )dt ; E X (t ) e At E X (0) t At 0 r (t ) I 2 (t ) Q 2 (t ); E r (t ) 2 (t ) E I 2 (t ) (t ) E X (t ) X T (t ) E I (t ) I (t ) E I (t ) I (t ) 2 E I (t ) From data: lim E r (t ) 2 X 2 lim E I 2 (t ) t Obtain {k,z,n} parameters t Chapter 5: Channel Simulations Flat-fading channel state-space realization in statespace + dWI ABCD X cos ct dWQ + ABCD Flat-fading channel X sin ct Chapter 5: Linear Channel State-Estimation State-Space Channel Estimation using Kalman filtering Considering flat-fading X (t ) A(t ) X (t ) F (t ) B(t )W (t ), X (0) N ( , ) y (t ) G (t ) X (t ) D(t )v(t ) W (t ) N (0, Q), (t ) N (0, R), W (t ), (t ) independent and also independent with X (0) Kalman filter for the state estimate is given by Xˆ (t ) A(t ) Xˆ (t ) F (t ) K (t ) y (t ) G (t ) Xˆ (t ) K (t ) standard Kalman filter gain Chapter 5: Linear Channel State-Estimation State-Space Channel Estimation using Kalman Filtering Inphase and Quadrature estimates Iˆ(t ) E I (t ) y ( s);0 s t CI Xˆ (t ) Qˆ (t ) E Q(t ) y ( s );0 s t CQ Xˆ (t ) Square-Envelop Estimate rˆ 2 (t ) Iˆ 2 (t ) Qˆ 2 (t ) e 2 (t ) e 2 (t ) I Q 2 2 2 ˆ ˆ e I (t ) E I (t ) I (t ) , eQ (t ) E Q(t ) Q(t ) Possible generalization to multi-path fading channel 2 N y (t ) Gi (t )X i (t ) D(t )v(t ) i 1 Chapter 5: Channel State-Estimation: Simulations Flat Fast Rayleigh Fading Channel, SNR = 10 dB, v = 60 km/h -14 2 2 1 1 -18 0 0 -20 -1 -1 SD(f) H(j) S (f), H(j)2 in dB D 2 -16 -22 -5 -4 -3 -2 -1 0 1 2 3 4 -2 5 0 50 100 150 200 [a] inphase. Time [ms] [a] v = 5 km/h, fc = 910 MHz, m = 400 ; Frequency Hz. -42 -2 250 10 0 50 100 150 200 [b] quadrature. Time [ms] 250 0 50 250 2 SD(f) H(j) -44 0 1 tan-1 [Q(t)/I(t)] S (f), H(j)2 in dB D 2 [dB] -10 -46 -20 -48 -50 -150 -1 -30 -100 -50 0 50 100 -40 150 0 50 100 150 200 [c] envelope. Time [ms] [b] v = 120 km/h, fc = 910 MHz, m = 400 ; Frequency Hz. -2 100 150 200 [d] phase. Time [ms] inphase MSE quadrature MSE -10 [dB] 1 0 -1 inphase estimate -20 -30 -40 -50 0 50 100 150 [a] Time [ms] 200 250 0 0.5 1 1.5 2 2.5 [a] Time [ms] 3 3.5 4 20 2 0 [dB] 1 0 -1 -2 250 0 2 -2 0 quadrature estimate -20 -40 -80 0 50 100 150 [b] Time [ms] 200 r2 -60 250 r2 estimate 0 50 100 150 [b] Time [ms] 200 250 Chapter 5: Channel State-Estimation: Simulations Frequency-Selective Slow Fading, SNR=20dB, v=60km/h 6 6 signal noiseless signal 4 4 2 2 0 0 1 2 [a] Time [ s] 3 8 0 0 1 2 [b] Time [ s] 3 6 noisy signal signal estimate 6 4 4 2 2 0 0 1 2 [c] Time [ s] 3 0 0 1 2 [d] Time [ s] 3 Chapter 5: Channel state-estimation: Conclusions For flat slow fading, I(t), Q(t), r2(t) show very good tracking at received SNR = -3 dB. For flat fast fading, I(t), Q(t), r2(t) show very good tracking when the received SNR = 10 dB. For frequency-selective slow fading, I(t), Q(t), r2(t) of each path show very good tracking, w.r.t. MSE, when the received SNR = 20 dB. Chapter 5: Channel state-estimation: References J.F. Ossanna. A model for mobile radio fading due to building reflections: Theoretical and experimental waveform power spectra. Bell Systems Technical Journal, 43:2935-2971, 1964. R.H. Clarke. A statistical theory of mobile radio reception. Bell Systems Technical Journal, 47:957-1000, 1968. M.J Gans. A power-spectral theory of propagation in the mobile-radio environment. IEEE Transactions on Vehicular Technology, VT-21(1):27-38, 1972. T. Aulin. A modified model for the fading signal at a mobile radio channel. IEEE Transactions on Vehicular Technology, VT-28(3):182-203, 1979. C.D. Charalambous, A. Logothetis, R.J. Elliott. Maximum likelihood parameter estimation from incomplete data via the sensitivity equations. IEEE Transactions on AC, vol. 5, no. 5, pp. 928-934, May 2000. C.D. Charalambous, N. Menemenlis. A state-state approach in modeling multi-path fading channels: Stochastic differential equations and OrnsteinUhlenbeck Processes. IEEE International Conference on Communications, Helsinki, Finland, June 11-15, 2001. Chapter 5: Channel state-estimation: References K. Miller. Multidimensional Gaussian Distributions. John Wiley & Sons, 1963. M.S. Grewal, A.P. Andrews. Kalman filtering – Theory and Practice, Prentice Hall, Englewood Cliffs, New Jersey 07632, 1993. D. Parsons. The mobile radio Propagation channel. John Wiley & Sons, 1995. R.G. Brown, P.Y.C. Hwang. Introduction to random signals and applied Kalman filtering: with MATLAB exercises and solutions, 3rd ed. John Wiley, 1996. G. L. Stuber. Principles of Mobile Communication. Kluwer Academic Publishers, 1997. P. E. Kloeden, E. Platen. Numerical Solution of Stochastic Differential Equations. Springer-Verlag, New York, 1999. Chapter 5: Channel Parameter Identification Consider the quasi-static multi-path fading channel model y (t ) ri cos c t i (t;t i ) sl (t t i ) (t ) K (t ) i 1 : noise i (t ;t ) i ( c di )t di t Given the observation process for each path find estimates for the channel parameters: r , i di , i : channel gain, Doppler spread and phase Chapter 5: Non-Linear Filtering-Sufficient Statistic Methodology: Use concept of sufficient statistics in designing nonlinear channel parameter estimator. Sufficient statistic: any quantity that carries the same information as the observed signal, i.e. conditional distribution. Chapter 5: Bayes’ Decision Criteria Detection criteria Chapter 5: Non-Linear Filtering Sketch of continuous-time non-linear filtering for parameter estimation. Derive a sufficient statistic and obtain the incomplete data likelihood ratio of multipath fading parameters (for flat and frequency selective channels) One parameter at a time while keeping others fixed, All parameters simultanously Consider the band-pass representation of the received signal y (t ) ri (t ) cos ( c di (t ))(t t i (t )) i S (t t i ) (t ) K i 1 h(t , x(t )) (t ) Chapter 5: Non-Linear Filtering Sketch of continuous-time non-linear filtering approach dx(t ) f t , x(t ) dt t , x(t ) dw(t ), x (0) x0 : state process dy (t ) h t , x(t ) dt N t dv (t ) : observation process w(t ) and v(t ) independent Brownian motions ˆ x(t ) E x(t ) ( z ) p t , z dt 0,t N pN t , z : normalized conditional density of x(t ) given 0,t y ( s );0 s t : observable events ˆ x(t ) : Least-squares estimate of x(t ) 0,t Non-linear filtering theory relies on successful computation of pN(.,.) Chapter 5: Non-Linear Filtering Continuous-time non-linear filtering Radon-Nikodym derivative (complete data likelihood ratio) dP dP Ft t exp hT ( s, x( s ))( N (t ) N T (t ))dy ( s ) 0 1 t T h ( s, x( s))( N (t ) N T (t )) h( s, x( s)) ds 2 0 (t ) where dx(t ) f t , x(t ) dt t , x(t ) dw(t ), x (0) x0 (, Ft , P ) dy (t ) N t dv(t ) (t ) : Complete data likelihood function Chapter 5: Non-Linear Filtering Continuous-time non-linear filtering; Bayes’ rule dP E x(t ) 0,t dP E x(t ) 0,t dP E 0,t dP n ( z ) p (t , z )dz n p (t , z )dz E : expectation under P p (, ) : unnormalized conditional density : sufficient statistic satisfies the forward Kolmogorov equation dp(t , x) L(t ) p (t , x)dt hT (t , x)( N (t ) N T (t )) 1 dy (t ), (t , x) 0, T p (0, x) p0 ( x), x n where 1 2 L(t ) (t ) Tr 2 ( (t , x) T (t , x)(t )) ( f (t , x)(t )) 2 x x n Chapter 5: Phase Estimation Problem 1: Flat-fading; phase estimation Given the observation process dy (t ) r (t ) cos ( c d (t ))(t t (t )) S (t t (t )) N t dv(t ) h t , , (t ) dt N t dv(t ) A fixed sample path (t ) r ( s, ), d ( s, );0 s t 1 , find 2 1. pN (t , ) and p(t , ) normalized and unnormalized conditional densities and : 0, 2 , with a priori density p0 ( ) given the sample path (t ) r ( s, ), d ( s, );0 s t ; ˆ (t ) 2. dP dP , for a fixed sample path (t ); 0,t 3. hˆ(t , ) E h(t , , (t )) 0,t , for a fixed sample path (t ); 4. ˆ(t ) E 0,t , for a fixed sample path (t ). Chapter 5: Phase Estimation Defintion: Flat-fading; phase estimation problem For t [0, Ts ], let z (t ) Vc (t ) Vs (t ) d y (t ) and define dt r (s) cos ( s (s))(s t (s)) S ( s t ( s))z ( s)ds r (s) sin ( s ( s))( s t ( s)) S ( s t ( s))z( s)ds t c 0 d t c 0 d Vs (t ) V (t ) V (t ) V (t ), (t ) tan V ( t ) c 1 t 2 Wc (t ) r ( s ) cos 2( c s d ( s ))( s t ( s )) S 2 ( s t ( s))N 2 ( s)ds 4 0 1 t 2 Ws (t ) r ( s ) sin ( c s d ( s))( s t ( s)) S 2 ( s t ( s))N 2 ( s) ds 4 0 2 2 1 Ws (t ) W (t ) Wc (t ) Ws (t ), (t ) tan W ( t ) c 2 c 2 s 1 Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation problem The unnormalized conditional density of given 0,t and the sample path (t ) r ( s, ), d ( s, );0 s t is given by 1 t 2 p (t , ) p0 ( ) exp r ( s ) S 2 ( s t ( s )) N 2 ( s )ds 4 0 exp W (t ) cos(2 (t )) exp V (t ) cos( (t )) 1 where p0 ( ) is the a priori density of , p0 ( ) , [0, 2 ] 2 and W (t ), V (t ), (t ), (t ) as above. Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation problem ˆ (t ), is given by The incomplete data likelihood ratio, 2 ˆ (t ) p (t , )d 0 1 t 2 exp r ( s) S 2 ( s t ( s)) N 2 ( s)ds exp W (t ) cos (t ) 4 0 (1) j 2 j k 1 h i V (t )W j k (t ) cos h (t ) sin i (t ) cos j (t ) sin k (t ) h ,i , j , k 0 h !i ! j ! k ! (h k )!(i 2 j k )! h i 2 j 2k i 2 j k h k h i 2 j 2 k ! ! !2 2 2 2 and W (t ), V (t ), (t ), (t ) as above. Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation problem The normalized conditional density, pN (t , ), is given by 1 1 t 2 pN (t , ) exp r ( s )S 2 ( s t ( s )) N 2 ( s )ds ˆ (t ) 2 4 0 exp W (t ) cos(2 (t )) exp V (t ) cos( (t )) 1 t 2 exp r ( s ) S 2 ( s t ( s )) N 2 ( s) ds exp W (t ) cos (t ) 4 0 and W (t ), V (t ), (t ), (t ) as above. The conditional least-squares estimate hˆ(t , ) is given by Chapter 5: Phase Estimation hˆ(t , ) 2 0 h(t , ) p (t , ) d 2 0 p (t , ) d 1 r (t ) cos c t S (t t (t )) exp W (t ) cos (t ) ˆ 2 (t ) (1) j 2 j k 1 h i V (t )W h ,i , j , k 0 h !i ! j ! k ! j k (t ) cos h (t ) sin i (t ) cos j (t ) sin k (t ) ( h k 2) 2 (i 2 j k 1) 2 ( h i 2 j 2k 3) 2 r (t ) sin c t S (t t (t )) exp W (t ) cos (t ) (1) j 2 j k 1 h i V (t )W h ,i , j , k 0 h !i ! j ! k ! j k (t ) cos h (t ) sin i (t ) cos j (t ) sin k (t ) ( h k 1) 2 (i 2 j k 2) 2 ( h i 2 j 2k 3) 2 and W (t ), V (t ), (t ), (t ) as above. Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation problem The least-squares estimate ˆ(t ) is given by 2 2 0 0 p(t , )d p(t , )d ˆ (t ) p ( t , ) d ˆ where (t ) p(t , ) d is computed as in theorem 2 ˆ(t ) 2 0 2 0 Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation Neglecting double frequency terms The unnormalized conditional density of given 0,t and the sample path (t ) r ( s, ), d ( s, );0 s t is given by 1 t 2 p (t , ) p0 ( ) exp r ( s ) S 2 ( s t ( s )) N 2 ( s )ds 4 0 exp V (t ) cos( (t )) 1 where p0 ( ) is the a priori density of , p0 ( ) , [0, 2 ] 2 and W (t ), V (t ), (t ), (t ) as above. Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation Neglecting double frequency terms ˆ (t ), is given by The incomplete data likelihood ratio, 2 ˆ (t ) p(t , )d 0 1 t 2 exp r ( s ) S 2 ( s t ( s )) N 2 ( s )ds I 0 V (t ) 4 0 where V (t ), as above and I 0 is the modified zero order Bessel function, 1 I0 x 2 exp x cos d . Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation Neglecting double frequency terms The normalized conditional density of given 0,t and the sample path (t ) r ( s, ), d ( s, ); 0 s t is given by pN (t , ) exp V (t ) cos( (t )) I 0 V (t ) The conditional least-squares estimate hˆ(t , , (t )) is given by I1 V (t ) ˆ cos ( c d (t ))(t t (t )) (t ) S (t t (t )) h(t , , (t )) r (t ) I 0 V (t ) where I 0 is the modified zero order Bessel function, and I1 is the modified first order Bessel function, 1 I1 x 2 cos exp x cos d . Chapter 5: Phase Estimation Solution of Problem 1: Flat-fading; phase estimation Neglecting double frequency terms The conditional least-squares estimate ˆ(t ) is given by ˆ(t ) 2 0 3 1 2 ( 1 2 k ) (t ) 2k pN (t , )d V (t ) 2 2 k 1 I 0 (V (t )) k 0 k !(2k )! (k !) 2 Chapter 5: Channel Estimation Same procedure for Gain Doppler Spread Joint Estimation of Phase, Gain, Doppler Spread Frequency Selective Channels Chapter 5: Simulations of Phase Estimation Phase estimation in continuous-time Chapter 5: Nonlinear Filtering Conclusions Conditional density is a sufficient statistic. Explicit but complicated expressions can be found for the various parameters of the channel. These estimations are very useful in subsequent design of various functions of a communications system. Chapter 5: Channel parameter estimation: References T. Kailath, V. Poor. Detection of stochastic processes. IEEE Transactions on Information theory, vol. IT-15, no. 3, pp. 350-361, May 1969. T. Kailath. A General Likelihood-ration formula for random signals in Gaussian noise. IEEE Transactions on Information theory, vol. 44, no. 6, pp. 2230-2259, October 1998. C.D. Charalambous, A. Logothetis, R.J. Elliott. Maximum likelihood parameter estimation from incomplete data via the sensitivity equations. IEEE Transactions on AC, vol. 5, no. 5, pp. 928-934, May 2000. S. Dey, C.D. Charalambous. On assymptotic stability of continuous-time risk sensitive filters with respect to initial conditions. Systems and Control Letters, vol. 41, no. 1, pp. 9-18, 2000. C.D. Charalambous, A. Nejad. Coherent and noncoherent channel estimation for flat fading wireless channels via ML and EM algorithm. 21st Biennial symposium on communications, Queen’s University, Kingston, Canada, June, 2002. C.D. Charalambous, A. Nejad. Estimation and decision rules for multipath fading wireless channels from noisy measurements: A sufficient statistic approach. Centre for information, communication and Control of Complex Systems, S.I.T.E., University of Ottawa, Technical report: 01-01-2002, 2002. Chapter 5: Channel parameter estimation: References P.M. Woodward. Probability and Information Theory with Applications to Radar. Oxford, U.K.: Pergamon, 1953. A.D. Whalen. Detection of signals in noise, Academic Press, New York, 1971. A. Leon-Garcia. Probability and Random Processes for Electrical Engineering. Addison-Wesley, New York, 1994. L.A. Wainstein, V.D. Zubakov. Extraction of signals from noise, Englewood Cliffs, Prentice-Hall, New Jersey, 1962. C.W. Helstrom. Statistical theory of signal detection. Pergamon Press, New York, 1960. M.S. Grewal, A.P. Andrews. Kalman filtering – Theory and Practice, Prentice Hall, Englewood Cliffs, New Jersey 07632, 1993. A.H. Jazwinski. Stochastic processes and filtering theory, Academic Press, New York, 1970. V. Poor. An Introduction to signal detection and estimation, Springer-Verlag, New York, 2000. Chapter 5: Stochastic power control for wireless networks: Probabilistic QoS measures Review of the Power Control Problem Probabilistic QoS Measures Stochastic Optimal Control Predictable Strategies Linear Programming Chapter 5: Power Control for Wireless Networks QoS Measures Review of the Power Control Problem M min ( p1 0, , pM 0) p; i 1 i pn g nn M j 1 p j g nj n n Chapter 5: Power Control for Wireless Networks QoS Measures Vector Form [Yates 1981] Then M min pi ; GI P GP p j , j 1, , M 0 i 1 Chapter 5: Power Control for Wireless Networks QoS Measures Probabilistic QoS Measures Define n I ( p) M j 1 p j g nj n 1 n pn g nn , The Constraints are equivalent to I n ( p) 0, n 1, ,M n 1, ,M Chapter 5: Power Control for Wireless Networks QoS Measures Decentralized Probabilistic QoS Measures Chapter 5: Power Control for Wireless Networks QoS Measures Chapter 5: Power Control for Wireless Networks Centralized Probabilistic QoS Measures Chapter 5: Power Control for Wireless Networks QoS Measures Stochastic optimal control Received signal yn (t ) j 1 u j (t ) s j (t ) exp kX nj (t ) d n (t ) M State-space representation yn (t ) j 1 u j (t ) s j (t )S nj (t ) d n (t ); M S nj (t ) exp kX nj (t ) dX nj (t ,t ) nj (t ,t ) nj (t ,t ) X nj (t ,t ) dt nj (t , t ) dBnj (t ) X nj (t0 ,t ) PL d (d ) dB nj Chapter 5: Power Control for Wireless Networks QoS Measures Pathwise QoS and Predictable Strategies define then where pi (t )dt; s j (t ) S S nk s j (t ) S nj (t ) Power control for short-term flat fading Pathwise QoS Measures and Predictable Strategies Mobile Base Station calculates t1 Mobile implements observe => calculate t-1 pm(t) Sm (t-1 pm(t) S(t1/2pm(t-1) t1/2 S(t1/2pm(t) pm(t+1) t t1/2 pm(t-1) Sm (t-1 => pm(t) t1 pm(t) Sm (t => pm(t+1) pm(t+1) Sm (t Power control for short-term flat fading Pathwise QoS Measures and Predictable Strategies Base Mobile Station Base Mobile implements calculates Station pm(t) Sm (t1 pm(t+1) Sm (t Value of signal pm(t+2) Sm (t1 desired pm(t-1) Sm (. pm(t) t1 pm(t+1) Sm (. pm(t) Sm (. pm(t+1) t Observe pm(t)Sm (t => calculate pm(t+1) t1 Chapter 5: Power Control for Wireless Networks QoS Measures Define sni (t )Sni (t ) Chapter 5: Power Control for Wireless Networks QoS Measures Predictable Strategies over the interval [tk , tk 1 ] M min pi (tk 1 ); p ( tk 1 )U ad i 1 p(tk 1 ) GI1 (tk , tk 1 ) G (tk , tk 1 ) p(tk 1 ) (tk 1 ) S nk (t )t[t Predictable Strategies Linear Programming k ,tk 1 ] Chapter 5: Power Control for Wireless Networks QoS Measures Chapter 5: Power Control for Wireless Networks QoS Measures (t k 1 ) sn (t )S n (t ) s j (t ) S j (t ) sn (t )S n (t ) sn S n Chapter 5: Power Control for Wireless Networks QoS Measures Generalizations Linear Programming S nk (t ) Stochastic Optimal Control with Integral/Exponential-of-Integral Constraints Snk (t ) Chapter 5: Power Control for Wireless Networks: Conclusions Predictable strategies and dynamic models linear programming Probabilistic QoS measures Stochastic optimal control linear programming References J. Zandler. Performance of optimum transmitter power control in cellular radio systems. IEEE Transactions on Vehicular Technology, vol. 41, no. 1, pp. 57-62, Feb. 1992. J. Zandler. Distributed co-channel interference control in cellular radio systems. IEEE Transactions on Vehicular Technology, vol. 41, no. 1, pp. 305311, Aug. 1992. R. Yates. A framework for uplink power control in cellular radio systems. IEEE Journal on Selected Areas in Communications, vol. 13, no. 7, pp. 1341-1347, Sept. 1995. S. Ulukus, R. Yates. Stochastic Power Control for cellular radio systems. IEEE Transaction on Communications, vol. 46, no. 6, pp. 784-798, Jume 1998. P. Ligdas, N. Farvadin. Optimizing the transmit power for slow fading channels. IEEE Transactions on Information Theory, vol. 46, no. 2, pp. 565576, March 2000. References C.D. Charalambous, N. Menemenlis. A state-space approach in modeling multipath fading channels via stochastic differntial equations. ICC-2001 International Conference on Communications, 7:2251-2255, June 2001. C.D. Charalambous, N. Menemenlis. Dynamical spatial log-normal shadowing models for mobile communications. Proceedings of XXVIIth URSI General Assembly, Maastricht, August 2002. C.D. Charalambous, S.Z. Denic, S.M. Djouadi, N. Menemenlis. Stochastic power control for short-term flat fading wireless networks: Almost Sure QoS Measures. Proceedings of 40th IEEE Conference on Decision and Control, volm. 2, pp. 1049-1052, December 2001. Chapter 5: Capacity, Optimal Encoding, Decoding The channel capacity is the most important concept of any communication channel because it gives the maximal theoretical data rate at which reliable data communication is possible We show an efficient method for computing the channel capacity of a single user time-varying wireless fading channels by means of stochastic calculus. We consider an encoding, and decoding strategy with feedback that is optimal in the sense that it achieves the channel capacity. Although the feedback does not increase the channel capacity, it is a tool for achieving the channel capacity Chapter 5: Channel Model and Mutual Information in Presence of Feedback , F , P is a complete probability space with filtration Ft t 0 and finite time t 0, T , T on which all random processes are defined X X t t 0 source signal t t 0 , t r (t ,t k ), d (t ,t k ), (t ,t k ) , is state channel process N N t t 0 Wiener process independent of X, representing thermal noise Chapter 5: Channel Model and Mutual Information in Presence of Feedback The received signal can be modeled as M dYt Z t( k ) At t k X , Y , dt dN t , Y0 0, (1) k 1 Z t k r (t ,t k ) cos ct d (t ) t t k (t ,t k ) t k : is a delay, d : is a Doppler shift, r is an amplitude, : is a phase, M is a number of resolvable paths, and c : is a carrier frequency At X , Y , : is the non-anticipatory functional representing encoding Chapter 5: Channel Model and Mutual Information in Presence of Feedback Also, we define the following measurable spaces associated with stochastic processes dX , Y , X , BX C 0, T ; R , B C 0, T ; R Y , BY C 0, T ; R , B C 0, T ; R 3 3 , B C 0, T ; R , B C 0, T ; R their filtrations F0,XT B C 0, T ; R , F0,YT B C 0, T ; R , F0,T B C 0, T ; R 3 and truncations of filtrations F0,Xt , F0,Yt , F0,t . Chapter 5: Channel Model and Mutual Information in Presence of Feedback The following assumptions are made 2 T M (k ) Pr Z t At t k X , Y , dt 1 0 k 1 (1) has the unique strong solution Definition 1: The set of admissible encoders is defined as follows Aad A : 0, T C 0, T ; R C 0, T ; R R; A At X , Y , ; t 0, T is progressively measurable, T 2 E At X , Y , dt 0 Chapter 5: Channel Model and Mutual Information in Presence of Feedback Theorem 1. Consider the model (1). The mutual information between the source signal X and received signal Y over the interval [0,T], conditional on the channel state , IT(X,Y|F Q), is given by the following equivalent expressions dPX ,Y | x, y | (i ) E X ,Y , log ( x, y ) dPX | x | dPY | y | 2 M 1 (k ) (ii) E E Z t At t k X , Y , 2 0 k 1 T (k ) ˆ Z t At t k Y , | dt , k 1 M 2 Aˆt t k Y , E At t k X , Y , | F0,Yt , Chapter 5: Channel Model and Mutual Information in Presence of Feedback (iii ) I X , Y | dP , T IT X , Y | E X ,Y dPX ,Y | x, y | ( x, y ) log dPX | x | dPY | y | Definition 2. Consider the model (1). The Shannon capacity of (1) is defined by 1 C sup IT X , Y | F ( X , A )X Aad T subject to the power constraint on the transmitted signal E At2 X , Y , | P (2) Chapter 5: Upper Bound on Mutual Information Theorem 2. Consider the model (1). Suppose the channel is flat fading. The conditional mutual information between the source signal X, and the received signal Y is bounded above by T 1 IT X , Y | F P E Z t2 dt 2 0 (3) It can be proved that this upper bound is indeed the channel capacity, by observing that there exists a source signal with Gaussian distribution dX t X t dt 2 PdWt , X 0 0 such that the mutual information between that signal X and received signal Y is equal to the upper bound (3). Chapter 5: Upper Bound on Mutual Information The capacity is T P 2 C E Z t dt 2T 0 Chapter 5: Optimal Encoding/Decoding Strategies for Non-Stationary Gaussian Sources We assume that the channel is flat fading (M=1), that it is known to both transmitter and receiver, that a source is Gaussian nonstationary, and can be described by the following differential dX t Ft X t dt Gt dWt (4) Ft and Gt are Borel measurable and bounded functions, Integrable and square integrable, respectively, Gt Gttr>0, t[0,T], W is a Wiener process independent of Gaussian random variable X0~ N X ,V Chapter 5: Optimal Encoding/Decoding Strategies for Non-Stationary Gaussian Sources Decoding. The optimal decoder in the case of mean square error criteria is the conditional expectation Xˆ t Y , Z E[ X t | F0,Yt , ] while the error covariance is 2 ˆ Vt Y , Z E[( X t X t Y , Z | F0,Yt , ] Encoding. The optimal encoder is derived by using equation for optimal decoder, and equation for power constraint (6). Chapter 5: Optimal Encoding/Decoding Strategies for Non-Stationary Gaussian Sources Definition 3. The set of linear admissible encoders Lad , where Lad Aad , is the set of linear non-anticipative functionals A with respect to source signal X, which have the form At X , Y , At0 (Y , ) At1 (Y , ) X t The received signal is then dYt Z t ( At0 (Y , ) At1 (Y , ) X t )dt dN t (5) The processes W, and N are independent, and the power constraint (2) becomes E[( At0 (Y , ) At1 (Y , ) X t ) 2 | ] P (6) Chapter 5: Optimal Encoding/Decoding Strategies for Non-Stationary Gaussian Sources Theorem 3 (Coding theorem for stochastic source). If the received signal is defined by the equation (5), the source by (4), then the encoding reaching the upper bound, C P E Z dt 2T optimal decoder, and corresponding error covariance are respectively given by T 2 t 0 At* X , Y * , P ˆ * Y * , Z X X t t Vt* Y * , Z dXˆ t* Y * , Z Ft Xˆ t* Y * , Z dt Z t PVt* Y * , Z dYt* Xˆ 0* Y * , Z X t t t t t * * 2 2 2 Vt Y , Z V exp 2 Fs ds Z s Pds Gs exp 2 Fu du Z u Pdu ds 0 s 0 s 0 V0* Y * , Z V . Chapter 5: Optimal Encoding/Decoding Strategies for Random Variable Sources Theorem 4 (Coding theorem for random variable source). If a source signal X, which is Gaussian random variable N X ,V is transmitted over a flat fading wireless channel, then the optimal encoding and decoding with feedback reaching the channel capacity are t P P * * 2 At X , Y , exp Z s ds X Xˆ t* Y * , Z V 2 0 dXˆ t* Y * , Z Z t Xˆ 0* Y * , Z X Pt 2 * PV exp Z s ds dYt 20 t * * 2 Vt Y , Z V exp P Z s ds 0 V0* Y * , Z V On Channel capacity: Conclusions We can use the new stochastic dynamical models developed to compute new results and get better insight on various computations of channel capacity which is a very important measure for transmission of information More information in the session, friday, nov. 13th. Chapter 5: Capacity, Optimal Encoding-Decoding: References C. Shannon. Channel with side information at the transmitter. IBM Journal, pp. 289-293, Oct. 1958. A. Goldsmith, P. Varaia. Capacity of Fading Channels with channel side information. IEEE Transactions on Information Theory, vol. 43, no. 6, pp. 1986-1992, Nov. 1997. G. Caire, S. Shamai. On the capacity of some channels with channel state information. IEEE Transactions on on Information Theory, vol. 45, no. 6, pp. 2007-2019, Sept. 1999. E. Bigliery, J. Proakis, S. Shamai. Fading Channels: Information theoretic and communication aspects. IEEE Transactions on Information Theory, vol. 44, no. 6, pp. 2619-2692, Oct. 1998. T. Cover. Elements of information theory. Future Work Robust Modeling Receiver Design Optimal Coding Decoding Joint Source and Channel Coding for Wireless Channels Computation of the Channel Capacity for MIMO Channels and Joint Source and Channel Coding Power Control for Wireless Networks