Advanced Topics in Information Theory

Similar documents
Lecture 22: Final Review

LECTURE 10. Last time: Lecture outline

Lecture 10: Broadcast Channel and Superposition Coding

Lecture 4 Channel Coding

ECE Information theory Final

Lecture 5 Channel Coding over Continuous Channels

Information Theory. Lecture 10. Network Information Theory (CT15); a focus on channel capacity results

EE 4TM4: Digital Communications II. Channel Capacity

Distributed Lossless Compression. Distributed lossless compression system

Information Theory for Wireless Communications. Lecture 10 Discrete Memoryless Multiple Access Channel (DM-MAC): The Converse Theorem

The Gallager Converse

TTIC 31230, Fundamentals of Deep Learning David McAllester, April Information Theory and Distribution Modeling

Lecture 6 I. CHANNEL CODING. X n (m) P Y X

Capacity of a channel Shannon s second theorem. Information Theory 1/33

(each row defines a probability distribution). Given n-strings x X n, y Y n we can use the absence of memory in the channel to compute

Lecture 3: Channel Capacity

UCSD ECE 255C Handout #14 Prof. Young-Han Kim Thursday, March 9, Solutions to Homework Set #5

ECE Information theory Final (Fall 2008)

EE5585 Data Compression May 2, Lecture 27

EECS 750. Hypothesis Testing with Communication Constraints

Chapter 4. Data Transmission and Channel Capacity. Po-Ning Chen, Professor. Department of Communications Engineering. National Chiao Tung University

Lecture 4 Noisy Channel Coding

The Method of Types and Its Application to Information Hiding

A Half-Duplex Cooperative Scheme with Partial Decode-Forward Relaying

On Optimum Conventional Quantization for Source Coding with Side Information at the Decoder

Multiterminal Source Coding with an Entropy-Based Distortion Measure

List of Figures. Acknowledgements. Abstract 1. 1 Introduction 2. 2 Preliminaries Superposition Coding Block Markov Encoding...

EE/Stat 376B Handout #5 Network Information Theory October, 14, Homework Set #2 Solutions

Lecture 2: August 31

UCSD ECE 255C Handout #12 Prof. Young-Han Kim Tuesday, February 28, Solutions to Take-Home Midterm (Prepared by Pinar Sen)

Quiz 2 Date: Monday, November 21, 2016

Sparse Regression Codes for Multi-terminal Source and Channel Coding

Lecture 15: Conditional and Joint Typicaility

Achievable Rates and Outer Bound for the Half-Duplex MAC with Generalized Feedback

EE5139R: Problem Set 7 Assigned: 30/09/15, Due: 07/10/15

18.2 Continuous Alphabet (discrete-time, memoryless) Channel

Interactive Hypothesis Testing with Communication Constraints

National University of Singapore Department of Electrical & Computer Engineering. Examination for

X 1 : X Table 1: Y = X X 2

Lecture 2. Capacity of the Gaussian channel

Midterm Exam Information Theory Fall Midterm Exam. Time: 09:10 12:10 11/23, 2016

The Fading Number of a Multiple-Access Rician Fading Channel

Hypothesis Testing with Communication Constraints

SOURCE CODING WITH SIDE INFORMATION AT THE DECODER (WYNER-ZIV CODING) FEB 13, 2003

Solutions to Homework Set #1 Sanov s Theorem, Rate distortion

Reliable Computation over Multiple-Access Channels

Capacity Bounds for Diamond Networks

EC2252 COMMUNICATION THEORY UNIT 5 INFORMATION THEORY

C.M. Liu Perceptual Signal Processing Lab College of Computer Science National Chiao-Tung University

(Classical) Information Theory III: Noisy channel coding

Channel combining and splitting for cutoff rate improvement

Second-Order Asymptotics for the Gaussian MAC with Degraded Message Sets

On Capacity Under Received-Signal Constraints

The Unbounded Benefit of Encoder Cooperation for the k-user MAC

On Common Information and the Encoding of Sources that are Not Successively Refinable

Exercises with solutions (Set D)

EE376A - Information Theory Final, Monday March 14th 2016 Solutions. Please start answering each question on a new page of the answer booklet.

A new converse in rate-distortion theory

LECTURE 15. Last time: Feedback channel: setting up the problem. Lecture outline. Joint source and channel coding theorem

Lecture 9 Polar Coding

Lecture 1: The Multiple Access Channel. Copyright G. Caire 12

Homework Set #3 Rates definitions, Channel Coding, Source-Channel coding

ECE598: Information-theoretic methods in high-dimensional statistics Spring 2016

A Comparison of Superposition Coding Schemes

Analyzing Large Communication Networks

Noisy channel communication

Distributed Hypothesis Testing Over Discrete Memoryless Channels

Solutions to Homework Set #3 Channel and Source coding

Lecture 8: Channel and source-channel coding theorems; BEC & linear codes. 1 Intuitive justification for upper bound on channel capacity

Chapter 2 Date Compression: Source Coding. 2.1 An Introduction to Source Coding 2.2 Optimal Source Codes 2.3 Huffman Code

Two Applications of the Gaussian Poincaré Inequality in the Shannon Theory

Polar Codes are Optimal for Write-Efficient Memories

Rematch and Forward: Joint Source-Channel Coding for Communications

3F1: Signals and Systems INFORMATION THEORY Examples Paper Solutions

Lecture 5: Channel Capacity. Copyright G. Caire (Sample Lectures) 122

4F5: Advanced Communications and Coding Handout 2: The Typical Set, Compression, Mutual Information

Network coding for multicast relation to compression and generalization of Slepian-Wolf

Classical codes for quantum broadcast channels. arxiv: Ivan Savov and Mark M. Wilde School of Computer Science, McGill University

Principles of Communications

Information Theory. Coding and Information Theory. Information Theory Textbooks. Entropy

Optimal Natural Encoding Scheme for Discrete Multiplicative Degraded Broadcast Channels

LECTURE 13. Last time: Lecture outline

On Scalable Coding in the Presence of Decoder Side Information

Notes 3: Stochastic channels and noisy coding theorem bound. 1 Model of information communication and noisy channel

ECS 452: Digital Communication Systems 2015/2. HW 1 Due: Feb 5

NETWORK INFORMATION THEORY

Performance of Polar Codes for Channel and Source Coding

Efficient Use of Joint Source-Destination Cooperation in the Gaussian Multiple Access Channel

Lossy Distributed Source Coding

EE376A: Homework #3 Due by 11:59pm Saturday, February 10th, 2018

Multimedia Communications. Scalar Quantization

Weak Flip Codes and its Optimality on the Binary Erasure Channel

1 Background on Information Theory

Lecture 11: Quantum Information III - Source Coding

Source and Channel Coding for Correlated Sources Over Multiuser Channels

Multicoding Schemes for Interference Channels

Distributed Hypothesis Testing Over Discrete Memoryless Channels

On Multiple User Channels with State Information at the Transmitters

Lecture 17: Differential Entropy

Transcription:

Advanced Topics in Information Theory Lecture Notes Stefan M. Moser c Copyright Stefan M. Moser Signal and Information Processing Lab ETH Zürich Zurich, Switzerland Institute of Communications Engineering National Chiao Tung University (NCTU) Hsinchu, Taiwan You are welcome to use these lecture notes for yourself, for teaching, or for any other noncommercial purpose. If you use extracts from these lecture notes, please make sure to show their origin. The author assumes no liability or responsibility for any errors or omissions. 3 rd Edition 2018. Version 3.3. Compiled on 12 December 2018. For the latest version see http://moser-isi.ethz.ch/scripts.html

E 2 E 1 B 2 B 3 B 1 B 5 B6 B 4 B 7 E 3

v 1 v 5 v 6 v 2 v v 4 v 3

p a p p b

p

y 1 1 x y 1 x

y y x x

Q F Q P(X) Q

q q q

q > 90 degrees q q

Q > 90 degrees Q Q F

Q > 90 degrees Q Q Q F

Q Q Q F

I II Q 2 Q 1 III M M c

F A Q P(X) Q

Q Q A F

Q Q A F

V U W

Error exponents R 0 E G (R) E B (R) 0 R crit R 0 C R

E 0 (1,Q) E 0 ( 1 2,Q) E 0 ( 1 3,Q) E 0 ( 1 4,Q) max {E 0(ρ,Q) ρr} 0 ρ 1 E 0 ( 1 8,Q) R

Error exponents E SP (R) R 0 E G (R) E (R) 0 R crit C R

Error exponents Error exponents R 0 E SP (R) R 0 E SP (R) E G (R) E G (R) E (R) E (R) 0 R R crit C R 0 R crit R C R

E G (R) R 0 E G (R) R 0 R crit R 0 C R C R

C I (E s ) line above C I (E s ) here a discontinuity is theoretically possible E s

Destination ˆM Dec. ψ n FB θ k Y k DMC Q Y X X k Enc. φ k M Uniform Source F k+1 Delay F k

1 δ 0 0 X δ δ? Y 1 1 δ 1

independent description 49 points dependent description 45 points

0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 4 3 2 1 0 1 2 3 4

ˆx 1 ˆx 2 θ 0.5 1 0.75 0.3578 1.3288 0.8433 0.3973 1.4011 0.8992 0.4202 1.4450 0.9326 0.4336 1.4714 0.9525 0.4414 1.4872 0.9643 0.4461 1.4966 0.9714 0.4488 1.5022 0.9755. 0.4528 1.5104 0.9816

0.4 0.35 0.3 0.25 0.2 0.15 0.1 0.05 0 4 3 2 1 0 1 2 3 4

1 D 0 0 ˆX D D 1 1 1 D X

R I (D) here a discontinuity is theoretically possible line below R I (D) D

rate information rate distortion function (with discontinuity) R I (D) R I (D+ǫ) D D+ǫ distortion

rate λ E [ d(x, ˆX) ] point ( E [ d(x, ˆX) ],I(X; ˆX) ) line of slope λ I(X; ˆX) E [ d(x, ˆX) ] distortion

rate R 0 (q,λ) R( ) λ E q [ d(x, ˆX) ] = λd I q (X; ˆX) = R(D) D distortion

rate R( ) R 0 (q,λ) achievable by q D distortion

rate R 0 (q,λ) R 0 (q,λ) R( ) achievable by q achievable by q distortion

R(D) R 0 (q,λ 1 ) R 0 (q,λ 2 ) slope discontinuities two different tangents with slopes λ 1 and λ 2 D

R(D) D

Û 1,...,ÛK Dest. Decoder Y 1,...,Y n X 1,...,X n U 1,...,U K DMC Encoder DMS

Encoder V 1...,V K Lossy Compressor U 1...,U K Binary DMS Destination ˆV 1,..., ˆV K Decoder Y 1,...,Y n DMC X 1,...,X n

DMC X 1,...,X n Channel Encoder W RD Encoder U 1...,U K Binary DMS Destination V 1,...,V K RD Decoder Ŵ Channel Decoder Y 1,...,Y n

σ 2 i σ 2 4 σ 2 1 σ 2 5 λ σ 2 7 σ 2 2 D 1 σ 2 3 D 4 D 5 D 7 D 2 D 3 σ 2 6 D 6 X 1 X 2 X 3 X 4 X 5 X 6 X 7

ˆx x x ˆx

the sources Q that do not work because for the given R and D: R( Q,D) > R inf D( Q Q) = D P(X) Q

W (1) Enc. φ (1) n Destination ˆX 1,..., ˆX n Dec. ψ (i) n X 1,...,X n Q W (2) Enc. φ (2) n

ˆX (2) Q ˆX(1), ˆX (2) X (, 0) 0 1 Q ˆX(1) X ( 0) ˆX (1) 0 3 2 2 2 1 2 2 1 2 1 0 2 1 Q ˆX(2) X ( 0) 2 2 2 1 ˆX (2) Q ˆX(1), ˆX (2) X (, 1) 0 1 Q ˆX(1) X ( 1) ˆX (1) 0 0 0 0 1 0 1 1 Q ˆX(2) X ( 1) 0 1

ˆX (2) Q ˆX(1), ˆX (2) (, ) 0 1 Q ˆX(1)( ) ˆX (1) 0 1 Q ˆX(2)( ) 3 2 2 2 2 1 2 2 1 2 2 2 1 2 1 2 2 2 1 2 2 2 2

Destination ˆX 1,..., ˆX n Dec. ψ n W Enc. φ n X 1,...,X n Q X,Y Y 1,...,Y n

codeword 1 codeword e nr bin 1 bin 2 bin 3 bin ( e nr 1 ) bin e nr

Destination ˆX 1,..., ˆX n Dec. ψ n W Enc. φ n X 1,...,X n BSS Y 1,...,Y n p p 1 p 1 p

W (1) Enc. φ (1) n X 1,...,X n Destination ˆX 1,..., ˆX n Ŷ 1,...,Ŷn Dec. ψ n W (2) Enc. φ (2) n Y 1,...,Y n Q X,Y

R (2) H(X,Y) separate compression and decompression H(Y) H(Y X) joint encoding H(X Y) H(X) H(X, Y) R (1)

Hsinchu X Taichung Y Q X,Y (, ) rain sun Hsinchu total rain 0.445 0.055 0.5 sun 0.055 0.445 0.5 Taichung total 0.5 0.5

Destination ˆM (1) ˆM (2) Dec. ψ n Y Channel Q Y X (1),X (2) X (1) X (2) Enc. φ (1) n Enc. φ (2) n M (1) M (2) Uniform Source 1 Uniform Source 2

0 X (1) 1 1 ǫ 1 ǫ 1 ǫ 1 0 1 ǫ 1 1 Y 1 ǫ 2 2 X (2) 0 ǫ 2 ǫ 2 1 ǫ 2 3 1

R (2) C (2) C (1) R (1)

R (2) 1 1 R (1)

1 2 0 X (2) 0 1 1 2 1 2 1 Y 1 2 2

R (2) 1 1 2 1 2 1 R (1)

R (2) I 3 I 2 D C I 1 0 E 0 B A R (1)

R (2) R (1)

R (2) I ( X (2) ;Y X (1)) D C I ( X (2) ;Y ) E I ( X (1) ;Y ) B A I ( X (1) ;Y X (2) ) R (1)

R (2) convex hull of C a C b C b C a R (1)

R (2) C [ϑ] R (1)

R (2) R (2) 20 inactive constraint 20 10 C a 10 C b 10 20 R (1) 10 20 R (1)

R (2) inactive constraint 20 10 C [1 2] 10 20 R (1)

R (2) ( [ϑ]) 0,I 2 D C ( [ϑ] ) I 3 I[ϑ] 2,I[ϑ] 2 E (0,0) B A ( [ϑ] I 1,0) ( [ϑ] ) I 1,I[ϑ] 3 I[ϑ] 1 R (1)

R (2) C D B E A R (1)

R (2) ( ) C E (2) σ 2 C ( ) R (1) +R (2) = C E (1) +E (2) σ 2 ( ) C E (2) E (1) +σ 2 B ( ) C E (1) E (2) +σ 2 ( ) C E (1) σ 2 R (1)

R (2) E E R (1)

R (2) α = 0 TDMA for α from 0 to 1 α = E(1) E (1) +E (2) α = 1 R (1)

Destination Û n 1 ˆV n 1 Dec. ψ Y n 1 MAC Q Y X (1),X (2) { (1)} n X k 1 { (2)} n X k 1 Enc. φ (1) Enc. φ (2) U n 1 V n 1 Q U,V

Encoder X (1) MAC Enc. 1 W (1) SW Enc. 1 U MAC Q Y X (1),X (2) Q U,V X (2) MAC Enc. 2 W (2) SW Enc. 2 V Destination Û ˆV SW Dec. Ŵ (1) Ŵ (2) Decoder MAC Dec. Y

U V 1 2 3 4 5 6 7 1 2 0 3 4 5 6 0 7

Q S S S Destination ˆM Dec. ψ Y Q Y X,S X Enc. φ M Uniform Source

1 q 0 0 X 1 q q Y 1 q 1

q 0 0 X q 1 q Y 1 1 1 q

1 p 0 0 X p p Y 1 1 1 p

λ α 0 0 1 2 1 α 1 2 S 1 2λ 2 U 1 α λ 1 2 1 α 1 1 2

Dest. 1 Dest. 2 ˆM (1) ˆM (0) ˆM (0) ˆM (2) Dec. ψ (1) Dec. ψ (2) Y (1) Y (2) BC Q Y (1),Y (2) X X Enc. φ M (1) M (0) M (2) Uniform Source 1 Uniform Source 0 Uniform Source 2

Q Y (2) Y (1) Y (1) Q Y (1) X X Y (2) Y (1)

V Z (1) + Y (1) + X Y (2) Y (1)

codeword X cloud center U

R (1) I ( X;Y (2)) +I ( X;Y (1) U ) R (1) = R (0) +I ( U;Y (2)) +I ( X;Y (1) U ) I ( X;Y (1) U ) I ( U;Y (2)) R (0)

R (2) = R (1) +I ( U (1) ;Y (1)) +I ( U (2) ;Y (2)) I ( U (1) ;U (2)) R (2) R (1) = I ( U (1) ;Y (1)) I ( U (2) ;Y (2)) A R (2) = I ( U (2) ;Y (2)) I ( U (2) ;Y (2)) I ( U (1) ;U (2)) B I ( U (1) ;Y (1)) I ( U (1) ;U (2)) I ( U (1) ;Y (1)) R (1)

X (1) f GP U (1) GP Enc. M (1) X f U (2) U (2) U (2) U (2) f: U (1) U (2) X Enc. 2 M (2)

R (0) +R (2) ( ) 1 2 log 1+ E σ(2) 2 α = 0 time-sharing α = 1 ( ) 1 2 log 1+ E σ(1) 2 R (1)

ˆM (1) X (1) Enc. φ (1) M (1) Uniform Source 1 Destination ˆM (0) ˆM (2) Dec. ψ Y MAC Q Y X (1),X (2) M (0) Uniform Source 0 X (2) Enc. φ (2) M (2) Uniform Source 2

R (0) R (2) R (1)

M (1) Terminal 1 M (2) ˆM (3) (2) X (1) ˆM (1) (5) ˆM(2) (5) Terminal 2 X (2) Y (5) Terminal 5 Y (2) DMN Channel ˆM (3) (5) ˆM(4) (5) X (3) Y (3) Y (4) X (4) Terminal 3 M (3) M (4) Terminal 4 ˆM (2) (4)

ˆM (1) ˆM (0) (1) ˆM (0) (2) ˆM (2) Terminal 1 Terminal 2 Y (1) Y (2) Broadcast Channel Q Y (1),Y (2) X (3) S 1 S 2 X (3) S 3 Terminal 3 M (1) M (0) M (2)

ˆM (1) ˆM (2) Terminal 3 Y (3) S 3 MAC Q Y (3) X (1),X (2) S 1 X (1) X (2) S 2 Terminal 1 Terminal 2 M (1) M (2)

S 2 Terminal 2 X (2) Y (2) S 1 ˆM Terminal 3 Y (3) Q Y (2),Y (3) X (1),X (2) X (1) Terminal 1 M

S 2 Terminal 2 S 4 X (2) Y (2) S 1 ˆM Terminal 4 Y (4) Q Y (2),Y (3),Y (4) X (1),X (2),X (3) X (1) Terminal 1 M X (3) Y (3) S 3 Terminal 3

Dest. 1 ˆM (1) Dec. ψ (1) Y (1) IC X (1) Enc. φ (1) M (1) Uniform Source 1 Dest. 2 ˆM (2) Dec. ψ (2) Y (2) Q Y (1),Y (2) X (1),X (2) X (2) Enc. φ (2) M (2) Uniform Source 2

1 ǫ 1 0 0 X (1) Y (1) ǫ 1 ǫ 1 1 1 1 ǫ 1 1 ǫ 2 0 0 X (2) Y (2) ǫ 2 ǫ 2 1 1 1 ǫ 2

R (2) C (2) C (1) R (1)

R (2) C (2) C (1) R (1)

ˆM (1) Terminal 3 Y (1) X (1) Terminal 1 M (1) IC Q Y (1),Y (2) X (1),X (2) ˆM (2) Terminal 4 Y (2) X (2) Terminal 2 M (2)

1.6 a 12 = 0.15, a 21 = 0.05 1.6 a 12 = 0.35, a 21 = 0.25 1.4 1.4 1.2 1.2 R (2) [bits] 1.0 0.8 0.6 R (2) [bits] 1.0 0.8 0.6 0.4 0.4 0.2 0.2 0 0 0.2 0.4 0.6 0.8 1.0 R (1) [bits] 1.2 1.4 1.6 0 0 0.2 0.4 0.6 0.8 1.0 R (1) [bits] 1.2 1.4 1.6 1.6 a 12 = 0.55, a 21 = 0.45 1.6 a 12 = 0.85, a 21 = 0.75 1.4 1.4 1.2 1.2 R (2) [bits] 1.0 0.8 0.6 R (2) [bits] 1.0 0.8 0.6 0.4 0.4 0.2 0.2 0 0 0.2 0.4 0.6 0.8 1.0 R (1) [bits] 1.2 1.4 1.6 0 0 0.2 0.4 0.6 0.8 1.0 R (1) [bits] 1.2 1.4 1.6 1.6 a 12 = 1.15, a 21 = 1.05 1.6 a 12 = 2.15, a 21 = 2.05 1.4 1.4 1.2 1.2 R (2) [bits] 1.0 0.8 0.6 R (2) [bits] 1.0 0.8 0.6 0.4 0.4 0.2 0.2 0 0 0.2 0.4 0.6 0.8 1.0 R (1) [bits] 1.2 1.4 1.6 0 0 0.2 0.4 0.6 0.8 1.0 R (1) [bits] 1.2 1.4 1.6

R (2) [bits] 3.5 3 2.5 2 1.5 1 R (1) +2R (2) 7.09 bits R (2) 3.26 bits R (1) +R (2) 4.19 bits 2R (1) +R (2) 7.09 bits 0.5 0 0 0.5 1 R (1) 3.26 bits 1.5 2 2.5 R (1) [bits] 3 3.5

d sym weak medium strong very strong 1 2 3 1 2 0 0 1 2 2 3 1 3 2 2 5 2 a