On Source-Channel Communication in Networks Michael Gastpar Department of EECS University of California, Berkeley gastpar@eecs.berkeley.edu DIMACS: March 17, 2003. Michael Gastpar: March 17, 2003.
Outline 1. Source-Channel Communication seen from the perspective of the separation theorem 2. Source-Channel Communication seen from the perspective of measure-matching Acknowledgments • Gerhard Kramer • Bixio Rimoldi • Emre Telatar • Martin Vetterli Michael Gastpar: March 17, 2003.
Source-Channel Communication Consider the transmission of a discrete-time memoryless source across a discrete-time memoryless channel. ˆ S X Y S ✲ ✲ ✲ ✲ Source F Channel G Destination G ( Y n ) = ˆ F ( S n ) = X n S n The fundamental trade-off is cost versus distortion , ∆ = Ed ( S n , ˆ S n ) Γ = Eρ ( X n ) What is the set of • achievable trade-offs (Γ , ∆) ? • optimal trade-offs (Γ , ∆) ? Michael Gastpar: March 17, 2003.
The Separation Theorem ˆ S X Y S ✲ ✲ ✲ ✲ Source F Channel G Destination G ( Y n ) = ˆ F ( S n ) = X n S n For a fixed source ( p S , d ) and a fixed channel ( p Y | X , ρ ) : A cost-distortion pair (Γ , ∆) is achievable if and only if R (∆) ≤ C (Γ) . A cost-distortion pair (Γ , ∆) is optimal if and only if R (∆) = C (Γ) , subject to certain technicalities. Rate-matching: In an optimal communication system, the minimum source rate is matched (i.e., equal) to the maximum channel rate. Michael Gastpar: March 17, 2003.
Source-Channel Communication in Networks Simple source-channel network: ˆ S 1 X 1 Y 1 S 11 ✲ F 1 ✲ ✲ G 1 ✲ Src 1 Dest 1 Channel S 12 , ˆ ˆ S 2 X 2 Y 2 S 22 ✲ F 2 ✲ ✲ G 12 ✲ Src 2 Dest 2 Trade-off between cost (Γ 1 , Γ 2 ) and distortion (∆ 11 , ∆ 12 , ∆ 22 ) . Achievable cost-distortion tuples? Optimal cost-distortion tuples? For the sketched topology, the (full) answer is unknown. Michael Gastpar: March 17, 2003.
These Trade-offs Are Achievable: For a fixed network topology and fixed probability distributions and cost/distortion functions: R 2 ✻ If a cost-distortion tuple satisfies R R (∆ 1 , ∆ 2 , . . . ) ∩ C (Γ 1 , Γ 2 , . . . ) � = ∅ , ❅ ❅ ❅ ❅ ❅ C then it is achievable. ❅ R 1 ✲ When is it optimal? Michael Gastpar: March 17, 2003.
Example: Multi-access source-channel communication S 1 F 1 X 1 ∈ { 0 , 1 } S 1 , ˆ ˆ ✲ Src 1 S 2 ❅ ❅ Y = X 1 + X 2 ❅ ❘ ♥ ✲ G 12 ✲ Dest ✒ � S 2 � ✲ F 2 X 2 ∈ { 0 , 1 } � Src 2 Capacity region of this channel is contained inside R 1 + R 2 ≤ 1 . 5 . Goal: Reconstruct S 1 and S 2 perfectly. R and C do not intersect. S 1 and S 2 are correlated: S 1 = 0 S 1 = 1 Yet uncoded transmission works. S 2 = 0 1 / 3 1 / 3 S 2 = 1 0 1 / 3 R 1 + R 2 ≥ log 2 3 ≈ 1 . 585 . This example appears in T. M. Cover, A. El Gamal, M. Salehi, “Multiple access channels with arbitrarily correlated sources.” IEEE Trans IT-26, 1980. Michael Gastpar: March 17, 2003.
So what is capacity? The capacity region is computed assuming independent messages. In a source-channel context, the underlying sources may be dependent. MAC example: Allowing arbitrary dependence of the channel inputs, the capacity is log 2 3 = 1 . 585 , ”fixing” the example: R ∩ C � = ∅ . Can we simply redefine capacity appropriately? Remark: Multi-access with dependent messages is still an open problem. T. M. Cover, A. El Gamal, M. Salehi, “Multiple access channels with arbitrarily correlated sources.” IEEE Trans IT-26, 1980. Michael Gastpar: March 17, 2003.
Separation Strategies for Networks In order to retain a notion of capacity: Discrete messages are transmitted reliably ˆ S 1 F ′ X 1 Y 1 G ′ S 11 ✲ F ′′ ✲ G ′′ ✲ ✲ ✲ ✲ Src 1 Dest 1 1 1 1 1 Channel S 12 , ˆ ˆ S 2 F ′ X 2 Y 2 G ′ S 22 ✲ F ′′ ✲ G ′′ ✲ ✲ ✲ ✲ Src 2 Dest 2 2 2 12 12 What is the best achievable performance for such a system? — The general answer is unknown. Michael Gastpar: March 17, 2003.
Example: Broadcast Z 1 ˆ Y 1 S 1 ❄ ✲ ♥ ✲ G 1 ✲ Dest 1 S X ✲ Z 2 Src F ˆ Y 2 S 2 ❄ ✲ ♥ ✲ G 2 ✲ Dest 2 ∆ ∗ 1 and ∆ ∗ S, Z 1 , Z 2 are i.i.d. Gaussian. 2 cannot be achieved si- multaneously by sending messages Goal: Minimize the mean- reliably: The messages disturb one squared errors ∆ 1 and ∆ 2 . another. Denote by ∆ ∗ 1 and ∆ ∗ 2 the single- But uncoded transmission achieves user minima. ∆ ∗ 1 and ∆ ∗ 2 simultaneously. This cannot be fixed by altering the definitions of capacity and/or rate-distortion regions. Michael Gastpar: March 17, 2003.
Alternative approach Eρ ( X n ) ≤ Γ ˆ S X Y S ✲ ✲ ✲ ✲ Source F Channel G Destination G ( Y n ) = ˆ F ( S n ) = X n S n A code ( F, G ) performs optimally if and only if it satisfies R (∆) = C (Γ) (subject to certain technical conditions). Equivalently, a code ( F, G ) performs optimally if and only if ρ ( x n ) = c 1 D ( p Y n | x n || p Y n ) + ρ 0 d ( s n , ˆ s n ) = − c 2 log 2 p ( s n | ˆ s n ) + d 0 ( s ) I ( S n ; ˆ S n ) = I ( X n ; Y n ) We call this the measure-matching conditions. Michael Gastpar: March 17, 2003.
Single-source Broadcast ˆ Y 1 S 1 ✲ G 1 ✲ Dest 1 S X ✲ Src F Chan ˆ Y 2 S 2 ✲ G 2 ✲ Dest 2 Measure-matching conditions for single-source broadcast: If the single-source broadcast communication system satisfies ρ ( x ) = c (1) 1 D ( p Y 1 | x || p Y 1 ) + ρ (1) = c (2) 1 D ( p Y 2 | x || p Y 2 ) + ρ (2) 0 , 0 s 1 ) = − c (1) s 1 ) + d (1) 2 log 2 p ( s | ˆ d 1 ( s, ˆ 0 ( s ) , s 2 ) = − c (2) s 2 ) + d (2) d 2 ( s, ˆ 2 log 2 p ( s | ˆ 0 ( s ) , I ( X ; Y 1 ) = I ( S ; ˆ I ( X ; Y 2 ) = I ( S ; ˆ S 1 ) , and S 2 ) , then it performs optimally. Michael Gastpar: March 17, 2003.
Sensor Network M wireless sensors measure physical phenomena characterized by S . ❅ � U 1 X 1 ✲ F 1 ❅ � U 2 X 2 ✲ F 2 ❅ � ˆ S Y S ✲ ✲ ✲ Source G Dest ❅ � U M X M ✲ F M Michael Gastpar: March 17, 2003.
Gaussian Sensor Network The observations U 1 , U 2 , . . . , U k are noisy versions of S . W 1 U 1 X 1 ✓✏ ❄ ✲ ✲ F 1 k =1 E | X k | 2 ≤ MP � M ✒✑ ❈ ❈ ❈ ❈ W 2 ❈ U 2 X 2 ✓✏ ❄ ❈ ✲ ✲ F 2 ❈ Z ✒✑ ❆ ❈ ❆ ❈ ❆ ❈ ❆ ❈ ❆ ❈❈ ˆ S Y ❲ S ✓✏ ❆ ❯ ❄ ✲ ✲ Source G Dest ✒✑ ✄✄ ✗ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ W M ✄ U M X M ✓✏ ❄ ✄ ✲ ✲ F M ✄ ✒✑ Michael Gastpar: March 17, 2003.
Gaussian Sensor Network: Bits Consider the following communication strategy: W 1 U 1 X 1 ✓✏ ❄ F ′ F ′′ ✲ ✲ Bits k =1 E | X k | 2 ≤ MP � M ✒✑ 1 1 ❈ ❈ ❈ ❈ W 2 ❈ U 2 X 2 ✓✏ ❄ ❈ F ′ F ′′ ✲ ✲ ❈ Bits Z 2 2 ✒✑ ❆ ❈ ❆ ❈ ❆ ❈ ❆ ❈ ❆ ❈❈ ˆ S Y ❲ S ✓✏ ❆ ❯ ❄ ✲ G ′′ G ′ ✲ Source Bits Dest ✒✑ ✄✄ ✗ ✄ ✄ ✄ ✄ ✄ ✄ ✄ ✄ W M ✄ U M X M ✓✏ ❄ ✄ F ′ F ′′ ✲ ✲ ✄ Bits ✒✑ M M Michael Gastpar: March 17, 2003.
Gaussian Sensor Network: Bits (1/2) Source coding part. CEO problem. See Berger, Zhang, Viswanathan (1996); Viswanathan and Berger (1997); Oohama (1998). S ∼ N c (0 , σ 2 W 1 S ) U 1 T 1 ✓✏ ❄ F ′ ✲ ✲ ✒✑ 1 and for k = 1 , . . . , M , W 2 U 2 T 2 ✓✏ ❄ W k ∼ N c (0 , σ 2 F ′ W ) ✲ ✲ ✒✑ 2 ˆ S S G ′ ✲ Src Dest For large R tot , the be- havior is D CEO ( R tot ) = σ 2 W . R tot W M U M T M ✓✏ ❄ F ′ ✲ ✲ ✒✑ M Michael Gastpar: March 17, 2003.
Gaussian Sensor Network: Bits (2/2) Channel coding part. Additive white Gaussian multi-access channel: � � 1 + MP R sum ≤ log 2 . σ 2 Z However, the codewords may be dependent. Therefore, the sum rate may be up to 1 + M 2 P � � R sum ≤ log 2 . σ 2 Z ⋆ ⋆ ⋆ Hence, the distortion for a system that satisfies the rate-matching condition is at least σ 2 W D rm ( M ) ≥ � � 1 + M 2 P log 2 σ 2 Z Is this optimal? Michael Gastpar: March 17, 2003.
Gaussian Sensor Network: Uncoded transmission Consider instead the following “coding” strategy: α 1 W 1 U 1 X 1 ✓✏ ❄ ✓✏ ❄ ✲ ❅ � k =1 E | X k | 2 ≤ MP � M � ❅ ✒✑ ✒✑ ❈ ❈ ❈ α 2 ❈ W 2 ❈ U 2 X 2 ✓✏ ❄ ✓✏ ❄ ❈ ✲ ❈ ❅ � Z � ❅ ✒✑ ✒✑ ❆ ❈ ❆ ❈ ❆ ❈ ❆ ❈ ❆ ❈❈ ˆ S Y ❲ S ✓✏ ❆ ❯ ❄ ✲ ✲ Source G Dest ✒✑ ✄✄ ✗ ✄ ✄ ✄ ✄ ✄ ✄ ✄ α M ✄ W M ✄ U M X M ✓✏ ❄ ✓✏ ❄ ✄ ✲ ✄ ❅ � � ❅ ✒✑ ✒✑ Michael Gastpar: March 17, 2003.
Gaussian Sensor Network: Uncoded transmission Strategy: The sensors transmit whatever they measure, scaled to their power constraint, without any coding at all. � M � � P � Y [ n ] = MS [ n ] + W k [ n ] + Z [ n ] . σ 2 S + σ 2 W k =1 If the “decoder” is the minimum mean-squared error estimate of S based on Y , the following distortion is incurred: Proposition 1. Uncoded transmission achieves σ 2 S σ 2 W D 1 ( MP ) = . M 2 W ) /P σ 2 S + σ 2 M +( σ 2 Z /σ 2 W )( σ 2 S + σ 2 W This is better than separation ( D rm ∝ 1 / log M ). In this sense, uncoded transmission beats capacity. Is it optimal? Michael Gastpar: March 17, 2003.
Recommend
More recommend