101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA...

28
101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman [email protected] University of Illinois DARPA ITMANET

description

Current Uses of Feedback Practice Feedback is noisy, used primarily for Robustness to channel uncertainty Estimation of channel parameters ARQ-style communication w/ erasures

Transcript of 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA...

Page 1: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

101

111

dfads

Using Feedback in MANETs: a Control Perspective

Todd P. [email protected] of Illinois

DARPA ITMANET

Page 2: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Current Uses of Feedback

Theory•Feedback modeled noiseless•Point-to-point: capacity unchanged •Significantly improved error exponents•Reduction in complexity

•MANETs: Enlargement of capacity region

Page 3: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Current Uses of Feedback

PracticeFeedback is noisy, used primarily for•Robustness to channel uncertainty•Estimation of channel parameters•ARQ-style communication w/ erasures

Page 4: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Current Uses of Feedback

PracticeFeedback is noisy, used primarily for•Robustness to channel uncertainty•Estimation of channel parameters•ARQ-style communication w/ erasures

But: Burnashev-style “forward error correction+ARQ” schemes are extremely fragile w/ noisy feedback (Kim, Lapidoth, Weissman 07)

Page 5: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

•Instantiate network feedback control algorithms for MANETs•Develop iterative practical schemes for noisy feedback?•Coding w/ feedback over statistically unknown channels?•Develop fundamental limits of error exponents with feedback w/ fixed block length

Applicability of Feedback in MANETs

dfads

101

111

Page 6: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

0 0.25 0.50 0.75 1.00

00 01 10 11

0 1

....011010]1,0[ W

Communication w/ Noiseless Feedback

Page 7: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

0 0.25 0.50 0.75 1.00

00 01 10 11

0 1

....011010]1,0[ W

Communication w/ Noiseless Feedback

Given an encoder’s Tx strategy, decoding is almost trivial (Baye’s rule)

Page 8: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

0 0.25 0.50 0.75 1.00

00 01 10 11

0 1

....011010]1,0[ W

Communication w/ Noiseless Feedback

Given an encoder’s Tx strategy, decoding is almost trivial (Baye’s rule)How do we select a (recursive) encoder

strategy for an arbitrary memoryless channel?

Page 9: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

A Control Interpretation of the Dynamics of the PosteriorColeman ’09: “A Stochastic Control Approach to ‘Posterior Matching’-style Feedback Communication Schemes”

Page 10: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

A Control Interpretation of the Dynamics of the PosteriorColeman ’09: “A Stochastic Control Approach to ‘Posterior Matching’-style Feedback Communication Schemes”

Page 11: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Fk-1

Controller

Z-1

P(Fk|Fk-1, uk)uk Fk

reference signalFw

*

A Control Interpretation of the Dynamics of the PosteriorColeman ’09: “A Stochastic Control Viewpoint on ‘Posterior Matching’-style Feedback Communication Schemes”

Page 12: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Xk

Fk

Fk+1

Fw*

D(F w* ||F k+1)

D(F w* ||F k)

Reward at any stage k is the reduction in

“distance” to target

Stochastic Control: RewardColeman ’09

Page 13: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Maximum Long-Term Average RewardColeman ’09

Page 14: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Maximum Long-Term Average Reward

(1),(2) hold w/ equality if:• a) Y’s all independent• b) Each Xi drawn

according to P*(x)

Coleman ’09

Page 15: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Maximum Long-Term Average Reward

(1),(2) hold w/ equality if:• a) Y’s all independent• b) Each Xi drawn

according to P*(x)

• Horstein ’63 (BSC)• Schalwijk-Kailath ’66 (AWGN)• Shayevitz-Feder ‘07, ‘08 (DMC)

Coleman ’09

Page 16: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

The Posterior Matching Scheme: an Optimal Solution

• Next input indep of everything decoder has seen so far, with capacity-achieving marginal distribution

• No forward error correction. Adapt on the fly.

Coleman ’09

Posterior matching scheme

Page 17: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

The Posterior Matching Scheme: an Optimal SolutionColeman ’09

• Next input indep of everything decoder has seen so far, with capacity-achieving marginal distribution

• No forward error correction. Adapt on the fly.

Posterior matching scheme

Page 18: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

0 1

0 1

0 1

Implications for Demonstrating Achievable Rates

0 1

1

Coleman ’09

Page 19: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Lyapunov Function

0 1

0 1

Posterior matching scheme:

Coleman ’09

Page 20: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Lyapunov Function (cont’d)

0 1

0 1

0 1

1

Coleman ’09

Page 21: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

ControlTheory

InformationTheory

Symbiotic Relationship

Converse Thms Give Upper Bounds on Average Long-Term Rewards for StochasticControl Problem

Coleman ’09: “A Stochastic Control Viewpoint on ‘Posterior Matching’-style Feedback Communication Schemes”

Page 22: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

ControlTheory

InformationTheory

Symbiotic Relationship

Converse Thms Give Upper Bounds on Average Long-Term Rewards for StochasticControl Problem

KL Divergence Lyapunov functions guarantee all rates achievable

Coleman ’09: “A Stochastic Control Viewpoint on ‘Posterior Matching’-style Feedback Communication Schemes”

Page 23: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Research Results with This Methodology•Interpret feedback communication encoder design as stochastic control of posterior towards certainty•Converse theorems specify fundamental performance bounds on a stochastic control problem related to controlling posterior.• An optimal policy implies the existence of a Lyapunov function, which is in essence a KL divergence •Lyapunov function directly implies achievability for all R < C Coleman ’09

Page 24: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

Research Results with This Methodology

Gorantla and Coleman ‘09: Encoders that achieve El Gamal 78: “Physically degraded broadcast channels w/ feedback“ capacity region in an iterative fashion w/ low complexity

•Interpret feedback communication encoder design as stochastic control of posterior towards certainty•Converse theorems specify fundamental performance bounds on a stochastic control problem related to controlling posterior.• An optimal policy implies the existence of a Lyapunov function, which is in essence a KL divergence •Lyapunov function directly implies achievability for all R < C Coleman ’09

Page 25: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

New Important Directions this Approach Enables

ControlTheory

Information

Theory

•Develop iterative low-complexity encoders/decoders for noisy feedback? Partially Observed Markov Decision Process

101

111

Page 26: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

New Important Directions this Approach Enables

ControlTheory

Information

Theory

•Develop iterative low-complexity encoders/decoders for noisy feedback? Partially Observed Markov Decision Process•Optimal coding w/ feedback over statistically unknown channels? Reinforcement learning from control literature

101

111

Page 27: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

New Important Directions this Approach Enables

ControlTheory

Information

Theory

•Develop iterative low-complexity encoders/decoders for noisy feedback? Partially Observed Markov Decision Process•Optimal coding w/ feedback over statistically unknown channels? Reinforcement learning from control literature•Develop fundamental limits of error exponents with feedback w/ fixed block length Lyapunov function enables a fundamental Martingale condition

101

111

Page 28: 101 111 Using Feedback in MANETs: a Control Perspective Todd P. Coleman University of Illinois DARPA ITMANET TexPoint fonts used.

New Important Directions this Approach Enables

ControlTheory

Information

Theory

•Develop iterative low-complexity encoders/decoders for noisy feedback? Partially Observed Markov Decision Process•Optimal coding w/ feedback over statistically unknown channels? Reinforcement learning from control literature•Develop fundamental limits of error exponents with feedback w/ fixed block length Lyapunov function enables a fundamental Martingale condition•Also: stochastic control approach provides a rubric to check tightness of converses via structure of optimal solution

101

111