The Interaction between Communication, Computation, and Control

Sekhar Tatikonda, Yale University

• Coordination and computation over wireless networks• Traditionally the design has been separated• Projects:

– Role of feedback in communication– Control with communication constraints– Distributed detection and compression in sensor networks– Advanced iterative decoding techniques

Feedback in Communication• Traditional information theory does not treat latency • Examples:

- streaming video - communication in the control loop

• Acks and channel measurements can increase capacity

• Output feedback can decrease latency: - Error decays at a double exponential rate over fading channels • Goals: - develop efficient feedback codes - practical issues: noisy feedback, what do we feedback

Control with Communication Constraints• Controls: x := Ax + Bu + w

• Consider communication between sensors and controller and between controller and actuators

• The communication requirements for stability:

C > log (det A)

• Similar results for other performance objectives

• Goals:

- Communication coordination between distributed controllers

plant

controller

Distributed Processing in Sensor Networks

• Sensor measurements are correlated and localized

• Low power constraints

• Minimize bits not messages

- computation much cheaper than communication

• Routing based on both correlation and geography

• Slepian-Wolf coding

Iterative Decoding Techniques

• LDPC decoding based on the iterative sum-product algorithm• Distributed computation based on message passing• This algorithm be applied to distributed estimation tasks as well• What if messages are corrupted?

• Goals:

- develop fault tolerant message passing schemes - develop feedback codes

p1

b4b1 b2 b3

p2