Paper detail

Coding With Action-dependent Side Information and Additional Reconstruction Requirements

Constrained lossy source coding and channel coding with side information problems which extend the classic Wyner-Ziv and Gel'fand-Pinsker problems are considered. Inspired by applications in sensor networking and control, we first consider lossy source coding with two-sided partial side information where the quality/availability of the side information can be influenced by a cost-constrained action sequence. A decoder reconstructs a source sequence subject to the distortion constraint, and at the same time, an encoder is additionally required to be able to estimate the decoder's reconstruction. Next, we consider the channel coding "dual" where the channel state is assumed to depend on the action sequence, and the decoder is required to decode both the transmitted message and channel input reliably. Implications on the fundamental limits of communication in discrete memoryless systems due to the additional reconstruction constraints are investigated. Single-letter expressions for the rate-distortion-cost function and channel capacity for the respective source and channel coding problems are derived. The dual relation between the two problems is discussed. Additionally, based on the two-stage coding structure and the additional reconstruction constraint of the channel coding problem, we discuss and give an interpretation of the two-stage coding condition which appears in the channel capacity expression. Besides the rate constraint on the message, this condition is a necessary and sufficient condition for reliable transmission of the channel input sequence over the channel in our "two-stage" communication problem. It is also shown in one example that there exists a case where the two-stage coding condition can be active in computing the capacity, and it thus can actively restrict the set of capacity achieving input distributions.