Ph.D Thesis

Ph.D StudentMaor Alina
SubjectTopics in Multiuser Source Coding
DepartmentDepartment of Electrical and Computer Engineering
Supervisor PROF. Neri Merhav


In recent years, the theory and practice of communication systems over distributed networks developed greatly. In this research we investigate three scenarios of information exchange under the average distortion constraints, from their source coding aspects and, for certain cases, we also treat joint source-channel extensions of the problem. Consider two users, which have access to correlated information and communicate interactively, in a two-way communication manner, exchanging the data in a number of successive alternating transmissions. Each user must reconstruct the data of the opponent with an accuracy that increases as the dialogue proceeds. We investigate an interactive hierarchical communication dialogue carried over stationary memoryless noisy channels and prove a separation theorem, asserting that there is no loss of asymptotic optimality in using separately source and channel encoders. The source-channel separation does not always hold even in non-interactive communication, and this result is non-trivial and is rather unexpected, since the flow of information is bi-directional. Next, we investigate the source coding problem of successive refinement (SR) of information when side information (SI) is available at the decoders causally. We characterize the achievable per-stage rates and distortions for this setup. We then extend the scenario to communication over discrete memoryless channels and show that here also a separation principle holds. For the two-way communication with causal SI at the decoders, we derive inner and outer bounds for the achievable rate-distortion region which differ only by the intermediate Markov conditions. Multi-user communication sometimes involves broadcast transmissions between users. Hence, in the last part of the research, we address the setup where a single encoder communicates via broadcast transmissions (over noise-free channels) with the same group of decoders in a number of successive steps. At all steps, either causal or non-causal SI is available at the decoders. This setup is of special interest since the problem of SR with non-causal SI and one decoder at each stage is fully solved only for degraded SI, and here we address and partially solve a generalized problem. For the problem of two-stage two-decoder SR with non-causal SI we derive inner and outer bounds for the rate-distortion region and show that these bounds are tight in some cases. For the case of causal SI at the decoders we derive a single-letter expression for the rate-distortion region. Finally, for a per-stage communication over memoryless broadcast channels with random parameter known causally at the encoder, we establish a coding theorem.