An Algebraic Framework for Multi-Terminal Communication.

Abstract

We consider the problem of developing coding techniques and characterizing information-theoretic achievable rate regions for the following three multi-terminal communication channels. Firstly, we study an interference channel with three transmitter receiver pairs (3-IC). Secondly, we consider a broadcast channel with three receivers (3-BC), wherein three independent information streams are to be communicated to the three receivers. Thirdly, we consider a two user multiple access channel (MAC) with channel state information distributed at the transmitters (MAC-DSTx). The above channels are assumed discrete, memoryless and used without feedback.

Current known coding technique for a general instance of these channels are based on independent unstructured codes. Recognizing the need for codes endowed with algebraic closure properties, we identify three ensembles of coset codes. We propose coding techniques based on these ensembles that exploit their algebraic closure property. We develop tools to characterize information-theoretic performance of the proposed coding techniques. These enable us derive achievable rate regions for a general instance of the above channels. The current known achievable rate regions can be enlarged by gluing together current known coding techniques and the ones proposed herein. Moreover, such an enlargement, as indicated below, is proven to be strict for certain instances.

We identify additive and non-additive instances of 3-IC for which the derived achievable rate region is analytically proven to be strictly larger than current known largest. Moreover, for these channels, the proposed coding techniques based on coset codes is capacity achieving. We also identify a vector 3-BC for which the achievable rate region derived herein is analytically proven to be strictly larger than the current known largest. This vector 3-BC is the first known broadcast channel, for which superposition and binning of unstructured independent codes, proposed over three decades ago, can be strictly improved upon. We also identify non-additive and non-symmetric instances of MAC-DSTx for which the proposed coding technique is verified, through computation, to yield strictly larger achievable rate regions.

Finally, we develop a coding technique based on nested coset codes to characterize a weaker set of sufficient conditions for the problem of computing sum of sources over a discrete memoryless MAC.