Achievable Schemes and Performance Bounds for Centralized and Distributed Index Coding

Index coding studies the efficient broadcast problem where a server broadcasts multiple messages to a group of receivers with side information. Through exploiting the receiver side information, the amount of required communication from the server can be significantly reduced. Thanks to its basic yet highly nontrivial model, index coding has been recognized as a canonical problem in network information theory, which is fundamentally connected with many other problems such as network coding, distributed storage, coded computation, and coded caching. In this thesis, we study the index coding problem both in its classic setting where the messages are stored at a centralized server, and also in a more general and practical setting where different subsets of messages are stored at multiple servers. In both scenarios the ultimate goal is to establish the capacity region, which contains all the communication rates simultaneously achievable for all the messages. While finding the index coding capacity region remains open in general, we characterize it through developing various inner and outer bounds. The inner bounds we propose on the capacity region are achievable rate regions, each associated with a concrete coding scheme. Our proposed coding schemes are built upon a two-layer random coding scheme referred to as composite coding, introduced by Arbabjolfaei et al. in 2013 for the classic centralized index coding problem. We first propose a series of simplifications for the composite coding scheme, and then enhance it through utilizing more flexible fractional allocation of the broadcast channel capacity. We also show that one can strictly improve composite coding by adding one more layer of random coding into the coding scheme. For the multi-server scenario, we generalize composite coding to a distributed version. The outer bounds characterize the fundamental performance limits enforced by the problem setup that hold generally for any valid coding scheme. The performance bounds we propose are based on Shannon-type inequalities. For the centralized index coding problem, we define a series of interfering message structures based on the receiver side information. Such structures lead to nontrivial generalizations of the alignment chain model in the literature, based upon which we propose a series of novel iterative performance bounds. For the multi-server scenario, our main result is a general outer bound built upon the polymatroidal axioms of the entropy function. This outer bound utilizes general groupings of servers of different levels of granularity, allowing a natural tradeoff between tightness and computational complexity. The security aspect of the index coding problem is also studied, for which a number of achievability and performance bounds on the optimal secure communication rate are established. To conclude this thesis, we investigate a privacy-preserving data publishing problem, whose model is inspired by index coding, and characterize its optimal privacy-utility tradeoff