Coding Techniques for Communication, Storage, and Computation in Distributed Networks

Abstract

As the demand for communication, storage, and computation of large data in a network consisting of massive devices increases, there has been a growing interest in designing distributed networks. This is mainly because existing centralized networks are not able to meet the requirements (e.g. low latency, massive connectivity) of increasing data and large number of nodes for emerging Internet of Things (IoT). In this dissertation, I propose distributed architectures for communication, storage, and computation of large data in a network focusing on scalability and security, which are one of the most important issues in the configuration of distributed networks. To support this, I analyze the performance of distributed networks in an information-theoretic sense, and demonstrate that the proposed architectures show the near-optimal performances on processing communication, storage, and computation of large data in a distributed network.

To begin with, I discuss the problem of interference management in distributed networks. Specifically, I focus on a blind interference alignment (IA) technique that exploits a reconfigurable antenna at the receiver sides, which does not require global channel state information at the transmitter sides. This technique can give the scalability of the network by reducing feedback overhead of channel state information in large-scale interference networks. I propose blind IA schemes for fully-connected and partially-connected interference networks, and derive the information-theoretic upper-bounds on the degrees-of-freedom (DoF) of the networks. I also consider a practical channel assumption that the coherence time of the channel is not long enough to perform a blind IA technique.

Secondly, I consider user's privacy and data security for information retrieval in distributed storage systems. I propose two secure distributed storage systems and their private information retrieval (PIR) schemes that keep user's privacy (concealing the index of the desired message) from each of the databases and preserve data security from an eavesdropper who can access to one of the databases or link between the database and the user to obtain information about messages. Moreover, It is shown that the PIR rates of the proposed schemes are within a constant multiplicative factor from the derived upper bound on the capacity of the PIR problem.

Lastly, I address a secure distributed computing problem in which a master wants to perform computation tasks (e.g. matrix multiplication, vector convolution) on confidential data with non-colluding workers in parallel, while not revealing information about confidential data to workers in an information-theoretic sense. To enhance the scalability of distributed computing networks, I provide a distributed computing scheme using polynomial codes that can efficiently cope with straggling effects. I demonstrate the performance of the proposed scheme on recovery threshold by showing that the achievable recovery threshold is within a constant multiplicative factor from the upper bound.