Cooperative Communication and Information Processing in Distributed Wireless Networks

Abstract

Large-scale wireless networked systems of intelligent devices are playing an increasingly important role in our life. In such systems, finding solutions in a collaborative and distributed fashion, in the absence of a central coordinator, is of great importance. In this dissertation, we explore some important problems in this area, making certain contributions to both theory and practice of this broad research topic.

Recent research has shown that cooperation of wireless nodes can achieve much better energy efficiency, which is known as a main concern for wireless ad-hoc and sensor networks. But whether cooperation benefits the total energy consumption or not highly depends on system demands and network topology. In Chapter 2 of this dissertation, we take an initial step to determine the switching criteria for non-cooperative transmission and some representative cooperative transmission strategies. Energy efficiency of relevant transmission strategies is studied both for wideband asymptotes and realistic system settings. General guidelines are presented for optimal transmission strategy selection in some typical scenarios involving system level metrics, aiming at minimum energy consumption with a target BER. We also address the criteria for choosing the optimal strategy according to instantaneous channel knowledge.

Belief propagation (BP) is considered as a prominent information processing framework for wireless networks recently. However, infeasible computation and communication requirement involved in applications entailing non-discrete distributions limits its use in practical situations. In Chapter 3, an effective approach to address the message representation/approximation problem in BP algorithms is studied, exploiting the recently proposed Gaussian particle filtering technique. The effectiveness of our approach is testified through the self-calibration problem in wireless networks, where the system dynamism, largely unexplored in the BP study, is explicitly considered.

Distributed and energy efficient in nature, message passing algorithms (such as belief propagation) are attractive for wireless applications. To this end, we propose a variational message passing framework for Markov random fields, with more energy and computation saved compared to the traditional belief propagation algorithm. Based on this framework, structured variational methods are explored to take advantage of the simplicity of approximation and the high accuracy of exact inferences. To investigate the asymptotic performance of this structured distributed inference framework, we first distinguish the intra- and inter-cluster inference algorithms as vertex and edge processes (corresponding to reversible and non-reversible Markov chains respectively). Their difference is illustrated, and convergence rate is derived for the intra-cluster inference procedure which is based on an edge process (the inter-cluster process has been well studied as reversible Markov chains). Then, viewed as a mixed vertex-edge process, the overall performance of structured variational methods is characterized via the coupling approach. The tradeoff between the complexity and performance of this algorithm is also addressed, which provides insights for network design and analysis. This constitutes the Chapter 4 of this dissertation.

The structured inference algorithms invoke the requirement of distributed node clustering. To tackle this problem, we also devise a novel distributed network decomposition algorithm with the aid of the factor graph model and the max-product algorithm. Formulating the network decomposition as an optimization problem, we derive a fully distributed procedure to cluster nodes and achieve the (approximate) minimum cut weight. Simple local operations and message forms also make it particularly suitable for wireless networks with limited capability and resource. Moreover, the proposed algorithm is readily extensible, thus providing a potentially powerful, data-independent clustering scheme for a wide range of data processing and networking applications.