Distributed Spiral Optimization in Wireless Sensor Networks without Fusion Centers

A distributed spiral algorithm for distributed optimization in WSN is proposed. By forming a spiral-shape message passing scheme among clusters, without loss of estimation accuracy and convergence speed, the algorithm is proved to converge with a lower total transport cost than the distributed in-cluster algorithm. Introduction: The most straight forward way to solve optimization problems in wireless sensor networks (WSN) is by centralized processing, where all nodes send observations to one central point that solves the problem using algorithms introduced by [1]. However, this approach is time-consuming, and is not energy-efficient. Recently, alternative decentralized approaches have been proposed, which adopt various distributed optimization algorithms. A brilliant solution has been originally introduced in [2] and further developed in [3]. Furthermore, the decentralized approaches can also be divided into in-network and in-cluster schemes. The former considers the network as a whole, and utilizes peer-to-peer communication to solve the problem [4]. And the latter, which is introduced in [5], is a hybrid form of both the centralized scheme and the in-network scheme. It can achieve higher estimation accuracy, lower transport cost, and lower latency than the distributed in-network schemes. But the drawback is the involvement of fusion centers and thus the low robustness of the system against fusion center failure and the high sensor-to-fusion-center transport costs. In this paper, we propose a distributed spiral algorithm, which achieves lower transport costs