Distributed Asynchronous Clustering for Self-Organisation of Wireless Sensor Networks

This paper presents a fully distributed asynchronous clust ering protocol for the self-organisation of a wireless sensor net work into an infrastructure of well separated cluster heads that supports in-network processing, routing, and deployment.In this protocol, nodes volunteer asynchronously for clusterhead duty and use a radio beacon to pre-emptively recruit members . Limited beacon range is used as the primary parameter for self-organisation. The resulting topology substantiallyreduces total transmission distance and the expected energy consum ed by radio communication. To further extend network lifetime and capability, well separated cluster heads may be easily located and replaced by more powerful devices 1. INTRODUCTION In this paper we present a hierarchical clustering protocol designed for a large scale Wireless Sensor Network (WSN) of thousands of power constrained data sources, one base station ("sink"), and a data gathering application that continuous ly monitors the network for an aggregate value such as the mean temperature recorded by all sensors. Sensor network design may be influenced by many factors (1), and each device may differ in terms of its capabilities (2). However, the fundamental challenge is to reduce the energy consumed by radio communication such that network lifetime is maximised given the application's quality of service (QoS) guarantee s. The Distributed Asynchronous Clustering (DAC) protocol provides an effective, low cost solution to an essential pro b- lem: how to generate a near optimal number of well separated cluster heads in a wireless deployment. DAC solves this problem by using local knowledge of the network and its environment to distribute cluster heads and terminate their generation. As a result, this protocol has many desirable properties. Fully distributed: Autonomous cluster heads use a limited range radio beacon to recruit members. By utilising local information, such as the relative strength of radio si gnals from nearby cluster heads, sensor nodes are able to make decisions that move network topology toward a global opti- mum. Such decentralised systems are more scalable and more robust against individual node or link failures (3). Adaptive: The probability of a node volunteering for cluster head duty depends on its power level, its beacon range depends on network density, and the shape of each cluster depends on