Reducing Energy Consumption for Distributed EM-based Clustering in Wireless Sensor Networks

Minimizing the energy consumption of wireless sensors is critical, yet a challenge for the design of wireless sensor networks (WSN). Energy is consumed in WSNs by sensing, communicating and processing. In various WSN applications, it is likely that communications are the major source of power consumption, rather than computation. Therefore, assuming that local processing is much less expensive than communicating, we focus on minimizing the number of transmissions for a distributed clustering problem in a sensor network. In our setup, each node in the network senses an environment that can be described as a mixture of Gaussians and each Gaussian component corresponds to one of the elementary conditions. For estimating the Gaussian components, we propose a distributed EM algorithm based on stochastic approximation (DEM-SA), and we study a trade-off between local processing and communication for the distributed clustering problem. DEM-SA reduces the traffic and contention in a WSN by keeping computations and communications local and avoiding the need for cycles through the network. Simulation results will be presented.