I. Introduction
The interest in wireless sensor networks (WSN) as a solution to distributed inference problems in the context of diverse monitoring applications has been steadily increasing in recent years [1]–[5]. The basic idea is to use remote sensor nodes that communicate over wireless links and are spatially distributed over the region to be monitored; these sensor nodes collect, process, and communicate information about the phenomenon of interest. The remote deployment and the unsupervised nature of the WSN make it difficult to recharge sensor node batteries or perform other maintenance tasks. Large network lifetime can thus be achieved only by designing the WSN as energy efficient as possible. This in turn requires to operate the WSN with small communication overhead and low computational complexity.