Distributed Sequential Estimation in Wireless Sensor Networks

This paper considers the problem of decentralized sequential estimation in dynamic wireless sensor networks. A coherent medium access control layer is considered, and optimal linear precoder and decoder matrices are designed to minimize the mean square error (MSE) in an online setting. Different from the state-of-the-art decentralized estimators, the proposed framework is flexible enough to handle time-varying parameters, channel gains, and power constraints. Although the general transceiver design problem is nonconvex, a fast block coordinate descent-based method is proposed that incurs very low complexity and yields near-optimal solutions. Motivated by the need to reduce the communication overhead incurred by the centralized schemes, two fully distributed transceiver design algorithms that make use of the constrained linear minimum MSE machinery are also advocated. The resulting approximate precoders are not only near optimal but can also be calculated locally at each sensor. Finally, the entire framework is generalized so as to allow tracking of parameters that follow a known state-space model. Extensive simulations are provided to demonstrate the efficacy of the proposed class of algorithms.