Online precoder design for parameter tracking in wireless sensor networks

This paper considers the problem of parameter tracking in wireless sensor networks. The sensor nodes observe a random vector source that varies according to a state-space model, and perform linear precoding on the observations, before coherently transmitting them to the fusion center. The fusion center then linearly decodes the received vector in order to recover the source vector. Compared to the state-of-the-art linear estimation approaches, the tracking requirement complicates the optimal precoder and decoder design, which must themselves be time-varying. Towards this end, the paper proposes an online, block-coordinate descent (BCD)-based algorithm that minimizes the mean-square error at every time slot. The proposed designs are not only near-optimal, but also provably convergent for some cases. Simulation results corroborate the performance enhancements provided by the proposed approach.