In this paper we consider integrated fusion and relaying algorithms for signal estimation based on noisy measurements over large-scale ad-hoc sensor networks. The algorithms we present are constructed locally at each node by exploiting locally available information about the network topology. They generate at each node a signal estimate via causal linear processing of a locally generated state sequence, and a dither-quantized version of the state sequence for broadcasting. The state sequence is generated from a local drive input and the locally available quantized messages that are broadcasted from directly connected nodes. We present distributed fusion algorithms, and evaluate their MSE performance in a signal-in-measurement noise problem, as a function of the number of quantization bits used for message broadcasting and the algorithmic processing rate.
展开▼
