Distributed estimation for nonlinear stochastic signals over sensor networks

This paper considers the problem of estimating the state of nonlinear stochastic processes observed by spatially distributed sensor nodes i.e, observations are taken by a network of sensor nodes. The measurement process of each node is assumed to be some nonlinear function of an unobservable process and is corrupted by gaussian noise. We refer to a scenario in which all nodes in the network wish to have near-optimal identical state estimate of the observed process and there is no centralized computation center. Sensor nodes do not have any global knowledge of the network topology and nodes are allowed to communicate with only their nearest neighbors. Each node applies a particle filtering algorithm to its own measurements to generate an individual state estimated signal. These nodes based estimated signals are then combined by using nonlinear distributed fusion rule to produce improved state estimated signal at each node. We demonstrate through numerical example that the performance of fused state estimated signal is superior to the performance of the state estimated signal generated by particle filtering algorithm.