Multi-SOI based particle filter for distributed estimation in ad hoc wireless sensor network

This paper deals with particle filters (PF) based on quantized innovation for distributed estimation (DE) in ad hoc wireless sensor network (WSN) within a robot tracking context. When only one bit is exchanged per instant between sensors the sign of innovation particle filter (SOI-PF) has a good tracking ability. However, we propose in this work to show that processing of more than one SOI information coming from a set of neighboring sensors, simultaneously, is more performant than using a multi level quantized innovation on the same number of bits. The tracking performance of the so resulted multi-SOI PF, presented here in the context of nonlinear robot´s motion tracking, are compared then to the multi level quantized innovation PF (MLQI-PF), which is derived in this work for a nonlinear and non gaussian tracking context. Simulations show that the MSOI-PF algorithm outperforms the MLQI-PF, moreover, the MSOI-PF´s performance with only two binary information are similar to that of the PF based on a non quantized innovation.