Particle-Based Message Passing Algorithm for Inference Problems in Wireless Sensor Networks

Optimal distributed estimation algorithms are usually not practical for wireless sensor networks (WSNs). This is because, in a general setup, these algorithms have high computational and data communication costs. Thus, sub-optimal algorithms that use quantized data and are based on linear and Gaussian approximations have been proposed in the literature. Such approximations are not always applicable. In this paper, we propose a distributed estimation method based on the well-known sum-product algorithm. To maintain a feasible complexity for WSNs, the sum-product update rules are reformulated using particle filtering. We consider the problem of distributed target tracking based on quantized data in a WSN. After deriving the factor graph representation of this tracking problem, we apply our proposed algorithm. We then study its performance based on the number of quantization bits, the number of particles and the measurement noise.