Abstract:
Many problems in signal processing requires the estimation of the state that changes over time using a sequence of noisy measurements made on the system. In reality most of the applications involve non-linear and non-Gaussian features. Estimating the state of the system in nonlinear non Gaussian environment is highly intractable. This includes the classical problem, of target tracking in wireless sensor network. In this report binary sensor networks are considered as a special case of wireless sensor networks for tracking the target. Unlike sensors considered in traditional tracking approaches, binary sensors provide only one bit of data indicating presence or absence of a target in the sensing range. The signals that reach the fusion center of these networks are therefore binary signals embedded in noise, and they pose challenging problems for recovering the sensed information by the sensors.
Particle filtering algorithm provides a numerical solution to the non-tractable recursive Bayesian estimation problem in case of non-linear and non-Gaussian systems like target tracking in binary sensor networks.
In this dissertation work, we have used the state space approach for deriving the particle filtering algorithm for non linear estimation problem. Various versions of particle filtering algorithms have been used for estimating the state of the system and have shown that the choice of auxiliary particle filter gives reasonably good results as compared to other particle filters when the process noise is equal to greater than the measurement noise.
Two particle filtering algorithms have been considered for processing of the binary data at the fusion center namely, auxiliary particle filtering and cost reference particle filtering. Unlike auxiliary particle filtering (APF), cost-reference particle filtering does not rely on any probabilistic assumptions about the dynamic system. Finally, the imperfect nature of the wireless communication channel between sensors and the fusion center is incorporated in the particle filter tracking algorithm known as channel aware particle filtering. For simulation MATLAB is used and it is demonstrated through simulation results that APF outperforms the cost reference particle filtering considerably in the presence of fading environment
