Distributed Estimation in Wireless Sensor Networks: Robust Nonparametric and Energy Efficient Environment Monitoring

Abstract

Wireless sensor networks estimate some parameters of interest associated with the environment by processing the spatio-temporal data. In classical methods the data
collected at different sensor nodes are combined at the fusion center(FC) through multihop communications and the desired parameter is estimated. However, this requires a large number of communications which leads to a fast decay of energy at the sensor nodes. Different distributed strategies have been reported in literature which use the computational capability of the sensor nodes and the estimated local parameters of the neighborhood nodes to achieve the global parameters of interest. However all these distributed strategies are based on the least square error cost function which is sensitive to the outliers such as impulse noise and interference present in the desired and/or input data. Therefore there is need of finding the proper robust cost functions which would be suitable for wireless sensor network in terms of communication and computational complexities. This dissertation deals with the development of a number of robust distributed algorithms based on the notion of rank based nonparametric robust cost functions to handle outliers in the (i) desired data; (ii) input data; (iii) in both input and desired data; and (iv) desired data in case of highly colored input data. Exhaustive simulation studies show that the proposed methods are robust against 50% outliers in the data, provide better convergence and low mean square deviation. Further this thesis deals with a real world application of energy efficient environment monitoring. A minimum volume ellipsoid is formed using distributed strategy covering those sensor nodes which indicate the event of interest. In addition a novel technique is proposed for finding the incremental path for regularly placed sensor nodes. It is shown mathematically that the proposed distributed strategy enhances the lifetime of the entire network drastically.