Search

Distributed estimation strategies over wireless sensor networks are one of the active areas of research due to the wide range of applications in a variety of fields ranging from agriculture to surveillance. The classical least squares based algorithms are simple and effective when the noise follows Gaussian distribution. In practical scenarios, the noise may have heavy tailed distributions unlike the Gaussian distribution. The presence of outliers or impulsive noise in distributed networks is unavoidable in practical scenarios due to atmospheric phenomena, electric machinery, saturation effects, link or node failures, etc. To handle such impulsive noise environments, robust cost functions derived from the theory of robust statistics and special robust functions of residual error should be utilized. The study of such robust techniques in distributed scenarios is not available in the literature, and such a literature review is much needed for the further development of novel robust algorithms to tackle outliers. A review of such robust distributed algorithms available in the literature for parameter estimation is presented in this manuscript.

Distributed signal processing is an important area of research in wireless sensor networks (WSNs) which aims to increase the lifetime of the entire network. In WSNs the data collected by nodes are affected by both additive white Gaussian noise (AWGN) and impulsive noise. The classical square error based distributed techniques used for parameter estimation are sensitive to impulse noise and provide inferior estimation performance. In this paper, novel robust distributed learning strategies are proposed based on the Wilcoxon norm and its variants. The Wilcoxon norm based learning strategy provides very slow convergence speed. In order to circumvent this improved distributed learning strategies based on the notion of the Wilcoxon norm are proposed for different types of environmental data. These algorithms require less computational complexity compared to previous ones. In addition these algorithms offer faster convergence rate in the presence of biased input data. Simulation based experiments demonstrate that the proposed techniques provide faster convergence speed than the previously reported techniques in both biased and unbiased input data. HighlightsThis paper deals with the development of some robust novel algorithms.These are robust against outliers in the desired data.Sign-regressor and sign-sign incremental Wilcoxon norm algorithms are proposed.A novel norm is also proposed.

Conventional distributed strategies based on least error squares cost function are not robust against outliers present in the desired and input data. This manuscript employs the generalized-rank (GR) technique as a cost function instead of least error squares cost function to control the effects of outliers present both in input and desired data. A novel indicator function and median based approach are proposed to decrease the computational complexity requirement at the sensor nodes. Further to increase the convergence speed a sign regressor GR norm is also proposed and used. Simulation based experiments show that the performance obtained using proposed methods is robust against outliers in the desired and input data.