A Robust Diffusion Adaptive Network Based on the Maximum Correntropy Criterion

Adaptive estimation over distributed networks has received a lot of attention due to its broad range of applications. A useful estimation strategy is diffusion adaptive network, where the parameters of interest can be well estimated from noisy measurements through diffusion cooperation between nodes. The conventional diffusion algorithms exhibit good performance in the presence of Gaussian noise but their performance decreases in presence of impulsive noise. The aim of the present paper is to propose a robust diffusion based algorithm that alleviates the effect of impulsive noise. To this end, we move beyond mean squared error (MSE) criterion and recast the estimation problem in terms of the maximum correntropy criterion (MCC). We use stochastic gradient ascent and useful approximations to derive an adaptive algorithm which is appropriate for distributed implementation. The resultant algorithm has the computational simplicity of the popular LMS algorithm, along with the robustness that is obtained by using higher order moments. We present some simulations results which show that the proposed algorithm outperforms existing alternative that rely MSE criterion.