Estimation of a Gaussian Source with Memory in Bursty Impulsive Noise

We consider the problem of estimating correlated Gaussian samples in (correlated) impulsive noise, through message-passing algorithms. This is a meaningful theoretical framework to model signal transmission on power-line communication systems. Due to the mixture of Gaussian variables (the samples) and Bernoulli variables (the impulsive noise switches), the complexity of messages increases exponentially with the number of samples. By adopting a Parallel Iterative Scheduling, with properly constrained messages, it turns out that each iteration of the proposed algorithm is equivalent to the parallel run of a classical Kalman Smoother and a binary sequence detection through the BCJR algorithm. Results demonstrate the effectiveness of the receiver along with its performance, in terms of mean square estimation error.