Skip to Main Content
Extensive work to develop and optimize signal processing for signals that are corrupted by additive Gaussian noise has been done so far mainly because of the central limit theorem and the ease in analytic manipulations. It has been observed that the algorithms designed for Gaussian noise typically perform poorly in the presence of non-Gaussian noise. This paper discusses an expectation maximization (EM) algorithm using kernel density estimates to improve channel estimation in a non-Gaussian noise environment. As a well known fact, expectation followed by maximization performed iteratively converges to local maxima for Gaussian mixtures. The noise probability density is assumed unknown at the receiver and is estimated by using the kernel density estimator. Thereby combining EM with the kernel density estimate provides a robust channel estimator for various non-Gaussian noise environments. The simulations for impulsive noise and cochannel interference with Gaussian noise confirms that a better estimate can be obtained by using the proposed technique as compared to the traditional least squares algorithm, which is considered optimal in the Gaussian noise environments.