Natural gradient non-parametric maximum likelihood algorithm for sparse channel estimation in non-Gaussian noise

Abstract

The stochastic gradient (SG) based non-parametric maximum likelihood (NPML) adaptive algorithm gives better channel estimation as compared to least squares in the presence of additive non-Gaussian noise. However, SG-NPML based channel estimator requires large number of iterations to converge for sparse channels. Natural gradient (NG) based adaptive algorithms are known to converge faster as compared to SG algorithms when the channel coefficients are warped into a known Riemannian structure with respect to Euclidean space, and performs well in sparse channel. In this paper, we propose a quadratic warping transformation on channel coefficients space and then calculate the Riemannian metric tensor that describes the local curvature of coefficient space. The proposed approach is applied for sparse channel estimation for orthogonal frequency division multiplexing (OFDM) based receiver in the presence of interference (modelled as non-Gaussian process) and simulation results show that the proposed NG based NPML algorithm converges faster than conventional SG-NPML for the same mean square error (MSE) floor. © 2016 IEEE.