Natural gradient blind deconvolution and equalization using causal FIR filters

Natural gradient adaptation is an especially convenient method for adapting the coefficients of a linear system in inverse filtering tasks such as blind deconvolution and equalization. Practical implementations of such methods require truncation of the filter impulse responses within the gradient updates. In this paper, we show how truncation of these filter impulse responses can create convergence problems and introduces a bias into the steady-state solution of one such algorithm. We then show how this algorithm can be modified to effectively mitigate these effects for estimating causal FlR approximations to doubly-infinite UR equalizers. Simulations indicate that the modified algorithm provides the convergence benefits of the natural gradient while still attaining good steady-state performance.