Adaptive equalisation of discrete channels via fast convolution techniques

Abstract

A discrete adaptive equaliser for synchronous data transmission is presented, based on fast convolution techniques and efficient Fourier transform computation algorithms. Mean-square error minimisation is automatically performed in the discrete frequency domain by using Rosen's gradient projection method. Algorithms are developed to enable the equaliser control to estimate both mean-square error gradient and correction step size and modify the filter tap weights directly in the discrete frequency domain. This discrete frequency-domain equaliser is compared with the conventional time-domain equaliser as regards ease of implementation and speed of convergence.