Linear recursive adaptive equalization utilizing a modified Gauss - Newton algorithm

Abstract

Listen

Translate article

A new tap update algorithm is developed for linear recursive adaptive equalizers. This algorithm is an extension of the Modified Gauss-Newton algorithm, a least-squares method, which has been successfully used to compute the coefficients for recursive digital filters [13], [19]. The new algorithm is an on-line technique but the taps are updated periodically rather than on a symbol-by-symbol basis. This block update procedure has the advantage of providing time to perform the calculations required to update the taps. The Modified Gauss-Newton algorithm is compared with recursive equalizers that attempt to minimize the mean-squared error by means of stochastic gradient update algorithms. It is shown, via Monte-Carlo computer simulation, that by utilizing more information than simply the gradient, improved performance is obtained. In particular, the algorithm is much less sensitive to the eigenvalue ratio corresponding to the distortion. Convergence speed and error performance of the new algorithm are favorably compared with the Least-Squares Adaptive Lattice Equalizer.