Jonas Rutström

Licentiate Thesis, Signals and Systems, Uppsala University, March 2005.

Thesis in pdf (10.2M)

Abstract:

This work considers tracking of time-varying parameters and automatic tuning of the step-size for the Simplified Wiener LMS algorithm (SWLMS). When tracking time-varying parameters in applications where the rate of change of the time-varying parameters and the noise level may change frequently, it is of interest to adjust the adaptation gain, or the step-size, on-line. The reason for this is that proper manual tuning of the step-size in these cases often is very time consuming, or maybe even impossible.

The purpose of this work has been to find a promising step-size updating algorithm to be used in combination with the SWLMS algorithm in order to create an almost self-tuning algorithm that can be used only with little help from the system designer. Various step-size candidates are evaluated and compared in different tracking scenarios.

In addition to the comparison of the different step-size algorithms, a small study concerning two other tracking issues is also performed. The first issue deals with the potential performance gain obtained by introducing individual step-size control of the different time-varying parameters. The second issue concerns the use of specific information, available to the designer, about the time-varying parameters and the characteristics of signals passed through the time-varying system, that maybe can be applied to improve the overall tracking performance.

In the end, a small case study is performed. Here the most promising algorithms are implemented in a realistic communication scenario. It is shown that the proposed methods are widely superior compared to the traditional constant gain algorithms.

References:

Conference paper in IEEE VTC 2002-Fall on gain adaptation in WLMS tracking algorithms.

Paper describing the Wiener LMS adaptation algorithm.