A study of convergence property of a multi-channel system identification algorithm

摘要

This paper derives conditions to reduce estimation error and optimum parameters to reduce cross-correlation in an adaptive algorithm to be able to identify unknown multi-channel systems even where the ratio of the independent component involved in each reference signal is very low. In the multi-channel system identification using the reference signals with such strong cross-correlation, the conventional adaptive algorithms cannot estimate the coefficients of adaptive filters. To solve this problem, the pre-processing techniques to increase the independent component involved in the reference signals fed into the unknown systems are studied. Unfortunately, the pre-processing distorts the reference signals; therefore the pre-processing of immoderately increasing the independent component cannot be applied. However, the little increase provides unsatisfied performance to the adaptive system. This paper applies an adaptive algorithm to enable the estimation even where the ratio of the independent components is about 40dB. In this adaptive algorithm, the pre-processing reducing the cross-correlation components is used only for the estimation; therefore, the pre-processed reference signals are not fed into the unknown systems.