A novel adaptive leakage factor scheme for enhancement of a variable tap-length learning algorithm

In this paper a new adaptive leakage factor variable tap-length learning algorithm is proposed. Through analysis the converged difference between the segmented mean square error (MSE) of a filter formed from a number of the initial coefficients of an adaptive filter, and the MSE of the full adaptive filter, is confirmed as a function of the tap-length of the adaptive filter to be monotonically non-increasing. This analysis also provides a systematic way to select the key parameters in the fractional tap-length (FT) learning algorithm, first proposed by Gong and Cowan, to ensure convergence to permit calculation of the true tap-length of the unknown system and motivates the need for adaptation in the leakage factor during learning. A new strategy for adaptation of the leakage factor is therefore developed to satisfy these requirements with both small and large initial tap-length. Simulation results are presented which confirm the advantages of the proposed scheme over the original FT scheme.