Abstract
In this work, we adopt the concept of 'variable tap length' in convolutive blind source separation. This relatively new concept has attracted attention in the field of adaptive signal processing, notably the least mean square (LMS) algorithm. However, to our best knowledge, this concept has not yet been introduced in the context of blind signal processing, where the desired signal is not available. Henceforth, we have applied the fractional variable tap length proposed by Gong et al to the convolutive Infomax to demonstrate its potential. In contrast to the approach of Gong et al, we define the optimum tap-length as the minimum tap-length that minimises the cross-terms or off-diagonal elements of the covariance matrix of the estimated sources. The simulation studies presented herein compare our approach with convolutive infomax with different tap lengths and confirms the effectiveness of the proposed approach. This technique can prove to be useful, whenever the filter length is longer than expected, for example, in approximating fractional delays.