engleski

Low Complexity Low Rank Transform Domain Adaptive Filtering

This paper introduces an efficient algorithm to solve the low rank transform domain adaptive filtering problem within the framework introduced in /9/. The method in /9/ extracts an underdetermined solution from an overdetermined least squares problem, using a unitary transformation. The optimal transforms as derived in /9/ dramatically improved performance, at the expense of increased complexity. The complexity bottleneck in the original method stems from the necessity to maintain the full length transform domain error vector, in order to identify the optimal low-rank transform to be used in each iteration. We introduce an alternate method to estimate this optimal transform using a statistical approach, involving the transform of the signal., but not of the error. This leads to the greatly reduced complexity. We also introduce a low rank mixed domain approach which further enhances performance and enables us to reduce the size of the transform. Using the acoustics echo-cancellation problem, this algorithm, with the DFT as the transformation, is shown to perform better than the affine projection algortihm of a high order.

Adaptive filtering; transform domain; least squares; affine projection; DFT

nije evidentirano