This paper addresses the numerical efficiency of adaptive filtering implemented in subbands. Our approach first focuses on oversampled GDFT (generalized DFT) filter banks and their potential benefits over other possible subband decompositions. Although the subband filters presented use complex arithmetic, the discussed method allows factorization into a real valued polyphase network, followed by a complex GDFT modulation, which can be mostly implemented via an FFT. Secondly we discuss the advantages and potential savings that can be gained by processing complex subband signals, with particular reference to adaptive system identification problems, for which we give demonstrations of the potential benefits of our GDFT approach compared to adaptive identification in both fullband and critically sampled DCT-IV based pseudo-QMF subbands.
展开▼