Efficient implementations of complex and real valued filter banks for comparative subband processing with an application to adaptive filtering

In this paper, we discuss efficient methods to implement correlation type algorithms in subbands. Based on a polyphase representation, a modulated GDFT filter bank with arbitrary integer decimation ratio can be performed at minimum cost, yielding complex subband signals. Real valued subbands can be achieved by appropriate postprocessing of the complex filter bank. Algorithm specific, complex or real implementation may be better suited, for which we present criteria. For the example of subband adaptive filtering as applied in acoustic echo cancellation, we compare a number of algorithms with respect to their best implementation.