SUBBAND FILTERING WITH PARAUNITARY LINEAR PHASE FILTER BANKS: APPLICATION TO ACOUSTIC ECHO CANCELLATION

Abstract

The purpose of this thesis is to study subband adaptive filtering and its application in acoustic echo cancellation (AEC). Adaptive filtering in subband is a new technique for the real-time identification of long impulse responses like the ones encountered in acoustic echo cancellation, with reduction in computation complexity and improved convergence behavior. For the critical sampling case, aliasing occurs due to the down-sampling process and cross filters have to be used in the subbands to improve the, otherwise degraded, performance. Adjacent subband approximation (ASA) hets been introduced to have only three subband adaptive digital filters (ADF) in each subband to make the subband filtering scheme really work. But, still, the taps of subband ADF’s are not independent, indicating that the problem of over-determination of the subband adaptive filtering system should be studied. The object of this thesis is to further simplify the subband adaptive filtering scheme, enabling even more reduction in computational complexity, hardware requirement and faster convergence rate. This is achieved by using paraunitary linear phase filter bank and a new single-cross filter (SCF) subband adaptive filtering scheme. The identical analysis/synthesis (up to a sign) filters provided by the paraunitary linear phase filter bank make the already tri-diagonal cross-filter matrix (CFM) a symmetric one. Hence it is sufficient to have only two subband ADF’s in each subband. This not only represents a significant saving in computation and hardware, but also implies improved convergence rate over the regular ASA scheme. Computer simulation confirms this theoretical prediction.