Juergen Cezanne
Anh-Tho Nguyen Pong
Acoust. Res. Dept., AT&T Bell Labs., 600 Mountain Ave., Murray Hill, NJ 07974
In a previous talk, ``A new adaptive differential microphone array'' by Elko and Pong, a differential microphone has been introduced that adapts its directivity pattern to the particular acoustic environment to provide for a good signal-to-noise ratio. There, the selected pattern remains more or less constant with respect to frequency. In this talk an approach is described that contains one more degree of freedom. The spectrum of the signals is partitioned in uniform subbands and different directivity patterns are adaptively chosen in each subband. This allows to cancel multiple noise sources with nonoverlapping spectra. An LMS-based algorithm will be derived with focus on a low computational load and a short delay for the desired signal. Consequences on the speed of adaptation are discussed. Further, experimental results of a first implementation with 33 subbands on a PC-based DSP32C board will be presented. The measurements verify the ability of the algorithm to cancel multiple noise sources with disjoint spectra without distorting the desired signal.