Julie E. Greenberg
Patrick M. Zurek
MIT Res. Lab. of Electron., Rm. 36-761, Cambridge, MA 02139
Previous work has shown that adaptive-array hearing aids based on the Griffiths--Jim beamformer perform poorly in reverberant environments. Although relatively long adaptive filters successfully reduce reverberant environments. Although relatively long adaptive filters successfully reduce reverberant interference, they also cause substantial cancellation of the desired (target) signal [Greenberg and Zurek, J. Acoust. Soc. Am. 91, 1662--1676 (1992)]. In this work, analysis of the adaptive noise canceller shows that target cancellation is eliminated when the delay in the primary channel (D) is shorter than the interval between the arrival of the direct wave and first reflection at the microphones. This generalizes the observation that setting D=0 prevents target cancellation due to reverberation [Hoffman, Ph.D. thesis, Univ. of Minnesota (1992)]. Computer simulations with several sets of source-to-microphone impulse responses show that target cancellation is most pronounced when there are a small number of reflections and they arrive at the array within D samples of the direct wave. Target cancellation is minimal when there are many reflections and only a small fraction arrive within D samples of the direct wave. Other benefits derived from using nonzero primary delays suggest that it is advisable to use a primary delay that is nonzero, but relatively small. [Work supported by NIDCD.]