XI TELEKOMUNIKACIONI FORUM TELFOR 2003, Beograd, Sava Centar, 25.-27.11.2003.g. ADAPTIVE MICROPHONE ARRAY IN ROOM WITH REVERBERATION

I INTRODUCTION Microphone arrays are powerful tool for the acoustic interference suppression. They usually utilize adaptive algorithm that adapts its beam pattern so to maximally suppress interferences, having unity gain for the desired signal. Its weightings are usually estimated applying minimum variance (MV) criterion utilizing one of the two beamformer types: Frost's beamformer [1] or generalized sidelobe canceller (GSC) [2]. The common drawback of MV based adaptive beamformers is theirs sensitivity to the room reverberation, steering and calibration error. Any of these disturbances cause cancellation and distortion of the desired signal. The great efforts were done to improve the performance of the microphone array and to make it robust against reverberation of the room. Some of the solutions utilize a beamformer type algorithm, followed by postprocessor. Claude Marro et al. [3] suggested splitting the microphone array into differentially equispaced subarrays. The proposed microphone array is " static " (nonadaptive). Its major advantage is robustness against acoustical environment. Unfortunately, its performance in suppression of the coherent noise components is less compared with adaptive algorithms such as GSC. Zoran Saric and Milorad Cetina [4] proposed partially adaptive GSC with spatial zeros placed on the directions of the known interference sources. The rest of the unknown interference sources are suppressing by the adaptive algorithm. Sharon Gannot et al. [5] proposed GSC solution, which is adapted to the general transfer function (TF) case. They suggested estimation of TFs by exploiting the nonstationary characteristics of the desired signal. In order to reduce desired signal cancellation, Greenberg et al. [6] proposed adaptive algorithm controlled by the cross correlation of the microphone signals. The algorithm is additionally improved by Osamu Hoshuyama et al. [7] utilizing adaptive blocking matrix. In this paper the influence of the room's reverberation to the performance of the GSC algorithm will be analyzed using general transfer functions (TF) model [5]. It will be shown that the desired signal cancellation is proportional to the cross correlation of the desired signal with its room reflections [8]. If the GSC weightings are estimated in the pauses of desired signal, there is no cancellation of the desired signal. The appropriate pause detection based algorithm in DFT domain will be proposed and tested by the simulations. II FULL ADAPTATION GSC Let us assume that the signals are processing in DFT domain. To simplify notation, we will omit frequency index f in all variables like this X≡X(f). Suppose that …