Noise-residual normalization

A new normalizer, called the noise-residual normalizer (NRN), is proposed for use in array signal processing applications. Whereas most time-spectral normalizers commonly handle source related bias reduction by examining power values in a small, moving, time-spectral window, the NRN technique exploits the phase information contained in the array to spatially model and subtract these source contributions at the array sensor level prior to beamforming. The remaining “residual” sensor data is beamformed and time-spectrally smoothed to obtain the noise estimate used in normalization. The performance of the NRN technique for narrowband, time-spectral normalization is investigated with respect to bias and variance. Results are compared with those obtained for the two-pass, split-window normalizer