Normalizing inter-band scaling in minimum variance (MV) beamformer estimates of variable-bandwidth filtered wideband signals

Abstract

Narrow-band minimum variance (MV) beamforming can be used to estimate a wideband signal when the signal is first filtered into several narrow-band signals. This paper identifies and solves a new problem for narrow-band MV beamforming of wideband signals. ‘‘Inter-band scaling’’ can occur when narrow-band MV beamforming is applied to a wideband signal filtered by a variable-bandwidth filterbank. This situation can occur when the wideband signal to be estimated is speech. Then, it is often desireable to use a filterbank with higher-frequency resolution in the low, than in the high-frequency bands since speech has more information in low frequency than high. In this case, the power out of the MV beamformer in each band can depend not only on the actual signal power in that band, but also on the bandwidth. The inter-band scaling normalization method presented here is tested with both white Gaussian noise, and a segment of a speech signal. The tests show that the method is effective and does not distort the speech signal. Finally, this method should be extendable to other problems (such as other adaptive array processing methods) that require estimates of statistical measures of variable-bandwidth filtered signals.