On limitations of minimum mean-square error deconvolution in deriving impulse responses of rooms

Abstract

Least-squares deconvolution filtering has been utilized extensively in both seismic processing and image restoration. Recently, Berkhout et al. [J. Acoust. Soc. Am. 68, 179–183 (1980)] have proposed the application of a particular seismic processing technique to room acoustics measurements, with the goal of computing the impulse response from the room’s response to a finite-length wavelet. We demonstrate mathematically and experimentally that the assumptions made in defining the method may preclude its usefulness in deriving the true impulse response of a room, showing that in most cases it will whiten the trace. In the common situations where the reflection factors are not constant as a function of frequency, the assumption of a ‘‘white’’ reflection series is not valid; we show how to remove this constraint. However, we also call into question the assumption of a stationary impulse response, and this is less susceptible to easy remedy. The use of a pseudoinverse filter based upon the premises of singular value decomposition is suggested as one way of avoiding these problems and its performance for synthetic and real data is briefly described.