1aSP5. Blind deconvolution based on common zeros of the z transforms of two observed signals.

Session: Monday Morning, December 2

Time: 11:15


Author: Ken'ichi Furuya
Location: Speech and Acoust. Lab., NTT Human Interface Labs., 3-9-11 Midori-cho Musashino-shi, Tokyo, 180 Japan
Author: Yutaka Kaneda
Location: Speech and Acoust. Lab., NTT Human Interface Labs., 3-9-11 Midori-cho Musashino-shi, Tokyo, 180 Japan

Abstract:

A new method is proposed for recovering an unknown source signal, which is observed through two unknown channels characterized by finite impulse response filters. Unlike conventional blind deconvolution methods, this method can recover not only the spectrum of the signal but also the phase characteristics. This method is based on a cost function that comes from a filter arrangement with a double-layer structure. The cost function is minimized when the common zeros of the z transforms of the two observed signals are extracted. If there are no common zeros between the system transfer functions of the two unknown channels, the common zeros of the observed signals represent the source signal and the noncommon zeros represent the characteristics of the two channels. Therefore, the source signal can be recovered by separating the common zeros from the other zeros, that is, by minimizing the cost function. Adaptive filters are used for this procedure. Computer simulations using room transfer functions as the two unknown channels demonstrate the effectiveness of this method.


ASA 132nd meeting - Hawaii, December 1996