A NEW IDENTIFICATION ALGORITHM FOR ALLPASS SYSTEMS BY HIGHER-ORDER STATISTICS

Based on a single cumulant of any order M greater than or equal to 3, a new allpass system identification algorithm with only non-Gaussian output measurements is proposed in this paper. The proposed algorithm, which includes both parameter estimation and order determination of linear time-invariant (LTI) allpass systems, outperforms other cumulant based methods such as least-squares estimators simply due to the more accurate model (allpass model) used by the former. It is applicable in channel equalization For the case of a phase distorted channel. Moreover, the well-known (minimum-phase) prediction error filter has been popularly used to deconvolve seismic signals where the source wavelet can be nonminimum phase and speech signals where the vocal-tract filter can be nonminimum phase. Therefore, the proposed algorithm can be used to remove the remaining phase distortion of the nonminimum-phase source wavelet and nonminimum-phase vocal-tract filter in predictive deconvolved seismic signals and speech signals, respectively. It is also applicable in the minimum-phase - allpass decomposition based ARMA system identification method. Some simulation results and experimental results with real speech data are provided to support the claim that the proposed algorithm works well.