New Algorithms for Broad-Band and Narrowband Source Localization and a Separable2-D IIR Filter Realization

摘要

(1) Optimal Design of ARMA (IIR) filters with arbitrary number of poles and zerosfrom Impulse Response Data has been developed. The general criterion derived in this report has never been found before. (2) Optimal synthesis of two dimensional IIR filters using one dimensional modules have been developed. (3) Optimal design of a class of two dimensional IIR filters from spatial domain data has been developed. (4) Optimal identification of Multivariable systems from Impulse response data is given. (5) A Periodogram-based Maximum Likelihood estimator of Narrowband frequencies requiring only off the shelf hardware/software has been developed. (6) A faster Simulated-Annealing method has been developed and applied to frequency estimation. (7) A coherent one-step angles of arrival estimator of multiple broadband sources has been developed. Existing coherent techniques can not localize well separated sources in one step. (8) An Order-Recursive approach has been given for AR-Bispectrum estimation. (9) A Time-Delay-Neural Network has been trained with LPC coefficients for Phoneme/Vowel recognition. (10) Parametric Non-linear prediction algorithms have been introduced for the first time for speech prediction/synthesis/coding.