BLIND SEPARATION OF CONVOLUTIVE MIXTURES USING RENYI’S DIVERGENCE By KENNETH E. HILD II A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy BLIND SEPARATION OF CONVOLUTIVE MIXTURES USING RENYI’S DIVERGENCE By Kenneth E. Hild II December 2003 Chair: Jose C. Principe Major Department: Electrical and Computer Engineering A new information-theoretic (IT) criterion is introduced, which is based on the nonparametric approximation of Renyi's quadratic entropy. Using this criterion, convolutive source separation is achieved by minimizing the mutual information between segments of processes. Three reasons are given why Renyi’s entropy is preferred to Shannon’s entropy. In addition, experimental evidence is provided that shows that the proposed method is very data efficient and that it performs well for noiseless synthetic mixtures. Perhaps more importantly, a supervised training method was found that estimates the demixing coefficients directly when the sources are recorded one at a time. This method provides an experimental approximate upper bound to the performance for real recordings. A comparison of several different methods using both real and synthetic mixtures indicates that feedforward, time-domain demixing structures are preferred, current BSS methods are