Date of Award

Spring 1994

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering - (M.S.)

Department

Electrical and Computer Engineering

First Advisor

Haimovich, Alexander

Second Advisor

Ansari, Nirwan

Third Advisor

Siveski, Zoran

Abstract

LMS and RLS type algorithms are suggested for decorrelation of multi-channel systems outputs. These algorithms act as signal separators when applied to unknown linear combinations of the inputs. The performance of the suggested algorithms is compared with that of the conventional LMS and RLS algorithms that minimize the mean square error. It. is shown that the correlation matrix eigenvalue spread associated with the LMS decorrelator is always smaller than the eigenvalue spread corresponding to the conventional LMS. resulting in faster convergence speed for the decorrelator. A new RLS type decorrelator algorithm is suggested. The RLS decorrelator is shown to be faster than the LMS decorrelator. not affected by the eigenvalue spread, and comparable in speed with the conventional RLS algorithm. Convergence analysis by simulation shows that the RLS algorithms and the LMS decorrelator have wider regions of convergence than the conventional LMS.

Share

COinS