Polynomial Matrix Approach to Independent Component Analysis: (Part I) Basics

Authors:	Sugimoto Kenji, Nara Institute of Science and Technology, Japan Nitta Masuhiro, Nara Institute of Science and Technology, Japan
Topic:	2.1 Control Design
Session:	Polynomial Design Methods: Theoretical Achievements
Keywords:	blind system identificationvector autoregressive modelindependent component analysispolynomial matrix analysis

Abstract

This paper proposes a method for blind system identification based on the independence of input signals. Under the assumption that the system is MIMO, square, and represented by a polynomial matrix fraction with constant numerator matrix, the method makes it possible to identify the system without observation of input signals. This rather challenging problem is solved by applying independent component analysis to an augmented state-space representation in order to estimate coefficients of the denominator polynomial matrix and the numerator matrix.