Cascaded Adaptive Filters in a Multilinear Approach for System Identification

Abstract

A recent approach has been proposed that introduces a specific structure in terms of a bilinear form, for identifying an unknown system by two cascaded filters. Following this idea, we develop the physical structures that can be described by a multilinear input-output relation. In this paper, we focus on the recursive least-squares (RLS) algorithm that exhibits the fast convergence rate with the cost of high computational complexity. Thereby, we introduce different N degree multilinear forms of the RLS algorithm, which come with a decrease in computational complexity and analyse their performance features. The proposed RLS multilinear algorithms achieve a good trade-off between computational complexity and performance criteria.