Some Implementation Aspects of Sliding Window Least Squares Algorithms

Abstract Sliding window is more suitable than exponential window for on-line processing of data with sharp changes. In this paper we propose a simple initialization method for sliding window recursive least squares algorithms. It does not require the implementation of a non recursive least squares algorithm and makes the algorithm more homogeneous between initialization and iterations. Then we discuss on the possible singularity of the downdating step in each iteration. We also formulate the factorized updating and downdating steps for a better numerical reliability. Finally, a variable window width algorithm with reduced memory requirement is presented.