Sparse identification of output error models using l-1 regularized least square

This paper presents the application of l-1 Regularized Least Square (H RLS)to Sparse identification of linear systems. The l-1 norm is the closest possible convex function to the function 1-0 norm and provides a convex optimization problem provided cost function without l-1 norm is convex. The sparse parameters of Output-Error (OE) model, which gives non-convex cost function, are estimated by combining Instrumental Variable method with 1-1 RLS resulting into a two stage algorithm. To support the speculation, the paper presents performance analysis using simulation results.