Recursive Nuclear Norm based Subspace Identification

Nuclear norm based subspace identification methods have recently gained importance due to their ability to find low rank solutions while maintaining accuracy through convex optimization. However, their heavy computational burden typically precludes the use in an online, recursive manner, such as may be required for adaptive control. This paper deals with the formulation of a recursive version of a nuclear norm based subspace identification method with an emphasis on reducing the computational complexity. The developed methodology is analyzed through simulations on Linear Time-Varying (LTV) systems particularly in terms of convergence rate, tracking speed and the accuracy of identification and it is shown to be computationally lighter and effective for such systems, with the considered rate of change of dynamics.