Computational Burden Reduction in Real-Time System Identification of Multi-Rail Power Converter by Re-using Covariance Matrix Approximation

Abstract

This paper presents an approach to significantly reduce the computational burden of typical recursive algorithms used for real-time system identification. Recursive algorithms, such as Affine Projection (AP) and Recursive Least Square (RLS), contain two important updates per iteration cycle; the Covariance Matrix Approximation (CMA) update and the gradient vector (or cost function) update. Usually, the computational effort of updating CMA is much higher than that of updating gradient vector. Therefore, reusing CMA, calculated from the last iteration cycle, for the current iteration can result in computational cost savings for real-time system identification. This technique is particularly suitable for system identification-based adaptive control of complex power converter architectures suffering enormous computational burden. In the paper, this technique is applied for AP and RLS algorithms, for the purpose of identifying the parameters of a three-rail power converter.