A new fast online identification method for linear time-varying systems

A novel online identification algorithm is proposed, which addresses the problem of convergence rate in dynamic parameter estimation in the presence of abrupt variations as well as noise in time-varying systems. The proposed identification technique optimizes a mean fourth error cost function by virtue of steepest descent (SD) method. It is proven that a unique solution for the optimal correcting gain of the SD update law exists and a closed-form solution is derived. To obtain high sensitivity to parameter variations, a block-wise version of the proposed technique, that incorporates only a finite length window of data, is developed. The performance of the proposed method is compared to those of two benchmark identification techniques.