An adaptive LMS algorithm with a quick tracking property for time-varying parameter estimation

This paper proposes a method which estimates time-varying parameters of a nonstationary signal such as a biological one in real time. This method is based on a local quasi-stationary concept in the block processing and on the standard LMS algorithm in the recursive processing. That is, we consider a first-order approximation to the time-varying parameters for a short period of time, and estimate coefficients of the constant and the first-order variation terms by the standard LMS algorithm. At the end of each period the estimated values are renewed. This approach does not require calculating matrix inversion and It uses only input values at each time to estimate time-varying parameters adaptively. Also, it can be implemented because of its simplicity. The most attractive point of this approach is that the coefficients of the first-order variation terms are estimated at each time, and thus quick tracking of time-varying parameters can be achieved. This improves the tracking performance drastically. A convergence property of the proposed algorithm is discussed theoretically. Finally, simulation results are presented to illustrate the tracking performance of the proposed algorithm compared with that of the standard LMS algorithms.