Stationary and nonstationary learning characteristics of the MMAXNLMS algorithm

In this paper a recently presented adaptive algorithm with reduced complexity is analysed for the white Gaussian input case. The new algorithm, which updates the weights corresponding to the element sizes of the data vector with the largest magnitude, is compared with the case where the updated weights are chosen randomly according to a uniform density function. This algorithm was previously analysed for the white Gaussian input case in a stationary environment and stability bounds were established along with the excess mean square error. Here, the previous analysis is extended to the nonstationary case. The results are verified via computer simulations