Title :
Analysis of a low complexity variable step size adaptive algorithm
Author :
Evans, Joseph B.
Author_Institution :
Dept. of Electr. & Comput. Eng., Kansas Univ., Lawrence, KS, USA
Abstract :
A number of variable-step-size algorithms have been proposed to improve the performance of stochastic gradient based adaptive systems. In the present work, the convergence and steady-state error performance of a low-implementation-complexity variable-step-size algorithm is analyzed. Iterative expressions for the evolution of the moments of the step size are derived and are used in conjunction with expressions for the mean square error to predict the learning curve. Expressions for the steady-state size, from which the steady-state mean square error can be found, are also developed. The analytical results are compared with simulation and good agreement is found.<>
Keywords :
adaptive filters; computational complexity; convergence of numerical methods; iterative methods; convergence; learning curve; low complexity; simulation; steady-state mean square error; stochastic gradient based adaptive systems; variable step size adaptive algorithm;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 1993. ICASSP-93., 1993 IEEE International Conference on
Conference_Location :
Minneapolis, MN, USA
Print_ISBN :
0-7803-7402-9
DOI :
10.1109/ICASSP.1993.319540
