Convergence Investigation of LMS Algorithms with Time-Varying Step Size
Loading...
Authors
Khomentrakarn, Chusak
Issue Date
1990-05
Type
Thesis
Language
en_US
Keywords
Least mean squared , Algorithms
Alternative Title
Abstract
An adaptive linear combiner has been used in many applications. One common adaptive algorithm for this field is the Least Mean Squared (LMS) algorithm which can be applied to both a stationary and a non-stationary input. This algorithm uses a fixed step size to train a process. A small step size provides the accuracy of an optimum value with less variance but slow convergence. Large step size gives fast convergence but is constrained by convergent conditions. There are a variety of strategies to vary step size according to these advantages and disadvantages. Three strategies are presented here: the Normalized Least Mean Squared algorithm, the Error Dependent Step-Size algorithm and the Independent Step-Size algorithm. The speed of convergence is compared by using the time constant of the mean squared error curve (the learning curve). Only stationary inputs are considered here. Simulation with both deterministic input and stochastic input is included.
Description
This thesis is being archived as a Digitized Shelf Copy for campus access to current students and staff only. We currently cannot provide this open access without the author's permission. If you are the author of this work and desire to provide it open access or wish access removed please contact the Wahlstrom Library to discuss permission.
Citation
C. Khomentrakarn, "Convergence Investigation of LMS Algorithms with Time-Varying Step Size", Masters thesis, Dept. of Engineering, Univ. of Bridgeport, Bridgeport, CT, 1990.