Abstract
Stochastic gradient algorithms were thoroughly investigated. Results reported in the literature as well as experiments we have carried out indicate that their performance can be improved by the addition of a gradient smoothing element. This motivated us to present and study a version of the algorithm, with constant adaptation coefficients, which we named Smoothed Least Mean Square (SLMS). The well-known LMS algorithm turns out to be a special case of the SLMS. We develop equations governing the behavior of first- and second-order statistics of this algorithm and conditions for its convergence. Our study indicates that improved steady-state performance can be achieved by the additional smoothing process.
Original language | English |
---|---|
Pages (from-to) | 265-276 |
Number of pages | 12 |
Journal | Signal Processing |
Volume | 11 |
Issue number | 3 |
DOIs | |
State | Published - 1 Jan 1986 |
Externally published | Yes |
Keywords
- Adaptive filters
- gradient algorithm
- least mean square (LMS)
ASJC Scopus subject areas
- Control and Systems Engineering
- Software
- Signal Processing
- Computer Vision and Pattern Recognition
- Electrical and Electronic Engineering