Abstract
Adaptive signal processing algorithms are often used in order to ″track″ an unknown time-varying parameter vector. Such algorithms are typically some form of stochastic gradient-descent algorithm. The Widrow LMS algorithm is apparently the most frequently used. This work develops an upper bound on the norm-squared error between the parameter vector being tracked and the value obtained by the algorithm. The upper bound illustrates the relationship between the algorithm step-size and the maximum rate of variation in the parameter vector. Finally, some simple covariance decay-rate conditions are imposed to obtain a bound on the mean square error.
Original language | English (US) |
---|---|
Pages | 466-469 |
Number of pages | 4 |
State | Published - 1980 |
Externally published | Yes |
Event | Unknown conference - Denver, CO, USA Duration: Apr 9 1980 → Apr 11 1980 |
Other
Other | Unknown conference |
---|---|
City | Denver, CO, USA |
Period | 4/9/80 → 4/11/80 |
ASJC Scopus subject areas
- General Engineering