DSP seminar

DATE: Tuesday, February 24
TIME: 4 p. m.
Location: Beckman Room 5602

SPEAKER: Robert Soni
TITLE: Projection Methods for Improved Performance in Adaptive Filtering Applications

Common adaptive filtering problems are often solved using the the least mean square (LMS) 
method.  These problems include channel equalization, echo cancellation, interference 
suppression, beamforming, and system identification.  Often, convergence rate or tracking
performance specifications can not be adequately met.  Particularly, LMS approaches will fail if
the existing specification changes to reflect increasing bandwidths or a more demanding (time-
varying or multi-path plagued) system.  Recursive least squares (RLS) methods can improve the
convergence rate performance at the expense of computational complexity and their ability to
track time-varying systems.  To address these issues, we present a "new" class of algorithms
loosely grouped as "projection" approaches which offer system designers the ability to tradeoff
computational complexity and performance (convergence and tracking ability) without changing
structure.  Often, we will see that significant performance improvement may be obtained with
 marginal increases in computation.  Simulation examples for the application examples listed
above and theoretical proof demonstrate the efficicacy of these approaches.