Analysis of Statistical Performance Measures
Abstract
Key accomplishments were fueled by the discovery of an inherent relationship between the Multi-Stage Wiener Filter (MWF) and the iterative search method of Conjugate-Gradients (CG) This lead to a host of new algorithmic implementations of CG-MWT that offer a number of important advantages over the original implementation of MWF The discovery that adding a "stage" to the MWF was equivalent to taking a "step" of a CG search has lead to substantial improvements to the MWF in terms of: (i) computationally efficient implementations that exploit structure in the data matrices or correlation matrices, such as Toeplitz, sparseness, etc.; (ii) amenability to real-time implementation; (iii) amenability to "smart" initialization (based on information learned or statistics estimated during the process of searching for a training sequence embedded in the data, for example), (iv) easy incorporation of a-priority information, constraints, and/or "past history" for accelerated convergence; and (v) amenability to complementary PC based rank reduction.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 30, 2002
- Accession Number
- ADA414936
Entities
People
- Michael D. Zoltowski
Organizations
- Purdue University