Transform Domain Adaptive Signal Processing

Abstract

Transform Domain Adaptive Signal Processing scheme using a combination of transforms and correlation cancellers implemented with plane rotations have been described. A recursive implementation of first order KLT by reconfiguring a lattice filter has been demonstrated. The usefulness of a deflation scheme for estimating eigen vectors has been shown. A study has been made of the acquisition behaviour of the rotation angle in Jacobi and Givens operations. The convergence behaviour of transforms such as DHT, DCT and low order KLT together with lattice stages and Jacobi/Givens plane rotation decorrelator has been studied. The study shows that for time series data best performance is obtained if transforms are preceded by lattice filter and followed by Jacobi stages.

Indexing terms:

• Transform domain adaptive signal processing
• Adaptive filters
• Pattern recognition
• System identification