On-Line Transform Domain LMS Algorithm Implemented with PCA Learning Chuan Wang, L-K Yen, and Jose C. Principe Computational NeuroEngineering Laboratory Electrical and Computer Engineering Department University of Florida Gainesville, FL 32611 Abstract An on-line transform domain Least Mean Square (LMS) algorithm based on a neural approach is proposed. A tempo- ral Principal Component Analysis (PCA) network is used as an orthonormalization layer in the transform domain LMS filter. Since PCA learning is an on-line learning algorithm, an on-line transform domain LMS filter can be easily implemented. Moreover, a modified Kalman estimation, which considers the sequence convergence property of the eigenvectors of PCA, is proposed to train the PCA network. The computation complexity of the PCA based LMS algorithm is then less than the conventional fast Fourier transform based algorithm. Corresponding author: Jose C. Principe CSE 405 Electrical and Computer Engineering Department University of Florida Gainesville, FL 32611 phone: 904-392-2662 fax: 904-392-0044 email: principe@synapse.ee.ufl.edu Presenter: Jose. C. Principe Electrical and Computer Engineering Department University of Florida Gainesville, FL 32611 Section: First choice, (10) Signal Processing Second choice, (14) Applications Oral presentation (overheads)