Proportionate Adaptation Paradigms and Application in Network Echo Cancellation

In this report, a new kind of adaptive filter paradigms, so-called proportionate adaptation, is introduced. Based on some recent work, particularly the proportionate normalized least-mean-squared (PNLMS) algorithm, some new proportionate adaptation paradigms are developed. Their theoretical issues of interest, such as H∞ optimality, convergence, computational complexity, etc. are addressed. The detailed information are referred to the other papers. The application of proportionate adaptation to network echo cancellation is discussed. Specifically, robust proportionate adaptation paradigms are particularly designed for specific network echo cancellation problem. Some real-life experiments taken in Bell Labs are presented. Finally, the implication of proportionate adaptation to neural network training is discussed.