Algorithms and structures for long adaptive echo cancellers

• Main Author: Zerguine, Azzedine
• Published: Loughborough University 1996
• Subjects: 621.3192; Circuits
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.321207

The main theme of this thesis is adaptive echo cancellation. Two novel independent approaches are proposed for the design of long echo cancellers with improved performance. In the first approach, we present a novel structure for bulk delay estimation in long echo cancellers which considerably reduces the amount of excess error. The miscalculation of the delay between the near-end and the far-end sections is one of the main causes of this excess error. Two analyses, based on the Least Mean Squares (LMS) algorithm, are presented where certain shapes for the transitions between the end of the near-end section and the beginning of the far-end one are considered. Transient and steady-state behaviours and convergence conditions for the proposed algorithm are studied. Comparisons between the algorithms developed for each transition are presented, and the simulation results agree well with the theoretical derivations. In the second approach, a generalised performance index is proposed for the design of the echo canceller. The proposed algorithm consists of simultaneously applying the LMS algorithm to the near-end section and the Least Mean Fourth (LMF) algorithm to the far-end section of the echo canceller. This combination results in a substantial improvement of the performance of the proposed scheme over both the LMS and other algorithms proposed for comparison. In this approach, the proposed algorithm will be henceforth called the Least Mean Mixed-Norm (LMMN) algorithm. The advantages of the LMMN algorithm over previously reported ones are two folds: it leads to a faster convergence and results in a smaller misadjustment error. Finally, the convergence properties of the LMMN algorithm are derived and the simulation results confirm the superior performance of this proposed algorithm over other well known algorithms.