Training of line echo canceller with PRBS signals

Abstract

The paper addresses the problem of convergence time reduction for the line echo cancellation problem in the context of VoIP applications. Training signals can be used for this purpose. Intuitively speaking, any decorrelated signal, such as white noise, can be used as a training signal. C. Antweiler and H.-G. Symanzik (see Proc. ICASSP-1995, p.3031-4, 1995) have shown that a certain signal, called perfect sequence, can be such a training signal. The method for generating such signals is not easy to use, neither can it be extended to arbitrary lengths. The authors propose another candidate signal, called maximum length pseudo random binary sequence (mlPRBS), to be a perfect sequence. The conditions on such a signal are further analyzed and highlighted. It is shown that the proposed training signal fulfils such requirements. The claims are backed by simulation results. The simulation elements used are available in the public domain.