Implementation And Improvement Of The Fast LMS/Newton Algorithm

Hands-free speech communication is indispensable in audio and video conference systems, hot-line telephones and videophones, mobile radio terminals, digital ISDN network etc. However, the control (cancellation) of the acoustic echo has always had a strong impact on the transmission quality in hands-free telecommunication~[1-3]Conventional methods of acoustic echo control (cancellation), such as echo suppression or gain control, may lead to the degradations in speech quality or make the speakers feel uncomfortable.Adaptive echo cancellation provides a comfortable solution whose concept is extremely simple: if an exact electric replica, parallel to the LRM (Loudspeaker-Room-Microphone) system is provided, the decoupling of loudspeaker and microphone can be achieved without any restrictions to the electric acoustic system and-more important-without any loss of convenience to the user.The implementation of such a system is extremely complicated due to three factors: the impulse response of a LRM system have a duration of several hundred milliseconds, the system has to be adaptive, and the adaptation has to be performed with a speech input The traditional adaptive algorithms such as LMS and NLMS cannot obtain the satisfied result in the real-time acoustic echo cancellation processing.Fast LMS/Newton algorithm has provided the possibility of the real-time acoustic echo cancellation~[4]. It combined with the simplicity of the LMS algorithm and the fast convergence rate of the Newton algorithm.Based on the Fast LMS/Newton algorithm, an improved Fast LMS/Newton algorithm has been presented in this paper. The improved algorithm has obtained a faster convergence rate and more stability by consuming a little more computation~[18] Furthermore, an acoustic echo canceller has been made using TMS320C542 (fix-point DSP). The Fast LMS/Newton algorithm has been implemented on this canceller in the real-time processing. It has been proved that Fast LMS/Newton algorithm can obtain a satisfied echo cancellation result~[17].The implementation of the improved Fast LMS/Newton algorithm is going on in the lab.