Acoustic echo cancellation in IP based voice transmission

In this study, a Fractional Fourier Transform (FrFT) domain adaptive filtering scheme employing least mean square (LMS) algorithm is proposed for acoustic echo cancellation in an IP based voice transmission system. In IP based systems usually conference type communication is realized and far-end speech signal from far-end site arrives to near-end site and an attenuated and delayed form of the far-end speech signal in the near end generates the echo. If this echo signal is not cancelled, it is added to the microphone signal and transmitted to the far-end site. Thus, a very low quality of service is obtained. To overcome this problem, echo cancellation algorithms are developed by applying adaptive filters through LMS, normalized LMS (NLMS), recursive least squares (RLS) and FrFT domain LMS algorithms in time and FrFT domains. The performance of the algorithms are analyzed and compared to each other. In these type of adaptive filter algorithms, far-end signal is the input of the filter to estimate the echo signal (desired signal). In each step, error signal is evaluated from the difference between the desired signal and its estimate. Filter coefficients are updated by previous filter coefficients, input and error signals. Better quality of service and best convergence are achieved by the selected adaptive filter. As a result, it is shown that, FrFT domain LMS-based adaptive filtering algorithm has the best performance in terms of Echo Return Loss Enhancement (ERLE) values.