Abstract
This thesis presents acoustic echo cancellation (AEC) algorithms designed in the frequency-domain. Least-mean-square (LMS) based method, such as Proportionate Normalized LMS (PNLMS) method, is used to model the echo path and to match the desired signal at the near end. To enhance cancellation performance of the frequency-domain algorithms, internal iterations and are employed with the coefficient update loop. In addition, microphone array beamforming, such as Delay and Sum (DAS) Fixed Beamformer (FBF) method, Minimum Variance Distortionless Response (MVDR) FBF method and Generalized Sidelobe Canceller (GSC) method, is utilized to receive local speech signals while suppressing echo and interference at undesired directions. Double-talk detector (DTD) is developed to ensure the robustness of the AEC in double-talk scenarios. The proposed AEC algorithm integrated with DTD and the beamformer is validated with simulations and experiments. For single talk AEC performance, the Echo-Return Loss Enhancement (ERLE) for LMS method is 10 dB, for iterative PNLMS method is 35 dB and for GSC method is 50 dB, which is the best method in all AEC algorithms.