Agenda

MSc SPS Thesis presentation

Dopple GroupChat (Dereverberation & Denoise in multi-channel full-duplex communication systems)

Long Xin

In response to the echo and noise interference problems faced in multi-user full-duplex voice communication scenarios, this work proposes a comprehensive solution encompassing signal modeling to beamforming optimization. First, by converting the Bluetooth channel into a virtual acoustic channel, a comprehensive acoustic transmission model covering both single-source and multi-source scenarios is established. Building on this foundation, various beamforming and filtering strategies are discussed, culminating in the proposal of the Multichannel Wiener Filter (MWF) beamforming scheme. Additionally, for the time-varying spatial domain, the update of the covariance matrix combined with a Voice Activity Detector (VAD) is addressed. Simulations demonstrate that the MWF scheme exhibits strong robustness and adaptability under different Bluetooth channel latencies and low Signal-to-Interference Ratio (SIR) environments. Furthermore, system performance is further enhanced when the user end is equipped with a sufficient number of microphones or in the presence of multiple bystanders.

Overview of MSc ME Thesis Presentation