M.S. (Master of Science)
Department of Electrical Engineering
Acoustic echo in teleconferencing systems and its cancellation by means of adaptive algorithms utilizing digital filter banks are presented in this work. Interleaving the theoretical algorithm research with practical implementation aspects to find efficient acoustic echo cancellers has been the essential and principal goal of this work. In order to achieve this, a comprehensive study of adaptive signal processing algorithms and multi-rate digital signal processing techniques has been done. Adaptive algorithms have been applied in digital filter bank structures to realize cost-efficient, high-performance acoustic echo cancellers. Starting with analyzing the performance of various adaptive acoustic echo cancellation algorithms (Normalized Least Mean Square (NLMS), Fast Newton Transversal Filter (FNTF) and Gradient Adaptive Lattice (GAL)) in full-band structure and understanding the computational and memory requirements of these algorithms, a comparative study is done in order to assess the feasibility of such algorithms in real-time implementations. NLMS algorithm belongs to the LMS class of algorithms and has poor convergence with speech signals. FNTF algorithm belongs to the Fast-RLS class. It has higher rate of convergence, but has high numerical sensitivity and computational complexity. GAL algorithm belongs to gradient lattice class and has suffers from inherent over-whitening of the input signal leading to poor performance with speech signals. Two new algorithms (Predictive-NLMS and Thin Adaptive Lattice) have also been formulated and implemented. P-NLMS algorithm yields better performance than NLMS algorithm due to the presence of a whitening filter. TAL algorithm is an improvement of GAL algorithm and has higher performance. An efficient subband structure employing polyphase-DFT filter banks was used in conjunction with the adaptive algorithms to realize computationally efficient subband acoustic echo cancellers (AECs). All the algorithms used in subband structure showed an improvement over their performance obtained when used in fullband structure. Real-time implementation of the subband AECs, using the TMS320C30 digital signal processor (DSP), is presented. The thin lattice algorithm has the least computational complexity and has shown superior performance over the rest of the algorithms. An efficient, cost-effective and high-performance subband thin lattice AEC is tested in real-time and an average cancellation of about 40 dB was observed.
Manapragada, Sai C., "Subband-based adaptive algorithms for acoustic echo cancellation" (1995). Graduate Research Theses & Dissertations. 5127.
xiv, 133 pages
Northern Illinois University
Rights Statement 2