The design and development of a twin fixed-point parallel DSP architecture for teleconferencing applications
MetadataShow full item record
Conventional telephones have grown into a new generation of telephone systems called hands-free telephones. In hands-free telephones, the speaker and microphone are widely separated according to application's need. Teleconferencing is one such application, where the acoustic path between the speaker and microphone is very long. This long acoustic path varies with time and causes an undesirable effect called the acoustic echo. Theoretically, the acoustic echo can be suppressed using long adaptive filters. However in real-time, long adaptive filters cannot converge fast enough to track the acoustic echo’s dynamic changes. One of the recent approaches is to use multirate filtering techniques to split the long adaptive filter into short subband adaptive filters. A new type of non-overlapping subband scheme has gained popularity in acoustic echo cancellation schemes. This scheme has high computational complexity, but can be easily implemented in real-time on floating-point hardware. In this thesis, a new twin fixed- point parallel DSP architecture has been proposed and designed to meet the complexity of the subband-based adaptive algorithm. The new DSP architecture utilizes a high-speed synchronous link to achieve fast data exchange and parallel execution of time-intensive applications. The simulation and real-time results confirm the feasibility of the algorithm in the new architecture. A pole-zero model along with an adaptive lattice filter structure is suggested for the new subband scheme. The new HR lattice model has been computer- simulated and the results are presented. The thesis also discusses some of the theoretical basics required for subband and lattice structures.