Implementation of a switched fast affine projection-normalized least mean square algorithm in the TMSC542 DSK for acoustic and line echo cancellation

• Main Author: Ruiz, Ramon Stephen L.
• Format: text
• Language: English
• Published: Animo Repository 2004
• Subjects: Switching theory; Algorithms; Adaptive signal processing; Adaptive filters
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/3166; https://animorepository.dlsu.edu.ph/context/etd_masteral/article/10004/viewcontent/CDTG003680_P.pdf
• Institution: De La Salle University

This study evaluates the performance of an echo canceller structure that employs two switched adaptive filter algorithms: the Fast Affine Projection (FAP) and the Normalized Least Mean Square (NLMS). The evaluation validates the observation that convergence rate of the FAP and the NLMS are comparable at error coefficients below -15dB. This observation suggests a new algorithm that initially employs an FAP and later switches to NLMS. This new structure presents comparable convergence rate as the FAP but with lower computational overhead. The new algorithm is implemented on Texas Instrument's TMS320C542 DSP starter kit (DSK) and evaluated against the NLMS and FAP algorithms in terms of echo return loss, convergence rate, and computational cost using simulated sources and echo paths. Using this new structure, acoustic and line echo cancellation, which is necessary in most telecommunications applications, can be implemented with improved convergence and lower computational overhead freeing processor resources for other tasks.