Implementation of the DCT-LMS Algorithm for Acoustic Echo Cancellation and its efficiency

conference paper

English

In the recent decades developers of telecommunication systems have adopted NLMS (Normalized Least/Mean Square), a simple but efficient adaptive algortihm, as a preferred method for noise cancellation. In 1993 ITU-T standardized the basic properties and criteria for acoustic echo cancellers through their recommendation G.167 and in 1997 they did the same for line-echo cancellers. Since then, however, each of these documents has been through various updates, and their development is still not over. In the context of these recommendations some researchers also tried to employ other alternatives to the NLMS adaptive algorithm, such as FLMS, FTF, GAL, etc. This article discusses an algorithm named DCT-LMS (Discrete Cosine Tranform - Least Mean Square), which has an advantage over the conventional NLMS in that it first "whitens" input speech samples before applying them to the basic mean-square estimator. By doing so, the NLMS should exhibit an improved convergence rate and a lower level of residual error. In order to decrease the computational complexity of the "whitening" operation a DCT transform is utilized which, in turn, is calculated recursively. It is anticipated that this algorithm will finally be implemented on a DSP (digital signal processor) in real-time and therefore achieving low computational complexity is the main concern. This article provides a thorough analysis of algorithm's structure and performance in two different acoustic echo cancellation environments including a comparison with the widely used NLMS.

adaptive, algortihm, speech, echo, processing, orthogonalization, DCT-LMS,

Vladimir Malenovsky, Ing. Ivan Koula, Ing.

2005

29. 6. 2005

STU Bratislava

Bratislava

80-227-2257-X

Proceedings EC-SIP-M, 5th EURASIP Conference focused on Speech and Image Processing, Multimedia Communications and Services

1

1

40

44

5