Nonparametric VSS-APA based on precise background noise power estimate
来源期刊：中南大学学报(英文版)2015年第1期
论文作者：WEN Hao-xiang(文昊翔) LAI Xiao-han(赖晓翰) CHEN Long-dao(陈隆道) CAI Zhong-fa(蔡忠法)
文章页码：251 - 260
Key words：adaptive algorithm; affine projection algorithm; echo cancellation; background noise power estimate; variable step-size affine projection algorithm
Abstract: The adaptive algorithm used for echo cancellation (EC) system needs to provide 1) low misadjustment and 2) high convergence rate. The affine projection algorithm (APA) is a better alternative than normalized least mean square (NLMS) algorithm in EC applications where the input signal is highly correlated. Since the APA with a constant step-size has to make compromise between the performance criteria 1) and 2), a variable step-size APA (VSS-APA) provides a more reliable solution. a nonparametric VSS-APA (NPVSS-APA) is proposed by recovering the background noise within the error signal instead of cancelling the a posteriori errors. The most problematic term of its variable step-size formula is the value of background noise power (BNP). The power difference between the desired signal and output signal, which equals the power of error signal statistically, has been considered the BNP estimate in a rough manner. Considering that the error signal consists of background noise and misalignment noise, a precise BNP estimate is achieved by multiplying the rough estimate with a corrective factor. After the analysis on the power ratio of misalignment noise to background noise of APA, the corrective factor is formulated depending on the projection order and the latest value of variable step-size. The new algorithm which does not require any a priori knowledge of EC environment has the advantage of easier controllability in practical application. The simulation results in the EC context indicate the accuracy of the proposed BNP estimate and the more effective behavior of the proposed algorithm compared with other versions of APA class.
WEN Hao-xiang(文昊翔)1, LAI Xiao-han(赖晓翰)2, CHEN Long-dao(陈隆道)2, CAI Zhong-fa(蔡忠法)2
(1. College of Physics and Mechanical & Electrical Engineering, Shaoguan University, Shaoguan 512005, China;
2. College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China)
Abstract:The adaptive algorithm used for echo cancellation (EC) system needs to provide 1) low misadjustment and 2) high convergence rate. The affine projection algorithm (APA) is a better alternative than normalized least mean square (NLMS) algorithm in EC applications where the input signal is highly correlated. Since the APA with a constant step-size has to make compromise between the performance criteria 1) and 2), a variable step-size APA (VSS-APA) provides a more reliable solution. a nonparametric VSS-APA (NPVSS-APA) is proposed by recovering the background noise within the error signal instead of cancelling the a posteriori errors. The most problematic term of its variable step-size formula is the value of background noise power (BNP). The power difference between the desired signal and output signal, which equals the power of error signal statistically, has been considered the BNP estimate in a rough manner. Considering that the error signal consists of background noise and misalignment noise, a precise BNP estimate is achieved by multiplying the rough estimate with a corrective factor. After the analysis on the power ratio of misalignment noise to background noise of APA, the corrective factor is formulated depending on the projection order and the latest value of variable step-size. The new algorithm which does not require any a priori knowledge of EC environment has the advantage of easier controllability in practical application. The simulation results in the EC context indicate the accuracy of the proposed BNP estimate and the more effective behavior of the proposed algorithm compared with other versions of APA class.
Key words:adaptive algorithm; affine projection algorithm; echo cancellation; background noise power estimate; variable step-size affine projection algorithm
