声反馈抑制需求鲁棒、 高效的自适应滤波器。该文提出一种基于Gauss-Seidel伪放射投影（Gauss-Seidel pseudo affine projection，GSPAP）的子带自适应声反馈消除算法。通过加权重叠相加滤波器组进行子带划分，子带上采用参考信号的自相关矩阵取代能量对滤波器的自适应步长进行归一化；在自相关矩阵的迭代公式中引入基于输入信号和参考信号能量最大值的自整定系数，增强算法鲁棒性；选用二阶 GSPAP 迭代法对自相关矩阵解算求逆，以平衡算法性能与复杂度。实验结果表明：在相同滤波器长度的条件下，该文方法获得11~22 dB的稳态增益增量，比时域归一化最小均方误差（normalized least mean square，NLMS）方法提升20%~55%。
Acoustic feedback cancellation needs robust and efficient adaptive filters. A Gauss-Seidel pseudo affine projection (GSPAP) sub-band adaptive filter based algorithm was developed using the auto-correlation matrix of the reference signal in the sub-band domain, instead of the power, to normalize the adaptation step size. A self-adjusting coefficient calculated from the maximum power of the input and output sub-band signals, was used to improve the robustness of the auto-correlation matrix. A two-order GSPAP algorithm was used to find the inverse of the auto-correlation matrix to balance the performance and complexity. Results from the added stable gain (ASG) experiment indicate that this algorithm gives 11 to 22 dB improved feedback cancellation for different types of hearing aids, which is 22% to 55% better than given by the time-domain normalized least mean square (NLMS) algorithm.
[1] Kates J M. Digital Hearing Aids [M]. San Diego, USA: Plural Publishing INC, 2008. [2] Nakagawa C R C, Nordholm S, YAN Weiyang. Feedback cancellation with probe shaping compensation [J]. IEEE Signal Processing Letters, 2014, 21(3): 365-369. [3] Hellgren J. Analysis of feedback cancellation in hearing aids with Filter-X LMS and the direct method of closed loop identification [J]. IEEE Transactions on Speech and Audio Processing, 2002, 10(2): 119-131. [4] MA Guilin. Feedback Suppression in Digital Hearing Instruments [D]. Ballerup, Denmark: Technical University of Denmark, 2010. [5] Wyrsch S, Kaelin A. Adaptive feedback cancellation in subbands for hearing aids [C]//Proceedings of the International Conference on Acoustic, Speech and Signal Processing (ICASSP'99). Phoenix, AZ, USA: IEEE, 1999: 921-924. [6] Gil-Cacho J M, Waterschoot V T, Moonen M, et al. Transform domain prediction error method for improved acoustic echo and feedback cancellation [C]//Proceedings of the 20th European Signal Processing Conference. Bucharest, Romania: EURASIP, 2012: 2422-2426. [7] Vanderkooy J. Aspects of MLS measuring systems [J]. Journal of the Audio Engineering Society, 1994, 42(4): 219-231. [8] Lee J, Park Y C, Youn D H. Robust pseudo affine projection algorithm with variable step-size [J]. Electronic Letters, 2008, 44(3): 250-252. [9] Sheikhzadeh H, Whyte K R L, Brennan R L. Partial update subband implementation of complex pseudo-affine projection algorithm on oversampled filterbanks [C]//Proceedings of the International Conference on Acoustic, Speech and Signal Processing (ICASSP 2005). Philadelphia, PA, USA: IEEE, 2005, 4: 373-376. [10] On Semiconductor. WOLA Filterbank Coprocessor: Introductory Concepts and Techniques[R/OL]. April, 2009. http://www.onsemi.com /pub/Collateral/AND8382-D.PDF. [11] Valente M. Hearing Aids: Standards, Options, and Limitations [M]. 2nd ed. New York, USA: Thieme Medical Publishers, 2002. [12] Frye Electronics INC. FONIX 8000 Hearing Aid Test System Operator's Manual Version 2.41[R/OL]. August, 2013. http://www.frye.com/wp/wp-content/uploads/2013b/manuals /8000v2.41.pdf. [13] 厦门莱亚特医疗器械有限公司. 一种测量助听器反馈路径的方法: 中国, CN201410685349[P]. 2015.01.28.Xiamen LA and Associates Medical Equipment Co Ltd. A Method for Measuring the Feedback Path of Hearing Aids: China, CN201410685349[P]. 2015.01.28. (in Chinese)
