| 摘 要: 为提升数字助听器回声消除算法的收敛速度及降低稳态误差,提出基于零吸引技术的SM-L0-IPNLMS算法。SM-L0-IPNLMS算法虽降低了计算复杂度,但它的小系数收敛速度还是较慢。为此,将零吸引函数与SM-L0-IPNLMS算法结合,在代价函数中引入零吸引项,使小幅值系数快速归0,仅迭代大幅值系数,以提高收敛速度。仿真结果显示,相比原算法,新算法在保留原算法特性的同时,以随机信号和真实语音为输入时,均方误差分别降低0.79、1.56dB,回声返回损失增强ERLE(Echo Return Loss Enhancement)值分别升高0.43、2.03dB,且在低信噪比输入下鲁棒性强,通过实验结果验证了新算法的可行性和性能优势。 |
| 关键词: 零吸引算法 回声消除 收敛速度 稳态误差 |
|
中图分类号:
文献标识码: A
|
| 基金项目: 江苏省自然科学基金项目(BK20171303) |
|
| Improved Echo Cancellation Algorithm Based on Zero-Attracting Technique |
|
QU Zongyi, MA Lingkun, YE Anlong
|
(School of Electronic Information and Artificial Intelligence, Shaanxi University of Science and Technology, Xi’an 710021, China)
1061767188@qq.com; mlk8685@outlook.com; a861829595@163.com
|
| Abstract: To enhance the convergence speed and reduce the steady-state error of echo cancellation algorithms in digital hearing aids, this paper proposes the SM-L0-IPNLMS algorithm incorporating zero-attracting techniques. While the SM-L0-IPNLMS algorithm reduces computational complexity, it still exhibits slow convergence for small coefficients. To address this limitation, a zero-attracting function is integrated into the algorithm. By introducing a zero-attracting term into the cost function, smal-l magnitude coefficients rapidly approach zero, allowing the algorithm to focus solely on updating large-magnitude coefficients and thereby accelerating convergence. Simulation results demonstrate that, compared with the original algorithm, the proposed algorithm retains the fundamental characteristics of its predecessor while achieving a reduction in Mean Square Error (MSE) by 0.79 dB and 1.56 dB for random signals and real speech inputs, respectively. Concurrently, the Echo Return Loss Enhancement (ERLE) values increase by 0.43 dB and 2.03 dB. The algorithm also exhibits strong robustness under low signa-l to-noise ratio (SNR) conditions.Experimental results validate the performance advantages and feasibility of the proposed algorithm. |
| Keywords: zero-attracting algorithm echo cancellation convergence speed steady-state error |
