文章摘要
崔海徽,王石刚,王高中,蒋志辉.基于前馈神经网络的自适应回声消除方法[J].声学技术,2004,(2):121~124
基于前馈神经网络的自适应回声消除方法
A novel adaptive method for echo cancellation based on feedforward neural networks
投稿时间:2003-10-28  修订日期:2004-03-04
DOI:
中文关键词: 回声消除  前馈神经网络  自适应滤波
英文关键词: echo cancellation  feed-forward neural networks  adaptive filter
基金项目:
作者单位
崔海徽 上海交通大学, 机电设计与自动化所, 上海, 200030 
王石刚 上海交通大学, 机电设计与自动化所, 上海, 200030 
王高中 上海交通大学, 机电设计与自动化所, 上海, 200030 
蒋志辉 三一重工股份有限公司, 长沙, 410100 
摘要点击次数: 1216
全文下载次数: 1638
中文摘要:
      回声消除常用的LMS算法收敛性差,而收敛性好的RLS算法计算量大。文章中提出一种全新的求解方法:基于前馈神经网络的自适应回声消除方法。把回声消除模型中求解滤波器系数的问题转化为前馈神经网络神经元权值的训练问题,并运用误差反向传播算法(BP算法)得出神经元权值的递推公式。经仿真计算,能较好地实现回声消除,与原传统算法LMS和RLS计算比较:该方法能得到非常高的计算精度和明显优越的收敛性能,而计算量只有RLS算法的一半。
英文摘要:
      The LMS algorithm widely used in echo cancellation performs poorly in terms of convergence, and the RLS algorithm with good convergence characteristics, on the other hand, needs considerably more computation overhead. A novel adaptive method based on feed-forward neural networks for echo cancellation is proposed in this paper. Computation of filter weights is converted into training of neural network weights. Recursive formulas of the neuron weights are derived with the BP algorithm. Simulations show that the algorithm has a better performance in echo cancellation. The results of echo calculation indicate that the precision and convergence are significantly improved compared to LMS and RLS algorithms, and the computation overhead is only half of the RLS algorithm.
查看全文   查看/发表评论  下载PDF阅读器
关闭