基于DSP声回声抵消系统的研究

【作者】 成利香；

【导师】 张桂新；

【作者基本信息】 中南大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 回声抵消问题已成为现代通信技术发展的一个瓶颈,系统中回声抵消器的好坏,直接影响到通话质量。现代数字网络通信是公认的最快捷最有效的通信手段,但数字网络通信系统引入更多的时延,使回声现象犹为突出,严重影响到网络通话的质量。回声抵消通常采用自适应技术,该技术也是解决回声问题最具前景的技术之一。回声抵消技术的关键是自适应算法和语音检测两个部分,自适应算法决定了整个系统的收敛的速度、稳态失调、计算复杂度等性能；语音检测的正确与否则影响到回声抵消的效果。本文首先研究了自适应滤波器的一些相关知识,包括自适应滤波器的基本概念、实现的网络结构及最具代表的LMS算法和RLS算法。针对现代数字网络通信中,大时延回声、长阶自适应滤波器等带来的计算量庞大的问题,为寻求一种在保证收敛速度和小稳态误差的情况下有效降低计算复杂度的方法,本文引入了块自适应滤波和快速自适应滤波。即在LMS算法的基础上,进行块处理,利用数字信号处理的相关理论,在频域上快速实现时域LMS。针对块算法带来的收敛速度放慢的特性,提出了采用频域内块数据的欧拉范数,对步长因子进行归一化调整的改进算法——步长归一化FBLMS算法。在Matlab下对步长归一化FBLMS算法进行了收敛速度和稳态失调的仿真测试,并与传统的LMS、时域块LMS和RLS算法进行了比较,验证了步长归一化FBLMS算法的优越性。最后在TMS320C5416实验箱上,基于Geigel算法的双端语音检测技术,设计了步长归一化的FBLMS算法的回声抵消系统,利用CCS Link在Matlab下对回声抵消器进行了测试。另外还在DSP硬件平台上进行了模拟回声的测试。测试结果表明,所设计的回声抵消器,符合ITU—G.168标准仿真测试条件下要求的性能指标。更多还原

【Abstract】 Offsetting of echoes has become the bottleneck of the modern comm-unication technology. An acoustic echo cancellation effects the quality o-f telephoning directly. Modern communications is that it’s the quickest a-nd most effective means of communication, but this system to introduce more time, the phenomenon of the features that will seriously affect the q-uality of the network.The adaptive technology, which is one of the development trend in t-he future,is applied in the acoustic echo cancellation.It includes two parts: the adaptive algorithm and the detection of the sound.The whole system o-f the convergence of speed, the imbalance of the stationary and the comp-utation complex etc are decided by the adaptive algorithm.The effect of t-he echoes offseting is the key to the detection of the sound.In this article,we studied to the adaptive filter of some of the relevan knowledge firstly,including to filter the basic concepts and implementat-ion of network and the representative of the LMS and RLS algorithm. In modern figures for internet communications, as for the echo,long taps to f-ilter for a computational large problems,find a kind of guarantee the con-vergence and a steady state the case,effectively reduce the degree of com-plexity.This article introduced block of the adaptive filter and fast adapta-tion to filter.On the basis of the LMS algorithm,we start with the researc-h for the bloch adaptive algorithm and the digital signal processing of rela-ted theory.The time-domain block LMS was implemented quieckly in th-e frequency domain Of the block algorithm for the convergence of the sp-eed slackened.In the paper the step size was adjusted by the frequency do-main of the data to improving the FBLMS algorithm.The simulation exa-mination for traditional time-domain LMS and frequeney-domain block LMS with normalized step size and RLS adaptive filtering algorithm wer-e completed in the platform of Matlab. While two algorithms were analyz-ed in the size of computational and convergence speed and stationary imb-alance,have provened superiority by FBLMS algorithm with normalize-d step size.Finally,an acoustic echo cancellation system has been made u -sing TMS320C5416DSP, the system was designed to the echo of the FB-LMS algorithms with normalized step size,based on the Geigel algorit-hm and analyed using CCSlink of Matlab.In the next,an acoustic echoc-ancellation was tested simulatly in the hardware platform of DSP.Testing results indicated that are designed to counteract the echoes meet the requi-rements of the ITU-G.168 actual performance indicators.更多还原