QQ客服 官方微博 反馈留言

题名	频域自适应滤波算法及其在声学回声抵消中的应用
作者	周名远
学位类别	博士
答辩日期	2008-05-29
授予单位	中国科学院声学研究所
授予地点	声学研究所
关键词	频域自适应滤波器 声学回声抵消 抗频域混叠规整 频域可变步长 残余回声抑制
其他题名	On Frequency-Domain Adaptive Filter Algorithms and Their Application to Acoustic Echo Cancellation
学位专业	信号与信息处理
中文摘要	声学回声不可避免地存在于远程会议系统、免持电话、蓝牙耳机式送受话器、助听器等语音通信终端中，若对其处理不当，则通话质量将受到严重影响。针对该问题，本文一方面深入研究了自适应滤波器这一声学回声抵消系统核心构件的性能改进，另一方面着力于将功率谱估计规整、步长控制、后处理滤波等和自适应滤波器紧密结合，来构建一个的实用的、可灵活配置的、完备的声学回声抵消系统。 自适应滤波器作为信号处理和自动控制的重要工具，广泛地应用于音频信号处理、通信、雷达、声纳和生物医学工程等领域。在数目众多的自适应滤波算法中，频域类算法凭借其快速的收敛速率和低计算复杂度获得了广泛关注，但是这类算法中存在的频域混叠、步长控制困难、传输时延、跟踪能力下降等突出问题限制了其在实际系统中应用。 依托声学回声抵消这一重要实际应用，本文深入研究了频域自适应滤波器（FDAF, frequency-domain adaptive filter）算法和分块频域自适应滤波器（PBFDAF, partitioned block frequency-domain adaptive filter）算法这两种典型的频域类算法。本文先是构造了性能优异的抗频域混叠规整因子，接着分析推导出了频域最优步长的理想值并给出了行之有效的估计方法，然后给出了维纳后处理滤波器的计算方法。将这三者和选定的频域自适应滤波器结合在一起便构成了一个实用的完备的频域声学回声抵消系统。针对频域类算法存在传输时延和跟踪能力下降的问题，本文给出了消除传输时延的方法，并进一步引入了时频联合更新的框架来加快其跟踪速率。本文另外还讨论了应用于蓝牙耳机式送受话器和助听器等短阶系统的无需自适应滤波器的声学回声抑制算法，以及语音增强中的先验信噪比的估计。本文最后简要地介绍了级联型频域自适应滤波器。 如何将本文针对单通道频域自适应声学回声抵消系统提出的抗频域混叠规整、频域可变步长、后处理滤波、传输延迟消除、时频联合更新、级联型结构等多项独创性技术扩展到频域自适应滤波算法所适用的多通道自适应声学回声抵消、有源噪声控制、自适应广义旁瓣相消等其它应用值得进一步研究。
英文摘要	Acoustic echo, which commonly exists in teleconference systems, hands-free phones, Bluetooth headsets, hearing aids, and many other voice communication terminals, would seriously degrade the voice communication quality if not properly handled. To solve this problem, we on the one hand focus on the improvement of the adaptive filter, which is the key component in acoustic echo cancellation (AEC), and on the other hand, we construct a practically applicable, flexible, and complete AEC system which integrates the regularization for power spectral estimation, step-size control, and postfiltering with the adaptive filter. The adaptive filter, a powerful tool for signal processing and automatic control applications, has been successfully used in the fields as diverse as audio processing, communications, rader, sonar, and biomedical engineering. Among quite a number of adaptive filter algorithms, the frequency-domain ones are well-known for their fast convergence and low computational complexity, but their applications in practical scenarios are restricted by the frequency-domain aliasing effects, lack of appropriate step-size control methods, transmission delay, and degraded tracking capability. In this paper, two typical frequency-domain adaptive filter algorithms, the frequency-domain adaptive filter (FDAF) and the partitioned block frequency-domain adaptive filter (PBFDAF), are thoroughly analyzed and evaluated in the context of AEC. We construct the anti-frequency-domain-aliasing regularization parameter in the first place, next we derive an optimal bin-wise variable step-size and propose an estimation method, and then we present a postfiltering scheme for residual echo suppression. Integrating the three components with a frequency-domain adaptive filter, we get a frequency-domain AEC system ready for practical applications. To solve the problems of transmission delay and degraded tracking capability, we further propose a method to eliminate the transmission delay and a time/frequency-domain unified adaptation scheme to accelerate the tracking speed. Besides the work mentioned above, an acoustic echo suppression algorithm without the need of adaptive filtering is proposed for short-tap systems, such as Bluetooth headsets and hearing aids, and the estimation of the A Priori signal-to-noise ratio in speech enhancement systems is discussed. The structure of a cascaded frequency-domain adaptive filter is briefly described in the end of the paper. Further research efforts will be devoted to extending the proposed novel techniques, including the anti-frequency-domain-aliasing regularization, variable step-size, postfiltering scheme, transmission delay elimination, time/frequency-domain unified adaptation, and cascaded structure, from the single-channel AEC application to the multi-channel AEC, active noise control, adaptive generalized sidelobe canceller, and other applications perferring frequency-domain adaptive filters.
语种	中文
公开日期	2011-05-07
页码	127
内容类型	学位论文
源URL	[http://159.226.59.140/handle/311008/422]
专题	声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文
推荐引用方式 GB/T 7714	周名远. 频域自适应滤波算法及其在声学回声抵消中的应用[D]. 声学研究所. 中国科学院声学研究所. 2008.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。