CAREER: Agile Speech and Audio Coding and Transmission over Heterogeneous Networks

在紧急情况下，通信网络的特点是链路故障和分组丢失方面的高度可变性。然而，当前的互联网协议语音（VoIP）方法在紧急情况期间提供不令人满意的性能，既缺乏适应苛刻网络条件的灵活性，又缺乏用于测量语音服务质量（QoS）的系统框架。 本研究的目标是提供一个VoIP的框架和方法，适用于苛刻的网络情况。通过提供在紧急情况或远程学习等其他要求较高的情况下实现语音通信的方法，这项研究将为社会提供不可估量的好处。 信道和网络的高度动态特性需要跨层设计方法，其中跨协议栈的所有层的参数被优化以有效地最大化应用QoS。为了将VoIP集成到跨层设计框架中，本研究提供了一种新的语音和音频编码范例以及几种新的错误控制和传输方法。特别是，这项研究提供了一个功能分层的方法来语音和音频编码，打破了编码器的操作成三个上升，分层：波形层，统计层，和预期失真层。在波形层，信号被分解为感知相关的对象，其随时间的演变在统计层被跟踪。在统计层执行量化，方向来自预期失真层。期望失真层计算各种编码器、差错控制和传输参数设置的语音QoS，允许基于自适应资源分配的跨层优化。这种模式产生了一个独特的统一模型的编码和数据包丢失隐藏，使感知自适应数据包保护，传输和流量工程随时间变化的异构有线和无线网络。

During emergencies, communication networks are characterized by their high degree of variability in terms of link failure and packet loss. However, current Voice over Internet Protocol (VoIP) methods deliver dissatisfactory performance during emergencies, lacking both the agility to adapt to demanding network conditions, and a systematic framework for measuring voice Quality-of-Service (QoS). The goal of this research is to provide a VoIP framework and methods suitable for demanding network situations. By furnishing methods for achieving voice communication during emergencies or other demanding situations such as distance-learning, this research will provide an immeasurable benefit to society. The highly dynamic nature of channels and networks necessitates cross-layer design approaches in which parameters across all layers of a protocol stack are optimized to efficiently maximize application QoS. To integrate VoIP into a cross-layer design framework, this research offers a new paradigm for speech and audio coding together with several novel error control and transmission methods. In particular, this research provides a functional layering approach to speech and audio coding, breaking up the operation of the coder into three ascending, hierarchical layers: the Waveform Layer, the Statistical Layer, and the Expected Distortion Layer. At the Waveform Layer, a signal is decomposed into perceptually relevant objects whose evolution through time is tracked at the Statistical Layer. Quantization is performed at the Statistical Layer with direction from the Expected Distortion Layer. The Expected Distortion Layer calculates voice QoS for various coder, error-control, and transmission parameter settings, allowing cross-layer optimization based adaptive resource allocation. This paradigm engenders a unique unified model for coding and packet loss concealment, enabling perceptually-adaptive packet protection, transmission, and traffic engineering over time-varying heterogeneous wired and wireless networks.