【作者】 丁贵广； 曾涛； 戴琼海；

【Author】 Ding Gui-Guang ZENG Tao DAI Qiong-Hai(Broadband Networks & Digital Media Lab of Automation Dept,Tsinghua University,Beijing 100084)

【机构】 清华大学自动化系宽带网数字媒体实验室；

【摘要】 在流媒体的传输中,客户端接收数据前的缓冲区可以起到平滑由网络环境随机变化所带来的抖动的作用。过分的缓冲尽管能够降低接收缓冲的下溢概率,但是却会导致更多的端到端延迟。自适应媒体播放(AdaptiveMediaPlayout,AMP)通过改变音视频媒体的播放速度来平缓缓冲的变化,从而获得下溢概率和延迟间的均衡。本文提出了一种新的基于最小均方学习算法(LMS)的自适应线性神经元(ADALINE)控制的方法,动态地改变媒体的播放速度,使缓冲长度的变化和播放速度变化维持在一定的范围之内,提高了播放缓冲的自适应能力和鲁棒性。最后的实验仿真说明了的ADALINE-LMS算法对播放缓冲的控制性能比原有的AMP方法有了提高。更多还原

【Abstract】 In streaming media, a client buffer can smooth out the jitter, which is due to unreliable network channel condition. An overladen buffer decreases the probability of the client buffer’s underflow, however, it also introduces more significant end-to-end latency. Adaptive media playout (AMP) slower the buffer’s change by varying media playout speed so as to achieve the buffer underflow versus delay trade-off. We present a new scheme with adaptive linear neuron (ADALINE) control based on least mean square (LMS) learn algorithm, changing media playout speed dynamically. With some additional control strategies, this maintains the variations of the buffer’s length and the playout speed in a small steady range, which improves buffer’s adaptivity and robustness. At the later, a simulation system shows an improvement of the control performance of ADALINE-LMS scheme on the original AMP schemes.更多还原