基于补偿因子自适应的NLMS多媒体播放算法

A Novel NLMS Media Playout Algorithm Based on Compensation Factor Adaption

基于线性预测的媒体播放算法由于其复杂度低、易实现,在网络多媒体应用中大量使用,其中采用最小均方差估计的NLMS算法精确度较高.首先对NLMS算法和改进的NLMS算法进行了分析,针对算法中补偿因子因网络抖动剧烈而导致补偿过大的问题,提出了β因子自适应的NLMS算法,采用补偿因子系数β随网络变化以及预测误差情况自适应调整的方法,对NLMS算法作了进一步改进,明显地减小了端到端延迟均值和丢包率.试验结果验证了算法改进的有效性.

Media playout algorithms based on the prediction of future packet's delay has been widely used in network multimedia applications. Linear prediction is favored for people because of its low complexity and easy realization. NLMS(normalized least mean square) algorithm is one of the methods with better accuracy. The factors affecting the performance of media playout algorithm involves two aspects: (1) the accuracy of delay prediction; (2) the property of the compensation factor. In almost all algorithms, the compensation factor coefficient β is a constant. So it may occur that the compensation factor is too large to adjust the total end to end delay in jitter situation. It will induce significant error between the predicted end-to-end delay and actual delay. In the ENLMS(enhanced-NLMS) method, although delay spike detection had been utilized, but the improvement is not significant. The reason for this is that the compensation redundancy is not match to adjust the delay prediction, when the network jitter is happening, the processing of as a constant is too simple. Therefore this paper presents a β-adaptive-NLMS method to improve the accuracy of delay prediction. The improved β is adapted according to the network circumstance and the error range between the predicted network delay and the actual delay. In the proposed β-adaptive method, β is decreased by 0.0625 when predicted value of delay is close enough to actual value (for example, prediction error range less than (β-1)v-i ). And β is increased by 0.125 when the estimation error is larger than βv-i. It makes the predicted value following actual delay adaptively. To guarantee the rate at which predicted value follows actual delay, the coefficient β is selected in the range of [3.5, 5.0]. Simulation results show the proposed adaptation method is effective in reducing the total end-to-end delay and packet loss. In the condition of the same average end-to-end delay, the packet loss ratio of β-adaptive-NLMS is decreased 10 to 90 percent comparing to the NLMS algorithm.