基于多层神经网络控制的动态自适应播放

Adaptive media playout based on multiplayer neural network control

基于流媒体客户端缓冲的被动变化可以通过自适应媒体播放（AMP）技术修改播放速度来调节，为了减轻AMP中速度突变对播放质量的影响，提高变速的性能，提出了一种基于多层神经网络控制的AMP方法，产生随当前缓冲动态变化的速度，并且保持缓冲稳定在一定的范围内。该方法引入了多层神经网络控制结构，并采用反向传播学习算法（BP）进行离线训练。仿真结果表明：该方法产生的速度平均幅值和变化增量可以减少0～100％，性能比原AMP方法更好。

Adaptive media playout （AMP） actively adjusts to the passive variations of the streaming client buffer＇s size by changing the playout speed to improve the stability. An AMP scheme based on the multilayer neural network control was developed to reduce the speed variations in previous AMP systems and improve the performance. The system dynamically exports the speed variations based on the current buffer＇s status and ensures that the buffer is stable inside a fixed range. The system uses a multilayer neural network architecture trained offline by a backpropagation （BP） learning algorithm. Extensive simulations confirmed that the scheme gives 0 - 100% reduction of the average speed variations, which is better than the original AMP method.