Adaptive Media Playout Based on Multiplayer Neural Network Control

ZENG Tao,DAI Qionghai
DOI: https://doi.org/10.3321/j.issn:1000-0054.2006.01.035
2006-01-01
Abstract: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.
What problem does this paper attempt to address?