Control-Theoretic Rate Control for Adaptive Packet Video Transmission over Time-Varying Channels

摘要

We present an adaptive rate control algorithm for VBR packet video transmission over time-varying channels. A new rate-quantization (R-Q) model is proposed and incorporated into the algorithm. The novelty of the algorithm has five folds: First, even in the presence of abrupt changes in video scene and transmission bandwidth from the network, it effectively maintains the encoder buffer occupancy in the vicinity of the target length. Thereby achieving high utilization of available bandwidth as well as minimizing the data loss due to encoder buffer overflow. Second, it hardly changes the quantizer_scale of the encoder unless the changes in scene or available bandwidth occur, periodic changes of picture types due to the motion estimation do not affect the video quality. Third, the R-Q model estimation not only effectively executes the rate allocation but is also computationally efficient so that it could allow for real-time encoding. Fourth, the algorithm is extended to maintain the packet delay at the encoder buffer either within a bound or in the vicinity of a target value in the average sense. Finally, unlike many existing algorithms, the performance of the algorithm has solid theoretical foundation; asymptotic stability and fast convergence is guaranteed. Through simulations we show that the proposed algorithm significantly outperforms the well-known TM-5 algorithm and is a promising solution to VBR video transmission over time-varying channels such as wireless fading channels, ATM ABR channels and TCP/IP channels.