TPMI:Accurate Throughput Prediction for Better Bitrate Selection in Adaptive Video Streaming
Jiang Liu, Jun Li, XiaoHan Yang, MengTing Sun · 2023
The ABR algorithm, as a crucial component of DASH technology, fundamentally aims to maximize the Quality of Experience (QoE) for users. It dynamically selects the bitrate of video chunks based on current network conditions, client-side cache status, and the processing capabilities of the device. Most existing ABR algorithms predict current throughput based solely on historical data; however, in wireless communication and mobile networks, where bandwidth is complex and variable, simple throughput prediction is inadequate. Inaccurate throughput information can severely impact the performance of ABR algorithms, thus affecting the user’s QoE. We have designed and implemented an accurate throughput predictor, named TPMI1, which can be integrated into any ABR algorithm that considers throughput as a factor, to enhance the performance of the ABR algorithm. This predictor is built on the informer model and considers a broader range of factors influencing throughput prediction as input features, such as historical throughput, block size, network connection type, signal strength, and player status. We validated its prediction performance in real network environments and compared it with prediction methods based on HM, MLR, and LSTM. Integrating TPMI into the MPC algorithm, we observed a 9.67% increase in QoE compared to the original MPC algorithm, and a 4.5% to 19.16% increase in average QoE over pensieve and BBA when using TPMI+MPC.