DeepMPC: a mixture ABR approach via deep learning and MPC

The leading adaptive bitrate (ABR) algorithm uses model predictive control (MPC) method to determine next chunks' video bitrate, while it heavily relies on the accuracy of throughput prediction, which thereby fails to perform well in all considered network scenarios. In this paper, we ask if deep learning can help taming the weakness of MPC to further improve the performance. We propose DeepMPC, which enhances MPC via two modules: DL-based Throughput Predictor (DTP), employing deep learning (DL) to predict future bandwidth, and Discounted Factor Optimizer (DFO), utilizing deep reinforcement learning (DRL) to determine the proper discounted factor γ. Using trace-driven experiments, we illustrate that DeepMPC outperforms existing ABR schemes in all considered network conditions. In particular, DeepMPC works better than the state-of-the-art ABR scheme Pensieve, with the improvements on average QoE of 4%-5.91%. Moreover, we implement DeepMPC in real-world network environments and extensive experimental results demonstrate the superiority of DeepMPC against existing state-of-the-art approaches.