ARMPC - ARIMA based prediction model for Adaptive Bitrate Scheme in Streaming

Abstract

With the boom of the internet, online streaming algorithms such as Adaptive bitrate streaming have gained prominence. The Adaptive Bitrate (ABR) scheme uses the Model Predictive Control (MPC) to determine the best possible bitrate for the given network conditions. Though this method works well, the major disadvantage is its heavy reliance on the throughput prediction error which makes it difficult to perform well in congested network conditions. Other methods such as DeepMPC have also been explored in this paper which use the Deep Learning algorithms to predict the bandwidth. These work better than the trivial harmonic predictor but demand high computational power. This paper proposes ARMPC which uses the Auto-Regressive Integrated Moving Average Technique (ARIMA) to predict the future bandwidth. Using trace-driven experiments, we have shown both mathematically and practically that the ARMPC can provide us with improvements in both the prediction and the computational points of view.