Adaptive rate decision algorithm for dash based on deep reinforcement learning

Abstract

Aiming at the problems of the existing HTTP-based Dynamic Adaptive Streaming over HTTP (DASH) bit rate adaptive algorithm in mobile scenes, wireless network environment, frequent bit rate switching and low average bit rate, a problem is proposed. DASH rate adaptive decision algorithm based on deep reinforcement learning. Using a combination of deterministic strategy gradient algorithm and deep neural network in reinforcement learning, experiments on real network trajectory data sets prove that this algorithm can obtain higher user experience quality than existing algorithms.