LD-ABR: An Adaptive Bitrate Algorithm for Video Transmission in Wireless Network

The popularity of mobile terminals has made video transmission in wireless scenarios important. However, achieving high viewing quality video transmission between base and mobile terminals is a difficult problem. The unpredictable fading and noise generated in wireless scenarios cause drastic fluctuations, making existing adaptive bit rate algorithms unable to adapt to the rapidly changing volatility and long tail problems in the network. In this paper, we introduce LSTM-D3QN Adaptive Bitrate Algorithm (LD-ABR), a new reinforcement learning Adaptive Bitrate Algorithm. LD-ABR uses long and short-term memory (LSTM) to predict throughput, and uses video bit rate, bit rate switching frequency, network speed and video pause time for bit rate selection to better cope with the complex changes in wireless networks. Finally, LD-ABR is compared with Comyco and Pensieve algorithms and the results show that LD-ABR has better performance in wireless network environment. Under the worst network conditions, Pensieve mode has a 16% chance of stopping, Comyco has a 7% chance, and LD-ABR mode has only a 1% chance, and its QoE index is more than 30% higher than Pensieve.