An Adaptive Video Compression Approach Utilizing Deep Reinforcement Learning for Next-Generation Wireless Networks

Abstract

High-resolution video transmission in next-generation wireless networks requires adaptive, error-resilient compression. In this work a Deep Reinforcement Learning-based Error Controlled Adaptive Transmission (DRL-ECAT) framework is proposed, which integrates low-rank signal modeling with a reinforcement learning agent for dynamic compression rank selection and bit allocation based on channel and buffer conditions. DRL-ECAT is evaluated on the Ultra Video Group (UVG) and High Efficiency Video Coding (HEVC) Class B datasets and compared with the Hierarchical Random-Access Coding (HRAC), Hyperprior-Driven Video Compression (HDVC), and Transformer-based Video Compression (TVC). The results show significant gains in Peak Signal-to-Noise Ratio (PSNR) and Multi-Scale Structural Similarity Index Measure (MS-SSIM), indicating improved fidelity and perceptual quality. On UVG, PSNR improves by 20.9%, 13.1%, and 7.9% over HRAC, HDVC, and TVC, while HEVC Class B shows 17.7%, 15.2%, and 11.6% gains. MS-SSIM improvements further confirm the proposed model's robustness under varying bitrates and channel conditions.