DeepWiVe: Deep-Learning-Aided Wireless Video Transmission

We present DeepWiVe , the first-ever end-to-end joint source-channel coding (JSCC) video transmission scheme that leverages power of deep neural networks (DNNs) to directly map signals channel symbols, combining compression, coding, and modulation steps into a single transform. Our DNN decoder predicts residuals without distortion feedback, which improves quality by accounting for occlusion/disocclusion camera movements. simultaneously train different bandwidth allocation frames allow variable transmission. Then, we network using reinforcement learning (RL) optimizes limited available among maximize overall visual quality. results show can overcome xmlns:xlink="http://www.w3.org/1999/xlink">cliff-effect is prevalent in conventional separation-based digital communication schemes, achieve graceful degradation with mismatch between estimated actual qualities. outperforms H.264 compression followed low-density parity check (LDPC) codes all conditions up 0.0485 terms multi-scale structural similarity index measure (MS-SSIM), H.265+ LDPC 0.0069 on average. also illustrate importance optimizing JSCC showing our optimal policy superior uniform as well heuristic benchmark.