Modified Loss Function Considering Outage Capacity for Deep Learning-Based OFDMA NOMA Video Transmission Resource Management

OFDMA—NOMA is promising for the fifth generation and beyond (B5G) communication systems. Cross physical and application layer resource allocation can achieve better video quality than physical layer only ones in today’s communications where most traffics are videos. Deep learning based OFDMA-NOMA cross layer resource allocation with post processing can satisfy the resource allocation constraints and reduce execution time compared to the conventional non deep learning based approaches. A prior work removed post-processing and further reduced the testing time by introducing one penalty in the loss function, but did not consider the outage capacity, which is a form of fairness and more important in B5G systems. In the paper, we propose a modified loss function with additional penalty (two penalties in total) considering the outage capacity. The result shows that the outage capacity (portion of user with satisfying PSNR, video quality) is higher and the percentage of resource allocation constraint violation is less than the prior work.