Cross-Layer Resource Management for Downlink BF-NOMA-OFDMA Video Transmission Systems and Supervised/Unsupervised Learning Based Approach

Ali
et al.
proposed a physical (PHY) layer resource management for single-carrier N-antenna beamforming (BF) non-orthogonal multiple access (NOMA) systems. Cross PHY/application (APP) layer resource allocation and adaptive source encoding rate for single-antenna NOMA/ orthogonal frequency division multiple access (OFDMA) video communications was also recently proposed. These prior works did not deal with joint subcarrier/ BF cluster assignment in a cross-layer fashion. In this paper, we propose a joint adaptive source encoding rate and iterative cross-PHY/APP layer resource management for BF-NOMA-OFDMA video communications to support 2N users on one subcarrier and improves the average video quality peak-signal-to-noise ratio (PSNR). The proposed scheme outperforms Ali
et al.
PHY layer resource management by 4.18dB in the numerical results. Considering the high computational complexity of the iterative resource allocation, we further propose replacing the iterative cross-layer resource allocation by a supervised learning-based approach. We propose a novel and problem-specific post-processing procedure to guarantee the constraints that every user holds at least one subcarrier and a subcarrier support 2N users. The proposed supervised learning-based approach reduces the execution time by 98% in a GPU-based platform, comparable with 91-99% execution time reduction of supervised learning based communication algorithms in the literature, at the cost of 4% PSNR loss. Finally, we propose a supervised/unsupervised learning -based approach with minus PSNR as the cost function for continuous learning in an unsupervised learning fashion after supervised learning is done. The numerical results show that the proposed hybrid supervised/unsupervised learning approach can reduce the PSNR loss to just 1%.