A BAC-NOMA Design for 6G umMTC With Hybrid SIC: Convex Optimization or Learning-Based?

Abstract

This paper presents a new backscattering communication (BackCom)-assisted non-orthogonal multipleaccess (BAC-NOMA) transmission scheme for device-to-device (D2D) communications. This scheme facilitates energy and spectrum cooperation between BackCom devices and cellular downlink users in 6G ultra-massive machine-type communications (umMTC) scenarios. Given its quasi-uplink nature, the hybrid successive interference cancellation (SIC) is applied to further improve performance. The data rate of BackCom devices with high quality of service (QoS) requirements is maximized by jointly optimizing backscatter coefficients and the beamforming vector. The use of hybrid SIC and BackCom yields two non-concave sub-problems involving transcendental functions. To address this problem, this paper designs and compares convex optimization-based and unsupervised deep learning-based algorithms. In the convex optimization, the closed-form backscatter coefficients of the first sub-problem are obtained, and then semi-definite relaxation (SDR) is utilized to design the beamforming vector. On the other hand, the second sub-problem is approximated by using a combination of sequential convex approximation (SCA) and SDR. For unsupervised deep learning-based optimization, a loss function is properly designed to satisfy constraints. Computer simulations show the following instructive results: i) the superiority of the hybrid SIC strategy; ii) the distinct sensitivities and efficacies of these two algorithms in response to varying parameters; iii) the superior robustness of the unsupervised deep learning-based optimization.