Non-orthogonal multiple access (NOMA) emerges as a superior technology for enhancing spectral efficiency compared to orthogonal multiple access. In NOMA networks, successive interference cancellation (SIC) plays a crucial role in decoding user signals sequentially. The challenge lies in the joint optimization of SIC ordering and power allocation, due to the factorial nature of ordering combinations. This study introduces an innovative solution, the Attention-based SIC Ordering and Power Allocation (ASOPA) framework, targeting an uplink NOMA network with dynamic SIC ordering. ASOPA aims to maximize weighted proportional fairness by employing deep reinforcement learning, strategically decomposing the problem into two manageable subproblems: SIC ordering optimization and optimal power allocation. Our approach utilizes an attention-based neural network, which processes instantaneous channel gains and user weights to determine the SIC decoding sequence for each user. Once the SIC ordering is established, the power allocation subproblem transforms into a convex optimization problem, enabling efficient calculation. Extensive simulations validate ASOPA's efficacy, demonstrating a performance closely paralleling the exhaustive method, with over 97% confidence in normalized network utility. Notably, ASOPA maintains a low execution latency of approximately 50 milliseconds in a ten-user NOMA network, aligning with static SIC ordering algorithms. Furthermore, ASOPA demonstrates superior performance in various NOMA network configurations, including scenarios with imperfect channel state information, multiple base stations, and multiple-antenna setups. Such results underscore ASOPA's robustness and effectiveness, highlighting its ability to excel across various NOMA network environments. The complete source code for ASOPA is accessible at https://github.com/Jil-Menzerna/ASOPA.

翻译：相较于正交多址接入，非正交多址接入技术因其在提升频谱效率方面的优越性而备受关注。在非正交多址网络中，串行干扰消除技术对于用户信号的顺序解码至关重要。由于排序组合具有阶乘级的复杂性，联合优化串行干扰消除排序与功率分配成为一项挑战。本研究针对具有动态串行干扰消除排序的上行非正交多址网络，提出了一种创新解决方案——基于注意力的串行干扰消除排序与功率分配框架。该框架旨在通过深度强化学习最大化加权比例公平性，并策略性地将原问题分解为两个可处理的子问题：串行干扰消除排序优化与最优功率分配。我们的方法采用基于注意力的神经网络，通过处理瞬时信道增益与用户权重来确定每个用户的串行干扰消除解码顺序。一旦串行干扰消除排序确定，功率分配子问题即转化为凸优化问题，从而实现高效求解。大量仿真实验验证了该框架的有效性，其性能与穷举法高度接近，在归一化网络效用方面达到超过97%的置信度。值得注意的是，在十用户非正交多址网络中，该框架的执行延迟保持在约50毫秒的低水平，与静态串行干扰消除排序算法相当。此外，该框架在多种非正交多址网络配置中均表现出优越性能，包括非理想信道状态信息、多基站以及多天线等场景。这些结果充分证明了该框架的鲁棒性与有效性，凸显了其在各种非正交多址网络环境中的卓越适应能力。该框架的完整源代码可在 https://github.com/Jil-Menzerna/ASOPA 获取。