This paper investigates uplink multiple access for the coexistence of enhanced mobile broadband+ (eMBB+) and massive machine-type communications+ (mMTC+) in terminal-centric cell-free massive MIMO (CF-mMIMO) systems. We propose a non-orthogonal scheme in which low-rate mMTC+ transmissions are spread across the time-frequency grid shared with eMBB+ users, enabling efficient resource reuse. In the presence of imperfect channel state information, we derive closed-form expressions for the achievable rates of both services based solely on statistical channel knowledge. For mMTC+ devices, the analysis also incorporates finite blocklength (FBL) modeling to capture short-packet transmissions. To support heterogeneous service requirements, we formulate a power-control problem that maximizes the minimum energy efficiency of mMTC+ devices subject to quality-of-service constraints on eMBB+ users. The resulting nonconvex problem is solved via sequential fractional programming, accounting for both the Shannon and FBL regimes. To enable real-time operation, we further propose a graph neural network (GNN) with multi-head attention to approximate the model-based solution. Constraint satisfaction during training is enforced via an augmented Lagrangian loss. Numerical results demonstrate effective multiplexing of the two data services and show that the proposed GNN algorithm achieves near-optimal performance with a significantly lower computational complexity.

翻译：本文研究了以终端为中心的无小区大规模MIMO（CF-mMIMO）系统中，增强型移动宽带+（eMBB+）与大规模机器类通信+（mMTC+）共存的上行多址接入问题。我们提出一种非正交方案，将低速率mMTC+传输扩展到与eMBB+用户共享的时频网格上，从而实现高效的资源复用。在信道状态信息不完美的条件下，我们仅基于统计信道知识推导出两种业务可达速率的闭式表达式。针对mMTC+设备，分析中还引入了有限块长（FBL）建模以捕捉短数据包传输特性。为支持异构业务需求，我们构建了一个功率控制问题，该问题在满足eMBB+用户服务质量约束的同时最大化mMTC+设备的最小能效。通过结合香农容量与有限块长机制，采用序列分数规划求解所得的非凸问题。为实现实时运算，我们进一步提出一种带多头注意力的图神经网络（GNN）来逼近基于模型的最优解。训练过程中通过增广拉格朗日损失函数强制执行约束满足条件。数值结果验证了两种数据业务的有效复用，并表明所提出的GNN算法能够以显著更低的计算复杂度实现接近最优的性能。