Both space and ground communications have been proven effective solutions under different perspectives in Internet of Things (IoT) networks. This paper investigates multiple-access scenarios, where plenty of IoT users are cooperatively served by a satellite in space and access points (APs) on the ground. Available users in each coherence interval are split into scheduled and unscheduled subsets to optimize limited radio resources. We compute the uplink ergodic throughput of each scheduled user under imperfect channel state information (CSI) and non-orthogonal pilot signals. As maximum-radio combining is deployed locally at the ground gateway and the APs, the uplink ergodic throughput is obtained in a closed-form expression. The analytical results explicitly unveil the effects of channel conditions and pilot contamination on each scheduled user. By maximizing the sum throughput, the system can simultaneously determine scheduled users and perform power allocation based on either a model-based approach with alternating optimization or a learning-based approach with the graph neural network. Numerical results manifest that integrated satellite-terrestrial cell-free massive multiple-input multiple-output systems can significantly improve the sum ergodic throughput over coherence intervals. The integrated systems can schedule the vast majority of users; some might be out of service due to the limited power budget.

翻译：在物联网网络中，空间通信和地面通信已从不同角度被证明是有效的解决方案。本文研究了多址接入场景，其中大量物联网用户由空间中的卫星和地面上的接入点（APs）协作服务。为优化有限的无线电资源，每个相干间隔内的可用用户被划分为调度子集和非调度子集。在不完美信道状态信息（CSI）和非正交导频信号条件下，我们计算了每个调度用户的上行遍历吞吐量。由于在地面网关和APs处本地部署最大比合并，上行遍历吞吐量以闭式表达式获得。分析结果明确揭示了信道条件和导频污染对每个调度用户的影响。通过最大化总吞吐量，系统可同时确定调度用户并基于交替优化的模型驱动方法或基于图神经网络的的学习驱动方法进行功率分配。数值结果表明，集成卫星-地面无蜂窝大规模多输入多输出系统可显著提高相干间隔上的总遍历吞吐量。该集成系统可调度绝大多数用户，但部分用户可能因功率预算有限而无法服务。