Learning-Oriented Feedback-Free Transmission and Resource Management in Space-Air-Ground Integrated FD-RAN

The integration of space, air, and ground networks into a fully decoupled radio access network (FD-RAN) is emerging as a promising approach for 6 G, driven by the need for seamless coverage, flexible spectrum allocation, and efficient collaboration among heterogeneous nodes. However, the inherent differences in wireless environments across terrestrial, aerial, and satellite segments pose challenges for traditional feedbackbased transmission. In response to these challenges, this paper introduces a space-air-ground integrated FD-RAN architecture where users can be served by multiple nodes with adaptive resource block (RB) allocation. For downlink transmissions in base stations (BSs), a feedback-free approach is developed by employing a deep learning-based channel state information (CSI) prediction framework, allowing BSs to perform multipleinput multiple-output (MIMO) transmissions using only user geolocation. To enhance cooperation and RB allocation among heterogeneous nodes, a many-to-one matching model is proposed, achieving stable matching with low complexity and fast convergence. Simulation results validate the effectiveness of the proposed framework, showing a 70 % improvement in spectrum efficiency compared to single-connection networks based on optimal path loss and round-robin resource scheduling.