Optimized Resource Allocation and Caching for Integrated Satellite–Terrestrial Networks Using Multiagent Deep Learning

ABSTRACT The amalgamation of terrestrial and satellite networks for nonorthogonal multiple access services presents a viable approach to improve the energy effectiveness and decrease the latency in communication systems. However, existing approaches face drawbacks such as high computational complexity, limited scalability, and difficulties in handling multiagent coordination, especially in large‐scale satellite–terrestrial networks. This research presents a hybrid optimization scheme that maximizes energy efficiency and minimizes delays based on vital considerations like base station (BS) satellite placement, caching strategy, user association, and transmission power control. A three‐phase approach is suggested to tackle these issues, relying on collaboration Q‐learning with convolutional neural network (C‐QCNN) and binary meerkat–hippopotamus optimization for power‐control enhancement, cache architecture, and user association. It is observed that the proposed C‐QCNN achieves the highest resource utilization of 99.23% when compared with existing techniques. This indicates that C‐QCNN effectively allocates and utilizes resources to the enhancement of overall network performance. Significant improvements in energy efficiency are found in integrated satellite–terrestrial networks, thereby enabling more reliable and efficient nonorthogonal multiple access services. The significance of the proposed approach lies in its emphasis on advanced techniques in resource allocation and network optimization for next‐generation communication systems.