With the intent of ubiquitous global and massive connectivity, non-terrestrial networks (NTN) has drawn extensive attention as one of the key technologies for the 6G mobile communication networks and beyond. In this thesis, the communication performance of NTN based on rate-splitting multiple access (RSMA) is optimized with the consideration of the imperfect channel state information (CSI) caused by the high altitude and rapid movement of aerial base stations (ABSs). On top of this, to capture more realistic scenarios in NTN, heterogeneous traffic demands of users due to the different geological conditions and user distributions, and the limited power of ABSs due to the unstable power supply and finite-sized battery are considered. To be specific, the RSMA-based rate-matching (RM) framework is proposed that minimizes the difference between traffic demands and actual offered rates in multibeam satellite communications (SATCOM). Channel phase perturbations arising from channel estimation and feedback errors are taken into account. To solve the non-convex formulated problem, it is converted into a tractable convex form via the successive convex approximation (SCA) approach. Simulation results show that RSMA flexibly arranges the powers of the common and private messages according to different traffic patterns between beams and users, efficiently satisfying non- uniform traffic demands. Second, to tackle the unstable supplied power and limited battery at ABSs, the joint power and beamforming framework in RSMA-based energy harvesting unmanned aerial vehicle (UAV) networks is proposed. A deep reinforcement learning (DRL) approach is utilized to allocate optimal trans- mission power at each time slot from harvested energy. The optimal power allocation strategy is determined according to the channel, harvested energy, and battery power status to maximize the sum rate from the long-term perspective. To design the RSMA precoder maximizing the sum-rate at each time slot with a given transmission power via DRL, sequential least squares programming (SLSQP) based on the Han–Powell quasi-Newton method is adopted.
