Imperfect-Spectrum-Sensing-Aware DRL-Based Dynamic Spectrum Access for Underlay Satellite Systems with Multiple Primary Transmit Power Levels Sensing

This paper investigates a dynamic spectrum access (DSA) scheme for the underlay scenario of coexistence between geostationary orbit (GSO) and nongeostationary orbit (NGSO) satellite systems, in which NGSO satellites can estimate the status of GSO satellites via spectrum sensing to avoid interfering with GSO systems when selecting a spectrum sharing strategy. However, in practice, an NGSO system cannot obtain completely correct information because of sensing errors. Therefore, the DSA problem considering imperfect spectrum sensing is regarded as a partially observable Markov decision process (POMDP) in this work. A deep reinforcement learning (DRL)- based algorithm is proposed to update the access strategy of the NGSO system every slot, including channel selection and power control. In addition, a more practical scenario in which the GSO satellites transmit with multiple power levels is considered in the proposed DSA scheme. We simulate the proposed algorithm in several sensing cases in which different capabilities and accuracies of NGSO spectrum sensing are considered, and the simulation results prove the effectiveness and superiority of the proposed DRL-based DSA algorithm.