Dynamic multi-beam optimization for space-based ADS-B reception via dual-layer deep reinforcement learning

The space-based Automatic Dependent Surveillance-Broadcast (ADS-B) system is vital for achieving global aviation surveillance, and one effective approach to mitigate co-channel interference within the system is the adoption of multi-beam reception. However, traditional static beam optimization methods struggle to maintain optimal reception performance under time-varying conditions due to the dynamic and non-uniform distribution of aircraft within satellite coverage areas. To address this challenge, this paper first establishes a dynamic optimization model for multi-beam reception, and presents a physics-consistent Markov Decision Process (MDP) formulation with an elaborate design of the state space and reward function based on space-based ADS-B characteristics. Then, a dual-layer predictive beamforming control (DPBC) method based on deep reinforcement learning (DRL) is proposed. It incorporates a State Prediction Network (SPN) to restore the Markov property by compensating the state spatiotemporal misalignment, and employs a capacity constrained clustering initialization strategy to embed geometric priors, thereby improve exploration efficiency in high-dimensional continuous beamforming control. Experiments using in-orbit data demonstrate that the DPBC method dynamically adjusts beam configurations to achieve superior reception performance compared to static methods across varying aircraft distributions, and can achieve full coverage with update intervals below 8 s compared to other DRL dynamic optimization approaches. Therefore, the proposed method is effective and adaptive for dynamic beam control in the space-based ADS-B system, meeting the required surveillance performance for air traffic control and having practical application potential in the growing aviation sector.