Thursday 27 March 2025
A recent study has shed light on a promising solution for maximizing the sum rate of terrestrial users in cognitive aerial-terrestrial networks (CATNs). The researchers proposed a safe deep reinforcement learning (DRL) based scheme, which effectively balances the trade-off between the sum rate and interference temperature constraints.
In CATNs, aerial users (AUs) such as airplanes and flying cars demand high-quality downlink communication. To alleviate interference from terrestrial base stations (BSs), user association and coordinated beamforming are essential. The problem is complex due to the dynamic nature of network conditions, including channel variations and handovers.
To address this challenge, the researchers formulated a non-cooperative partially observable Markov game (NCPOMG) with cost functions derived from the received interference power of AUs. They then developed a safe DRL algorithm for each BS agent to maximize reward while satisfying safety constraints.
The proposed scheme consists of two stages: user association and beamforming optimization. In the first stage, each terrestrial user (TU) selects its associated BS based on the received signal strength. The second stage involves optimizing the beamforming vectors of BSs to minimize interference and maximize the sum rate of TUs.
Simulation results demonstrate that the proposed scheme outperforms existing optimization-based methods in terms of computational complexity and communication overhead. The average received interference power of AUs is generally below the threshold, ensuring reliable communication for aerial users.
The study’s findings have significant implications for the development of CATNs, which are expected to play a crucial role in future wireless networks. By leveraging deep reinforcement learning, this scheme offers a promising solution for optimizing network performance while meeting stringent safety constraints.
One of the key advantages of the proposed scheme is its ability to adapt to dynamic network conditions. As channel conditions change or handovers occur, the BS agents can adjust their beamforming vectors and user associations in real-time to optimize network performance.
The researchers also explored the impact of different parameters on the system’s performance, including the maximum transmit power of BSs and interference temperature limits. Their results highlight the importance of carefully tuning these parameters to achieve optimal network performance.
Overall, this study represents a significant step forward in the development of CATNs, which will be essential for enabling high-quality wireless communication services in future aerial applications. By combining deep reinforcement learning with safe optimization techniques, the proposed scheme offers a powerful tool for balancing competing objectives and ensuring reliable communication in these complex networks.
Cite this article: “Optimizing Cognitive Aerial-Terrestrial Networks via Safe Deep Reinforcement Learning”, The Science Archive, 2025.
Cognitive Aerial-Terrestrial Networks, Deep Reinforcement Learning, Safe Optimization, Terrestrial Users, Aerial Users, Coordinated Beamforming, Interference Temperature Constraints, Non-Cooperative Partially Observable Markov Game, User Association, Beamforming Optimization.
