Robust Beamforming Based on Graph Attention Networks for IRS-Assisted Satellite IoT Communications

Area of Science:

• Wireless Communication
• Satellite Networks
• Artificial Intelligence

Background:

• Internet of Remote Things (IoRT) applications increasingly rely on satellite communication.
• Intelligent Reflecting Surfaces (IRS) offer potential for enhancing signal power in Low Earth Orbit (LEO) satellite networks.
• Challenges include acquiring channel state information (CSI) and optimal IRS phase shifts due to LEO mobility and passive IRS elements.

Purpose of the Study:

• To maximize the sum-rate of terrestrial users in an IRS-assisted LEO satellite communication network.
• To jointly optimize the satellite's precoding matrix and the IRS's phase shifts.
• To address the limitations of conventional algorithms in dynamic scenarios with complex CSI acquisition.

Main Methods:

• Proposed a robust beamforming design using Graph Attention Networks (RBF-GAT).
• Established a direct mapping from received pilots and network topology to satellite and IRS beamforming.
• Employed an unsupervised learning approach for offline training of the RBF-GAT model.

Main Results:

• The RBF-GAT approach demonstrated effective beamforming design for satellite and IRS.
• Achieved over 95% of the performance of the upper bound.
• Exhibited low computational complexity compared to conventional methods.

Conclusions:

• The proposed RBF-GAT is a computationally efficient and effective solution for IRS-assisted LEO satellite networks.
• This AI-driven approach overcomes challenges associated with CSI acquisition and dynamic network conditions.
• Enables enhanced performance for Internet of Remote Things (IoRT) applications via optimized beamforming.