BeamNet: Unsupervised Beamforming for ISAC Systems Under Imperfect CSI

Area of Science:

• Electrical Engineering
• Computer Science
• Signal Processing

Background:

• Integrated Sensing and Communication (ISAC) systems merge radar sensing and wireless communication for enhanced efficiency.
• Transmitting beamforming is crucial for optimizing performance in dual-function systems.
• Existing methods often require perfect channel state information (CSI) or complex optimization solvers.

Purpose of the Study:

• To propose BeamNet, an unsupervised deep learning framework for transmit beamforming in ISAC systems.
• To enable joint optimization of communication rate (CR) and sensing rate (SR) under general fading and imperfect CSI.
• To learn the CR-SR Pareto frontier without requiring beamforming labels or embedded solvers.

Main Methods:

• Developed BeamNet, an unsupervised deep learning framework mapping noisy channel estimates to beamforming vectors.
• Trained BeamNet end-to-end by maximizing a weighted sum of CR and SR.
• Evaluated performance in Rayleigh, Nakagami-m, and Rician fading channels with varying CSI quality.

Main Results:

• BeamNet accurately reproduced analytical Pareto-optimal solutions in perfect CSI scenarios.
• Characterized CR-SR trade-offs across different fading parameters and assessed robustness to distribution mismatch.
• Demonstrated superior performance under imperfect CSI compared to closed-form beamformers, recovering performance loss from estimation errors.

Conclusions:

• Unsupervised learning provides a flexible and robust approach for ISAC beamforming in fading environments.
• BeamNet effectively handles imperfect channel state information, offering a practical solution for future wireless networks.
• The framework learns the CR-SR trade-off efficiently, outperforming traditional methods in challenging conditions.