Loss Function Design for Enhanced Joint Beamforming in RIS-Aided ISAC Systems

Reconfigurable Intelligent Surface (RIS)-aided Integrated Sensing and Communication (ISAC) systems are important for next generation wireless networks (e.g., 6G), because it combines RISs capability to intelligently map wireless propagation environments and ISACS capability to realize radar sensing and data transmission at the same frequency band for higher spectrum use and lower hardware cost. The joint beamforming design for these systems is challenging: traditional optimization-based approaches are very time-consuming and cannot be used in real-time applications, while current learning-based methods typically involve large amounts of labeled data, which are expensive and hard to obtain in practice. We present a better unsupervised learning framework, whose primary focus is on designing a new loss function with 3 tunable coefficients {{}}}, , . In particular, coefficient {{}} scales the overall loss gain to keep the optimization running, focuses on the sensing gain term to guide the model to prioritize the echo power for sensing performance, and weights the communication gain term in order to explicitly guarantee user-side link quality. In this way, the proposed design can adaptively balance the sensing signal-to-noise ratio (SNRs) and communication SNRc during training, rather than sacrificing one for the other. Through extensive simulations, we show that compared with the original unsupervised learning scheme, the modified loss function not only achieves a better trade-off between sensing and communication, but also significantly improves SNRs and ensures SNRc meets practical thresholds, while maintaining low computational overhead.