Model-Based Online Learning for Active ISAC Waveform Optimization

This paper proposes a Model-Based Online Learning (MBOL) framework for waveform optimization in integrated sensing and communications (ISAC) systems. In particular, the MBOL framework is proposed to enhance the ISAC performance under dynamic environmental conditions. Unlike Model-Free Online Learning (MFOL) methods, our approach leverages a rich structural knowledge of sensing, communications, and radio environments, offering better explainability and sample efficiency. This paper establishes a theoretical analysis of the proposed class of MBOL methods, showing essential performance conditions and convergence rates. This theoretical analysis is critical for understanding the potential of MBOL in active waveform optimization tasks. We demonstrate the proposed MBOL framework in multicarrier ISAC systems, focusing on the sub-carrier selection and power allocation problem. Via numerical experiments, we show that the proposed MBOL method outperforms the MFOL method in terms of sample efficiency. The results underline the potential of MBOL for improving the active waveform optimization performance in ISAC systems, particularly when sample efficiency and explainability are critical.