Unsupervised Learning-Based ISAC Waveforms

Waveform designs for integrated sensing and communication (ISAC) via classical optimization theory may become computationally exhaustive due to trade-offs between sensing and communication metrics. To circumvent this, an unsupervised learning-based ISAC waveform design is proposed. The ISAC trade-offs are captured via a custom loss function subject to constraints. The trade-offs between the user rates, probability of detection, and Cramér-Rao bound are used to evaluate and compare performance gains. Our results reveal the potential of learning-based techniques in balancing the performance and complexity of ISAC waveform designs.