Spectrum monitoring on regenerative payloads using versal AI engines

Spectrum monitoring from space can be beneficial for assessing the nature of electromagnetic emissions from ground stations located across the beam footprint of the satellite. In particular, for regenerative payloads, this can enable dynamic spectrum access across designated transmission bands to maximize spectral usage and bandwidth efficiency within permissible regulatory limits. However, to perform accurate detection of various signals from space at lower signal-to-noise thresholds, it is beneficial to use neural network-based techniques with well-crafted feature engineering. In the work presented in this paper, both CNN-based and YOLO-based detection techniques are implemented and analyzed on a Versal FPGA with AI Engines for a future 6G space mission. The performance of the original design and quantized versions are compared for both model types.