Dataset Creation Strategies for Future AI-based Radio Spectrum Sensing Applications

The rapid growth of connected IoT devices, driven by advances in low-power wide area networks, presents significant challenges related to spectrum congestion and quality of service. Many IoT communication protocols risk increasing packet error rates due to collisions as large-scale deployments connect billions of devices competing for the same radio spectrum resources. Radio spectrum sensing technologies offer a viable solution by enhancing radio channel utilization through advanced interference mitigation and channel analytics planning strategies. This paper addresses and summarizes the different dataset formats that can be used in spectrum sensing applications and highlights two strategies for designing synthetic datasets that incorporate simulated channel impairments and real-world datasets captured using software-defined radios. The proposed configurations are ranked based on performance level, signal acquisition methods, dataset type used and integrated AI architecture. The main contribution of this work is to provide guidelines for dataset generation methodologies that can provide a roadmap for future AI-based spectrum sensing applications and use cases.