Spectrum Occupancy Detection Supported by Federated Learning
Łukasz Kułacz
TL;DR
The paper tackles reliable spectrum occupancy detection for dynamic spectrum access under distributed sensing. It applies federated learning, specifically coefficient averaging via FedAvg, to train detectors across sensors while exchanging only model parameters. Using real USRP/GNURadio data with GMSK signals, results show that federated learning improves reliability and approaches centralized ML performance, particularly when some sensors are faulty. The work demonstrates a practical, privacy-preserving approach for distributed spectrum sensing and motivates further study of environmental variation and multi-hardware deployments.
Abstract
Dynamic spectrum access is essential for radiocommunication and its limited spectrum resources. The key element of dynamic spectrum access systems is effective spectrum occupancy detection. In many cases, machine learning algorithms improve detection effectiveness. Because of the recent trend of using federated learning, a federated learning algorithm is presented in the context of distributed spectrum occupancy detection. The results of the work presented in the paper are based on actual signal samples collected in the laboratory. The proposed algorithm is effective, especially in the context of a set of sensors with faulty sensors.