Large‐scale Wireless Spectrum Monitoring
Sreeraj Rajendran, Sofie Pollin · 2020
Automated monitoring of wireless spectrum over frequency, time, and space is still a difficult research problem. This chapter discusses the main challenges of large-scale wireless spectrum monitoring. It concentrates on two very concrete spectrum analysis examples where the use of machine learning is promising. To emphasize the power of such a framework, the chapter concentrates on two major applications: anomaly detection and wireless signal classification. A crowdsourced spectrum monitoring framework, which leverages state-of-the-art software-defined radio and big data architectures, can help to address these challenges and democratize spectrum awareness. A semi-supervised deep learning setup based on the latest deep learning research which achieves performance close to fully supervised models with only 20% of the labeled samples is discussed. The chapter analyzes the performance of quantized models and shows that, in principle, considerable computational performance can be achieved at a cost of 10% classification accuracy loss.