Deep Learning Classification of 3.5 GHz Band Spectrograms with Applications to Spectrum Sensing

W. Max Lees,Adam Wunderlich,Peter Jeavons,Paul D. Hale,Michael R. Souryal

DOI: https://doi.org/10.1109/TCCN.2019.2899871

2018-12-05

Abstract:In the United States, the Federal Communications Commission has adopted rules permitting commercial wireless networks to share spectrum with federal incumbents in the 3.5~GHz Citizens Broadband Radio Service band. These rules require commercial systems to vacate the band when sensors detect radars operated by the U.S. military; a key example being the SPN-43 air traffic control radar. Such sensors require highly-accurate detection algorithms to meet their operating requirements. In this paper, using a library of over 14,000 3.5~GHz band spectrograms collected by a recent measurement campaign, we investigate the performance of thirteen methods for SPN-43 radar detection. Namely, we compare classical methods from signal detection theory and machine learning to several deep learning architectures. We demonstrate that machine learning algorithms appreciably outperform classical signal detection methods. Specifically, we find that a three-layer convolutional neural network offers a superior tradeoff between accuracy and computational complexity. Last, we apply this three-layer network to generate descriptive statistics for the full 3.5~GHz spectrogram library. Our findings highlight potential weaknesses of classical methods and strengths of modern machine learning algorithms for radar detection in the 3.5~GHz band.

Signal Processing,Machine Learning

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is to improve the accuracy of detecting SPN - 43 radar signals within the 3.5 GHz frequency band. Specifically, the paper explores the use of deep - learning methods to classify spectrograms in the 3.5 GHz frequency band for more efficient spectrum sensing. These methods are mainly used to support spectrum sharing between commercial wireless networks and US federal agencies in this frequency band. In particular, when military radar operations are detected, commercial systems are required to evacuate this frequency band in a timely manner. The paper evaluates the performance of thirteen different detection algorithms, including classical signal detection theory methods, machine - learning methods, and multiple deep - learning architectures, on the SPN - 43 radar detection task. The study found that deep - learning methods significantly outperform traditional methods in detection performance. In particular, the three - layer convolutional neural network (CNN) provides the best balance between accuracy and computational complexity. In addition, the paper also uses the optimal deep - learning model to conduct a descriptive statistical analysis of the spectrogram library of the entire 3.5 GHz frequency band, revealing the potential weaknesses of traditional methods and the advantages of modern machine - learning algorithms.