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

Area of Science:

• Wireless communication
• Signal processing
• Machine learning

Background:

• Federal Communications Commission rules allow commercial wireless networks to share the 3.5 GHz Citizens Broadband Radio Service (CBRS) band with federal incumbents.
• Commercial systems must vacate the band upon detection of U.S. military radars, such as the SPN-43 air traffic control radar.
• Accurate detection algorithms are critical for the reliable operation of these sensing systems.

Purpose of the Study:

• To evaluate thirteen different methods for detecting the SPN-43 radar within the 3.5 GHz band.
• To compare the performance of classical signal detection techniques against machine learning and deep learning approaches.
• To identify the most effective algorithms for radar detection in the CBRS band.

Main Methods:

• Utilized a dataset of over 14,000 spectrograms from the 3.5 GHz band, collected during a recent measurement campaign.
• Compared thirteen detection algorithms, including classical signal detection theory methods, machine learning algorithms, and deep learning architectures.
• Focused on evaluating accuracy and computational complexity trade-offs.

Main Results:

• Machine learning algorithms demonstrated significantly superior performance compared to classical signal detection methods.
• A three-layer convolutional neural network (CNN) emerged as the top performer, offering an optimal balance between accuracy and computational cost.
• The developed CNN was applied to analyze the entire 3.5 GHz spectrogram library, yielding descriptive statistics.

Conclusions:

• Modern machine learning algorithms, particularly deep learning models like CNNs, show substantial promise for radar detection in the 3.5 GHz CBRS band.
• Classical detection methods may have limitations in this dynamic spectrum sharing environment.
• The findings support the adoption of advanced algorithms for robust radar sensing in shared spectrum bands.