Application of deep learning algorithms to 3.5 GHz spectrograms to characterize incumbent federal radar emissions.
This project compared various detection algorithms for federal incumbent radar signals in the 3550-3650 MHz band using a set of over 14,000 spectrograms collected by the completed Waveform Measurements of Radars Operating in the 3.5 GHz Band project. The results demonstrated that training deep learning algorithms with real-world measurements can outperform traditional energy detection methods. Moreover, they demonstrated the utility of machine learning for estimating spectrum occupancy and ambient power distributions.
For more information on NIST research concerning the 3.5 GHz Citizens Broadband Radio Service (CBRS) Band, please see the NIST CBRS project webpage.
W. Lees, A. Wunderlich, P. Jeavons, P. Hale, M. Souryal, “Deep learning classification of 3.5 GHz band spectrograms with applications to spectrum sensing,” IEEE Transactions on Cognitive Communications and Networking, 5(2), pp. 224-236, June 2019.
W. Lees, A. Wunderlich, P. Jeavons, P. Hale, M. Souryal, “Spectrum Occupancy and Ambient Power Distributions for the 3.5 GHz Band Estimated from Observations at Point Loma and Fort Story,” NIST Technical Note 2016, Sept 2018.