Dynamic Spectrum Access by Learning Primary Network Topology

Jump to: navigation, search

Project summary


This project focuses on developing cooperative algorithms for the identification of spectrum holes and classification of primary network activity by geographically large-scale cognitive radio networks. The project consists of three parts, namely localization, footprint learning and network topology learning.

The knowledge of PU location is required in a cognitive radio (CR) network for advanced spectrum sharing, e.g., location-based interference management. Our proposed Cyclic Weighted Centroid Localization (WCL) algorithm uses the distinct cyclostationary properties of the PUs in order to estimate the location of each. The proposed Cyclic WCL reduces the localization error by 5x as compared to traditional WCL when the interferer's transmit power is 40 dB higher than the target.

When multiple CRs and PUs coexist in the same spectrum, different sets of CRs may receive signals from each PU. Learning these footprints of each PU is necessary for disambiguating the PU signals which, in turn, is necessary for computing individual PUs’ radio environment map, location, and for signal classification. To increase versatility of our proposed algorithms, we use only the received energy measurements at the CRs as input. Furthermore, they do not rely on knowledge of the channel propagation models or the location of any radios, PUs or CRs. Their ability to distinguish the signals from each PUs can be used to extend existing algorithms, such as WCL, designed for single PU systems.

The learned footprints and the detected spectrum occupancy are used to analyze the PU activity over time. The PU activity is then used to identify sets of PUs that are part of the same network. Learning the network topology improves the spatial resolution of the spectrum holes by identifying the receiver of each PU transmission. Our methodology includes identification of transmitters based on their unique radio fingerprints and modulation signal types. We are applying tools from deep neural networks to achieve robustness against radio imperfections and channel uncertainties while exploit features related to radio circuits and signal cyclostationarity.


  • NSF grant title: NeTS: Small: Dynamic Spectrum Access by Learning Primary Network Topology
  • NSF Awards number: 1527026 [Link]


Broader Impacts

  • D. Cabric, Symposium Chair, "Machine Learning for Communications", IEEE International Conference on Networking and Communications, (ICNC 2019)
  • D. Cabric, Keynote speaker, "Revisiting Spectrum Sharing Techniques for Unlicensed Massive IoT", CROWNCOM 2019, Ghent, Belgium, September 2018.
  • D. Cabric, Plenary talk, "Learning Wireless Networks Footprints and Topologies in Shared Spectrum", IEEE International Conference on Computing, Networking and Communications (ICNC), March 2018.
  • D. Cabric, Invited Panelist, "How relevant is Communication Theory to Spectrum Policy? Should communication theorists be more involved in policy debates or leave this for the lawyers?", IEEE Communication Theory Workshop, June 2017.
  • D. Cabric, Keynote talk "Localization in Cognitive Radio Networks",International Conference on Localization and GNSS, Barcelona, Spain, June 28, 2016.


Data management

Research results and data from our works are available upon request.