검색
  • RAMO

Spectrum Sharing System: Visualized Interference Classification and CNN Approach

최종 수정일: 3월 16일



Authors: Hyelin Nam, Kywon Han, Jihoon Park, Han Cha, Seong-Lyun Kim


In USA, the Citizens Broadband Radio Service (CBRS) is launched in 3.55-3.7GHz by Federal Communications Commission (FCC)[1]. The CBRS system consists of 3 hierarchical users: the Incumbent, Priority Access Licenses (PAL), Generally Authorized Access (GAA). According to the rule of the CBRS system, the PAL operators should protect an authorized incumbent user transmission such as federal radio-location system, aeronautical system, and the high-powered defense radar systems of the Department of Defense (DoD), while the GAA should also avoid using overlapped spectrum band with the PAL user [2]. To comply with this rule, the PAL operator should observe the activity of incumbent users.


Recently, many studies have been using deep learning techniques for spectrum sensing. Deep learning-based sensing methods with sufficient data outperform traditional methods. However, since the rules of the CBRS system do not allow collision cases, the data of collision cases rarely happens. The lack of collision data is the main bottleneck of deep learning-based DSA system, particularly in the training phase.


To tackle this, we generate a radar signal mimicking the Navy's active sensing signal through Universal Software Radio Peripheral (USRP) and LTE signal that follows the CBRS rule by making custom LTE base station. From this, we could artificially acquire the data of the collision cases in the CBRS system. We implement a demo system that the base station avoids the incumbent user and operates through the Spectrum Access System (SAS). We uploaded our dataset to Kaggle [3].


We demonstrate the deep learning-based spectrum sharing system that detects the channel usage of primary users and controls secondary users through CNN in real-time. Our testbed is implemented with the following structure.

  1. BS starts operating on Ch1.

  2. After a certain period of time, Radar operates in Ch1. At this time, the sensor detects the collision and sends the channel information to SAS.

  3. BS stops using at ch1 and asks SAS for available channels.

  4. SAS informs BS that Ch2 is available, and BS operates on Ch2.

  5. Radar operates in Ch1 and Ch2.

  6. The sensor detects a collision, and SAS commands BS to stop using Ch2.

A video of our scenario can be watched below [4].


Full Paper: H. Nam, K. Han, J. Park, H. Cha and S. -L. Kim, "A Spectrum Sharing System: Visualized Interference Classification and CNN Approach," 2021 IEEE International Symposium on Dynamic Spectrum Access Networks (DySPAN), 2021, pp. 33-40, doi: 10.1109/DySPAN53946.2021.9677198.


Acknowledgement

This work was supported by Institute of Information \& communications Technology Planning \& Evaluation (IITP) grant funded by the Korea government(MSIT) (No. 2018-11-1864, Scalable Spectrum Sharing for Beyond 5G Communication and No. 2021-0-00347, 6G Post-MAC)


Reference

[1] Federal Communications Commission, "Auction 105 Long-Form Application Granted," Public Notice (DA 21-824), July 2021, [Online]. Available: https://www.fcc.gov/auction/105

[2] E. Drocella, R. Sole, and N. LaSorte, “NTIA TR-20-546: Technical feasibility of sharing federal spectrum with future commercial operations in the 3450-3550 MHz band,” National Telecommunications and Information Administration (NTIA) Technical Report, Jan. 2020.

[3] [Online]. Available: https://www.kaggle.com/hyelinnam/vic-dataset-iq-signal-visualization-for-cbrs.

[4] [Online]. Available: https://youtu.be/KnHBo2FkKrc.


댓글 0개