Guo, Yijia, Zhang, Junqing 
ORCID: 0000-0002-3502-2926 and Hong, Y-W Peter
(2023)
Deep Learning-Enhanced Physical Layer Authentication for Mobile Devices.
In: GLOBECOM 2023 - 2023 IEEE Global Communications Conference, 2023-12-4 - 2023-12-8, Kuala Lumpur, Malaysia.
|
Text
GC2023_CSI_Authentication_Accepted.pdf - Author Accepted Manuscript Available under License Creative Commons Attribution. Download (1MB) | Preview |
Abstract
The Internet of Things (IoT) is ubiquitous thanks to the rapid development of wireless technology. However, the broadcast nature of wireless transmission results in great challenges to the security authentication for large-scale IoT. In this paper, we propose a novel physical layer authentication approach for mobile scenarios employing deep learning and channel state information (CSI). Specifically, the convolution neural network (CNN) is designed to learn the temporal and spatial similarity between CSIs and output a score to measure the difference between the input CSIs. Device authentication is achieved by comparing the score to an empirically obtained threshold. We build a WiFi-based testbed and carry out a comprehensive experimental evaluation. The performance of using the CSI magnitude and real & imaginary parts is compared. The effect of the distance between legitimate and rogue devices on authentication performance is studied. The generalization performance of the CNN model in different test scenarios is also evaluated. Experiment results demonstrate the effectiveness of the proposed CNN-based authentication over conventional correlation-based authentication schemes.
| Item Type: | Conference or Workshop Item (Unspecified) |
|---|---|
| Divisions: | Faculty of Science and Engineering > School of Electrical Engineering, Electronics and Computer Science |
| Depositing User: | Symplectic Admin |
| Date Deposited: | 29 Aug 2023 08:32 |
| Last Modified: | 27 Mar 2024 13:32 |
| DOI: | 10.1109/globecom54140.2023.10437299 |
| Related URLs: | |
| URI: | https://livrepository.liverpool.ac.uk/id/eprint/3172384 |
Dimensions
Dimensions
