Article
Enhanced Security Authentication Based on
Convolutional-LSTM Networks
Xiaoying Qiu
1,
* , Xuan Sun
1
and Monson Hayes
2
Citation: Qiu, X.; Sun, X.; Hayes, M.
Enhanced Security Authentication
Based on Convolutional-LSTM
Networks. Sensors 2021, 21, 5379.
https://doi.org/10.3390/s21165379
Academic Editors: Tommaso
Pecorella and Benedetta Picano
Received: 1 July 2021
Accepted: 6 August 2021
Published: 9 August 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
College of Information Management, Beijing Information Science and Technology University,
Beijing 100192, China; sunxuan@bistu.edu.cn
2
Department of Electrical and Computer Engineering, George Mason University, Fairfax, VA 22030, USA;
hayes@gmu.edu
* Correspondence: 20192329@bistu.edu.cn
Abstract:
The performance of classical security authentication models can be severely affected by
imperfect channel estimation as well as time-varying communication links. The commonly used
approach of statistical decisions for the physical layer authenticator faces significant challenges in a
dynamically changing, non-stationary environment. To address this problem, this paper introduces
a deep learning-based authentication approach to learn and track the variations of channel charac-
teristics, and thus improving the adaptability and convergence of the physical layer authentication.
Specifically, an intelligent detection framework based on a Convolutional-Long Short-Term Memory
(Convolutional-LSTM) network is designed to deal with channel differences without knowing the
statistical properties of the channel. Both the robustness and the detection performance of the learn-
ing authentication scheme are analyzed, and extensive simulations and experiments show that the
detection accuracy in time-varying environments is significantly improved.
Keywords: physical layer security; wireless networks; classification algorithms; deep learning
1. Introduction
Innovations in communication technologies and artificial intelligence (AI) over the
past two decades have not only brought about tremendous new smart applications, but also
significantly increased serious security risks imposed on wireless devices, owing to the
openness of radio signal propagation [
1
–
3
]. An explosive growth in the number of Internet-
of-Thing (IoT) terminals provide abundant opportunities for adversaries to intercept trans-
missions and commit undetected spoofing attacks. In addition, the lack of standardization
in security protocols for IoT and intermittent communications is detrimental to the per-
formance of wireless communication systems. Moreover, the complex dynamic network
environments and the “on–off” transmissions of resource-constrained devices make it more
difficult to authenticate and identify illegal transmissions in wireless networks. There-
fore, a proper authentication control mechanism is essential for wireless communication
networks, especially considering the continuous integration between the wireless infras-
tructure and the development of smart industries supported by IoT.
While digital key-based cryptographic schemes are employed for network security,
they are based on an assumption that spoofing attackers lack the computational and storage
capabilities to successfully attack the network [
4
]. A fundamental drawback of key-based
cryptographic techniques is that a user will either pass a one-time static authentication or
fail by a binary security check [
5
]. Because of the rapid growth in the computing power
of smart communication nodes, it is becoming increasingly more likely that a potential
intruder will be able to crack privacy keys from received information [
4
]. Although
repeated security checks may be achieved with conventional key-based cryptographic
methods by repeatedly logging into the network, this is not a feasible solution since the time
delay and computational overhead put a strain on resource-constrained sensor nodes [
6
,
7
].
Sensors 2021, 21, 5379. https://doi.org/10.3390/s21165379 https://www.mdpi.com/journal/sensors
