A Convolutional Attention Based Deep Learning Solution for 5G UAV Network Attack Recognition over Fading Channels and Interference Joseanne Viana †‡ , Hamed Farkhari ∗† , Luis Miguel Campos ∗ , Pedro Sebastião †‡ , Katerina Koutlia ¶ , Sandra Lagén ¶ , Luis Bernardo §‡ , Rui Dinis §‡ , † ISCTE – Instituto Universitário de Lisboa, Av. das Forças Armadas, 1649-026 Lisbon, Portugal ∗ PDMFC, Rua Fradesso da Silveira, n. 4, Piso 1B, 1300-609, Lisboa, Portugal ‡ IT – Instituto de Telecomunicações, Av. Rovisco Pais, 1, Torre Norte, Piso 10, 1049-001 Lisboa, Portugal § FCT – Universidade Nova de Lisboa, Monte da Caparica, 2829-516 Caparica, Portugal ¶ CTTC - Centre Tecnològic de Telecomunicacions de Catalunya (CERCA); Emails: joseanne_cristina_viana@iscte-iul.pt, Hamed_Farkhari@iscte-iul.pt, luis.campos@pdmfc.com, pedro.sebastiao@iscte-iul.pt, {kkoutlia, slagen}@cttc.es, rdinis@fct.unl.pt Abstract—When users exchange data with Unmanned Aerial Vehicles - (UAVs) over Air-to-Ground - (A2G) wireless commu- nication networks, they expose the link to attacks that could increase packet loss and might disrupt connectivity. For example, in emergency deliveries, losing control information (i.e., data related to the UAV control communication) might result in accidents that cause UAV destruction and damage to buildings or other elements. To prevent these problems, these issues must be addressed in 5G and 6G scenarios. This research offers a Deep Learning (DL) approach for detecting attacks on UAVs equipped with Orthogonal Frequency Division Multiplexing - (OFDM) receivers on Clustered Delay Line (CDL) channels in highly complex scenarios involving authenticated terrestrial users, as well as attackers in unknown locations. We use the two observable parameters available in 5G UAV connections: the Received Signal Strength Indicator (RSSI) and the Signal to Interference plus Noise Ratio (SINR). The developed algorithm is generalizable regarding attack identification, which does not occur during training. Further, it can identify all the attackers in the environment with 20 terrestrial users. A deeper investigation into the timing requirements for recognizing attacks shows that after training, the minimum time necessary after the attack begins is 100 ms, and the minimum attack power is 2 dBm, which is the same power that the authenticated UAV uses. The developed algorithm also detects moving attackers from a distance of 500 m. Index Terms—Cybersecurity, Convolutional Neural Networks, Deep Learning, Jamming Detection, Jamming Identification, Unmanned Aerial Vehicles, 5G; I. I NTRODUCTION Unmanned Aerial Vehicles - (UAVs) will integrate into 5G and 6G networks to provide delivery services, security, general and risky inspections, emergency services, and other functions inside and outside the network. The logistics industry will benefit first from using UAVs in their ecosystem, followed by all other vertical industries. In addition to coverage, high throughput, and low latency requirements, there is an increas- ing demand for secure and reliable connections with powerful data protection [1]. We expect that emergency and high-value transportation, whose success depends on the capacity to com- municate reliably and securely, will employ UAVs to provide high-quality services at lower costs [3]. Due to their aerial nature, UAVs provide faster and more flexible network services at higher data rates since they have complete control over their movement and a high probability of establishing robust Line- of-Sight (LoS) communication links. However, the vulnerabil- ity of wireless Air-to-Ground - (A2G) communication links make UAVs susceptible to attacks that increase packet loss or, even worse, completely lose communication. In order to keep UAV communications safe, it is crucial to detect potential risks and implement countermeasures. There is extensive research on Anti-Jamming techniques. Two established approaches to identify jamming are: analyzing the packet delivery ratio and the received signal strength. Both mechanisms deal with a high amount of lost information before detecting the attack. In ultra-dense networks, the overall amount of connected devices might hide the presence of local jammers. Finding other ways to address security issues in UAV networks is vital. Currently, researchers are adopting machine learning tech- niques for sequence prediction problems with spatial inputs and pattern recognition [2]. As a part of machine learning, Deep Learning (DL) research exploits algorithms to make models with high-level data abstractions by using multi- ple processing layers with complex structures. Deep Neural Networks (DNNs) such as Convolutional Neural Networks (CNNs) [3], [4] with Long Short-Term Memory (LSTM) or attention layers are used for temporal modeling, and to define universal functions in complex wireless scenarios. [5] [4]. These characteristics make them suitable for applications that deal with time series and spatial data, such as interference identification in wireless networks. The signal under analy- sis uses specific features to detect anomalies. The authors in [6] add an attention layer in their CNN to track long temporal variations in the time domain gradients. Some pre- trained networks do not require re-design because they use transfer learning methods to learn classification procedures. For example, the authors in [7] use pre-trained networks (i.e., AlexNet, VGG-16, ResNet-50) to identify jamming using spectral images of the received signal in the UAV. These networks can be vast and require extensive processing to sort information, making them unsuitable for use by UAVs. Even though embedded deep network techniques in the cloud or edge can monitor and evaluate channel degradation due to interference, fading, and jamming attacks, anti-jamming procedures and non-traditional approaches to avoid jamming are the focus of most research on this topic rather than recognizing attacks. As a result, there is a lack of publicly 978-1-6654-5468-1/22/$31.00 ©2022 IEEE 2022 IEEE 96th Vehicular Technology Conference (VTC2022-Fall) | 978-1-6654-5468-1/22/$31.00 ©2022 IEEE | DOI: 10.1109/VTC2022-Fall57202.2022.10012726 Authorized licensed use limited to: UNIVERSITAT POLITECNICA DE CATALUNYA. Downloaded on March 02,2023 at 09:41:07 UTC from IEEE Xplore. Restrictions apply.