Project 1: RF-Based Fingerprint Classification for UAV Cybersecurity

UAVs rely on wireless communications for control and telemetry to function properly, hence, establishing and maintaining secure communications is essential to achieving their mission. As the threat of malicious attacks on UAVs becomes more relevant with attackers having access to advanced communication hardware and algorithms, the physical layer becomes a valuable resource for cybersecurity [1] [2] [3]. The goal of this REU project is to study RF-based UAV cybersecurity techniques such as fingerprinting and to develop an experimental framework for the measurement of RF fingerprints in the laboratory. The study focuses on features extracted from the RF signal spectral signature, and form measured raw I/Q data, to identify unique fingerprints from commercially available UAV communication wireless systems. Students will create the datasets in the laboratory using a signal analyzer (up to 26.5 GHz), study the raw I/Q and frequency spectrum data for feature extraction, and then explore different machine learning (ML) algorithms such as decision trees, random forest, and Naïve Bayes, to provide the students an understanding of the nature of the RF signals and physical layer, but also the different factors involved in achieving a robust ML algorithm.

Project 2: Adversarial Classifications on UAV Machine Learning through Performance Metrics

Adversarial attacks, or an attack committed by an adversary to hijack a system, are finding their way into the Machine Learning/Deep Learning domain. Particularly, adversarial attacks on Deep Learning come in the form of perturbed images. Typically, these are cyber-based attacks aimed at forcing the model to misclassify and compromise the affected system [4]. Attacks on the physical world are not as easily generated and come in the form of natural perturbations (e.g., ImageNet-B, a natural perturbation dataset) [5]. Additionally, training data for natural perturbations is not common and does not cover handmade datasets. Tensorflow, an open-source Machine Learning platform, includes a Profiler to observe system and hardware resources to resolve model performance bottlenecks [6]. The goal of this project is to discover if adversarial attacks affect the inference performance capabilities compared to a normal input inference. This can increase the model security because an adversarial image can be detected without human-in-the-loop intervention and solely relying on the model prediction. By increasing model security, the UAV’s overall security is protected from adversarial system attacks. This project will utilize a UAV capable of implementing onboard Machine Learning, i.e., MobileNet or any Tensorflow-based Deep Learning image classifier. The baseline experiment will utilize Tensorflow’s Profiler; however, to accurately identify adversarial classifications the TensorFlow profiler will be extended to include more metrics seen on other performance benchmarks like MLPerf which extends to inference execution time or latency bounded throughput [7]. The project includes the design of visual classes that the drone can operate around to the server as input to the Deep Learning onboard the UAV. Using the Profiler or a derivative, performance metrics of the Deep Learning model will be calculated for inferences on the visual classes. Furthermore, the project will also include experimentation to test adversarial images. The project necessitates participants to research techniques to implement natural perturbations/adversarial data on the initial object class, i.e., a screen mesh on the object.

Project 3: Surveillance Mission Modelling with Swarms of Micro Aerial Vehicles

Reconnaissance mission success within hostile territory requires fast and reliable data collection and carries significant risk. In recent years, the UAV design and development have evolved towards miniaturizing the aerial systems and cooperation among a swarm of micro aerial vehicles (MAV) [8]. This research focuses on the performance analysis of MAV swarm deployments in hostile territory. The reconnaissance mission success is given by the percentage of data collection within timing, distributed data storage, collision avoidance, and swarm MAV mission integrity constraints. The MAV in the swarm are modeled as individual agents with a set of defined behaviors as follows: (1) Swarm goal behavior, to collect data on the entire territory, (2) MAV goal behavior, to collect data on the assigned territory section, (3) Avoid behavior, to handle potential collisions, (4) Collaboration behavior, to distribute surveillance workload among the swarm, (5) Protection behavior, to make multiple copies of collected data, and (6) Coordination behavior, to maintain swarm integrity within an assured level of performance. This research is expected to provide solutions for the deployment of MAV swarms in hostile environments and identify the balance between goal achievement and the size of MAV swarms under different operational scenarios.

Project 4: Decentralized Quality Assessment for Crowdsensed UAS Data

Unmanned aerial systems (UASs) have unprecedented potential in various fields, from surveillance to packet delivery [9]. The swarms formed by multiple UAS increase the number of potential UAS applications as they can be utilized for complicated tasks [10] [11] [12]. In this project, we consider a crowdsensing application for weather monitoring using a UAS fleet and we focus on data quality assessment as it is critical for weather related applications [13]. Crowdsourcing systems use data quality assessment as they utilize intelligence to solve complex tasks [14]. However, centralized data quality control methods have problems such as reliability and single point of failure due to their centralized nature. In addition to trust related issues, creating an automated and scalable data quality measurement system is a vexing challenge. To solve these issues, a decentralized blockchain-based framework will be explored using data mining techniques. The overall goal is to tackle challenges such as unfair payment, unreliable data, negative participant work, and cooperative cheating in UAS-based crowdsensing applications.

Project 5: Secure UAV Authentication Using Channel State Information

Low-cost UAVs and their swarms are increasingly being explored as tools to achieve critical tasks such as disaster relief, search, and rescue operations, etc. For completing these mission projects, mutual authentication of UAV(s) is crucial to prevent spoofing and impersonating attacks preserving communication and data integrity. Traditional crypto-based authentication solutions usually involve high computational overhead and require the challenging task of managing secret keys. In this project, we focus on implementing a lightweight (non-cryptographic) authentication scheme that requires no priori key exchange nor full channel knowledge. Specifically, unique noise variations in communication modules (radios) [15] within UAVs are extracted for authentication by analyzing fine-grained channel state information (CSI) data. Each UAV computes the hardware signatures between themselves and other UAVs using just a few CSI data samples, signal processing algorithms, and neural network model. We will evaluate the performance and effectiveness of the proposed approach via prototyping implementation on commercial UAVs and conducting a wide range of controlled experiments in the Anechoic chamber and real-world outdoor settings. We will further investigate the stability of signatures/fingerprints with environmental changes and test the tamper-resistant property of signatures against remote adversary attacks.

[1] B. Chatterjee, D. Das, S. Maity and S. Sen, "RF-PUF: Enhancing IoT Security Through Authentication of Wireless Nodes Using In-Situ Machine Learning," IEEE Internet of Things Journal, vol. 6, no. 1, pp. 388-398, 2019.

[2] M. Rostami, F. Koushanfar and R. Karri, "A Primer on Hardware Security: Models, Methods, and Metrics," Proceedings of the IEEE, vol. 102, no. 8, pp. 1283-1295, August 2014.

[3] K. Subramani, G. Volanis, M. Bidmeshki, A. Antonopoulos and Y. Makris, "Trusted and Secure Design of Analog/RF ICs: Recent Developments," in IEEE International Symposium on On-Line Testing and Robust System Design (IOLTS), Rhodes, Greece, 2019.

[4] X. Wang, J. Li, X. Kuang, Y.-a. Tan and J. Li, "The security of machine learning in an adversarial setting: A survey," Journal of Parallel andd Distrubuted Computing, pp. 12-23, 2019.

[5] V. Shankar, A. Dave, R. Roelofs, D. Ramanan, B. Recht and L. Schmidt, "A Systematic Framework for Natural Pertubations from Videos," in ICML Deep Phenomena, 201.

[6] Tensorflow, A profiling and performance analysis tool for Tensorflow, https://github.com/tensorflow/profiler, 2021.

[7] V. J. Reddi, C. Cheng, D. Kanter, P. Mattson, G. Schmuelling, C.-J. Wu, B. Anderson, M. Breughe, M. Charlebois, W. Chou, R. Chukka, C. Coleman, S. Davis, P. Deng, G. Diamos, J. Duke, D. Fick, J. S. Gardner, I. Hubara, S. Idgunji, T. B. Jablin, J. Jiao, T. St. John, P. Kanwar, D. Lee, J. Liao, A. Lokhmotov, F. Massa, P. Meng, P. Micikevivius, C. Osborne, G. Pekhimenko, A. T. R. Rajan, D. Sequeira, A. Sirasao, F. Sun, H. Tan, M. Thomson, F. Wei, E. Wu, L. Xu, K. Yamada, B. Yu, G. Yuan, A. Zhong, P. Zhang and Y. Zhou, "MLPerf Inference Benchmark," in arxiv.org, 2020.

[8] R. B. R. S. R. Koeneke, "Target area surveillance optimization with swarms of autonomous micro aerial vehicles," in IEEE Internationa Systems Conference, Orlando, Florida, 2019.

[9] E. Y. a. R. M. S. Hayat, "Survey on Unmanned Aerial Vehicle Networks for Civil Applications: A Communication Viewpoint," in IEEE Communications Surveys & Tutorials, vol. 18, no. 4, pp. 2624–2661, 2016.

[10] T. N. a. M. I. Akbas, "SOSUAS: Stability Optimized Swarming for Unmanned Aerial Systems.," in Accepted to the AIAA/IEEE Digital Avionics Systems Conference (DASC), September, 2022.

[11] S. M. K. N. M. C. M. I. A. P. M. R. S. S. I. Muna, "Air Corridors: Concept, Design, Simulation, and Rules of Engagement," in In MDPI Sensors, 21(22):7536, November, 2021.

[12] J. Rentrope and M. I. Akbas, "Spatially Adaptive Positioning for Molecular Geometry Inspired Aerial Networks," in Proceedings of the ACM International Symposium on Design and Analysis of Intelligent Vehicular Networks and Applications (DIVANet), 2017.

[13] M. I. A. a. M. C. K. Adkins, "Real-Time Urban Weather Observations for Urban Air Mobility," International Journal of Aviation, Aeronautics, and Aerospace (IJAAA), 7(4), , November, 2020.

[14] A. C. F. B. K. L. F. D. K. a. P. B. Capponi, "A survey on mobile crowdsensing systems: Challenges, solutions, and opportunities," IEEE communications surveys & tutorials 21, no. 3 (2019): 2419-2465..

[15] Z. Z. a. S. Y. L. N. Kandel, "Exploiting CSI-MIMO for Accurate and Efficient Device Identification," in IEEE Global Communications Conference (GLOBECOM), 2019.