Gabriele Oligeri

Saeif Al-Hazbi, Ahmed Hussain, Savio Sciancalepore et al.

Radio Frequency Fingerprinting (RFF) techniques promise to authenticate wireless devices at the physical layer based on inherent hardware imperfections introduced during manufacturing. Such RF transmitter imperfections are reflected into over-the-air signals, allowing receivers to accurately identify the RF transmitting source. Recent advances in Machine Learning, particularly in Deep Learning (DL), have improved the ability of RFF systems to extract and learn complex features that make up the device-specific fingerprint. However, integrating DL techniques with RFF and operating the system in real-world scenarios presents numerous challenges, originating from the embedded systems and the DL research domains. This paper systematically identifies and analyzes the essential considerations and challenges encountered in the creation of DL-based RFF systems across their typical development life-cycle, which include (i) data collection and preprocessing, (ii) training, and finally, (iii) deployment. Our investigation provides a comprehensive overview of the current open problems that prevent real deployment of DL-based RFF systems while also discussing promising research opportunities to enhance the overall accuracy, robustness, and privacy of these systems.

Gabriele Oligeri, Savio Sciancalepore, Simone Raponi et al.

Wireless devices resorting to event-triggered communications have been proved to suffer critical privacy issues, due to the intrinsic leakage associated with radio-frequency (RF) emissions. In this paper, we move the attack frontier forward by proposing BrokenStrokes: an inexpensive, easy to implement, efficient, and effective attack able to detect the typing of a pre-defined keyword by only eavesdropping the communication channel used by the wireless keyboard. BrokenStrokes proves itself to be a particularly dreadful attack: it achieves its goal when the eavesdropping antenna is up to 15 meters from the target keyboard, regardless of the encryption scheme, the communication protocol, the presence of radio noise, and the presence of physical obstacles. While we detail the attack in three current scenarios and discuss its striking performance--its success probability exceeds 90% in normal operating conditions--, we also provide some suggestions on how to mitigate it. The data utilized in this paper have been released as open-source to allow practitioners, industries, and academia to verify our claims and use them as a basis for further developments.

Savio Sciancalepore, Omar Adel Ibrahim, Gabriele Oligeri et al.

We propose PiNcH, a methodology to detect the presence of a drone, its current status, and its movements by leveraging just the communication traffic exchanged between the drone and its Remote Controller (RC). PiNcH is built applying standard classification algorithms to the eavesdropped traffic, analyzing features such as packets inter-arrival time and size. PiNcH is fully passive and it requires just cheap and general-purpose hardware. To evaluate the effectiveness of our solution, we collected real communication traces originated by a drone running the widespread ArduCopter open-source firmware, currently mounted on-board of a wide range (30+) of commercial amateur drones. We tested our solution against different publicly available wireless traces. The results prove that PiNcH can efficiently and effectively: (i) identify the presence of the drone in several heterogeneous scenarios; (ii) identify the current state of a powered-on drone, i.e., flying or lying on the ground; (iii) discriminate the movements of the drone; and, finally, (iv) enjoy a reduced upper bound on the time required to identify a drone with the requested level of assurance. The effectiveness of PiNcH has been also evaluated in the presence of both heavy packet loss and evasion attacks. In this latter case, the adversary modifies on purpose the profile of the traffic of the drone-RC link to avoid the detection. In both the cited cases, PiNcH continues enjoying a remarkable performance. Further, the comparison against state of the art solution confirms the superior performance of PiNcH in several scenarios. Note that all the drone-controller generated data traces have been released as open-source, to allow replicability and foster follow-up. Finally, the quality and viability of our solution, do prove that network traffic analysis can be successfully adopted for drone identification and status discrimination.

Saeif Alhazbi, Ahmed Mohamed Hussain, Gabriele Oligeri et al.

As Large Language Models (LLMs) become increasingly integrated into many technological ecosystems across various domains and industries, identifying which model is deployed or being interacted with is critical for the security and trustworthiness of the systems. Current verification methods typically rely on analyzing the generated output to determine the source model. However, these techniques are susceptible to adversarial attacks, operate in a post-hoc manner, and may require access to the model weights to inject a verifiable fingerprint. In this paper, we propose a novel passive and non-invasive fingerprinting technique that operates in real-time and remains effective even under encrypted network traffic conditions. Our method leverages the intrinsic autoregressive generation nature of language models, which generate text one token at a time based on all previously generated tokens, creating a unique temporal pattern like a rhythm or heartbeat that persists even when the output is streamed over a network. We find that measuring the Inter-Token Times (ITTs)-time intervals between consecutive tokens-can identify different language models with high accuracy. We develop a Deep Learning (DL) pipeline to capture these timing patterns using network traffic analysis and evaluate it on 16 Small Language Models (SLMs) and 10 proprietary LLMs across different deployment scenarios, including local host machine (GPU/CPU), Local Area Network (LAN), Remote Network, and Virtual Private Network (VPN). The experimental results confirm that our proposed technique is effective and maintains high accuracy even when tested in different network conditions. This work opens a new avenue for model identification in real-world scenarios and contributes to more secure and trustworthy language model deployment.

Muhammad Usman, Simone Raponi, Marwa Qaraqe et al.

The massive deployment of IoT devices being utilized by home automation, industrial and military scenarios demands for high security and privacy standards to be achieved through innovative solutions. This paper proposes KaFHCa, a crypto-less protocol that generates shared secret keys by combining random frequency hopping collisions and source indistinguishability independently of the radio channel status. While other solutions tie the secret bit rate generation to the current radio channel conditions, thus becoming unpractical in static environments, KaFHCa guarantees almost the same secret bitrate independently of the channel conditions. KaFHCa generates shared secrets through random collisions of the transmitter and the receiver in the radio spectrum, and leverages on the fading phenomena to achieve source indistinguishability, thus preventing unauthorized eavesdroppers from inferring the key. The proposed solution is (almost) independent of the adversary position, works under the conservative assumption of channel fading (σ = 8dB), and is capable of generating a secret key of 128 bits with less than 564 transmissions.

Gabriele Oligeri, Simone Raponi, Savio Sciancalepore et al.

Physical-layer security is regaining traction in the research community, due to the performance boost introduced by deep learning classification algorithms. This is particularly true for sender authentication in wireless communications via radio fingerprinting. However, previous research efforts mainly focused on terrestrial wireless devices while, to the best of our knowledge, none of the previous work took into consideration satellite transmitters. The satellite scenario is generally challenging because, among others, satellite radio transducers feature non-standard electronics (usually aged and specifically designed for harsh conditions). Moreover, the fingerprinting task is specifically difficult for Low-Earth Orbit (LEO) satellites (like the ones we focus in this paper) since they orbit at about 800Km from the Earth, at a speed of around 25,000Km/h, thus making the receiver experiencing a down-link with unique attenuation and fading characteristics. In this paper, we propose PAST-AI, a methodology tailored to authenticate LEO satellites through fingerprinting of their IQ samples, using advanced AI solutions. Our methodology is tested on real data -- more than 100M I/Q samples -- collected from an extensive measurements campaign on the IRIDIUM LEO satellites constellation, lasting 589 hours. Results are striking: we prove that Convolutional Neural Networks (CNN) and autoencoders (if properly calibrated) can be successfully adopted to authenticate the satellite transducers, with an accuracy spanning between 0.8 and 1, depending on prior assumptions. The proposed methodology, the achieved results, and the provided insights, other than being interesting on their own, when associated to the dataset that we made publicly available, will also pave the way for future research in the area.

Ahmed Mohamed Hussain, Gabriele Oligeri, Thiemo Voigt

We present a new machine learning-based attack that exploits network patterns to detect the presence of smart IoT devices and running services in the WiFi radio spectrum. We perform an extensive measurement campaign of data collection, and we build up a model describing the traffic patterns characterizing three popular IoT smart home devices, i.e., Google Nest Mini, Amazon Echo, and Amazon Echo Dot. We prove that it is possible to detect and identify with overwhelming probability their presence and the services running by the aforementioned devices in a crowded WiFi scenario. This work proves that standard encryption techniques alone are not sufficient to protect the privacy of the end-user, since the network traffic itself exposes the presence of both the device and the associated service. While more work is required to prevent non-trusted third parties to detect and identify the user's devices, we introduce Eclipse, a technique to mitigate these types of attacks, which reshapes the traffic making the identification of the devices and the associated services similar to the random classification baseline.

Gabriele Oligeri, Savio Sciancalepore, Roberto Di Pietro

In this paper, we study the privately-own IRIDIUM satellite constellation, to provide a location service that is independent of the GNSS. In particular, we apply our findings to propose a new GNSS spoofing detection solution, exploiting unencrypted IRIDIUM Ring Alert (IRA) messages that are broadcast by IRIDIUM satellites. We firstly reverse-engineer many parameters of the IRIDIUM satellite constellation, such as the satellites speed, packet interarrival times, maximum satellite coverage, satellite pass duration, and the satellite beam constellation, to name a few. Later, we adopt the aforementioned statistics to create a detailed model of the satellite network. Subsequently, we propose a solution to detect unintended deviations of a target user from his path, due to GNSS spoofing attacks. We show that our solution can be used efficiently and effectively to verify the position estimated from standard GNSS satellite constellation, and we provide constraints and parameters to fit several application scenarios. All the results reported in this paper, while showing the quality and viability of our proposal, are supported by real data. In particular, we have collected and analyzed hundreds of thousands of IRA messages, thanks to a measurement campaign lasting several days. All the collected data ($1000+$ hours) have been made available to the research community. Our solution is particularly suitable for unattended scenarios such as deserts, rural areas, or open seas, where standard spoofing detection techniques resorting to crowd-sourcing cannot be used due to deployment limitations. Moreover, contrary to competing solutions, our approach does not resort to physical-layer information, dedicated hardware, or multiple receiving stations, while exploiting only a single receiving antenna and publicly-available IRIDIUM transmissions. Finally, novel research directions are also highlighted.

Simone Raponi, Isra Ali, Gabriele Oligeri

Classifying a weapon based on its muzzle blast is a challenging task that has significant applications in various security and military fields. Most of the existing works rely on ad-hoc deployment of spatially diverse microphone sensors to capture multiple replicas of the same gunshot, which enables accurate detection and identification of the acoustic source. However, carefully controlled setups are difficult to obtain in scenarios such as crime scene forensics, making the aforementioned techniques inapplicable and impractical. We introduce a novel technique that requires zero knowledge about the recording setup and is completely agnostic to the relative positions of both the microphone and shooter. Our solution can identify the category, caliber, and model of the gun, reaching over 90% accuracy on a dataset composed of 3655 samples that are extracted from YouTube videos. Our results demonstrate the effectiveness and efficiency of applying Convolutional Neural Network (CNN) in gunshot classification eliminating the need for an ad-hoc setup while significantly improving the classification performance.

Omar Adel Ibrahim, Savio Sciancalepore, Gabriele Oligeri et al.

Universal Serial Bus (USB) Flash Drives are nowadays one of the most convenient and diffused means to transfer files, especially when no Internet connection is available. However, USB flash drives are also one of the most common attack vectors used to gain unauthorized access to host devices. For instance, it is possible to replace a USB drive so that when the USB key is connected, it would install passwords stealing tools, root-kit software, and other disrupting malware. In such a way, an attacker can steal sensitive information via the USB-connected devices, as well as inject any kind of malicious software into the host. To thwart the above-cited raising threats, we propose MAGNETO, an efficient, non-interactive, and privacy-preserving framework to verify the authenticity of a USB flash drive, rooted in the analysis of its unintentional magnetic emissions. We show that the magnetic emissions radiated during boot operations on a specific host are unique for each device, and sufficient to uniquely fingerprint both the brand and the model of the USB flash drive, or the specific USB device, depending on the used equipment. Our investigation on 59 different USB flash drives---belonging to 17 brands, including the top brands purchased on Amazon in mid-2019---, reveals a minimum classification accuracy of 98.2% in the identification of both brand and model, accompanied by a negligible time and computational overhead. MAGNETO can also identify the specific USB Flash drive, with a minimum classification accuracy of 91.2%. Overall, MAGNETO proves that unintentional magnetic emissions can be considered as a viable and reliable means to fingerprint read-only USB flash drives. Finally, future research directions in this domain are also discussed.

Simone Raponi, Savio Sciancalepore, Gabriele Oligeri et al.

Assisted-navigation applications have a relevant impact on our daily life. However, technological progress in virtualization technologies and Software-Defined Radios recently enabled new attack vectors, namely, road traffic poisoning. These attacks open up several dreadful scenarios, which are addressed in this contribution by identifying the associated challenges and proposing innovative countermeasures.

Maurantonio Caprolu, Simone Raponi, Gabriele Oligeri et al.

A new cybersecurity attack,where an adversary illicitly runs crypto-mining software over the devices of unaware users, is emerging in both the literature and in the wild . This attack, known as cryptojacking, has proved to be very effective given the simplicity of running a crypto-client into a target device. Several countermeasures have recently been proposed, with different features and performance, but all characterized by a host-based architecture. This kind of solutions, designed to protect the individual user, are not suitable for efficiently protecting a corporate network, especially against insiders. In this paper, we propose a network-based approach to detect and identify crypto-clients activities by solely relying on the network traffic, even when encrypted. First, we provide a detailed analysis of the real network traces generated by three major cryptocurrencies, Bitcoin, Monero, and Bytecoin, considering both the normal traffic and the one shaped by a VPN. Then, we propose Crypto-Aegis, a Machine Learning (ML) based framework built over the results of our investigation, aimed at detecting cryptocurrencies related activities, e.g., pool mining, solo mining, and active full nodes. Our solution achieves a striking 0.96 of F1-score and 0.99 of AUC for the ROC, while enjoying a few other properties, such as device and infrastructure independence. Given the extent and novelty of the addressed threat we believe that our approach, supported by its excellent results, pave the way for further research in this area.

Roberto Di Pietro, Gabriele Oligeri

Jamming techniques require just moderate resources to be deployed, while their effectiveness in disrupting communications is unprecedented. In this paper we introduce several contributions to jamming mitigation. In particular, we introduce a novel adversary model that has both (unlimited) jamming reactive capabilities as well as powerful (but limited) proactive jamming capabilities. Under this powerful but yet realistic adversary model, the communication bandwidth provided by current anti-jamming solutions drops to zero. We then present Silence is Golden (SiG): a novel anti jamming protocol that, introducing a tunable, asymmetric communication channel, is able to mitigate the adversary capabilities, enabling the parties to communicate. For instance, with SiG it is possible to deliver a 128 bits long message with a probability greater than 99% in 4096 time slots in the presence of a jammer that jams all the on-the-fly communications and the 74% of the silent radio spectrum---while competing proposals simply fail. The provided solution enjoys a thorough theoretical analysis and is supported by extensive experimental results, showing the viability of our proposal.