Radio Frequency Fingerprinting (RFF) exploits inherent hardware-level imperfections of wireless transmitters as unclonable identifiers for device identification. These unique signatures, concealed in transmitted signals, inevitably experience complex distortions during wireless propagation (i.e., coupled with ambient noise and channel fading), making it extremely challenging for reliable extraction. Despite substantial research efforts dedicated to advancing effective fingerprint extraction techniques, current approaches still struggle in handling fingerprint robustness under distance variations, leading to severe SNR fluctuations and complex multipath effects. To address this gap, we propose the first unsupervised framework for distance-invariant radio frequency fingerprinting, eliminating dependence on labeled target domain data. Specifically, we first preprocess raw RF samples by confining them within a specified variation range and filtering noisy high-frequency components while avoiding aliasing. For source domain data, we then propose a set of physics-inspired data augmentation techniques designed to emulate realistic wireless signal propagation effects. Building on this, we introduce a dual alignment contrastive learning method to explicitly decouple identity-discriminative features, ensuring the model focuses on device-specific traits. Furthermore, we incorporate a pseudo-labeling-based domain adaptation module to refine the model for the unlabeled target domain, enhancing its generalization to unseen distances. Extensive experiments on public datasets show that our method achieves the identification accuracy outperforming state-of-the-art approaches by 40%, while maintaining computational efficiency suitable for edge deployment.

The computational and memory demands of deep neural networks for vision tasks remain a critical barrier to their deployment on resource-constrained edge devices. Although knowledge distillation (KD) effectively transfers over-parameterized models’ knowledge into compact students, its efficacy diminishes substantially when a significant capacity gap exists between them. Current approaches often impose linear mapping constraints between output distributions, an assumption that becomes prohibitively restrictive under such capacity gaps. This paper proposes a fundamental relaxation of alignment requirements. Specifically, rather than enforcing strict parametric relationships, we experimentally validate that preserving monotonic rank correlation between teacher and student outputs suffices for effective knowledge transfer. To operationalize this insight, we introduceMonotonic Rank Knowledge Distillation, a novel framework that leverages differentiable approximations of Kendall’s rank correlation coefficient to measure and optimize rank-order consistency. Our methodology further decomposes rank correlation into inter-class and intra-class components, ensuring the student network retains both global discriminative patterns and fine-grained categorical distinctions inherent to the teacher’s outputs. Extensive experiments across CIFAR-100 and ImageNet-1K benchmarks validate the effectiveness of our approach, demonstrating consistent performance gains over state-of-the-art distillation methods. The proposed framework achieves superior generalization across diverse architectures, including CNN-based, MLP-based, and ViT-based, with particular efficacy in various compression scenarios.

Trajectory similarity query can retrieve a set of trajectories similar to the user's query from the database and is widely used in various fields such as travel recommendations. Previous studies mainly focused on accelerating trajectory similarity search in plaintext. However, with the increasing concern about privacy protection in outsourced cloud environments, conducting trajectory similarity queries while preserving privacy becomes a significant challenge. This paper proposes efficient privacy-preserving top-$k$and range similarity queries over trajectory data. We leverage Discrete Synchronous Euclidean Distance (DSED) to measure the spatio-temporal similarity of trajectory data, and employ a filter-then-refine strategy to enhance efficiency. Specifically, Hilbert curve-based filtering is first applied to exclude a large portion of dissimilar trajectories, followed by homomorphic encryption-based refinement to retrieve precise results. Security analysis demonstrates that our schemes protect the privacy of trajectory data, query requests, and query results. Finally, extensive experimental results indicate that the proposed methods achieve a trade-off between data availability and privacy, achieving over 99% average precision while initially filtering out 90% of dissimilar trajectories, and improving query efficiency by at least an order of magnitude.

Physical layer secret key generation has emerged as a promising approach for secret key establishment in wireless networks. Unlike traditional quantization-based methods, recent studies have explored matching the patterns of segmented channel samples of equal length for key agreement. However, equal-length segmentation either suffers from inconsistencies between users for short segments or a reduced key generation rate for long ones. To address these issues, we propose a Variable-length Segment Matching-based Secret Key Generation method, VSM-SKG, which adaptively partitions channel samples into variable-length segments to enhance overall matching accuracy, key generation rate, and encryption strength. Specifically, we introduce a dissimilarity-enhanced segmentation and calibration strategy that partitions channel samples into variable-length segments to enlarge segment-wise dissimilarity. To achieve consistent key recovery between users, we develop a dynamic path-aware key generation method that identifies potential segmentation patterns and generates agreed-upon secret keys using a recursive approach combined with a fast retrieval mechanism. Theoretical analyses and real-world experiments validate the attributes of VSM-SKG in terms of accuracy, efficiency, and security in key generation.

The widespread adoption of wrist wearables featuring a range of sensors has led to a substantial user base. Among these sensors, Photoplethysmography (PPG) sensors have gained prominence for their affordability and non-intrusive nature, particularly in the context of vital signs monitoring. However, PPG sensors, with their potential for gesture recognition, also have the ability to extract sensitive private information from users. As the authentication method for many critical scenarios, once a PIN is compromised, it can lead to unbearable consequences. In this study, we introduce PPGLogger, a new side-channel attack that leverages PPG sensors for PIN inference. PPGLogger effectively separates signals originating from heartbeats and keystrokes to minimize noise interference. Additionally, we devise a keystroke detection technique capable of identifying and segmenting individual keystroke signals within the waveforms. For the keystroke inference component, we employ an Rocket-based classifier to achieve precise keystroke recognition. Through extensive real-world experiments, our findings are compelling. PPGLogger achieves an impressive accuracy rate of 76.3% in recognizing 10 digits. Furthermore, we demonstrate that PPGLogger can attain over 50% accuracy in classifying single keystrokes with a minimal dataset of just 28 samples, equivalent to merely 7 PIN entries. This underscores the efficacy and potential threat posed by PPGLogger in compromising user security.

Optical image stabilization (OIS), powered by a special micro-electromechanical structure in the camera lenses to compensate for the optical distortion caused by camera shakes, has become an indispensable feature in many smartphones. However, we discover that this seemingly benign component can be exploited to eavesdrop on nearby audio signals, posing a significant threat to people's privacy during conversations or phone calls. Specifically, the OIS component can be influenced by external acoustic stimuli leading to slight vibrations, and at the same time, the coil and magnetized components inside the OIS induce electromagnetic leakage as they vibrate, according to Faraday's Law of Electromagnetic Induction. This electro-magnetic leakage contains voice information that can be used to recover the audio signals if intercepted by individuals with malicious intent. Inspired by the above discovery, we propose OISMic, a new acoustic eavesdropping attack that takes advantage of sound-induced OIS vibrations on smartphones. Unlike other existing acoustic eavesdropping attacks, eavesdropping exploiting OIS vibrations not only overcomes the constraints imposed by system permissions for many sensor-based approaches but is also immune to ultrasonic jammer that hinders the methods relying on microwave or light reflections to sense sound-induced vibrations. To execute this non-trivial attack in practical scenarios, we developed a prototype circuit that has a compact design capable of capturing the electromagnetic leakage caused by OIS vibrations. After converting the collected leaked electromagnetic signals into audio signals, a software-based phase-locked loop (PLL) method is developed to enhance the representation of voice components. Meanwhile, to reconstruct the weak audio signals, we also designed a diffusion-based neural network to learn the distribution of electromagnetic noise within the audio spectrum. Extensive experiments indicate that OISMic can accurately reconstruct voice under various scenarios, achieving an average word correct rate of 90.57 % across different devices.

Personal Identification Number (PIN), as one of the primary means of protecting digital properties and privacy on mobile devices, has been suffering from shoulder surfing attacks and weak password guessing for the long term. Recent years witness the growing interest in two-factor authentication that takes advantage of two different ways for mutual verification, thereby strengthening user authentication's accuracy and reliability. Especially with the popularity of smartwatches, more physiological signals are readily available to facilitate two-factor authentication. This paper presents a lightweight and unobtrusive two-factor authentication scheme, P2Auth, integrating the PIN and unique keystroke-related Photoplethysmography (PPG) measurement on wearables. Specifically, we propose the transformation of the multivariate PPG signal induced by the keystrokes to extract reliable biometric features. We develop short-time energy-based methods to identify the input cases, thus enabling support the authentication for both one-handed and two-handed input cases. Furthermore, we also consider the situation where there is no fixed PIN and design a new enhanced privacy scheme by combining the PPG measurements of different keystrokes to improve authentication security. The experiments involving 15 volunteers demonstrate that our prototype system can achieve an average authentication accuracy of over 95% for one-handed cases and over 90% for two-handed cases.