Advanced wireless communication systems adopt deep learning (DL) approaches to achieve automatic modulation recognition (AMR) for spectrum monitoring and management, especially in the spectrum bands supporting diverse co-existing wireless protocols. In practical wireless environments, wireless signals can easily get compromised by malicious noise, intentional interference, and adversarial attacks, reducing the effectiveness of AMR. By exploiting DL model vulnerabilities, an undetectable perturbation added to the wireless signal can cause misclassification, resutling in serious consequences including decoding errors, throughput degradation, and communication disruption. Facing the limitations of existing works on defending against wireless adversarial attacks, this work innovates the Transformer model to design an adversarial robust AMR driven by exploring temporal correlation in time-sequence wireless signals. Instead of directly applying the Vision Transformer (ViT), we first innovate a feature extraction module specifically for radio frequency (RF) signals from both the time and frequency domains, together with an adaptive positional embedding to the Transformer encoder for enhancing AMR accuracy. To mitigate the noise effect in practical wireless communication, we then propose a noise-adaptive adversarial training scheme on the developed Transformer-based model using adversarial examples crafted by white-box attackers. To show the scheme's efficiency, effectiveness, and robustness, our proposed design has been thoroughly evaluated via a self-collected real-world dataset consisting of over 30 million wireless signal data samples with 21 modulation schemes in both indoor and outdoor scenarios. Our results reach a maximum accuracy of 94.17% in AMR classification and 71.2 % under adversarial attacks. Besides, for the first time, we demonstrate the robustness of our design under a real wireless adversarial attack in real-time. Datasets and code available in https://github.com/coulsonlee/Robust-ViT-for-AMR-SP2025.

Memory Electromagnetic Radiation (EMR) allows attackers to manipulate the DRAM of infiltrated systems to leak sensitive secret information. Although most of the existing works have demonstrated its feasibility, practical concerns, such as the ideal electromagnetic environment and stationary attacking layout, make the covert channel attack less convincing, especially in vulnerable sites such as offices and data centers. This work removes the above impractical assumptions to uncover the potential of memory EMR by proposing the first parallel EMR covert communication protocol. Our design reshapes the current "1-to-1" covert communication mode to "n-to-1" mode via a novel pattern-based 2-dimensional symbol encoding scheme, allowing multiple victim computers to simultaneously perform data exfiltration to one attacker (the receiver) without mutual interference. Meanwhile, this novel scheme design also enables the very first mobile attacker, i.e., a smartphone connected to a software-defined radio (SDR) dongle, to capture parallel memory EMR signals in a volatile environment. Extensive experiments are conducted to verify the performance in a volatile environment with different parameter configurations, distances, motion modes, shielding materials, orientations, hardware configurations, and SDR platforms. Our experimental results demonstrate that FreeEM can support up to 4 parallel memory EMR transmissions to achieve an overall throughput of 625Kbps and a decoding accuracy of 96.88%. The maximum communication distance can reach up to 20 meters.

Human faces have been widely adopted in many applications and systems requiring a high-security standard. Although face authentication is deemed to be mature nowadays, many existing works have demonstrated not only the privacy leakage of facial information but also the success of spoofing attacks on face biometrics. The critical reason behind this is the failure of liveness detection in biometrics. This work advances most biometric-based user authentication schemes by exploring dynamic biometrics (human facial activities) rather than traditional static biometrics (human faces). Inspired by observations from psychology, we propose the mmFaceID to leverage humans' dynamic facial activities when performing word reading for achieving robust, highly accurate, and effective user authentication via mmWave sensing. By addressing a series of technical challenges of capturing micro-level facial muscle movements using a mmWave sensor, we build a neural network to reconstruct facial activities via estimated expression parameters. Then, unique features can be extracted to enable robust user authentication regardless of relative distances and orientations. We conduct comprehensive experiments on 23 participants to evaluate mmFaceID in terms of distances/orientations, length of word lists, occlusion, and language backgrounds, demonstrating an authentication accuracy of 94.7%. We also extend our evaluation in a real IoT scenario. By speaking real IoT commends, the average authentication accuracy can reach up to 92.28%.

Indoor localization has gained significant attention in recent years due to its various applications in smart homes, industrial automation, and healthcare, especially since more people rely on their wireless devices for location-based services. Deep learning-based solutions have shown promising results in accurately estimating the position of wireless devices in indoor environments using wireless parameters such as Channel State Information (CSI) and Received Signal Strength Indicator (RSSI). However, despite the success of deep learning-based approaches in achieving high localization accuracy, these models suffer from a lack of generalizability and can not be readily-deployed to new environments or operate in dynamic environments without retraining. In this paper, we propose meta-learning-based localization models to address the lack of generalizability that persists in conventionally trained DL-based localization models. Furthermore, since meta-learning algorithms require diverse datasets from several different scenarios, which can be hard to collect in the context of localization, we design and propose a new meta-learning algorithm, TB-MAML (Task Biased Model Agnostic Meta Learning), intended to further improve generalizability when the dataset is limited. Lastly, we evaluate the performance of TB-MAML-based localization against conventionally trained localization models and localization done using other meta-learnina algorithms.

The wide deployment of IoT devices has resulted in a critical shortage of spectrum resources. Many IoT devices coexist on the same spectrum band, where the network performance is always degraded. As a promising solution, the Cross-Technology Communication (CTC) enables the direct communication among heterogeneous IoT devices. Unfortunately, the emerging cross-technology attacks have demonstrated their high success rates in terms of spoofing the end IoT devices or jamming the communication channels. In this paper, we investigate a novel cross-technology jamming issue for a distributed heterogeneous IoT system. Compared with traditional jamming methods, the cross-technology jammer has a much higher jamming power, wider jamming bandwidth, and stronger stealthiness, all of which deserve a complete re-thinking of defensive mechanisms. Therefore, we propose a hybrid anti-jamming scheme that jointly considers frequency hopping and power control techniques. Specifically, we model the anti-jamming process as a Markov Decision Process (MDP) and adopt Deep Q-Network (DQN) to find the optimal strategy. Extensive real-world experiments show that the goodput (payload data) of our anti-jamming scheme can achieve up to 2X and 1.39X than the passive and random anti-jamming approaches, respectively. In particular, our anti-jamming scheme provides 78% of goodput with the presence of a cross-technology jammer, outperforming existing passive and random anti-jamming scheme designs at 37.6% and 54.1%.