Wireless sensing technologies, leveraging ubiquitous sensors such as acoustics or mmWave, can enable various applications such as human motion and health tracking. However, the recent trend of incorporating deep learning into wireless sensing introduces new challenges, such as the need for extensive training data and poor model generalization. As a remedy, data augmentation is one solution well-explored in other fields such as computer vision; yet they are not directly applicable due to the unique characteristics of wireless signals. Hence, we propose a custom data augmentation framework, WixUp, tailored for wireless human sensing. Our goal is to build a generic data augmentation framework applicable to various tasks, models, data formats, or wireless modalities. Specifically, WixUp achieves this by a custom Gaussian mixture and probability-based transformation, making any data formats capable of an in-depth augmentation at the dense range profile level. Additionally, our mixing-based augmentation enables un-supervised domain adaptation via self-training, allowing model training with no ground truth labels from new users or environments in practice. We extensively evaluated WixUp across four datasets of two sensing modalities (mmWave, acoustics), two model architectures, and three tasks (pose estimation, identification, action recognition). WixUp provides consistent performance improvement (2.79%-84.25%) across these various scenarios and outperforms other data augmentation baselines.

The surging popularity of home assistants and their voice user interface (VUI) have made them an ideal central control hub for smart home devices. However, current form factors heavily rely on VUI, which poses accessibility and usability issues; some latest ones are equipped with additional cameras and displays, which are costly and raise privacy concerns. These concerns jointly motivate Beyond-Voice, a novel high-fidelity acoustic sensing system that allows commodity home assistant devices to track and reconstruct hand poses continuously. It transforms the home assistant into an active sonar system using its existing onboard microphones and speakers. We feed a high-resolution range profile to the deep learning model that can analyze the motions of multiple body parts and predict the 3D positions of 21 finger joints, bringing the granularity for acoustic hand tracking to the next level. It operates across different environments and users without the need for personalized training data. A user study with 11 participants in 3 different environments shows that Beyond-Voice can track joints with an average mean absolute error of 16.47mm without any training data provided by the testing subject.

The surging popularity of home assistants and their voice user interface (VUI) have made them an ideal central control hub for smart home devices. However, current form factors heavily rely on VUI, which poses accessibility and usability issues; some latest ones are equipped with additional cameras and displays, which are costly and raise privacy concerns. These concerns jointly motivate Beyond-Voice, a novel high-fidelity acoustic sensing system that allows commodity home assistant devices to track and reconstruct hand poses continuously. It transforms the device into an active sonar system using its existing onboard microphones and speakers. By feeding a high-resolution range profile to the deep learning model, we can localize 21 finger joints in 3D, bringing the granularity for acoustic hand tracking to the next level. A user study with 11 participants in 3 different environments shows that Beyond-Voice can track joints with an average mean absolute error of 16.47mm for unseen environments and users.

Online fraud is the well-known dark side of the modern Internet. Unsupervised fraud detection algorithms are widely used to address this problem. However, selecting features, adjusting hyperparameters, evaluating the algorithms, and eliminating false positives all require human expert involvement. In this work, we design and implement an end-to-end interactive visualization system, FDHelper, based on the deep understanding of the mechanism of the black market and fraud detection algorithms. We identify a workflow based on experience from both fraud detection algorithm experts and domain experts. Using a multi-granularity three-layer visualization map embedding an entropy-based distance metric ColDis, analysts can interactively select different feature sets, refine fraud detection algorithms, tune parameters and evaluate the detection result in near real-time. We demonstrate the effectiveness and significance of FDHelper through two case studies with state-of-the-art fraud detection algorithms, interviews with domain experts and algorithm experts, and a user study with eight first-time end users.

RFID is an increasingly ubiquitous technology widely adopted in both the industry and our daily life nowadays. But when it comes to eavesdropping, people usually pay attention to devices like cameras and mobile phones, instead of small-volume and battery-free RFID tags. This work shows the possibility of using prevalence RFIDs to capture and recognize the acoustic signals. To be specific, we attach an RFID tag on an object, which is located in the vicinity of the sound source. Our key innovation lies in the translation between the vibrations induced when the sound wave hits the object surface and the fluctuations in the tag's RF signals. Although the inherent sampling rate of commercial RFID devices is deficient, and the vibrations are very subtle, we still extract characteristic features from imperfect measurements by taking advantage of state-of-the-art machine learning and signal processing algorithms. We have implemented our system with commercial RFID and loudspeaker equipment and evaluated it intensively in our lab environment. Experimental results show that the average success rate in detecting single tone sounds can reach as high as 93.10%. We believe our work would raise the attention of RFID in the concern of surveillance and security.

Opportunistic networks have emerged as prospective network architecture for smart mobile devices related applications. The main obstacle for designing routing protocol in opportunistic networks is their inherent dynamic nature. Unknown future link patterns lead to blind and inefficient packet forwarding behavior. Lots of efforts have focused on the future link prediction problem, but contact patterns among node pairs are too different to be predicted by one single method. To this end, we propose a combination link prediction method for opportunistic network routing named Combo-Pre. Combo-Pre applies periodic pattern mining methods to predict frequent and periodic contacts, decision tree method to predict frequent but non-periodic contacts and Adamic-Adar methods in complex networks to predict infrequent contacts. Experimental results show that Combo-Pre outperforms state-of-the-art link prediction methods in opportunistic networks. These results can be applied in routing protocol design to decrease routing cost and promote delivery rate.