Leveraging wearable devices for motion reconstruction has emerged as an economical and viable technique. Certain methodologies employ sparse Inertial Measurement Units (IMUs) on the human body and harness data-driven strategies to model human poses. However, the reconstruction of motion based solely on sparse IMU data is inherently fraught with ambiguity, a consequence of numerous identical IMU readings corresponding to different poses. In this paper, we explore the spatial importance of sparse sensors, supervised by text that describes specific actions. Specifically, uncertainty is introduced to derive weighted features for each IMU. We also design a Hierarchical Temporal Transformer (HTT) and apply contrastive learning to achieve precise temporal and feature alignment of sensor data with textual semantics. Experimental results demonstrate our proposed approach achieves significant improvements in multiple metrics compared to existing methods. Notably, with textual supervision, our method not only differentiates between ambiguous actions such as sitting and standing but also produces more precise and natural motion.

Deep learning excels in advanced inference tasks using electronic neural networks (ENN), but faces energy consumption and limited computation speed challenges. To mitigate this, optical neural networks (ONNs) were developed, utilizing light for computations. However, their high manufacturing costs limited accessibility. In this work, we first introduce the binary optical neural network (BONN) - a streamlined ONN variant with binarized weights, which significantly reduces fabrication complexities and costs. Specifically, we address (i) the development of a binarization weight function aligned with backward-error propagation, and (ii) a simulation-based training for extra-large neural networks housing millions of neurons. We prototype six BONNs, each comprising four 0.8 × 0.8mm2 layers with one million 800 nm diameter neurons. Costs are cut to 0.13 USD per layer, marking a substantial decrease of 769× from previous ONNs. Experimental results reveal BONNs consume 2, 405× less power than leading ENNs while maintaining an average recognition accuracy of 74% across six datasets.

Although billions of battery-free backscatter devices (e.g., RFID tags) are intensively deployed nowadays, they are still unsatisfying in the two major performance limitations (i.e., short reading range and high miss reading rate) resulting from the current harvesting inefficiency. The classic beamforming technique is regarded as the most promising solution to address the issue. However, applying it to backscatter systems meets the deadlock start problem, i.e., without enough power, the backscatter cannot wake up to provide channel parameters; but, without channel parameters, the system cannot form beams to provide power. In this work, we propose a new paradigm calledtransfer beamforming(${\sf TBF}$), namely, the beamforming strategies can be transferred from reference tags with known positions to power up other unknown neighbor tags of interest. In short, transfer beamforming (is accomplished) via (launching) beamforming (to reference tags first) for (the purpose of) transfer. To do so, we adopt the semi-active tags as the reference tags, which can be powered up with a normal reader in a wide range. Then the beamforming is initiated and transferred to power up the low-sensitive but cost-effective passive tags surrounded by reference tags. A prototype evaluation of${\sf TBF}$with 8 transmitting antennas presents a 99.9% inventory coverage rate in a crowded warehouse with 2,160 RFID tags. Our comprehensive evaluation reveals that${\sf TBF}$can improve the power transmission by 6.9 dB and boost the inventory speed by 2× compared with state-of-art methods.

Although billions of battery-free backscatter devices (e.g., RFID tags) are intensively deployed nowadays, they are still unsatisfying in performance limitations (i.e., short reading range and high miss-reading rate) resulting from power harvesting inefficiency. However, applying classic beamforming technique to backscatter systems meets the deadlock start problem, i.e., without enough power, the backscatter cannot wake up to provide channel parameters; but, without channel parameters, the system cannot form beams to provide power. In this work, we propose a new beamforming paradigm called transfer beamforming (TBF), namely, beamforming strategies can be transferred from reference tags with known positions to power up unknown neighbor tags of interest. Transfer beamforming (is accomplished) via (launching) beamforming (to reference tags firstly) for (the purpose of) transfer. To do so, we adopt semi-active tags as reference tags, which can be easily powered up with a normal reader. Then beamforming is initiated and transferred to power up passive tags surrounded by reference tags. A prototype evaluation of TBF with 8 antennas presents a 99.9% inventory coverage rate in a crowded warehouse with 2,160 RFID tags. Our evaluation reveals that TBF improves the power transmission by 6.9 dB and boosts the inventory speed by 2 × compared with state-of-art methods.

Physical-layer identification aims to identify wireless devices during RF communication by exploiting the imperfections of their radio circuitry, i.e., hardware fingerprint. Previous work proposed several hardware fingerprints for RFIDs (e.g., TIE, ABD, PSD, etc). However, these proposed fingerprints suffer from either unscalability or acquisition inefficiency. This work presents RF-DNA, a new hardware fingerprint composed of millions of Dual Natural Attributes (DNA) organized in a helical structure, where a pair of DNA represents a tag's intrinsic response at some frequency. We take advantage of the frequency agnostic phenomenon that a commercial RFID tag can respond within a wider band than the regulated, to acquire 10X more features than previous fingerprints. At the heart of this work are the context-free acquisition approach to extracting DNA from backscatter signals; and the accurate DNA matching algorithm for verifying a tag's identity. A total of 160,000 RF-DNA instances were collected from 16,000 tags using a customized automatic acquisition system. We subsequently carried out large-scale experiments to test the identification accuracy of RF-DNA and previously proposed fingerprints. Our comprehensive evaluation reveals that RF-DNA can achieve a mean accuracy of 95.98%. In contrast, those of previous fingerprints fall to 60% below when in face of thousands of tags.