Growing interest in wireless sensing, a cornerstone of the Artificial Intelligence of Things (AIoT), stems from its ability to gauge target states through nearby wireless signals. However, the escalating count of AIoT nodes escalates redundant data flow and exacerbates energy usage in AI cloud infrastructures. This amplifies the urgency for machine learning techniques that function in proximity to, or directly within, sensors. In light of this, we present the Radio-Frequency Neural Network (RFNN), a novel architecture that uses cost-effective transmissive intelligent surfaces to mimic the functions of a traditional neural network near (or in) sensors, transforming sensory nodes into intelligent terminals primed for machine learning. We first devised a unique training algorithm to mitigate the issues arising from unmodelable error-backward propagation; secondly, we incorporated contrastive learning to address the issue of blind labels stemming from environmental uncertainties. Our RFNN prototype, resonating at a 5 GHz WiFi bandwidth, has been honed across nine varied sensing tasks. The rigorous evaluation shows that it achieves a mean accuracy of 91.5% while consuming only 67.2 μJ of energy. This positions RFNN as a match in inferencing prowess to its electronic neural network counterparts but with significantly diminished energy demands.

Cross-Technology Communication (CTC) is an emerging technology that enables physical-layer direct communication from a WiFi sender to other Internet of Things (IoT) receivers via waveform emulation. Previous research has primarily relied on the reverse engineering to identify the suitable WiFi payloads capable of emulating waveforms similar to desired IoT packets (e.g., ZigBee). However, this approach has several limitations, including irreversibility, scalability challenges, symbol misalignment, and an over-reliance on empirical methods. In this work, we present XiTuXi, a one-size-fits-all solution to automatically achieve the CTC by taking advantage of the neural machine translation (NMT). Inspired by the task comparability between CTC and homophony-based cross-linguistic communication, we employ a well-known NMT model called Transformer to learn the rationale behind translating bit sequences for CTC without human intervention. Specifically, we introduce forward engineering as a solution to tackle the challenge of acquiring training datasets. By utilizing XiTuXi, we effortlessly achieved CTC across 30 protocol combinations (including 802.11b, g, n, ax, ah → Zig-Bee, Bluetooth, LoRa, and Sigfox), ultimately freeing experts from the tedious tasks they faced previously.

Wireless sensing has attracted considerable attention because it can sense the state of the targets by analyzing the surrounding wireless signals, which has become the key role of the artificial intelligence of things (AIoT). As the number of sensory nodes increases, large amounts of redundant data are exchanged between sensory terminals and the AI cloud. To process such large amounts of data efficiently and decrease power consumption, a machine-learning approach that operates close to or inside sensors must be developed. To this end, we present the radio-frequency neural network (RFNN), a physical neural network taking advantage of a group of transmissive intelligent surfaces (i.e., metasurfaces) to mimic the computations of a fully-connected neural network. The design is spurred by the capability of RFNNs to perform expensive multiplication and additions at the speed of light, with ultra-low power consumption. We prototype RFNN at 5 GHz for WiFi sensing regarding nine wireless sensing tasks. Extensive evaluations demonstrate the comparably equivalent inference ability as the conventional electronic neural networks while consuming less energy.

Cross-Technology Communication (CTC) is an emerging technology that enables physical-layer direct communication from a WiFi sender to other Internet of Things (IoT) receivers via waveform emulation. The previous works use the reverse engineering to find the appropriate WiFi payload that can emulate the waveform similar to the desired IoT packet in the format of the IoT protocol (e.g., ZigBee). Unfortunately, the reverse engineering approach suffers from many limitations, such as being non-reversible and unscalable, misaligning symbols, and over-relying on empiricism. In this work, we present XiTuXi, a one-size-fits-all solution to automatically achieve the CTC by taking advantage of the neural machine translation (NMT), inspired by the task comparability between CTC and homophony-based cross-linguistic communication. We employ a well-known NMT model called Transformer to learn the bit-sequence to bit-sequence translation rationale behind the CTC without human intervention. Particularly, we introduce the forward engineering to address the dilemma of acquiring training datasets. By using XiTuXi, we achieved the CTC with 30 protocol combinations (ie., 802.11b, g, n, ax, ah Å ZigBee, Bluetooth, LoRa, and Sigfox) effortlessly, which ultimately liberates the experts from previous tedious tasks.

Obtaining fine-grained spatial information is of practical importance in Radio Frequency Identification (RFID)-based systems for enabling multi-object identification. However, as high-precision positioning remains impractical in commercial-off-the-shelf (COTS)-RFID systems, researchers propose to combine computer vision (CV) with RFID and turn the positioning problem into a matching problem. Promising though it seems, current methods fuse CV and RFID through converting traces of tagged objects extracted from videos by CV into phase sequences for matching, which is a dimension-reduced procedure causing loss of spatial resolution. Consequently, they fail in harsh conditions like small tag intervals and low reading rates. To address the limitation, we propose TagFocus to achieve fine-grained multi-object identification with visual aids in RFID systems. The key observation is that traces generated through different methods shall be compatible if they are of one identical object. Accordingly, a Transformer-based sequence-to-sequence (seq2seq) model is trained to generate a simulated trace for each candidate tag-object pair. And the trace of the right pair shall best match the observed trace directly extracted by CV. A prototype of TagFocus is implemented and extensively assessed in lab environments. Experimental results show that our system maintains a matching accuracy of over 91% in harsh conditions, outperforming state-of-the-art schemes by 27%.

Obtaining fine-grained spatial information is of practical importance in RFID-based applications. However, high-precision positioning remains a challenging task in commercial-off-the-shelf (COTS) RFID systems. Inspired by progress in the computer vision (CV) field, researchers propose to combine CV with RFID systems and turn the positioning problem into a matching problem. Promising though it seems, current methods fuse CV and RFID through converting traces of tagged objects extracted from videos by CV into phase sequences for matching, which is a dimension-reduced procedure causing loss of spatial resolution. Consequently, they fail in more harsh conditions such as small tag intervals and low reading rates of tags. To address the limitation, we propose TagFocus, a more robust RFID-enabled system for fine-grained multi-object identification and tracking with visual aids. The key observation of TagFocus is that traces generated by different methods shall be compatible if they are acquired from one identical object. Leveraging this observation, an attention-based sequence-to-sequence (seq2seq) model is trained to generate a simulated trace for each candidate tag-object pair. And the trace of the right pair shall best match the observed trace directly extracted by CV. A prototype of TagFocus is implemented and extensively assessed in lab environments. Experimental results show that our system maintains a matching accuracy of over 89% in harsh conditions, outperforming state-of-the-art schemes by 25%.