Abstract Millimeter-wave radar has shown significant potential in privacy-preserving human activity recognition. However, the lack of diverse radar datasets across various scenarios poses a challenge to the robustness and generalization of deep learning models. To address this limitation, existing works mainly focus on synthesizing micro-doppler data from video, range-doppler data, which provides an extra dimension, has been overlooked due to challenges caused by signal offsets. In this paper, we propose a comprehensive approach for synthesizing range-doppler data from videos by leveraging computer vision techniques and principles of camera imaging. Furthermore, we implement a map enhancement and classification model to facilitate human activity recognition. Our approach is validated on a custom dataset, where the proposed range-doppler synthesis method and classification model achieve an accuracy of 97.3% for activity recognition tasks. This performance is comparable to that of vision-based HAR methods, demonstrating the effectiveness of our proposed scheme in achieving privacy-preserving human activity recognition.

Human semantic segmentation facilitates the recognition of different parts of the human body and is essential for applications such as sports analysis and fall detection. To integrate human semantic segmentation into the domain of radio front-end sensing, this article introduces mmSeg, an innovative system that leverages commercial millimeter-wave radar for human semantic segmentation. However, the inherent propagation characteristics of mmWave signals often result in highly sparse point clouds with limited semantic information and the entanglement of temporal-topological features, making human semantic segmentation a challenging task. To address these challenges, mmSeg (i) first introduces a radar cross-section (RCS) calculation method suitable for commercial millimeter-wave radar to enhance the semantic information of radar point clouds at a coarse granularity; (ii) further designs a temporal-topological decoupling network to obtain the fine-grained human semantic segmentation results; (iii) constructs an efficient loss function for end-to-end training, based on an adjacency matrix graph to improve the segmentation performance. We evaluate mmSeg on our self-built millimeter-wave dataset HSS and a public dataset MM-Fi. mmSeg achieves an average point cloud segmentation accuracy of 87.74% on the HSS dataset and 84.18% on the MM-Fi dataset, outperforming the existing methods in both cases.

The surge in location-based services has driven the demand for accurate indoor localization techniques, with WiFi based localization emerging as a promising solution due to its extensive coverage in indoor environments. This paper presents BFI-L10N, an indoor localization method that leverages beam forming feedback information (BFI) obtained from standard multi-user multiple-input multiple-output (MU-MIMO) WiFi operations. Unlike traditional channel state information (CSI) based methods that require vendor-specific firmware patches or restricted device support, BFI leverages standardized MU-MIMO operations, enabling compatibility with off-the-shelf WiFi devices. BFI-L10N processes the BFI data collected during the beam forming process through a deep learning framework and uses a BERT model for localization. Compared to CSI-based systems, BFI-L10N offers advantages such as reduced overhead, enhanced sensitivity, compatibility with standard devices, and real-time predictions. Our experimental results in two distinct indoor environments demonstrate that BFI-L10N achieves average localization accuracies of 10.7 cm and 15.5 cm in a research laboratory and a conference room, respectively, outperforming the state-of the-art CSI techniques by 28%. Moreover, the BERT model can be fine-tuned after pre-training across multiple locations, which enhances the versatility of BFI-L10N. This paper presents a novel perspective on WiFi sensing and lays the foundation for practical indoor localization using standard WiFi infrastructure.

We present HEScatter, a high-efficiency ambient backscatter system that simultaneously improves carrier, power, and transmission efficiency. To improve carrier efficiency, we choose cellular signal as the carrier due to its continuous transmission nature. Specifically, to ensure low power, we design low-power periodic template matching based on the periodicity of cellular signals to trade time for synchronization accuracy. Further, we calibrate the drift introduced by Sampling Frequency Offset (SFO) to increase carrier utilization. In addition, we exploit Reference Signal (RS)-based demodulation to demodulate tag and ambient data from backscattered signals alone in various traffic patterns for efficient transmission. We prototype HEScatter using off-the-shelf FPGAs and SDRs. Extensive experiments show that HEScatter performs well in carrier utilization, power consumption and data transmission. The carrier utilization rate of HEScatter is as high as 99.97%, which is 3.0x higher than the counterpart of SyncLTE. In end-to-end transmission, the energy efficiency of HEScatter is 1.6x and 19.2x higher than LScatter+ and SyncLTE, while LScatter suffers from transmission failures. We also demonstrate the high transmission efficiency of HEScatter, as its aggregate goodput is 1.5x and 3.8x better than LScatter+ and SyncLTE respectively.

Real-time semantic recognition is crucial for virtual reality (VR) applications, but the efficient fusion of multi-modal data poses significant challenges under resource-constrained VR scenarios. While integrating millimeter-wave (mmWave) radar point clouds with vision data offers a promising solution, existing methods often suffer from excessive data overhead and degraded accuracy due to redundant and noisy information. To address this limitation, this paper presents VR-PCT, a multi-modal transformer for edge-client collaborative VR semantic recognition that fuses mmWave radar point cloud and vision data for VR applications. VR-PCT introduces a novel collaborative design where VR clients perform lightweight semantic region detection while VR edge processes multi-modal VR semantic recognition. Through efficient edge-client collaboration, VR-PCT optimizes the transmission of mmWave point cloud and vision data by transmitting only the VR semantic region of vision data instead of the entire video. Additionally, it incorporates adaptive cross-modal data selection and fusion strategies to achieve real-time semantic recognition while significantly reducing data redundancy. Across 22 participants engaged in four experimental scenes utilizing VR devices from three different manufacturers, our evaluation demonstrates that VR-PCT achieves 97.6% recognition accuracy while reducing transmission overhead by 81.5% compared to existing approaches. These results highlight the effectiveness of VR-PCT in enabling efficient and accurate multi-modal VR semantic recognition for VR applications. The code and data of VR-PCT are released onhttps://github.com/luoyumei1-a/VR-PCT.

Nowadays, there is a growing trend to deploy machine learning (ML) models on edge devices. To cope with the increasing resource requirements of current ML models, multi-accelerator edge devices that integrate CPU, GPU, NPU, or TPU in a single SoC gain popularity. However, we observe that existing ML inference serving frameworks are poor in utilizing the unique hardware architecture of these edge devices. In this paper, we present InFSCALER, an efficient ML inference serving framework tailored for multi-accelerator edge devices. InfSCALER discovers the architectural bottleneck of ML models and designs a bottleneck-aware asymmetric auto-scaling technique to facilitate efficient resource allocation for ML models on the edge. Furthermore, InfScaler capitalizes on the hardware’s unified memory feature inherent to edge devices, ensuring efficient data sharing between the asymmetrically scaled model partitions. Our experimental results show that InfScaler achieves up to $\mathbf{1 2 6. 5 9 \%}$ throughput improvement and $\mathbf{2 7. 3 2 \%}$ resource reduction while satisfying the latency requirements compared with the state-of-the-art inference serving approaches.

Pulse modulation based communication techniques have enabled various Internet of Things (IoT) applications, such as smart meters and automotive systems. However, the existing pulse modulators rely on platform-specific hardware components, leading to limited extensibility and hardware dependency when supporting diverse variants such as Pulse Position Modulation (PPM) and Pulse Amplitude Modulation (PAM). This paper introduces NN-Pulse, an innovative neural networkdefined pulse modulator designed to enhance extensibility and flexibility, ensuring compatibility with multiple pulse modulation schemes. Specifically, NN-Pulse realizes the pulse modulation process using fundamental neural network modules with carefully tailored weights, via the proposed spike neural network (SNN)-based position selection module and transposed convolutional layers for phase modulation. Evaluations show that NNPulse generates PPM and PAM signals with bit error ratios of 0.6% and 0.2%, respectively. Moreover, the time consumption of modulating a pulse symbol via NN-Pulse is only$1.4 \mu \mathrm{s}$, outperforming traditional methods by 47 times.

Serverless edge computing is attracting increasing attention due to its management-free deployment and on-demand resource provision characteristics. However, the limited resources on edge servers make efficient task scheduling crucial. Specifically, the diverse and dynamic requests in edge scenarios and the intertwined networking and computing process in serverless edge make the existing one-for-all CPU scheduling strategy fall flat. To address the CPU scheduling problem above, we introduce EdgeSched, an adaptive user-space scheduling framework for serverless functions in edge computing. EdgeSched proposes an enclave-based hybrid scheduling technique that separates network I/O tasks from diverse computational tasks efficiently to different enclaves and applies the best-fit scheduling policy on each enclave. Furthermore, EdgeSched leverages deep reinforcement learning to adaptively schedule CPU resources into different enclaves and adapts scheduling different algorithms in the user space. Through experiments on real-world edge devices and workloads, we observe a more than$2 \times$performance improvement compared to state-of-the-art methods.

Spatio-temporal prediction focuses on jointly modeling spatial correlations and temporal evolution and has a wide range of applications. Due to the heterogeneity of spatio-temporal data, accurate prediction relies on effectively integrating topological structures and sequential patterns. Although recurrent graph learning methods excel at capturing dynamic graph patterns, explicitly inferring future snapshots from historical dynamic graphs remains a significant challenge. Recently, prompt-based graph learning has shown the potential to improve future snapshot inference by leveraging node or task-specific prompts. However, these methods fail to fully capture edge information resulting in incomplete and less accurate representations of future snapshot structures. To bridge this gap, we propose ProST, a framework that Prompts future snapshots on dynamic graphs for Spatio-Temporal prediction, which leverages dynamic graph pre-training to generate a premise graph containing historical graph information and then employs prompts on the premise graph to infer explicit future snapshots. Specifically, this framework comprises three steps: Firstly, dynamic graph pre-training is performed using multi-granularity evolution graph convolution to obtain the premise graph with both local and global features of dynamic graphs. Secondly, prompt subgraphs are used to prompt node pairs and edge features within the premise graph. The subgraph prompt aggregation mechanism propagates this information to generate future snapshots. Finally, we freeze the parameters of the pre-trained model and update the subgraph prompt parameters using meta-learning to adapt to downstream spatio-temporal prediction tasks. Extensive experiments on real-world datasets validate that ProST achieves state-of-the-art performance.

Cross-Technology Communication (CTC) is a cornerstone for seamless interoperability in heterogeneous wireless environments, enabling diverse devices to coexist and cooperate effectively. In this paper, we present Wi-Fi-to-X, designed to leverage deep learning techniques to generate waveforms that are compatible with multiple communication protocols, allowing seamless data transmission between Wi-Fi and other wireless technologies such as ZigBee, LoRa, and Bluetooth. This approach enables devices operating under different wireless standards to communicate effectively without requiring hardware modifications or protocol standardization. By training a specialized neural network on simulations of Orthogonal Frequency Division Multiplexing (OFDM) and Quadrature Amplitude Modulation (QAM), we have improved the efficiency and reliability of signal processing in CTC, Wi-Fi-to-X achieves robust signal modulation and demodulation across disparate technologies, enabling communication from Wi-Fi to other IoT devices, including ZigBee, LoRa, and Bluetooth. We evaluated both USRP and commodity devices, demonstrated that Wi-Fi-to-X can achieve concurrent wireless communication from Wi-Fi to other IoT devices.

Occupancy sensing is essential for vehicle safety and security applications such as seat belt reminders, airbag deployment, intrusion detection, and child-left-behind alerts. This paper presents UMusic, a novel in-car occupancy sensing system that reuses the ultra-wideband (UWB) devices already installed for access control in modern vehicles. However, due to the compact size and metal structure, the in-car environment is full of reflected propagation paths, which cannot be precisely resolved even with UWB's wide-bandwidth feature. To overcome this challenge, UMusic introduces a reflected-path decomposition technique to extract a high-resolution power delay profile (PDP) from the channel impulse response (CIR) provided by commodity UWB devices, enabling precise environmental perception. By comparing PDPs in empty and occupied conditions, UMusic is able to detect the occupancy status in both a sedan and an SUV with multiple passengers across various scenarios. Our results show that UMusic achieves a 90.2% detection rate using a single CIR measurement collected within 50 ms, outperforming the state-of-the-art by 15.7%. When aggregating six consecutive CIR measurements, UMusic reaches 99.4% accuracy, demonstrating its effectiveness for real-world deployment.

Wi-Fi is a very common means for providing wireless access to the Internet, e.g., using the 2.4GHz Industrial, Scientific, and Medical (ISM) band and more recently also the 6 GHz band via Wi-Fi 6E. Thanks to a chip recently launched by Semtech, in the same 2.4GHz band now can also operate Long Range (LoRa), which is widely used in Internet of Things (IoT) applications due to its low power consumption and wide coverage range. To allow for data interchange among these technologies, multi-radio gateways are needed, which introduce additional costs, complexities, and potential points of failure. To address this challenge, we propose the concept of Wireless to LoRa (WiLo) to make directional communication from Wi-Fi to LoRa. WiLo uses physical-layer (PHY) communication and dedicated input chips in the 2.4 GHz band to transmit information. To overcome the modulation technique differences between Wi-Fi and LoRa, WiLo leverages narrow-band communication, a technique that generates ultra-narrowband signals using single-tone sinusoidal signals by manipulating the payload of Wi-Fi devices. These signals can be detected by LoRa Wide Area Network base stations due to their high receiver sensitivity for long-range communication. Our experiments, which make use of both Universal Software Radio Peripheral (USRP) and commodity devices, demonstrate that WiLo can achieve concurrent wireless communication over a distance of 500 m, from commercial Wi-Fi chips to a LoRaWAN, with more than 96% frame reception rate. These findings show the effectiveness of WiLo in enabling reliable and efficient wireless communication over long distances, making it particularly relevant for applications such as remote monitoring systems, sensor networks, and smart cities.

LoRaWAN is a prominent technology for Low Power Wide Area Networks (LPWAN). However, the increasing network size has introduced a significant challenge: packet collisions resulting from concurrent transmissions in LoRaWAN. Previous studies either overlooked the issue by examining limited features or tackled it with intricate receivers employing up to eight antennas. To achieve a more favorable balance between implementation cost and system performance, we introduce$\text{Hi}^{2}\text{LoRa}$—a solution utilizing highly dimensional and accurate features for LoRa concurrent decoding, implemented with only two receiving antennas. The feature dimensions are expanded through an exploration of various hardware imperfections and inherent channel state information specific to each transceiver pair. To enhance feature accuracy, low pass filters and BiLSTM networks are applied to capture and learn their temporal patterns. Additionally, an efficient collision suppression strategy is introduced to mitigate feature corruption from concurrently transmitted packets. Extensive real-world testbed evaluations demonstrate that the achievable concurrency in$\text{Hi}^{2}\text{LoRa}$approaches that of state-of-the-art approaches with significantly higher complexity (e.g., utilizing eight antennas) or exceeds prior work by a factor of 2.7 with comparable complexity (e.g., using two antennas).

Mobile applications based on machine learning (ML) are increasingly relying on offloading to the edge devices for low-latency, resource-efficient computation. Applying serverless computing for these ML applications on the edge offers a promising solution for handling dynamic workloads while meeting user-specified latency service-level objectives (SLOs). However, existing serverless frameworks, with their coarse-grained data parallelism and rigid model partitioning, are inadequate for ML inference on widely adopted edge System-on-Chip (SoC) devices. This paper presents FluidEdge, an edge-native serverless inference framework. FluidEdge identifies bottleneck operators in ML models and addresses them through a novel fine-grained intra-function latency-sensitive auto-scaling approach that dynamically scales inference bottlenecks during online serving. Additionally, it employs inter-function scaling to further prevent latency SLO violations and leverages the unified memory of edge SoCs for efficient data sharing during inference. Experimental results demonstrate that FluidEdge achieves a 37.4% latency improvement and 67.3%-87.6% SLO violation reduction compared to best-performed state-of-the-art serverless inference frameworks.

Cross-technology communication (CTC) facilitates seamless interaction between different wireless technologies. Most existing methods use reverse engineering to derive the required transmission payload, generating a waveform that the target device can successfully demodulate. However, traditional approaches have certain limitations, including reliance on specific reverse engineering algorithms or the need for manual parameter tuning to reduce emulation distortion. In this work, we present NNCTC, a framework for achieving physical layer cross-technology communication through explainable neural networks, incorporating relevant knowledge from the wireless communication physical layer into the neural network models. We first convert the various signal processing components within the CTC process into neural network models, then build a training framework for the CTC encoder-decoder structure to achieve CTC. NNCTC significantly reduces the complexity of CTC by automatically deriving CTC payloads through training. We demonstrate how NNCTC implements CTC in WiFi systems using OFDM and CCK modulation. On WiFi systems using OFDM modulation, NNCTC outperforms the WEBee and WIDE designs in terms of error performance, achieving an average packet reception ratio (PRR) of 92.3% and an average symbol error rate (SER) as low as 1.3%. In WiFi systems using OFDM modulation, the highest PRR can reach up to 99%.

In many real-life applications, wireless chargers are deployed outdoor or in public area or even unattended environment such as hotels, restaurants, retail stores. They are exposed to various risks and malicious attacks that may break them down and further incur significant cost (e.g., battery replacement and maintenance) or performance degradation. Hence, we consider the problem ofFault-tolerant wIreless chaRger placeMent (FIRM): given a set of wireless chargers and a set of tasks to be collaboratively conducted by a set of rechargeable devices, determining where to deploy the chargers to maximize the worst-cast overall task charging utility subject to the constraint that up to$\tau$chargers may break down. FIRM is a non-linear combinatorial two-level optimization problem. We first consider a relaxed version of FIRM (FIRM-R for short) corresponding to the inner optimization problem in FIRM. To address FIRM-R, we first propose an area discretization scheme to convert the infinite solution space into finite candidate positions. We then devise a power allocation method, based on which we prove that FIRM-R falls into the realm of maximizing a monotone submodular function under a uniform constraint. We then propose a constant-factor approximation algorithm to solve FIRM-R. Taking the above approximation algorithm as a subroutine, we further develop an approximation algorithm that solves FIRM with a constant-factor approximation ratio. Our extensive simulations and field experiments demonstrate that the overall charging utility of our proposed algorithm FIRM considering fault tolerance by greedy removal of$\tau$chargers outperforms the that of FIRM-R without considering fault tolerance by greedy removal of$\tau$chargers by at least 119.89%.

Backscatter communication has attracted significant attention due to its low power consumption and energy efficiency. Enabling concurrent backscatter allows multiple tags to operate simultaneously, and their data can be recovered from collided signals. This capability is crucial for enhancing management efficiency in smart logistics and mitigating multi-tag collisions in Internet-of-Things (IoT) scenarios where multiple tags work collaboratively. However, existing concurrent backscatter schemes are vulnerable to noise and asynchronous signals, causing limited performance. To address these challenges, we introduce Ortho-CodeA, a backscatter scheme that enables reliable concurrent backscatter communication despite high noise levels and asynchronous signals. The underlying concept is to take advantage of coding mechanisms to combat noise and employ time-shifted orthogonal codes to mitigate the effects of asynchronous signals. Specifically, we design a set of time-shifted orthogonal codes that maintain code orthogonality despite asynchronous signals. Built upon the designed codes, we develop a multi-tag decoding scheme to recover data from each tag. We theoretically analyze the feasibility of our scheme and validate its performance through extensive experimental simulations. The results demonstrate that Ortho-CodeA achieves a BER of about 0.0036% in the case of 7 tags with an SNR of 10 dB and a maximum time delay of$1 \,\mu \text{s}$.

In on-demand delivery,online orders are delivered by couriers from merchants to customers within a short time (e.g., 45 minutes). An important task is to provide an efficient order dispatching solution. Existing studies focus on scenarios with stable routing behavior using pre-determined courier-order matching before delivery while ignoring real-time dynamics during delivery. In this work, we leverage courier-courier encounter events as an opportunity to enable cooperative order dispatching (i.e., conducting order transfers among couriers during delivery) for better delivery efficiency. However, it is non-trivial to conduct encounter-aware cooperative order dispatching in real-time dynamics due to two major challenges: (i) the dynamic nature of encounters in diverse real-world scenarios, and (ii) global delivery efficiency optimization by local order transfers. To address the above challenges, we design a detection-driven cooperative dispatching framework, called DECO. Specifically, we design (i) a Received Signal Strength Indicator (RSSI) variance-based state encoder to model encounter dynamics, (ii) an encounter event selector to choose encounter scenarios, (iii) a time-constrained order mask module to filter unsuitable orders, and (iv) an encounter-aware order transfer scheduler to make detailed order transfer decisions. Extensive experiments on real-world data from two large companies (i.e., JD Logistics, Eleme) show that DECO outperforms other baselines.Real-world deployment results at JD Logistics show that DECO improves the order overdue rate by 4.8%.

Trajectory data is a valuable asset for service management and spatio-temporal mining in transportation and logistics systems. However, due to equipment failure, network delay, and energy constraints, some trajectory point may be missed, which makes it difficult for trajectory-based management. Some researchers have focused on recovering sparse trajectories from road networks and historical trajectory data, but these methods are ineffective when the road network is incomplete. Recent research works have explored learning-based methods to recover trajectories in free space but lack user movement behavior modeling and efficient feature extraction on sparse long-range trajectories. Our work exploits the periodic behavior of couriers and fine-grained Area of Interest (AOI) data for sparse trajectory recovery in last-mile delivery. However, we face challenges with AOI access sequence deviations due to GPS inaccuracies and abnormal courier behaviors, as well as the complex, dynamic relationships within and between courier routes due to uncertain pick-up demands. To address these challenges, we design a graph-based multi-task learning framework, focusing on multi-scale attention fusion for end-to-end free space trajectory recovery. Our approach starts with a behavior-aware graph network that generates detailed spatial features. Following this, we propose a multi-scale attention fusion mechanism to extract intra- and inter-trajectory features. Finally, we design a multi-task learning module that predicts both coarse-grained spatial access sequences and fine-grained trajectory points. We evaluate the model with six-month data involved with more than 360,000 trajectory segments and more than 7.2 million waybills collected from one of the largest logistic companies in China. Extensive experiments on real-world datasets demonstrate that our method outperforms state-of-the-arts in multiple metrics.

LoRa has been widely adopted in Internet of Things (IoT) due to its long distance and low cost. With the largescale deployment of LoRa devices, it is not uncommon that multiple nodes transmit concurrently, leading to packet collisions and degraded performance. Previous studies have focused on examining single or multiple features in the received raw signals, named as chirp-level features, to separate collided packets for parallel decoding. However, the accuracy of feature extraction turns out to be vulnerable to interference, which can lead to incorrect packet decoding as network concurrency increases. Our study reveals that, other than the chirp-level features, a standard LoRa packet encoder introduces coding correlations across symbols of the same packet and thus leaves packet-level features to the symbols that can be utilized to disentangle symbols of collided packets in a new dimension. In this work, we introduce DeepDetangle, a Deep-learning-based feature fusion framework that efficiently fuses both packet-level and chirp-level features of symbols to enhance LoRa parallel decoding. DeepDetangle utilizes Complex-CNN and LSTM structures to capture multidimensional chirp-level features, their joint distributions, and temporal patterns. An MLP is employed to aggregate the chirplevel features with the packet-level features, thereby enabling the decoding of symbols on a per-block basis. By integrating features from both levels, erroneous symbols that cannot be recovered with chirp-level features alone can now be effectively corrected with other valid ones in the same block. Therefore, it demonstrates a strong capability to combat interference from other packets. Extensive evaluations have been conducted to assess the performance of DeepDetangle. The results indicate that DeepDetangle achieves$\mathbf{1 8. 1 \%}$to$\mathbf{8 0. 5 \%}$higher network throughput compared to existing works of similar complexity.

Millimeter-wave (mmWave) radar is emerging as a promising sensor for various human sensing tasks. Deep learning is frequently applied in radar-based applications, which typically require extensive data collection and labeling. In this demo, we present a low-cost hardware setup and a cross-platform software pipeline that automatically captures radar data of human activities, labels ground truth, and tests inference models in real time. The effectiveness of the testbed is demonstrated through real-time human pose estimation.

Tracking the human arm is essential for a variety of applications, including medical rehabilitation, sports analysis, and human-computer interaction. Current vision-based and wearable sensor-based approaches either struggle with occlusion, poor lighting conditions, and privacy concerns or result in intrusive user experiences. This paper introduces mmHAT, a novel 3D arm trajectory tracking method using mmWave radar. mmHAT proposes an end-to-end neural network design to address two major challenges: the lack of arm semantic information and dynamic variations in the mmWave point clouds. Firstly, mmHAT incorporates a multi-task joint learning framework, where the primary task is 3D arm tracking and the auxiliary task is gesture recognition. This aims to leverage the auxiliary task to guide the network in developing a deeper understanding of the user's arm movements. Secondly, for dynamic mmWave point clouds, mmHAT incorporates a new spatial-temporal feature encoder that aggregates the features of the arms point cloud from a global perspective. We collect ~320K frames of daily arm activity data for experimental validation. The results show that mmHAT achieves an average joint location error of 1.67 cm and angle estimation error of 4.23° for arm joints (i.e., elbow, wrist), while delivering excellent performance with only 2.67 ms latency.

To improve the efficiency of multi-agent systems in the Internet of Things, multi-agent reinforcement learning (MARL) has been extensively studied. Although transformer-based models now treat decision-making in MARL as a sequence problem and achieve advanced performance, the action of agents often depends on previous states without direction in information sharing. To address this issue, we propose a Bi-directional Sequence Inference framework for Multi-Agent Reinforcement Learning (BSI-MARL), consisting of three components: Action-Observation Processing Module, Sequence Inference Module, and Policy Optimization Module. The Action-Observation Processing Module defines the state space, action space, and reward function for the agents. Based on the Transformer model, BSI-MARL designs an encoder-decoder module for bi-directional sequence inference to generate action sequences for multi-agent decisions. Additionally, the policy gradient optimization module broadens the action sampling window, improving training efficiency. The experiments conducted on Mujoco-Half Cheetah and Google Research Football demonstrate that BSI-MARL exhibits excellent performance in multi-agent decision-making and good stability and generalization ability.

With the explosive proliferation of wireless devices, our lives are improved by various applications supported by heterogeneous wireless technologies, such as WiFi and ZigBee. However, the coexistence of WiFi and ZigBee also results in the degradation of the network performance, which cannot be avoided if the WiFi devices are even unaware of the ambient ZigBee devices. To better accommodate the heterogeneous wireless devices, this paper presents X-Disco, the first cross-technology neighbor discovery mechanism, for a WiFi device to detect ZigBee neighbors, without modification to hardware or firmware. With the help of the recently proposed cross-technology communication, X-Disco enables a commodity WiFi device to trigger responses, containing ZigBee neighbor information, from the ambient ZigBee coordinators (including routers). Through exploring the WiFi PHY-layer information accessible by WiFi driver, X-Disco decodes the responded ZigBee messages and obtains the ZigBee neighbor information. To improve X-Disco's reliability, we also propose ZigBee neighbor validation and interruption mitigation to exclude hidden node terminals and mitigate the interference caused by the ambient WiFi traffic respectively. The evaluation of X-Disco is performed on the commodity devices (TP-Link WDR 4300 WiFi router, TelosB motes) and USRP B210. The results demonstrate X-Disco successfully detects nine ZigBee neighbors within 70ms in the office.

Recent research on physical layer cross technology communication (PHY-CTC) brings a timely answer for escalated wireless coexistence and open spectrum movement. PHY-CTC achieves direct communication among heterogeneous wireless technologies (e.g.,WiFi, Bluetooth, and ZigBee) in physical layer and thus brings communication support for coexistence service such as spectrum management and IoT device control. To put PHY-CTC into service, however, there still exists a gap due to its transmission failure and asymmetric link (i.e., one-way PHY-CTC) issues. In this paper, we propose NetCTC – the first networking support design for PHY-CTC to establish feedbacks (e.g., ACKs) and thus meet the upper layer networking requirements in heterogeneous unicast, multicast and broadcast. The core design of NetCTC is a real-time interaction mechanism which achieves reliable, transmission efficient and concurrent interactive communication among heterogeneous devices. We implement and evaluate NetCTC on commodity devices and the USRP-N210 platform. Our extensive evaluation demonstrates that NetCTC achieves reliable bidirectional cross technology communication under a full range of wireless configurations including stationary, mobile and duty-cycled settings.

LoRa as a representative of Low-Power Wide Area Networks (LPWAN) technologies has emerged as an attractive communication platform for the Internet of Things. Since its dense deployment, signal collisions at base stations caused by concurrent transmissions degrade network performance. Existing approaches utilize the signal feature, e.g., frequency, to separate packets from collisions. They do not work well in burst traffic networks because the feature is not stable or fine-grained enough and the information for directed signal separation is not sufficient. In this paper, we leverage multidimensional information and propose a novel PHY layer approach called SCLoRa to decode collided LoRa transmissions. SCLoRa utilizes cumulative spectral coefficient, which integrates both frequency and power information, to separate symbols in the overlapped signal. The practical factors of channel fading, similar symbol boundary, and spectrum leakage are taken into account. The SCLoRa design requires neither hardware nor firmware changes in commodity devices – a feature allowing fast deployment on LoRa base stations. We implement and evaluate SCLoRa on USRP B210 base stations and commodity LoRa devices (i.e., SX1278). The experiment results in different scenarios with different radio parameters show that the throughput of SCLoRa is 3× than the state-of-the-art.

With the explosive increase in wireless devices, physical-layer signal analysis has become critically beneficial across distinctive domains including interference minimization in network planning, security and privacy (e.g., drone and spycam detection), and mobile health with remote sensing. While SDR is known to be highly effective in realizing such services, they are rarely deployed or used by the end-users due to the costly hardware ~1K USD (e.g., USRP). Low-cost SDRs (e.g., RTL-SDR) are available, but their bandwidth is limited to 2-3 MHz and operation range falls well below 2.4 GHz - the unlicensed band holding majority of the wireless devices. This paper presents SDR-Lite, the first zero-cost, software-only software defined radio (SDR) receiver that empowers commodity WiFi to retrieve the In-phase and Quadrature of an ambient signal. With the full compatibility to pervasively-deployed WiFi infrastructure (without any change to the hardware and firmware), SDR-Lite aims to spread the blessing of SDR receiver functionalities to billions of WiFi users and households to enhance our everyday lives. The key idea of SDR-Lite is to trick WiFi to begin packet reception (i.e., the decoding process) when the packet is absent, so that it accepts ambient signals in the air and outputs corresponding bits. The bits are then reconstructed to the original physical-layer waveform, on which diverse SDR applications are performed. Our comprehensive evaluation shows that the reconstructed signal closely reassembles the original ambient signal (>85% correlation). We extensively demonstrate SDR-Lite effectiveness across seven distinctive SDR receiver applications under three representative categories: (i) RF fingerprinting, (ii) spectrum monitoring, and (iii) (ZigBee) decoding. For instance, in security applications of drone and rogue WiFi AP detection, SDR-Lite achieves 99% and 97% accuracy, which is comparable to USRP.

Multi-user MIMO (MU-MIMO) is a widely-known, fundamental technique to significantly improve the spectrum efficiency. While there is a great demand for spectrum efficiency and massive scalability under explosively increasing IoT, hardware limitations make it particularly challenging for the mechanism to be transferred to the IoT (e.g., ZigBee) domain. This paper presents X-MIMO, a zero-cost, software-only cross-technology MU-MIMO for commodity ZigBee. As the first work to shed the light on the feasibility of MU-MIMO on commodity IoT, X-MIMO leverages on cross-technology communication (CTC) to turn the pervasively-deployed WiFi AP into MU-MIMO transmitter, delivering different packets to multiple ZigBees in parallel. X-MIMO uniquely exploits WiFi CSI to extract the accurate physical layer signal of the ZigBee packet and the WiFi-ZigBee channel coefficient. Rigorous derivation shows that X-MIMO's precoding is inherently immune to the uncertainties of the commodity devices, making X-MIMO highly reliable in practice. Lastly, spectrum-efficient emulation is proposed to maximize the spectrum reuse. We implement and comprehensively evaluate the performance of X-MIMO on commodity devices (Atheros AR9334 WiFi NIC and TelosB CC2420) as well as on USRP B210 for in-depth analysis. Results reveal that X-MIMO achieves 495 Kbps with <1% symbol error rate (SER) and 704.24 Kbps with 6.1% SER for two and three streams, respectively. Near-linear increase of the throughput effectively demonstrates the feasibility of X-MIMO.

To mitigate the issue of cross-technology interference (CTI) under dense wireless, cross-technology communication (CTC) was recently proposed, which enables direct communication among heterogeneous wireless technologies. We present SymBee, a novel ZigBee to WiFi CTC with symbol-level encoding for performance breakthrough from packet-level state-of-the-arts. SymBee is uniquely built on the new insight on ZigBee-WiFi physical layer cross-observability - i.e., the output on WiFi when fed with ZigBee signal (due to frequency overlap). This is analyzed experimentally and theoretically through rigorous derivations, from which the key innovation in SymBee design, i.e., payload encoding, stems; Conveying data across technologies is as simple as putting specific symbols in ZigBee packet payload, such that they yield unique and easily detectable patterns when cross-observed at WiFi. This symbol-level encoding is fully compatible with any commodity ZigBee device. Decoding at WiFi is a light-weight function that recycles the output from idle listening, thereby minimizing the computation while keeping compatibility to WiFi standard. SymBee is extensively evaluated both theoretically and experimentally through testbed evaluations on six distinct locations including outdoor. The result demonstrate that SymBee reaches the throughput of up to 31.25kbps, 145.4× faster than the state-of-the-art.

Cross-technology interference (CTI) from dense and prevalent wireless has become a primary threat to low-power IoT. This paper presents G-Bee, a CTI avoidance technique that uniquely places ZigBee packet on the guard band of ongoing WiFi traffic, which effectively safeguards the packet from WiFi interference. Such design ensures reliable ZigBee communication even under saturated WiFi traffic where traditional ZigBee is considered inoperable. Technical highlight is in lighweight WiFi guard band capture mechanism using ZigBee PHY layer samples directly accessible in various commercial ZigBee chip. Another exclusive feature of G-Bee is spectrum-synchronized low duty cycling - by utilizing guard bands of periodic WiFi beacons, active slots are effectively synchronized to spectrum availability (i.e., guard band) for significant delay improvement. Extensive evaluations on our prototype system demonstrates G-Bee PRR over 95% where legacy ZigBee drops to below 15% under significant interference with hundreds WiFi users and reduction of low duty cycle delay by 87.5%, all of which are achieved with a light computational overhead of 0.3%.