Recently, Prof. Zhang Daqing's research group at Peking University's School of Computer Science made a series of significant advancements in the field of intelligent wireless sensing based on WiFi and radar. They explored fundamental theoretical issues, including the Fresnel zone sensing model, mapping relationships between Doppler velocity and speed, speed and direction to position orientation, sensing boundaries in wireless perception, perception limits, perception signal quality, and perception models for mobile scenarios. Six extended research papers were published in top-tier academic journals in the field of computer network systems, such as the IEEE Journal on Selected Areas in Communications (JSAC), and in the field of ubiquitous computing at the prestigious international conference ACM UbiComp 2022. Additionally, one extended research paper was published at ACM MobiCom 2022, a leading international conference in the field of mobile computing, and received the Best Paper Award Runner-up.
Ubiquitous wireless sensing refers to the technology that enables contactless sensing of people and objects in the environment using omnipresent wireless radio frequency signals. These wireless signals originate from various wireless communication devices deployed in our surroundings, including Wi-Fi, 4G/5G, RFID, and radar, among others. By utilizing these ubiquitous wireless signals, analyzing the fluctuation patterns of wireless signals affected by human activities and extracting signal features, it is possible to identify human movement behaviors, monitor respiration and heart rate, sleep conditions, as well as implement applications like intrusion detection, indoor positioning, and human trajectory tracking.
One of the most important signal features in wireless sensing, Doppler velocity, can be extracted from WiFi signals received at the receiver end, thereby supporting applications like trajectory tracking and gesture recognition. However, there are two important theoretical questions that remain unanswered when estimating the motion velocity of a target using WiFi signals: (1) Is the precision of target velocity estimation the same at any position within the WiFi sensing range? (2) Since the placement of WiFi devices in reality is arbitrary, how can accurate target velocity estimation be achieved? The paper titled 'Rethinking Doppler Effect for Accurate Velocity Estimation With Commodity WiFi Devices' addresses these questions. Professor Zhang Daqing's team started with the basic concept of the relativistic Doppler effect and, based on their proposed Fresnel zone perception model, established the relationship between Doppler frequency shift and target motion velocity in a WiFi sensing system. They quantitatively analyzed how the accuracy of velocity estimation (including speed and direction) in WiFi non-contact sensing is influenced by factors such as the target's position relative to WiFi transceivers and the target's motion direction. They provided guiding principles for achieving accurate velocity estimation. Finally, the paper, with applications in gesture recognition and human trajectory tracking, provides examples of how to extract velocity features that are not affected by position and motion direction. This research achievement was published in the CCF Class A international journal IEEE JSAC 2022, with Dr. Kai Niu, a member of the team, as the first author of the paper.
Figure 1: The physical interpretation of the relationship between Doppler frequency shift and the target's motion velocity.
In ubiquitous wireless sensing systems, the accuracy of sensing distance, range, and signal parameter estimation is often influenced by signal noise. Professor Zhang Daqing's team achieved significant progress in both the theoretical foundations and applications of ubiquitous wireless sensing by introducing a model for sensing signal-to-noise ratio (SNR) into the field. Previous WiFi sensing work lacked research and exploration of the sensing range, leading to a lack of theoretical guidance and a reliance on trial and error for system performance in real-world deployments. The paper, 'Placement Matters: Understanding the Effects of Device Placement for WiFi Sensing,' introduced and studied the concept of 'sensing SNR.' It was the first internationally to provide a formula for the boundary of the sensing range in ubiquitous wireless sensing systems, revealing the shape of the sensing range boundary changes with the distance between the transceivers. As the distance between WiFi devices increases, the sensing range changes from a small ellipse into a larger one, then it forms an inward dent, and finally becomes two smaller ellipses surrounding the transceivers (see Figure 2). When WiFi sensing systems are deployed in real environments, merely placing the transceivers according to the theoretical guidance solves two common WiFi sensing problems: a limited sensing range and interference from distant targets. For example, by appropriately placing transmitters and receivers, the sensing range of a human tracking system can be increased by 200%. By increasing the distance between transmitters and receivers, even when an interfering object is 0.5 meters away from the sensing target, the system can accurately detect fine-grained respiration of the target. The related work was published in the CCF A-ranked international conference UbiComp 2022 (ACM IMWUT), with the first author of the paper being doctoral student Xuanzhi Wang.
Figure 2: The boundaries of the sensing coverage area at different LoS lengths, with the red and black circles representing the positions of the Tx and Rx.
In scenarios like gesture recognition, the signal quality can vary due to different factors such as device placement, gesture position, and motion direction. Treating the signal quality corresponding to user gestures as constant and uniformly preprocessing it can lead to unstable accuracy in gesture recognition. In reality, when the hand is in different positions and orientations, sometimes the signal change is much greater than the noise, and sometimes the signal change is close to the noise. The corresponding perception signal-to-noise ratio and signal perception quality change accordingly. To address the issue of inconsistent signal perception quality for hand movements in different positions and orientations, the paper 'Towards Robust Gesture Recognition by Characterizing the Sensing Quality of WiFi Signals' proposes a signal perception quality measurement model. This model characterizes the relative relationship between the effective perception signal and noise signal during hand movements, allowing real-time quantification of WiFi-CSI signal quality at different times and positions. Based on the signal perception quality derived from wireless signals, the paper further introduces a signal preprocessing framework based on the perception quality indicator. Specifically, this framework classifies signals of different perception quality for processing. For signals with good perception quality that can be used for recognition tasks, the framework utilizes multi-carrier information to further enhance signal quality. For signals with poor perception quality that cannot be used for recognition tasks, the framework infers hand movement information at that time based on prior knowledge. This ultimately enables accurate extraction of Doppler velocity information for hand movements in different positions and orientations, significantly improving the performance and robustness of the gesture recognition system. This work was published in the CCF Class A international conference UbiComp 2022 (ACM IMWUT), with the first author being doctoral student Ruiyang Gao.
Figure 3: Different hand movements have varying signal qualities, and by measuring the signal quality, a stable gesture recognition system can be constructed.
Based on the signal-to-noise ratio model, Professor Zhang Daqing's team has also made significant progress in improving the wireless sensing range. In traditional wireless sensing systems, weak reflected signals are easily drowned out by noise when the target is far from the sensing device, rendering the sensing system inoperable. Addressing this common challenge, the paper titled "DiverSense: Maximizing Wi-Fi Sensing Range Leveraging Signal Diversity" introduces a time-frequency-space domain-based signal-to-noise ratio enhancement technique, greatly extending the wireless signal sensing range and expanding the WiFi sensing coverage. In a WiFi sensing system, the channel state information (CSI) collected by wireless devices has a deterministic relationship with the motion of the target and its position relative to the wireless device, while random noise roughly follows a Gaussian distribution. According to the law of large numbers, by sampling and superimposing signals with independent and identically distributed noise from multiple moments, frequencies, and spatial domains, the signal will converge to its expected value. Based on this principle, taking WiFi signals collected in the time (multiple moments), frequency (multiple carriers), and space (multiple antennas) domains as multiple samples and aligning and fusing these signals can significantly improve the signal-to-noise ratio, allowing for the reconstruction of real human movement information. To demonstrate this, the team implemented the DiverSense system, focusing on respiration detection applications. Using commercial WiFi equipment, they achieved a new record of extending human respiration detection distance from the current farthest 11 meters to 40 meters in a corridor environment. This long-distance respiration detection work based on WiFi was published at the CCF A-class international conference UbiComp 2022, with the paper's first author being doctoral student Yang Li.
Figure 4: The corridor's 40-meter respiration sensing experimental scenario and the enhanced signal-to-noise ratio CSI signals, as well as the extracted respiration waveforms.
An existing critical missing feature in wireless non-contact sensing is the inability to achieve sensing functionality in mobile device scenarios. The challenge lies in the fact that the signal variations caused by irregular device movement overlap with the signal changes induced by the movement of the sensed targets, making it difficult to separate them. The paper "Mobi2Sense: Empowering Wireless Sensing with Mobility" introduces a new model and technology for non-contact wireless sensing in mobile scenarios. This scenario has arisen due to the miniaturization and cost reduction of radar chips in recent years, as well as their integration into various mobile devices such as handheld devices and mobile robots, significantly increasing the demand for wireless sensing in mobile scenarios. Our proposed Mobi2Sense mobile sensing system cleverly uses the reflection signals from static objects in the environment to characterize device movement patterns. It selects the reflection signals from static reference objects as reference signals, and then divides the received signal by the reference signal to eliminate the device's motion component from the received signal, accurately restoring the original motion signal of the target. Extensive experiments demonstrate that Mobi2Sense can capture subtle movements of sensed targets in mobile scenarios, "hearing" the vibrations of a speaker to reproduce sound, "observing" the respiratory status of the human body, and "recognizing" multi-target gestures with high precision. This system can be used for scenarios such as doctors using handheld radar devices to sense patients' vital signs in emergency situations, and mobile robots tracking and monitoring the health status of elderly people in their homes, significantly expanding the applications of wireless sensing. This research achievement was published at the top conference in mobile computing systems, ACM Mobicom 2022, and received the Best Paper Award Runner-up. This work was a collaboration between Peking University, the Institute of Software of the Chinese Academy of Sciences, and the University of Massachusetts in the United States, and the first author, Fusang Zhang, is an associate researcher at the Institute of Software of the Chinese Academy of Sciences (formerly a postdoctoral researcher).
Figure 5: The Mobi2Sense mobile sensing system can be used with handheld devices or installed on mobile robot platforms to accurately sense target activities.
Brief introduction of the corresponding author Professor Zhang Daqing:
Zhang Daqing, Chair Professor at Peking University's School of Computer Science, a Fellow of the European Academy of Sciences, an IEEE Fellow, and the Chair of the CCF Ubiquitous Computing Committee. Since joining Peking University in 2014, he has been dedicated to the research of ubiquitous wireless sensing theory and technology. His team has made significant contributions to this field on an international level. They were the first to propose the wireless sensing theory based on the Fresnel zone, revealing the relationship between the activity location, orientation, size of sensed targets, the position of WiFi transceivers, and the received wireless signals. Additionally, they introduced concepts such as the CSI business model, speed-Doppler velocity model, and various essential properties of indoor wireless signal propagation. By introducing the concept of signal-to-noise ratio, they answered fundamental theoretical questions in the field of WiFi sensing, including sensing limits, sensing boundaries, and signal quality. This work has provided a novel theoretical foundation for wireless sensing based on WiFi, 4G/5G signals, and radar signals. Furthermore, their research has achieved international recognition, particularly in applications like non-contact respiration monitoring, indoor positioning, activity tracking, and intrusion detection, consistently delivering top-tier performance.
Since 2016, Professor Zhang Daqing's team has published over 30 papers at the top international academic conference in the field of ubiquitous computing, ACM UbiComp. The cumulative number of papers published by the team has consistently ranked first on the international stage. Notably, during the period from 2016 to 2019, four of their long papers in the field of wireless sensing are among the top three most cited papers at the respective UbiComp conferences. In the last two years, their contributions were further recognized with the Distinguished Paper Award at ACM UbiComp 2021 and the Best Paper Award Runner-up at ACM MobiCom 2022. In 2022, Professor Zhang Daqing's team, in collaboration with companies including Huawei, led the way in China by successfully implementing WiFi sensing for commercial use. These achievements underscore Peking University's position at the forefront of research in the fields of ubiquitous computing and wireless sensing.
