RF-ECG - Heart Rate Variability Assessment based on COTS RFID Tag Array

RF-ECG:Heart Rate Variability Assessment Based on COTS RFID Tag Array CHUYU WANG,Nanjing University,CHN LEI XIE*,Nanjing University,CHN WEI WANG,Nanjing University,CHN YINGYING CHEN,Rutgers University,USA YANLING BU,Nanjing University,CHN SANGLU LU,Nanjing University,CHN As an important indicator of autonomic regulation for circulatory function,Heart Rate Variability(HRV)is widely used for general health evaluation.Apart from using dedicated devices(e.g.ECG)in a wired manner,current methods search for a ubiquitous manner by either using wearable devices,which suffer from low accuracy and limited battery life,or applying wireless techniques(e.g..FMCW),which usually utilize dedicated devices(e.g.,USRP)for the measurement.To address these issues,we present RF-ECG based on Commercial-Off-The-Shelf(COTS)RFID,a wireless approach to sense the human heartbeat through an RFID tag array attached on the chest area in the clothes.In particular,as the RFID reader continuously interrogates the tag array,two main effects are captured by the tag array:the reflection effect representing the RF-signal reflected from the heart movement due to heartbeat;the moving effect representing the tag movement caused by chest movement due to respiration.To extract the reflection signal from the noisy RF-signals,we develop a mechanism to capture the RF-signal variation of the tag array caused by the moving effect,aiming to eliminate the signals related to respiration. To estimate the HRV from the reflection signal,we propose a signal reflection model to depict the relationship between the 85 RF-signal variation from the tag array and the reflection effect associated with the heartbeat.A fusing technique is developed to combine multiple reflection signals from the tag array for accurate estimation of HRV.Experiments with 15 volunteers show that RF-ECG can achieve a median error of 3%of Inter-Beat Interval(IBI),which is comparable to existing wired techniques. CCS Concepts:.Networks-Sensor networks;Mobile networks;.Human-centered computing-Mobile devices; Additional Key Words and Phrases:RFID:Heart rate:HRV ACM Reference Format: Chuyu Wang,Lei Xie,Wei Wang.Yingying Chen,Yanling Bu,and Sanglu Lu.2018.RF-ECG:Heart Rate Variability Assessment Based on COTS RFID Tag Array.Proc.ACM Interact.Mob.Wearable Ubiquitous Technol.2,2,Article 85 (June 2018),26 pages. https:/doiorg/10.1145/3214288 "Lei Xie is the corresponding author,Email:Ixie@nju.edu.cn. Authors'addresses:Chuyu Wang,wangcyu217@dislab.nju.edu.cn,Nanjing University,State Key Laboratory for Novel Software Technology, 163 Xianlin Ave,Nanjing,210046,CHN;Lei Xie,Ixie@nju.edu.cn,Nanjing University,State Key Laboratory for Novel Software Technology, 163 Xianlin Ave,Nanjing,210046,CHN;Wei Wang,Nanjing University,State Key Laboratory for Novel Software Technology,163 Xianlin Ave,Nanjing.210046,CHN:Yingying Chen,Rutgers University,WINLAB,North Brunswick,NJ,08902,USA;Yanling Bu,Nanjing University. State Key Laboratory for Novel Software Technology,163 Xianlin Ave,Nanjing,210046,CHN;Sanglu Lu,Nanjing University,State Key Laboratory for Novel Software Technology,163 Xianlin Ave,Nanjing.210046.CHN. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page.Copyrights for components of this work owned by others than ACM must be honored.Abstracting with credit is permitted.To copy otherwise,or republish,to post on servers or to redistribute to lists,requires prior specific permission and/or a fee.Request permissions from permissions@acm.org. 2018 Association for Computing Machinery. 2474-9567/2018/6-ART85$15.00 https:/doi.org/10.1145/3214288 Proc.ACM Interact.Mob.Wearable Ubiquitous Technol,Vol.2,No.2,Article 85.Publication date:June 2018

85 RF-ECG: Heart Rate Variability Assessment Based on COTS RFID Tag Array CHUYU WANG, Nanjing University, CHN LEI XIE∗ , Nanjing University, CHN WEI WANG, Nanjing University, CHN YINGYING CHEN, Rutgers University, USA YANLING BU, Nanjing University, CHN SANGLU LU, Nanjing University, CHN As an important indicator of autonomic regulation for circulatory function, Heart Rate Variability (HRV) is widely used for general health evaluation. Apart from using dedicated devices (e.g, ECG) in a wired manner, current methods search for a ubiquitous manner by either using wearable devices, which suffer from low accuracy and limited battery life, or applying wireless techniques (e.g., FMCW), which usually utilize dedicated devices (e.g., USRP) for the measurement. To address these issues, we present RF-ECG based on Commercial-Off-The-Shelf (COTS) RFID, a wireless approach to sense the human heartbeat through an RFID tag array attached on the chest area in the clothes. In particular, as the RFID reader continuously interrogates the tag array, two main effects are captured by the tag array: the reflection effect representing the RF-signal reflected from the heart movement due to heartbeat; the moving effect representing the tag movement caused by chest movement due to respiration. To extract the reflection signal from the noisy RF-signals, we develop a mechanism to capture the RF-signal variation of the tag array caused by the moving effect, aiming to eliminate the signals related to respiration. To estimate the HRV from the reflection signal, we propose a signal reflection model to depict the relationship between the RF-signal variation from the tag array and the reflection effect associated with the heartbeat. A fusing technique is developed to combine multiple reflection signals from the tag array for accurate estimation of HRV. Experiments with 15 volunteers show that RF-ECG can achieve a median error of 3% of Inter-Beat Interval (IBI), which is comparable to existing wired techniques. CCS Concepts: • Networks → Sensor networks; Mobile networks; • Human-centered computing → Mobile devices; Additional Key Words and Phrases: RFID; Heart rate; HRV ACM Reference Format: Chuyu Wang, Lei Xie, Wei Wang, Yingying Chen, Yanling Bu, and Sanglu Lu. 2018. RF-ECG: Heart Rate Variability Assessment Based on COTS RFID Tag Array. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 2, 2, Article 85 (June 2018), 26 pages. https://doi.org/10.1145/3214288 ∗Lei Xie is the corresponding author, Email: lxie@nju.edu.cn. Authors’ addresses: Chuyu Wang, wangcyu217@dislab.nju.edu.cn, Nanjing University, State Key Laboratory for Novel Software Technology, 163 Xianlin Ave, Nanjing, 210046, CHN; Lei Xie, lxie@nju.edu.cn, Nanjing University, State Key Laboratory for Novel Software Technology, 163 Xianlin Ave, Nanjing, 210046, CHN; Wei Wang, Nanjing University, State Key Laboratory for Novel Software Technology, 163 Xianlin Ave, Nanjing, 210046, CHN; Yingying Chen, Rutgers University, WINLAB, North Brunswick, NJ, 08902, USA; Yanling Bu, Nanjing University, State Key Laboratory for Novel Software Technology, 163 Xianlin Ave, Nanjing, 210046, CHN; Sanglu Lu, Nanjing University, State Key Laboratory for Novel Software Technology, 163 Xianlin Ave, Nanjing, 210046, CHN. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from permissions@acm.org. © 2018 Association for Computing Machinery. 2474-9567/2018/6-ART85 $15.00 https://doi.org/10.1145/3214288 Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

85:2·C.Wang et al. 1 INTRODUCTION 1.1 Motivation Heart Rate Variability(HRV)represents the variation of the time interval between adjacent heartbeats [7].As an important indicator of autonomic regulation for circulatory function,HRV reflects how the cardiovascular regulatory system responds to demands,stress and illness [35].Fine-grained HRV information is usually important to quantitatively measure physiological and mental changes during treatment.For example,the frequent detection of the premature beats indicates some kinds of heart problems about the person.Currently,reduced HRV has been used as a marker of aging,decreased autonomic activity,diabetic neuropathy,and increased risk of sudden cardiac death.According to the news from WHO [2],more than 17 million people die annually from cardiovascular disease(CVD),an estimated 31%of all deaths worldwide.Efficient HRV assessment can not only provide accurate information about the heart states for the treatment of heart diseases and the adjustment of mental status,but also timely detect the heart problem in the early stage.Therefore,accurate and fine-grained HRV assessment is regarded as one of the most effective methods for the general health evaluation [29]. 1.2 Limitation of Prior Art HRV analysis is normally based on the Inter-Beat Interval(IBI)measurements,which accurately describes the time interval between the adjacent heartbeats.As a traditional approach,the electrocardiogram(ECG)is regarded as the standard way to measure the IBI [6],which can accurately measure the IBI information from the tiny electrical changes on the skin.But the user is usually tethered to the electrodes in a wired manner,which imposes restrictions on the range of daily activities of the user.Moreover,it requires direct skin contact,indicating some people need to remove the chest hair to achieve better signal quality.Some optical absorption methods leverages the photoplethysmograph(PPG)[12,18]to estimate the IBI information.But it usually relies on the reflection of infrared light,leading to the loss of waveform detail and time accuracy.Commercial devices(e.g.,wristbands[5]) try to integrate the sensors(electrodes or visual sensors)into fabrics for wearable applications.However,they are only designed to estimate the average heart rate within a duration,which works as a training aid for exercises. So the fine-grained heart problems,such as premature beats and heart arrhythmia,cannot be effectively detected, which requires the accurate estimation of the IBI information.Moreover,these wearable devices are mostly constrained by their limited battery life,requiring frequent battery recharging.This limits the efficient HRV monitoring for elderly people and infants,who may forget to recharge the wearable devices.To eliminate the dependency on such sensing devices,a number of device-free methods are proposed to perceive the heartbeat via wireless channel(e.g.,Wi-Fi and FMCW)[10,11,42].However,they usually depend on the dedicated devices Sleep monitor SANT Heartbeat Elder monitor RFID tag arr变 Tag array Antenna (a)Illustrations of applications of RF-ECG (b)Illustration of working flow of RF-ECG Fig.1.RF-ECG system scenario. Proc.ACM Interact.Mob.Wearable Ubiquitous Technol..Vol 2.No.2.Article 85.Publication date:lune 2018

85:2 • C. Wang et al. 1 INTRODUCTION 1.1 Motivation Heart Rate Variability (HRV) represents the variation of the time interval between adjacent heartbeats [7]. As an important indicator of autonomic regulation for circulatory function, HRV reflects how the cardiovascular regulatory system responds to demands, stress and illness [35]. Fine-grained HRV information is usually important to quantitatively measure physiological and mental changes during treatment. For example, the frequent detection of the premature beats indicates some kinds of heart problems about the person. Currently, reduced HRV has been used as a marker of aging, decreased autonomic activity, diabetic neuropathy, and increased risk of sudden cardiac death. According to the news from WHO [2], more than 17 million people die annually from cardiovascular disease (CVD), an estimated 31% of all deaths worldwide. Efficient HRV assessment can not only provide accurate information about the heart states for the treatment of heart diseases and the adjustment of mental status, but also timely detect the heart problem in the early stage. Therefore, accurate and fine-grained HRV assessment is regarded as one of the most effective methods for the general health evaluation [29]. 1.2 Limitation of Prior Art HRV analysis is normally based on the Inter-Beat Interval (IBI) measurements, which accurately describes the time interval between the adjacent heartbeats. As a traditional approach, the electrocardiogram (ECG) is regarded as the standard way to measure the IBI [6], which can accurately measure the IBI information from the tiny electrical changes on the skin. But the user is usually tethered to the electrodes in a wired manner, which imposes restrictions on the range of daily activities of the user. Moreover, it requires direct skin contact, indicating some people need to remove the chest hair to achieve better signal quality. Some optical absorption methods leverages the photoplethysmograph (PPG) [12, 18] to estimate the IBI information. But it usually relies on the reflection of infrared light, leading to the loss of waveform detail and time accuracy. Commercial devices (e.g., wristbands [5]) try to integrate the sensors (electrodes or visual sensors) into fabrics for wearable applications. However, they are only designed to estimate the average heart rate within a duration, which works as a training aid for exercises. So the fine-grained heart problems, such as premature beats and heart arrhythmia, cannot be effectively detected, which requires the accurate estimation of the IBI information. Moreover, these wearable devices are mostly constrained by their limited battery life, requiring frequent battery recharging. This limits the efficient HRV monitoring for elderly people and infants, who may forget to recharge the wearable devices. To eliminate the dependency on such sensing devices, a number of device-free methods are proposed to perceive the heartbeat via wireless channel (e.g., Wi-Fi and FMCW) [10, 11, 42]. However, they usually depend on the dedicated devices RFID tag array


Elder monitor Sleep monitor (a) Illustrations of applications of RF-ECG Tag array Antenna Heartbeat Reflection effect Moving effect (b) Illustration of working flow of RF-ECG Fig. 1. RF-ECG system scenario. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018.

RF-ECG:Heart Rate Variability Assessment Based on COTS RFID Tag Array.85:3 (e.g.,USRP [10]and RFID with conductive fabric electrodes [11]),and thus are expensive compared with the commercial devices.Additionally,some professional skills are required to operate the dedicated devices,which is not suitable for applications in a ubiquitous environment.Moreover,when multiple people are within the sensing range,the device-free property prevents them from effectively distinguishing multiple human subjects and performing HRV monitoring simultaneously.Therefore,a new solution based on Commercial-off-the-Shelf (COTS)devices is desirable to the above limitations in a lightweight and distinguishable approach. 1.3 Our Approach The emerging RFID technology [15,25,27,34,37,38,40]has brought new opportunities for HRV monitoring in a more convenient and accurate approach,as the RFID tag can be regarded as an extremely lightweight sensor and its nature of identification can be used to effectively and easily distinguish different human subjects.In this paper,we propose RF-ECG,an RFID-based approach to perform HRV monitoring on human subjects,which aims to recover the IBI information with a similar accuracy as the traditional ECG based on the COTS RFID devices. Specifically,instead of designing a dedicated platform [11],we attach a set of COTS RFID tags on the chest area in the clothes of the human subject,which forms a tag array for comprehensive sensing.We deploy a COTS RFID reader to continuously interrogate these tags by issuing a continuous wave,and collect their backscattered RF-signals within the effective scanning range(1 ~3m).Similarly to the traditional ECG-based approaches,we focus on the HRV monitoring with a relatively quite environment,where the users are supposed to keep still during the measurement.Figure 1(a)illustrates the deployment of tag array of RF-ECG,where 6 tags form a 2 x 3 tag array.Through the RF-signal from the tag array,RF-ECG is able to facilitate two main applications for HRV monitoring in daily life.Firstly,by deploying the antenna above the bed,RF-ECG is able to monitor the user's heart status during sleeping.which can further enable the treatment of sleep apnea [1]and sleep stage detection [21]. Such mode can be extended to infant monitoring,because infants spend most of the time sleeping in the cribs. Secondly,since elderly people are usually less active and spend lots of time in performing more stationary activities(e.g,watching TVs),we can deploy the antenna in the living room and bedroom for efficient HRV monitoring.Such HRV monitoring can offer rapid and effective diagnostic clues for the general health evaluation. Toward these monitoring applications,Figure 1(b)further presents the diagram of our HRV monitoring approach. As the heart beats behind the human chest,the continuous wave reflected from the heart movement could be captured by the tag array,and then backscattered to the RFID reader via the RF-signals [41],which we call the reflection effect of the tag array.Meanwhile,the RF-signals are also affected by the chest movements due to human respiration,which we call the moving effect of the tag array.Hence,the RF-signals received by the RFID reader consist of heartbeat signal,respiration signal and the ambient noise from the environment.In this paper,we investigate the possibility of extracting such tiny reflection signals corresponding to the heartbeat, while eliminating other signal interferences from the human respiration and the ambient noise in the multi-path environment.In particular,according to the RF-signals received from the tag array,we develop a mechanism to capture the chest movement,aiming to cancel the respiration signal corresponding to the moving effect.Then,by using the wavelet-based signal denoising,we further extract the heartbeat signals corresponding to the reflection effect.Finally,we build a model to depict the reflection effect on RF-signals of different tags from the tag array, and extract the Inter-Beat Interval(IBI)for HRV monitoring. 1.4 Challenges There are three main challenges in performing the HRV assessment via the RFID based approach.The first challenge is to detect and extract weak heartbeat signals from RFID tags among multiple interferences caused by human respiration and ambient noises.In particular,the signal variation captured by the reflection effect from heartbeat is much weaker than the signal variation caused by the moving effect from respiration.To address Proc.ACM Interact.Mob.Wearable Ubiquitous Technol.Vol.2.No.2.Article 85.Publication date:June 2018

RF-ECG: Heart Rate Variability Assessment Based on COTS RFID Tag Array • 85:3 (e.g., USRP [10] and RFID with conductive fabric electrodes [11]), and thus are expensive compared with the commercial devices. Additionally, some professional skills are required to operate the dedicated devices, which is not suitable for applications in a ubiquitous environment. Moreover, when multiple people are within the sensing range, the device-free property prevents them from effectively distinguishing multiple human subjects and performing HRV monitoring simultaneously. Therefore, a new solution based on Commercial-off-the-Shelf (COTS) devices is desirable to the above limitations in a lightweight and distinguishable approach. 1.3 Our Approach The emerging RFID technology [15, 25, 27, 34, 37, 38, 40] has brought new opportunities for HRV monitoring in a more convenient and accurate approach, as the RFID tag can be regarded as an extremely lightweight sensor and its nature of identification can be used to effectively and easily distinguish different human subjects. In this paper, we propose RF-ECG, an RFID-based approach to perform HRV monitoring on human subjects, which aims to recover the IBI information with a similar accuracy as the traditional ECG based on the COTS RFID devices. Specifically, instead of designing a dedicated platform [11], we attach a set of COTS RFID tags on the chest area in the clothes of the human subject, which forms a tag array for comprehensive sensing. We deploy a COTS RFID reader to continuously interrogate these tags by issuing a continuous wave, and collect their backscattered RF-signals within the effective scanning range (1 ∼ 3m). Similarly to the traditional ECG-based approaches, we focus on the HRV monitoring with a relatively quite environment, where the users are supposed to keep still during the measurement. Figure 1(a) illustrates the deployment of tag array of RF-ECG, where 6 tags form a 2 × 3 tag array. Through the RF-signal from the tag array, RF-ECG is able to facilitate two main applications for HRV monitoring in daily life. Firstly, by deploying the antenna above the bed, RF-ECG is able to monitor the user’s heart status during sleeping, which can further enable the treatment of sleep apnea [1] and sleep stage detection [21]. Such mode can be extended to infant monitoring, because infants spend most of the time sleeping in the cribs. Secondly, since elderly people are usually less active and spend lots of time in performing more stationary activities (e.g., watching TVs), we can deploy the antenna in the living room and bedroom for efficient HRV monitoring. Such HRV monitoring can offer rapid and effective diagnostic clues for the general health evaluation. Toward these monitoring applications, Figure 1(b) further presents the diagram of our HRV monitoring approach. As the heart beats behind the human chest, the continuous wave reflected from the heart movement could be captured by the tag array, and then backscattered to the RFID reader via the RF-signals [41], which we call the reflection effect of the tag array. Meanwhile, the RF-signals are also affected by the chest movements due to human respiration, which we call the moving effect of the tag array. Hence, the RF-signals received by the RFID reader consist of heartbeat signal, respiration signal and the ambient noise from the environment. In this paper, we investigate the possibility of extracting such tiny reflection signals corresponding to the heartbeat, while eliminating other signal interferences from the human respiration and the ambient noise in the multi-path environment. In particular, according to the RF-signals received from the tag array, we develop a mechanism to capture the chest movement, aiming to cancel the respiration signal corresponding to the moving effect. Then, by using the wavelet-based signal denoising, we further extract the heartbeat signals corresponding to the reflection effect. Finally, we build a model to depict the reflection effect on RF-signals of different tags from the tag array, and extract the Inter-Beat Interval (IBI) for HRV monitoring. 1.4 Challenges There are three main challenges in performing the HRV assessment via the RFID based approach. The first challenge is to detect and extract weak heartbeat signals from RFID tags among multiple interferences caused by human respiration and ambient noises. In particular, the signal variation captured by the reflection effect from heartbeat is much weaker than the signal variation caused by the moving effect from respiration. To address Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

85:4·C.Wang et al. this challenge,we develop a mechanism to depict the RF-signal variation of the tag array caused by the chest movement,helping to eliminate the part of the respiration signals corresponding to the moving effect.We further build a model to depict the reflection effect on RF-signals of different tags from the tag array.By fusing the RF-signals from multiple tags,we are able to strengthen the heartbeat signals while suppressing the signal interferences from the others. The second challenge is to achieve a fine-grained heart beat estimation for the HRV assessment according to the reflection effect.Instead of simply calculating the heart rate,which could be estimated through the Fast Fourier Transform results,the fine-grained HRV assessment requires to estimate the beat-to-beat interval by performing accurate beat segmentation.To address this challenge,after eliminating the respiration signals,we apply the wavelet-based denoising method to further filter out the ambient noise signals outside the frequency band of heart rate.The filtered signals thus can show a clear periodic pattern to facilitate fine-grained IBI segmentation.Finally, we propose a PCA-based scheme to derive a template to depict the inter-beat signals and use it to iteratively perform the IBI segmentation. The third challenge is to understand the sensing mechanism of RFID tag array and leverage the RFID tag array to perform accurate sensing on HRV.Specifically,as multiple tags are deployed on the human body,these tags can be regarded as a non-rigid body array to perceive the moving effect,since the tags may have relative displacement during the process of respiration;they can also be regarded as a rigid body array to perceive the reflection effect,after eliminating the respiration signal from the received RF-signal.To fuse multiple RF-signals from the tag array for accurate sensing,we capture the relationship between the RF-signals from the tag array and the corresponding movement from the heartbeat or respiration,via our reflection effect model and the moving effect mechanism,respectively.Based on the above techniques,we are able to perform data fusion over RF-signals from the RFID tag array for accurate sensing on HRV. 1.5 Contributions This paper makes four contributions:First,to the best of our knowledge,this is the first work that investigates the feasibility of Heart Rate Variability assessment based only on the COTS RFID.We leverage the RFID tag array to perform accurate sensing on HRV assessment in a lightweight and distinguishable approach.Second,we have conducted in-depth investigation on the sensing mechanism of RFID tag array.We develop a reflection effect model and a moving effect mechanism,respectively,to capture the relationship between the RF-signals from the tag array and the corresponding movement from the heart beat or respiration.Third,we design novel algorithms to extract the HRV from the RF-signals mixed with heartbeat signals,respiration signals,and ambient noises.We use wavelet-based signal denoising and signal fusion from tag array to remove the interferences to extract the IBI from the heartbeat signals.Fourth,we implement a system prototype for HRV assessment in the practical environment,and evaluate the performance with extensive experiments.Experiment results show that RF-ECG can achieve a median IBI error of 24ms,i.e,about 3%error of a normal IBI value,which is compatible to existing wired techniques. 2 RELATED WORK Sensor-based heartbeat detection:Photoplethysmogram(PPG)sensors are widely used for heart rate estima- tion by using a pulse oximeter,which illuminates the skin and measures the changes in light absorption [8]. Current wearable devices(e.g.,smartwatches and fitness trackers [3-5])usually use the PPG-based technology to measure the heart rate.However,body motion can easily distort the waveform of PPG,which causes large noise for heart rate sensing.To address this issue,recent research employs the inertial measurement units(IMU)to improve the overall accuracy of PPG [12,18].Besides PPG,Zou et al.develop a heart rate monitoring system using nanofiber-based strain sensors,which could be more compliant and comfortable [43].However,these Proc.ACM Interact.Mob.Wearable Ubiquitous Technol..Vol 2.No.2.Article 85.Publication date:lune 2018

85:4 • C. Wang et al. this challenge, we develop a mechanism to depict the RF-signal variation of the tag array caused by the chest movement, helping to eliminate the part of the respiration signals corresponding to the moving effect. We further build a model to depict the reflection effect on RF-signals of different tags from the tag array. By fusing the RF-signals from multiple tags, we are able to strengthen the heartbeat signals while suppressing the signal interferences from the others. The second challenge is to achieve a fine-grained heart beat estimation for the HRV assessment according to the reflection effect. Instead of simply calculating the heart rate, which could be estimated through the Fast Fourier Transform results, the fine-grained HRV assessment requires to estimate the beat-to-beat interval by performing accurate beat segmentation. To address this challenge, after eliminating the respiration signals, we apply the wavelet-based denoising method to further filter out the ambient noise signals outside the frequency band of heart rate. The filtered signals thus can show a clear periodic pattern to facilitate fine-grained IBI segmentation. Finally, we propose a PCA-based scheme to derive a template to depict the inter-beat signals and use it to iteratively perform the IBI segmentation. The third challenge is to understand the sensing mechanism of RFID tag array and leverage the RFID tag array to perform accurate sensing on HRV. Specifically, as multiple tags are deployed on the human body, these tags can be regarded as a non-rigid body array to perceive the moving effect, since the tags may have relative displacement during the process of respiration; they can also be regarded as a rigid body array to perceive the reflection effect, after eliminating the respiration signal from the received RF-signal. To fuse multiple RF-signals from the tag array for accurate sensing, we capture the relationship between the RF-signals from the tag array and the corresponding movement from the heartbeat or respiration, via our reflection effect model and the moving effect mechanism, respectively. Based on the above techniques, we are able to perform data fusion over RF-signals from the RFID tag array for accurate sensing on HRV. 1.5 Contributions This paper makes four contributions: First, to the best of our knowledge, this is the first work that investigates the feasibility of Heart Rate Variability assessment based only on the COTS RFID. We leverage the RFID tag array to perform accurate sensing on HRV assessment in a lightweight and distinguishable approach. Second, we have conducted in-depth investigation on the sensing mechanism of RFID tag array. We develop a reflection effect model and a moving effect mechanism, respectively, to capture the relationship between the RF-signals from the tag array and the corresponding movement from the heart beat or respiration. Third, we design novel algorithms to extract the HRV from the RF-signals mixed with heartbeat signals, respiration signals, and ambient noises. We use wavelet-based signal denoising and signal fusion from tag array to remove the interferences to extract the IBI from the heartbeat signals. Fourth, we implement a system prototype for HRV assessment in the practical environment, and evaluate the performance with extensive experiments. Experiment results show that RF-ECG can achieve a median IBI error of 24ms, i.e., about 3% error of a normal IBI value, which is compatible to existing wired techniques. 2 RELATED WORK Sensor-based heartbeat detection: Photoplethysmogram (PPG) sensors are widely used for heart rate estima￾tion by using a pulse oximeter, which illuminates the skin and measures the changes in light absorption [8]. Current wearable devices (e.g., smartwatches and fitness trackers [3–5]) usually use the PPG-based technology to measure the heart rate. However, body motion can easily distort the waveform of PPG, which causes large noise for heart rate sensing. To address this issue, recent research employs the inertial measurement units (IMU) to improve the overall accuracy of PPG [12, 18]. Besides PPG, Zou et al. develop a heart rate monitoring system using nanofiber-based strain sensors, which could be more compliant and comfortable [43]. However, these Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

RF-ECG:Heart Rate Variability Assessment Based on COTS RFID Tag Array.85:5 Water RFID tag bag array RFID antenna Enlarged RFID tag array Fig.2.Empirical study setup. dedicated sensors are fairly expensive and battery-hungry,so it is not suitable for large-scale usage and long-term deployment.HB-phone [22]estimates the human heart rate based on the vibration sensors mounted on the bed,which provides the heartbeat monitoring during sleeping.However,it only provides the average heart rate during a time window without the fine-grained IBI information.Recently,researchers try to measure the heart rate by using the smartphone cameras [19,24,31],which require the user to place her fingertip on the camera.Such vision-based approaches heavily rely on the user's pose as well as the illumination condition; moreover,as the signals are collected from the weak pulse,the accuracy is relatively lower than the approaches of obtaining the signals directly from the heartbeat.In contrast,our solution measures the heartbeat using the signal reflected from the heart,which could be more accurate than the pulse-based measurement.In addition,our solution leverages the battery-free RFID tags for heartbeat detection,which is more lightweight,scalable and not limited by the battery life. RF-based heartbeat detection:Recent research has shown that RF-signals are sensitive to the changes of the multi-path environments [9,14,23,39,41],thus both the heartbeat and breathing can be detected according to the variation of the RF-signals without requiring the user to hold or wear any device.Radar-based approaches, such as FMCW [10,42],doppler radar [17,29],are accurate at measuring such tiny environmental changes However,they usually require dedicated hardware and incur high cost for daily heartbeat monitoring.Nguyen et al [30]try to estimate the respiration rate and heart rate based on a radio transceiver and a radar navigator.But it requires specific motion devices for navigation,and it only provides the coarse-grained heart rate estimation. Wi-Fi based approaches focus on estimating the vital signs using commercial off-the-shelf(COTS)Wi-Fi devices. Specifically,they mainly leverage the Channel State Information(CSI)from both the time and frequency domain to estimate the breathing rate and heart rate [26,28,36].However,both the radar and Wi-Fi based techniques cannot label the subject,due to their device free characteristic.Therefore,it is difficult to distinguish and monitor multiple users simultaneously,especially when users share similar breathing or heartbeat patterns.To address this problem,Adib et al.proposed a radar technique to monitor the vital signs of multiple people simultaneously, by separating reflectors into different buckets depending on the distance between these people and the device [10].Hence,it requires the human subjects to be separated with a considerable distance for efficient distinction In contrast to the previous work,in this paper,we propose a novel approach for heartbeat sensing via RFID tag array,which can be regarded as an extremely lightweight sensor.Moreover,its nature of identification can be used to effectively distinguish multiple human subjects,even if these human subjects are very close to each other. 3 UNDERSTANDING HEART RATE VARIABILITY 3.1 Measurement of Periodic Signal via RFID Tag Array In order to systematically study how to use an RFID tag array to passively sense a periodic signal,e.g.,the heartbeat signal,we first use a controlled experiment to investigate the reflection influence of the periodic signal on the tag array,which is reflected from a signal source without attaching an RFID tag.Specifically,we manually Proc.ACM Interact.Mob.Wearable Ubiquitous Technol,Vol.2,No.2,Article 85.Publication date:June 2018

RF-ECG: Heart Rate Variability Assessment Based on COTS RFID Tag Array • 85:5 Enlarged RFID tag array RFID antenna RFID tag array Water bag Fig. 2. Empirical study setup. dedicated sensors are fairly expensive and battery-hungry, so it is not suitable for large-scale usage and long-term deployment. HB-phone [22] estimates the human heart rate based on the vibration sensors mounted on the bed, which provides the heartbeat monitoring during sleeping. However, it only provides the average heart rate during a time window without the fine-grained IBI information. Recently, researchers try to measure the heart rate by using the smartphone cameras [19, 24, 31], which require the user to place her fingertip on the camera. Such vision-based approaches heavily rely on the user’s pose as well as the illumination condition; moreover, as the signals are collected from the weak pulse, the accuracy is relatively lower than the approaches of obtaining the signals directly from the heartbeat. In contrast, our solution measures the heartbeat using the signal reflected from the heart, which could be more accurate than the pulse-based measurement. In addition, our solution leverages the battery-free RFID tags for heartbeat detection, which is more lightweight, scalable and not limited by the battery life. RF-based heartbeat detection: Recent research has shown that RF-signals are sensitive to the changes of the multi-path environments [9, 14, 23, 39, 41], thus both the heartbeat and breathing can be detected according to the variation of the RF-signals without requiring the user to hold or wear any device. Radar-based approaches, such as FMCW [10, 42], doppler radar [17, 29], are accurate at measuring such tiny environmental changes. However, they usually require dedicated hardware and incur high cost for daily heartbeat monitoring. Nguyen et al. [30] try to estimate the respiration rate and heart rate based on a radio transceiver and a radar navigator. But it requires specific motion devices for navigation, and it only provides the coarse-grained heart rate estimation. Wi-Fi based approaches focus on estimating the vital signs using commercial off-the-shelf (COTS) Wi-Fi devices. Specifically, they mainly leverage the Channel State Information (CSI) from both the time and frequency domain to estimate the breathing rate and heart rate [26, 28, 36]. However, both the radar and Wi-Fi based techniques cannot label the subject, due to their device free characteristic. Therefore, it is difficult to distinguish and monitor multiple users simultaneously, especially when users share similar breathing or heartbeat patterns. To address this problem, Adib et al. proposed a radar technique to monitor the vital signs of multiple people simultaneously, by separating reflectors into different buckets depending on the distance between these people and the device [10]. Hence, it requires the human subjects to be separated with a considerable distance for efficient distinction. In contrast to the previous work, in this paper, we propose a novel approach for heartbeat sensing via RFID tag array, which can be regarded as an extremely lightweight sensor. Moreover, its nature of identification can be used to effectively distinguish multiple human subjects, even if these human subjects are very close to each other. 3 UNDERSTANDING HEART RATE VARIABILITY 3.1 Measurement of Periodic Signal via RFID Tag Array In order to systematically study how to use an RFID tag array to passively sense a periodic signal, e.g., the heartbeat signal, we first use a controlled experiment to investigate the reflection influence of the periodic signal on the tag array, which is reflected from a signal source without attaching an RFID tag. Specifically, we manually Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

85:6·C.Wang et al. -35 -Tagl (ue!ped) -Tagl 40 Tag2 ap) Tag3 Tag2 Tag4 00a00a090a Tag3 Tag5 -Tag4 -47 Tag6 Tag5 Tag6 .50 2 4 6 2 6 Time (s) Time (s) (a)RSSI trend in bag swing (b)Phase trend in bag swing Fig.3.Preliminary study:Measurement of periodic signal on the RFID tag array. generate a periodic signal source by swinging a water bag as a plummet and sense the period of the water bag from a pre-deployed tag array.As shown in Figure 2,we hang a water bag to emulate the periodic signal and investigate the signal changes as we swing the water bag.A 2 x 3 tag array is deployed on the surface of a box, which is 10cm away from the water bag,to sense the periodic signal of the water bag.The size of the tag array is designed based on the size of human chest,so that the same tag array can be deployed on the chest area in the clothes of the human subject.An RFID antenna is deployed 2m away from the tag array to continuously interrogate the tags on the array. We observe that both the phase and RSSI waves have clear periodic patterns during the swing process of the water bag,but the periodic shapes are different from each tag.In particular,we swing the water bag as a pendulum to generate the periodic signal.The length of the pendulum is 25cm so that the period is about 1Hz in our setup. We show both the RSSI and phase trend of the selected tags from the tag array,respectively,in Figure 3(a)and Figure 3(b).We can clearly observe the periodical pattern in the waves of RSSI and phase trend during the swing process of water bag.The reason is that,the swing periodically changes the position of the reflection surface, thus the propagation paths(i.e.,multi-path effect)of the RF-signal are changed as well,which further leads to the periodical variation in the RF-signals.In regard to the difference of the absolute RSSI value,it is caused by the different positions of each tag.Moreover,even if all the tags present the same cycle time,the exact shape of their waveforms are different among the tags.This indicates that different tags on the tag array have different sensitivities to the reflection effect of the swing bag.We will build a theoretical model to explain the phenomenon later in Section 3.3. 3.2 Measurement of HRV in Real Settings We further investigate how the actual heartbeat affects the RF-signals in real settings.Specifically,we attach a 2 x 3 tag array on the chest area in the clothes of the human subject.We first let the human subject hold the breath for 20s,indicating that the chest movement of breathing can be negligible.We then let the human subject breathe normally for 20s,and thus there exists obvious chest movement of breathing.We respectively collect the phase/RSSI values of these two sets.We select an arbitrary tag from the tag array and present the results in Figure 4. According to Figure 4(a),we can observe weak but fairly clear periodic heartbeat patterns from the phase sequences,since the chest movement of breathing can be negligible.According to Figure 4(b),we can observe obvious periodic respiration patterns for the chest movement of breathing,as the moving effect due to the chest movement clearly changes the phase values.However,the periodic heartbeat patterns can hardly be detected Proc.ACM Interact.Mob.Wearable Ubiquitous Technol,Vol.2,No.2,Article 85.Publication date:June 2018

85:6 • C. Wang et al. Time (s) 0 2 4 6 8 RSSI (dBm) -50 -45 -40 -35 Tag1 Tag2 Tag3 Tag4 Tag5 Tag6 (a) RSSI trend in bag swing Time (s) 0 2 4 6 8 Phase (Radian) -2 -1 0 1 2 Tag1 Tag2 Tag3 Tag4 Tag5 Tag6 (b) Phase trend in bag swing Fig. 3. Preliminary study: Measurement of periodic signal on the RFID tag array. generate a periodic signal source by swinging a water bag as a plummet and sense the period of the water bag from a pre-deployed tag array. As shown in Figure 2, we hang a water bag to emulate the periodic signal and investigate the signal changes as we swing the water bag. A 2 × 3 tag array is deployed on the surface of a box, which is 10cm away from the water bag, to sense the periodic signal of the water bag. The size of the tag array is designed based on the size of human chest, so that the same tag array can be deployed on the chest area in the clothes of the human subject. An RFID antenna is deployed 2m away from the tag array to continuously interrogate the tags on the array. We observe that both the phase and RSSI waves have clear periodic patterns during the swing process of the water bag, but the periodic shapes are different from each tag. In particular, we swing the water bag as a pendulum to generate the periodic signal. The length of the pendulum is 25cm so that the period is about 1Hz in our setup. We show both the RSSI and phase trend of the selected tags from the tag array, respectively, in Figure 3(a) and Figure 3(b). We can clearly observe the periodical pattern in the waves of RSSI and phase trend during the swing process of water bag. The reason is that, the swing periodically changes the position of the reflection surface, thus the propagation paths (i.e., multi-path effect) of the RF-signal are changed as well, which further leads to the periodical variation in the RF-signals. In regard to the difference of the absolute RSSI value, it is caused by the different positions of each tag. Moreover, even if all the tags present the same cycle time, the exact shape of their waveforms are different among the tags. This indicates that different tags on the tag array have different sensitivities to the reflection effect of the swing bag. We will build a theoretical model to explain the phenomenon later in Section 3.3. 3.2 Measurement of HRV in Real Settings We further investigate how the actual heartbeat affects the RF-signals in real settings. Specifically, we attach a 2 × 3 tag array on the chest area in the clothes of the human subject. We first let the human subject hold the breath for 20s, indicating that the chest movement of breathing can be negligible. We then let the human subject breathe normally for 20s, and thus there exists obvious chest movement of breathing. We respectively collect the phase/RSSI values of these two sets. We select an arbitrary tag from the tag array and present the results in Figure 4. According to Figure 4(a), we can observe weak but fairly clear periodic heartbeat patterns from the phase sequences, since the chest movement of breathing can be negligible. According to Figure 4(b), we can observe obvious periodic respiration patterns for the chest movement of breathing, as the moving effect due to the chest movement clearly changes the phase values. However, the periodic heartbeat patterns can hardly be detected Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

RF-ECG:Heart Rate Variability Assessment Based on COTS RFID Tag Array.85:7 2.2 -52 -525 10 a的人心 10 10 15 Time(s) Time(s) Frequency(Hz) (a)The phase/RSSI sequence without breath-(b)The phase/RSSI sequence with breathing(c)Frequency domain analysis of the phase ing sequence Fig.4.Preliminary study:measurement of HRV with and without breathing. anymore,since the reflection effect is orders of magnitude smaller than the moving effect in the tag array.Besides, we cannot detect any periodic patterns from the RSSI sequences in both situations,since the resolution of RSSI is rather coarse-grained.We further perform FFT analysis on the phase sequences from the frequency domain. Figure 4(c)shows the experiment results in both situations.For the situation without breathing,we can clearly find a small peak at 1.355Hz,which is corresponding to the heartbeat frequency band.This asserts the reflection effect exists but relatively small.For the situation with breathing,we can clearly detect a peak at 0.2Hz,which corresponds to the respiration frequency band,besides,we can only detect a very small peak at 1.3Hz,which is almost buried in the noises.Therefore,to obtain the precise heartbeat signal,it is essential to first remove the influence of respiration and then strengthen the heartbeat reflection. 3.3 Modeling HRV via Tag Array Sensing In this subsection,we model the relationship between the RF-signals from the tag array and the heart displacement in the reflection effect,which is the periodic signal of the heartbeat. 3.3.1 Extracting the Reflection Signal.To understand how the heart displacement affects the reflection effect,it is essential to extract the reflection signal from the received RF-signal,i.e.,the RF-signal reflected from the heart to the tags.Hence,we use a signal propagation model to depict the RF-signal transmission with the reflection effect.As shown in Figure 5,we use A,T,B and C to denote the RFID antenna,RFID tag,reflection object and background environment,respectively.First,the antenna A sends the continuous wave to activate the tags.Due to the multi-path effect,the tagT receives a superposed signal,which contains the line-of-sight(LOS)signal SaT(blue line),the reflection signal Sa from the reflection object B(red line),as well as the reflection signal Sac from the background environment (green line).Then,after the specified tag is successfully activated,it backscatters the signal to the antenna with necessary data modulation.Hence,the raw signal received by the antenna A can be represented as: S,=hT+ah(S-+T+SA-B-T+S领-→+C→T), (1) where h represents the signal attenuation due to propagation path loss and h is the reflection coefficient of the tag.Since both the LOS signal and the reflection signal from the background environment are usually stable during the whole propagation,we combine them as Sr.0=hTah(Sa-T+SA→c→T), (2) and denote the remained reflection signal from the reflection object as Sr.1=hT-→ahrSa→B-→T (3) Proc.ACM Interact.Mob.Wearable Ubiquitous Technol,Vol.2,No.2,Article 85.Publication date:June 2018

RF-ECG: Heart Rate Variability Assessment Based on COTS RFID Tag Array • 85:7 Time (s) 0 5 10 15 20 Phase (Radian) 1.8 2 2.2 RSSI (dBm) -54 -53 -52 (a) The phase/RSSI sequence without breath￾ing Time (s) 0 5 10 15 20 Phase (Radian) 1.7 1.9 2.1 RSSI (dBm) -54.5 -53.5 -52.5 (b) The phase/RSSI sequence with breathing Frequency (Hz) 0 0.5 1 1.5 2 FFT 0 50 100 150 Phase without breath Phase with breath (c) Frequency domain analysis of the phase sequence Fig. 4. Preliminary study: measurement of HRV with and without breathing. anymore, since the reflection effect is orders of magnitude smaller than the moving effect in the tag array. Besides, we cannot detect any periodic patterns from the RSSI sequences in both situations, since the resolution of RSSI is rather coarse-grained. We further perform FFT analysis on the phase sequences from the frequency domain. Figure 4(c) shows the experiment results in both situations. For the situation without breathing, we can clearly find a small peak at 1.355Hz, which is corresponding to the heartbeat frequency band. This asserts the reflection effect exists but relatively small. For the situation with breathing, we can clearly detect a peak at 0.2Hz, which corresponds to the respiration frequency band, besides, we can only detect a very small peak at 1.3Hz, which is almost buried in the noises. Therefore, to obtain the precise heartbeat signal, it is essential to first remove the influence of respiration and then strengthen the heartbeat reflection. 3.3 Modeling HRV via Tag Array Sensing In this subsection, we model the relationship between the RF-signals from the tag array and the heart displacement in the reflection effect, which is the periodic signal of the heartbeat. 3.3.1 Extracting the Reflection Signal. To understand how the heart displacement affects the reflection effect, it is essential to extract the reflection signal from the received RF-signal, i.e., the RF-signal reflected from the heart to the tags. Hence, we use a signal propagation model to depict the RF-signal transmission with the reflection effect. As shown in Figure 5, we use A, T, B and C to denote the RFID antenna, RFID tag, reflection object and background environment, respectively. First, the antenna A sends the continuous wave to activate the tags. Due to the multi-path effect, the tag T receives a superposed signal, which contains the line-of-sight (LOS) signal SA→T (blue line), the reflection signal SA→B→T from the reflection object B (red line), as well as the reflection signal SA→C→T from the background environment (green line). Then, after the specified tag is successfully activated, it backscatters the signal to the antenna with necessary data modulation. Hence, the raw signal received by the antenna A can be represented as: Sr = hT→AhT(SA→T + SA→B→T + SA→C→T), (1) where hT→A represents the signal attenuation due to propagation path loss and hT is the reflection coefficient of the tag. Since both the LOS signal and the reflection signal from the background environment are usually stable during the whole propagation, we combine them as Sr,0 = hT→AhT(SA→T + SA→C→T), (2) and denote the remained reflection signal from the reflection object as Sr,1 = hT→AhTSA→B→T. (3) Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

85:8·C.Wang et al. Reflection Object(Heart) Sr =hT-AhT (SA+B→T+SAT+Sc→T) dA-B △dBT≈△h cos pr Ah PT D'D C SA-C-T Fig.5.Reflection model of heart displacement. To extract the reflection signal S.1 from the superposed signal S,we could subtract the complex signal S.o[41]. which can be measured simply by removing the reflection object. Figure 6 shows the RSSI and phase variation of the reflection signals from periodic signal of the water bag according to the subtraction Sr.1=S,-Sr.0.We note that,for each reflection signal,the RSSI variation is irregular and its value is usually less than -60dBm,as the power of the reflection signal is rather weak and easy to be affected.In fact,since the reflection object,ie.,the water bag,only moves a small distance,theoretically the RSSI variation should be very small according to Friis Equation [16].Meanwhile,the phases of the reflection signal from different tags all have obvious periodical patterns in the waveforms.Moreover,they not only have some similarities in the waveform contours,but also have differences in the waveform details among each other. Therefore,this implies that all the tags are subjected to similar reflection influence,but the performance in sensitivity are different among different tags. 3.3.2 Estimating Heart Displacement via a Single Tag.After extracting the reflection signal,it is essential to further figure out the relationship between the heart displacement and the reflection RF-signal.Since the phase of reflection signal is more sensitive to the heart displacement than RSSI,we use the phase as a metric for RF-signal to depict the corresponding relationship.According to Eq.(3),let Abe the wave length of RF-signal,the phase of reflection signal S,.can be calculated from the reflection path length as: 8g=2x(女月+d-8+d87+0eon)mod2元， (4) 入 where the superscript'indicates the reflection signal and Ocons is the constant phase deviation due to reflection. Since d is fixed during the whole propagation and econs is constant,when the reflection object moves from BtoB',the phase change△9only depends on the change of path length△da-→sand△dg-T,as: △85=2mAdA-8+Adg-7 mod 2. (5) 入 Without loss of generality,we first consider the path length change Adg.Assume that the points D and D'are respectively the projection of the points B and B'on the horizontal line of T in Figure 5,then the edge length dsp =dg'p.When the reflection object moves from B to B',the reflection path changes from ds to dgT,which are,respectively,the hypotenuses of two right triangles ABDT and AB'D'T.Let or and to denote the angle /BTD and /B'T D',respectively.Then,the path length change Ads-can be calculated as follows: △dg→T=dgT-dsT= dsp dsD (6) sin sinor Proc.ACM Interact.Mob.Wearable Ubiquitous Technol.,Vol.2,No.2,Article 85.Publication date:June 2018

85:8 • C. Wang et al. B A T B‘ !" #$ #$ % !&'() * !" +,- #) ./ 0 ")(1") 2.1('() 3 .1() 3 .4()5 D’ D Reflection Object (Heart) #1 C Fig. 5. Reflection model of heart displacement. To extract the reflection signal Sr,1 from the superposed signal Sr , we could subtract the complex signal Sr,0 [41], which can be measured simply by removing the reflection object. Figure 6 shows the RSSI and phase variation of the reflection signals from periodic signal of the water bag according to the subtraction Sr,1 = Sr −Sr,0. We note that, for each reflection signal, the RSSI variation is irregular and its value is usually less than −60dBm, as the power of the reflection signal is rather weak and easy to be affected. In fact, since the reflection object, i.e., the water bag, only moves a small distance, theoretically the RSSI variation should be very small according to Friis Equation [16]. Meanwhile, the phases of the reflection signal from different tags all have obvious periodical patterns in the waveforms. Moreover, they not only have some similarities in the waveform contours, but also have differences in the waveform details among each other. Therefore, this implies that all the tags are subjected to similar reflection influence, but the performance in sensitivity are different among different tags. 3.3.2 Estimating Heart Displacement via a Single Tag. After extracting the reflection signal, it is essential to further figure out the relationship between the heart displacement and the reflection RF-signal. Since the phase of reflection signal is more sensitive to the heart displacement than RSSI, we use the phase as a metric for RF-signal to depict the corresponding relationship. According to Eq.(3), let λ be the wave length of RF-signal, the phase of reflection signal Sr,1 can be calculated from the reflection path length as: θ r T = 2π( dT→A + dA→B + dB→T λ + θcons) mod 2π, (4) where the superscript r indicates the reflection signal and θcons is the constant phase deviation due to reflection. Since dT→A is fixed during the whole propagation and θcons is constant, when the reflection object moves from B to B ′ , the phase change ∆θ r T only depends on the change of path length ∆dA→B and ∆dB→T, as: ∆θ r T = 2π ∆dA→B + ∆dB→T λ mod 2π. (5) Without loss of generality, we first consider the path length change ∆dB→T. Assume that the points D and D′ are respectively the projection of the points B and B ′ on the horizontal line of T in Figure 5, then the edge length dBD = dB′D′. When the reflection object moves from B to B ′ , the reflection path changes from dB→T to dB′→T, which are, respectively, the hypotenuses of two right triangles △BDT and △B′D′T. Let ϕT and ϕ ′ T to denote the angle ∠BT D and ∠B ′T D′ , respectively. Then, the path length change ∆dB→T can be calculated as follows: ∆dB→T = dB′T − dBT = dBD sinϕ ′ T − dBD sinϕT . (6) Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

RF-ECG:Heart Rate Variability Assessment Based on COTS RFID Tag Array.85:9 -60 -Tagl Tagl -Tag2 Tag2 Tag3 Tag3 Tag4 0 Tag4 Tag5 Tag5 80 -Tag6 Tag6 -90 2 6 0 2 6 8 Time (s) Time (s) (a)The RSSI of the reflection signal (b)The phase of the reflection signal Fig.6.The extracted reflection signal. Let Ah be the displacement of the reflection object(i.e.,the heart displacement dgs),it is essentially equal to △h=dgg=dpD= dgD」 dsD tan (7) tano By combining the two equations to remove dsD,we obtain sino-sin △dg-T=△ (8) sin(or-) If we define the angle deviation Aor=or-o,then o=or-Aor,thus △dg-T=△h(cos+sin 1-cos△9虹). (9) sin△pr In regard to the heartbeat,the reflection object(ie.,the heart)actually moves a rather small distance Ah,thus the angle deviation.Hence=tan.Therefore,Eq()can be simplified as follows: △dgT≈△h cos r. (10) Similarly,for the path length change Ads,let denote the angle between AB and the horizontal line,we also obtain△d→s≈△hcosoa, Therefore,by combining Eq.(10)and Eq.(5),we obtain △8g≈2 Ah(cos97+cosp》 mod 2. (11) Hence,for any arbitrary tag T.given the phase change of the reflection signal A0,the heart displacement Ah can be estimated as follows: (△G+2πk)n △h≈ 2π(cosr+cosoa) ,where k=·,-1,0,1, (12) Here,k represents the periods of the phase values.Since the heart displacement is rather small,k equals to 0 in our problem. Proc.ACM Interact.Mob.Wearable Ubiquitous Technol,Vol.2,No.2,Article 85.Publication date:June 2018

RF-ECG: Heart Rate Variability Assessment Based on COTS RFID Tag Array • 85:9 Time (s) 0 2 4 6 8 RSSI (dBm) -90 -80 -70 -60 Tag1 Tag2 Tag3 Tag4 Tag5 Tag6 (a) The RSSI of the reflection signal Time (s) 0 2 4 6 8 Phase (Radian) -4 -2 0 2 4 Tag1 Tag2 Tag3 Tag4 Tag5 Tag6 (b) The phase of the reflection signal Fig. 6. The extracted reflection signal. Let ∆h be the displacement of the reflection object (i.e., the heart displacement dB′B), it is essentially equal to ∆h = dB′B = dD′D = dBD tanϕ ′ T − dBD tanϕT . (7) By combining the two equations to remove dBD, we obtain ∆dB→T = ∆h sinϕT − sinϕ ′ T sin(ϕT − ϕ ′ T ) . (8) If we define the angle deviation ∆ϕT = ϕT − ϕ ′ T , then ϕ ′ T = ϕT − ∆ϕT, thus ∆dB→T = ∆h(cosϕT + sinϕT 1 − cos ∆ϕT sin ∆ϕT ). (9) In regard to the heartbeat, the reflection object (i.e., the heart) actually moves a rather small distance ∆h, thus the angle deviation ∆ϕT → 0. Hence, 1−cos ∆ϕT sin ∆ϕT = tan ∆ϕT 2 → 0. Therefore, Eq (9) can be simplified as follows: ∆dB→T ≈ ∆h cosϕT. (10) Similarly, for the path length change ∆dA→B, let ϕA denote the angle between AB and the horizontal line, we also obtain ∆dA→B ≈ ∆h cosϕA . Therefore, by combining Eq. (10) and Eq. (5), we obtain ∆θ r T ≈ 2π∆h(cosϕT + cosϕA) λ mod 2π. (11) Hence, for any arbitrary tag T , given the phase change of the reflection signal ∆θ r T , the heart displacement ∆h can be estimated as follows: ∆h ≈ (∆θ r T + 2πk)λ 2π(cosϕT + cosϕA) , where k = · · · , −1, 0, 1, · · · (12) Here, k represents the periods of the phase values. Since the heart displacement is rather small, k equals to 0 in our problem. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018

85:10·C.Wang et al.. fy Ti(xiyiO) Pit X 0-O-@ A(xA,yA,ZA） B(xB,yB,ZB) 项○ 中 (a)Compute the parameters cos;and cos in the 3D coordinate (b)The phase change of reflection signal for different system tags Fig.7.Heart displacement estimation via tag array. 3.3.3 Estimating Heart Displacement via Tag Array.In Section 3.3.2,we derive the relationship between the heart displacement and the reflection signal from a single tag,according to the model in Eq.(12).However,due to the issues such as the multi-path effect and the ambient noises,the reflection signal from a single tag can be distorted if it is deployed in a position with interferences and noises,which further leads to the errors in the heart displacement estimation.Therefore,it is essential to further investigate the reflection signals of multiple tags from different positions of the tag array to improve the sensing performance. Suppose we build a 3-dimensional coordinate system by setting the center of the tag array as the origin O,the X and Y-axis are parallel to the tag array,whereas the Z-axis is orthogonal to the tag array.Then,each tag Ti from the tag array can be denoted as a point with the coordinate(xi,yi.0).Assume that the reflection object(e.g., the heart)can be regarded as a virtual point B according to the overall reflection effect,thus it can be denoted with the coordinate(xB,yB,zg).Besides,the antenna A can be denoted with the coordinate (xA,yA,ZA).Then, as shown in Figure 7(a),for an arbitrary tag 7i from the tag array,the parameters cos ;(i.e.,cos o for tag 7i) and cosa from the model in Eq.(12)can be depicted as follows: C0s9:= VxB-xP+(yB-1P+z话 (13) C0sφA= V(xB-xA)2+(yB-yA)+(zB-zA)2 According to the model of phase change in Eq.(11),in regard to different tags from the tag array,the parameter a is only related to the reflection object,so it should be consistent for all tags;the parameter actually depends on the exact position of the tag Ti,so it should be different for different tags.According to Eq.(11),the smaller the value of i is,the larger phase change of the tag Ti is obtained.This implies that the reflection object causes larger signal influences to the tags which is more close in position.Based on this property,it is possible to further locate the position of the reflection object(i.e.,the heart),according to the signal variances from the tag array. We further validate this hypothesis with an empirical study.The experiment setup is the same as Section 3.1.We deploy a 2 x 3 tag array to detect the reflection of the swinging water bag,and label each tag as(x,y) according to its order in the specified dimension.The center of the water bag is located around the coordinate (2.5,1.5).We show the reflection phase range of all the 6 tags in Figure 7(b).It is found that the tag at(2,1)has the largest reflection phase range,whereas the other tags gradually decreases,as they are away from the tag at (2,1).Moreover,based on the amplitude of the reflection phase,we can infer the position of the water bag around (2,1)and (3,1),which is consistent to the groundtruth and the model in Eq.(11).In addition to the position of water bag,we also estimate the displacement of the water bag based on the reflection phase matrix.In Figure 7(b), Proc.ACM Interact.Mob.Wearable Ubiquitous Technol,Vol.2,No.2,Article 85.Publication date:June 2018

85:10 • C. Wang et al. x z y !"#$" % &" % '( )#$)% &)% *)( +" ,#$,% &,% *,( +) O (a) Compute the parameters cos ϕi and cos ϕA in the 3D coordinate system 3 2 X 1 1 2 Y 10 5 0 Reflection phase range (Radian) (b) The phase change of reflection signal for different tags Fig. 7. Heart displacement estimation via tag array. 3.3.3 Estimating Heart Displacement via Tag Array. In Section 3.3.2, we derive the relationship between the heart displacement and the reflection signal from a single tag, according to the model in Eq.(12). However, due to the issues such as the multi-path effect and the ambient noises, the reflection signal from a single tag can be distorted if it is deployed in a position with interferences and noises, which further leads to the errors in the heart displacement estimation. Therefore, it is essential to further investigate the reflection signals of multiple tags from different positions of the tag array to improve the sensing performance. Suppose we build a 3-dimensional coordinate system by setting the center of the tag array as the origin O, the X and Y-axis are parallel to the tag array, whereas the Z-axis is orthogonal to the tag array. Then, each tag Ti from the tag array can be denoted as a point with the coordinate (xi ,yi , 0). Assume that the reflection object (e.g., the heart) can be regarded as a virtual point B according to the overall reflection effect, thus it can be denoted with the coordinate (xB,yB, zB). Besides, the antenna A can be denoted with the coordinate (xA,yA, zA). Then, as shown in Figure 7(a), for an arbitrary tag Ti from the tag array, the parameters cosϕi (i.e., cosϕT for tag Ti ) and cosϕA from the model in Eq.(12) can be depicted as follows:    cosϕi = |zB | √ (xB −xi ) 2+(yB −yi ) 2+z 2 B cosϕA = |zB −zA | √ (xB −xA) 2+(yB −yA) 2+(zB −zA) 2 (13) According to the model of phase change in Eq.(11), in regard to different tags from the tag array, the parameter ϕA is only related to the reflection object, so it should be consistent for all tags; the parameter ϕi actually depends on the exact position of the tag Ti , so it should be different for different tags. According to Eq.(11), the smaller the value of ϕi is, the larger phase change of the tag Ti is obtained. This implies that the reflection object causes larger signal influences to the tags which is more close in position. Based on this property, it is possible to further locate the position of the reflection object (i.e., the heart), according to the signal variances from the tag array. We further validate this hypothesis with an empirical study. The experiment setup is the same as Section 3.1. We deploy a 2 × 3 tag array to detect the reflection of the swinging water bag, and label each tag as (x,y) according to its order in the specified dimension. The center of the water bag is located around the coordinate (2.5, 1.5). We show the reflection phase range of all the 6 tags in Figure 7(b). It is found that the tag at (2, 1) has the largest reflection phase range, whereas the other tags gradually decreases, as they are away from the tag at (2, 1). Moreover, based on the amplitude of the reflection phase, we can infer the position of the water bag around (2, 1) and (3, 1), which is consistent to the groundtruth and the model in Eq.(11). In addition to the position of water bag, we also estimate the displacement of the water bag based on the reflection phase matrix. In Figure 7(b), Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 2, No. 2, Article 85. Publication date: June 2018