VECTOR:Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices HAORAN WAN,Nanjing University,China LEI WANG,Nanjing University,China TING ZHAO,Nanjing University,China KE SUN,University of California,San Diego,USA SHUYU SHI,Nanjing University,China HAIPENG DAl,Nanjing University,China GUIHAI CHEN,Nanjing University,China HAODONG LIU,Huawei,China WEI WANG,Nanjing University,China Ambient temperature distribution monitoring is desired in a variety of real-life applications including indoors temperature 144 control and building energy management.Traditional temperature sensors have their limitations in the aspects of single point/item based measurements,slow response time and huge cost for distribution estimation.In this paper,we introduce VECTOR,a temperature-field monitoring system that achieves high temperature sensing accuracy and fast response time using commercial sound playing/recording devices.First,our system uses a distributed ranging algorithm to measure the time-of-flight of multiple sound paths with microsecond resolution.We then propose a dRadon transform algorithm that reconstructs the temperature distribution from the measured speed of sound along different paths.Our experimental results show that we can measure the temperature with an error of 0.25C from single sound path and reconstruct the temperature distribution at a decimeter-level spatial resolution. CCS Concepts:Human-centered computing-Ubiquitous and mobile computing systems and tools. Additional Key Words and Phrases:Temperature monitoring.Acoustic signals,Wireless sensing. ACM Reference Format: Haoran Wan,Lei Wang,Ting Zhao,Ke Sun,Shuyu Shi,Haipeng Dai,Guihai Chen,Haodong Liu,and Wei Wang.2022. VECTOR:Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices.Proc.ACM Interact.Mob.Wearable Ubiquitous Technol 6,3,Article 144(September 2022),28 pages.https://doiorg/10.1145/3550336 1 INTRODUCTION Ambient temperature monitoring is vital to a variety of ubiquitous computing applications,from warehouse monitoring [1]to building energy management [2]and greenhouse temperature control [3].With the ever Authors'addresses:Haoran Wan,wanhr@smail.nju.edu.cn,Nanjing University,Nanjing,Jiangsu,China,210023;Lei Wang,wang_l@pku. edu.cn,Nanjing University,Nanjing.Jiangsu,China,210023;Ting Zhao,zhaoting@smailnju.edu.cn,Nanjing University,Nanjing,Jiangsu, China,210023;Ke Sun,kesun@eng.ucsd.edu,University of California,San Diego,La Jolla,California,USA,92093;Shuyu Shi,ssy@nju.edu.cn, Nanjing University,Nanjing,Jiangsu,China,210023;Haipeng Dai,haipengdai@nju.edu.cn,Nanjing University,Nanjing.Jiangsu,China, 210023;Guihai Chen,gchen@nju.edu.cn,Nanjing University,Nanjing,Jiangsu,China,210023;Haodong Liu,liuhaodong@huawei.com, Huawei,Shanghai,Shanghai,China,300060;Wei Wang.ww@nju.edu.cn,Nanjing University,Nanjing.Jiangsu,China,210023. 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. 2022 Association for Computing Machinery. 2474-9567/2022/9-ART144$15.00 https:/doi.org/10.1145/3550336 Proc.ACM Interact.Mob.Wearable Ubiquitous Technol.,Vol.6,No.3,Article 144.Publication date:September 2022
144 VECTOR: Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices HAORAN WAN, Nanjing University, China LEI WANG, Nanjing University, China TING ZHAO, Nanjing University, China KE SUN, University of California, San Diego, USA SHUYU SHI, Nanjing University, China HAIPENG DAI, Nanjing University, China GUIHAI CHEN, Nanjing University, China HAODONG LIU, Huawei, China WEI WANG, Nanjing University, China Ambient temperature distribution monitoring is desired in a variety of real-life applications including indoors temperature control and building energy management. Traditional temperature sensors have their limitations in the aspects of single point/item based measurements, slow response time and huge cost for distribution estimation. In this paper, we introduce VECTOR, a temperature-field monitoring system that achieves high temperature sensing accuracy and fast response time using commercial sound playing/recording devices. First, our system uses a distributed ranging algorithm to measure the time-of-flight of multiple sound paths with microsecond resolution. We then propose a dRadon transform algorithm that reconstructs the temperature distribution from the measured speed of sound along different paths. Our experimental results show that we can measure the temperature with an error of 0.25◦C from single sound path and reconstruct the temperature distribution at a decimeter-level spatial resolution. CCS Concepts: • Human-centered computing → Ubiquitous and mobile computing systems and tools. Additional Key Words and Phrases: Temperature monitoring, Acoustic signals, Wireless sensing. ACM Reference Format: Haoran Wan, Lei Wang, Ting Zhao, Ke Sun, Shuyu Shi, Haipeng Dai, Guihai Chen, Haodong Liu, and Wei Wang. 2022. VECTOR: Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 6, 3, Article 144 (September 2022), 28 pages. https://doi.org/10.1145/3550336 1 INTRODUCTION Ambient temperature monitoring is vital to a variety of ubiquitous computing applications, from warehouse monitoring [1] to building energy management [2] and greenhouse temperature control [3]. With the ever Authors’ addresses: Haoran Wan, wanhr@smail.nju.edu.cn, Nanjing University, Nanjing, Jiangsu, China, 210023; Lei Wang, wang_l@pku. edu.cn, Nanjing University, Nanjing, Jiangsu, China, 210023; Ting Zhao, zhaoting@smail.nju.edu.cn, Nanjing University, Nanjing, Jiangsu, China, 210023; Ke Sun, kesun@eng.ucsd.edu, University of California, San Diego, La Jolla, California, USA, 92093; Shuyu Shi, ssy@nju.edu.cn, Nanjing University, Nanjing, Jiangsu, China, 210023; Haipeng Dai, haipengdai@nju.edu.cn, Nanjing University, Nanjing, Jiangsu, China, 210023; Guihai Chen, gchen@nju.edu.cn, Nanjing University, Nanjing, Jiangsu, China, 210023; Haodong Liu, liuhaodong@huawei.com, Huawei, Shanghai, Shanghai, China, 300060; Wei Wang, ww@nju.edu.cn, Nanjing University, Nanjing, Jiangsu, China, 210023. 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. © 2022 Association for Computing Machinery. 2474-9567/2022/9-ART144 $15.00 https://doi.org/10.1145/3550336 Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 3, Article 144. Publication date: September 2022
144:2·Wan et al.. increasing user demands,the capability of monitoring the temperature distribution within a given space,e.g.,in a room or in a car,becomes important for the next generation HVAC(Heating.Ventilation,and Air Conditioning) systems.Instead of a single temperature reading given by traditional temperature sensing systems,temperature distribution monitoring systems provide fine-grained spatial temperature variation information of the target area. For warehouse monitoring,temperature requirements for food storage are strict and vary for different types of food that are stored in different places of the same warehouse,e.g.,32~40F(0~4.4 in Celsius)for refrigerated storage and 50F(10C)for dry food storage [4].For indoor air conditioning,precise temperature monitoring and controlling are vital given that city folk spend 80~90%of their time indoors [5,6].Different users may have different temperature preferences and improper indoor temperature would cause low productivity and even sickness [7].Different parts of the same room may also have different heating conditions due to sunlight from windows or room occupation.Moreover,air conditioners and electric fans have accounted for 10%of all global electricity consumption [8].With precise temperature control,the energy consumption could be reduced by more than 30%while maintaining the thermal comfort for users [9].As the premise for precise temperature distribution control,the capability of monitoring the temperature of different locations in the same space becomes an important research issue. Widely used temperature sensors cannot satisfy the demands for temperature distribution monitoring.First, most temperature sensors only measure the temperature at a single location.To reconstruct the temperature distribution,we have to densely deploy temperature sensors over the space.While the cost of temperature sensors are low,the wiring and deployment cost could be far more than the sensor hardware,even for small spaces such as in the car or in a room.Existing works try to infer the ambient temperature distribution using physical model [10]with the assumption that temperature sensor's readings represent the average in a room or a confined space.Such estimations are unreliable and coarse-grained,since our experiments show that the temperature difference can be as large as 3C within the limited space in a car.While infrared cameras can capture temperature distribution [11],they are still too expensive for environment monitoring applications.Second,most temperature sensors are based on thermistors or thermocouples,which measures the temperature of the sensor's probe instead of the air.Therefore,the material of the sensor need to be heated/cooled when the temperature changes so that these sensors have large response delays.Such extra delays often lead to difficulties in designing a stable fine-grained temperature control algorithm.With ubiquitous mobile devices and wireless signals(acoustic signal and electromagnetic signal including mmWave and Wi-Fi signals)surrounding us,it's natural to come to the idea of reusing these existing devices and signals to conduct temperature distribution estimation.However, most existing works focus on approximating and replacing the traditional temperature sensors with mobile devices [12]or RFID tags [13,14]and only measure the temperature for single point or item [15]. In this paper,we develop a system called VECTOR,(Velocity based Temperature-field Monitoring),that can reconstruct the temperature distribution with a small number of low-cost ubiquitous acoustic devices.Our design is based on the fact that the speed of sound is physically related to the air temperature along the sound propagation path.Therefore,we can infer the temperature along a given path using the time-of-flight(ToF) measurements of sound signals.As illustrated in Fig.1(a),a pair of acoustic devices can monitor multiple acoustic paths passing through different regions in a car.Temperature changes on different regions incur different phase variations determined by the ToF along specific segments of the sound paths.Therefore,we can reconstruct the temperature distribution using the different phases changes of these line-of-sight (LoS)paths and reflected paths. In our experiments,VECTOR can measure the temperature along the sound path with an accuracy of 0.25C by sensing slight changes in the speed of sound,when the distance between two devices is known.Moreover, VECTOR incurs minimal hardware cost,as it can reuse the built-in audio systems that are already widely deployed in indoor environments or in cars.Compared with traditional temperature sensors,the sound-based scheme directly measures the temperature in the air instead of the temperature of the sensor.Therefore,VECTOR can detect human perceivable temperature fluctuations within a few seconds,while the latency of traditional sensor Proc.ACM Interact.Mob.Wearable Ubiquitous Technol..Vol 6.No.3.Article 144.Publication date:September 2022
144:2 • Wan et al. increasing user demands, the capability of monitoring the temperature distribution within a given space, e.g., in a room or in a car, becomes important for the next generation HVAC (Heating, Ventilation, and Air Conditioning) systems. Instead of a single temperature reading given by traditional temperature sensing systems, temperature distribution monitoring systems provide fine-grained spatial temperature variation information of the target area. For warehouse monitoring, temperature requirements for food storage are strict and vary for different types of food that are stored in different places of the same warehouse, e.g., 32 ∼ 40◦F (0 ∼ 4.4 in Celsius) for refrigerated storage and 50◦F (10◦C ) for dry food storage [4]. For indoor air conditioning, precise temperature monitoring and controlling are vital given that city folk spend 80 ∼ 90% of their time indoors [5, 6]. Different users may have different temperature preferences and improper indoor temperature would cause low productivity and even sickness [7]. Different parts of the same room may also have different heating conditions due to sunlight from windows or room occupation. Moreover, air conditioners and electric fans have accounted for 10% of all global electricity consumption [8]. With precise temperature control, the energy consumption could be reduced by more than 30% while maintaining the thermal comfort for users [9]. As the premise for precise temperature distribution control, the capability of monitoring the temperature of different locations in the same space becomes an important research issue. Widely used temperature sensors cannot satisfy the demands for temperature distribution monitoring. First, most temperature sensors only measure the temperature at a single location. To reconstruct the temperature distribution, we have to densely deploy temperature sensors over the space. While the cost of temperature sensors are low, the wiring and deployment cost could be far more than the sensor hardware, even for small spaces such as in the car or in a room. Existing works try to infer the ambient temperature distribution using physical model [10] with the assumption that temperature sensor’s readings represent the average in a room or a confined space. Such estimations are unreliable and coarse-grained, since our experiments show that the temperature difference can be as large as 3 ◦C within the limited space in a car. While infrared cameras can capture temperature distribution [11], they are still too expensive for environment monitoring applications. Second, most temperature sensors are based on thermistors or thermocouples, which measures the temperature of the sensor’s probe instead of the air. Therefore, the material of the sensor need to be heated/cooled when the temperature changes so that these sensors have large response delays. Such extra delays often lead to difficulties in designing a stable fine-grained temperature control algorithm. With ubiquitous mobile devices and wireless signals (acoustic signal and electromagnetic signal including mmWave and Wi-Fi signals) surrounding us, it’s natural to come to the idea of reusing these existing devices and signals to conduct temperature distribution estimation. However, most existing works focus on approximating and replacing the traditional temperature sensors with mobile devices [12] or RFID tags [13, 14] and only measure the temperature for single point or item [15]. In this paper, we develop a system called VECTOR, (Velocity based Temperature-field Monitoring), that can reconstruct the temperature distribution with a small number of low-cost ubiquitous acoustic devices. Our design is based on the fact that the speed of sound is physically related to the air temperature along the sound propagation path. Therefore, we can infer the temperature along a given path using the time-of-flight (ToF) measurements of sound signals. As illustrated in Fig. 1(a), a pair of acoustic devices can monitor multiple acoustic paths passing through different regions in a car. Temperature changes on different regions incur different phase variations determined by the ToF along specific segments of the sound paths. Therefore, we can reconstruct the temperature distribution using the different phases changes of these line-of-sight (LoS) paths and reflected paths. In our experiments, VECTOR can measure the temperature along the sound path with an accuracy of 0.25◦C by sensing slight changes in the speed of sound, when the distance between two devices is known. Moreover, VECTOR incurs minimal hardware cost, as it can reuse the built-in audio systems that are already widely deployed in indoor environments or in cars. Compared with traditional temperature sensors, the sound-based scheme directly measures the temperature in the air instead of the temperature of the sensor. Therefore, VECTOR can detect human perceivable temperature fluctuations within a few seconds, while the latency of traditional sensor Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 3, Article 144. Publication date: September 2022
VECTOR:Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices.144:3 TV姓 HVAC contro Cold Aisle fot Aisle Cold area (a)In-car temperature-field reconstruction us- (b)Providing feedback for DC energy man- (c)Smart home/buildings with flexible tem- ing sound velocity along multiple paths. agement systems. perature demands. Fig.1.Application scenarios. is at a scale of tens of seconds.The sensitivity of VECTOR allows next-generation HVAC systems to recognize different types of heat sources and take timely reactions.With an array of distributed acoustic devices,we can reconstruct the temperature distribution with comparable resolution to infrared cameras,as shown in Fig.7(c). We face three key technical challenges when developing VECTOR.The first challenge is to precisely measure the ToF of the sound signal along each paths.For a 60 cm path,temperature change of 1C at room temperature of 25C yields a 3.02 us difference in ToF,which is far less than the 20 us sampling interval of widely used sound sampling frequency of 48 kHz on commercial devices.To achieve precise ToF measurement,we design an Orthogonal Frequency-Division Multiplexing(OFDM)sensing signal that can measure both the coarse- grained cross-correlation estimation and the fine-grained phase estimation.Our coarse-grained correlation scheme measures ToF at the sampling interval level(20us),while fine-grained phase estimation achieves sub- microsecond time resolution using the phase of the carrier frequency at 19 kHz.By removing the ambiguity of phase measurement using the coarse-grained correlation results,we can achieve a ToF accuracy of 0.371us,which is enough to capture temperature change of 0.12C along a 60 cm path.The second challenge is to reconstruct the temperature distribution using the ToF measurements.Intuitively,traditional Radon transform measures the signal attenuation from multiple angles to reconstruct the image of the object [16]and we can reuse it onto our cause to reconstruct the temperature distribution in the same manner.However,the ToF is reciprocally related to the speed of the sound and the temperature so that our physical model is different to the traditional Radon transform.To address this challenge,we propose the dRandon transform algorithm by transforming the temperature term using Taylor series expansion and use the relative phase changes to reconstruct the temperature distribution.The third challenge is to reconstruct the temperature distribution with limited acoustic devices.In real-world scenarios,we cannot get the acoustic paths in all desired angles with a small number of devices.To address this challenge,we utilize reflected paths to increase the number of phase measurements and train a linear model to reconstruct the temperature distribution as shown in Fig.1(a). Our experimental results show that VECTOR can measure the temperature on the LOS path with an error of 0.25C and reconstruct the temperature distribution with a decimeter-level spatial resolution.By monitoring multiple reflection paths in a car,VECTOR can measure distinct temperatures of all four seats with an average error of 0.44C using only one pair of devices. 2 MOTIVATION AND APPLICATION SCENARIOS 2.1 Motivation The motivation of using acoustic signal as the medium of temperature sensing is twofold: Proc.ACM Interact.Mob.Wearable Ubiquitous Technol.,Vol.6,No.3,Article 144.Publication date:September 2022
VECTOR: Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices • 144:3 Hot area Cold area Device A Device B (a) In-car temperature-field reconstruction using sound velocity along multiple paths. Cold Aisle Cold Aisle Hot Aisle Hot Aisle Microphones Microphones Speakers Speakers (b) Providing feedback for DC energy management systems. TV set Voice Assistant Voice Assistant TV set Cold area Cold area Hot area Hot area HVAC control HVAC control FeedBack FeedBack (c) Smart home/buildings with flexible temperature demands. Fig. 1. Application scenarios. is at a scale of tens of seconds. The sensitivity of VECTOR allows next-generation HVAC systems to recognize different types of heat sources and take timely reactions. With an array of distributed acoustic devices, we can reconstruct the temperature distribution with comparable resolution to infrared cameras, as shown in Fig. 7(c). We face three key technical challenges when developing VECTOR. The first challenge is to precisely measure the ToF of the sound signal along each paths. For a 60 𝑐𝑚 path, temperature change of 1 ◦C at room temperature of 25◦C yields a 3.02 𝜇𝑠 difference in ToF, which is far less than the 20 𝜇𝑠 sampling interval of widely used sound sampling frequency of 48 𝑘𝐻𝑧 on commercial devices. To achieve precise ToF measurement, we design an Orthogonal Frequency-Division Multiplexing (OFDM) sensing signal that can measure both the coarsegrained cross-correlation estimation and the fine-grained phase estimation. Our coarse-grained correlation scheme measures ToF at the sampling interval level (20𝜇𝑠), while fine-grained phase estimation achieves submicrosecond time resolution using the phase of the carrier frequency at 19 𝑘𝐻𝑧. By removing the ambiguity of phase measurement using the coarse-grained correlation results, we can achieve a ToF accuracy of 0.371𝜇𝑠, which is enough to capture temperature change of 0.12◦C along a 60 𝑐𝑚 path. The second challenge is to reconstruct the temperature distribution using the ToF measurements. Intuitively, traditional Radon transform measures the signal attenuation from multiple angles to reconstruct the image of the object [16] and we can reuse it onto our cause to reconstruct the temperature distribution in the same manner. However, the ToF is reciprocally related to the speed of the sound and the temperature so that our physical model is different to the traditional Radon transform. To address this challenge, we propose the dRandon transform algorithm by transforming the temperature term using Taylor series expansion and use the relative phase changes to reconstruct the temperature distribution. The third challenge is to reconstruct the temperature distribution with limited acoustic devices. In real-world scenarios, we cannot get the acoustic paths in all desired angles with a small number of devices. To address this challenge, we utilize reflected paths to increase the number of phase measurements and train a linear model to reconstruct the temperature distribution as shown in Fig. 1(a). Our experimental results show that VECTOR can measure the temperature on the LOS path with an error of 0.25◦C and reconstruct the temperature distribution with a decimeter-level spatial resolution. By monitoring multiple reflection paths in a car, VECTOR can measure distinct temperatures of all four seats with an average error of 0.44◦C using only one pair of devices. 2 MOTIVATION AND APPLICATION SCENARIOS 2.1 Motivation The motivation of using acoustic signal as the medium of temperature sensing is twofold: Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 3, Article 144. Publication date: September 2022
144:4·Wan et al. Performance:The key performance advantages of acoustic temperature sensing are in its low latency and long-range measurement capabilities.First,the speed of acoustic signal is physically determined by the air temperature,which incurs almost no delay.In many scenarios,air temperature could change quickly due to air conditioning,sunshine,or workload of servers in data centers,where such change could be captured by acoustic sensing.In contrast,traditional temperature sensors measure the temperature of their sensor probes,which could be different from the air temperature due to thermal conduction process.Second, with acoustic sensing,we can measure the average temperature along a long distance and use distributed devices to cover a large indoor area.Traditional sensors can only capture the temperature of a given point, and they are often placed on the wall or near the roof that are far way from the target region. Cost:Sound devices are ubiquitous in daily life,e.g.,voice assistants in home and car,speakers and microphones for electronic devices,so that acoustic sensing can reach the region of interest(living or working area)with no extra cost.Deploying traditional temperature sensors to provide acceptable coverage of the target regions may incur extra cost and inconvenience for daily activities.Therefore,our acoustic sensing method provides a low-cost solution to upgrade the temperature measurement performance. 2.2 Application Scenarios With the advantages of line coverage,low latency,and low cost,VECTOR can enable the following new application scenarios: In-car temperature sensor:In future smart vehicles,the electronics would make up 35%of a car's cost [17],which also make the wiring harness replacement/repair labor and cost skyrocket [18].VECTOR can reuse the built-in microphone and speaker in the car to provide fine-grained temperature readings,as shown in Fig.1(a).By measuring temperatures of individual seats and react to thermal condition changes with low-latency,VECTOR can improve the thermal comfort of occupancy with no extra hardware cost Furthermore,replacing traditional sensors with VECTOR by existing acoustic hardware can largely reduce the cost for both manufacturers and customers. Front-end of data center thermal management system:Data Centers(DCs)consume oceans of energy, e.g.,in 2014,DCs in U.S.consumed 1.8%of country's electricity consumption and 40%of the energy is used for temperature management[19].In Singapore,this ratio was 7%due to the tropical climate [20].There are a series of research works in both academia [21-23]and industry [24]in designing the HAVC control systems in DCs using temperature sensors as the feedback signal.As shown in Fig.1(b),VECTOR can provide the air temperature for multiple hot/cold aisles and racks with lower feedback latency and higher granularity compared with traditional sensors.Generally,feedback with lower latency can reduce the response time for a control system [25]and given the huge energy consumption of DCs,shorter response time means saving more energy. Cooperation with smart home/buildings:Thermal design of smart home/buildings aims at providing thermal demand flexibility with least energy consumption [26].Heating power loss coefficient(HPLC) are used to evaluate the thermal efficient of houses [27]and VECTOR can detect the heating sources without connection to the heating device and provide the heating periods data required by HPLC [26,27]. In addition,VECTOR can be easily integrated with existing thermal efficiency systems such as Google Nest Thermostats[28].Another important research issue for smart buildings is demand flexibility for thermal comfort [29-31],where VECTOR can provide temperature management systems with the temperature distribution using a small number of existing acoustic devices,as shown in Fig.1(c). Proc.ACM Interact.Mob.Wearable Ubiquitous Technol.,Vol.6,No.3,Article 144.Publication date:September 2022
144:4 • Wan et al. • Performance: The key performance advantages of acoustic temperature sensing are in its low latency and long-range measurement capabilities. First, the speed of acoustic signal is physically determined by the air temperature, which incurs almost no delay. In many scenarios, air temperature could change quickly due to air conditioning, sunshine, or workload of servers in data centers, where such change could be captured by acoustic sensing. In contrast, traditional temperature sensors measure the temperature of their sensor probes, which could be different from the air temperature due to thermal conduction process. Second, with acoustic sensing, we can measure the average temperature along a long distance and use distributed devices to cover a large indoor area. Traditional sensors can only capture the temperature of a given point, and they are often placed on the wall or near the roof that are far way from the target region. • Cost: Sound devices are ubiquitous in daily life, e.g., voice assistants in home and car, speakers and microphones for electronic devices, so that acoustic sensing can reach the region of interest (living or working area) with no extra cost. Deploying traditional temperature sensors to provide acceptable coverage of the target regions may incur extra cost and inconvenience for daily activities. Therefore, our acoustic sensing method provides a low-cost solution to upgrade the temperature measurement performance. 2.2 Application Scenarios With the advantages of line coverage, low latency, and low cost, VECTOR can enable the following new application scenarios: • In-car temperature sensor: In future smart vehicles, the electronics would make up 35% of a car’s cost [17], which also make the wiring harness replacement/repair labor and cost skyrocket [18]. VECTOR can reuse the built-in microphone and speaker in the car to provide fine-grained temperature readings, as shown in Fig. 1(a). By measuring temperatures of individual seats and react to thermal condition changes with low-latency, VECTOR can improve the thermal comfort of occupancy with no extra hardware cost. Furthermore, replacing traditional sensors with VECTOR by existing acoustic hardware can largely reduce the cost for both manufacturers and customers. • Front-end of data center thermal management system: Data Centers (DCs) consume oceans of energy, e.g., in 2014, DCs in U.S. consumed 1.8% of country’s electricity consumption and 40% of the energy is used for temperature management [19]. In Singapore, this ratio was 7% due to the tropical climate [20]. There are a series of research works in both academia [21–23] and industry [24] in designing the HAVC control systems in DCs using temperature sensors as the feedback signal. As shown in Fig. 1(b), VECTOR can provide the air temperature for multiple hot/cold aisles and racks with lower feedback latency and higher granularity compared with traditional sensors. Generally, feedback with lower latency can reduce the response time for a control system [25] and given the huge energy consumption of DCs, shorter response time means saving more energy. • Cooperation with smart home/buildings: Thermal design of smart home/buildings aims at providing thermal demand flexibility with least energy consumption [26]. Heating power loss coefficient (HPLC) are used to evaluate the thermal efficient of houses [27] and VECTOR can detect the heating sources without connection to the heating device and provide the heating periods data required by HPLC [26, 27]. In addition, VECTOR can be easily integrated with existing thermal efficiency systems such as Google Nest Thermostats [28]. Another important research issue for smart buildings is demand flexibility for thermal comfort [29–31], where VECTOR can provide temperature management systems with the temperature distribution using a small number of existing acoustic devices, as shown in Fig. 1(c). Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 3, Article 144. Publication date: September 2022
VECTOR:Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices.144:5 20 -12 2C同 1000 500 cy (Hz) 500 1000 20000 10000 020 150017形o20d (a)Baseband signal for odd subcarriers in (b)Modulated odd subcarriers in frequency (c)CIR of odd subcarriers with rich environ- frequency domain. domain mental multipaths. Fig.2.Key intermediate steps of signal processing. 3 SIGNAL DESIGN In the section,we first introduce the physical relationship between temperature and the speed of sound.We then design an OFDM sound signal to accurately measure the ToF along a sound path to derive the temperature. 3.1 Background The speed of sound in the air depends on environmental variables such as temperature,humidity,and air pressure.Within the normal room temperature range,the speed of sound can be approximated as c=331.3+0.606×T m/s, (1) where the temperature T is in degrees Celsius(C).While there are better approximations that relate the speed of sound to both temperature and air pressure [32],we use Eq.(1)as it is accurate enough for our system. We observe that the speed of sound increases by around 0.2%when the air temperature raises by one degree Celsius at room temperature.As an example,for two devices that are separated by a distance of 60 cm,the ToF measurement will decrease by a small amount of 3.02 us based on Eq.(1).Under the widely supported sampling rate of 48 kHz for sound playing/recording,the interval between consecutive samples is 20.8us,which is far greater than the small change in ToF.Therefore,traditional correlation-based ranging schemes cannot reliably detect such small changes in ToF,which is less than the sampling interval.To this end,we use an OFDM modulated signal to capture both the coarse-grained cross-correlation measure and the fine-grained phase measurement to detect microsecond-level changes in ToF. Phase-based ToF measurement provides high-resolution and reliable ToF results.The phase change for a specific path p,p,is related to the speed of sound by p=-2dpfe/c,where dp is the length of the path and fe is the carrier frequency of the signal.In the following discussion,we use the carrier frequency of fe=19 kHz if not specified.For two devices that are separated by a distance of 60 cm,the phase change will decrease by an amount of 0.360 in radian when the temperature raises by one degree at room temperature.Such phase increase can be reliably measured using OFDM signals [33].However,as phase changes are limited in the range of 0~2m, it cannot determine whether the phase changes by 0.5 or 2.5.We use coarse-grained cross-correlation results to resolve the ambiguity in phase measurements.As a phase change of 2m at 19 kHz carrier is equivalent to 52.6 us in ToF,we can use the cross-correlation result that has a resolution of 20.8 us to determine how many 2m the phase has been changed.Therefore,we design an OFDM signal that can measure both the phase and the cross-correlation offset at the same time. Proc.ACM Interact.Mob.Wearable Ubiquitous Technol.,Vol.6,No.3,Article 144.Publication date:September 2022
VECTOR: Velocity Based Temperature-field Monitoring with Distributed Acoustic Devices • 144:5 -1000 -500 0 500 1000 Frequency (Hz) 0.5 1.0 1.5 2.0 Magnitude ZCodd[n] Zero at central frequency bin. (a) Baseband signal for odd subcarriers in frequency domain. -20000 -10000 0 10000 20000 Frequency (Hz) 0.5 1.0 1.5 2.0 Magnitude ZC ZCodd[n] ∗ odd[n] (b) Modulated odd subcarriers in frequency domain. 0 250 500 750 1000 1250 1500 1750 2000 Sample Points 0.0 0.2 0.4 0.6 0.8 1.0 Normalized Magnitude ϕpositive = 1.714 ϕnegative = −1.427 (c) CIR of odd subcarriers with rich environmental multipaths. Fig. 2. Key intermediate steps of signal processing. 3 SIGNAL DESIGN In the section, we first introduce the physical relationship between temperature and the speed of sound. We then design an OFDM sound signal to accurately measure the ToF along a sound path to derive the temperature. 3.1 Background The speed of sound in the air depends on environmental variables such as temperature, humidity, and air pressure. Within the normal room temperature range, the speed of sound can be approximated as 𝑐 = 331.3 + 0.606 ×𝑇 𝑚/𝑠, (1) where the temperature 𝑇 is in degrees Celsius (◦C). While there are better approximations that relate the speed of sound to both temperature and air pressure [32], we use Eq. (1) as it is accurate enough for our system. We observe that the speed of sound increases by around 0.2% when the air temperature raises by one degree Celsius at room temperature. As an example, for two devices that are separated by a distance of 60 𝑐𝑚, the ToF measurement will decrease by a small amount of 3.02 𝜇𝑠 based on Eq. (1). Under the widely supported sampling rate of 48 𝑘𝐻𝑧 for sound playing/recording, the interval between consecutive samples is 20.8𝜇𝑠, which is far greater than the small change in ToF. Therefore, traditional correlation-based ranging schemes cannot reliably detect such small changes in ToF, which is less than the sampling interval. To this end, we use an OFDM modulated signal to capture both the coarse-grained cross-correlation measure and the fine-grained phase measurement to detect microsecond-level changes in ToF. Phase-based ToF measurement provides high-resolution and reliable ToF results. The phase change for a specific path 𝑝, 𝜙𝑝 , is related to the speed of sound by 𝜙𝑝 = −2𝜋𝑑𝑝 𝑓𝑐/𝑐, where 𝑑𝑝 is the length of the path and 𝑓𝑐 is the carrier frequency of the signal. In the following discussion, we use the carrier frequency of 𝑓𝑐 = 19 𝑘𝐻𝑧 if not specified. For two devices that are separated by a distance of 60 𝑐𝑚, the phase change will decrease by an amount of 0.360 in radian when the temperature raises by one degree at room temperature. Such phase increase can be reliably measured using OFDM signals [33]. However, as phase changes are limited in the range of 0 ∼ 2𝜋, it cannot determine whether the phase changes by 0.5𝜋 or 2.5𝜋. We use coarse-grained cross-correlation results to resolve the ambiguity in phase measurements. As a phase change of 2𝜋 at 19 𝑘𝐻𝑧 carrier is equivalent to 52.6 𝜇𝑠 in ToF, we can use the cross-correlation result that has a resolution of 20.8 𝜇𝑠 to determine how many 2𝜋 the phase has been changed. Therefore, we design an OFDM signal that can measure both the phase and the cross-correlation offset at the same time. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol., Vol. 6, No. 3, Article 144. Publication date: September 2022