Air Quality Monitoring with Mosaic: Revisiting sensornets for practical applications. Dr. Wei Dong Zhejiang University

Air Quality Monitoring with Mosaic: Revisiting sensornets for practical applications Dr. Wei Dong Zhejiang University dongw@zju.edu.cn 1

Self introduction Associate Professor in ZJU Leads the Embedded Networked Systems group (EmNets) http://www.emnets.org Research interests IoT and sensornets Wireless and mobile computing Network measurement Homepage: http://www.emnets.org/dongw 2

2000 David Culler: Prof of UCB [ASPLOS00] System architecture directions for networked sensors (citation: 4404) sensors CPU radio battery 3

2002 Ian F. Akyildiz: Prof of GIT [COMNET12] Wireless sensor networks: a survey (citation: 14897) The sensornet vision: a large number of sensor nodes, densely deployed self-organizing capabilities In-network processing 4

IPSN & SenSys IPSN: Information Processing in Sensor Networks, 2002 Feng Zhao, MSR SenSys: Embedded Networked Sensor Systems, 2003 Ian Akyildiz, GIT Deborah Estrin, UCLA Victor Bahl, Sigmobile Representive Craig Partridge, Sigcomm Representive Taieb Znati, tznati@nsf.gov 5

Sensornet topics Comm & Networking MAC, routing, congestion Systems: Platforms, OS, programming, storage, simulation Energy Sensing & tracking Acoustic, mobile, and wireless sensing Time & location Deployment & apps 6

Topics of ACM SenSys 26 24 22 20 18 # of papers 16 14 12 10 8 6 4 2 0 03 04 05 06 07 08 09 10 11 12 13 14 15 Year Sys Comm Sensing Apps Proc Time & loc Energy 7

2010 Solved MAC protocols Dissemination Routing protocols Localization Data collection OS design Time sync Unsolved Energy management Security Network management Fault tolerance Programming models 8

Sensornet researchers David Culler John Stankovic Matt Welsh Phil Levis Kamin Prabal WhitehouseDutta Fred Jiang John Regehr Chenyang Lu Tian He Gang Zhou Kannan Srinivasan Guoliang Xing 9

Wireless researchers Dina Katabi Shin Kang Sachi Katti Shyam Gollakota Tarek Abdelzaher Xinyu Zhang 10

Cross-components research Battery & radio E.g. Ambient/WiFi backscatter Sensor & radio E.g. Wireless sensing New considerations Radio is costly? Tradeoff between comm., sensing, and energy? 11

Cross-disciplinary research Volcano monitoring Environment monitoring Water monitoring Air quality monitoring 12

Hangzhou, China

2014-05-24

AQI (Air Quality Index) Impacting factors:,, 2, 2, 2 AQI Levels of Health Concern Colors 0-50 Good Green 51-100 Moderate Yellow 101-150 Unhealthy for Sensitive Groups Orange 151 to 200 Unhealthy Red 201 to 300 Very Unhealthy Purple 301 to 500 Hazardous Maroon 15

PM (Particulate Matters) 16

Remote sensing using satellites 17

Measurement stations 18

UrbanAir: urbanair.msra.cn Yu Zheng, MSRA U-Air: When Urban Air Quality Inference Meets Big Data, KDD 2013 Forecasting Fine-Grained Air Quality Based on Big Data, KDD 2015 19

Monitoring using low-cost sensors? Pushing the Spatio-Temporal Resolution Limit of Urban Air Pollution Maps, PerCom 2014. Best Paper Mobile, dedicated high-cost sensors AirCloud: A Cloud-based Air-quality Monitoring System for Everyone, SenSys 2014. low-cost sensors, but not for mobile 20

Intel PAM 检测 PM2.5 PM10 使用 PPD42N 粉尘传感器 Arduino 本地处理后, 上传云端 云端收集所有 PAM 数据 将位置相近的数据统计为一类 在每一类中, 利用天气 温湿度 气压 海拔等信息对数据进行校准 21

墨迹天气空气果 分为室内版和室外版 室内版拥有温度 湿度 PM2.5 和 CO 2 传感器室外版拥有温度 湿度 PM2.5 和气压传感器 22

Mosaic PerCom 2014 Mobile, dedicated high-cost sensors SenSys 2014 low-cost sensors, but not for mobile => Can we design a low-cost sensors for mobile sensing? Mosaic: Mobile sensor networks for air quality monitoring in cities 23

Mosaic architecture 24

Mosaic node 25

Mosaic node 26

PM2.5 Sensor: PPD42NS Light-scattering Particles scatter light Light sensor senses scattered light Convert light sensor readings to particle concentration Stable airflow is essential for accurate sensing A 100Ω resistor generates heat for stable airflow Use Dylos for calibration 27

Key research problems 1. Accurate sensing How to achieve accurate sensing using PPD42NS in mobile environment? 2. Sensor deployment How to deploy sensors in mobile vehicles (buses) to achieve maximum coverage (utility)? 28

Key research problems 1. Accurate sensing How to achieve accurate sensing using PPD42NS in mobile environment? 2. Sensor deployment How to deploy sensors in mobile vehicles (buses) to achieve maximum coverage (utility)? 29

Accurate sensing: challenge 1 Challenge 1: unstable airflow on moving vehicles causes inaccurate readings 30

Accurate sensing: solution 1 Solution: Remove the resistor and use the relative airflow constructively Use speed and acceleration for data filtering 31

Accurate sensing: challenge 2 Challenge 2: ANN-based calibration is prone to go to local minima with finite data sets: it tends to fit the trained data well while has poor performance in predicting the new unseen data. 32

Accurate sensing: solution 2 Solution: Try to use Multi-SVMs as SVM-based model is less prone to over-fitting Classify the raw data into different discrete levels Combine Multi-SVMs and ANN ANN better SVM better 33

Accurate sensing: overall solution 34

Results ANN: 0.669 Ours: 0.753 35

Key research problems 1. Accurate sensing How to achieve accurate sensing using PPD42NS in mobile environment? 2. Sensor deployment How to deploy sensors in mobile vehicles (buses) to achieve maximum coverage (utility)? 36

Sensor deployment: formulation Divide the interested area into grids. Input Grids with their importance A set of buses and its trajectories # of selected buses Output A set of selected buses, i.e. # of nodes. Goal Maximize the total utility of covered grids 37

Sensor deployment: example 1 1/2 1/8 1/18 1/20 1 1/2 1/8 1/18 1/20 1/2 1/4 1/10 1/8 1/10 1/4 1/2 1/4 1/2 1/4 Route 1 1/2 1/4 1/10 1/8 1/10 1/4 1/2 1/4 1/2 1/4 1/2 1 1/2 1 1/2 1/2 1 1/2 1 1/2 1/4 1/2 1/4 1/2 1/4 1/4 1/2 1/4 1/2 1/4 Route 2 Route 3 38

Sensor deployment: results Comparison Random walk algorithm Evolutionary algorithm [JAIHC14] Mosaic 39

Lessons learnt Vibration-absorbing is necessary Pay attention to the shutting down of GPRS module Discard sampling data during start and stop of buses Pay attention to remote node errors String constants waste precious RAM 40

More info about Mosaic Mosaic: Towards City Scale Sensing with Mobile Sensor Networks, ICPADS 2015, invited paper. Mosaic: A Low-Cost Mobile Sensing System for Urban Air Quality Monitoring, INFOCOM 2016. 41

The future of Mosaic Hybrid deployment Online calibration E.g. Reducing Multi-Hop Calibration Errors in Large-Scale Mobile Sensor Networks, IPSN 2015, Best Paper. Air quality prediction, tracking 42

几个思考 : 关于硬件 1. Is the node hardware ready for practical apps? 硬件能力 摩尔定律预测的曲线 2004 传感器节点发展曲线 时间 43

Arduino: https://www.arduino.cc/ GitHub search results Keywords Repositories Arduino 46,940 Telos 110 Mica 496 Wall avoiding robot Make a UAV Spyplane Using the Arduino 44

关于系统软件 2. Is the system software reusable? TinyOS, Contiki, SenSpire OS, Arduino: setup() loop() Consideration Simple vs sophisticated? Specific vs general? Software verification CTP+FTSP CTP for high rate collection Sniffer 45

关于无线通信 3. Where is the design space for protocols? Sophisticated technology for practical apps Various new wireless comm systems? Full duplex Back scatter Visible light 46

Vertical thinking vs horizontal thinking 47

Vertical vs horizontal Horizontal thinking: Sensornet protocols QoF [INFOCOM11], STAIRS [INFOCOM14] CorModel [INFOCOM15], CARE [ICNP15] Vertical thinking: protocol stack? Vertical thinking: Link tomography Scalpel [ICNP14], OMA [ToN revision], k- identifiability [INFOCOM16] Horizontal thinking: can be applied elsewhere? 48

Short-term vs long-term 杭州天才少女 : 郭文景 百米冲刺马拉松接力赛 49

Thank you Q & A dongw@zju.edu.cn 50