Signals, Instruments, and Systems W12 Environmental Sensor. Real Deployments

Transcription

1 Signals, Instruments, and Systems W12 Environmental Sensor Networks Algorithms and Real Deployments 1

2 Outline Wireless sensor networks in the field: the Sensorscope project Main problems and baseline algorithms Practical issues Mobile sensor network for air quality monitoring: the OpenSense project Goals and research questions Lausanne and Zürich deployments 2

3 Motivation for Sensor Networks What if we could monitor events which have a large spatial and temporal distribution require in-situ measurements take place in hard to access places generate data which hneed dto be available in real-time 3

4 Motivation for Sensor Networks What would we need for that? A device which is cheap so we can distribute many of it is reliable so we can measure for a long time uses little power battery/solar cell powered has a radio so it can communicate can potentially move so it can potentially relocate 4

5 Building a Sensor Network: Key Concepts 5

6 Introduction Temperature Humidity Light 6

7 Topology? 7

8 Topology 8

9 Topology 9

10 Topology GPRS 10

11 Topology Pros Very simple! No restrictions in sensor locations Cons The server may be quite far from the stations A long-range g link may consume a lot of energy! 11

12 Topology 700! A]! 400 Power Consumption [m 14x times the XE Sensor MSP430 XE1205 GPRS 12

13 Topology 700! A]! 400 Power Consumption [m 14x times the XE Sensor MSP430 XE1205 GPRS 13

14 Topology Assuming four AA batteries, 1.2 V, 2000 mah Sensor: 167 days MSP430: 28 days Short range radio: 1.7 days Long range radio: 8 hours 14

15 Topology 15

16 Topology Short range GPRS Sink 16

17 Topology Friis law Example: To transmit over 5 Km we can using 868 MHz we can: One hop of 5 km: P L = 106 db Two hops of 2.5 km: P L = 99 db Five hops of 1 km: P L = 92 db Energy is the main issue!!! 17

18 Multi-hop WSNs GPRS 18

19 Multi-hop WSNs Pros Only one car battery in the network The WSN has extended monitoring coverage Multiple l routes for stations ti to communicate with the sink Auto configurable network (robustness) Cons Significantly more complicated Dt Data rate is not tincreased Unable to use directional antennas 19

20 Multi-hop WSNs Implementation: Neighborhood discovery Time synchronization Data routing Duty-cycling (radio management) 20

21 Neighborhood Hello messages (Beacons) are one common method: 1. Node A sends a HELLO message to its neighbors (B, C, and D). 2. Nodes B, C, and D update their neighborhood table. 3. Node B sends a HELLO message to its neighbors (A, C, and D). 4. B A D C 21

22 Neighborhood Hello messages (Beacons) are one common method: 1. Node A sends a HELLO message to its neighbors (B, C, and D). 2. Nodes B, C, and D update their neighborhood table. 3. Node B sends a HELLO message to its neighbors (A, C, and D). 4. A (A) B D (A) C (A) 22

23 Neighborhood Hello messages (Beacons) are one common method: 1. Node A sends a HELLO message to its neighbors (B, C, and D). 2. Nodes B, C, and D update their neighborhood table. 3. Node B sends a HELLO message to its neighbors (A, C, and D). 4. A (B) (A) B D (A, B) C (A, B) 23

24 Neighborhood Hello messages (Beacons) are one common method: 1. Node A sends a HELLO message to its neighbors (B, C, and D). 2. Nodes B, C, and D update their neighborhood table. 3. Node B sends a HELLO message to its neighbors (A, C, and D). 4. A (B, C, D) (A, C, D) B D (A, B, C) C (A, B, D) 24

25 Neighborhood What information do we need about our neighbors? Distance to sink Last time heard Link quality F 2h hops A H 3 hops B 1 hop G 2 hops E 2 hops C 1 hop D 1 hop 25

26 Neighborhood Node E s neighborhood table Id Age Distance Quality F 2 hops A B 2 min 1 hop 87% H 3 hops B 1 hop C 2 min 1 hop 98% F 4 min 2 hops 74% G 1 min 2 hops 93% G 2 hops E 2 hops C 1 hop D 1 hop A few remarks: Only the distance to the sink is stored. Neighborhood discovery can t be done only once! We need to estimate link qualities! 26

27 Neighborhood Variations of simple schema: Each node sends X beacons per minute. Number of beacons received per minute are stored. Quality is estimated over the past Y minutes by counting losses. t 4 t 3 t - 2 t Quality = 0.8 t 4 t 3 t - 2 t - 1 Example (X = 10;Y = 4): Quality = 0.71 t 4 t 3 t -2 t Quality =

28 Time Synchronization Weather conditions, especially temperature and humidity, may have a significant effect on hardware 8 Air tempe erature [ C] Oct Oct 31 Crystal oscillators are highly impacted by temperature! 28

29 Time Synchronization ] Air temp perature [ C Indoor Outdoor Freezer [ms] Time drift Day 29

30 Time Synchronization Nodes need to know the time to: Timestamp packets Synchronize actions (e.g., taking samples, transmitting data) How do we get time: Locally: Every node gets the time itself Distributed: Time is propagated from reference nodes 30

31 Time Synchronization Every node gets the time: Atomic clock receivers: Cheap (both energy and $) Complexity Limited coverage GPS: Coverage Complexity High cost (energy and $) GPRS: same as GPS with less coverage What about a distributed approach? 31

32 Time Synchronization F 2h hops A H 3 hops B 1 hop G 2 hops E 2 hops C 1 hop D 1 hop 32

33 Field Deployment of Sensor Networks: The SensorScope Project 33

34 SensorScope Microcontrollers are inside hermitically sealed boxes, attached hdon a mounting pole with up to seven external sensors. Price: $

35 SensorScope 35

36 SensorScope Shockfish h TinyNode with TinyOS 2.x MSP bit 8MHz 48KB ROM, 10KB RAM, and 512KB flash memory Semtech XE1205 radio 868MHz, 76Kbps 36

37 SensorScope 162x140mm 140 solar panel 12Ah NiMH rechargeable battery 37

38 38

39 SensorScope Many previous successful deployments 97 stations ti deployed d at EPFL (one year) 39

40 SensorScope Many previous successful deployments 16 stations deployed at Le Génépi to monitor conditions leading to dangerous mudslides (two months) 40

41 Mobile Sensor Networks The OpenSense Project 41

42 OpenSense Community-driven, large-scale air pollution measurement in urban environments Mobile sensors (parasitic, uncontrolled mobility) on public transportation vehicles Sensorscope Static wireless sensing and communication infrastructure Permasense 42

43 Motivation Urban population will double in next decades > 50% of world population already lives in cities rural population expected to stagnate or drop Urban air pollution 2% of all deaths (1.2 million people) 0.6% of burden of disease (DALY) World Urbanization Prospects, U. N Global Health Risks, WHO

44 Motivation Air pollution is highly locationdependent traffic chokepoints urban canyons industrial installations Fine resolution air quality data is needed! Enabling research in: Public service & education Human exposure enable private users to make Air Pollution Engineering informed decisions Urban Planning raising popular awareness Environmental Justice Public Policy 44

45 Traditional Air Monitoring Systems Sparse networks of ground stations Example: Switzerland s NABEL ( Station locations 16 stations specially selected sites urban with traffic urban residential suburban rural, etc. resolution: high temporal low spatial Ozone concentration Mission: monitor air pollution on national level & gauge impact of environmental policies Public data access: 45

46 Traditional Air Monitoring Systems Satellite-based remote sensing Examples: Measurements of Pollution in the Troposphere (MOPITT on Terra satellite) Ozone Measurement Instrument (OMI on Aura satellite) Features: daily scans large coverage homogeneous quality sensitive to cloud coverage low resolution 46

47 Proposed System mobile sensor network parasitic mobility: anchored to existing mobility sources - vehicle energy supply public transport - predictable mobility - single point maintenance low-cost, light-weight chemical sensors (CO, CO 2, NO 2, O 3 ) intelligent integration & control to mitigate demanding constraints 47

48 Proposed System NANO SENSING SYSTEM From many wireless, mobile, heterogeneous, unreliable raw measurements INFORMATION SYSTEM to reliable, understandable and Web accessible real time information TERA mobile nodes static nodes Nabel station Zürich GPRS GPS sensor network control optimization of data acquisition information dissemination 48

49 Value of Dense Measurements Traditional approach Recent results Few stations Low resolution interpolated estimates of pollutant concentrations across massive regions Massive deployment of stations (150) at street-level (2008/2009 New York City Community Air Quality Survey) Pollutants of interest heavily concentrated along roads with high traffic densities 49

50 Challenges Global questions: More data, more noise, but also more redundancy Can we produce better quality data? Case study for other environmental phenomena: Radiation, noise, energy Research directions: Wireless Sensor Network control Community sensing When/Where to sample? privacy protection What/To whom to transmit? trustworthiness of data, relevance of data gathered and Sensor Node design information produced Sampling System Modeling Localization sensor, device and mobility models Software & hardware architecture air quality models Mechanical integration privacy, trust & activity models 50

51 Gas Sampling System Problem: Chemical sensors have very slow dynamics (example: Telaire 6613 CO2 sensor step response <2min) Open sampling Closed sampling sensors directly exposed to environmental measurand Benefits: simple & slim solution continuous sampling Drawbacks: no absolute concentration values noisy signal (sensitive to environment variations: pressure, humidity) Typical response: sensors exposed to measurand inside controlled chamber 3-phase strategy Benefits: absolute measurements noise due to environment filtered Drawbacks: complex & bulky non-continuous sampling Typical response: [Lochmatter 2010] [Trincavelli 2010] Idea: Combine these two approaches to get the benefits of both systems. 51

52 Gas Sampling System Current deployment passive active open controlled flow [Lochmatter et al. 2010] closed unclean air clean air [Gonzalez-Jimenez et al. 2011] Smart sampling module possibly hybrid single/multi-chamber wind sensing Anemometer [Lochmatter et al. 2010] 52

53 Localization Robust localization prerequisite for adaptive control exploits commercial state of the art u-blox LEA-6R GPS + dead reckoning (DR) module augmentation with additional sensor modalities (compass, accelerometer, gyro) GPS only GPS + DR 53

54 Localization large set of rich data: location parameters (geographical coordinates, heading, odometer, speed, acceleration etc.) vehicle context data Next stop: Sallaz doors open Current stop: Sallaz Next stop: Valmont 54

55 Deployment Status Lausanne 3 mobile plus 2 static stations Measured parameters NO2, CO, CO2, O3, relative humidity, temperature, location Power Solar panel (stationary stations) Bus power (mobile station) Data Transmission via GPRS to a central server 10 additional mobile stations to be deployed over summer Static stations near local NABEL roadside station necessary for sensor calibration Station 1059 since September 2011 Station 1215 since March NABEL station on Cesar Roux 55

56 Deployment Status Lausanne Mobile stations (on buses) since June 2011 testing mechanical & electronic integration failed in November 2012 currently online since March 2013 improved mechanical design includes enhanced localization module 56

57 Electric Car Deployment 100% electric > no self contamination of measurements flexible mobility in the short to medium run: testbed for multiple sampler designs in parallel (final passive sampler, active sniffer, wind sensing etc.) targeted investigation tool (increasing sample rate on particular links, stop/measure/go scenarios etc.) i th in the llong run OpenSense O S supernode calibrating other nodes controlled decrease of field uncertainty at key points 57

58 Electric Car Deployment Speed pulse signal directly from rear wheel sensor Sensor box & GPS antenna on roof Loggers & localization secured in the boot Laptop mount for efficient field work 58

59 Deployment Status Zürich Static station by NABEL in Dübendorf 1 station (varying sensors) Long-term ozone sensor testing (since April 2011) Testing new sensors (e.g., combined CO/NO2 sensor) Mobile deployments On top of LuftiBus (Lungenliga) 1 station (O 3 and UFP sensors) Installed in March 2013, covers whole Switzerland On top of trams in Zurich 10 stations (O 3, CO, and UFP sensors) > 1 year of measurements and d30mi Mio data points 59

60 Open Data Access Lausanne Zürich

61 Conclusion 61

62 Take Home Messages Sensor networks enable environmental monitoring in remote locations and of difficult-to-measure processes Real-world deployments may be highly unpredictable! Mobile sensor networks can increase coverage and spatial resolution of measured data Increasing the resolution of air pollution data is necessary for understanding health impact. Whether data extracted from poor quality measurements can be processed to obtain useful data on air pollution is an important research question. Other questions: How to design the node? How to control the network? Using existing mobility sources holds important benefits, but achieving a reliable system integration is non-trivial. 62