Wireless sensor networks for environmental monitoring

Download Report

Transcript Wireless sensor networks for environmental monitoring

Wireless sensor networks for
environmental monitoring
Kris Steenhaut
Yann-Aël Le Borgne
ICCE 2010
11th of August, 2010
0
Agenda
1) Technology and applications
2) Information processing for WSN
3) Case studies
1
Part 1:
Technology
and
Applications
2
Wireless Sensor Networks
Wireless sensors: Latest trend of Moore’s law (1965)
 Computing devices get
 Smaller
 Cheaper
3
Smart Dust (Berkeley, 2000)
Image © 2000 Peter Menzel/Robo sapiens
4
Wireless Sensor Networks
Sensor nodes can collect, process and communicate data
[Warneke et al., 2001; Akyildiz et al., 2002]
 Sensors: Light, temperature,
humidity, pressure, acceleration
 Radio: ~100 kbps, ~10 meters
 Microprocessor: ~10 MHz
 Memory: ~100 KB
5
Wireless Sensor Networks
 Applications: Medical, interactive arts, ecology,
industry, disaster prevention....
6
Research trend 2000-2009
7
Network stack
8
Physical layer: Hardware
Muneeb Ali,9 2008
Physical layer: Hardware
CPU
Memory
Radio
Mica-2
Tmote Sky
Imote2
Waspmote
2002
2005
2007
2009
ATmega128L
TI MSP430
Intel PXA271
ATmega1281
8-bit, 8MHz
16-bit, 8MHz
32-bit, 13-416MHz
8MHz
36 µW asleep
15 µW asleep
390 µW asleep
2 µW asleep
60 mW active
5.4 mW active
>31 mW active
27 mW active
4KB RAM
10KB RAM
32 MB RAM
8 KB RAM
128 KB Flash
48 KB Flash
32 MB Flash
128 KB Flash
CC1000
CC2420
Xbee-802.15.4
76 Kbps
250 Kbps
250 Kbps
100 µW sleep
60 µW sleep
<30 µW sleep
36 mW receive
63 mW receive
150 mW receive
75 mW xmit
57 mW xmit
135 mW xmit
2ms setup
1ms setup
2ms setup
10
Companies
Crossbow
TelosB, MICAZ, Eko nodes
Sentilla (formerly MoteIV)
Tmote, JCreate
Sensinode
Nanostack and nanorouter
Libelium
Waspmotes and Meshlium.
11
Network stack
12
Great Duck Island (2002)
Photo: Peg Skorpinski
Photo: Peter Scott
13
Tungurahua (2005)
http://fiji.eecs.harvard.edu/Volcano
14
Ocean Tracking Network
(2010)
15
http://oceantrackingnetwork.org/
Network stack
16
Challenges: Energy
17
Challenges: Energy
The radio is the most energy consuming module.
If run continuously with the radio, the lifetime is about 5 days.
18
Challenges: Wireless
19
Research Challenges
Time synchronisation
Multihop routing
Aggregation
Sleeping...
20
Network stack
21
Optimizing the MAC layer
Communication unit -> put to sleep if no
communication needed
MAC protocol -> enables communication
between neighbor nodes
22
MAC layer
XMAC: Nodes’ radio unit periodically wakes
up for short period (non synchronised)
Poll is for me, I send an
Ack and wait for data
B has data for A
Poll is not for me,
I go back to sleep
23
MAC layer
Classic paper S-MAC (UCLA)
24
MAC layer
25
MAC layer
26
Read : MAC survey by Koen Langendoen
Multiple channels
If there’s a lot of traffic on the 802.11
wireless links, a lot of channels on the
802.15.4 spectrum will be unavailable
27
Multiple Channels?
Multiple channels  more bandwidth
Parallel communications without
interference
28
MuChMAC design
Frequency hopping scheme
Dynamic hopping scheme
Supports Broadcast & Unicast
Strict time synchronization??
Can be relaxed!
29
+ Low power
Duty cycle
+ TDM (random wake-up moments in slot)
30
Drift compensation
31
Network stack
32
Routing layer
Routing tree
Standard approach:
‘min’ route
33
Routing layer
34
Directed diffusion
Data centric routing based on Interests
Type=temperature
Location=[0,0,15,35]
Time= 00:00:00, 10:00:00
Interval=30s
Id=324
Type=temperature
Value=25.3°C
Location=[10,25]
Time=04:23:30
Id=323
Type=temperature
Value=23.4°C
Location=[40,25]
Time=04:23:30
35
Directed diffusion
a) Query dissemination
b) Gradient setup
c) Path reinforcement
UCLA, 2000
36
TinyDB (Berkeley,2002)
SQL-based querying
SELECT temperature FROM sensors
WHERE location=[0,0,15,35]
DURATION= 00:00:00, 10:00:00
EPOCH DURATION 30s
Id=324
Type=temperature
Value=25.3°C
Location=[10,25]
Time=04:23:30
Id=323
Type=temperature
Value=23.4°C
Location=[40,25]
Time=04:23:30
37
Tiny Aggregation
1
2
3
38
Berkeley, 2005
Tiny Aggregation
Aggregation: Fuse data along the routing tree
Example: Average of the measurements
SELECT AVG(temperature) FROM sensors
WHERE location=[0,0,15,35]
DURATION= 00:00:00, 10:00:00
EPOCH DURATION 30s
39
Data aggregation
 Let si[t] be the measurement of node i at time t, 1<i<S
 Average:
40
Network load
41
Network load
42
Data aggregation
Advantages of aggregation
Reduce communication
Network load does not depend on network size:
scalability
Changes load distribution: root node is no
longer the bottleneck
43
Data aggregation
Operators:
Count, sum, average, min, max (SQL operators,
[Madden et al, 2005])
Distributed regression [Guestrin et al, 2004]
Principal component analysis [Le Borgne et al., 2009]
Middleware:
Tiny diffusion [Intanagonwiwat et al, 2000],
Tiny Aggregation (TAG) [Madden et al, 2005],
Dozer [Burri, 2007], ...
44
Operating systems
TinyOS
Berkeley (1999)
NesC language
Based on components
Contiki
SICS (2003)
C language
Based on protothreads
45
Part 2:
Information processing
for WSN
46
Machine learning
47
Machine learning:
methodology
48
Modeling sensor data
49
The model approximates
Modeling sensor data
the measurements with just
one parameter
50
Replicated models
51
Replicated models
52
Replicated models
53
Replicated models
54
Replicated models
55
Replicated models
Pros
Guarantees the observer with ɛ accuracy
Simple or complex models can be used
Cons
In most cases, no a priori information available.
Which model to choose?
56
Replicated models
57
Racing
At first, all models are in competition.
58
Racing
As time passes, h1 statistically outperforms h6.
59
Racing
h3 then outperforms h5.
60
Racing
h3 is finally selected as the best one.
61
Replicated models
Communication savings depends on accuracy
In practice, 50% communication savings for high
accuracy. Up to 95% savings for rough
approximations.
62
Replicated models
Summary
Send model parameters instead of
measurements
Guarantees accuracy and communication
savings
Very low compuational overhead. Implemented
in TinyOS for Tmote Sky.
63
Principal component aggregation
64
Routing tree
65
Data collection
66
Data collection
67
Data collection
68
Data collection
69
Data collection
Communication is much more important at
nodes close to the base station
The batteries of these nodes expire first
The root node is the bottleneck
Rest of the network disconnected!
70
Data aggregation
 Solution: Remember aggregation
 Example: average
71
Beyond average:
Principal component analysis
72
Principal component analysis
PCA: Versatile technique for:
Compression,
Noise filtering,
Event detection,
Event classification
Particularly appropriate when data are
correlated (as in sensor networks)
73
Principal component analysis
No noise
SNR=1
 3 phenomena, appearing
and disappearing
 The network is a 10*10 grid
of nodes
QuickTime™ and a
decompressor
are needed to see this picture.
QuickTime™ and a
decompressor
are needed to see this picture.
1 PC
QuickTime™ and a
decompressor
are needed to see this picture.
3 PCs
QuickTime™ and a
decompressor
are needed to see this picture.
 Dramatic reduction of dimensionality! Here, S=100, q=3
 Performs noise filtering
100 PCs
QuickTime™ and a
decompressor
are needed to see this picture.
74
Principal component analysis
75
Principal component
aggregation
76
Principal component
aggregation
77
Intel Laboratory monitoring
54 sensors recording temperature, humidity and light
2 months deployment.
78
Intel Laboratory monitoring
79
Intel Laboratory monitoring
80
Intel Laboratory monitoring
81
Intel Laboratory monitoring
82
Part 3:
Case studies
ETRO/VUB
83
Areas of Competence
Environmental sound monitoring
Data Aggregation
Embedded computing
OS porting – SW-HW tools
MAC and Routing protocols
Reconfigurable computing
Low power design
Contiki – Tiny OS
Cross layer optimization for
low power
wireless sensor nodes
Enabler for novel applications
84
Current project:
IWT-SBO project
Intelligent
Distributed
Environmental
Assessment
85
IDEA
brings
More information
at a cheaper price.
air (UFP)
tiny noise
reference
noise
86
IDEA
brings
Bottom up: alerting functionality
and instant validation
sensor error
car
truck
train
nature
talking
87
industry
IDEA
brings
Top down:
interpolation and detailed querying
i
LAden = 64 dBA
details
origin
PM10 = 34  g/m3
details
origin
UFP = ...
details
origin
Origin:
Quality : medium
Interpolation based on traffic data
Closest measurement : 120 m
Closest high-end : 890 m
Time stamp : 1-2-2009 to 1-3-2009
...
88
IDEA
brings
Top down:
interpolation and detailed querying
Tracing sensor details
Start: 6/3/2009 11:22
Location: //serv1.gent.be/tr23.xml
Tracing sensor details
Start: 6/3/2009 11:23
Location: //serv1.gent.be/tr24.xml
89
IDEA: Main work areas
 Work will concentrate on 3 main work
areas:
1. Hardware level enhancements for mobile
wireless sensor nodes
2. Network level enhancements and protocol
design for self organizing wireless networks
3. Setup of a demonstrator as proof of concept
90
Hardware Enhancements
 Single Board Computer, minimum
processing capability, minimum price/node
 System runs Windows Embedded,
Windows XP or Linux (Voyage: slimmed
down Debian)
91
Outdoor tests of microphones
92
Self-Testing microphone
Mic + circuitry
Schematics of aluminum casing,
finished parts and finished product.
Micro Speaker
Microphone design
with screw able cap,
10mm micro speaker
inside cap.
93
Field test of Self-Testing Microphone
Ref Mic
Test Mic
Added windshield for
outdoor testing. Results
compared with reference
microphone.
Self-Testing system first
version has a separate jack
for the microphone
and
94
speaker.
System Architecture
95
System Architecture
96
ISEM
Intelligent Sensorwebs for
Environmental Monitoring
IWOIB project: Prospective Research for Brussels
Start date: March 2010
97
Challenges
 A breakthrough in the creation of Urban
sound-level monitoring on a big scale with:




High spatial and temporal accuracy
Heterogeneous nodes
Ultra low-cost autonomous nodes (20€ to 50€)
Use public RF band to lower communication
cost
98
Integrating smaller nodes:
ISEM
250 W
5W
0.05 W
0.05 W
+
99
Adding computing power
to the Sentilla motes
Static Hybrid RC systems
Two phase operation
Configuration
Computation
100
Adding embedded energy
scavenging to the motes
101
Adding embedded energy
scavenging to the motes
Harvesting Method Power Density
Solar Cells
15 mW/cm3
Piezoelectric
330uW/cm3
Vibration
116uW/cm3
Thermoelectric
40 uW/cm3
102
ISN: Applications for WSN’s
 Interoperable Sensor Networks
 Explore/improve existing standards in
WSN world and build application test
cases
 ITEA Consortium
 O&O project IWOIB, start 1st July 2010
103
PROJECT DRIVERS&GOALS
ISN/EU OBJECTIVES
Simplify deployment
and integration
Simplify application
development
Simplify monitoring and
management
Enable efficient use
of data
104
ISN-Brussels
WirelessHART
ISN
Box
6loWPAN
ZigBee
Gateway
Vertical markets
• Energy efficiency and comfort in building
• Renewable energy
...
105
VUB Greenhouse Monitoring
30 Tmote Sky
ISN
Box
Gateway
Expected lifetime: > 6 months
Measured parameters: temperature, humidity,
light
106
Acoustic Comfort Monitoring
ISN Box
Measured parameters:
•Background noise
•Reverberation time
•Speech intelligibility
•etc.
107
Acoustic Comfort Monitoring
108
Photovoltaic Monitoring
ISN Box
Measured parameters:
•Irradiation
•Module temperature
•Ambient temperature
•Wind speed
•etc.
109
Photovoltaic Monitoring
110
Conclusions
WSN today
Technology available on the market
 Time for deployment!
WSN future
Technology?
Applications?
111
Thank you for your
attention!
Question?
112