Title: Intelligent Sensing and Sensor Web: Design and Deployment Experiences
1Intelligent Sensing and Sensor Web Design and
Deployment Experiences
IEEE SIMA 2008 Nov. 18, 2008
WenZhan Song Washington State University
2Background - Sensorweb
3Sensor Web Enablement Framework
4Sensor Web Desires
- Quickly discover sensors (secure or public) that
can meet my needs location, observables,
quality, ability to task - Obtain sensor information in a standard encoding
that is understandable by me and my software - Readily access sensor observations in a common
manner, and in a form specific to my needs - Task sensors, when possible, to meet my specific
needs - Subscribe to and receive alerts when a sensor
measures a particular phenomenon
5OASIS System OverviewOptimized Autonomous Space
In-situ Sensorweb
- OASIS will have two-way communication capability
between ground and space assets, use both space
and ground data for optimal allocation of limited
power and bandwidth resources on the ground, and
use smart management of competing demands for
limited space assets. - 1. In-situ sensor-web autonomously determines
network topology, bandwidth and power allocation.
- 2. Activity level rises causing self-organization
of in-situ network topology and a request for
re-tasking of space assets. - 3. High-resolution remote-sensing data is
acquired and fed back to the control center. - 4. In-situ sensor-web ingests remote sensing data
and re-organizes accordingly. Data are publicly
available at all stages.
6Why Study Volcano?
- Volcanoes are everywhere - on Earth and beyond
- Magmatism is of fundamental importance to
planetary evolution and essential to life as we
know it - On Earth, volcanic risk is increasing rapidly as
human population increases - Volcanic Earthquakes
- Directed Blast
- Tephra
- Volcanic Gases
- Lava Flows
- Debris Avalanches, Landslides, and Tsunamis
- Pyroclastic Surge
- Pyroclastic Flows
- Lahars
7Mount St. Helens an active volcano
8Mount St. Helens an active volcano
9Volcano Crater a harsh environment
10Volcano Crater a harsh environment
Camera and gas sampler spider shown
pre-positioned at Sugar Bowl on 14 January 2005.
Shortly after this picture was taken, spider was
deployed within 100 m of extrusion site.
So we need smarter sensors and networks to ensure
continuous, spatially dense monitoring in
hazardous areas (OASIS)
11Old Stations Technology
- Several types of systems in place
- Dual frequency GPS with digital store and forward
telemetry when polled not continuous! - Short period seismic stations with geophones and
analog telemetry not digital - Broad band seismic stations with digital
telemetry cost above 10K and several days to
deploy - Microphones for explosion detection added to the
short period seismic stations
12Application Characteristics
- Challenging environment
- Extreme weathers temperature (baking/freezing),
wind, snow, rain, - Dynamic environment rock avalanche, land
sliding, gas/steam emissions, volcanic eruptions,
earthquake - Battery is the only reliable energy source. Solar
panel is possible in summer, but frequently
covered by ashes - Stations are frequently destroyed, some hot spot
can only be accessed through air drop - Low signal noise ratio of both communication and
sampling - High data rate, and require network synchronized
sampling - Seismic sensor 100-200Hz, 16 bit/sample
- Infrasonic sensor 100-200Hz, 16 bit/sample
- Lightning sensor 1Hz, 16 bit/sample
- GPS raw data 200-300 bytes/10 seconds
13Outline
- System Design
- System Deployment
- Lessons
14System Design
15OASIS System Overview
16Ground Segment Overview
17Ground Segment Overview
- In-Situ Sensor Web architecture
- 1. Sensor-nodes are self-organized to form
network data flow forms a dynamic data diffusion
tree rooted at gateway. - 2. Smart bandwidth and power management according
to environmental changes and mission needs. - 3. Remote control center manages network and
data, and interacts with space assets and
Internet.
18802.15.4 antenna
GPS antenna
old spider design
accelerometer in sandbox as a seismometer
New spider design
19Hardware Design
iMote2
UBlox GPS
MDA320
- Seismic
- Infrasonic
- Lightning
20Hardware Design
- Controller Intel Mote2
- CPU PXA271 13-416MHz with Dynamic Voltage
Scaling. 13MHz operates at a low voltage (0.85V) - Storage 256kB SRAM, 32MB SDRAM, 32MB Flash
- 802.15.4 radio CC2420
- Other Hardware Components
- Seismic low noise MEMS accelerometer (Silicon
Designs Model 1221J-002) - Infrasonic low range differential pressure
sensor (All Sensors's Millivolt Output Pressure
Sensors Model 1 INCH-D-MV) - Lightning (for ash detection) custom USGS/CVO
RF pulse detector - GPS (for deformation measurement) L1 GPS (Ublox
model LEA-4T) - Customized SmartAmp 2.4GHz, 250mW, amplify -3dBm
input to 20dBm output. - Antenna 12 dB omni, withstand extreme wind
speeds in excess of 130 MPH - Battery a bundle of Cegasa air-alkaline
industrial batteries
21Lab Environment Setup
22Lab Environment Setup
GPS receiver and distributor
DAC for data input
Lab Mini-net
23Ground Segment Software Components
- In-Situ Sensor Web Components Relationship
- 1. Topology Management Routing, Bandwidth
Management and Power Management runs autonomously
to meet mission needs and optimize the resource
usage. - 2. Network Management module enable network
status monitoring and external command control
from Space element, users or scientific agent
softwares. - 3. Situation Awareness middleware learns
environment situations and allocates network
communication and computation resources
accordingly..
24Data Management
- Port to USGS existing tools (e.g., EARTHWORM,
VALVE, SWARM) - database for real-time data storage
- web tools for seismic, GPS data visualization
- Sensor web services for Ground lt-gt Space Linkage
and future inter-geo-system data sharing
25Network Management
- Monitor the network status
- Network topology
- Node battery status
- Packet delivery ratio
- Sensor status and sampling rate
- Command Control
- Inject feedback from space asset, user, and
scientific agent software - Network adjust resource allocation accordingly
- Programming over the air
- Upgrade software without risky and expensive
field retrieval
26Node Software Architecture
27Intelligent Sensing Features
- Online Configurable Sensing
- Environment Situation Awareness
- Node Situation Awareness
- Situation-driven Synchronization
- Situation-driven Networking
28Online Configurable Sensing
29Online Configurable Sensing
- Configurable Parameters
- Change sampling rate
- Add/Delete sensor
- Change data priority
- Change node priority
- Parameter will be saved to Flash in case node
reboots. - Platform Independent
- Support for different hardware platforms
- Support for different applications
30Online Configurable Sensing
- Configurable Data Processing Tasks
- RSAM calculation
- STA/LTA event
- detection
- Threshold based
- event detection
- Prioritization
- Compression
31Online Configurable Sensing
- Light-weight Remote Procedure Call Mechanism
- Module designers decide which interface or
command to be allowed to call remotely, by simply
adding _at_rpc() - interface SensingConfig _at_rpc()
- It will translated to XML
- ltSmartSensingM.SensingConfig.setSamplingRate
commandID"23" componentName"SmartSensingM"
functionName"setSamplingRate" functionType"comma
nd" interfaceName"SensingConfig"
interfaceType"SensingConfig" numParams"2"
provided"1" signature" command result_t
SmartSensingM.SensingConfig.setSamplingRate (
uint8_t type, uint16_t samplingRate ) "gt - ltparamsgt
- ltparam0 name"type"gt
- lttype typeClass"unknown"
typeDecl"uint8_t" typeName"uint8_t" /gt - lt/param0gt
- ltparam1 name"samplingRate"gt
- lttype typeClass"unknown"
typeDecl"uint16_t" typeName"uint16_t" /gt - lt/param1gt
- lt/paramsgt
- ltreturnType typeClass"unknown"
typeDecl"result_t" typeName"result_t" /gt - lt/SmartSensingM.SensingConfig.setSamplingRategt
32Online Configurable Sensing
- Compiling time XML file generation
- Run-time memory address access
- Run-time function call
Marionette, IPSN 2006
33Online Configurable Sensing
- Cascades reliable fast flooding protocol
- Opportunistic broadcast flow
- Parent-children monitoring
- Explicit and implicit ACK
- Retry and request
34Online Configurable Sensing
- 1000 packets injected
- 2 sec interval
- 15 nodes
- 967 vs. 1735
35Online Configurable Sensing
- 1000 packets injected
- 2 sec interval
- 15 nodes
- 303.4 ms vs. 344.1 ms
36Environment Situation Awareness
- Learn environment situations through data
processing and correlations, with the guide of
domain sciences - Changes of volcanic activity will cause sensors
to adjust sampling autonomously. For instance,
active seismic activity trigger sampling rate
increases of gas sensors - Situation-driven data prioritization
- Priority of sensor types. For instance, during
eruption, gas sensor reading is more useful than
seismic reading on the other hand, during
background activity level, seismic reading is
more important. - Priority of node locations. If we can not get
data from all sensors, we shall be able to
automatically identify the minimum set of sensors
that will provide mission critical data.
37Environment Situation Awareness
38Environment Situation Awareness
- Threshold based event detection
- Compare raw data to threshold
- Lightning sensor data
- STA/LTA event detection
- Short-Term Average and Long-Term Average data are
compared - Event is triggered when ratio is over threshold
39Environment Situation Awareness
- RSAM value calculation - Real-Time
Seismic-Amplitude Measurement
- RSAM period 1 sec
- STA window 8 sec
- LTA window 30 sec
- Trigger ratio 2
40Environment Situation Awareness
41Environment Situation Awareness
- Prioritization
- Assigning priorities based on data and event type
- Assigning retransmission opportunities based on
priorities
42Node Situation Awareness
- Fault Detection
- Sensor board disconnection
- All value are 0xFFFF
- Sensor broken (including GPS)
- Sensing value are much different from other nodes
- System error
- Low battery, low memory space
43Node Situation Awareness
- Watchdog mechanism to restart nodes
- If watchdog timer was not reset for certain time
periods. - If radio did not send or receive for 5 minutes
(when the network data rate is high). - If some memory buffer is full and never get
cleared for 5 minutes. - Those are crucial for a long-term deployed
system - Our network has been continuously running after
deployed, never died till now.
44Situation-driven Time Synchronization
- Design goal
- Synchronize with UTC time
- Synchronized sampling different nodes sample
channels at same time point, 1ms resolution - Hybrid Time Synchronization
- Stay synchronized with GPS if GPS is good
- Switch to modified FTSP (Flooding Time
Synchronization Protocol, Maróti, Sensys 2004)
when GPS is disconnected
45Situation-driven Time Synchronization
- Modified FTSP
- Dynamic root selection
- Smallest node id with GPS connected
- Smallest node id if none has GPS
2
1
root
1
1
root
2
3
2
3
6
6
5
4
5
4
7
7
46Situation-driven Time Synchronization
- GPS Time Synchronization
- Not trivial due to OS delays
47Situation-driven Time Synchronization
48Situation-driven Networking
- Self-organizing and Self-healing routing
- Link metrics based on cc2420 LQI (Link Quality
Indicator) - Fast switch and fast recovery
- Qos-aware packet scheduling
- VWFQ (Variable Weighted Fair Queue)
- Energy-efficient, traffic-adaptive (supports
high-throughput), and scalable MAC protocol - TreeMAC localized TDMA for energy-efficient
high-rate data collection
49Situation-driven Networking TreeMAC
- Most of existing sensor MAC protocol assume low
duty-cycle applications (e.g., sample every 5
minutes or so) - Periodic wakeup, sense and sleep to reduce idle
listening energy waste - Until recently, some attention has been paid to
high data-rate application needs - Z-MAC (SenSys 2005)
- Funneling-MAC (SenSys 2006)
- In fact, many applications have high data (102
to 105 Hz sampling rate), including industrial
monitoring, civil infrastructure, medial
monitoring, industrial process control,
structural health monitoring, fluid pipelining
monitoring, geological environment monitoring. - Needs an Energy-efficient, traffic-adaptive, and
scalable MAC protocol for data collection - Support high-throughput if traffic is heavy
- Conserve energies if traffic is light
50Situation-driven Networking TreeMAC
- TreeMAC is inspired from the following key
observations - The majority data flows are from sensors to sink
or from sink to sensors, not random any-to-any
communication tree-based routing.
- Existing MAC protocols (including CSMA-based and
Z-MAC) tend to give every node equal channel
access opportunity, which is not fair for data
collection the node closer to the gateway
forwards more data, and shall gain more channel
access opportunity.
51Situation-driven Networking TreeMAC
- Time divided to cycles
- Cycle divided to N frames (predefined, e.g., 16
in the illustration). - Frame divided to 3 slots
- Slot 0 red Slot 1 green Slot 2 blue
- Each node picks slot according to its tree-level
only (e.g., level3) - Each node can send only in picked slot per frame,
listen/sleep in the other two - Mitigate vertical 2-hop interference
- Each node assign its childrens frames in the
beginning of each cycle, such that childrens
frames does not overlap/conflict with each other. - Mitigate horizontal 2-hop interference
52Situation-driven Networking TreeMAC
1
0
1
0
0
1
1
1
2
2
2
0
0
0
0
1
Every node get number of slots proportional to
its bandwidth demand
53Situation-driven Networking TreeMAC
54Deployment Experiences
- On Oct. 15, 2008, 5 stations air-dropped into the
crater of Mount St. Helens
55(No Transcript)
56SEP
NED
VALT
http//van-database.vancouver.wsu.edu
57(No Transcript)
5810/15/08
59Deployment Experiences
- On Oct. 16, Node 15 disappears
60Node 15 disappears in 18 hours, because
Node 15
15 minutes after the helicopter left, it
disappears again
10/22/08
61Fix it by repairing the voltage regulator circuit
and added more stones to prevent
2 days later, volcano cowboys hike to find the
mystery
10/24/08
62Wind speed peaks at 120 miles/hour
Infrasonic sensor records the unusual gust
63Weather affects link quality
6448-Hour Period Starting 5PM 11/4 Forward 2
Days
65Magnitude 1 Earthquake
- Mount St. Helens
- 3 km depth
- November 4, 2008
66OASIS 2 Hz geophone
67OASIS MEMS accelerometer
68CVO 2 Hz geophoneanalog
69CVO piezo accelerometer
70CVO digital broadband
71Deployment Lessons
- Hardware verification is as important as software
verification, and shall be done earlier too. - One month before the deployment, we start to look
at how to make the transmission range longer than
300 meters - CC2420 radio has at most 0 dBm output, some nodes
only achieve -3dBm after signal attenuation by
circuit. There is no commercial 802.15.4
amplifier supports input less than 0 dBm. - Customized SmartAmp 2.4GHz, 250mW, accept -3dBm
input.
72Deployment Lessons
- Quantitative measurement is essential, do not
just trust experiences - Some articles saying there is a correlation
between RSSI and LQI it is NOT true! - After we added RF amplified, the measurement
device (e.g., Anritsu S332D Site Master) shows it
has strong RSSI even 400 meters away, but the
network just does not connect well. - We decided to write our RSSI and LQI measurement
program TOSBaseLQI it shows LQI is below 70
(90-110 means good), though RSSI is -30 dBm (well
above -90 dBm receiver sensitivity threshold).
73Deployment Lessons
- Open for any possibility need critical thinking
skills. - RSSI is high but LQI is low with the
TOSBaseLQIs help, we tried to change everything,
almost think nothing is trustable ? Eventually,
we found that the small L-connector between
CC2420 radio and amplifier adds sufficient noises
to make S/N ratio too low before amplification. - Some iMote2 with 4-digit serial number will die
after 8 hours running unknown reasons, while
those with 7-dgit serial number are okay. - One node (node 10)s signal quality will decrease
during 1PM-6PM sunny days (when temperature is
high) - Unbelievable after we changed the high-quality
antenna cables (LMR_at_-400-ULTRAFLEX COAXIAL CABLE
TIMES MICROWAVE SYSTEMS) to some lower quality
lab cables (BELDEN 8262M17/155-00001 MIL-C-17
16428 2137 1922 ROHS), the problem is gone. This
problem does not happen in other nodes. - During deployment, USGS replaced the cables back
to the high quality one the problem did not
repeat in field so far. - Still do not know exact reasons though it must
be related to temperature!
74Sensor Data Quality
VALT BHZ
VALT 24 bit broad-band (the gold standard) in a
quiet location. It costs more than 10K and is
deeply buried in the ground, which takes several
days to deploy this station.
SEP EHZ2 Hz geophoneanalogburied, vertical
NED EHZpiezo accelerometerburied, vertical
75Sensor Data Quality
N10 EHZ2Hz geophones, buried, vertical
N14 EHZ MEMS accelerometers in sandbox
N15 EHZMEMS accelerometers, buried, vertical
76Sensor Data Quality
N16 EHZMEMS accelerometers, buried, vertical
N18 EHZ 2Hz geophones, buried, vertical
77Management Lessons
78Management Lessons
- Developing a complete system for 1-year running
is very different from writing research paper - Management overheads
- Wiki, SVN, Mailinglist, Google docs and Citeulike
are useful - Testbed setup, simulation and debug tools need to
be invented. - Not just simulations
- Not just piece by piece when put together,
things are different - Black box test and end-to-end performance test.
79Management Lessons
- Benefit from the defined design philosophy for
the team - TinyOS 1.x open source sensor network OS.
Learning curve is sharp, due to few
documentations - Programming philosophy and tips need to be
accumulated and shared - http//sensorweb.vancouver.wsu.edu/wiki/index.php/
Tips - Lacks of test in real deployment lots of bugs
to fix - http//sensorweb.vancouver.wsu.edu/wiki/index.php/
Bugs - Sanity check is necessary everywhere code
review among group is helpful - link layer checksum not working
- byte-alignment problems without _packed
attribute) - Array bound check, null pointer check, even you
are sure it will not happen. - Backup message header before it sends to radio
we found sometime radio will corrupt messages
when sending. - Not well designed for high-data rate applications
80Acknowledgement
- NASA ESTO AIST and USGS Volcano Hazard Program
- A multidisciplinary team involves
- computer scientists (Washington State University)
- WenZhan Song (Assistant Professor, WSU)
- Behrooz Shirazi (Chair Professor, EECS director,
WSU) - Students team
- space scientists (Jet Propulsion Laboratory)
- Steve Chien (Principal Computer Scientist, JPL)
- Sharon Kedar (Geophysicist, JPL)
- Frank Webb (Principal Manager, JPL)
- Danniel Tran (Software Engineer, JPL)
- Joshua Doubleday (Software Engineer, JPL)
- earth scientists (USGS Cascade Volcano
Observatory) - Rick LaHusen (Senior Instrumentation Engineer,
CVO) - Cynthia Gardener (Science-in-charge, CVO)
- John Pallister (Geologist, CVO)
- Dan Dzurisin (Geophysicist, CVO)
- Seth Moran (Seismologist , CVO)
- Mike Lisowski (Geophysicist , CVO)
81IEEE SIMA 2008 Nov. 18, 2008
Thank You!WenZhan SongEmail songwz_at_wsu.edu
Deployment video http//www.youtube.com/watch?vIb
CpioUlF0I More information, visit http//sensorw
eb.vancouver.wsu.edu
82SEP
NED
VALT
83Ground Segment Overview
84(No Transcript)