Wireless Sensor Networks for High Fidelity Sampling

1
Wireless Sensor Networks for High Fidelity
Sampling

• Sukun Kim
• Qualifying Examination
• Dec 1, 2005

2
High Fidelity Sampling

• Three classes of WSN applications
• Monitoring environments
• Great duck island 11, Redwood forest 12
• Focus on low-duty cycle and low power consumption
• Monitoring objects High Fidelity Sampling
• machine health monitoring 13, condition-based
  monitoring, earthquake monitoring 14,
  structural health monitoring 15
• Focus on fidelity (quality) of sample
• Interactions with space and objects
• Lighting control 16
• Focus on control

3
Structural Health Monitoring
Challenges

• High Fidelity Data
• High Frequency Sampling with Low Jitter
• Time Synchronized Sampling
• Large-scale Multi-hop Network
• Reliable Command Dissemination
• Reliable Data Collection

FTSP 8
Mint 9
Drip 10
4
Reliable Data Collection- Problem Statement

• Every data from every node needs be collected to
  PC over a multi-hop network without loss
• High throughput
• Small number of packet injections to network
• Overcome interference
• Assumptions
• Powerful receiver, resource constrained sender
• Receiver (PC) can arbitrate flow
• Low congestion
• Low loss rate

5
Hypothesis (Proposed Solution)

• High Fidelity Data Low-cost low-power MEMS
  accelerometer board with proper signal processing
  and calibration, produces data of meaningful
  fidelity
• High Frequency Sampling with Low Jitter WSN
  mote and TinyOS with guaranteed worst-case
  jitter, provide real time operation of meaningful
  level
• Reliable Data Collection Rate-based
  alternating-flow protocol with complex receiver
  and simple sender and pipelining, achieve
  reliable collection efficiently

6
Table of Contents

• Related Work
• High Fidelity Data
• High Frequency Sampling with Low Jitter
• Reliable Data Collection
• Future Work

7
Structural Health Monitoring System
Spencer, et al 5 Lynch, et al 6 Wisden 7
High Fidelity Data 2.0µG/vHz 0.5mG/vHz ?
High Frequency Sampling with Low Jitter ? 977Hz 160Hz
Time Synchronized Sampling ? lt1µs no
Large-scale Multi-hop Network ? no yes
Reliable Command Dissemination ? yes N/A
Reliable Data Collection ?B/s ?B/s 250B/s
Preliminary customized systems Kruger, et al
1, Qiang, et al 2, Engel, et al 3, Caicedo,
et al 4
8
TCP on Wireless Networks

• Blind link-level-retransmission (LLR) can
  decrease throughput DeSimone, et al 17
• Support for mobile host
• I-TCP, Balakrishnan, et al 18
• Support for wireless ad-hoc network WTCP, ATP
• Rate-based transmission
• Selective ACK contains congestion information
• No sender timeout for retransmission WTCP

9
Reliable Transfer on WSN

• Reliable diffusion
• PSFQ, RMST, Garuda, Drip, Deluge
• Congestion Control
• ESRT, CODA, Fusion, Ee, et al 19
• Better best-effort convergence
• RBC
• Reliable convergence
• Wisden
• Sender sends data at static rate
• In a routing tree, mote sends NACK to get missing
  packet from child for efficiency
• PC sends NACK to source mote for e2e reliability
• Incorrectly tuned rate and topology change make
  the network collapse
• Compared to hardware, low bandwidth

10
Table of Contents

• Related Work
• High Fidelity Data
• In IWSHM 05
• High Frequency Sampling with Low Jitter
• Reliable Data Collection
• Future Work

11
Accelerometer Board

• Two accelerometers for two axis
• Thermometer, 16bit ADC, Low-pass filter

ADXL 202E
Silicon Designs 1221L

• Signal processing averaging in software
• Calibration for manufacturing variation and
  temperature
• System noise floor 30(µG/vHz)
• Gives desired quality in static, dynamic test

12
Table of Contents

• Related Work
• High Fidelity Data
• High Frequency Sampling with Low Jitter
• In IWSHM 05
• Reliable Data Collection
• Future Work

13
Analysis of Jitter
Sampling
Other jobs like EEPROM write
Time
Non-preemptible portion
Preemptible portion
Probability

• 1. Remove unnecessary blocking atomic section,
  interrupts
• Turn off unnecessary components
• 2. Verify maximum blocking section is small enough

P2/T2
Jitter
P1/T1
0
C
W
T1C
T2C
14
Verification of Jitter (6.67KHz)
10µs
0µs
0µs
10µs
Time Series
Histogram

• Jitter is within 10µs (6.67), 0.2 at 200Hz
• Tradeoff turning off radio
• WSN mote and TinyOS are not inherently limited in
  real time operation
• It is a matter of the hardness of real time
  requirement and the tradeoff for the loss of
  functionality

15
Table of Contents

• Related Work
• High Fidelity Data
• High Frequency Sampling with Low Jitter
• Reliable Data Collection
• Straw
• Future Work

16
Overview
PC
Mote
Application
Application
read(dest, start, size)
Straw
Straw
Multi-hop Routing
Routing layer is assumed to deliver packets
end-to-end

• PC application arbitrates flow
• Determines who sends when
• Triggers one flow at a time
• Adjust RTT, adjust transmission rate to avoid
  interference
• Cross-layer information

17
Protocol
1. Data Request
PC
Mote
Complex
Simple
2. Data Transfer
3. Request missing holes
4. Transfer missing holes
Selective NACK
Straw
Straw

• Selective NACK
• No need for flow control, rate-based transmission
• No congestion control
• Pipelining, no link level retransmission
• Alternating flow, no concurrent bidirectional flow

Rate if (Depth lt Interference Radius) then
(UART Delay) Depth (Radio Delay) else
(Interference Radius) (Radio Delay)
18
Optimization

• Transfer the checksum of the whole data to
  guarantee the integrity
• Parallelize reading from the memory and sending
  to the network

Send
Network
Read
Memory
19
Test Result
93.2 299B/s
91.8 304B/s

• 10KB of data
• 500 packets
• Mica2dot, 36 bytes/pkt
• Comparison to routing layer
• 630B/s for 1 hop
• Up to 91.4 efficiency
• 352B/s for 2 hops
• Up to 86.4 efficiency

91.4 296B/s
95.6 560B/s
96.6 576B/s
End-to-end Raw Reliability Effective
Bandwidth (Byte/s)
20
Channel Capacity Utilization

• Hardware capacity limit
• UART 57.6Kbps
• Radio 19.2Kbps
• 1 hop 14.4Kbps
• Measured actual capacity usage
• UART 27.8Kbps
• Radio 9.74Kbps
• Routing 5.46Kbps (1 hop)
• Reliable 4.7Kbps (1 hop)

33
Mica2, 36bytes/pkt
21
212.7µs
33 data transfer
24 overhead for transferring data
43 header transfer overhead
Mica2, 36bytes/pkt
22
Effect of Packet Size on Bandwidth
RAM

• Doubled packet size 36B ? 72B
• Payload 20B ? 56B (2.8 times)
• Packets/sec 29.4 ? 20.9 (71)
• Bandwidth doubled 588B/s ? 1172B/s (1.99 times)
• RAM usage jump from 3437B to 4733B for SHM
  application (Sentri)
• 36 packet buffers
• Basic services (Comm TimeSync Routing Bcast
  Reliable) can go beyond 4KB of RAM

Loss rate was 0.2
23
Why so much RAM (packet buffer)?
2. Forwarding Queue (20 out of 36)
1. At least one at each end Component (12 out of
36)
Cmpnt5
Cmpnt6
Forward
Cmpnt2
Cmpnt3
Cmpnt4
Straw
Forward
Routing
Cmpnt1
Drip
Bcast
Forward
Forward
QueuedSend
Sharing packet buffer?
GenericComm
Reliability of system versus Efficient use of
resource
Sensornet Network Layer
24
Reliable Data Collection- Problem Statement
Revisited

• Every data from every node needs be collected to
  PC over a multi-hop network without loss
• High throughput
• Small number of packet injections to network
• Overcome interference
• Assumptions
• Powerful receiver, resource constrained sender
• Receiver (PC) can arbitrate flow
• Low congestion
• Low loss rate

IF
25
Testbed
In SECON 04
Source
Destination
5 hops, 26.28 link loss rate (78.23 E2E), 300
packets, separated by 1 sec, on BVR
26
8 original messages
27
8 original messages
Overhead End-to-end retransmission (Straw)
8.6 at 96.6(1hop) success rate 13.6 at
91.8(2hops) success rate Erasure code 12.5,
25,
28
Table of Contents

• Related Work
• High Fidelity Data
• High Frequency Sampling with Low Jitter
• Reliable Data Collection
• Future Work

29
Link Level Retransmission Pipelining

• Link level retransmission is effective when loss
  rate is high
• Pipelining is effective for long path
• Combining two can intensify interference

- higher correlated losses
Throughput (e2e success rate) (pkts/s at
sender)
30
Congestion Control

• In case Straw is used together with constant
  upstream traffic, congestion control will be
  needed
• Congestion control from the receiver
• Include congestion information in NACK packet
• Sender adjusts rate using congestion information

31
Deployment at Footbridge
Using Sentri (structural health monitoring
toolkit)
32
Plots of calibrated data
33
Model Properties
Match with SAP bridge model
First Vertical Mode of Vibration
34
Timeline

• Mar 2006 for SenSys Deployment on the Golden
  Gate Bridge
• April 2006 Study correlation between (link
  level retransmission pipelining) and
  interference
• Jul 2006 Implement link level retransmission
  pipelining
• Dec 2006 Congestion control with rate adjustment

35
Questions and Discussions
36
Backup Slides
37
References

• 1 M. Kruger and C. U. Grosse. Structural health
  monitoring with wireless sensor networks.
  Otto-Graf-Journal, 157790, 2004.
• 2 P. Qiang, G. Xun, and Z. Chang-you. A
  wireless structural health monitoring system in
  civil engineering. The Third International
  Conference on Earthquake Engineering (3ICEE),
  Nanjing, China, October 18-20, 2004.
• 3 J. M. Engel, L. Zhao, Z. Fan, J. Chen, and C.
  Liu. Smart brick - a low cost, modular wireless
  sensor for civil structure monitoring.
  International Conference on Computing,
  Communications and Control Technologies (CCCT
  2004), Austin, TX USA, August 14-17, 2004.
• 4 J. M. Caicedo, J. Marulanda, P. Thomson, and
  S. J. Dyke. Monitoring of bridges to detect
  changes in structural health. the Proceedings of
  the 2001 American Control Conference, Arlington,
  Virginia, June 2527, 2001.
• 5 B. S. Jr., M. Ruiz-Sandoval, and N. Kurata.
  Smart sensing technology Opportunities and
  challenges. Journal of Structural Control and
  Health Monitoring, in press, 2004.
• 6 J. P. Lynch. Overview of wireless sensors for
  real-time health monitoring of civil structures.
  Proceedings of the 4th International Workshop on
  Structural Control (4th IWSC), New York City, NY,
  USA, June 10-11, 2004.
• 7 N. Xu, S. Rangwala, K. Chintalapudi, D.
  Ganesan, A. Broad, R. Govindan, and D. Estrin. A
  wireless sensor network for structural
  monitoring. the Proceedings of the ACM Conference
  on Embedded Networked Sensor Systems, November
  2004.

38
References (continued)

• 8 A. DeSimone, M. C. Chuah, O. Yue, Throughput
  performance of transport-layer protocols over
  wireless LANs. In Proceedings of IEEE Globecom
  93, Houston, USA, 1993.
• 9 A. Bakre, B. R. Badrinath, I-TCP indirect
  TCP for mobile hosts, Proceedings of the 15th
  International Conference on Distributed Computing
  Systems (ICDCS'95).
• 10 H. Balakrishnan, S. Seshan, and R. H. Katz,
  Improving reliable transport and handoff
  performance in cellular wireless networks. ACM
  Wireless Networks, December 1995.

39
(No Transcript)
40
Reliable Data Collection- Problem Statement

• Every data from every node needs be collected to
  PC over a multi-hop network without loss in a way
  that gives high throughput with small number of
  packet injections
• The collection must overcome interference with
  the flow in the same and the opposite direction

41
Reliable Data Collection- Problem Statement

• Every data from every node needs be collected to
  PC over a multi-hop network without loss in a way
  that gives high throughput with small number of
  packet injections
• The collection must overcome interference with
  the flow in the same and the opposite direction

Data
ACK or NACK
42
(No Transcript)
43
Accelerometer Board Design

• Two accelerometers for two axis
• Thermometer
• 16bit ADC

ADXL
Silicon Designs
ADXL 202E Silicon Designs 1221L
Range -2G 2G -0.1G 0.1G
System noise floor 200(µG/vHz) 30(µG/vHz)
Price 10 150
44
Signal Processing

• As an analog signal processing low-pass filter is
  used, which filters high frequency noise
• Low-pass filter with threshold frequency 25Hz is
  used
• As a digital signal processing, averaging is used
• If noise follows Gaussian distribution, by
  averaging N numbers, noise decreases by a factor
  of sqrt(N)

45
Sensor Calibration
46
Temperature Calibration
F
C
27.3
81.1
19.5
67.1
11.7
53.0
Temperature
3.9
39.0
mG
27.5
0
Thanks to Crossbow
-27.5
Acceleration
47
Power Consumption
Operation Mode Consumption (mW)
Board Only 240.3
Idle 358.2
One LED On 383.4
Erasing Flash 672.3
Sampling 358.2
Transferring Data 388.8

• 3 of Tadiran 5930 (lithium-ion, 3.6V, 19Ah, 17,
  D size) are used

48
Power Consumption (cont)

• With optimal sleeping, 30 days
• Board itself consumes significant amount of
  energy

Power source
Power source
Switch
Switch
Sensor
Mote
Mote
ADC
Sensor
ADC
49
(No Transcript)
50
Verification of Jitter Time Series (1KHz, 5KHz,
6.67KHz)
10µs
0µs

• Peak to Peak is time to fill up buffer
• Spiky portion is time to write buffer to flash
• Can sample as long as the former is larger than
  the latter

51
Verification of Jitter Histogram (1KHz, 5KHz,
6.67KHz)
0µs
10µs

• Jitter is within 10µs
• Peak at 625ns Wakeup time from sleep mode

52
Real-time System

• Use separate MCU for sensor board, or two motes

Sensor
MCU
MCU
Buffer
Radio
53
(No Transcript)
54
Constraints and Options
( of pkts received) Psuccess ( of pkts sent)

• Sources of Failure
• Link Failure
• Mote Failure

• How to obtain reliability? (possible options)
• Add redundancy of information
• Retransmission link-level, end-to-end
• Data redundancy duplication, erasure code
• Path redundancy Use thick path
• Increase success rate
• Alternative Route
• Congestion Control

55
Problem Statement

• Goal
• Reliable communication in multi-hop Wireless
  Sensor Networks
• Assumption
• Wireless communication
• Resource-constrained mote

56
Erasure Code
57
Systematic Code

• Benefit if receiver has codes containing
  original messages
• Encoding, Decoding are faster
• Even if receiver get less than 8 packets, we
  dont lose every message

58
Systematic Code
Encoding one code word takes 1.7ms Decoding time
varies from 0 to 27ms Real time processing is
possible
In MICA2
59
Alternative Route Discovery
Find Alternative Route
What if
And if
Get 6 best candidates for the next hop from
routing table. And try from the best
60
8 original messages
61
8 original messages
62
(No Transcript)
63
(No Transcript)
64
(No Transcript)
65
(No Transcript)
66
Design Space Routing Layers
Convergence
Implementation on Beacon Vector Routing However,
solutions are applicable to all routing layers
67
8 original messages
68
8 original messages
69
Given a threshold reliability requirement, what
is the retransmission and redundancy combination
that has the smallest overhead?
Decomposing causes of failures (5 ret, no RF)
70
Findings

• Link-level retransmission efficiently handle
  transient link failure
• Route fix cure stale routing table problem,
  increase usefulness of erasure code
• Erasure code relieve the burden of last a few
  packets, which is very expensive
• Some options addresses some problems efficiently,
  but not all failures
• Combining options would provide a sweet spot

71
Run Length Encoding (RLE)

• 94720, 94704, 94715, 94708 becomes 947 20, 04,
  15, 08
• Exception
• 94720, 94704, 92345, 94708 becomes
• 947 20, 04, \92345, 08
• Run simulation on footbridge vibration data

Threshold 2
Fragment Size 4
72
High Resolution Footbridge data
73
Low Resolution Footbridge data
74
Analysis
High Resolution Low Resolution
RLE 66 45
gzip 68 49
Theory 56.25 (9 dynamic bits) 37.5 (6 dynamic bits)

• Algorithm of RLE fits better to sensor data
• Basic algorithm of gzip utilizes repetition of
  same pattern
• Compression ratio is sensitive to parameters
  (even go above 100)
• Selecting RLE parameter (either statically or
  dynamically) is critical

Windows zip showed 0.64 increase
75
Analysis (continued)

• There exists room for lossless or lossy
  compression
• Compression ratio is sensitive to parameters
  (even go above 100)
• Selecting RLE parameter (either statically or
  dynamically) is critical

76
(No Transcript)
77
State Diagram of Sender

• Send everything once, and fill holes
• Read depth in routing tree, and adjust
  transmission rate, packet interval, RTT estimate

Start
Request / Set Timer
More
No / Stop Timer
Yes / Read
Timer Fired / Send

• Simple (intelligence in receiver)
• Interface is simple
• read(start, size, buffer)

78
State Diagram of Receiver
Start
Send Network-Info Request
Timeout / FAIL
Receive / Send Transfer Request, Set Timer
Receive not Last / Set Timer
Receive Last Timeout / count 0
No
More in Round
Yes / count
Yes
Receive / count 0
Yes / Send Read Request, Set Timer
No / FAIL
More
count lt threshold
Timeout
No / SUCCESS
79
Mica2, 36bytes/pkt
80
Channel Capacity Utilization

• Hardware capacity limit
• UART 57.6Kbps 200pkts/s
• Radio 19.2Kbps 66.7pkts/s
• 1 hop 14.4Kbps 50pkts/s
• Measured actual capacity usage
• UART 27.8Kbps 120pkts/s
• Radio 9.74Kbps 42pkts/s
• Routing 5.46Kbps 31pkts/s (1 hop)
• Reliable 4.7Kbps 29.4pkts/s (1 hop)

Mica2, 36bytes/pkt
81
Mica2, 36bytes/pkt
82
Effect of header is considered here
Mica2, 36bytes/pkt
83
For Mica2, packet size 36 bytes
Top, Left Packet Size Bottom, Right pkts/sec
84
RAM space

• From 3437 to 4733 (SHM application Sentri)
• 36Bytes RAM increase per 1Byte increase in packet
  size
• Reason Packet buffer space
• 4 below GenericComm
• 3 in TimeSync
• 16 in Routing
• 4 in Bcast
• 2 5 in Reliable
• 2 in application
• Basic services (Comm TimeSync Routing Bcast
  Reliable) can go beyond 4KB RAM with packet
  size 72Bytes

85
Findings

• Reliable Data Collection (Straw)
• 5.2 decrease in packet throughput
• 13.8 decrease in bandwidth
• Packet is small compared to the size of header,
  so doubling packet size doubles bandwidth
• RAM limit due to many packet buffers
  Reliability of system versus Efficient use of
  resource

86
(No Transcript)
87
Loss Rate at Footbridge
Mica2, 36bytes/pkt
88
Loss Rate at Golden Gate Bridge
Mica2, 36bytes/pkt
89
Time-plots of calibrated data
90
Frequency-plots of calibrated data
91
Second Vertical Mode of Vibration
0.94
1.00
0.68
0.33
0.41
Frequency 1.78 Hz Damping Ratio 1
92
Custody Transfer

• Efficient for unreliable media
• Benefit from small packet header
• Pipelining gets complicated
• How to do custody transfer without interfering
  pipelining?

93
Low-power Reliable Transfer

• Lessons from NEST FE
• Duty cycle affects retransmission timeout
  timeout should consider duty cycle
• Aggressive transfer depletes battery mote stop
  responding, causes transfer failure
• Power information from lower layer are very
  useful for proper and efficient operation
• Power stack duty cycle info, warning for low
  energy budget