CS263
Wireless Sensor Networks
Lecture 1: Introduction
Prof. Matt Welsh
Harvard University
January 29, 2009
Introduction: Wireless Sensor Networks
TMote Sky
(Sentilla)
Tiny, low-power, wireless sensors
Minimal CPU, memory, and radio
MicaZ (Crossbow)
Typically 8 Mhz CPU, 10 KB RAM
100 m radio range, IEEE 802.15.4
Extremely low power
A pair of AA batteries can power a mote
for months or years!
WeC (Berkeley)
2009 Matt Welsh Harvard University
iMote2 (Intel)
Rene (Berkeley)
Key WSN Hardware Characteristics
Limited CPU
Limited memory
10 KB of RAM and 60 KB of program ROM.
Much of this taken up by system software.
Potentially lots of storage
Slow (8 MHz) -- No floating point computation.
512-point FFT takes 450 ms, IFFT takes 144 ms.
Some designs support up to 2 GB of MicroSD flash
But, expensive to access: 13 ms to read/write a 512-byte block; ~ 25 mA.
Low-power radio
802.15.4 best case performance: 100 Kbps or so (single node transmitting, no
interference, short range)
Approx 50 m range, and very unreliable!!
2009 Matt Welsh Harvard University
Wireless Technologies Comparison
Complexity/power/cost
802.11a
802.11b
802.11g
11 Mbps
54 Mbps
Bluetooth
720 kbps
802.15.4
Zigbee
CC1000
250 kbps
38.4 kbps
2009 Matt Welsh Harvard University
Data rate
4
Wireless Technologies Comparison
Type
Data rate
Transmit pwr
Range (approx)
Cost
802.11b
11 Mbps
100 mW
100' 300'
~$100
802.11g
54 Mbps
100 mW
< 802.11b
~$100
802.11a
54 Mbps
100 mW
80'
~$100
Bluetooth
720 kbps
1 mW / 30 mW
30' / 300'
~$5
802.15.4
250 kbps
1 mW
30 225'
~$5
2009 Matt Welsh Harvard University
Power Consumption
Type
Current (receive)
Current (transmit)
802.11b
170-350 mA
285-490 mA
Bluetooth
35 300mA active
35 300 mA active
802.15.4
19.7 mA
17.4 mA
2009 Matt Welsh Harvard University
WSN Research Phenomenon
LEAP
zeevo BT
WINS
(UCLA/ROckwell)
Intel
rene
LWIM-III
(UCLA)
SmartDust
WeC
Intel/UCB
dot
Rene
BTNode
Intel
cf-mica
Eyes
trio
Mica
XBOW
rene2
Intel
MOTE2
Intel
iMOTE
Telos
XBOW
mica
XBOW
cc-dot
Bosch
cc-mica
XBOW
mica2
XBOW
micaZ
digital sun
rain-mica
Dust Inc
blue cc-TI
97 98
99
00
01
2009 Matt Welsh Harvard University
02
03
04
05
06
07
7
Environmental Monitoring
UCLA, UC Berkeley, many others
2009 Matt Welsh Harvard University
Gunshot Detection
PinPtr, Vanderbilt
2009 Matt Welsh Harvard University
Monitoring Volcanic Eruptions
Volcan Reventador, Ecuador, July/Aug 2005
Radio modem
Konrad
Four-channel
sensor node
GPS receiver
Solar panels for charging
car battery (used by
FreeWave and GPS only)
2009 Matt Welsh Harvard University
10
Glacier Monitoring
Glacsweb, Univ. Southampton
2009 Matt Welsh Harvard University
11
Forest Fire Detection
FireWxNet, Univ. Colorado
2009 Matt Welsh Harvard University
12
Emergency Medical Care and
Disaster Response
CodeBlue, Harvard
2009 Matt Welsh Harvard University
13
Neuromotor disease assessment
Mercury, Harvard
Data acquisition
controlled by laptop
SHIMMERs
Raw data
Master node
Feature extraction
SHIMMER in armband
Classification
(e.g., UPDRS scores)
2009 Matt Welsh Harvard University
14
Urban-Scale Monitoring
CitySense, Harvard
2009 Matt Welsh Harvard University
15
The Macroscope
For the first time, sensor networks allow us to:
1) Observe the world (environment, buildings, people, etc.) at very
high spatial resolutions;
2) Make these observations continuously; and
3) Collect the observations in digital form.
Some have referred to this concept as a macroscope -a scientific instrument that observes entire systems.
2009 Matt Welsh Harvard University
16
Intelligent Instrumentation
Sensor networks are not just passive instruments!
We can push processing and intelligence into the network.
Processing can happen at many levels:
On individual sensor nodes.
At aggregation points within the network.
At the base station or gateway.
Sensor networks fundamentally change the notion of
scientific observation from a passive process to an active one.
This has a deep impact on many aspects of science.
2009 Matt Welsh Harvard University
17
Fundamental Research Questions
Low-power wireless networking
Operating system design
Dealing with complexities of RF propagation not a disc model
Limited bandwidth, very expensive to transmit, receive, and even listen!
Every node is a router addressing, route selection, reliable transfers
Motes have ~10 KB of RAM. Can't run Linux.
What are the right abstractions for concurrency, power management, communication?
Distributed network services
Nodes in a WSN don't exist in isolation. They must coordinate their behavior.
Localization how do you know where nodes are? Use RF signals? Ultrasound?
Time synchronization how do nodes agree on a global clock?
2009 Matt Welsh Harvard University
18
Fundamental Research Questions (2)
In-network sensor data processing
Communication is expensive: Sending one packet costs same
energy as thousands of CPU cycles.
Always better to process the data closer to its source
Example: aggregation nodes can collect data locally, compute aggregates (mean,
max min, etc.) rather than sending raw data
Tracking: Sensors can collaborate to detect, localize, and track a target (tank or animal)
Mobile, acoustic, and camera-based sensing
Very different sensing modalities and challenges
Acoustic and vision sensors require substantial computational horsepower
Mobile sensing involves (possibly unpredictable) variations in radio connectivity
How do we deal with noisy and intermittent measurements of the world?
2009 Matt Welsh Harvard University
19
What will this class be like?
This is a graduate research seminar.
We will mostly be reading and discussing research papers
Roughly 4 papers a week
Prerequisites:
Must either be a CS grad student or have taken either CS161 or CS143.
Must feel comfortable programming in C.
One programming assignment
Introduce you to programming sensor networks using the Pixie OS and NesC language
Run on the Harvard MoteLab sensor network testbed
Research project
You pick the topic, write a proposal, do the project, give presentation, write final report
2009 Matt Welsh Harvard University
20
Readings and Reviews
You are responsible for completing assigned readings before lecture
Email a short review of the reading to cs263-staff@eecs
Usually 2 papers for each class
Review is due before beginning of lecture
A couple of paragraphs about the reading
Highlight the main take away point of the reading
Provide a short critique of the work as well
Be concise, critical, and thoughtful
Reviews constitute 25% of your course grade
You are allowed to miss two classes of paper reviews over the term
2009 Matt Welsh Harvard University
21
Course Blog
http://harvard-cs263.blogspot.com
Blogging class discussions
Each class, one person will blog the discussion and post it later that day
You are welcome to post comments, thoughts, musings, etc. as comments
Or, you can blog anything else you want (related to the course material).
This blog is public so be technically accurate and respectful!
2009 Matt Welsh Harvard University
22
Programming Assignment
There is one programming assignment for the course
Main goal: Get experience programming a real sensor network
You will use this experience for your course project
Project will involve designing a multihop routing protocol, running on
the Pixie OS, on the Harvard MoteLab sensor testbed
You should be comfortable programming in C
2009 Matt Welsh Harvard University
23
Research Project
Main goal of this course: Do some research
Work individually or in pairs (pairs preferred)
Select a juicy research problem that fits the theme of this course
Use the project to further your own research goals
Ideal project is one that fits in with your own thesis topic in some way
Focus of project need not be on systems and networks
e.g., theory, AI, languages, hardware design, etc. are all valid
As long as it ties into the course topic in some way
2009 Matt Welsh Harvard University
24
Project Requirements
Project Proposal
Research presentations (last two days of class)
Short (4 pages max) on what you propose to do, why the project is interesting, and how
you plan to get started
Should include rough schedule of project milestones
Short project update due midway through semester short email on where you are and
how you plan to finish up your project
Give a short, fun talk telling us what you did
Learn from each other's experiences
Research papers
Conference-style research paper (12 pages max) detailing your project
Goal is to get used to writing these things it's important
I can work with you afterwards to to turn it into a conference/journal submission
2009 Matt Welsh Harvard University
25
Project Ideas
Develop an adaptive time-sync protocol that tunes packet
transmission rates based on energy availability
Develop a sensor duty-cycling algorithm that accounts for energy
drain and energy collect (e.g., using solar panels)
Develop a tool to characterize and visualize energy and bandwidth
consumption across a sensor network, use to identify hotspots and
load imbalance
Design a new sensor scripting language that includes resource
constraints as a primitive
Develop a technique to automatically detect and diagnose software
and communication failures in a sensor network
2009 Matt Welsh Harvard University
26
Course staff and administrivia
Instructor: Matt Welsh (mdw@eecs)
TF: Bor-rong Chen (brchen@eecs)
Office: Maxwell Dworkin 233
Office hours: Thursdays, 10am 12pm
Office: Maxwell Dworkin 238
Office hours: TBD
General course consulting and help with programming assignment
All papers, due dates, etc. on course web page:
http://www.eecs.harvard.edu/~mdw/course/cs263/
2009 Matt Welsh Harvard University
27
Syllabus
http://www.eecs.harvard.edu/~mdw/course/cs263
Primarily research papers from the last few years of key conferences
in the area: SenSys and IPSN in particular.
Most papers about 14 pages in length.
2009 Matt Welsh Harvard University
28
Other Policies
Enrollment will be limited to 15 students
No laptops!
Preference given to grad students in CS, then grad students in other disciplines, then
undergrads in CS, then undergrads in other disciplines.
Unless you are blogging that week.
No pass/fail grading option for this course.
2009 Matt Welsh Harvard University
29
Grading
25% - Class participation and discussion
Come to class, participate in the discussion, ask questions, speak up!
25% - Paper summaries
Allowed to miss two days' worth of summaries during the term
10% - Programming assignment
40% - Final project
Graded on original proposal, final report, and in-class presentation
2009 Matt Welsh Harvard University
30
Next lecture
Two papers to read for next lecture:
System architecture directions for networked sensors
Jason Hill et al., ASPLOS 2000
Analysis of a Large Scale Habitat Monitoring Application
Robert Szewczyk, SenSys 2004
Send reviews to cs263-staff@eecs before class!
Come prepared to talk!!!
2009 Matt Welsh Harvard University
31
