CSE590: CSE590: Algorithms for wireless sensor networks Algorithms - - PowerPoint PPT Presentation

▶
cse590 cse590 algorithms for wireless sensor networks
SMART_READER_LITE
LIVE PREVIEW

CSE590: CSE590: Algorithms for wireless sensor networks Algorithms - - PowerPoint PPT Presentation

Feb 11, 2024 571 likes •1.02k views

CSE590: CSE590: Algorithms for wireless sensor networks Algorithms for wireless sensor networks Jie Gao Computer Science Department Stony Brook University 9/7/05 Jie Gao, CSE590-fall06 1 Computer networking Computer networking

slide-1
SLIDE 1

9/7/05 Jie Gao, CSE590-fall06 1

CSE590: CSE590: Algorithms for wireless sensor networks Algorithms for wireless sensor networks

Jie Gao

Computer Science Department Stony Brook University

slide-2
SLIDE 2

9/7/05 Jie Gao, CSE590-fall06 2

Computer networking Computer networking

  • Internet

– Enable efficient communication (Email, skype). – Share computing/storage resources (RAID, grid computing). – “Network of information”: Distributed information publishing, storage, and indexing (e.g., google).

  • Sensor networks

– Connect the Internet with the physical world.

slide-3
SLIDE 3

9/7/05 Jie Gao, CSE590-fall06 3

A generic sensor node A generic sensor node

  • CPU.
  • On-board flash memory or external memory
  • Sensors: thermometer, camera, motion, light

sensor, etc.

  • Wireless radio.
  • Battery.
slide-4
SLIDE 4

9/7/05 Jie Gao, CSE590-fall06 4

Centralized Centralized v.s v.s. distributed sensing . distributed sensing

  • Centralized sensing:

– a few number of powerful sensors.

  • Distributed sensing:

– a large number of inexpensive, less powerful sensors.

  • Advantages of sensor networks:

– System robustness. – Easy to deploy. – Fine-grained data collection or environment monitoring.

slide-5
SLIDE 5

9/7/05 Jie Gao, CSE590-fall06 5

Applications of sensor networks Applications of sensor networks

  • Fine-grained data collection.

– Agriculture: monitor soil moisture. – Science: volcanoes, birds, glacier.

  • Traditional approach:

– A few sensors connected by wires. – Not sufficient for dense monitoring, e.g., sample every meter in a forest. – Wires are messy, easy to break.

slide-6
SLIDE 6

9/7/05 Jie Gao, CSE590-fall06 6

Into Deep Ice Into Deep Ice

  • Monitor glacier behavior, for the understanding
  • f the dynamics of glaciers as well as global

warming.

  • “Sensors are placed in, on and under glaciers

and data collected from them by a base station

  • n the surface. Measurements include

temperature, pressure, stress, weather and subglacial movement.”

slide-7
SLIDE 7

9/7/05 Jie Gao, CSE590-fall06 7

Into Deep Ice Into Deep Ice

  • http://leo.ecs.soton.ac.uk/glacsweb/plotter.php
  • A java applet for on-line data query
slide-8
SLIDE 8

9/7/05 Jie Gao, CSE590-fall06 8

Applications of sensor networks Applications of sensor networks

  • Ad hoc networking: easy to deploy

– Disaster rescue. – Military applications.

  • Real-time environment monitoring.

– Alert system. – Health care.

  • RFID tags

– Warehouse management, library book management – Smart shopping.

slide-9
SLIDE 9

9/7/05 Jie Gao, CSE590-fall06 9

From a philosophical point of view From a philosophical point of view

  • Swarm intelligence: “systems of non-intelligent robots

exhibiting collectively intelligent behavior” [Beni, 89].

Ants forming a bridge Shortest path routing

slide-10
SLIDE 10

9/7/05 Jie Gao, CSE590-fall06 10

Networked sensors can be intelligent Networked sensors can be intelligent

  • Local decisions, global optimal behaviors.

The "V" formation of the flock enables each individual bird to save about 23% energy.

slide-11
SLIDE 11

9/7/05 Jie Gao, CSE590-fall06 11

“ “A world full of sensors A world full of sensors” ” is not a fantasy is not a fantasy

  • There are already many sensors deployed
  • ut there.

– Cell phones. – Surveillance cameras. – GPS receivers. – Motion and light sensors.

  • Now let’s connect them into a network.
slide-12
SLIDE 12

9/7/05 Jie Gao, CSE590-fall06 12

Outline Outline

  • Challenges of wireless sensor networks.
  • Course overview.
slide-13
SLIDE 13

9/7/05 Jie Gao, CSE590-fall06 13

Major goals Major goals

  • How to organize the network?
  • How to retrieve, store, and index data from

sensors?

  • Shift interest from “network” to “data”.
  • Intertwine data processing with data

delivery.

slide-14
SLIDE 14

9/7/05 Jie Gao, CSE590-fall06 14

Algorithmic challenges Algorithmic challenges

  • Resource constraints:

– Computation, communication and energy.

  • Dynamic environment:

– Network topology is dynamic. – Inexpensive nodes have high failure rate.

  • Robust data-processing algorithms:

– Sensor data is noisy. Sensors malfunction.

  • Distributed algorithms preferred.
slide-15
SLIDE 15

9/7/05 Jie Gao, CSE590-fall06 15

Energy constraints Energy constraints

  • Battery-powered devices.
  • Load balancing

– avoid overloading any particular node.

  • Communication is much more energy

consuming than computation.

– Transmitting 1 bit costs as much energy as running about 1,000 instructions.

  • In-network processing

– Compress raw data in the network.

slide-16
SLIDE 16

9/7/05 Jie Gao, CSE590-fall06 16

Ad hoc networking Ad hoc networking

  • Ad hoc multi-hop network:

– Nodes relay messages for each other. – Save energy: energy consumption is 1/rα, where α=2~5.

  • Ad hoc deployment, no fixed or

predefined topology.

  • Highly dynamic:

– Sensors die, links come and go. – Wireless broadcasting, interference.

slide-17
SLIDE 17

9/7/05 Jie Gao, CSE590-fall06 17

Difficult calibration Difficult calibration

  • Localization

– Data integrity. – Location information helps network

  • rganization.
  • Synchronization

– No global sync server. – Important for in-network reasoning such as target tracking

slide-18
SLIDE 18

9/7/05 Jie Gao, CSE590-fall06 18

Information processing Information processing

  • Two major challenges:

– Massive amount of data. – Raw sensor readings.

  • Techniques to be developed:

– Low-level sensor readings high-level semantic reports. – Data aggregation (suppress redundant data) and compression (by exploring spatial correlation).

slide-19
SLIDE 19

9/7/05 Jie Gao, CSE590-fall06 19

Information storage, indexing, query Information storage, indexing, query

  • New query engine: “google” the physical

world.

– Where is the data stored? – How is the data indexed in a distributed fashion? – How does a user retrieve his desired data?

slide-20
SLIDE 20

9/7/05 Jie Gao, CSE590-fall06 20

Distributed, localized, collaborative Distributed, localized, collaborative protocols protocols

  • Measurements are local, computing

and communication are distributed;

  • Achieve globally optimal objectives.
  • The local/global interaction is one of

the most mysterious phenomena in nature that we don’t understand.

slide-21
SLIDE 21

9/7/05 Jie Gao, CSE590-fall06 21

Outline Outline

  • Challenges of wireless sensor networks.
  • Course overview.
slide-22
SLIDE 22

9/7/05 Jie Gao, CSE590-fall06 22

Course overview Course overview

  • We study robust algorithmic solutions for

– Network organization. – Information processing.

  • Basic network setup. (topology control and

discovery).

  • Where is the data generated? (Localization)
  • How to transfer data? (routing)
  • How to summarize and query the data?

(Data storage, compression, replication, indexing, query, etc).

slide-23
SLIDE 23

9/7/05 Jie Gao, CSE590-fall06 23

Course information Course information

  • http://www.cs.sunysb.edu/~jgao/CSE590-fall06/
  • T/Th 12:50pm-2:10pm at Social Behavior Science

S218.

  • My email: jgao@cs.sunysb.edu. My office hour:

1415 CS building, Tuesday/Thursday 3:00pm - 4pm or by appointment.

  • Interactive class: ask questions whenever you want.
slide-24
SLIDE 24

9/7/05 Jie Gao, CSE590-fall06 24

Course materials Course materials

  • Research papers.

– Required reading (covered by lectures): please read these papers before class. – Additional reading. – Use “google”

  • Recommended textbook.

Wireless sensor networks: An information processing approach By Feng Zhao and Leonidas Guibas Elsevier/Morgan-Kaufmann, 2004.

  • On 2-hours reserve in CS library.
slide-25
SLIDE 25

9/7/05 Jie Gao, CSE590-fall06 25

Course requirement and grading Course requirement and grading

  • 20%: class participation

– Group of pairs. – A 30min presentation of a paper in class.

– A critique (about 1 page long, two pages at most).

  • Strength and limitation of this paper.
  • ways to improve over the results in the

paper

  • ther related open problems motivated

by this paper.

slide-26
SLIDE 26

9/7/05 Jie Gao, CSE590-fall06 26

Course requirement and grading Course requirement and grading

  • 80%: research project in groups of 2 or 3.

– Theory track

  • Choose a topic of interest.
  • Prove something, come up with an algorithm…
  • If significant progress is made, then you get an A

automatically.

  • Otherwise, evaluation is based on a survey paper
  • f at most 15 pages on related work, possible

techniques, your observations and thoughts, and future directions.

– Applied track

  • Simulation or implementation of an existing or

new algorithm and performance comparison.

  • A report on your discovery.
slide-27
SLIDE 27

9/7/05 Jie Gao, CSE590-fall06 27

Projects Projects

  • Milestone:

– By early-Oct, find your groupmate(s), find a topic of interests, do some preliminary reading. – Mid-term project presentation 10/10 and 10/12 (tentative): present to the class your project idea and get feedback. – Project presentation on 12/12 and 12/14 (last class).

  • Start early.
  • If you want to discuss your idea, come to my office

hour or email me for an appointment.

  • Talk (or email) to me for questions on Latex,

software, etc.

slide-28
SLIDE 28

9/7/05 Jie Gao, CSE590-fall06 28

Projects Projects

  • Good projects can turn out to be a research paper.

– In fall05, two projects get published in top conferences such as Mobihoc and mobicom.

  • Suggested project ideas will be handed out in class in

a couple of weeks. You are also highly encouraged to come up with your own.

  • What I am looking for in research projects: creative,

innovative ideas.

  • Have fun.
slide-29
SLIDE 29

9/7/05 Jie Gao, CSE590-fall06 29

Questions or comments? Questions or comments?

slide-30
SLIDE 30

9/7/05 Jie Gao, CSE590-fall06 30

Modeling sensor networks Modeling sensor networks

Jie Gao

Computer Science Department Stony Brook University

slide-31
SLIDE 31

9/7/05 Jie Gao, CSE590-fall06 31

Required reading Required reading

  • [Ganesan02] D. Ganesan, B. Krishnamachari, A.

Woo, D. Culler, D. Estrin and S. Wicker. Complex behavior at scale: An experimental study of low-power wireless sensor networks. Technical Report UCLA/CSD-TR 02-0013, UCLA, 2002.

slide-32
SLIDE 32

9/7/05 Jie Gao, CSE590-fall06 32

Modeling node distribution Modeling node distribution

  • Arbitrary distribution.
  • Random distribution.

– Inside a fixed-size region, randomly throw nodes.

  • Controlled density.

– Guarantee sufficient coverage. – At least one node inside every disk of radius r.

slide-33
SLIDE 33

9/7/05 Jie Gao, CSE590-fall06 33

Modeling link connectivity Modeling link connectivity

  • Wireless communication characteristics is

complex.

slide-34
SLIDE 34

9/7/05 Jie Gao, CSE590-fall06 34

Modeling link connectivity Modeling link connectivity

Transmission power

slide-35
SLIDE 35

9/7/05 Jie Gao, CSE590-fall06 35

Link connectivity Link connectivity

  • In general far away nodes can not

communicate and nearby nodes can.

  • Directionality: approximately 5-15% of all

links are asymmetric.

  • Heavy tail: long links of good quality exist.
slide-36
SLIDE 36

9/7/05 Jie Gao, CSE590-fall06 36

Unit disk graph model Unit disk graph model

  • Two nodes with distance less than 1 have

a link in between.

– The simplest model. – Widely used in theoretical analysis of algorithm performance. – Many interesting geometric properties. – Algorithms based on UDG need to be re- examined in practice.

slide-37
SLIDE 37

9/7/05 Jie Gao, CSE590-fall06 37

More general models More general models

  • Quasi-UDG model.

– A link exists if inter-distance < r. – A link does not exist if inter-distance >R. – Uncertain otherwise.

  • More general than UDG
  • Still not quite practical.

– r might be approaching 0.

r R

slide-38
SLIDE 38

9/7/05 Jie Gao, CSE590-fall06 38

Graphs with bounded growth rate Graphs with bounded growth rate

  • A graph has growth rate k and density d if

the number of nodes within r-hop from any node is at most drk.

  • UDGs or quasi-UDGs with constant

density have growth rate 2.

r-hop neighborhood

slide-39
SLIDE 39

9/7/05 Jie Gao, CSE590-fall06 39

Mostly seen simulation settings Mostly seen simulation settings

  • Random geometric graph.

– Random distribution + UDG.

  • Perturbed grid + UDG.

– Each grid point is perturbed by a random Gaussian or uniform noise. – Simulates manual deployment.

slide-40
SLIDE 40

9/7/05 Jie Gao, CSE590-fall06 40

Random geometric graph Random geometric graph

slide-41
SLIDE 41

9/7/05 Jie Gao, CSE590-fall06 41

Random geometric graph Random geometric graph

slide-42
SLIDE 42

9/7/05 Jie Gao, CSE590-fall06 42

Random geometric graph Random geometric graph

slide-43
SLIDE 43

9/7/05 Jie Gao, CSE590-fall06 43

Which model is the best? Which model is the best?

  • No clear winner.
  • For a specific problem,

– Extract the essential feature needed to solve the problem. – Choose the most general model possible.

  • Link dynamics is not captured yet.
slide-44
SLIDE 44

9/7/05 Jie Gao, CSE590-fall06 44

Questions or comments? Questions or comments?

  • Next class: localization.