Wireless Sensor Networks
MOBICOM 2002 Tutorial
(Deborah Estrin, Mani Srivastava, Akbar Sayeed)
2006.11.01
Young Myoung,Kang (INC lab)
([email protected])
Contents





Part I : Introduction
Part II : Sensor Node Platforms & Energy Issues
Part III: Time & Space Problems in Sensor Networks
Part IV: Sensor Network Protocols
Part V : Collaborative Signal Processing
SNU INC lab.
2
Part IV
Sensor Network Protocols
SNU INC lab.
3
Introduction

WSN protocols
– Primary theme
• long-lived
• massively-distributed

Minimize duty cycle and communication
–
Adaptive MAC
–
Adaptive Topology
–
Routing
SNU INC lab.
4
MAC in Sensor Nets

Important attributes of MAC protocols
–
–
–
–
–
–
–
Energy efficiency
Collision avoidance
Scalability in node density
Latency
Fairness
Throughput
Bandwidth utilization
SNU INC lab.
5
Identifying the Energy Consumers

Major source of energy waste
–
–
–
–
Idle listening when no sensing events
Collisions
Control overhead
Overhearing
SNU INC lab.
6
Sensor-MAC(SMAC)

Major components of S-MAC
–
–
–
–

Periodic listen and sleep
Collision avoidance
Overhearing avoidance
Message passing
Periodic listen and sleep
listen
sleep
listen
sleep
– Turn off radio when sleeping
– Reduce duty cycle to ~10% (200 ms on/2s off)
– Increased latency for reduced energy
SNU INC lab.
7
SMAC - Collision Avoidance

Collision Avoidance
– Problem:
• Multiple senders want to talk
– Solution: Similar to IEEE 802.11 ad hoc mode (DCF)
•
•
•
•
Physical and virtual carrier sense
Randomized backoff time
RTS/CTS for hidden terminal problem
RTS/CTS/DATA/ACK sequence
SNU INC lab.
8
Adaptive Topology

Goal:
–
–
–

Exploit high density (over) deployment to extend system lifetime
Provide topology that adapts to the application needs
Self-configuring system that adapts to environment
How many nodes to activate?
SNU INC lab.
9
ASCENT : Adaptive Self-Configuring
sEnsor Networks Topologies

The nodes can be in active or passive state.
– Active nodes
• forward data packets
– Passive nodes
• do not forward any packets but may sleep or collect network
measurements.
Data Message
Help
Messages
Source
Sink
Neighbor
Announcements
Messages
Source
Passive Neighbor
(a) Communication Hole
Data
Message
Sink
Sink
Active Neighbor
(b) Self-configuration transition
SNU INC lab.
Source
(c) Final State
10
STEM : Sparse Topology and Energy
Management

Major Concept
– Need to separate Wakeup and Data Forwarding Planes
– Chosen two separate radios for the two planes
– Use separate radio for the paging channel to avoid interference with
regular data forwarding
– Trades off energy savings for path setup latency
Wakeup plane: f1
Data plane: f2
SNU INC lab.
11
Routing

Goal
– To disseminate data from sensor nodes to the sink node in
energy-awareness manner, hence, maximize the lifetime of
the sensor networks.

Problem Description
– Given a topology, how to route data?
– Traditional Ad hoc routing protocols doesn’t fit

Classification of Routing Protocols
– Data Centric Protocols
• SPIN , Directed Diffusion
– Hierarchical Protocols
• LEACH , TEEN
– Location Based Protocols
• GAF , GEAR
SNU INC lab.
12
Data Centric Routing



The ability to query a set of sensor nodes
Attribute-based naming
Data aggregation during relaying
SNU INC lab.
13
Directed Diffusion




Sink node floods named “interest” with larger update interval
Sensor node sends back data via “gradients”
Sink node then sends the same “interest” with smaller update
interval
Query-driven
SNU INC lab.
14
Energy Efficient Routing

Possible Route
• Route 1: Sink-A-B-T, total PA = 4, total α = 3
• Route 2: Sink-A-B-C-T, total PA = 6, total α = 6
• Route 3: Sink-D-T, total PA = 3, total α = 4
• Route 4: Sink-E-F-T, total PA = 5, total α = 6
Maximum PA route: 4
Minimum hop route: 3
Minimum energy route: 1
SNU INC lab.
15
Database Centric Approach

Traditional Approach
– Data is extracted from sensors and stored on a front-end server
– Query processing takes place on the front-end

Sensor Database System
– Distributed query processing over a sensor network
Sensor
DB
Sensor
DB
Sensor
DB
Warehouse
Front
End
Sensor
DB
Sensor
DB
Front
End
SNU INC lab.
16
Sensor DB Architecture
SNU INC lab.
17
Part II
Collaborative Signal Processing
SNU INC lab.
18
Introduction

Sensor Network from SP perspective
– Provide a virtual map of the physical world:
• Monitoring a region in a variety of sensing modalities
• (acoustic, seismic, thermal, …)

Two key components:
– Networking and routing of information
– Collaborative signal processing (CSP) for extracting and
processing information from the physical world
SNU INC lab.
19
Space-Time sampling

Space
Space

Sensors sample the spatial signal field in a particular
modality (e.g., acoustic,seismic)
Sensor field decomposed into space-time cells to
enable distributed signal processing (multiple nodes
per cell)
Time
Time
Uniform space-time cells
Non-uniform space-time cells
SNU INC lab.
20
Single Target Tracking
Initialization: Cells A,B,C and D are put
on detection alert for a specified period
Five-step procedure:
1. A track is initiated when a target is
detected in a cell (Cell A – Active cell).
Detector outputs of active nodes are sent
to the manager node
2. Manager node estimates target
location at N successive time instants
using outputs of active nodes in Cell A.
3. Target locations are used to predict
target location at M<N future time
instants
4.Predicted positions are used to create
new cells that are put on detection alert
5.Once a new cell detects the target it
becomes the active cell
SNU INC lab.
21
Why CSP?

More information about a phenomenon can be
gathered from multiple measurements
– Multiple sensing modalities (acoustic, seismic, etc.)
– Multiple nodes

Limited local information gathered by a single node
– Inconsistencies between measurements
– malfunctioning nodes

Variability in signal characteristics and environmental
conditions
– Complementary information from multiple measurements can
improve performance
SNU INC lab.
22
Various Forms of CSP

Single Node, Multiple Modality (SN, MM)
–

Simplest form of CSP: no communication burden
• Decision fusion
• Data fusion (higher computational burden)
Multiple Node, Single Modality (MN, SM)
–
x2
x 2,1
x1,1
Higher communication burden
• Decision fusion
• Data fusion (higher computational burden)

x1
Manager x 3,1
node
Multiple Node, Multiple Modality (MN, MM)
–
Highest communication and computational burden
• Decision fusion across modalities and nodes
• Data fusion across modalities, decision fusion across nodes
• Data fusion across modalities and nodes
x1,1 x1, 2
x 3,1 x 3, 2
SNU INC lab.
x 2,1 x 2 , 2
Manager
node
23
Event Detection

Simple energy detector
– Detect a target/event when the output exceeds an adaptive
threshold (CFAR)

Detector output:
– At any instant is the average energy in a certain window
– Is sampled at a certain rate based on a priori estimate of target
velocity and signal bandwidth

Output parameters for each event:
– max value (CPA – closest point of approach)
– time stamps for: onset, max, offset
– time series for classification

Multi-node and multi-modality collaboration
SNU INC lab.
24
Constant False Alarm Rate (CFAR)
Detection




Energy detector is designed to maintain a CFAR
Detector threshold is adapted to the statistics of the
decision variable under noise hypothesis
Let x[n] denote a sensor time series
Energy detector:
W 1
Target present (H1 )
 N( s ,  2 )
2
e[n ]   x[n  k ] ~ 
 N( n ,  2 )
k 0
(H 0 )
Target absent
W is the detector window length

Detector decision:
e[n ]  
e[n ]  
Target present
Target absent
SNU INC lab.
25
Single Measurement Classifier
M=3 classes
P(x | 1 )
x
Event feature
vector
P(x | 2 )
C(x)=2
P(x | 3 )
Class likelihoods
SNU INC lab.
Decision
(max)
26
Multiple Measurement Classifier
Data Fusion
M=3 classes
x1
x2
Event feature
vectors from 2
measurements
x1 
x 
x 2 
P(x | 1 )
P(x | 2 )
C(x)=3
P(x | 3 )
Concatenated
event feature Class likelihoods
vector
SNU INC lab.
Decision
(max)
27
Multiple Measurement Classifier –
Soft Decision Fusion
P(x1 | 1 )
x1
P(x1 | 2 )
Comb.
P(x1 | 3 )
P(x2 | 1 )
x2
Event feature
vectors from 2
measurements
P(x2 | 2 )
C(x)=1
Comb.
Comb.
Component Final
Decision
decision
(max)
combiner
P(x 2 | 3 )
SNU INC lab.
28
Multiple Measurement Classifier –
Hard Decision Fusion
M=3 classes
x1
x2
x3
Event feature
vectors from 3
measurements
C1 (x1 )
C2 (x2 )
1
C(x)=1
3
Majority vote
1
Final
decision
C3 ( x 3 )
Component
hard decisions
SNU INC lab.
29
Summary

WSN protocols
– MAC
– Routing

WSN CSP
– Data Fusion
– Decision Fusion
SNU INC lab.
30