how to make a sensor network live longer?

How to make a sensor network live longer?Presentator: Yibo SunCourse prof.: Kyoung-Don Kang

Agenda

The definition of Lifetime of sensor networks

To make the whole network live longer -- Energy balancing strategy To make an application live longer -- Energy optimization for target area

What is the lifetime?

The time till all nodes die out? The average of nodes’ lifetime? The time for the system to work

properly? The time for an application to work

properly?

Some definition of Lifetime

[Blough,et al. 02], defines the Lifetime of the sensor network as

min {t1,t2,t3} t1 is the time it takes for the cardinality of the larges

t connected components to drop below c1*n(t), where n(t) is the number of alive nodes at time t

t2 is the time it takes for n(t) to drop below c2*n(0) t3 is the time it takes for the area covered to drop be

low c3*L^d 0 <= c1 ,c2, c3 <= 1

setting c1 =0 and c2 = 1 corresponds to defining lifetime as the time it takes for the first node to die

setting c1 =1 and c2 =0 corresponds to defining lifetime as the time to network disconnection.

E.GNode#1 is the only sinkNode#6’s lifetime = The Lifetime

Definition by coverage ratio

By setting c1=c2=0, and c3=q, then we obtain the definition of network lifetime given in [Zhang,et al. 04] , which defines lifetime as:

the entire interval in which at least q portion of the region R is covered by at least one sensor node. (q =1 indicates full coverage)

Definition by target node/area [Duarte-Melo, et al. 02] defines the lifetime of a sens

or network as the expected lifetime of any given sensor in the network. In a densely deployed sensor network this definition is extend to be the time until a certain percentage of the sensors died.

In [Ye, et al. 02] The lifetime is defined as the time it takes for the coverage (defined as the ratio of the area covered by working nodes to the total area) to drop below, and never exceed again, a pre-determined threshold.

“The Lifetime”

Hence, different application has different lifetime definition

“The Lifetime” here is defined by how long the target function unit works properly.

i.e. The lifetime of a subgraph G’(V) in whole graph G(V)

Target Function Unit The nodes near the event together

with the notes on the route (red line)

The target nodes’ lifetime = The lifetime

How to make a whole network live longer? Main ideas:

Reduce the total power consumptionEfficiently transmit the data packetsSynchronously consuming all nodes’ ene

rgy [Joongseok, et al, 2005] Maximum lifet

ime is NP-hard!

Minimum Energy vs. Maximum Lifetime

Lifetime under minimum energy routing is 67% of that under maximum lifetime routing

Some assumptions

Energy consumption model: consider packet sending and receiving

only Simplified Radio Model:

Radio transceiver, Micro Controller, Energy Source

In fixed transmit radio power level, energy used to send/receive one bit is

Transmiting a k-bit packet a distance d costs

Receive a k-bit packet costs

In real propagation model, E = k*dc, 2<c<4

bitnJEelec /502//100 mbitpJamp

2) *),()(,( dkkEdkEkEdkE electxamptxelectx

kEkEkE elecrxelecrx *)(( )

An experiment on lifetime of routing protocols of different category Direct communication protocol Minimum-transmission-energy

routing protocol Clustering

Direct communication can be generated to protocols based on

multiple base stations or anchor nodes Energy used on sending one k-bit message

is: 2) *),()(,( dkkEdkEkEdkE electxamptxelectx

MTE Shortest path GPSR Energy used on sending one k-bit message is:

Each node consumes:)'(**))',()((*',( 2

) dEkndkEkEndkE elechopstxamptxelechopstx

)'*(*)',()(',( 2) dEkdkEkEdkE electxamptxelectx

Clustering Energy consumed when sending a k-bit

message:

Each node consumes:

)*(**)),()((*,(2''''

)'' dEkmdkEkEmdkE elecsclusterhoptxamptxelecsclusterhoptx

)*(*),()(,(2''''

)'' dEkdkEkEdkE electxamptxelectx

The application dies long before the last node dies!

How about energy awaring routing? Minimum Total Transmission Power Routing

(MTPR) Attempts to reduce the total transmission power per

packet. Prefers routes with more hops with short transmission

ranges than few hops with longer transmission rage.

Problem: cost more extra energy , delay, not scalable Min-Max Battery Cost Routing (MMBCR)

Consider the remaining battery power of nodes as metric. Nodes with high residual capacity participate in routing

more often. And prefers to choose a path whose weakest node has the maximum remaining power.

Conditional MMBCR

Set a parameter as threshold, if no node in the chosen route with MMBCR algorithm, whose battery capacity is lower than , MMBCR applied, else, use MTPR.

Seems end-to-end optimization is not practical.

Hop-to-Hop Optimization

Take energy consumption in routing metric

NADV [Seungjoon et al, MobiHoc’05] Select neighbors with optimal trade-off be

tween proximity and link cost

Advance: advantage by greedy option

Normalized Advance: Cost(n) = fraction of successful data tra

nsmission to neighbor n

GEAR (Geographically and Energy Aware Routing) Each node maintains a neighbor table

Energy levels and locations of each neighbor

Cost to transmit to each neighbor Packet is forwarded to neighbor with s

mallest cost

Why not balance the energy? Make clusters to be balanced in member

Balanced k-clustering [Soheil, et,al.Sensors 2002]

Make every node to be key nodeLow-Energy Adaptive Clustering Hierarchy

(LEACH) Combine direct communication and MTE

[Martin Haegnni, ISCAS '03 ]

Balanced K-clustering

Minimum cost flow question, O((n+k)3)

LEACH

Using randomization to distribute the energy load evenly

Break up operation into rounds Set-up phase

Cluster-head Advertisement Cluster Set-Up Transmission schedule creation

Steady-state phase Data transmission to cluster head Signal prosessing (Data fusion) Data transmission to the base station

Combine direct communication and MTE Node i transmits locally generated pac

ket to next neighbor with probability ai

,and directly to the sink with bi = 1 - ai .

Incoming packet always forward to the neighbor

Goal: choose ai to achieve energy balancing

Assume distance between node is the same

By solving

In 5 nodes: b1…b5 are 0.0301,0.0438,0.0694,0.1250,0

In 10 nodes: 14% lifetime increased with an extra energy consumption: 60%~80%

To make all the nodes live same shorter???

1

11

)1)1(()1(i

iki

i

kkii bbiNiabiNE

How to make an application live longer? Here, “an application” implies it car

es more about a set of nodes instead of allTurn off the redundant nodes down or ma

ke them to sleep…can be a choiceSet priority to different nodes and conside

r as a factor in routing. Thus to divide a subgraph from the graph.

An application-oriented GPSR version Set a VIP-rate p (0<=p<=1) to every node,

initially 0 Set a threshold h (not carefully

considered!) When event arises, nodes near the spot

are set their priority to a higher value, say 0.8And it send a VIP-awareness packet to its

neighbor

When an intermediate node want forward a packet

if there exists a greedy option, it compares the VIP-ratesIf higher, then do greedy forwardIf lower, then do primeter forward

If no greedy option, follow right hand rule

Maintenance all nodes maintain a single-hop neighbor table

At source: mode = greedy

Intermediate node: if (mode == greedy) {

greedy forwarding;if (no_greedy_option||greedy_option_VIPrate - this.VIPrat

e>h) mode = perimeter;

}if (mode == perimeter) {

if (have left local maxima && local maxima’s VIPrate - this.VIPrate<h) mode = greedy;

else (right-hand rule);}

Original Networks

Earthquake happen0.8

0.8

What we need is to optimize lifetime of this

subgraph

Spread the VIPrate or not?0.8

0.8

This node still suffers from large traffic

Earthquake moves…0.8

0.8

0.8

0.8

What we need is to optimize lifetime of this

subgraph

After optimization0.8

0.8

0.8

0.8

Work in lower cycle

A analysis by hand Before

optimization

23

1

7

5

1

1

5

9

1 3 21

1

7

1

3

3

Send 1 packet

Receive 1 packetForward 1 packet

Send 1 pakcet

Energy consume: 55

Whole energy: 92

After optimization

23

11

7

5

1

1

3

13

1 21

3

11

1

1

7

Energy consume: 49

Whole energy: 100Save for subgraph:

6Extra energy: 8

11

11

17

15

1

11

3

9

1

7

1

5

3 9

3

Energy consume: 32Whole energy: 98

Save for subgraph: 24Extra energy: 6

Thanks