decentralized reactive clustering in sensor networks yingyue xu april 26, 2015

33
Decentralized Reactive Clustering in Sensor Networks Yingyue Xu March 21, 2022

Upload: damaris-merithew

Post on 14-Dec-2015

213 views

Category:

Documents


0 download

TRANSCRIPT

Decentralized Reactive Clustering in Sensor Networks

Yingyue XuApril 18, 2023

2

Wireless Sensor Networks

Consists of battery operated sensor nodesDeployed randomly on the field in large numbers Resource constrained sensor nodes

– power, computation, communicationNo networking infrastructure

3

MOTETwo Board Sandwich– CPU/Radio board– Sensor Board: temperature, light

Size– Mote: 11 in– Pocket PC: 5.23.1 in

CPU– Mote: 4 MHz, 8 bit– Pocket PC: 133 MHz, 32 bit

Memory– Mote: 512 B RAM; 8K ROM– Pocket PC: 32 MB RAM; 16 MB ROM

Radio– 900 Hz, 19.2 kbps– Bluetooth: 433.8 kbps (symmetric)

Lifetime (Power) Mote: 3-65 days Pocket PC: 8 hrs

Cost Mote: $100 Pocket PC: $400

4

Motivation: CSIP in sensor networks

Collaborative Signal and Information Processing (CSIP)Each node senses the event individually, need to collaborate with other nodes to exchange information and get final and accurate resultsChallenges– Energy Efficiency– Scalability– Reliability– Real-time Performance

How?

5

Architectural overview of the CSIP system

Wireless Communication

Routing Layer

Clustering & Distributed Computing Paradigm

Collaborative Information processing

6

Computing paradigms in WSN - Client/Server based paradigm

Client/Server paradigmCentralized processingPowerful central serverNot energy efficientNetwork traffic – highDifficult to reconfigure

7

Computing paradigms in WSN - Mobile agent based paradigm

Agent travels to each nodeNo centralized nodeReduced network trafficScalable networkAdaptive to dynamic WSN environment like node failures, etc.

8

Using Network Simulator 2 (ns-2)Randomly deployed in a 10m by 10m areaMAC layer protocol: 802.11Routing protocol: DSDVMetrics:

- Execution Time - Energy Consumption

8 experiments are designedResults: When the number of nodes is large, the mobile agent paradigm performs better; But when the number is small, the client/server paradigm is better

Performance Evaluation

9

Hybrid Computing Paradigms

Scheme A: mobile agent paradigm within clusters

and client/server paradigm between

cluster heads

Scheme B: client/server paradigm within clusters

and mobile agent paradigm between cluster

heads

10

Clustering

Partition the network into clusters– Achieving scalability– Easer to management– Increase lifetime and energy efficiency

Nodes communicate with clusterheads, clusterheads communicate with processing center

11

State of Arts and Problems in Sensor Networks

SPAN, GAF– Need to know position of the node in advance

LEACH– Random rotation of cluster head– Need time synchronized in advance

Two level Clustering Algorithm– Estrin’s method– Not event driven, Proactive

12

Problems of Existing Clustering Protocols

Proactive clustering results in– unnecessary radio transmission– large transmission power to reach cluster head, lack of

intermediate nodes acting as routing nodes,– energy inefficiency event

w ake up nodes

sleeping nodes

cluster head

13

Benefits of DRC

Reactive clustering - driven by eventsLocalized protocolEnergy-efficientLess transmission power to reach cluster head

event

w ake up nodes

sleeping nodes

cluster head

14

Benefits of DRCevent event

(A) A predefined clustering (B) Clustering after DRC

15

Message Format

TYPE: the type of message which can be REQUEST, REPLY, JOIN, JOIN-FORWARD and END.Power Level: the transmission power the node currently uses.Destination ID: the destination node identification and we use all 1's as the broadcast address.Source ID: the address of current node. Cluster ID: the cluster head address of the cluster the current node belongs to and we use 0 if the node is unclustered.Energy: the remaining energy of the node.Signal Energy: the signal energy sensed by the current node emitted by the potential target.

TYP E(4 b its)

P owerLeve l

(4 b its)

D estina tionID

(2 bytes)

S ource ID(2 bytes)

C luster ID(2 bytes)

E nergy(4 bytes)

S igna lE nergy

(4 bytes)

Messages are exchanged only locally

16

Desirable Features of DRC

Reactive clustering driven by the eventsUses power control technique to minimize the transmission powerA localized clustering protocol that simple local node behavior achieves a desired global objective

17

Outline of DRC

Post-deployment PhaseCluster Forming Phase– Wait for a time– Broadcast REQUEST– Increase transmission power and rebroadcast if

receive no message in a certain time– 4 scenarios in determine the clusterhead– A timer in cluster head determines the end of this

phase

18

Outline of DRC

Intra-cluster Data Processing PhaseCluster Head to Processing Center Phase

19

Different Scenarios

Node A, B unclusteredChoose one with higher energy as cluster head

Node A clustered, node B unclustered

B join the cluster A belongs to

Node A unclustered, node B clustered

A join the cluster B belongs to

Node A, B clusteredB discard the message

20

Simulation of DRC

In Java3 Classes– Simu.java (main class)– Node.java (node object)– Message.java (message object)

Simple communication model– No routing, transmission error

21

States of Nodes

Sleep

Idle

Undecide

ClusterHead

Member

22

Flow chart

Parameters initialize

Generate nodes

Nodes deployment

Calculate distance

Run protocol (DRC, LEACH or Fix)

Change target position

Display results

23

Parameters in Simulation

30 by 30 areaTransmit power level: 8Initial energy: 36 joulesData size: 8000 bitsMobile agent size: 800 bitsMessage size: 152 bits

24

Node deployment

RandomGrid

25

LEACH

Generate random number

Calculate threshold

Determine clusterhead

Use client/server to transfer data

Divide the network into random number of clustersUse client/server paradigm in each clusterNodes are driven by events and go back to sleep after sending data

Change target position

26

Fix clustering

Determine cluster head

Member sends data to cluster head

Change target position

Divide the network into 4 fixed clustersUse client/server paradigm in each clusterUse highest level of transmission powerNodes are driven by events and go back to sleep after sending data

27

State Machines

(A) Operation of fix clustering (B) Operation of LEACH and DRC

28

Output results

Each node’s ID, status, energy and clusterIDTotal energy consumptionLifetime of the network

29

Number of Nodes

DRC consumes less energy and has longer lifetime than LEACH and predefined

clustering

(A) Energy Consumption (B) Network Lifetime

30

Target Speed

DRC performs better than LEACH and predefined clustering

(A) Energy Consumption (B) Network Lifetime

31

Signal Range

More nodes will wake up as signal range increases, thus consume more energy

(A) Energy Consumption (B) Network Lifetime

32

Number of Events

More events will cause more nodes to wake up and thus consume more energy

(A) Energy Consumption (B) Network Lifetime

33

Future Directions

Implement DRC into UCB’s Mote nodes– Energy consumption– Network lifetime

Setup a testbed of Motes and WINSNG nodesDo experiments and measure results– Energy consumption– Network lifetime