l 3 (live & let live)- increasing longevity in sensor networks ee 228a professor walrand

Post on 02-Jan-2016

16 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

DESCRIPTION

L 3 (Live & Let Live)- Increasing Longevity in Sensor Networks EE 228A Professor Walrand Contributors: Tanya Roosta Anshuman Sharma. Outline. Introduction Problem Definition Existing Approaches Our Approach Future Work Conclusion Q&A. Outline. Introduction Problem Definition - PowerPoint PPT Presentation

TRANSCRIPT

LL3 3 (Live & Let Live)-(Live & Let Live)- Increasing Increasing Longevity in Sensor NetworksLongevity in Sensor Networks

EE 228A

Professor Walrand

Contributors:

Tanya Roosta

Anshuman Sharma

Introduction

Problem Definition

Existing Approaches

Our Approach

Future Work

Conclusion

Q&A

Outline

IntroductionIntroduction

Problem Definition

Existing Approaches

Our Approach

Future Work

Conclusion

Q&A

OutlineOutline

IntroductionIntroduction

What are sensor networks Networks comprised of hundreds to thousand

nodes, where each node is a sensor Examples of use include guidance and control,

data collection and aggregation Sensor nodes are designed to be

– Low cost– Non obtrusive– Dynamically reprogrammable

Introduction

Problem DefinitionProblem Definition

Existing Approaches

Our Approach

Future Work

Conclusion

Q&A

OutlineOutline

Problem DefinitionProblem Definition

Sensors must be lightweight and compact Limited Power Supply Replenishing power is not an option Important to minimize power consumption of each

node to maximize battery life and lifetime of entire network

Existing network protocols stress on QoS (high throughput and low delay) and high bandwidth efficiency

Problem Definition (cont…)Problem Definition (cont…)

Energy Consumption Energy consumption occurs in three domains: sensing,

data processing and communication. In a wireless sensor network, communication is the

major consumer of energy Example

For ground to ground transmission, it costs 3J to transmit 1 Kb over a distance of 100m. However, a general-purpose processor with 100 MIPS processing capability executes 300 million instructions for the same amount of energy

Problem Definition (cont…)Problem Definition (cont…)

Design ChallengesThree main classes

– Hardware– Wireless Networking– Application

Problem Definition (cont…)Problem Definition (cont…)

Routing in Wireless Networks: Revisited Direct Communication Protocol:

Each sensor sends its data directly to the base station

Multi-hop routing protocol (MTE)Nodes route data destined to the base station through intermediate nodes

At first look it seems that a multi-hop approach would be able conserve more power

Problem Definition (cont…)Problem Definition (cont…)

Multi-hop Routing Protocols Table-driven (proactive)

– Destination-Sequenced Distance-Vector Routing– Cluster Gateway Switch Routing– Wireless Routing Protocol

Source-initiated (Reactive)– Ad Hoc On-Demand Distance Vector Routing– Dynamic Source Routing– Temporally-Based Routing– Signal Stability Routing

Introduction

Problem Definition

Existing ApproachesExisting Approaches

Our Approach

Future Work

Conclusion

Q&A

OutlineOutline

Existing ApproachesExisting Approaches

Power-Aware Routing: Metrics Minimize energy consumed/packet: Minimizes the

total energy consumed over n nodes Maximize Time to Network Partition: A load

balancing problem so that the response time is minimized

Minimize Cost/Packet: Assigns a cost function to each node and minimizes the total cost of routing a packet from that node

Existing Approaches (cont…)Existing Approaches (cont…)

Routing in Clustered Multi-hop Networks Aggregate nodes into clusters controlled by a

cluster-head Clustering on the basis of either lowest-ID

distributed clustering algorithm or highest-connectivity algorithm

Within a cluster, a cluster-head controlled token protocol used to allocate channel.

Cluster Routing Protocol

Total system energy dissipated for the 100-node random network

Existing Approaches (cont…)Existing Approaches (cont…)

Adaptive Energy-Conserving Routing BECA

– Turn of radio power– Involvement of application layer information– Can increase latency and packet loss

AFECA– All the nodes do not need be involved– Exploiting node density– Can interchange nodes for routing purposes

Existing Approaches (cont…)Existing Approaches (cont…)

Adaptive Energy-Conserving Routing (cont…)

BECA– Nodes are in three possible states:

sleeping, listening, active.– Start in sleeping state. Radio is off.– After a certain time, transition to

listening state– If a node has data to transmit it

transitions to active state

Existing Approaches (cont…)Existing Approaches (cont…)

Adaptive Energy-Conserving Routing (cont…)

AFECA– Used in densely-populated networks– Each node estimates its neighborhood– Each node increases its sleeping time proportional

to the number of nodes in its neighborhood

BECA versus AODV for different values of sleeping time

The latency for unmodified AODV is fixed The latency grows roughly linearly The growth is slightly lower at higher traffic rates

Percentage of energy saved is (Er - Es) / Er

Less saving for higher traffic rates since more nodes in active mode High values of sleeping time give no energy improvement

PE is the loss rate PE=P/E where P is the size of data delivered and E is the total energy

consumed by all nodes We can use PE to determine an optimal value for the sleeping time

Assumption: Unlimited amount of energy in the nodes

As expected AFECA and BECA do worse in terms of latency and packet loss than unmodified AODV

AFECA has a better energy consumption than BECA as expected

AFECA aggressive power savings result in the consistently highest efficiency

BECA protocol is about 20% longer and AFECA is about 55% longer than unmodified AODV when the energy in the nodes is limited.

Assumption: The nodes have limited amount of power

OutlineOutline

IntroductionProblem DefinitionExisting ApproachesOur ApproachOur ApproachFuture WorkConclusionQ&A

Our ApproachOur Approach

Insight Computation is much cheaper than communication Use of distributed approach to reduce

– Total number of transmissions– Energy dissipated in the network

Application-level/ higher layer feedback is important

Establish trade-offs (complexity vs. performance improvement, etc)

Our Approach (cont…)Our Approach (cont…)

Radio Model (First Order)

ETx(k,d) =ETx-elec(k) + ETx-amp(k,d)

=Eelec*k + amp*k*d2

ERx(k) =ERx-elec(k)

=Eelec*k

ETx-elec = ERx-elec = Eelec (Energy dissipated to run Rx/Tx)

amp (Energy dissipated for amplifying to get good gain)Source: Energy-Efficient Communication Protocol for Wireless Microsensor Networks: MIT

Our Approach (cont…)Our Approach (cont…)

Additions to Radio Model Does not consider energy consumption while

radios are idle Inclusion of idle time based on experiments with

WaveLAN radios Most of the time the radio is idle, hence idle time

dominates energy consumption Add term idle (idle energy expended per unit

time)

Our Approach (cont…)Our Approach (cont…)

Important to determine critical transmission range Let there be

– n total nodes– k cliques that we intend to form

Use modified Prim algorithm to form cliques of at least 3 nodes

Why the magic number 3?

Our Approach (cont…)Our Approach (cont…)

Pick k nodes at random (for each of the k cliques) k nodes are temporary cluster-heads Start with some minimum radius of discovery - Goal is to discover a minimum of 3 nodes for each

clique Increments of , if cannot find any node in the

periphery After the first node is discovered it tries to look

for another node, incrementing by each time

Our Approach (cont…)Our Approach (cont…)

All three nodes then adjust their transmission power to reach other

This results in a Hamiltonian Cycle If more than 3 nodes are possible without

increasing power then OK to have > 3 nodes in clique

After forming cliques, use TDMA to allocate time-slots for nodes to be cluster-head.

The nodes also use TDMA to schedule updates to cluster-head (intra-clique communication).

Our Approach (cont…)Our Approach (cont…)

The other nodes are put to sleep (turn-off radios) when not communicating, similar to PAMAS

A cluster head is responsible for discovering other cliques and sharing information within the clique.

Possibility of adding multiple hierarchies depending upon the trade-off between complexity and advantages

Our Approach (cont…)Our Approach (cont…)

Considerations GPS is available but might not be viable Next generation design of Low power ICs can

make adjusting duty cycle easy Exploring node density as a measure of reducing

computation and communication overhead CDMA codes allow efficient use of the channel

bandwidth

OutlineOutline

IntroductionProblem DefinitionExisting ApproachesOur ApproachFuture WorkFuture WorkConclusionQ&A

Future WorkFuture Work

Evaluating model through simulations Tuning density to trade operational quality against

lifetime Using multiple sensor modalities to obtain robust

measurements Exploiting fixed environmental characteristics Using a more comprehensive radio model that

takes into account time to wake up from sleep cycles

Exploring of various benchmarks for “lifetime” of a network

OutlineOutline

IntroductionProblem DefinitionExisting ApproachesOur ApproachFuture WorkConclusionConclusionQ&A

ConclusionConclusion

Our model is based on work that has already been done

We exploit characteristics of proven approaches Simulations would provide a measure of

advantages incurred by using our approach

OutlineOutline

IntroductionProblem DefinitionExisting ApproachesOur ApproachFuture WorkConclusionQ&AQ&A

Q&AQ&A

top related