Reducing Energy Consumption in Wireless Sensor Networks · Reducing Energy Consumption in Wireless...

Carlo FischioneReducing Energy Consumption

in Wireless Sensor Networks

Reducing Energy Consumptionin Wireless Sensor Networks

CHESS seminarDecember 11, 2007

Carlo Fischione

University of California at Berkeleye-mail: [email protected]



Introductory Overview

Minimum Energy Coding

Radio Power Control

Bit Error Rate

Energy Consumption

Numerical Results

Conclusions and Future Work

UCB December 11, 2007

Carlo Fischione

Definition of WSNsDefinition of WSNs

“Wireless Sensor Networks (WSNs) are networks forcommunication, control, sensing and actuation by smallnodes communicating through wireless links”

Sensor Networks

Gateway

Network Infrastructure

Application

Reducing Energy Consumptionin Wireless Sensor NetworksUCB December 11, 2007

Carlo Fischione

ApplicationsApplications

Tremendous space of applications:

¬ Monitoring space: ocean water, pollution, ...¬ Monitoring things: robots, human body,…


Carlo Fischione

• WSNs are an area of active research from electronic to computer sciencesince few years.

• Many industries are now investing in this new technology:¬ ABB

¬ Fiat

¬ Pirelli

¬ Siemens

¬ United Technology

¬ …


DARPA DSN node (~1960) Mica node (~2000)

Tmote-sky node (~2003)

History of WSNsHistory of WSNs


Carlo Fischione

Design Principles of WSNsDesign Principles of WSNs


• Coding • Radio Power• Routing• Modulation•…

Communication

• Optimization• Hybrid Systems• Networked Control• System Identification•…

Control• Embedded Software• Middleware• Operating Systems•…

Computer Science

WSN

• Formal Verification• Logic Synthesis• Microprocessors•…

CAD


Carlo Fischione

Some Challenges...Some Challenges...

♣ How to schedule the link each node should select to forward packets?

♣ How to quantize the measurements to transmit the minimum amount ofinformation in the presence of uncertainties of the network?

♣ How to do radio power and rate control, when neither accurate channelmodels and channel state information are available?

♣ How to access to the channel?


COSI project: Synthesis of Embedded Networks for Building Automation and Control

A. Pinto & A. Sangivanni-Vincentelli, UC Berkeley

Carlo Fischione

Design Principles of WSNsDesign Principles of WSNs……

♣ Design of WSNs includes several techniques (distributed computation,distributed source coding, distributed control,...).

♣ Cross-layer approaches are necessary to take into account severalinteracting techniques and protocols.

PlatformDesign-Space

Export

PlatformMapping

Architectural Space

Application Space

Application Instance

Platform Instance

System (Software + Hardware)Platform

CognitiveRadio

Cross Layer Design

Platform BasedDesign

A. Sangiovanni-Vincentelli et al.






Radio Power Control

Bit Error Rate

Energy Consumption

Numerical Results



Carlo Fischione

Minimum Energy CodingMinimum Energy Coding

♣ Performance analysis of minimum energy coding schemes in CodedDivision Multiple Access (CDMA) wireless sensor networks:

1. Minimum Energy coding (ME);

2. Modified Minimum Energy coding (MME).

♣ Detailed models of the energy consumption and bit error rate:1. coding schemes;2. wireless channel;3. power control;4. hardware characteristics of the transceivers.


Carlo Fischione

System DescriptionSystem Description

♣ Data sensed by a node are coded with a Minimum Energy coding scheme.

♣ The bits of the ME coded data are OOK modulated: only bits having value 1are transmitted after a DS-CDMA spreading operation.

Source

Tx 1

Rx 1 1

Tx 2

Rx 2 2

Tx K

Rx K K

Transmitter-receive pairs

Tx i

Rx i i


100111 1000010000Source Codeword ME Codeword


Carlo Fischione

ME CodingME Coding

♣ In ME Coding (Erin and Asada: “Energy Optimal Codes for WirelessCommunications” in 38th IEEE CDC 99) a source codeword is mappedinto a new codeword having larger length but less number of 1 (or high)bits.

♣ Source codewords having large probability of occurrence are associatedto ME codewords with less high bits.

♣ Only high bits are transmitted with OOK modulation.

Source codeword

ME codeword


Tx i

Rx i i


Carlo Fischione

♣ Only bits having value 1 are transmitted

♣ is the average transmit time per codeword.

OOK ModulationOOK Modulation

• Lower Activity• Lower interference• Higher bit error rate

Tx i

Rx i i


Carlo Fischione

ME Coding (2)ME Coding (2)

♣ Energy consumption per ME codeword

♣ ME coding increases the value of three systemparameters:1. the codeword length;

2. the Transmitter active time (it is negligible with respect to the radiopower consumption);

3. the Receiver active time: it is not negligible, the power spent toreceive is approximately the same as that used to transmit.


Carlo Fischione

♣ MME coding exploits a structure of the codeword that allows thereceiver to go in a sleep state: Kim and Andrews: “An EnergyEfficient Source Coding and Modulation Scheme for Wireless SensorNetworks”. In IEEE WSPAWC, 2005.

♣ In MME coding, energy is saved not only at the transmitter as MEcoding, but also at the receiver.

MME CodingMME Coding


Tx i

Rx i i


Carlo Fischione

MME coding: the ME codewords are partitioned into sub-frames starting with anindicator bit.

Indicator bit = 1 => there are not high bits in the sub-frame, there is no need fordecoding, and the receiver goes to sleep.

Indicator bit = 0 => there are high bits in the sub-frame, so the decodingoperation must be performed, and the receiver cannot go to sleep.

MME Coding (2)MME Coding (2)

Tx i

Rx i i

MME codeword

Source codeword


Carlo Fischione

MME Coding (3)MME Coding (3)

♣ MEE energy consumption per codeword:

♣ = average receiver activity (it is a function of the numberof sub-frames, and bit error rate)


Carlo Fischione

Energy of ME and MMEEnergy of ME and MME


Tx i

Rx i i

Tx i

Rx i i






Radio Power Control

Bit Error Rate

Energy Consumption

Numerical Results



Carlo Fischione

Radio Power ControlRadio Power Control

♣ Efficient power control algorithms for wireless sensor networks(WSNs) are crucial to reduce energy consumption.

¬ Lifetime: in scenario where recharging is not possible, radio powercontrol is very important to increase the network lifetime.

¬ Collisions: power control is beneficial to reduce packet collisions(more retransmitted packets means wasting energy), and increasethroughput.

♣ Energy for radio power is responsible from 37% up to 85% ofthe total Energy consumption for off-the-shelf sensor nodes.


Carlo Fischione

Radio Power Control (2)Radio Power Control (2)

Node K

Node 1TX

SINRi(t)

Node i

Multiple Access Interference

Power-Rate updating

SINR Estimation

+

Delay

xSINR(t)

+

Node i

-~SINRSINR

th

Quantization

Channel attenuation

Tx 1Rx 1

1

Tx 2Rx 2

2

Tx KRx K

K


Tx i

Rx i i


Carlo Fischione


♣ Received signal

Desired Signal

MAI

Thermal Noise

Transmitter Activity (ME or MME)


Tx 1Rx 1

1

Tx 2Rx 2

2

Tx KRx K

K


Tx i

Rx i i

Delay

SINR Estimation

+x

+ Node i

Controllingnode

-+Quantization

Power updating

Power to transmit data packets.

Carlo Fischione

♣ The SINR of the signal of a node is expressed as

The SINRThe SINR

Vector of Binary on/off Processes:ME and MME coding

t

SINR(t)

SINR threshold

Outages

The Signal to Interference + Noise Ratio

UCB December 11, 2007Reducing Energy Consumption


Carlo Fischione

The Optimization ProblemThe Optimization Problem

♣Mixed integer-real optimization problem


SINR

Outages

• K-L. Hsiung, S-J. Kim, S. Boyd, “Power Control in Lognormal Fading Wireless Channel with Uptime Probability Specifications via Robust Geometric Programming”, ACC 2005.• J. Papandriopoulos, J. Evans, S. Dey, “Outage-Based Power Control for Generalized Multiuser Fading Channels”, IEEE Trans. Comm. 2006.

Delay

SINR Estimation

+x

+ Node i

Controllingnode

-+Quantization

Power updating


Carlo Fischione

The Optimization Problem (2)The Optimization Problem (2)

♣ How to express the constraints on the outage probability?

♣ How to solve efficiently the problem?¬ good precision of the solution (otherwise waste of

resources);¬ computationally affordable (WSNs have limited

processing capabilities).



Delay

SINR Estimation

+x

+ Node i

Controllingnode

-+Quantization

Power updating


t

SINR(t)

SINR threshold

Outages

The Signal to Interference + Noise Ratio

Carlo Fischione

♣ The SINR is a combination of log-normal processes weighted with on/offbinary random processes, it is approximated with an overall Log-normal:

Outage Probability ConstraintOutage Probability Constraint

Interference Function

Fischione, Graziosi, Santucci,IEEE Trans. Comm., October 2007


SINR

Outages

Carlo Fischione

♣ Main Problem The Relaxation Problem

♣ Any pair of vectors and that solves the relaxation problem, with a costfunction less then , provides a feasible solution of the original problem.

♣ Proposition 1: if then for all i and .

♣ Theorem: If the pair is a solution of the relaxation problem, then it verifiesthe following equations:

Relaxation ProblemRelaxation Problem



Carlo Fischione

♣ Proposition 2: if is feasible, the pair can be a solution of theoriginal problem only if

The proposition says that, given a feasible vector , the search for thefeasible solutions can be restricted to the set

♣ Proposition 3: if is unfeasible, then any is unfeasible aswell.

The proposition says that, if is unfeasible, the set of unfeasible solutionscan be reduced as follows:

Branch-and-bound SolutionBranch-and-bound Solution


Carlo Fischione


♣ T Relaxation Problem in solved by contraction mappings (Fischione,Butussi, IEEE ICC 2007)

♣ A branch-and-bound algorithm builds the set of feasible andunfeasible rates by solving the Relaxation Problem, collecting feasiblesolutions, and reducing the candidate solutions via Proposition 2 and3.

♣ It can be proved that the algorithm converges within a random timeshorter then an exhaustive search over the initial set of possible rates.


Relaxation Problem

Carlo Fischione

An exhaustive search over all rate vectors would have required solutions of the relaxation problem.

Numerical Results for PowerNumerical Results for PowerControlControl

Convergence of the Relaxation Problem

Number of rate vectors explored (solutions of the Relaxation Problem) to get the optimal solution with our proposed method:







Radio Power Control

Bit Error Rate

Energy Consumption

Numerical Results



Carlo Fischione

Bit Error RateBit Error Rate

♣ Decision variable at the receiver

♣ The decision variable is erroneously decoded:

¬ as a high bit, when a low bit was transmitted:

¬ as a low bit when a high bit was transmitted:


Tx 1Rx 1

1

Tx 2Rx 2

2

Tx KRx K

K


Tx i

Rx i i

Carlo Fischione

Bit Error Rate (2)Bit Error Rate (2)

♣ The probabilities are computed adopting the usual standard Gaussianapproximation, where is modelled as a Gaussian random variableconditioned to the distribution of the channel coefficients and coding.


Carlo Fischione

Bit Error Rate (4)Bit Error Rate (4)

♣ The Bit error probability is computed using the Log NormalSINR expression, and the Stirling approximation for thestatistical Expectation:

♣ It can be proved that the BER is a convex function.¬ The gradient method is used to derive the optimal decision

threshold.Reducing Energy Consumption

in Wireless Sensor NetworksUCB December 11, 2007

Carlo Fischione

Bit Error Rate of ME and MMEBit Error Rate of ME and MME

The characterization of can be used tocompute the BER of

1. ME:

2. MME: the coding structure,


Carlo Fischione

♣ Introductory Overview

♣ Main Contribution

♣ System Description

♣ Radio Power Control

♣ Bit Error Rate

♣ Energy Consumption

♣ Numerical Results

♣ Conclusions and Future Work


Carlo Fischione

Average ME Energy ConsumptionAverage ME Energy Consumption

Tx 1Rx 1

1

Tx 2Rx 2

2

Tx KRx K

K


Tx i

Rx i i


Carlo Fischione

Average Energy MMEAverage Energy MMEConsumptionConsumption

♣The energy is function of the BER

MME Energy Gain

Receiver activity per MME codeword


Carlo Fischione

♣ Introductory Overview

♣ Main Contribution

♣ System Description

♣ Radio Power Control

♣ Bit Error Rate

♣ Energy Consumption

♣ Numerical Results

♣ Conclusions and Future Work


Carlo Fischione

Numerical ResultsNumerical Results

CC2420 was taken as reference for the energyparameters Number of transmitter receiver pairs

Path loss at reference distance

Spreading gain

Noise Variance

SINR Threshold

Bit rate


Carlo Fischione

Radio Power OptimizationRadio Power Optimization

♣ Convergence of the power minimization algorithm. On thex-axis is reported the number of iterations.

Tx 1

Rx 1 1

Tx 2

Rx 2 2

Tx K

Rx K K


Tx i

Rx i i


Relaxation Problem

Carlo Fischione

BERBER

♣ Bit error probability for the ME and MME cases (note thatthey overlap).

Tx i

Rx i i


Carlo Fischione

Energy ConsumptionEnergy Consumption

♣ MME Energy Gain as function of the sub-frame length, fordifferent values of the transmitter activity

Tx i

Rx i i


Carlo Fischione

Conclusions and Future WorkConclusions and Future Work

♣ A general framework for accurate analysis of the Energy Consumptionof ME and MME coding has been proposed.

¬ Accurate Energy model and wireless propagation scenario.

¬ Distributed power minimization strategy.

¬ Optimal decision thresholds.

♣ Future studies¬ delay characterization.¬ distributed estimation.¬ simpler power control (geometric programming...)


Carlo Fischione

AcknowledgementsAcknowledgements

♣ M. Butussi

♣ K. H. Johannson

♣ F. Graziosi

♣ A. Sangiovanni-Vincentelli

♣ F. Santucci

♣ B. Zurita Ares


Carlo Fischione

QuestionsQuestions

Thank you for your attention!

Any questions?


Reducing Energy Consumption in Wireless Sensor Networks · Reducing Energy Consumption in Wireless...

Documents

Transcript of Reducing Energy Consumption in Wireless Sensor Networks · Reducing Energy Consumption in Wireless...