Page 1

IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS, VOL. 28, NO. 7, JULY 2009 1017

A Framework for Energy-Consumption-Based

Design Space Exploration for

Wireless Sensor Nodes

Sonali Chouhan, Student Member, IEEE, Ranjan Bose, Member, IEEE, and M. Balakrishnan, Senior Member, IEEE

Abstract—Inthispaper,wefirstestablishthat,inwirelesssensor

networks, operating over “small” distances, both computation

energy and radio energy influence the battery life. In such a sce-

nario, to evaluate the utility of error-correcting codes (ECCs) from

an energy perspective, one has to consider the energy consumed

in encoding–decoding and transmitting additional “redundant”

bits vis-à-vis the energy saved due to coding gain. This paper

presents a framework for evaluating various ECCs based on a

comprehensive energy model of a sensor node. The framework

supports exploration of sensor node design space with application-

and deployment-related parameters, like distance, bit error rate,

path loss exponent, as well as the modulation scheme and ECC

parameters. The exploration results show that, as compared to

the uncoded-data transmission, the energy-optimal ECC saves

15%–60% node energy for the given parameters.

Index Terms—Communication systems, design methodology,

energy management, error correction coding, modeling.

I. INTRODUCTION

A

to deploy, harmless, multiple use, and economical. The WSN

is being used in many applications today, e.g., habitat moni-

toring, disaster management, inventory management, medical

diagnosis, structural monitoring, and agriculture. In most of the

applications, sensor nodes are battery driven, and the energy

consumption of the sensor node determines the battery life.

The battery-driven nodes are easy to deploy, but at the same

time, changing batteries frequently is neither convenient nor

economical.Thepower andtheenergyconstraintarethecritical

issues in a wireless sensor node.

For different applications, the channel conditions and op-

erating frequency of the network are different as the WSN

deployment environment varies. Also, for every application, the

internode distance and required bit error rate (BER) could dif-

fer significantly. To achieve the desired BER, error-correcting

WIRELESS sensor network (WSN) is a network of

small sensor nodes, characterized by features like easy

Manuscript received August 22, 2008; revised January 6, 2009. Current

version published June 17, 2009. This paper was presented in part at the

ISLPED’08 Conference in August 2008 [1]. This paper was recommended by

Associate Editor V. Narayanan.

S. Chouhan and R. Bose are with the Department of Electrical Engineering,

Indian Institute of Technology Delhi, New Delhi 110 016, India (e-mail:

sonali@cse.iitd.ac.in; rbose@ee.iitd.ac.in).

M. Balakrishnan is with the Department of Computer Science and Engineer-

ing, Indian Institute of Technology Delhi, New Delhi 110 016, India (e-mail:

mbala@cse.iitd.ac.in).

Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.

Digital Object Identifier 10.1109/TCAD.2009.2018865

codes (ECCs) are widely used. While transmitting the encoded

data, the required signal-to-noise ratio (SNR) is less as com-

pared to sending the uncoded data. This saving in signal power

is known as coding gain. The overheads of ECC are encoding

and decoding of data and transmission of “redundant” bits.

We can trade off these energy overheads against the energy

gain due to the coding gain of ECC. With this tradeoff in

energy, a design space is created for selecting an appropriate

ECC, which not only maintains BER but may also reduce

the energy consumption of the sensor node. ECC, as well as

the modulation scheme, determines the transmission quality

in terms of BER. The energy consumption of a sensor node

is a complex function of internode distance, desired BER,

channel conditions, operating frequency, modulation schemes,

and ECCs. In this paper, we present a systematic approach to

explore such a design space and find an energy-optimal ECC.

Closely deployed nodes is an inherent characteristic of WSN,

and that makes it different from conventional wireless net-

works. At short internode distances, the signal transmission

energy required is not very high and is comparable to the energy

consumed in circuit components in transmitting and receiving

the signal. Interestingly, the computation energies of the en-

coder and decoder are also comparable to the transmit signal

energy at short distances. Thus, for a comprehensive analysis of

energy consumption in WSN, all the three energy components,

namely, the signal, the circuit, and the computation, need to be

considered in an integrated manner.

Keeping in view the energy limitations for WSNs, many

researchers have explored existing ECCs with a particular mod-

ulation scheme for WSN. Balakrishnan et al. [2] have explored

the energy efficiency of various ECCs with field-programmable

gate-array (FPGA) and application-specific integrated-circuit

(ASIC) implementations of encoder–decoder. The authors have

found that power consumption with ASIC implementation is an

order of magnitude less than that with FPGA implementation.

Howard et al. [3] have analyzed the distance at which an

ECC becomes energy efficient for different environments and

operating frequencies. They have considered different decoders

implemented in ASIC. Although encoding or decoding con-

sumes an order of magnitude less power in ASICs than in

processors, but in commercial sensor nodes, low-power proces-

sors or microcontrollers are used [4] due to their versatility and

cost effectiveness. In processor-based studies, Lettieri et al. [5]

have proposed the use of forward error correction with

ARQ for low-power design. They have calculated the energy

0278-0070/$25.00 © 2009 IEEE

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on September 1, 2009 at 03:16 from IEEE Xplore. Restrictions apply.

Page 2

1018IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS, VOL. 28, NO. 7, JULY 2009

Fig. 1.Wireless sensor node design space exploration.

based on the number of instructions executed for a particular

encoding/decoding. Min and Chandrakasan [6] have proposed

a framework for energy-efficient communication, which con-

siders an empirical energy model for Viterbi decoding of

convolution codes (CCs). Shih et al. [7] have measured the

energy consumption of processor for CCs and found that, due

to large decoding overheads, CCs are not suitable for WSNs.

Thisdirect-measurement-based methodgives usgoodaccuracy,

but to explore a large design space, direct measurement is not

practical.

In this paper, a system-level methodology is presented for

exploring various ECCs with different modulation schemes

while taking into account the transmit signal energy, the circuit

energy, the encoder energy, as well as the decoder energy.

Based on the energy tradeoffs between radio and computation

functions, we find an energy-optimal ECC that reduces node

energy consumption and hence increases the node life. For this

exploration, a framework and a sensor node energy model have

been developed. In this framework, all significant parameters

to which a sensor node energy is sensitive to are taken into

consideration. Different possible application-driven scenarios

are considered. The overall methodology and the proposed

exploration framework have been described in Section II. The

models used for computing the different energy components are

dealtwithinSectionIII.Usingthismethodology, exploration of

various ECCs with different modulation schemes in different

operating conditions is presented in Section V. Section VI

summarizes the conclusions.

II. METHODOLOGY

To design an energy-efficient network, we need a framework

that allows us to explore the design space with various design

parameters. Fig. 1 shows the wireless sensor node design space

exploration methodology. Our design space parameters are

derived from the WSN application and associated constraints.

The specified application gives us a deployment range in terms

Fig. 2. Exploration framework.

of area, number of nodes, network topology, and channel

conditions. Constraints are the minimum quality of service

to be maintained and SNR. A library provides various op-

tions to the configuration selector for channel coding schemes,

compression techniques, modulation, and processor model. On

the basis of these inputs, the configuration selector decides

a particular configuration. For this system configuration, an

energy simulator calculates energy consumption. Feedback of

energy consumption pattern helps the configuration selector to

choose a better configuration that suits the needs. This process

repeats until we get the optimal energy for the system, under

the given application parameters and constraints.

A. Exploration Framework

To explore the design space for finding an energy-optimal

ECC for a sensor node, we propose a comprehensive explo-

ration framework (Fig. 2). In this framework, the optimal ECC

and modulation scheme is chosen on the basis of the node

energy per bit for the candidate configuration. To calculate the

sensor node energy per bit, we need an energy model of the

sensor node. To develop the sensor node energy model, we

identify the major energy components of a sensor node. In

any sensor node, energy is consumed by the power supply, the

sensor, the computation unit, and the radio unit. However, with

different applications and environments, power consumption

varies significantly only in the computation and the radio unit,

i.e., the power consumption of other units is assumed to be

relatively constant. In the computation unit, energy is spent

in performing encoding and decoding of information bits. The

radio unit energy is contributed by signal energy consumption

while communicating these encoded data and energy consumed

in the circuit components of the radio unit.

The radio energy model (RadioEM) and the computation

energy model (CompEM) are the two major modules of the

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on September 1, 2009 at 03:16 from IEEE Xplore. Restrictions apply.

Page 3

CHOUHAN et al.: FRAMEWORK FOR DESIGN SPACE EXPLORATION FOR WIRELESS SENSOR NODES1019

design space exploration framework. RadioEM consists of ra-

dio transceiver circuit energy and transmit signal energy. For a

particular ECC, the performance model provides the required

SNR for the desired BER. Specific channel models are used to

simulate different channel conditions. RadioEM takes distance,

operating frequency, transmitter–receiver antenna gains, packet

length, channel bandwidth, and modulation scheme as inputs.

All these inputs are either provided by the user as application

constraints or selected by the configuration selector, depending

on the target application.

For a given ECC, encoding and decoding functions are

implemented in a processor. Using the cycle-accurate processor

model and a processor simulator [8], energies for encoding and

decoding are computed in CompEM. Depending on the net-

work topology, encoder and decoder energies may be combined

appropriately. To cater to these situations, CompEM computes

encoder and decoder energies separately.

The exploration framework, using energy models and envi-

ronmental conditions, calculates the energy consumed per bit

of an ECC. Given a set of ECCs, from the framework, we get

the minimum-energy-consuming ECC defined as the energy-

optimal ECC.

III. ENERGY MODELS

A. Radio Energy Model (RadioEM)

As stated earlier, the variation in energy consumption of

a radio unit is mainly contributed by the following two

components:

1) energy consumed in transmitting the signal;

2) energy consumed by the radio circuit.

Our energy model incorporates these two components.

Energy consumption is controlled by keeping the transceiver

on, only when sending or receiving data and keeping it off at

all other times. With this arrangement, radio energy Eradiois

calculated as the sum of ON-, SLEEP-, and TRANSIENT-mode

energies. TRANSIENT-mode energy is the energy consumed

in switching radio between ON and SLEEP modes. The radio

energy per bit, on transmitting L bits, can be written as

Eradio=PonTon+ PspTsp+ PtrTtr

L

(1)

where Pon, Psp, and Ptrare the powers consumed during trans-

ceiver ON, SLEEP, and TRANSIENT modes, and Ton, Tsp, and

Ttrare the transceiver ON-, SLEEP-, and TRANSIENT-mode

durations, respectively. The power consumed in radio during

ON mode is the summation of transmit signal power Psigand

circuit power consumption Pckt_tot. As in CMOS technology,

leakage current is small, the power consumed in SLEEP mode

is approximated to zero. This assumption is not valid for future-

generation CMOS circuits that would be built with smaller

geometries but can still be easily taken into account. During

TRANSIENT mode, we can assume that the energy consumed is

negligible as the duration of switching radio from ON to OFF

and vice versa is very small.

Thus, radio energy consumption is

Eradio=(Psig+ Pckt_tot)Ton

L

.

(2)

Psigand Pckt_totare calculated in the following sections.

1) Transceiver Circuit Energy Model: The main compo-

nents of a typical transmitter circuit are digital-to-analog

converter (DAC), low-pass filter (LPF), mixer, frequency syn-

thesizer (FS), power amplifier (PA), and band-pass filter (BPF).

The receiver circuit components are mainly BPF, low-noise

amplifier (LNA), mixer, FS, intermediate-frequency amplifier

(IFA), LPF, and analog-to-digital converter (ADC). Transceiver

circuit power consumption Pckt_totis contributed by PA power

(PPA) and the rest of the circuit element power (Pckt) of the

transceiver circuit

Pckt_tot= PPA+ Pckt.

(3)

Here

Pckt= PDAC+ 2(PLPF+ PFS+ PBPF)

+ PLNA+ PIFA+ PADC

(4)

where PPA, PDAC, PLPF, PFS, PBPF, PLNA, PIFA, and PADC

are the power consumptions in the respective components. The

power consumption in PA is dependent on signal transmit

power as PPA= αPsig, where constant α is related to drain

efficiency η of RF PA, with η = 1/(1 + α) [9].

2) Transmit Signal Energy Model: The transmitted power

required for sending L bits can be expressed as given by

Proakis [10]

?4π

where d is the distance between the transmitter and receiver;

λ is the wavelength of the transmitted signal; Pris the received

power; Gt and Gr are the transmitter and receiver antenna

gains, respectively; and n is the path loss exponent. The re-

ceived power for an additive white Gaussian noise (AWGN)

channel is [10]

Psig=

λ

?2

dn Pr

GrGt

(5)

Pr= SNRuncodedbBN0

2NF

(6)

where SNRuncoded is the SNR per bit for transmitting the

uncoded data, b is the number of bits per modulation symbol,

B is the bandwidth, N0/2 is the noise spectral density for an

AWGN channel, and NF is the receiver noise figure.

For the coded data, the received power is

Pr= SNRcodedbBN0

2NF

(7)

where SNRcodedis the SNR per bit for transmitting N encoded

bits for K data bits. In the case of nonbinary ECCs, N and K

represent the numbers of symbols.

Thus, the radio energy per bit is

Eradio=(1 + α)PsigTon+ PcktTon

L

.

(8)

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on September 1, 2009 at 03:16 from IEEE Xplore. Restrictions apply.

Page 4

1020IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS, VOL. 28, NO. 7, JULY 2009

The transceiver “on” duration and the required SNR per bit

depend on the modulation scheme used. In this paper, we

consider M-ary phase-shift keying (MPSK), M-ary quadrature

amplitude modulation (MQAM), and M-ary frequency-shift

keying (MFSK) modulation schemes.

For MPSK, Tonand SNRcodedare expressed as in [10]

Ton=L

bB

⎧

⎩

(9)

SNRcoded=

⎨

N

K

?erfc−1(2ps)?2,

b(sin

b = 1

N

K

(erfc−1(ps))

2

π

2b)

2 ,

otherwise

(10)

where psis the channel symbol error probability corresponding

to the desired coded symbol error probability or coded bit error

probability and computed while computing the coding gain

for ECC.

For MQAM, the expressions are again derived from

Proakis [10]

Ton=

L

2bB

(11)

SNRcoded=N

K

2(2b− 1)

3b

⎛

⎜

⎜

⎝erfc−1

ps

2

?

1 −

1

(2b)

√

?

⎞

⎟

⎟

⎠

2

.

(12)

For MFSK, Tonis [10]

Ton=2bL

bB

(13)

and SNRcodedis upper bounded as

SNRcoded≤2N

bK

?

erfc−1

?

2ps

2b− 1

??2

.

(14)

B. Computation Energy Model (CompEM)

To estimate the computation energy, we are using Sim-

Panalyzer [8], which is a cycle-accurate power simulator for

the ARM instruction set architecture. Specifically, it simulates

the StrongArm SA-1100 processor.

The power model of the simulator comprises several compo-

nents, which models distinct parts of a processor: cache power

models, datapath and execution unit power models, clock tree

power models, and I/O power models. Sim-Panalyzer computes

the energy dissipation in a program, based on counting the

number of transitions in these parts of the processor. To make

CompEM closer to a real sensor node, we have configured the

simulator to include the external memory for data and instruc-

tion. The simulator takes the encoding (or decoding) function,

written in C, and after compiling it for the target architecture, it

estimates the overall energy spent by the processor in executing

these functions. The total computation energy per bit is

Ecomp=Eenc+ Edec

L

(15)

where Eencand Edecare the encoder and decoder computation

energies, respectively.

TABLE I

PARAMETERS FOR SIMULATIONS

C. Node Energy

For the coded system, the node energy per bit Enode_codedis

the summation of radio energy and computation energy. Using

(8) and (15)

Enode_coded=(1 + α)PsigTon+ PcktTonN

K+ LEcompN

K

L

.

(16)

IV. EXPERIMENTAL SETUP

Using our framework, we have explored and analyzed the

energy consumption of a sensor node with and without using

ECC and with various modulation schemes. In this paper,

we have considered various configurations of Hamming code,

Reed–Solomon (RS) code, and CC.

Decoding of coded data may be done in two ways in a sensor

network. First, the coded data are transmitted from the node to

the base station, and they are decoded at the base station. In

such a type of network, the nodes are simply collecting data,

and data analysis and decision making are done at the base

station. As the base station does not have any energy constraint,

at the node, we consider only encoding energy overheads.

Second, the encoded data are decoded at another node. In such

cases, the nodes are not only collecting data but also taking

local decisions. In this situation, we have to consider encoding,

as well as decoding, energy overheads. Therefore, we have

considered the following two types of nodes:

Type 1) in which decoding is not done at the node but at the

non-energy-constrained end;

Type 2) in which decoding is also done at the battery-powered

sensor node.

Radio energy Eradioiscalculated using(8),withparameters be-

ing shown inTable I.Sim-Panalyzer’s energy model and the cir-

cuit component power [11]–[14] listed in Table I are both based

on 0.18-μm technology. Sim-Panalyzer is configured for a node

with a StrongArm SA-1100 processor operating at 200 MHz,

8-kB data cache, 16-kB instruction cache, and 1-MB RAM.

MPSK and MQAM modulations require a linear PA, so a class-

AB PA is used [15]. For MFSK, linearity is not a constraint,

so a class-C amplifier with drain efficiency of 75% is used.

The coding gains for Hamming and RS codes are calculated

for hard-decision decoding, whereas for CCs, the coding gain

is taken for soft-decision Viterbi decoding [16].

V. RESULTS

In this section, we present the exploration results for the node

energy for a variety of ECCs and modulation schemes. In a

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on September 1, 2009 at 03:16 from IEEE Xplore. Restrictions apply.

Page 5

CHOUHAN et al.: FRAMEWORK FOR DESIGN SPACE EXPLORATION FOR WIRELESS SENSOR NODES1021

Fig. 3. Effect of decoding energy in a sensor node.

given set of ECCs, the ECC with minimum energy is referred

to as the energy-optimal ECC.

A. Significance of Energy Components

In WSN, when considering a node energy with an ECC, it

is important to consider the computation energy along with

the radio energy. At short transmission distances, typical in

WSNs, the computation energy becomes comparable to the

radio energy, so it cannot be neglected. In Fig. 3, the node

energy with its energy components is plotted for different RS

codes, Hamming codes, and CCs with QPSK modulation. As

discussed in Section III, the PA energy is a part of the radio

circuit energy, but it is proportional to the transmit signal

energy. In this figure, we have shown the PA energy separately.

To analyze the significance of energy components in Fig. 3,

we consider the energy of different RS codes. If we do not

consider the encoder–decoder energy, which is typically the

case in conventional wireless networks, then RS (31, 19, 13)

is the scheme with the lowest energy. For Type-1) nodes, the

optimal ECC is RS (31, 27, 5) and is actually consuming

11% less energy than RS (31, 19, 13). When Type-2) nodes

are taken into consideration, RS (31, 29, 3) is the energy-

optimal ECC in this set. As compared to RS (31, 19, 13), this

optimal scheme is consuming 78% less energy per bit. Similar

trends are observed for Hamming codes and CCs. Hence, for

every type of node, the computation energy must be taken into

account.

1) Computation–Radio-Energy Tradeoff: In different ECC

codes, computation–radio-energy tradeoffs are different. A

change in radio energy is due to the coding gain of the par-

ticular ECC and redundant bits. As coding gain increases, less

signal power is required to achieve the same performance. The

computation energy also varies for different ECC schemes as

the computation requirement of each scheme is different. For

example, in Fig. 3, for RS codes with higher coding gains, the

computation energy increases considerably, while in the case

of Hamming codes with higher code word length, the gain

in SNR is not very high, and also, the computation energy

does not change much. In the case of CCs on increasing the

constraint length, the coding gain increases considerably, but

the computation overhead increases exponentially.

In Fig. 3, on considering Type-1) nodes, for certain ECCs, for

example, for RS (31, 29, 3), RS (31, 19, 13), and CC-1/2(k =

3), the node energy per bit with encoding is less than the

uncoded node energy per bit. When we consider the Type-2)

nodes, the computation energy is the sum of energy of encoding

and decoding a bit. Inclusion of decoding function on a node

still shows the energy gain in RS (31, 29, 3). After including

the decoding energy, the node energy shows different trends for

various ECCs. From Fig. 3, we observe that, for the encoding

energy only, the CC with code rate 1/2 and constraint length

k = 4 energy is optimal. When the decoding energy is also

included, RS (31, 29, 3) turns out to be the optimal encoding

scheme. The decoding energy of the CCs dominates over the

other energy components, which makes them unsuitable for

Type-2) nodes. We observe that RS codes are suitable for both

types of nodes and have flexibility to choose different values of

code word length N and error-correcting capability t. In further

discussions, we consider variants of RS codes.

B. Effect of WSN Deployment on Node Energy

Apart from the computation energy, the amount of redun-

dancy added, and the coding gain, other application-specific pa-

rameters, like distance, BER, and path loss exponent, also affect

the energy of a node. Our exploration framework takes these

parameters into account while exploring an energy-optimal

ECC. Energy variations in the Type-1) node for the RS code

with N = 31 are considered with changes in distance, BER,

and path loss exponent. Similar results are observed for other

schemes also.

1) Distance: Distances between sensor nodes vary from a

few meters to hundreds of meters for different applications. For

example, in a WSN deployed for glacial environment monitor-

ing [17], nodes were kept 20–25 m apart, whereas for volcano

monitoring [18], nodes were deployed 200–400 m apart.

Due to significant energy overheads at small distances, an

energy-efficient ECC changes with change in operating dis-

tance. Fig. 4 shows ECCs for a range of distances with BFSK

modulation. As the distance varies, the low-energy-consuming

ECC also changes. The distances where the optimal scheme

changes are marked as crossover points. The optimal code

changes from uncoded to RS (31, 29, 3) at 91 m and, again,

to RS (31, 27, 5) at 102 m.

2) BER and Path Loss Exponent: The optimal node energy

is also influenced by BER and path loss exponent. The BER

requirement changes with the type of application. Table II lists

BERs for different types of applications.

As BER decreases, the node energy with an ECC increases

(Fig. 5). Along with this, the relative energy among the schemes

also changes, so the optimal scheme is not the same for all

BERs. In Fig. 5, the energy saving with the lowest energy RS

code with respect to uncoded at a BER of 10−4is 13%, whereas

for a BER of 10−7, it increases to 38%.

Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY DELHI. Downloaded on September 1, 2009 at 03:16 from IEEE Xplore. Restrictions apply.