Fuzzy-based Routing Metrics Combination for RPL

Patrick Olivier Kamgueu

1,2

, Emmanuel Nataf

2,3

, Thomas Djotio

and Olivier Festor

2,3

University of Yaound´e I, LIRIMA - Masecness project, Yaound´e, Cameroon

Universit´e de Lorraine, Nancy, France

INRIA - Madynes project, Nancy, France

1 INTRODUCTION

The routing protocol is a key functionality of any

communication network, it must support effective

transport of data from one point to another in the net-

work. Due to the many opportunities that they offer,

these recent years, Wireless Sensor Networks (WSN)

are the subject of a growing interest for industrial and

scientiﬁc communities. Such environment must face

with severe constraints, such as fragile nodes with

low energy capabilities, low data rate over a sharing

medium with unstable and lossy transmission links.

WSN protocol designers must take into account many

parameters and challenges different from those of tra-

ditional wired networks.

It is in this context that the Internet Engineering

Task Force

roll

working group was chartered a few

years ago, to design a routing protocol for Low-Power

and Lossy Network (Winter et al., 2012). Very re-

cently, the later has started to provide their results:

the IPv6 Routing Protocol for Low-Power and Lossy

Network (RPL) and other companiondocuments. The

protocol was designed with the purpose of separating

the packet processing and forwarding from the rout-

ing optimization goal. A number of metrics (Vasseur

et al., 2012) are intended to be included by the pro-

tocol during the network topology building phase,

implemented as an objective function. So far, only

two objective functions are speciﬁed: one using the

hop-count as routing metric also called

of0

(Thubert,

2012), the other making use of the expected number

of transmission needed to successfully send a packet

to its destination (ETX) namely

Minimum Rank with

Hysteresis Objective Function

(Gnawali and

Levis, 2012). Use of other deﬁned criteria are left

to implementer, also the possibility of combining sev-

eral metrics into one, to ensure quality of service or

meet application requirements, is not well-addressed.

This problem, including for RPL protocol has al-

ready been investigated in the literature (Karkazis

et al., 2012), and falls into two forms of combina-

tion: additive and lexicographic. In this paper we

propose to take advantage of fuzzy logic to solve

it. This approach is motivates by the fact that, us-

ing this paradigm, with a small memory footprint,

we can seek for a halfway between several criteria,

even antagonistic. Contrary, with the two aforemen-

tioned method, metrics must follow the same direc-

tion (growing or decreasing).

The remainder of the paper is organized as fol-

lows. Section 2 presents related works on combining

metrics for routing in WSN, as well as works that use

fuzzy inference system for routing design. Section 3

describes an overview of RPL protocol. In section

4, we detail the proposed objective function design,

followed by implementation parameters, experiment

results and discussions in section 5. Finally, we con-

clude and discusses future directions in section 6.

2 RELATED WORK

The diversity of applications that WSN are called to

support imposes different requirements on the under-

lying network with respect to delay, loss and energy

criteria. That is why (Karkazis et al., 2012) propose

to use additive and lexicographic composition to op-

timize more than one performance aspect. In the ad-

ditive approach, the composite metric is written as a

linear combination of basic metrics (M =

∑

The main shortcoming of this scheme is that the basis

metrics must necessarily be deﬁned on the same order

relation, thereby restricting the type of metrics to be

considered. The lexicographic approach overcomes

this, but its drawback is that basic metrics are evalu-

ated sequentially, the next is taken into account only

to break the tie. So that, some metrics are frequently

not take into consideration.

For instance, Link Estimation and Parent Selec-

tion protocol (Yan and Sun, 2007) uses hop-count and

Link Quality Indicator (LQI) in lexicographic man-

ner to select the next hop. The source node selects

the neighbouring node which has the minimum hop

to the sink node as its next hop node. If there is more

Kamgueu P., Nataf E., Djotio T. and Festor O..

Fuzzy-based Routing Metrics Combination for RPL.

 2014 SCITEPRESS (Science and Technology Publications, Lda.)

than one node having the minimum hops, the one

which has the largest LQI is chosen. The main dis-

advantage of this protocol is the early death of nodes

and the unbalanced energy dissipation. To overcome

this, RPLRE (Yan et al., 2009) based as the previ-

ous on LQI and hop-count, suggests to take into ac-

count probability selection and additionally residual

energy when choosing next hop node. This improves

the latter by avoiding to select the same node more

often. The result is a more balanced energy consump-

tion among potential parent of a given node, and the

network lifetime is thereby extended.

(Aslam et al., 2004) propose a composite metric

that uses multiple parameters to ﬁnd optimal route,

given the QoS constraints for Optimized Link State

Routing (OLSR) protocol. This routing is natively

based on hop-count. The objective of this composite

metric is to ﬁnd an optimal path with maximum avail-

able bandwidth and minimum delay and jitter. The

latter is computed as a linear combination of the given

parameters.

In (Heo et al., 2009), authors proposed EARQ, a

novel routing protocol for wireless industrial sensor

networks. The protocol provides real-time, reliable

delivery of a packet while considering energy aware-

ness. The path with the lower energy cost is selected

according to a probabilistic model, in addition only

paths that may deliver packets in time are selected to

achieve real-time requirement. Source node may send

a redundant packet via alternate path if the reliabil-

ity is not meet. Contrary to previous methods, EARQ

consider these three criteria (energy, latency and relia-

bility) separately, instead of a single combined metric.

Moreover, protocol supposes that every node knows

it location and rely on a GPS mechanism or location

process for that, but this not always the case for many

WSN deployment.

There are many growing interest of integration

of artiﬁcial intelligence technologies like fuzzy rule-

based systems to design protocols for WSN. GAFO

(Ghataoura et al., 2009) uses a genetic adaptive fuzzy

hop selection scheme, to make optimal choices for ro-

bust packet transmission in WSN involved in varying

channel conditions. The paper describes a fuzzy sys-

tem engine that takes signal to noise ratio and outage

probability as input, to determine the possibility of a

neighbour node to be selected as the next hop for data

forwarding. Experiments show that in the same con-

ditions, this protocol outperforms the crisp approach

on average by 20% for reliability and 15% for total

energy consumption.

A cluster head election method using fuzzy logic

has been introduced by (Gupta et al., 2005) to over-

come the defect of LEACH (Heinzelman et al., 2000),

a popular cluster head selection technique. The main

idea of LEACH protocol is that node are elected de-

pending on a stochastic model and uses localized

clustering. The consequence is that some cluster

heads may be very close to each other or may be

located in the edge of the WSN. This careless clus-

ter heads distribution could not maximize energy ef-

ﬁciency. Other schemes (Basirnezhad and Torshiz,

2011; Ran et al., 2010) were proposed to improve

election process involved on LEACH. Those have

proved that the network lifetime can be efﬁciently

prolonged using fuzzy variables (concentration, en-

ergy and node centrality). Unfortunately, LEACH is

not applicable to networks deployed in large regions,

since it uses single-hop routing where node can trans-

mit data directly to cluster head, afterwards the latter

transmit data to the sink.

EDARP (Zeynali et al., 2009) is another work tar-

geting the routing plane. Contrary to LEACH were

each cluster head directly sends data to sink, proto-

col establishes a Fuzzy Spanning Tree that uses the

energy and distance to construct a routing path over

all cluster heads. These two criteria is used to gener-

ate a fuzzy election number and lead to the selection

of the best parent into the routing tree. EDARP bal-

ance energy consumption among all nodes by keeping

rotation in cluster head election, and parent’s node se-

lection in whole lifespan of the network.

Unlike the previous protocols that uses a clus-

tering mechanism to build a hierarchical topology,

FEAR (Almomani and Saadeh, 2011), directly build

a logical tree topology between network nodes. They

aim to enhance the existing tree-based routing in

terms of reducing the number of hops and solving

the problem of node/link failure. The protocol uses

a ranking system based on fuzzy inference so that

nodes rank their neighbours considering both neigh-

bours depth and power. The fuzzy ranking system

is used to construct and maintain the tree topology.

Compared to RPL, FEAR generates more control

messages which implies a greater power consump-

tion, since communication operations are the more

power consumer. In addition, the protocol uses a node

identiﬁcation (ID) construction model where a node’s

ID is computed based on the node’s parent ID. Thus

when a parent node dead, all nodes in its subtree must

recomputed their ID as soon as a new parent is found,

involving more processing and communication oper-

ations.

SENSORNETS2014-DoctoralConsortium

3 RPL OVERVIEW

RPL (Winter et al., 2012) is a distance vector rout-

ing protocol optimized for low power and lossy net-

work where multipoint-to-pointis the dominant trafﬁc

pattern. The protocol also support point-to-multipoint

trafﬁc pattern using destination advertisement mecha-

nism, and provide a basic structure for point-to-point

route. The network topology is organized as one

or more Destination Oriented Direct Acyclic Graph

(DODAG) each rooted at a single point, that act as

sinks for the topology. Three new types of ICMP mes-

sages are deﬁned and manipulated:

• DODAG Information Object (DIO) used to create

and maintain upwards routes.

• DODAG Destination Advertisement Object

(DAO) used to install downwards routes.

• DODAG Information Solicitation (DIS) actively

used by a node wishing to join the network or ask-

ing for more recent informations.

The topology building starts at the root (initially, the

only router which is part of a DODAG), that sends

DIO messages in its neighbourhood. This message

contains all common conﬁguration parameters, in-

cluding root ID, mode of operation, timers values, etc.

Upon receipt of a number of such messages, neigh-

bour nodes may participate in the DODAG according

to the objectivefunction (

), select theirs parents and

then start emitting their own DIO messages. This pro-

cess spreads gradually to cover the whole network as

new nodes join the DODAG. Only one node among

parent’s nodes (the preferred parent) acts as the next-

hop on the path towards the root.

RPL pro-actively creates and maintains the topol-

ogy, by regularly sending ICMP control messages

in the vicinity. The frequency of these exchanges

are governed by the trickle algorithm (Levis et al.,

2011), that reduce the overhead induced by control

messages. This is done by sending DIO less often

when the topology is steady, but reacts and propagates

rapidly informations on topology change or when in-

consistencies are detected.

An important point is when a node received more

than one consistent DIO, each from a different neigh-

bour and must choose which is the preferred parent.

This choice is governed by the objective function that

speciﬁes how the node selects the best parent into the

parent set, and calculates its own rank (the relative po-

sition with respect to the root) from the parent’s rank.

Different criteria also called routing metrics are de-

ﬁned (Vasseur et al., 2012) to capture link or node

characteristics on the path for parent selection. The

rank calculation is derived from the set of these se-

lected metrics, and must monotonically decrease as

we move toward the root. This last property enables

the routing structure to maintain its acyclic nature and

helps to avoid routing loops.

Unlike existing

s (Gnawali and Levis, 2012;

Thubert, 2012) that use only one metric to construct

the DODAG, we want to integrate QoS into RPL by

taking into account more than one performance as-

pect. The next section shows how we design such

using fuzzy inference system.

4 ROUTING METRIC DESIGN

4.1 Fuzzy Inference System

Fuzzy logic reasoning allows us to transform sev-

eral input variables (delay, ETX and energy) into one

(Quality). The fuzzy inference system consists on

several steps.

• Fuzziﬁcation: take a crisp value input and deter-

mine its degree of membership (fuzziness) for the

appropriate fuzzy sets.

• Fuzzy inference: Apply combination rules to

fuzziﬁed inputs and compute a fuzzy output.

• Aggregation: If an output depends on more than

one rule, this step uniﬁes all values into one.

• Defuzziﬁcation: Convert the fuzzy output ob-

tained at the previous step into a crisp value.

In this paper, due to its simplicity and efﬁciency, we

use the most common fuzzy inference method named

Mamdani model (Mamdani, 1977).

4.2 Composite Metric Design

In order to illustrate the mechanism of fuzzy logic

composition, we consider the network topology de-

picted in the ﬁgure 1. Node N, in order to send data to

the destination S, must select the next hop (between

P1 and P2) as parent. This choice is governed by re-

ceived informations as shown, according to the fuzzy

inference engine.

4.2.1 Linguistic Variables

Node’s performances knowledge are represented as

linguistic variables:

• ETX - The expected number of required transmis-

sions before a packet reaches the destination.

• Delay - The average time for a packet to reach its

destination.

Fuzzy-basedRoutingMetricsCombinationforRPL

ETX=4

ENERGY=75%

ETX=2

ENERGY=70%

DELAY=1000

DELAY=700

Figure 1: Parent Selection Process.

• Energy - The energy cost of the path, also energy

of the node having the smallest remaining battery

level on the path.

We use cross-layer mechanisms to retrieve ETX and

delay from data link and network layers. Node’s en-

ergy is estimated based on the real-time energy es-

timation model described by (Rahm´e and Fourthy,

2010) and implemented by (Nataf and Festor, 2013).

4.2.2 Fuzziﬁcation Process

To avoid the complexity of combining directly three

variables into one, we perform the fuzziﬁcation pro-

cess in two stages, as shown in the ﬁgure 2.

Second Stage

Fuzzy Inference

Engine

First Stage

Fuzzy Inference

Engine

Delay

ETX

Quality

QoS

Energy

Figure 2: Fuzzy Inference Engine.

First Stage Fuzziﬁcation

On the ﬁrst stage, we combine delay and ETX as in-

put to compute QoS which is taken in its turn as in-

put for the next stage. The linguistic variables used

to represent delay are divided into

small, average

and

high

membership functions, and ETX fall into

short, average

and

long

. Figure 3 depicts their

membership functions parameterized by the number

of hops (hc) upwards to the sink, as delays (respec-

tively ETX) are comparable only at the same hop

count.

Table 1 illustrates the relationship between these

two linguistic variables for the computation of QoS.

So, short is the ETX and small is the delay, better is

the QoS to consider.

ETX

short longaverage

0.5

Delay

0.5

small average high

3*hc 6*hc 9*hc 12*hc

600*hc 1200*hc 1800*hc 2400*hc

Figure 3: Membership functions.

Table 1: QoS Output Metric.

ETX / Delay small average high

short very fast fast average

average fast average slow

long average slow very slow

For instance, considering a crisp value ETX, for-

mula 1 indicates how is its the level of membership

in the

average

fuzzy set, for one hop (ie. hc = 1).

Similar formulas establish the level of membership

for others ETX fuzzy sets, as well as delay and en-

ergy linguistic variables.

average(etx) =











0 if etx ≤ 3

etx−3

6−3

if 3 < etx < 6

1 if 6 ≤ etx ≤ 9

etx−12

9−12

if 9 < etx < 12

0 if etx ≥ 12

(1)

For the example given in ﬁgure 1, node N com-

putes as level of membership

short(etx)=0.66,

average(etx)=0.33

, and

long(etx)=0

for the

parent node P1. Same types of computations for P1’s

delay allow us to determine as respective fuzzy sets

small

average

and

high

, the values

0.83, 0.16

and

Since QoS relates to ETX and delay, the previously

computed membership functions are combined ac-

cording to table 1. The Mamdani model allows

us to use the minimum operator as the composition

function, and maximum as the aggregation operator.

For instance, formula 2 indicates how to compute

average(QoS)

fuzzy set from inputs. In the same

way, we establish formulas for fuzzy sets ranging

from

very fast

very slow

avg(QoS) = max





min(long(etx), small(dly)

min(avg(etx),avg(dly))

min(short(etx,high(dly))





(2)

SENSORNETS2014-DoctoralConsortium

For the topology in ﬁgure 1, node N computes

three non-zero QoS membership functions con-

cerning neighbour P1:

very fast(QoS)=0.66,

fast(QoS)=0.33

, and

average(QoS)=0.16

. These

values are defuzziﬁed (following procedure in §4.2.3)

into a single QoS output (

QoS=0.78

), and then used

in the next fuzziﬁcation stage.

Second Stage Fuzziﬁcation

As the second stage of the fuzzy inference system,

we combine the previously computed QoS with the

energy linguistic variable to provide Quality. For a

given node, energy could be

low

medium

full

and the output values for Quality is divided into seven

levels ranging from

awful

excellent

. Table 2

shows how to derive Quality based on QoS and en-

ergy.

Table 2: Quality Output Metric.

QoS / Energy low medium full

very slow awful bad average

slow bad degraded average

average degraded average acceptable

fast average acceptable good

very fast average good excellent

4.2.3 Defuzziﬁcation Process

746660

4638

0.5

awful

accept.

good

excellentbad

degr.

avg

Figure 4: Quality defuzziﬁcation.

All fuzzy values obtained after aggregation step

are converted into a single crisp output. The most

common and accurate defuzzication method uses

the centroid, where the result is the center of gravity

of the polygon drawn using fuzzy values of the

output membership function. Figure 4 illustrates the

defuzziﬁcation process for Quality linguistic variable.

Output values range from 0 to 100 and indicate how

is the level of quality to choose a neighbour as the

next hop, according to the selected metrics.

For the proposed topology, three member-

ship functions are ﬁred as P1 Quality output,

accept(Quality)=0.25

good(Quality)=0.70

and

excellent(Quality)=0.30

. The center of

gravity for the depicted region is

. Similar compu-

tations produce the value

for P2. So, the best next

hop for N according to three criteria (ETX, delay and

energy) is P1.

5 EXPERIMENTS RESULTS

5.1 Network Model and Assumptions

To evaluate the proposed combined metric, sim-

ulations were performed on Cooja (Osterlind and

Dunkels, 2006) that emulate real nodes running Con-

tiki OS code. Twenty nodes are randomly scattered

in the environment of interest (60 × 60 m

2D-grid)

without any isolation. All sensor nodes have the same

capabilities, the same transmission range set to 20 m,

with 30 m of interference range, and limited power

resources. Sensor nodes, located in ﬁxed places with-

out mobility, sense data and transmit it to the sink for

processing at regular time interval. Sink node is as-

sumed to have unconstrained resources. We use as

layer 2, ContikiMAC (Dunkels, 2011) that operated

on top of IEEE 802.15.4 with non-persistent CSMA

and provides power efﬁciency by nodes keeping their

radios turned off for roughly 99% of the time. Table 3

summarizes other setups parameters.

Table 3: Network Setups.

Settings Value

Wireless channel model Unit Disk Graph Medium w/ Distance Loss

Communication range 20 m (Tx/Rx), 30 m (Interf.)

Mote type Tmote sky

Transport Layer UDP

Network Layer µIPv6 + 6LoWPAN

MAC Layer non-persistent CSMA + ContikiMAC

PHY + Radio Layer IEEE 802.15.4 w/ CC2420

5.2 Performance Evaluation

We have performed a data collection application. Two

scenarios are developed, in the ﬁrst the RPL routing

topology is organized according to the native ETX

metric, while the second scenario uses the proposed

combined metric according to the fuzzy inference en-

gine. Data are sent to the sink periodically, several

tests are launch at various data emission rates, rang-

ing from one to twenty application data packets per

minute. Collect application is run for 48 hours, all

sensor nodes start with the same battery level. We are

interested in the network lifetime by looking at energy

depletion of nodes. So the latter is considered as the

time on which the ﬁrst node has completely exhausted

its battery (Dietrich and Dressler, 2009). We also es-

timate the packet reception rate at the sink node, as it

measures the transmission accuracy.

Fuzzy-basedRoutingMetricsCombinationforRPL

62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89

Nodes (%)

Remaining Energy (%)

Throughput: 12pkts/min

ETX

Fuzzy

72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92

Nodes (%)

Remaining Energy (%)

Throughput: 5pkts/min

ETX

Fuzzy

(a) (b)

Figure 6: Battery Distribution.

100

1 5 12 20

Receive ratio (%)

Emission interval (pkt/min)

ETX

Fuzzy

Figure 5: Reception ratio at sink node.

5.2.1 Packet Reception Ratio

Experiments show that for high transmission rates,

the combined metric scenario obtain better reception

ratio, as depicted by ﬁgure 5 at a rate of 12 and 20

packets per minute. When transmission rate is low

(one or ﬁve packets per minute), both scenarios be-

have roughly in the same way, although the combined

metric scenario obtain slightly better results. This

is justiﬁed by the fact that, at high transmission rate

nodes are faced with more data trafﬁc, thus there is

more noise and contention in the network. The com-

bined metric scenario, as ETX scenario, selects nodes

with high expected transmission count, but in addition

the former favors the least overloaded nodes, since la-

tency takes into account buffer size and contention

at the MAC layer. To see this observation in more

details, table 4 presents packet loss rates for various

emission interval. One can note that, at 20 packets per

minute, the single metric scenario loses packets more

than twice compared to the combined metric scenario.

Table 4: Packet Loss Ratio.

Scenario / Freq. 1pkt/min 5pkt/min 12pkt/min 20pkt/min

ETX 2.32% 3.05% 9.40% 46.26%

Fuzzy combined 2.20% 2.64% 6.06% 18.29%

5.2.2 Energy Exhaustion

We assess the energy distribution on the network dur-

ing the routing topology construction. Ideally, we

would like to delay as much as possible when the

ﬁrst network node dies, this means for the routing

protocol to better balance the energy distribution of

nodes (Kamgueu et al., 2013). As shown by the ﬁg-

ure 6, at 12pkts/min (ﬁg. 6-a) the combined metric

scenario have nodes whose batteries levels range be-

tween 69% − 78%, while the same range is 62% −

89%. Furthermore, 15% of nodes in the single met-

ric routing have their remaining power lower than the

weakest node in the second scenario. These obser-

vations are roughly the same at 5pkts/min (ﬁg. 6-b).

The combined metric routing achieved the goal, since

node’s energy dispersion is lower than the native ETX

scenario, also it better delay the ﬁrst node death. This

observation is more accentuated for higher packet

emission rates, as nodes exhaust their battery more

quickly. This result is not surprising since the combi-

nation in addition to others criteria, takes energy into

account while the native objective function don’t care

about that.

6 CONCLUSIONS

In this paper we designed and implemented a new

objective function for IPv6 routing protocol for low

power and lossy network. The proposed solution aims

SENSORNETS2014-DoctoralConsortium

to optimize more than one network performance as-

pects. We used fuzzy inference system to combined

expected transmission count, delay and node’s re-

maining power into one unique value. Experiment

results show that the combined metric objective func-

tion obtains better results, compared to the ETX sce-

nario, both for network energy distribution and packet

reception rates. These results are more highlighted as

soon as the data trafﬁc is heavy in the network.

Currently we perform intensive simulations for

longer durations, we aim to measure more precisely

the inﬂuence of latency and jitter on the routing. In

our future work, we envision to parameterize the con-

tribution of each metric to fuzzy combination and as-

sess its impacts on the routing. Moreover we plan

to implement other forms of metric combinations

(namely lexicographic and additive approaches) and

compare their simulation results with those obtained

with the fuzzy logic.

REFERENCES

Almomani, I. M. and Saadeh, M. K. (2011). Fear: fuzzy-

based energy aware routing protocol for wireless sen-

sor networks. International Journal of Communica-

tions, Network and System Sciences, 4:403–415.

Aslam, N., Phillips, W., and Robertson, W. (2004). Com-

posite metric for quality of service routing in olsr. In

Proc. of Canadian Conference on Electrical and Com-

puter Engineering, pages 759 – 762, Niagara Falls

Ontario, Canada.

Basirnezhad, M. and Torshiz, M. N. (2011). Energy efﬁ-

cient cluster head election using fuzzy logic in wire-

less sensor networks. International Journal of Com-

puter Science and Information Security, 9(5):255–

260.

Dietrich, I. and Dressler, F. (2009). On the lifetime of wsn.

ACM: Trans. on Sensor Networks, 5.

Dunkels, A. (2011). The contikimac radio duty cycling pro-

tocol. Technical Report ISBN 1100-3154, SICS.

Ghataoura, D. S., Yang, Y., and Matich, G. (2009). Gafo:

genetic adaptive fuzzy hop selection scheme for wire-

less sensor network. In Proc. of 5th IEEE Interna-

tional Wireless Communication and Mobile Comput-

ing conference (IWCMC), Leipzig, Germany.

Gnawali, O. and Levis, P. (2012). The Minimum Rank with

Hysteresis Objective Function. RFC 6719 (Proposed

Standard).

Gupta, I., Riordan, D., and Sampalli, S. (2005). Cluster-

head election using fuzzy logic for wireless sensor

networks. In Proc. of 3rd IEEE Annual Communi-

cation Networks and Services Research Conference,

Halifax, Canada.

Heinzelman, W., Chandrakasan, A., and Balakrishnan, H.

(2000). Energy-efﬁcient communication protocol for

wireless microsensor networks. In Proc. of 33rd

Hawaii International Conference on System Sciences

(HICSS), pages 3005–3014.

Heo, J., Hong, J., and Cho, Y. (2009). Earq: Energy aware

routing for real-time and reliable communication in

wireless industrial sensor networks. IEEE Trans. on

Industrial Informatics, 5(1):3 – 11.

Kamgueu, P., Nataf, E., Djotio, T., and Festor, O. (2013).

Energy-based metric for the routing protocol in low-

power and lossy network. In Proc. of 2nd Sensornets,

Barcelona, Spain.

Karkazis, P., Leligou, H. C., Trakadas, P., and Velivas-

saki, T. H. (2012). Design of primary and compos-

ite routing metrics for rpl-compliant wireless sensor

networks. In Proc. of International Conference on

Telecommunications and Multimedia (TEMU), pages

13 – 18, Chania, Greece.

Levis, P., Clausen, T., Hui, J., Gnawali, O., and Ko, J.

(2011). The Trickle Algorithm. RFC 6206 (Proposed

Standard).

Mamdani, E. H. (1977). Application of fuzzy logic to ap-

proximate reasoning using linguistic synthesis. IEEE

Transaction on Computing, C-26(12):1182 – 1191.

Nataf, E. and Festor, O. (2013). Online estimation of battery

lifetime for wireless sensors network. In Proc. of 2nd

Sensornets, Barcelona, Spain.

Osterlind, F. and Dunkels, A. (2006). Cross-level sensor

network simulation with cooja. In Proc. of 31st IEEE

Conf. SenseApp, pages 641 – 648, Tampa, Florida.

Rahm´e, J. and Fourthy, N. (2010). Energy management for

battery-powered embedded systems. In Proc. of IEEE

WCNC, pages 277 – 324, Sydney.

Ran, G., Zhang, H., and Gong, S. (2010). Improving on

leach protocol of wireless sensor networks using fuzzy

logic. Journal of Information & Computational Sci-

ence, 7(3):767–775.

Thubert, P. (2012). Objective Function Zero for the Routing

Protocol for Low-Power and Lossy Networks (RPL).

RFC 6552 (Proposed Standard).

Vasseur, J., Kim, M., Pister, K., Dejean, N., and Barthel,

D. (2012). Routing Metrics Used for Path Calcula-

tion in Low-Power and Lossy Networks. RFC 6551

(Proposed Standard).

Winter, T., Thubert, P., Brandt, A., Hui, J., Kelsey, R.,

Levis, P., Pister, K., Struik, R., Vasseur, J., and

Alexander, R. (2012). RPL: IPv6 Routing Protocol

for Low-Power and Lossy Networks. RFC 6550 (Pro-

posed Standard).

Yan, C., Hu, J., Shen, L., and Song, T. (2009). Rplre: A

routing protocol based on lqi and residual energy for

wireless sensor networks. In Proc. of International

Conference on Information Science and Engineering

(ICISE), pages 2714 – 2717, Nanjing, China.

Yan, T. and Sun, L. (2007). Principle and performance eval-

uation of routing protocol in tinyos. Trans. on Com-

puter Engineering, 33(1):112 – 114.

Zeynali, M., Khanli, L. M., and Mollanejad, A. (2009).

Edarp : novel energy and distance-aware routing pro-

tocol in wireless sensor network. In IEEE, editor, 2nd

ICIS : Information Technology, Culture and Human,

Seou, Korea.

Fuzzy-basedRoutingMetricsCombinationforRPL