A Heuristic Decision Maker Algorithm for Opportunistic Networking in

C-ITS

Rodrigo Silva

1,2

, Christophe Couturier

, Jean-Marie Bonnin

1,2

and Thierry Ernst

IMT Atlantique, IRISA, Inria, Univ. Rennes, Rennes, France

YoGoKo, Rennes, France

Keywords:

ISO TC 204, ETSI TC ITS, C-ITS, Decision Making, ACO, AD4ON.

Abstract:

The number of connected devices is growing worldwide and connected and cooperative vehicles should be a

major element of such ecosystem. However, for ubiquitous connectivity it is necessary to use various wireless

technologies, such as vehicular WiFi (ITSG5, and DSRC), urban WiFi (e.g., 802.11 ac,g,n), 802.15.4, cellular

(3G, 4G, and 5G under preparation). In such an heterogeneous access network environment, it is necessary

to provide applications with transparent decision making mechanisms to manage the assignment of data ﬂows

over available networks. In this paper, we propose the Ant-based Decision Maker for Opportunistic Network-

ing (AD4ON), a Decision Maker (DM) algorithm capable to manage multiple access networks simultaneously,

attempting to choose the best access network for each data ﬂow. Moreover, the AD4ON is capable to increase

decision’s stability, to reduce the ping-pong effect and to manage decisions ﬂow by ﬂow while maximizing

ﬂow’s satisfaction.

1 INTRODUCTION

The number of connected devices is growing expo-

nentially around the world. Connected Internet of

Things (IoT) devices are expected to surge to 125 bil-

lion by 2030 (Howell, 2018).

According to Gartner research company, con-

nected cars will be a major element of the IoT (Gart-

ner, 2018). Once vehicles are capable to exchange in-

formation with others devices and the infrastructure,

they become cooperative and an ecosystem of appli-

cations and services can be developed around them.

In this context, users, devices and vehicles need

to be connected anywhere, anytime with anything.

Such an environment is characterized by its hetero-

geneity. Each service has speciﬁc communication re-

quirements. There are a wide variety of mobile de-

vices, each one with speciﬁc capabilities in terms of

storage, processing and communication. Moreover,

users can have speciﬁc preferences.

However, a single access technology to connect all

these services and devices is impractical. For ubiq-

uitous connectivity it is necessary to use heteroge-

nous wireless technologies, such as vehicular WiFi

(ITSG5, and DSRC), urban WiFi (e.g., 802.11 ac,g,n),

802.15.4 or cellular (3G, 4G, and future 5G) (ETSI,

2010; IEEE, 2006; Chakrabarti et al., 2017). Due

to complementary characteristics of such networks,

more connectivity opportunities are available. Mo-

bile devices equipped with multiple communication

capabilities can use opportunistically these multiple

access technologies in order to maximize ﬂows satis-

faction (e.g., maximizing communication bandwidth,

and/or reducing latency) and to satisfy communica-

tion requirements (e.g., security, monetary cost, traf-

ﬁc load balancing, and others).

In such an heterogeneous and dynamic access net-

work environment, applications and services cannot

take into account all technology particularities, un-

less they explicitly need it. It is preferable to pro-

vide applications with a communication architecture

that hides the heterogeneity of underlying access tech-

nologies, providing seamless communications inde-

pendently of radio access technology.

Standardization bodies have worked to establish

an harmonized communication-centric architecture.

International Organization for Standardization (ISO)

and European Telecommunications Standards Insti-

tute (ETSI) proposed an ITS-S reference architec-

ture (ISO, 2014), which is capable to manage hetero-

geneous wireless access technologies while hides to

the application the underlying differences of access

network.

In our previous work (Silva et al., 2017a), we pro-

578

Silva, R., Couturier, C., Bonnin, J. and Ernst, T.

A Heuristic Decision Maker Algorithm for Opportunistic Networking in C-ITS.

DOI: 10.5220/0007799005780585

In Proceedings of the 5th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2019), pages 578-585

ISBN: 978-989-758-374-2

 2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reser ved

posed an architecture which is compatible with ISO

and ETSI (ISO, 2014) standards and reserves space

for a DM algorithm in charge of managing in real time

the placement of applicative ﬂows over available net-

works.

Based on our research, on ISO/ETSI standards

and surveys, we identiﬁed some important properties

for decision making in the vehicular environment that

are not addressed by existing DM algorithms, like in-

crease the stability of decisions to avoid the ping-pong

effect, and avoid full recalculation when few network

parameters change.

Therefore, to meet such properties, we propose in

this paper the AD4ON, a DM algorithm capable to

manage multiple access networks simultaneously, at-

tempting to choose the best access network for each

data ﬂow. This algorithm is designed to increase de-

cision’s stability while maximize ﬂow’s satisfaction.

We performed simulations to compare the

AD4ON algorithm with three other DM algorithms:

the well-known Technique for Order Preference by

Similarity to Ideal Solution (TOPSIS), a modiﬁed

TOPSIS (mTOPSIS) and a commercial DM used in

most of smartphones, in which decisions are based on

predeﬁned network priorities (e.g., connect to WiFi if

available or to 3G/4G otherwise).

Simulations demonstrate that AD4ON algorithm

outperforms the other algorithms, increasing the total

ﬂow satisfaction, increasing decision’s stability and

reducing the overall monetary cost.

The rest of this paper is organized as follows.

Section 2 reviews some related work. Section 3 de-

scribes the needs for DM in vehicular environment.

The AD4ON algorithm is described in Section 4. In

section 5 we discuss about simulation results and Sec-

tion 6 concludes the paper.

2 RELATED WORK

Several works have concentrated on decision making

algorithms for multiple attributes problems. Most of

them are based on well-known Multi-Attribute Deci-

sion Making (MADM) techniques like Simple Ad-

ditive Weighting (SAW), Multiplicative Exponential

Weighting (MEW) and, mainly TOPSIS (Hwang and

Yoon, 1981).

In TOPSIS, decision calculations are based on a

matrix where lines represent available networks and

decision attributes are set on columns. A quadratic

vector normalisation is applied on columns in order

to homogenize the weight of each attribute. This re-

sults in a hight sensitivity to extreme values which

leads to unstable decisions when a mobile approaches

the limit of range of a given network (where the net-

work can appear or disappear between two steps of

calculations). To mitigate this so called rank rever-

sal behaviour, authors of (Senouci et al., 2016) pro-

pose to replace the classical quadratic normalization

by a new approach based on utility functions. This

way, applications can present a speciﬁc utility func-

tion for each attribute. Simulations show that this ap-

proach eliminates the rank reversal and increases the

ranking quality by fulﬁlling the application require-

ments. However, this paper considers all ﬂows over

only one access network at a time. We compared the

AD4ON with a version of this approach (which we

named mTOPSIS).

In (Bouali et al., 2016) authors develop a fuzzy

MADM methodology to combine application Qual-

ity of Service (QoS) requirements with context com-

ponents (e.g., monetary cost or network power con-

sumption), in order to make context-aware network

selection for each application. First they use Fuzzy

Logic Controllers (FLC) that consider network pa-

rameters and application QoS requirements to deter-

mine the QoS suitability level of each network. Then,

they use MADM SAW algorithm to combine the pre-

vious calculated QoS suitability level with context

components. The network alternatives are ranked by

their context suitability level. Finally, they choose the

best network for each application, i.e., the network

that maximizes the context suitability level. Since this

paper consider the SAW MADM technique, it is sus-

ceptible to suffer from “ping-pong” effect (Trianta-

phyllou, 2000).

Paper (Zarin and Agarwal, 2017) proposes an hy-

brid decision approach in which decision making is

distributed between user devices and the Central Con-

troller Network (CCN), a centralized controller in the

network side. First of all, user devices scan available

networks based on both the Received Signal Strength

(RSS) and user mobility. For example, a mobile user

with high speed does not consider networks with low

coverage. If the received signal strength is higher than

a predeﬁned threshold, the device reports its input

parameter (application bandwidth, user mobility and

battery constraint) to the CCN. Thus, the CCN uses

the MADM MEW algorithm to rank the networks and

provides an associated set of networks for each appli-

cation. This approach suffers from scalability issues.

Due the growing number of user devices and their mo-

bility, the amount of information exchanged between

user devices and the CCN management overhead.

In paper (Rayana and Bonnin, 2009) authors de-

scribe a DM framework for network management,

in which a combination of MADM SAW and Ant

Colony Optimization (ACO) based algorithms is used

A Heuristic Decision Maker Algorithm for Opportunistic Networking in C-ITS

579

to select access networks for each ﬂow communica-

tion. First, the SAW algorithm calculates a utility

score for each feasible ﬂow - network solution. Such

score indicates the matching degree between ﬂow re-

quirements and network characteristics. In a second

step, the ACO-based algorithm adds the network costs

(i.e., power consumption and network load) to the

previous utility function in order to ﬁnd solutions tak-

ing into account the whole system satisfaction. In this

work, the decision stability is addressed by changing

the weight of network costs so that currently enforced

solutions are privileged. However, choosing the best

weight is not a trivial task.

3 PROBLEM DESCRIPTION

Due to the dynamic environment of connected ve-

hicles, we work on a DM mechanism for oppor-

tunistic networking in heterogeneous access network

environment. In our previous work (Silva et al.,

2017a), we proposed a DM architecture for oppor-

tunistic communication. This architecture is based on

the ISO/ETSI ITS-S communication architecture due

the latter’s capability to manage heterogeneous access

technology (ISO, 2014).

At the heart of this architecture, we need a DM al-

gorithm to take smart and ﬁne-grained decisions. In

previous researches (Silva et al., 2017b) we identiﬁed

some properties for such DM mechanism in the ve-

hicular environment. These properties are summed

up bellow.

3.1 Expected Properties

The DM algorithm should present the following prop-

erties. 1) It is necessary to manage decisions ﬂow by

ﬂow, choosing the access network that better match

the communication requirements for each ﬂow. 2)

It is necessary to manage multiple attributes from

different actors (e.g., application requirements, user

preferences, administrators and regulators rules). 3)

The DM should manage multiple objectives simulta-

neously, which can be contradictory. For example,

increase the communication QoS (data rate, latency)

while reduce the overall monetary cost. 4) It is nec-

essary to increase the stability of decisions, avoiding

“ping-pong” effect. 5) Moreover, the DM algorithm

should avoid full recalculation when only few net-

work parameters change.

3.2 Motivation

Existing algorithms in the literature do not meet all

our identiﬁed needs and current commercial ones

do not consider multiple attributes: they are usually

based on static and predeﬁned decisions.

A large number of research studies have concen-

trated on the development of DM algorithms based

on MADM methods. Among these algorithms we can

highlight the TOPSIS, which is the most used in the

literature. Despite the MADM methods present ad-

vantages such as relative low computation complex-

ity, this approach has some limitations. They require

to make full recalculation even if only a given network

parameter change and suffer from ranking abnormal-

ity (i.e. they are very sensitive to small changes of

inputs).

Therefore we propose AD4ON, a DM algorithm

able to manage multiple ﬂows and multiple access

networks simultaneously, attempting to choose the

best access network for each data ﬂow while increas-

ing decision stability, reducing the ping-pong effect

and managing decisions ﬂow by ﬂow to maximize

ﬂow requirements satisfaction. The AD4ON algo-

rithm is explained in section 4.

4 THE AD4ON ALGORITHM

AD4ON is based on the ACO, a swarm intelligence

class of algorithms based on the collective and co-

operative behavior of ants, which are capable to ﬁnd

high-quality solutions for complex combinatorial op-

timization problems in a reasonable time.

The ACO algorithms present some properties that

can be explored to meet our needs. For example, since

ants drop pheromone based on the solution qualities,

and decisions are driven by pheromone concentration,

solutions are created smoothly over time. This tends

to ﬁlter transient effects and thus offers a better sta-

bility. Moreover, since it is a memory-based algo-

rithm (i.e. new solution can take into account previous

status of the network environment), we can prevent

full recalculation when only few network parameters

change.

We modeled our ﬂow to network assignment prob-

lem as a bipartite graph G(F, N, E), where F corre-

spond to data ﬂows, N correspond to available net-

works and E = { f : i 7→ j | i ∈ F, j ∈ N}, i.e., E is

the union between the sets F and N, if ﬂows in F

can be assigned to networks in N. The AD4ON takes

into account requirements and preferences from dif-

ferent actors (e.g., applications, users, administrators

and regulators), as well as information about access

VEHITS 2019 - 5th International Conference on Vehicle Technology and Intelligent Transport Systems

580

networks conditions (e.g., data rate, latency) in order

to construct this graph.

Once the graph with all potential solutions is gen-

erated, the algorithm establishes the ﬂow to network

assignment as described in Algorithm 1.

• First, we set the values of parameters α, β and ρ

that respectively determine the relative inﬂuence

of the pheromone trail, the heuristic information

and the evaporation coefﬁcient of pheromone. We

also initialize the “PS” and “FL” variables that

will respectively store the non-dominated solu-

tions and the list of ﬂows (lines 1 - 3).

• The DM receives the graph G(F, N, E) that repre-

sents all possible solutions (line 4).

• The ants are randomly scattered through the exist-

ing ﬂows in the graph G.

• Once the ants are distributed, they start to con-

struct solutions by randomly exploring the graph.

For each visited ﬂow, ant chooses one network

among all possible networks, i.e., a path in the

Algorithm 1: ACO algorithm.

1 Set values of ACO parameters (e.g., α, β and ρ)

2 PS = null ; // Initialize Pareto Set

(PS) as empty

3 FL ← list of ﬂows

4 G(F, N, E) ← “Rank Alternatives” module

5 while stop condition do

6 for k = 1 → NumberO f Ants do

/* Construct a solution */

7 Sort the ﬂow list FL

8 while remains not visited ﬂow in FL do

9 for each possible network for such

f low do

10 calculate the probability of

choosing that network

according to Equation 1

11 end

12 choose the network to be mapped

13 end

14 end

/* Evaporation */

15 apply the pheromone updating according to

Equation 3

/* Evaluation */

16 Calculate the value of objective function

for each solution in current ant population

(Equation 4)

17 Update the Pareto set solutions (PS)

18 end

19 Return the Pareto Set PS

graph G between current ﬂow and potential net-

works (line 10). The probability (P

i, j

) for an ant

to choose the path i, j, i.e., the path between ﬂow

i and network j is given by Equation 1.

i, j

[τ

i j

]

[η

i j

]

∑

k ∈ V

[τ

]

[η

]

(1)

where τ

i j

is the amount of pheromone present be-

tween ﬂow i and network j, V

is the set of avail-

able networks for the ﬂow i, η

i j

is the heuristic

information and it is given by Equation 2.

i j

∑

n=1

∗ uP

(i, j)

) (2)

where uP

(i, j)

is the utility function of a given pa-

rameter n (e.g., data rate, latency, monetary cost)

between the ﬂow i and the network j, N is the

number of decision parameters, and w

is the

weight of each utility parameter function, such

that

∑

= 1.

• In order to better meet ﬂow requirements, a suit-

able utility function was deﬁned for each param-

eter (i.e., uP

), as shown on Figure 1. The utility

for Data Rate (DR), Packet Delivery Ratio (PDR)

and Received Signal Strength Indication (RSSI)

is deﬁned by function “a”, Latency is deﬁned by

function “b” and the monetary utility is deﬁned

by function “c”. The min and max values rep-

resent the RSSI threshold to get a good wireless

connection or the minimum and maximum ﬂow

requirement for the other parameters.

Figure 1: Utility functions.

• Once an ant found a complete solution, i.e., an ac-

cess network for each ﬂow from the graph, such

ant update the pheromone table (line 15). Besides

the pheromone deposition, it is necessary to apply

a pheromone evaporation rule. Such pheromone

evaporation prevents the convergence of the ACO

A Heuristic Decision Maker Algorithm for Opportunistic Networking in C-ITS

581

algorithm to a locally optimum solution while en-

ables ants to “forget” low quality solutions. The

pheromone update mechanism is given by Equa-

tion 3.

i, j

(t +1) = (1 − ρ)τ

i, j

(t) +

∑

k=1

∆τ

i, j

(3)

where ρ is the pheromone evaporation rate (0 <

ρ < 1), m is the number of ants and ∆τ

i, j

is the

amount of pheromone deposited by ant k on the

edge i, j.

• After all ants have visited the graph G, we should

evaluate the found solutions. In decision making,

utility refers to the satisfaction that a solution pro-

vides to the decision maker. Therefore, we pro-

pose an utility function that calculates a score rep-

resenting the matching degree of each solution in

the current ant colony (line 16). The utility func-

tion is deﬁned by Equation 4.

U =

∑

f =1

(4)

where η

is the heuristic information between a

given ﬂow and its network solution. F is the num-

ber of ﬂows in the graph G.

• Each solution not dominated by both other solu-

tions in the current colony and the non-dominated

solutions already in the Pareto set PS, should be

added to PS. And all solutions dominated by the

added one should be eliminated from PS (line 17).

Then, the ants are scattered again though the ex-

isting ﬂows in the graph G. And all process restart

from the line 5, until a stop condition is satisﬁed.

5 EVALUATION OF THE AD4ON

ALGORITHM

In order to test the AD4ON algorithm, we performed

simulations and compared the results with three other

DM algorithms: the traditional TOPSIS, a modi-

ﬁed version of TOPSIS (mTOPSIS) like the one

of (Senouci et al., 2016), and a version of current

commercial DM (Commercial DM (CM)). To eval-

uate simulation results we deﬁned some Key Perfor-

mance Indicator (KPI).

5.1 Key Performance Evaluation

According to the literature, performances are usu-

ally evaluated using an objective function (utility or

cost function) regardless of whether or not it satis-

ﬁes the application needs (i.e. the higher is the utility

value of a decision, the better the solution). However,

this evaluation does not reﬂect the actual applications

needs. For example, if we consider a ﬂow, requiring

300 kbps of maximum bandwidth that is sent through

a WiFi network called WiFi-1 that offers 1 Mbps of

bandwidth; if another WiFi (WiFi-2) offering 2 Mbps

of bandwidth appears and if we do not consider other

parameters more than bandwidth, the decision maker

based only on the objective function is supposed to

move the ﬂow over WiFi-2. However, both WiFi net-

works satisfy 100% of ﬂow requirement and it would

be better to maintain the ﬂow through the WiFi-1, in

order to avoid packet loss or increased latency due to

the new network association.

Therefore, in order to compare the AD4ON algo-

rithm with others, we consider three KPI:

• The ﬂow satisfaction (FS): is the percentage of

meeting ﬂow requirements. We consider that a

given ﬂow is completely satisﬁed if all its require-

ments are 100% satisﬁed by the chosen network.

For example, a ﬂow that requires a maximum data

rate DR

f low

and a minimum latency sensibility

min

is 100% satisﬁed by a network N, if such net-

work is capable to supply the ﬂow with a data rate

net

, such that DR

net

≥ DR

f low

, and with a la-

tency L

net

, such that L

net

≤ L

min

. If the chosen

network satisfy only the minimum value for all

parameters required by a given ﬂow, such a ﬂow

satisfaction will be the minimum one, i.e., 10% as

considered in this work.

• The stability of decision (DS): frequent changes of

network can increase the packet loss and the com-

munication latency. Therefore, we aim to reduce

the number of network switching. To calculate

this indicator, we consider the average of network

switching performed by each DM algorithm in all

scenarios.

• Monetary utility (MU): we aim at ﬁnding solu-

tions that offer the lowest monetary cost for users

(i.e., higher monetary utility). We assume that the

user informs the DM algorithm of the maximum

price he or she is willing to pay for data commu-

nication. Based on this information, the DM can

calculate a monetary utility as being the ratio be-

tween the communication cost and the maximum

price the user is willing to pay.

Finally, we deﬁne a cost function that is the aver-

age of the three KPI (see Equation 5). In this way, the

algorithm with the best performance is the one that

VEHITS 2019 - 5th International Conference on Vehicle Technology and Intelligent Transport Systems

582

ﬁnds solutions with lowest total cost (TC).

Total cost =

∑

i=1

(i)

∗ (1 − KPI

(i)

) (5)

where w

(i)

is the weight of KPI

(i)

, such that

∑

= 1.

5.2 Description of Simulation Scenarios

To implement the simulation scenarios, we consider

a vehicle capable to connect with multiple access net-

works simultaneously, which is being driven in a zone

covered by four access networks (two urban WiFi net-

works, one cellular network and one vehicular WiFi).

In each scenario, the vehicle moves along a route

of 1000 meters long, while it experiences different

ﬂow demands and network conditions. Each access

network is described by ﬁve parameters: data rate, la-

tency, PDR, RSSI and monetary cost (i.e., the pricing

for mobile data services).

For the sake of statistical analysis, each simulated

scenario was executed 5 times. Indeed, due to the

stochastic property of ACO algorithms, the results

of AD4ON may vary between two executions of the

same scenario.

We divided our simulation into two steps: 1) a ﬁrst

step composed by a simple scenario, which is based

on real testbed; and 2) a second step composed by

multiples scenarios commonly used by literature. In

the following, we describe the simulation scenarios

for each of these steps.

Simulation Scenario for the First Step

In the ﬁrst step we deﬁned the scenario 1, a simple

scenario composed by one application (App1) with

constant requirements and four access networks. The

objective of this step is to show the output and the

key performances for each DM algorithm separately,

while the vehicle moves along the deﬁned route. The

application and network parameter values observed

by the vehicle along this route are showed on Fig-

ure 2. They are taken from our database of real mea-

surements on the ﬁeld.

Simulation Scenario for the Second Step

In the second step, we used scenarios commonly used

by most of literature works, i.e., with input values ran-

domly chosen from a range of predeﬁned values. In

this step, we deﬁned 50 new scenarios. Each one in-

volving four data ﬂows (conversational, streaming, in-

teractive and security) and four access networks (cel-

lular, vehicular WiFi (ITS-G5), and two urban WiFi

(a) RSSI

(b) Data rate

(d) PDR

Figure 2: Inputs for scenario 1.

(802.11n)). Each data ﬂow represents a given appli-

cation with speciﬁc requirements in terms of data rate,

latency and PDR.

The ﬂows requirements and networks conditions

are randomly generated using the range of values

given in Table 1 and Table 2.

Table 1: Range of ﬂow requirements.

Flow

Name

Data rate

(Mbps)

Latency

(s)

Packet

Loss (%)

Conver-

sational

[0.1 .. 0.5] [0.03 .. 0.4] [5 .. 15]

Streaming [0.5 .. 1.9] [0.5 .. 10] [5 .. 20]

Interactive [0.004 .. 0.5] [0.5 .. 4] [5 .. 30]

Security [0.002 .. 0.5] [0 .. 0.1] [5 .. 10]

Table 2: Range of Network Parameters.

Parameters

Cellular ITS-G5 WiFi

DR (Mbps)

0 .. 14 0 .. 22 0 .. 22

Latency (ms)

0 .. 250 0 .. 200 0 .. 200

PDR (%)

90 .. 100 80 .. 100 40 .. 100

RSSI (dBm)

-120 ..-65 -110 ..-45 -110 ..-45

Cost ($/MB)

0.1 .. 0.4 0 0

The network priorities deﬁned for the CM were

based on the monetary cost of networks, i.e., free net-

works were privileged. Therefore, we deﬁned the fol-

lowing descending priority’s order: urban WiFi (WiFi

1 and WiFi 2), ITS-G5 and Cellular.

We performed simulation of the AD4ON with dif-

ferent parameter values (α, β, etc), and we chosed the

values that gave better results for the simulated sce-

narios. Selected parameter values are showed on Ta-

ble 3.

A Heuristic Decision Maker Algorithm for Opportunistic Networking in C-ITS

583

Table 3: AD4ON parameters.

Parameter Values Description

α 2.0 pheromone inﬂuence

β 3.0 heuristic inﬂuence

ρ 0.3 pheromone evaporation

ants 10 number of ants

iterations 50 stop condition (Algorithm 1)

5.3 Simulation Results and Discussion

Simulation results are compared based on the previ-

ous deﬁned KPIs and TC function. Depending on ap-

plications, the total cost evaluation based on the same

weight may not be suitable in some cases. However,

in this work we considered the same weight for all

KPIs.

Results for the First Step

Simulation results of the four algorithms on the sce-

nario 1 (the one of the ﬁrst simulation step) are

showed on Figure 3.

(a) TOPSIS (b) CM

(d) AD4ON

Figure 3: Results for scenario 1.

Figure 3 shows the network chosen for “App1”

by each DM algorithm, while the vehicle moves

along the route. As expected, the traditional TOPSIS

present more ping-pong effect than the others, as we

can observe on Figure 3a.

The KPI (FS, DS and MU), as well as the TC for

each DM algorithm, are shown on Table 4.

This speciﬁc scenario favors the CM algorithm.

The RSSI distribution along the route, as showed on

Figure 2a, enables a smooth network switching, re-

ducing the ping-pong effect. Moreover, the network

parameters follow the RSSI distribution, i.e., good

RSSI levels coincide with good values of DR, Latency

Table 4: Key performance results for scenario 1.

KPI

TOPSIS mTOPSIS CM AD4ON

FS (%)

63.79 63.78 59.79 62.22

MU (%)

74.41 74.41 70.48 73.76

DS (%)

6.67 33.34 46.67 66.67

TC (%)

51.71 42.83 41.02 32.45

and PDR. Therefore, in this speciﬁc scenario, the CM

presents good performances.

The solutions found by the AD4ON present a

ﬂow satisfaction of 62.22% and a monetary utility of

73.76%. This means that, in average, the solution

satisﬁed 62.22% of “App1” requirements and that for

73.76% of the time, the algorithm selected the access

network with the lowest monetary cost. Since TOP-

SIS tries to ﬁnd the best utility regardless to ping-

pong effect, for these two KPI the AD4ON offers

slightly lower performances than TOPSIS (around 1%

difference), which occupies the ﬁrst place for this

speciﬁc scenario. However, such slight underper-

formance is compensated by AD4ON stability, i.e.,

avoiding ping-pong effects.

Concerning decision stability, the AD4ON is bet-

ter than the others, performing 65% less network

switching than TOPSIS and 50% less than mTOPSIS.

Analyzing the total cost (calculated by Equa-

tion 5), we observe that the AD4ON outperforms the

other algorithms, meaning that proposed solutions of-

fer the best compromise between the three indicators.

Results for the Second Step

The results for the second step of the simulations are

shown below. We only show the total cost for each

ﬂow, calculated as described by Equation 5.

(a) Streaming

(b) Conversational

Figure 4: Total Cost.

VEHITS 2019 - 5th International Conference on Vehicle Technology and Intelligent Transport Systems

584

Figure 4 shows the total cost of solutions found by

each DM algorithm. Analyzing this ﬁgure we observe

that the CM has the worst performances. Since its so-

lutions are based only on the RSSI, we observed that

it frequently ﬁnds no feasible solution, i.e., it chooses

network that has good RSSI, but does not meet the

minimum ﬂow requirements (e.g., in terms of PDR,

DR or Latency). This behavior impacts negatively the

KPI, and consequently increases the TC.

The AD4ON algorithm outperforms the other al-

gorithms. It found better solutions for streaming, con-

versational and interactive ﬂows in all simulated sce-

narios, as shown on Figures 4a, 4b, and 4c. For the

safety ﬂow (Figure 4d), the AD4ON outperforms the

others in most of scenarios, and in the worst case,

presents the same quality as TOPSIS.

6 CONCLUSION

In this paper, we proposed the AD4ON, an ACO-

based DM algorithm to solve the problem of assigning

multiple data ﬂows over heterogeneous access net-

works in real time.

We compared the AD4ON algorithm with three

others: the TOPSIS, a variation of TOPSIS and the

one used in most of smartphones (CM). Simulations

results have demonstrated that the AD4ON outper-

forms the other algorithms by increasing the total ﬂow

satisfaction, limiting the ping-pong effect and conse-

quently increasing the decision stability. This shows

that ACO algorithms are good candidates for the im-

plementation of such decision algorithm in routers

used to manage vehicular communications.

This work presents a reactive algorithm, (i.e. one

that ﬁnds new solutions by reacting to the observa-

tion of network conditions). However, vehicles can

move at high speed frequently changing network en-

vironment. Due to such highly dynamic mobility,

it is desired a DM capable to make proactive deci-

sions. Therefore, as future work we aim to enable the

AD4ON to take into account the near future predic-

tion of network environment in its decision process.

REFERENCES

Bouali, F., Moessner, K., and Fitch, M. (2016). A Context-

Aware User-Driven Framework for Network Selection

in 5G Multi-RAT Environments. In 2016 IEEE 84th

Vehicular Technology Conference (VTC-Fall), pages

1–7.

Chakrabarti, S., Montenegro, G., and Droms, R. (2017).

IPv6 over Low-Power Wireless Personal Area Net-

work (6LoWPAN) ESC Dispatch Code Points and

Guidelines. RFC 8066.

ETSI (2010). ETSI ES 202 663 European proﬁle standard

for the physical and medium access control layer of

Intelligent Transport Systems operating in the 5 GHz

frequency band. Technical report.

Gartner ([Accessed: September 2018]). Gart-

ner Says By 2020, a Quarter Billion Con-

nected Vehicles Will Enable New In-Vehicle

Services and Automated Driving Capabilities.

http://www.gartner.com/newsroom/id/2970017.

Howell, J. ([Accessed: September 2018]). Number of

Connected IoT Devices Will Surge to 125 Billion by

2030, IHS Markit Says. https://technology.ihs.com/

596542/number-of-connected-iot-devices-will-surge-

to-125-billion-by-2030-ihs-markit-says.

Hwang, C. L. and Yoon, K. (1981). Methods for Multi-

ple Attribute Decision Making. In Multiple Attribute

Decision Making: Methods and Applications A State-

of-the-Art Survey, Lecture Notes in Economics and

Mathematical Systems, pages 58–191. Springer.

IEEE (2006). IEEE 802.15.4 – Local and metropolitan area

networks– Speciﬁc requirements– Part 15.4: Wireless

Medium Access Control (MAC) and Physical Layer

(PHY) Speciﬁcations for Low Rate Wireless Personal

Area Networks (WPANs). Technical report.

ISO (2014). ISO 21217 Intelligent Transport Systems —

Communications Access for Land Mobiles (CALM)

— Architecture. Technical report.

Rayana, R. B. and Bonnin, J. M. (2009). A Smart Manage-

ment Framework for Multihomed Mobile Nodes and

Mobile Routers. In IEEE 20th International Sympo-

sium on Personal, Indoor and Mobile Radio Commu-

nications, pages 2881–2885.

Senouci, M. A., Hoceini, S., and Mellouk, A. (2016). Util-

ity function-based TOPSIS for network interface se-

lection in Heterogeneous Wireless Networks. In 2016

IEEE International Conference on Communications

(ICC), pages 1–6.

Silva, R., Bonnin, J. M., and Ernst, T. (2017a). An ITS-

based Architecture for Opportunistic Networking. In

VEHICULAR 2017: The Sixth International Confer-

ence on Advances in Vehicular Systems, Technologies

and Applications, pages 91–96. IARIA.

Silva, R., Bonnin, J. M., and Ernst, T. (2017b). Oppor-

tunistic Networking for ITS. In Intelligent Transport

Systems: Past, Present and Future Directions. Nova

publisher.

Triantaphyllou, E. (2000). More Cases of Ranking Abnor-

malities when some MCDM Methods are used. In

Multi-Criteria Decision Making Methods: A Com-

parative Study, number 44 in Applied Optimization,

pages 213–233. Springer US.

Zarin, N. and Agarwal, A. (2017). A hybrid network se-

lection scheme for heterogeneous wireless access net-

work. In 2017 IEEE 28th Annual International Sym-

posium on Personal, Indoor, and Mobile Radio Com-

munications (PIMRC), pages 1–6.

A Heuristic Decision Maker Algorithm for Opportunistic Networking in C-ITS

585