The document proposes a vertical handover management scheme for vehicular cloud computing systems that considers vehicle velocity and current network connection. It involves a two-step process: 1) a vertical handover initiation process evaluates the need for handover; and 2) if needed, a network selection process selects the best alternative network based on service requirements and operator policies. Performance evaluation shows the scheme ensures the best connection for vehicles while outperforming other handover management methods.

1. A Vertical Handover Management Scheme for
VANET Cloud Computing Systems
Emmanouil Skondras1, Angelos Michalas2, Aggeliki Sgora1, Dimitrios D. Vergados1
1
Department of Informatics, University of Piraeus, Piraeus, Greece, Email: {skondras, asgora, vergados}@unipi.gr
2
Department of Informatics Engineering, Technological Education Institute of Western Macedonia,
Kastoria, Greece, Email: amichalas@kastoria.teiwm.gr
Abstract—Vehicular Cloud Computing (VCC) systems use
heterogeneous network access technologies in order to fulfill
the requirements of modern services. Therefore, the need for
efficient Vertical Handover (VHO) management schemes must
be addressed. In this paper a VHO management scheme for
VCC systems is described. The proposed method takes into
account the vehicles velocity as well as its current connection
type and applies a two step VHO algorithm to reduce operations’
costs and optimize mobility management. Accordingly, as a first
step a VHO initiation process evaluates the necessity to perform
handover and subsequently a network selection process selects the
appropriate network alternative considering both vehicular ser-
vice requirements and operators’ policies. The proposed scheme
is applied to a VCC system which includes Long Term Evolution
(LTE) Macrocells and Femtocells as well as 802.11p Road Side
Units (RSUs). Performance evaluation shows that the suggested
algorithm ensures the Always Best Connection (ABC) principle,
while at the same time outperforms existing VHO management
schemes.
I. INTRODUCTION
Vehicular Cloud Computing (VCC) combines the operating
principles of both Vehicular Ad Hoc Networks (VANETs)
and Cloud computing. In a typical VCC system, vehicles are
equipped with On-Board Units (OBUs) with computational,
storage and communication resources. Vehicles communicate
with each other as well as with Road Side Units (RSUs).
Additionally, each RSU interacts with a Cloud infrastructure,
which offers a variety of modern services with strict Quality
of Service (QoS) requirements.
Heterogeneous network access technologies, such as 3GPP
Long Term Evolution (LTE) [1] macrocells and femtocells
as well as IEEE 802.11p Wireless Access for Vehicular
Environment (WAVE) [2] RSUs, could be used for the in-
teraction between the VANET and the Cloud infrastructure.
In a such heterogeneous network environment, the vehicular
users should always obtain connectivity to the best network
access technology, according to their services requirements as
well as operators’ policies. Therefore, the design of efficient
vertical handover (VHO) management schemes is required.
Indicatively, in [3] an algorithm called User Centric Context
Aware (UCCA) is proposed. More specifically, it considers
the estimated time that a vehicle will remain connected to its
current network, in order to decide whether a VHO must be
performed. Thereafter, network selection is performed, while
the alternative networks are ranked using a utility function
and the best network is selected. In [4] a two step decision
algorithm for VHO is proposed. During the first step, a
pre-decision process is performed, where the user’s current
network is evaluated to verify whether it satisfies the minimum
requirements of user services. In case the performance of the
user’s network lies above a predefined threshold, the algorithm
progresses to the second step, where network selection is
performed using a Multi-Criteria Decision Making (MCDM)
method. In [5] a fuzzy logic based VHO management scheme
is described. More specifically, a Mamdani Fuzzy Inference
System (FIS) [6] combines both network and user related cri-
teria in order to perform VHO management between WWAN
and WLAN networks. Finally, in [7] a scheme for handover
between LTE macrocells and femtocells is proposed, where
the user velocity, its current connection type as well as its
service type are considered.
This paper describes a VHO management scheme for VCC
systems, which takes into account the vehicle’s velocity as well
as its current connection type. Initially the fact that a vehicle
with high velocity will remain for a limited time inside the
communication range of a femtocell, is considered. Following,
a two step method is applied. In the first step a VHO initiation
process evaluates the user’s necessity to perform handover,
while during the second step a network selection process
selects the appropriate network alternative considering both
vehicular service requirements and operators’ policies.
The remainder of the paper is as follows: Section II de-
scribes the proposed scheme, while section III presents the
simulation setup and the evaluation results. Finally, section
IV concludes the discussed work.
II. THE PROPOSED VHO MANAGEMENT SCHEME
The proposed scheme considers the vehicle velocity as well
as its current connection type in order to determine the set
of alternative networks and the process that must be executed
next. More specifically, in a way similar to [7], the following
states are defined:
• The vehicle velocity is greater than 30 km
h and is con-
nected to a femtocell: In this case, since the vehicle will
remain for a limited time inside the femtocell coverage,
the VHO initiation process is bypassed and network
2017 IEEE Symposium on Computers and Communications (ISCC)
978-1-5386-1629-1/17/$31.00 ©2017 IEEE

2. selection is executed, while no femtocells are considered
as alternatives.
• The vehicle velocity is greater than 30 km
h and it is
not already connected to a femtocell: Accordingly, the
VHO initiation will be executed, while no femtocells are
considered as alternatives.
• The vehicle velocity is less than 30 km
h : Then, the
VHO initiation will be executed, while all the available
networks will be considered as alternatives.
A. VHO initiation
During the VHO initiation process the Su,i parameter is
defined, determining the satisfaction grade of user u from
his current network i. If the satisfaction grade is less than
a predefined Sth threshold value, then the network selection
process will be executed. More specifically, Su,i is calculated
using a Mamdani FIS which considers two parameters, the
RSSu,i and the Qu,i. The RSSu,i represents the Received
Signal Strength, while the Qu,i represents the quality of the
user’s services offered from the current network. Qu,i is
calculated using formula 1, where thu,i,k, du,i,k, ju,i,k and
plu,i,k represent the throughput, the delay, the jitter and the
packet loss ratio respectively, obtained by user u for the service
k. Additionally, the wth,k, wd,k ,wj,k and wp,kl represent the
weights of the aforementioned parameters, while N represents
the number of the parameters considered and K the number
of the available services.
Qu,i =
K
k=1
(wth,k · thu,i,k + wd,k ·
1
du,i,k
+
wj,k ·
1
ju,i,k
+ wpl,k ·
1
plu,i,k
)/N /K
(1)
Both RSSu,i and Qu,i are normalized in order to have
values within the range [0, 1]. Also, the MFRSS, MFQ,
MFS membership functions (MF) are defined, indicating the
linguistic terms and the corresponding Interval Valued Trape-
zoidal Fuzzy Numbers (IVTFN) for the fuzzy representation
of the RSSu,i, Qu,i and Su,i respectively. Thus, for each
crisp value, two membership degrees are determined in the
corresponding MF, one for the upper trapezoid and one for the
lower trapezoid. The Mamdani FIS implements the following
methods:
a) Fuzzy rule (or knowledge) base: A set R of fuzzy
rules is defined, where each rule r ∈ R is a simple if-then
statement with a condition and a conclusion. The rule’s con-
dition consists of MFRSS and MFQ membership functions,
while its conclusion indicates an MFS membership function.
b) Fuzzification: The RSSu,i and Qu,i crisp values are
converted to degrees of membership indicated as RSSu,i and
Qu,i by a lookup in MFRSS and MFQ membership functions
respectively.
c) Combining the fuzzified inputs (Fuzzy Operations):
A Zu,i,r degree is calculated considering the rule’s condition,
indicating the strength of the rule. Furthermore, in case that a
rule’s condition has multiple parts, the fuzzy operators ‘AND’
and ‘OR’ may be used to combine more than one conditions.
The ‘Algebraic product’ and the ‘Algebraic sum’ are applied
for the ‘AND’ and the ‘OR’ operators respectively. In our case,
the ‘Algebraic product’ is calculated using formula 2, while
the ‘Algebraic sum’ is calculated using formula 3.
Zu,i,r = RSSu,i,rˆ·Qu,i,r = RSSu,i,r · Qu,i,r (2)
Zu,i,r = (RSSu,i,r + Qu,i,r) − (RSSu,i,r · Qu,i,r) (3)
d) Implication method: The implication method esti-
mates the consequence MFSc
r
of the rule conclusion, con-
sidering both the rule conclusion MFSr and the rule strength
Zu,i,r. More specifically, the MFSr
height is trimmed with
respect to the Zu,i,r degree, using formula 4, which applies
the Min method.
MFSc
rHeight
= min{MFSrHeight
, Zu,i,r} (4)
e) Aggregation method: The aggregation method com-
bines the R rules’ consequences to calculate the SA
u,i fuzzy
set, using formula 5, which applies the Max method.
SA
u,i = MFSc
r
(r = 1)∪MFSc
r
(r = 2)∪...∪MFSc
r
(r = R)
(5)
f) Defuzzification: During the defuzzification, the SA
u,i
fuzzy set is transformed to the crisp value Su,i. Formula 6
that applies the Weighted Average method is used, where µr
is the height and hr is the centroid of each rule obtained from
the SA
u,i. Also, symbols U and L represent the upper and the
lower trapezoid of each rule respectively .
Su,i =
R
r=1
(µU
r · hU
r + µL
r · hL
r )
R
r=1
(hU
r + hL
r )
(6)
B. Network selection
The network selection is performed using the Trapezoidal
Fuzzy Topsis (TFT) [8] algorithm. TFT is based on the
concept that the best alternative should have the shortest
distance from the positive ideal solution and the longer dis-
tance from the negative ideal solution. Also, it assumes that
the linguistic values of criteria attributes are represented by
interval-valued trapezoidal fuzzy numbers. More specifically,
suppose AL = {AL1, AL2, . . . , ALn} is the set of possible
alternatives, CR = {CR1, CR2, . . . , CRn} is the set of
criteria and w1, w2, . . . , wm are the weights of each criterion.
The steps of the method are as follows:
a) Construction of the decision matrix: Each gij element
of the n × m decision matrix DM is an interval-valued
trapezoidal fuzzy number which expresses the performance
of alternative i for criterion j. Thus
DM =
CR1 . . . CRm
AL1 g11 . . . g1m
...
...
...
...
ALn gn1 . . . gnm
(7)
where gij = (gL
ij1, gL
ij2, gL
ij3, gL
ij4, vL
ij), (gU
ij1, gU
ij2, gU
ij3, gU
ij4, vU
ij) .
2017 IEEE Symposium on Computers and Communications (ISCC)

3. In the case that there are D decision makers the decision
matrix and the criteria weights include the average of the
performance values and weights respectively, of the decision
makers. Hence, assuming that for the kth
decision maker xijk
is the performance of alternative i for criterion j, and wjk
is the importance weight for criterion j, the average of the
performance values and weights are given by formula 8 and
9, respectively.
gij =
1
D
D
d=1
gijd
(8)
wj =
1
D
·
D
d=1
wjd (9)
b) Normalization of the decision matrix: Consider that
Ωb is the set of benefits attributes and Ωc is the set of
costs attributes. Then, the elements of the normalized decision
matrix are calculated using either formula 10 or 11, where
bj = maxi gU
ij4 for each j ∈ Ωb and cj = mini gL
ij4 for each
j ∈ Ωc.
gij =
gL
ij1
bj
,
gL
ij2
bj
,
gL
ij3
bj
,
gL
ij4
bj
, vL
ij ,
gU
ij1
bj
,
gU
ij2
bj
,
gU
ij3
bj
,
gU
ij4
bj
, vU
ij
(10)
gij =
cj
gL
ij4
,
cj
gL
ij3
,
cj
gL
ij2
,
cj
gL
ij1
, vL
ij ,
cj
gU
ij4
,
cj
gU
ij3
,
cj
gU
ij2
,
cj
gU
ij1
, vU
ij
(11)
c) Construction of the weighted normalized decision ma-
trix: The weighted normalized decision matrix is constructed
by multiplying each element of the normalized decision matrix
gij with the respective weight wj according to the formula 12.
uij = g L
ij1 · wj, g L
ij2 · wj, g L
ij3 · wj, g L
ij4 · wj, v L
ij ,
g U
ij1 · wj, g U
ij2 · wj, g U
ij3 · wj, g U
ij4 · wj, vU
ij
(12)
d) Determination of the positive and negative ideal so-
lution: The positive ideal solution is defined in 13, where
i
≡ maxi in case j ∈ Ωb and
i
≡ mini in case j ∈ Ωc.
Correspondingly, the negative ideal solution is defined in 14,
where i ≡ mini in case j ∈ Ωb and i ≡ maxi in case
j ∈ Ωc.
G+
= g+L
ij1 , g+L
ij2 , g+L
ij3 , g+L
ij4 , v+L
ij , g+U
ij1 , g+U
ij2 , g+U
ij3 , g+U
ij4 , v+U
ij
(13)
G−
= g−L
ij1 , g−L
ij2 , g−L
ij3 , g−L
ij4 , v−L
ij , g−U
ij1 , g−U
ij2 , g−U
ij3 , g−U
ij4 , v−U
ij
(14)
e) Measurement of the distance of each alternative from
the ideal solutions: The distances of each alternative from the
positive ideal solution are evaluated using formulas 15 and 16.
Likewise the distances of each alternative from the negative
ideal solution are estimated using formulas 17 and 18.
p+
i1 =
m
j=1
1
4
uL
ij1 − g+L
ij1
2
+
uL
ij2 − g+L
ij2
2
+ uL
ij3 − g+L
ij3
2
+ uL
ij4 − g+L
ij4
2
1
2
(15)
p+
i2 =
m
j=1
1
4
uU
ij1 − g+U
ij1
2
+
uU
ij2 − g+U
ij2
2
+ uU
ij3 − g+U
ij3
2
+ uU
ij4 − g+U
ij4
2
1
2
(16)
p−
i1 =
m
j=1
1
4
uL
ij1 − g−L
ij1
2
+
uL
ij2 − g−L
ij2
2
+ uL
ij3 − g−L
ij3
2
+ uL
ij4 − g−L
ij4
2
1
2
(17)
p−
i2 =
m
j=1
1
4
uU
ij1 − g−U
ij1
2
+
uU
ij2 − g−U
ij2
2
+ uU
ij3 − g−U
ij3
2
+ uU
ij4 − g−U
ij4
2
1
2
(18)
Consequently, similar to [9] the alternatives distance from the
positive and negative ideal solutions are expressed by intervals
such as [p+
i1, p+
i2] and [p−
i1, p−
i2], instead of single values, while
in this way less information is lost.
f) Calculation of the relative closeness: The relative
closeness of the distances from the ideal solutions are calcu-
lated using formula 19. Subsequently, the compound relative
closeness is obtained using formula 20.
RCij =
p−
ij
p+
ij + p−
ij
j = 1, 2 (19)
RCi =
RCi1 + RCi2
2
(20)
g) Alternatives ranking: The alternative networks are
ranked according to their RCi values, while the best alternative
is that with the higher RCi value.
III. SIMULATION SETUP AND RESULTS
In our experiments we consider the Software Defined
VANET Cloud topology presented in figure 1, which includes
a heterogeneous network environment and a cloud infrastruc-
ture. The network environment consists of one LTE Macrocell,
two LTE Femtocells and two 802.11p RSUs, with radius
equal to 250m, 120m and 15m respectively. Accordingly, the
Cloud infrastructure includes a set of Virtual Machines (VMs)
providing services such as Navigation Assistance (NAV),
Voice over IP (VoIP), Conversational Video (CV), Buffered
Streaming (BS) and Web Browsing (WB). Furthermore, a Soft-
ware Defined Network (SDN) controller provides centralized
control of the entire system.
Each access network supports at least one of the afforemen-
tioned Cloud services, while three Service Level Agreements
(SLAs) are defined. Each SLA determines the provided speci-
fication per network for each service type, in terms of through-
put, delay, jitter, packet loss ratio, price, security and service
reliability. Service reliability determines the ability for service
constraints satisfaction and optimization of performance when
a network is congested. SLA1 has the higher service priority,
while SLA3 has the lower one. Additionally, SLA1 supports
2017 IEEE Symposium on Computers and Communications (ISCC)

4. Cloud
SDN
controller
VM
Services
VM
Services
VM
Services
VM
Services
VM
Services
VM
Services
VM
Services
VM
Services
...
...
...
... ...
...
...
...
LTE Macro
802.11p RSU1802.11p RSU1 802.11p RSU2802.11p RSU2
LTE
Femto2
LTE
Femto2
LTE
Femto1
LTE
Femto1
Fig. 1: The simulated topology.
all service types and provides the best values for QoS and
policy decision criteria. SLA2 supports less service types and
slightly worse decision criteria values. Finally, SLA3 supports
only the WB service and provides acceptable decision criteria
values. The linguistic terms for the criteria attributes are
represented by interval-valued trapezoidal fuzzy numbers as
shown in table I.
TABLE I: Lingustic terms and the corresponding interval-valued trapezoidal
fuzzy numbers used for the criteria attributes.
Linguistic term Interval-valued trapezoidal fuzzy number
Absolutely Poor (AP) [(0.0, 0.0, 0.0, 0.0, 0.8), (0.0, 0.0, 0.0, 0.0, 1)]
Very Poor (VP) [(0.01, 0.02, 0.03, 0.07, 0.8), (0.0, 0.01, 0.05, 0.08, 1)]
Poor (P) [(0.04, 0.1, 0.18, 0.23, 0.8), (0.02, 0.08, 0.2, 0.25, 1)]
Medium Poor (MP) [(0.17, 0.22, 0.36, 0.42, 0.8), (0.14, 0.18, 0.38, 0.45, 1)]
Medium (M) [(0.32, 0.41, 0.58, 0.65, 0.8), (0.28, 0.38, 0.6, 0.7, 1)]
Medium Good (MG) [(0.58, 0.63, 0.8, 0.86, 0.8), (0.5, 0.6, 0.9, 0.92, 1)]
Good (G) [(0.72, 0.78, 0.92, 0.97, 0.8), (0.7, 0.75, 0.95, 0.98, 1)]
Very Good (VG) [(0.93, 0.98, 1, 1, 0.8), (0.9, 0.95, 1, 1, 1)]
Absolutely Good (AG) [(1, 1, 1, 1, 0.8), (1, 1, 1, 1, 1)]
We consider the case where 10 vehicles are moving inside
the heterogeneous network environment and need to be con-
nected to a network which satisfies the requirements of their
services and at the same time comply with their respective
SLA agreements. For each vehicle, table III presents its SLA
and the services used. Also, the current access network, the
candidate networks as well as the next process (VHO initiation
or network selection) that must be executed, according to the
proposed algorithm, are shown.
A. VHO initiation
During the VHO initiation process, the Analytic Hierarchy
Process (AHP) [10] method is applied in order to estimate the
services weights. The criteria used include throughput, delay,
TABLE II: The available networks of each SLA.
SLA Service Network Throughput Delay Jitter
Packet
Loss
Price
Service
Reliability
Security
SLA1
NAV
LTE
Macro
VG
(300 Kbps)
AG
(25 ms)
AG
(15 ms)
VG
(10−3)
VP AG VG
LTE
Femto1
G
(290 Kbps)
MG
(40 ms)
VG
(25 ms)
AG
(10−4)
VP VG AG
LTE
Femto2
AG
(305 Kbps)
AG
(25 ms)
VG
(22 ms)
VG
(10−3)
AP G G
802.11p
RSU1
MG
(280 Kbps)
MG
(40 ms)
G
(30 ms)
VG
(10−3)
MP MG MG
802.11p
RSU2
M
(270 Kbps)
M
(45 ms)
MG
(35 ms)
VG
(10−3)
MP MG MG
VoIP
LTE
Macro
MG
(210 Kbps)
MG
(90 ms)
VG
(22 ms)
VG
(10−4)
VP AG VG
LTE
Femto1
AG
(250 Kbps)
AG
(40 ms)
AG
(15 ms)
MG
(10−2)
VP VG AG
LTE
Femto2
AG
(240 Kbps)
VG
(50 ms)
G
(28 ms)
G
(10−3)
MP AG VG
802.11p
RSU1
MG
(210 Kbps)
MG
(90 ms)
MG
(30 ms)
AG
(10−5)
P MG G
802.11p
RSU2
G
(220 Kbps)
G
(80 ms)
VG
(20 ms)
AG
(10−5)
MP MG G
CV
LTE
Macro
MP
(8 Mbps)
MG
(60 ms)
VG
(35 ms)
AG
(10−5)
AP AG VG
LTE
Femto1
AG
(9.5 Mbps)
AG
(45 ms)
AG
(25 ms)
VG
(10−4)
AP VG AG
LTE
Femto2
G
(9 Mbps)
VG
(50 ms)
AG
(25 ms)
AG
(10−5)
MP VG G
802.11p
RSU1
MG
(8.5 Mbps)
MG
(60 ms)
G
(40 ms)
AG
(10−5)
MP MG VG
802.11p
RSU2
MP
(8 Mbps)
MP
(70 ms)
MG
(45 ms)
AG
(10−5)
MP MG G
BS
LTE
Macro
M
(8 Mbps)
G
(60 ms)
VG
(40 ms)
VG
(10−6)
AP AG VG
LTE
Femto1
VG
(9 Mbps)
VG
(55 ms)
AG
(35 ms)
AG
(10−7)
AP VG AG
LTE
Femto2
G
(8.5 Mbps)
G
(60 ms)
VG
(40 ms)
AG
(10−7)
MP VG G
802.11p
RSU1
VG
(9 Mbps)
AG
(50 ms)
VG
(40 ms)
AG
(10−7)
MP MG VG
802.11p
RSU2
G
(8.7 Mbps)
VG
(55 ms)
G
(45 ms)
AG
(10−7)
MP MG G
WB
LTE
Macro
AG
(3.2 Mbps)
AG
(150ms)
AG
(80 ms)
AG
(10−5)
VP AG VG
LTE
Femto1
MG
(2.5 Mbps)
M
(190ms)
G
(90 ms)
VG
(10−4)
MP VG AG
LTE
Femto2
VG
(3 Mbps)
G
(170ms)
M
(100ms)
AG
(10−5)
MP VG G
802.11p
RSU1
AG
(3.2 Mbps)
G
(170ms)
G
(90 ms)
AG
(10−5)
P MG MG
802.11p
RSU2
G
(2.8 Mbps)
M
(190ms)
AG
(80 ms)
AG
(10−5)
MP MG G
SLA2
CV
LTE
Macro
MP
(8 Mbps)
M
(65 ms)
MG
(45 ms)
G
(10−3)
MP G G
LTE
Femto1
G
(9 Mbps)
G
(55 ms)
VG
(35 ms)
VG
(10−4)
M G G
LTE
Femto2
MG
(8.5 Mbps)
MG
(60 ms)
AG
(30 ms)
VG
(10−4)
MP G MG
802.11p
RSU1
MP
(8 Mbps)
MP
(70 ms)
M
(50 ms)
VG
(10−4)
M G MG
802.11p
RSU2
MP
(8 Mbps)
P
(75 ms)
P
(60 ms)
VG
(10−4)
M P M
BS
LTE
Macro
M
(8 Mbps)
G
(60 ms)
MG
(50 ms)
VG
(10−6)
MP G G
LTE
Femto1
M
(8 Mbps)
MG
(65 ms)
AG
(35 ms)
AG
(10−7)
MP G G
LTE
Femto2
MG
(8.2 Mbps)
M
(70 ms)
VG
(40 ms)
AG
(10−7)
M MG M
802.11p
RSU1
G
(8.5 Mbps)
VG
(55 ms)
G
(45 ms)
AG
(10−7)
M MP G
802.11p
RSU2
P
(6.7 Mbps)
MP
(80 ms)
M
(60 ms)
VG
(10−6)
MP P M
WB
LTE
Macro
G
(2.8 Mbps)
M
(190ms)
M
(100ms)
AG
(10−5)
MP MG M
LTE
Femto1
M
(2.3 Mbps)
MP
(200ms)
MG
(95 ms)
VG
(10−4)
M G G
LTE
Femto2
MG
(2.5 Mbps)
M
(190ms)
M
(100ms)
AG
(10−5)
M M M
802.11p
RSU1
G
(2.8 Mbps)
M
(190ms)
M
(100ms)
AG
(10−5)
MP MG MP
802.11p
RSU2
MG
(2.5 Mbps)
MP
(200ms)
MG
(95 ms)
AG
(10−5)
M M MG
SLA3 WB
LTE
Macro
MP
(2 Mbps)
P
(210ms)
M
(100ms)
G
(10−3)
VG P P
LTE
Femto1
M
(2.3 Mbps)
M
(190ms)
G
(90 ms)
VG
(10−4)
VG P MP
LTE
Femto2
VP
(1.8 Mbps)
P
(220ms)
P
(120ms)
AG
(10−5)
VG VP VP
802.11p
RSU1
AP
(1.1 Mbps)
VP
(260ms)
P
(120ms)
VG
(10−4)
AG VP AP
802.11p
RSU2
AP
(1.0 Mbps)
AP
(300ms)
VP
(140ms)
G
(10−3)
AG AP AP
jitter and packet loss. Figure 2 depicts the estimated VHO
initiation weights for each service.
The linguistic terms for the RSSu,i, Qu,i and Su,i are
represented by interval-valued trapezoidal fuzzy numbers as
shown in table IV. The RSSu,i and Qu,i values are combined
using a Mamdani Fuzzy Inference System (FIS) [6], which
2017 IEEE Symposium on Computers and Communications (ISCC)

5. TABLE III: The simulated vehicles.
Vehicle SLA Velocity Services Current Network (RSS)
Candidate
Networks
Next process
1 1 10 kmh VoIP, CV LTE Femto2 (-85dBm) All VHO initiation
2 1 50 kmh
NAV,
VoIP
80211p RSU1 (-68 dBm)
All except
femtocells
VHO initiation
3 1 80 kmh CV LTE Femto1 (-88 dBm)
All except
femtocells
Network selection
4 1 10 kmh NAV, WB LTE Femto2 (-105 dBm) All VHO initiation
5 2 40 kmh BS 80211p RSU2 (-70 dBm)
All except
femtocells
VHO initiation
6 2 20 kmh CV, WB LTE Macro (-107 dBm) All VHO initiation
7 2 10 kmh CV 80211p RSU2 (-77 dBm) All VHO initiation
8 2 5 kmh BS, WB 80211p RSU1 (-60 dBm) All VHO initiation
9 3 70 kmh WB 80211p RSU1 (-85 dBm)
All except
femtocells
VHO initiation
10 3 30 km WB 80211p RSU2 (-75 dBm) All VHO initiation
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
Navigation Assistance VoIP Conversational Video Buffered Streaming Web
Weight
VHO initiation weights
Throughput Delay Jitter Packet loss
Fig. 2: Criteria weights per service for the VHO initiation.
calculates the SA
u,i fuzzy set using the rules presented in table
V. Afterwards, the SA
u,i is defuzzified in order to extract
TABLE IV: Linguistic terms and the corresponding interval-valued
trapezoidal fuzzy numbers used for RSSu,i, Qu,i and Su,i.
RSSu,i membership functions.
Linguistic term Interval-valued trapezoidal fuzzy number
Too Bad (TB) [(0.0, 0.0, 0.0, 0.0, 0.8), (0.0, 0.0, 0.0, 0.0, 1)]
Bad (B) [(0.1, 0.2, 0.3, 0.4, 0.8), (0.01, 0.15, 0.38, 0.5, 1)]
Enough (EN) [(0.47, 0.59, 0.69, 0.75, 0.8), (0.4, 0.52, 0.74, 0.82, 1)]
More than Enough (ME) [(0.72, 0.78, 0.92, 0.97, 0.8), (0.7, 0.75, 0.95, 0.98, 1)]
Excellent (EX) [(1, 1, 1, 1, 0.8), (1, 1, 1, 1, 1)]
Qu,i membership functions.
Linguistic term Interval-valued trapezoidal fuzzy number
Absolutely Poor (AP) [(0.0, 0.0, 0.0, 0.0, 0.8), (0.0, 0.0, 0.0, 0.0, 1)]
Very Poor (VP) [(0.01, 0.02, 0.03, 0.07, 0.8), (0.0, 0.01, 0.05, 0.08, 1)]
Poor (P) [(0.04, 0.1, 0.18, 0.23, 0.8), (0.02, 0.08, 0.2, 0.25, 1)]
Medium Poor (MP) [(0.17, 0.22, 0.36, 0.42, 0.8), (0.14, 0.18, 0.38, 0.45, 1)]
Medium (M) [(0.32, 0.41, 0.58, 0.65, 0.8), (0.28, 0.38, 0.6, 0.7, 1)]
Medium Good (MG) [(0.58, 0.63, 0.8, 0.86, 0.8), (0.5, 0.6, 0.9, 0.92, 1)]
Good (G) [(0.72, 0.78, 0.92, 0.97, 0.8), (0.7, 0.75, 0.95, 0.98, 1)]
Very Good (VG) [(0.93, 0.98, 1, 1, 0.8), (0.9, 0.95, 1, 1, 1)]
Absolutely Good (AG) [(1, 1, 1, 1, 0.8), (1, 1, 1, 1, 1)]
Su,i membership functions.
Linguistic term Interval-valued trapezoidal fuzzy number
Absolute Unsatisfactory (AU) [(0.0, 0.0, 0.0, 0.0, 0.8), (0.0, 0.0, 0.0, 0.0, 1)]
Very Unsatisfactory (VU) [(0.01, 0.02, 0.03, 0.07, 0.8), (0.0, 0.01, 0.05, 0.08, 1)]
Unsatisfactory (U) [(0.04, 0.1, 0.18, 0.23, 0.8), (0.02, 0.08, 0.2, 0.25, 1)]
Slightly Unsatisfactory (SU) [(0.08, 0.14, 0.26, 0.3, 0.8), (0.04, 0.12, 0.32, 0.38, 1)]
Less than Acceptable (LA) [(0.17, 0.22, 0.36, 0.42, 0.8), (0.14, 0.18, 0.38, 0.45, 1)]
Slightly Acceptable (SA) [(0.32, 0.41, 0.58, 0.65, 0.8), (0.28, 0.38, 0.6, 0.7, 1)]
Acceptable (A) [(0.44, 0.52, 0.61, 0.75, 0.8), (0.37, 0.45, 0.68, 0.83, 1)]
More than Acceptable (MA) [(0.58, 0.63, 0.8, 0.86, 0.8), (0.5, 0.6, 0.9, 0.92, 1)]
Slightly Satisfactory (SS) [(0.72, 0.78, 0.92, 0.97, 0.8), (0.7, 0.75, 0.95, 0.98, 1)]
Satisfactory (S) [(0.83, 0.87, 0.95, 0.98, 0.8), (0.74, 0.77, 0.98, 0.99, 1)]
Very Satisfactory (VS) [(0.93, 0.98, 1, 1, 0.8), (0.9, 0.95, 1, 1, 1)]
Absolute Satisfactory (AS) [(1, 1, 1, 1, 0.8), (1, 1, 1, 1, 1)]
the crisp value Su,i expressing the satisfaction of user u for
his current network i. Figure 3 illustrates the complete set of
possible Su,i values as a function of the initial RSSu,i and
Qu,i values. Indicatively, when the RSSu,i and Qu,i values
are too low, the produced Su,i value is too low as well. On
the contrary, when the RSSu,i and Qu,i values are close to 1,
the produced Su,i value is also high, indicating that the user is
fully satisfied. Furthermore, when only one of the RSSu,i or
the Qu,i values is close to 0, the user satisfaction is in quite
low levels.
Table VI presents the minimum acceptable values for
RSSSLA and QSLA per SLA as well as the evaluated Sth,SLA
TABLE V: The FIS rules for SA
u,i calculation.
Rule MFRSS Operator MFQ MFS
1 TB or AP AU
2 B and VP AU
3 EN and VP VU
4 ME and VP VU
5 EX and VP U
6 B and P AU
7 EN and P VU
8 ME and P U
9 EX and P SU
10 B and MP AU
11 EN and MP SU
12 ME and MP LA
13 EX and MP SA
14 B and M VU
15 EN and M LA
16 ME and M SA
17 EX and M A
18 B and MG VU
19 EN and MG SA
20 ME and MG A
21 EX and MG SS
22 B and G U
23 EN and G A
24 ME and G SS
25 EX and G S
26 B and VG U
27 EN and VG MA
28 ME and VG S
29 EX and VG VS
30 B and AG SU
31 EN and AG S
32 ME and AG VS
33 EX and AG AS
Fig. 3: The S values range as obtained using the FIS.
thresholds, which are obtained from the Mamdani FIS. Finally,
table VII presents the VHO initiation results for each vehicle.
As it can be observed, the VHO initiation process is ignored
for the vehicle 3, due to the fact that it moves with high
velocity while at the same time it is connected to a femtocell.
TABLE VI: The RSSSLA, QSLA and Sth,SLA thresholds per SLA.
SLA RSSSLA QSLA Sth,SLA
1 0.8 0.9 0.67287
2 0.7 0.7 0.49000
3 0.6 0.35 0.24076
TABLE VII: VHO initiation results.
Vehicle RSSu,i Qu,i Su,i Sth,SLA VHO required
1 1.000000 0.704539 0.84523 0.67287 No
2 0.733333 0.538458 0.41351 0.67287 Yes
3 - - - -
Yes
(according to velocity)
4 0.428571 0.773342 0.32516 0.67287 Yes
5 0.666667 0.697471 0.48699 0.49000 Yes
6 0.371429 0.844627 0.11667 0.49000 Yes
7 0.433333 0.661852 0.36116 0.49000 Yes
8 1.000000 0.986799 0.96375 0.49000 No
9 0.166667 0.408965 0.032391 0.24076 Yes
10 0.500000 0.345476 0.23878 0.24076 Yes
2017 IEEE Symposium on Computers and Communications (ISCC)

6. 0
0,1
0,2
0,3
0,4
0,5
Navigation
Assistance
(SLA1)
VoIP
(SLA1)
Conversational
Video
(SLA1)
Buffered
Streaming
(SLA1)
Web
(SLA1)
Conversational
Video
(SLA2)
Buffered
Streaming
(SLA2)
Web
(SLA2)
Web
(SLA3)
Weight
Network Selection weights for each SLA
Throughput Delay Jitter Packet loss Price Service Reliability Security
Fig. 4: Criteria weights per service and SLA for the Network Selection.
LTEFemto2
802.11pRSU1
LTEFemto1
LTEFemto2
802.11pRSU2
LTEMacro
802.11pRSU2
802.11pRSU1
802.11pRSU1
802.11pRSU2
LTEFemto2
LTEMacro
LTEMacro
LTEMacro
LTEMacro
LTEFemto1
LTEFemto1
802.11pRSU1
LTEMacro
LTEFemto1
LTEFemto2
802.11pRSU1
LTEFemto1
LTEMacro
802.11pRSU2
LTEFemto1
LTEFemto1
802.11pRSU1
LTEFemto1
802.11pRSU2
LTEFemto1
LTEMacro
LTEFemto1
LTEFemto2
LTEMacro
LTEMacro
802.11pRSU2
LTEMacro
802.11pRSU2
LTEMacro
0
0,02
0,04
0,06
0,08
0,1
0,12
0,14
Vehicle 1
v=10kmh
Vehicle 2
v=50kmh
Vehicle 3
v=80kmh
Vehicle 4
v=10kmh
Vehicle 5
v=40kmh
Vehicle 6
v=20kmh
Vehicle 7
v=10kmh
Vehicle 8
v=5kmh
Vehicle 9
v=70kmh
Vehicle 10
v=30kmh
TFTrank
TFT ranking of each VHO scheme
Current RSU Proposed Scheme UCCA Two-step
Fig. 5: Proposed VHO management scheme’s results.
B. Network selection
During the network selection process initially the Analytic
Network Process (ANP) [11] method is applied in order to
estimate the decision weights per service type and SLA,
considering the ANP network model proposed in [8]. The
criteria used include throughput, delay, jitter, packet loss,
price, service reliability and security. The decision weights
per service and SLA are obtained as presented in figure 4.
Subsequently, the TFT algorithm selects the best network
for each vehicle considering the vehicle service requirements
(Table III).
Figure 5 compares the results of the proposed scheme with
the ones obtained using the UCCA [3] and the Two-step [4]
VHO management schemes. In this figure, for each vehicle
the current network as well as the target network connection
estimated by each of the three schemes are presented. Ad-
ditionally, the TFT ranking of each network is given. From
the obtained results it is clear that the suggested algorithm
outperforms the existing schemes since it selects as target
networks for vehicles the ones with the best TFT ranks. Also,
in special cases where the velocity of vehicles is high (eg. for
vehicles 2, 3, 5 and 9) the proposed scheme considers only
the wide coverage candidate networks as alternatives avoiding
the handovers to femtocell networks.
IV. CONCLUSION
This paper proposes a VHO management scheme for VCC
systems. The vehicle velocity and its current network con-
nection type are considered. Subsequently, the VHO initiation
process estimates whether a VHO must be performed, while
the network selection algorithm selects the most appropriate
network, with respect to both vehicle’s services and operator’s
policies. Simulation results showed that the ABC principle is
fully ensured, while at the same time the proposed scheme
outperforms existing VHO management algorithms.
ACKNOWLEDGEMENTS
The publication of this paper has been partly supported by
the University of Piraeus Research Center (UPRC).
REFERENCES
[1] “TS 36.300 (V13.2.0): Evolved Universal Terrestrial Radio Access
Network (E-UTRAN) (Release 13),” Technical Specification, 3GPP,
2016.
[2] “1609.3-2016 - ieee standard for wireless access in vehicular environ-
ments (wave) – networking services,” IEEE Standard, 2016.
[3] Q.-T. Nguyen-Vuong, N. Agoulmine, and Y. Ghamri-Doudane, “A user-
centric and context-aware solution to interface management and access
network selection in heterogeneous wireless environments,” Computer
Networks, vol. 52, no. 18, pp. 3358–3372, 2008.
[4] C. Liu, Y. Sun, P. Yang, Z. Liu, H. Zhang, and X. Wen, “A two-step
vertical handoff decision algorithm based on dynamic weight compen-
sation,” in Communications Workshops (ICC), 2013 IEEE International
Conference on. IEEE, 2013, pp. 1031–1035.
[5] D. J. Rani, J. A. Sneha, T. T. M. Delsy, J. A. Glenn, and S. Salih, “A new
technology using decision control algorithm with adaptive multi criteria
vertical handover for hwn,” in Electrical, Electronics, and Optimization
Techniques (ICEEOT), International Conference on. IEEE, 2016, pp.
4863–4868.
[6] F. Riaz, R. Asif, H. Sajid, and M. A. Niazi, “Augmenting autonomous
vehicular communication using the appreciation emotion: A mamdani
fuzzy inference system model,” in Frontiers of Information Technology
(FIT), 13th International Conference on. IEEE, 2015, pp. 178–184.
[7] H. Zhang, X. Wen, B. Wang, W. Zheng, and Y. Sun, “A novel handover
mechanism between femtocell and macrocell for lte based networks,”
in Communication Software and Networks, 2010. ICCSN’10. Second
International Conference on. IEEE, 2010, pp. 228–231.
[8] E. Skondras, A. Sgora, A. Michalas, and D. D. Vergados, “An analytic
network process and trapezoidal interval-valued fuzzy technique for
order preference by similarity to ideal solution network access selection
method,” International Journal of Communication Systems, vol. 29,
no. 2, pp. 307–329, 2016.
[9] B. Ashtiani, F. Haghighirad, A. Makui et al., “Extension of fuzzy topsis
method based on interval-valued fuzzy sets,” Applied Soft Computing,
vol. 9, no. 2, pp. 457–461, 2009.
[10] Z. Shi and Q. Zhu, “Network selection based on multiple attribute
decision making and group decision making for heterogeneous wireless
networks,” The Journal of China Universities of Posts and Telecommu-
nications, vol. 19, no. 5, pp. 92 – 114, 2012.
[11] M. Lahby, C. Leghris, and A. Adib, “New multi access selection method
using differentiated weight of access interface,” in Communications and
Information Technology (ICCIT), 2012 International Conference on.
IEEE, 2012, pp. 237–242.
2017 IEEE Symposium on Computers and Communications (ISCC)