Received December 28, 2017, accepted February 15, 2018, date of publication March 12, 2018, date of current

version April 18, 2018.

Digital Object Identifier 10.1109/ACCESS.2018.2811047

Adaptive Hysteresis Margin Based on Fuzzy

Logic for Handover in Mobile Networks
With Dense Small Cells
KETYLLEN DA COSTA SILVA 1,2 , ZDENEK BECVAR 2, (Senior Member, IEEE),
AND CARLOS RENATO LISBOA FRANCES 1
1 Post -Graduation Program in Electrical Engineering, Federal University of Pará, Belém 66075110, Brazil
2 Department of Telecommunication Engineering, Czech Technical University in Prague, 166 27 Prague, Czech Republic
Corresponding author: Zdenek Becvar ([email protected])
This work was supported in part by the Coordination for the Improvement of Higher Education Personnel-CAPES/Pro-Amazônia and
National Council for Scientific and Technological Development–CNPq and in part by the Czech Technical University in Prague, under
Grant SGS17/184/OHK3/3T/13.

ABSTRACT To satisfy requirements on future mobile network, a large number of small cells should be
deployed. In such scenario, mobility management becomes a critical issue in order to ensure seamless
connectivity with a reasonable overhead. In this paper, we propose a fuzzy logic-based scheme exploiting
a user velocity and a radio channel quality to adapt a hysteresis margin for handover decision in a self-
optimizing manner. The objective of the proposed algorithm is to reduce a number of redundant handovers
and a handover failure ratio while allowing the users to exploit benefits of the dense small cell deployment.
Simulation results show that our proposed algorithm efficiently suppresses ping pong effect and keeps it at
a negligible level (below 1%) in all investigated scenarios. Moreover, the handover failure ratio and the total
number of handovers are notably reduced with respect to existing algorithms, especially in scenario with
high number of small cells. In addition, the proposed scheme keeps the time spent by the users connected
to the small cells at a similar level as the competitive algorithms. Thus, the benefits of the dense small cell
deployment for the users are preserved.

INDEX TERMS Handover, hysteresis margin, mobile networks, small cells, fuzzy logic, self-optimization.

I. INTRODUCTION handover process is a core element of the mobile networks

As demands on mobile traffic are increasing exponentially in terms of a support of the user’s mobility. The handover
due to new mobile devices, services, and applications, mobile also impacts on the overall network performance as shown,
networks should be prepared for a mobile traffic growth over e.g., in [2] or [3]. One of the challenges in mobile networks
the next decade. A powerful technique to address the mobile concerns the need to offer the best possible experience with an
traffic growth is a network densification, i.e., provisioning of infinite number of rapidly proliferating devices. An increase
more base stations to serve a geographical area. With net- in both size and complexity of the current mobile networks
work densification, the system throughput can be enhanced leads to a more complex network management. The recent
exploiting spatial reuse of the spectrum as the access network deployment of the SCeNBs in order to provide higher capac-
is brought closer to the user [1]. The network densification ity brings a significant increase in the number of network
is implemented through a massive deployment of small cell elements, making configuration and maintenance of the net-
base stations (SCeNBs), i.e., base stations with a small cov- work more complicated. In this case, a classic manual and
erage due to a low transmission power. The 5G networks are field trial-based approaches for network planning and man-
expected to deal with many SCeNBs deployed dynamically agement may lead to suboptimal solutions even in case of
and heterogeneously across the network. a minor change of the environment [4]. In the light of this,
In the mobile networks, continuous connection during a it is necessary to assess the effects of the growing number
user’s movement is ensured by handover of a User Equip- of the SCeNBs deployed in the network in terms of mobility
ment (UE) between two base stations (denoted as eNB). The management.

2169-3536
2018 IEEE. Translations and content mining are permitted for academic research only.
17178 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 6, 2018
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

Technical challenges, which must be solved to fully exploit of the art handover decision algorithms (incl. our prior
the dense SCeNB deployment are described, e.g., in [5]–[8]. work [15]) and we show that the ping pong effect
Regarding the mobility management, one of the main prob- can be almost eliminated by the proposed algorithm.
lems is to find a compromise between an elimination of We investigate the performance for various numbers
redundant handovers caused by the dense deployment of of the SCeNBs to confirm robustness of the proposed
SCeNBs and a high utilization of the available communica- algorithm. We also confirm that the time spent by the
tion resources of the SCeNBs [9]. UEs connected to the SCeNB is kept at a similar level
Several schemes addressing mobility management are as for the competitive algorithms guaranteeing that the
described in, e.g., [10] and [11] where the authors investigate benefits of the dense SCeNB deployment are preserved.
the impact of various parameters on the handover and pro- • Fourth, in order to better understand the main sources
pose an adaptive adjustment of these parameters, respectively. of the performance gain, we evaluate an impact of the
Furthermore, in [12], Kuang et al. combine an inter-cell inputs of the proposed fuzzy logic system on the perfor-
interference coordination (ICIC) techniques with the han- mance of the proposed algorithm.
dover process to allow an efficient reduction of interference. The rest of the paper is organized as follows. Next section
Nevertheless, it has negative impact on the ping pong rate presents an overview of related works in the context of
mainly in scenarios with high speed UEs. In [2], interference- handover optimization. In section III, the fuzzy logic scheme
aware and energy-efficient handover decisions exploiting two is outlined and our proposal is described in details. Then,
handover hysteresis margins is proposed to mitigate redun- in section IV, the simulation models and scenarios are spec-
dant handovers. All the schemes in [2], [5]–[12] are able to ified. The simulation results are presented and discussed in
improve specific handover performance metrics, but it is at Section V. Finally, the last section summarizes the major
the cost of a worsening other handover metric(s). conclusions and outlines potential future research directions.
Another option for handover decision optimization is based
on fuzzy logic. In [13], the fuzzy logic is considered for II. RELATED WORKS
an optimization of the handover decision parameters. The This section provides thorough overview of the work related
paper shows promising performance of the fuzzy logic based to optimization of the handover procedure in the mobile
solutions for the handover optimization in a general macro networks.
cell scenario. The fuzzy logic is then exploited also in [14] The main challenges related to handover procedure are
where fuzzy rules and membership functions are adapted studied in [3]. The authors discuss the handover decision
according to previous handovers. Based on these two papers, algorithms and provide a comparison of existing algorithms.
in [15], we outline and evaluate a self-tuning handover algo- The paper points out that the optimization of the handover
rithm (STHA) to diminish the ping pong effect and a han- decision parameters is the most prominent challenge. The
dover failure ratio (i.e., the ratio of handovers not completed authors conclude the values of the handover decision param-
successfully) exploiting knowledge of the UE’s velocity and eters should be changed dynamically and should be adapted
the radio channel quality combined into a fuzzy-logic system in line with the UEs preferences. The handover decision
that represents an additional step in the handover decision based on an adaptation of the handover decision parameters
algorithm. In the STHA [15], first, the conventional handover is discussed, e.g., [16]–[18].
condition should be fulfilled. Then, the level of signal from In [16], Becvar and Mach present a novel handover proce-
the serving eNB is compared with the fuzzy-based generated dure based on an estimation of the UE’s throughput gain. The
threshold for handover initiation gain in throughput is derived from the estimated evolution of
In this paper, we take advantage of findings from our the signals levels of all involved cells and from an estimated
former work presented in [15] and we modify and extend the time spent by the UEs in femtocells. The handover is initiated
STHA it in the following directions: only if the estimated gain in UE’s throughput exceeds a
• First, the fuzzy logic is integrated to the conventional predefined threshold. The results show high efficiency of
handover decision via a dynamically adjusted hysteresis the proposed decision in mitigation of redundant handovers.
margin. This means both signal level from the serving However, the proposed idea is only limited to cells with a
and target eNBs are put into the context of a dynamically very small radius where the estimation is accurate enough and
adjusted hysteresis margin, which is an output of a fuzzy cannot be applied to general small cells.
logic based system. A self-optimization scheme that adjusts the handover
• Second, we propose a new fuzzy inference scheme parameters to minimize Radio Link Failures (RLFs) for
tailored for the handover optimization purposes and dynamic small cell networks is presented in [17]. The scheme
suitable for the proposed handover decision algorithm. first detects the types of RLF and then adjusts the han-
This new inference scheme derives the hysteresis margin dover parameters accordingly. Simulation results show that
dynamically according to the actual UE’s velocity and the scheme can eliminate RLF. Nevertheless, the convergence
the radio channel quality. time of the algorithm is high, and the analysis of the ping pong
• Third, we demonstrate superior performance of the pro- effect is neglected. Another approach with aim to mitigate
posed handover decision scheme with respect to state mobility problems is showed in [18] where a data-driven

VOLUME 6, 2018 17179

 K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

handover optimization approach is proposed. The authors general handover mechanism in [21]. The proposed system
evaluate five types of handovers: too-late handover, too-early allows dynamic optimization considering the probability of
handover, handover to wrong cell, ping pong handover and outage and the probability of handover. The results shown that
unnecessary handover. The proposed approach collects data the GELS outperforms existing handovers. However, the gain
from the signal level measurements and provides a model to is obtained at the cost of high complexity.
estimate the relationship between the Key Performance Indi- The paper [22] proposes a self-optimizing handover hys-
cators (KPI), represented by a weighted average of the five teresis scheme with dual mobile relay nodes for wireless
different mobility problem ratios, and features from the col- networks in high-speed mobile environments. The proposed
lected dataset. Based on the model, the handover parameters, mechanism adapts the hysteresis margin and cell individual
including the Hysteresis Margin and time-to-trigger (TTT), offset based on the velocity of the vehicle and the handover
are optimized to minimize the KPI. Simulation show that performance indicator of the cell characterized as a sum-
the proposed approach could effectively mitigate mobility mation of the handover failures, handover ping pong ratio,
problems. However, the neural network considered to esti- and RLF indicators. The results show superior performance
mate the KPI function requires a large diversity of training with respect to the conventional schemes and confirms that
for real-world operation, which is a notable constraint for the fixed parameter setting is adequate for common scenar-
the mobile networks with dense SCeNBs operating in a self- ios. However, in a high-speed environment, a more flexible
optimizing manner. scheme is necessary. In [23], Xenakis et al. target to optimize
Lee et al. [19] propose a cost-based adaptive handover hys- the handover by employing interference-aware and energy-
teresis scheme that focuses on a performance improvement efficient handover decisions exploiting two handover hystere-
in terms of the handover failure ratio in real time. The cost sis margins. The first hysteresis margin is considered to avoid
function for vertical handover in heterogeneous network is cells that can compromise service continuity, e.g., due to poor
provided as a weighted sum of normalized functions by fol- channel conditions. The second hysteresis margin identifies
lowing dominant factors: a load difference between the target the cell with the minimum requirements in terms of RSRP.
and serving eNBs, UE’s velocity, and the service type. The The simulation results indicate that the proposed algorithm
simulation results show a lower handover failure ratio than allows to double the macrocell offloading ratio, enhance the
the compared schemes, but other important handover perfor- uplink capacity, and reduce the interference level at UEs.
mance indicators, such as the number of handovers or ping On the other hand, handover probability is increased.
pong ratio are not considered. The papers [24] and [12] show that LTE-based networks
Besides hysteresis margin, also other parameters can be can be efficiently managed by algorithms based on fuzzy
considered for adjustment of the handover procedure. In [10], logic. The idea of fuzzy logic is exploited also for han-
an impact of various offsets and timers on the efficiency of dover control. For example, in [13], a fuzzy system-based
the handover in heterogeneous networks is investigated. The handover decision is proposed. This system modifies the
analysis shows that large and positive offset values, which handover decision-related parameters to adjust the service
are typical for macro-only deployments, lead to a higher area of an eNB. Thus, the coverage of the congested eNB
number of handover failures in heterogeneous networks. is reduced while coverage of the adjacent less loaded eNBs
On the other hand, small and negative offset values is increased. The fuzzy system is improved by a Q-Learning
result in a more frequent ping pong (i.e., repeated non- to select the most appropriate action targeting either the load
beneficial handovers performed between two eNBs). In [12], balancing or the handover optimization. The user dissatisfac-
an impact of three handover-related parameters, i.e., time- tion in terms of the call blocking ratio and call dropping ratio
to-trigger, hysteresis margin, and Reference Signal Received keep values similar to baseline schemes. However, omitting
Power (RSRP), is analyzed in terms of the handover failure mobility related metrics and parameters (such as velocity
ratio and the ping pong ratio in an environment close to a of the UE) from the handover decision leads a performance
real world. The authors also investigate an impact of cell degradation if the small cells are deployed.
range expansion (CRE) and ICIC techniques on the mobil- Another similar handover decision algorithm based on
ity support efficiency. The paper reveal that the impact of fuzzy logic is presented in [14]. The algorithm enables to
CRE and ICIC depends on the handover type (i.e., macro-to- adapt fuzzy rules and membership functions according to
pico, pico-to-macro, pico-to-pico and macro-to-macro). The historical data available within a tracking area. Three inputs
authors further propose a dynamic ICIC mechanism assisting are considered for this algorithm: RSRP, Block Error Rate and
the handover process. The proposed combination of han- Quality of Service (QoS). The implementation demonstrates
dover with ICIC allows an efficient reduction of interference. that this approach minimizes operating expenses and the
Nevertheless, it increases the ping pong ratio mainly in high number of unnecessary handovers by 20% when comparing
speed scenarios. to the standard LTE handover. An advancement of the fuzzy
In [20], Fischione et al. propose a Generalized Extended logic–based handover is presented in [25] where Aibinu et al.
Last Squares Handover (GELS) to select and optimize the propose a hybrid artificial intelligent handover decision. The
hysteresis margin. This work is further enhanced by a new RSS is considered as a trigger of the handover procedure
modeling of the handover process taking into account a and it is accompanied by a neural system forecasting the

17180 VOLUME 6, 2018

K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

number of users in the network. The neural network is used The RSRP is defined as a linear average of the powers
for a determination of the model coefficients for an effective received at specific resources (reference signals) spanned
prediction of the RSS level in the handover decision manage- over whole frequency bandwidth. The handover is triggered
ment system. However, as in the previous paper, the mobility on the basis of the measurement reports received by the eNB
related parameters are not taken into account. Moreover, from the UE. In its simple form, the handover decision is
the algorithm is of very high complexity so its suitability for initiated if the following condition prevails for a particular
the dense SCeNBs deployment is limited. period of time:
Furthermore, in [26], Hussein et al. propose a Fuzzy
Multiple-Criteria Cell Selection (FMCCS) to optimize the RSRPNeNB > RSRPSeNB + 1HM (1)
handover procedure. The FMCCS method considers a fuzzy
where RSRPNeNB and RSRPSeNB are the levels of RSRP
system integrated with a Technique for Order Prefer-
from the neighboring and serving eNBs, respectively; and
ence by using Similarity to Ideal Solution (TOPSIS). The
1HM stands for the hysteresis margin. The purpose of 1HM
fuzzy-based TOPSIS is employed as a Multiple-Criteria
is to avoid redundant ‘‘ping pong’’ handovers when the UE is
Decision-Making (MCDM) process that performs ratings
continuously handed over between two eNBs.
and weighting of criterions represented by linguistic vari-
ables. The handover decision considers a combination of
B. PROPOSED HANDOVER DECISION WITH DYNAMIC
S-criterion, where S refers to signal quality in the downlink
HYSTERESIS MARGIN
(defined as RSRP), availability of resource blocks for data
In our previous work [15], the handover decision is based
transfer, and Signal to Interference plus Noise Ratio (SINR).
on a new added condition that is evaluated once common
The results show that the FMCCS reduces frequency of ping
handover decision condition (1) is fulfilled. The decision
pong handovers and handover failure ratio. The scheme is
condition imposes that the handover is performed only if all
evaluated only in macrocells scenarios and there is no assur-
the inputs of the fuzzy algorithm indicate that the serving
ance of flexibility for dense environments.
connection is not sufficient to provide the required throughput
In [15], we introduce a self-tuning handover algorithm for
and handover to one of the neighboring cells is more advan-
mobile networks with dense small cells. The proposed STHA
tageous. In contrast, now, we derive a dynamic hysteresis
allows to estimate if the handover to a new cell is efficient
margin determined by the multi-criteria fuzzy logic system
or not and avoid handovers, which are not seen beneficial.
improving the handover decision in a self-optimizing manner.
The STHA extends the common handover decision by a new
The fuzzy logic is efficiently used in the context of com-
fuzzy logic-based handover condition comparing the signal
plex ill-defined processes and it is suitable to handle a large
level from the serving eNB with a dynamic threshold of signal
number of imprecise parameters involved in the handover
level. The simulation results, presented in [15], show that the
decision. In this sense, we propose a mechanism capable to
proposed algorithm reduces handover failure ratio and ping
reduce the number of redundant handovers while keeping
pong effect.
a high utilization of SCeNBs. The general principle of our
With respect to all-above mentioned papers, the novelty
proposal is as follows.
of our proposal consists in design of a new handover deci-
In the first step, the RSRP of the serving eNB is compared
sion scheme, which is capable of dynamic adaptation of the
with the RSRP of the neighboring eNBs. Then, if any neigh-
hysteresis margin based on a new fuzzy logic system. The
boring eNB offers the RSRP of a higher level than the serving
system is designed so that it allows to minimize the number
eNB, the fuzzy system determines the value of the dynamic
of redundant handovers, ping pong handovers, and handover
hysteresis margin. In the third step, the dynamic hysteresis
failure ratio. Simultaneously, the time when the UEs are
margin derived by the fuzzy logic system is considered for
connected to the SCeNBs remains unimpaired so the benefits
the handover decision. These main steps of the proposed
of dense SCeNBs deployment are preserved.
algorithm are summarized as follow:
III. PROPOSED FUZZY LOGIC HANDOVER DECISION 1) Check a preliminary condition for triggering the pro-
In this section, a novel fuzzy logic-based scheme for dynamic cess of handover decision assessment:
adjustment of the hysteresis margin is presented. To this
end, we first summarize basic principle of a general existing RSRPNeNB > RSRPSeNB (2)
handover decision algorithm based on the hysteresis margin.
Then, we give a high-level overview of the proposed handover This condition avoids redundant assessment of the han-
decision algorithm incorporating dynamic fuzzy logic control dover decision condition, so it can be omitted without
of the hysteresis margin. Last, we provide details of the any impact on the handover performance.
proposed algorithm including realization of the developed 2) If (2) is fulfilled, determine thedynamic hysteresis
fuzzy logic system. margin (1HM ,d ) by a new fuzzy logic-based system.
3) Perform a common handover decision considering the
A. CONVENTIONAL HANDOVER DECISION new fuzzy logic-based hysteresis margin, i.e.,:
The conventional handover decision is based on a compari-
son of RSRP from the serving and neighboring eNBs [27]. RSRPNeNB > RSRPSeNB + 1HM ,d (3)

VOLUME 6, 2018 17181

 K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

The first and the third steps are common for many han-
dover decision algorithms. Thus, the main novelty lays in
the second step, fuzzy logic-based system for determina-
tion of the 1HM ,d , which is described in the following
subsection.

C. FUZZY LOGIC SYSTEM FOR DETERMINATION OF

DYNAMIC HYSTERESIS
The fuzzy logic systems have been developed to manage
vagueness and uncertainty in a reasoning process of an intel- FIGURE 2. Illustration of the membership functions of a general fuzzy
ligent system, such as a knowledge based system, an expert logic system.

system, or a logic control system [28]. The fuzzy logic

systems are very useful for automatic network parameter opti-
mization, which is composed of three basic steps: collecting A triangular fuzzy number à can be denoted by a triplet
data, evaluation of the data, and performing a control action. (a1, a2, a3). The mathematical form is expressed as follows:

In Fig. 1, an architecture of the proposed fuzzy system is  0 x ≤ a1

presented. To easy understanding of the proposed concept, 
 x − a1
a1 < x ≤ a2


we first explain an architecture of general fuzzy logic sys- 
− a1
tems, and then, we explain its application to our handover µÃ(x) = a2 a3 − x (4)
decision problem.

 a2 < x ≤ a3
a3 − a3



x > a3

0

The next block of the fuzzy logic system, the Rules Base,
comprises all possible relationships among the system inputs
and output. An example of a fuzzy system with two inputs x1
and Center of Gravity x2 (antecedents) and a single output y
(consequent) is described by a collection of r linguistic
IF–THEN propositions in the form:
IF x1 is Ãk1 and IF x2 is Ãk2 THEN y is B̃K for k=1, 2, . . . , r
FIGURE 1. Illustration of general fuzzy logic architecture with inputs and
output of the proposed fuzzy system for the handover decision algorithm. (5)
where Ãk1 and Ãk2 are the fuzzy sets representing k-th inputs
The first step performed in the fuzzy logic system is the (antecedents) and B̃k is the fuzzy set representing the k-th
fuzzification process. In this step, the crisp inputs are trans- output (consequent).
lated into linguistics variables (e.g., low, medium,high) and a The key part of each fuzzy logic system is the Inference
membership function is calculated for each input of the fuzzy engine, which identifies rules to be triggered and calculates
system. The membership function is a curve that defines how the fuzzy values of the output variables using a max–min
each point in the input space is mapped to a membership inference method [31]. This method tests the magnitudes of
value (or degree of membership). Typically, the membership each rule and selects the highest one. The main advantage of
functions are expressed in a form of mathematic functions. the max-min method is its computational simplicity. In the
The fuzzification process also involves transformation of the Inference engine, a fuzzy implication operator is applied to
values of input variables and a scale mapping. The scale obtain a new fuzzy set based on the consequent of each rule
mapping translates the range of the inputs values into (a fuzzy set) and obtained antecedent value. Then, the outputs
corresponding universes of discourse finding the fuzzy rep- obtained for each rule are combined into a single fuzzy set
resentation of non-fuzzy input values. using a fuzzy aggregation operator. In other words, the rule
The second block, Data Base, defines the fuzzy member- with the highest degree of truth is selected and then, the con-
ship functions that allow to assign the grades of membership sequent membership function to be activated is determined.
to the fuzzy sets. Such an assignment is built from concepts, The output (consequent) is given by:
which are subjectively defined and based on expert knowl- h h ii
edge. The fuzzy set à in a universe of discourse X is described µB̃k (y) = max min µÃk (VEL), µÃk (RSRP), µÃk (RSRQ)
k 1 2 3
by the membership function µÃ(x) where x ∈ h0, 1i and for k = 1, 2, . . . , r (6)
x ∈ <. The function value µÃ(x) is denoted as the grade of
the membership of x in Ã. As shown in Fig. 2, we adopt a tri- Finally, the Defuzzification process, which has the oppo-
angular membership function for all the inputs and the output, site meaning of the Fuzzification, provides a non-fuzzy
as this function is suitable for real-time operation due to their control action from an inferred fuzzy control action. This
simplicity in modeling and easy interpretation [29], [30]. step consists in a transformation of the aggregated fuzzy

17182 VOLUME 6, 2018

K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

FIGURE 3. Membership functions for inputs: (a) µVEL, (b) µRSRP, and (c) µRSRQ.

set µB̃k (y) into one single crisp number. This transformation considered as the third input. Fig. 3c shows the sets
corresponds to the determination of the Center of Grav- of µRSRQ defined in the following way: Poor (from
ity (COG) [32]. The weighted average of the membership −60 to −18 dB), Good (from −22 to −12 dB), Very Good
function or the center of gravity of the area bounded by the (from −14 to −6 dB), and Excellent (from −10 to +20 dB)
membership function is computed as follows: with respect to definition of the range of RSRQ in 3GPP.
N The intervals and granularity of all input parameters are
µB̃(y)y
P
defined based on the ranges of values commonly expected in
i=1 mobile networks as assumed in 3GPP [34].
µB̃k (y) = (7)
N All three inputs are combined by the Inference Engine
µB̃(y)
P
i=1
into the output of the proposed fuzzy system represented by
1HM ,d . Four fuzzy sets are defined for the output 1HM ,d
where µB̃(y)y is the centroid of each symmetric membership
to achieve a reasonable granularity in the output space (see
function.
Fig. 4): Very low (from 1 to 4.5 dB), Low (from 3.5 to 7 dB),
In our proposal, we consider three inputs in the Fuzzifica-
Average (from 6 to 9.5 dB), and High (from 8.5 to 12 dB).
tion process: UE’ velocity, RSRP, and RSRQ. The first input
parameter is the UE’s velocity denoted as µVEL. High-speed
UEs may pass through the SCeNB in a short time interval,
causing frequent handover leading to a massive handover
signaling overhead to the network [33]. In Fig. 3a, we express
the characterization of the three sets of the UE’s velocity:
Slow (from 0 to 10 km/h), Moderate (from 8 to 50 km/h),
and Fast (from 45 to 80 km/h). It is important to notice
that the membership functions are overlapping due to the
smooth transition boundary, which is an underlying charac-
teristic of the fuzzy sets; i.e., the precise input values during
fuzzification process can belong to more than one fuzzy set
with the different degree of membership shown in individual
membership functions of each parameter. FIGURE 4. Membership function for the output of the proposed fuzzy
As the second input, we consider a received signal power logic system 1HM,d .
represented by RSRP, denoted as µRSPR, considering fol-
lowing three states (defined based on a common range We formulate 36 fuzzy rules (the number of rules is deter-
of RSRP in 3GPP): Weak (−160 to −95 dBm), Moder- mined by the combination of all possible states of all three
ate (−100 to −73 dBm), and Strong (−80 to −20 dBm), input variables, i.e., 3 x 3 x 4). The rules are defined consid-
as shown in Fig. 3b. With the aim to define a suitable eNB ering following aspects and requirements:
for handover, the RSRP measurement provides an indication • If µVel is Low the 1HM ,d should be set to lower values to
of the eNB coverage, and the received signals strengths that provide a freedom to find the most suitable eNBs. On the
is required for the handover decision. other hand, if the µVel is High the 1HM ,d should be set
Furthermore, considering scenarios with the dense SCeNB to higher values to avoid the premature handover. In this
deployment, an efficiency of the radio communication case, it is preferred to temporary tolerate a suboptimal
depends not only to the signal level from the serving eNB, connection rather than perform unnecessary handovers.
but also on noise and interference caused by the neighbor- • If µRSRP is Weak the 1HM ,d should be set to lower
ing eNBs. Thus, parameter RSRQ (denoted as µRSRQ) is values to facilitate the handover, while if the µRSRP is

VOLUME 6, 2018 17183

 K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

TABLE 1. Rules formulated for determination of µ1HM,d . Algorithm 1 Proposed Handover Decision Algorithm With
Fuzzy Logic-Based Determination of Dynamic Hysteresis
Margin
1. IF RSRPNeNB > RSRPSeNB
2. Converts {µVEL, µRSRP and µRSRQ} to fuzzy sets
3. Calculate the degree of truth for each fuzzy rule acc.
to (4)
4. Computes the antecedent of each k rule by implication
operator: h i
µB̃k (y) = min µÃk (VEL), µÃk (RSRP), µÃk (RSRQ)
1 2 3
5. Calculate the outputs of each triggered rule acc. to the
proposed rule base and the membership functions of
the output defined in TABLE 1
6. Aggregate outputs obtained for each rule into a single
fuzzy set:
µ1̃HM ,d = max µBk˜ (y)
 
k
7. Transform the output value of µ1̃HM ,d into a crisp
value 1HM ,d by center of gravity method acc.to (7)
8. IF RSRPNeNB > RSRPSeNB + 1HM ,d
9. PERFORM handover
10. ELSE DO NOT PERFORM handover
11. END

performance metrics are defined. Then, the last subsection

describes the competitive state of the art algorithms consid-
ered for the performance comparison.

Strong the 1HM ,d should be set to higher values to keep

FIGURE 5. Example of the simulation deployment with eNBs represented
the current connection. by red triangles, SCeNBs by black crosses, and UEs by blue circles.
• If µRSRQ is Poor the 1HM ,d should be set to lower
values to facilitate the handover since there are no advan-
tages of insisting on a bad connection. On the other hand, A. SIMULATION MODELS AND SCENARIOS
if the µRSRQ is Excellent the 1HM ,d should be set to The simulations are performed in MATLAB. We assume an
higher values to maintain the connection experience. area with a size of 1000 m x 1000 m. Within this area,
TABLE 1 shows the rules and consequent definition of two eNBs, up to 200 SCeNBs, and 50 UEs are deployed
the µ1HM ,d . as depicted on an example in Fig. 5. The SCeNBs and
Individual steps of our proposed fuzzy logic based the UEs are randomly dropped with a uniform distribution
handover decision algorithm (by means of determination at the beginning of the simulation while both eNBs are
of adaptive hysteresis margin 1HM ,,d are summarized in located at predefined positions close to the area’s corners
Algorithm 1. so that these eNBs can provide coverage for the whole
simulation area. The UEs move according to the random
IV. PERFOMANCE EVALUTION METHODOLOGY waypoint mobility model (see [35] for more details). For
In this section, models, scenarios, and deployment used variability, the speed is randomly selected in a range from
for performance evaluations are presented. Afterwards, the 0 to 80 km/h.

17184 VOLUME 6, 2018

K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

TABLE 2. Simulation parameteres. In our case, the handover failure events are determined acord-
ing to the downlink SINR. As in LTE-A, we assume that
when SINR is lower than the threshold Qout, a bad channel
condition is indicated and the T310 times is started. The
handover failure is declared when T310 expires.
Then, the handover failure (HF) ratio, is defined as a
ratio between the number of handover failures (Nfail ) and the
number of all handover attempts. The number of handover
attempts is given by the sum of the number of the failed
handover and the number of successful handovers (Nsuc ):
Nfail
HF = (9)
Nfail + Nsuc
The third metric, the handover ping pong (HPP) ratio,
The signal propagation from the base stations to the UEs is is defined as follows. If a connection is handed over to a
derived according to the models recommended by Small Cells new (SC)eNB and handed back to the original (SC)eNB
Forum. Hence, Okumura-Hata and ITU-R P.1238 models in less than a critical time, denoted as minimum time-of-
are used for the signal propagation from the eNBs and the stay (tMTS ), the handover is considered as the ping pong han-
SCeNBs, respectively. The major simulation parameters are dover. The ping pong handover ratio represents the number
summarized in TABLE 2. of ping pong handovers (NPP ) divided by the total number of
handovers including: i) the number of ping pong handovers,
B. PERFORMANCE METRICS ii) the number of handovers without ping pong (NnPP ),
Following metrics are considered for the performance assess- i.e., with stay longer than tMTS , and iii) the number of failed
ment and comparison of the performance with competi- handovers (Nfail ). Then, the HPP is formulated as:
tive solutions: Average number of performed handovers, NnPP
Handover failure ratio, Ping pong ratio, and Time spent in HPP = (10)
NnPP + NPP + Nfail
SCeNBs.
The average number of handovers (NHO,AVG ) is calculated Last, the relative time spent in small cells (tSCeNB ) is under-
as a sum of the number of handovers performed by all stood as an average duration of the connection of the UEs to
UEs (NHO ) over the total number of the UEs in the simula- the SCeNBs (tconnSC ) over the simulation time (tsim ):
tion (NUE ). P 
tconnSC
NHO NUE
NHO,AVG = (8) tSCeNB = (11)
NUE tsim

C. COMPETITIVE ALGORITHMS
In our simulations, the proposed handover algorithm is com-
pared with three competitive schemes and with our previous
work to demonstrate the superiority of the proposal. The
following state of the art algorithms implemented for the
performance comparison are considered:
• Best Connection (BC) representing the case when the
UE is always connected to the (SC)eNB providing the
highest RSRP.
• Conventional LTE handover (in figures denoted as LTE)
is implemented according to 3GPP as defined in [37].
FIGURE 6. Illustration of radio link monitoring process and handover
process for determination of handover failure [36]. • Fuzzy Multiple-Criteria Cell Selection (FMCCS)
defined in [26] is based on fuzzy logic integrated with
For modelling of the handover failure ratio, the handover TOPSIS.
procedure is divided into three states according to LTE [36], • Self-Tuning Handover Algorithm (STHA) defined in [15]
as shown in Fig. 6. The Stage 1 is defined by the instant that adds a new fuzzy-based handover condition on the top
precedes the Event A3 condition in 3GPP. In Stage 2, the UE of convention handover condition to improve handover
triggers the measurement reporting if the Event A3 condition decision. This algorithm is our former work, which
holds throughout the time-to-trigger duration. Stage 3 occurs serves as a basement for the algorithm developed in this
when the UE successfully receives handover command from paper. Thus, we include it to demonstrate superiority of
the serving eNB and starts the handover execution process. the new proposal with respect to STHA.

VOLUME 6, 2018 17185

 K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

V. PERFORMANCE EVALUATION
In this section, the results of simulations are presented to
provide a comparison of the performance with respect to
the competitive approaches. This section is divided into two
subsections. In the first subsection, the performance of the
proposed and competitive algorithms is compared and dis-
cussed. Then, we evaluate the impact of fuzzy logic system
inputs on the performance of our algorithm.

A. COMPARISON OF PERFORMANCE WITH

COMPETITIVE ALGORITHMS
First, we investigate an impact of the number of SCeNBs
on the average number of handovers performed by the UEs
(see Fig. 7). As can be expected, the number of handovers
increases with the number of SCeNBs, because more cell FIGURE 8. Impact of the number of SCeNBs on the handover failure ratio.
edges appear in the network and the UEs are forced to per-
form handover more often to avoid connection losses. The
proposed algorithm significantly outperforms all compared
schemes thanks to the smart adaptation of the hysteresis
margin allowing to mitigate redundant handovers. The gain
introduced by the proposed algorithm increases with the
number of SCeNBs and reaches 18%, 22%, 33% and 52%
improvement comparing to STHA, FMCCS, LTE and BC
algorithms, respectively, for 200 SCeNBs deployed in the
simulation area.

FIGURE 9. Impact of the number of SCeNBs on the handover ping pong

ratio for tMTS = 2s.

algorithm is still able to keep low HF ratio (around 3%) even

for 200 SCeNBs.
An impact of the number of SCeNBs on the ping pong
effect (represented by the HPP ratio) is illustrated in Fig. 9 for
tMTS = 2s. This figure shows superiority of the proposed
algorithm, which keeps the HPP ratio always below 0.5%
FIGURE 7. Impact of the number of SCeNBs on the average number of for all investigated numbers of SCeNBs. This eminent perfor-
handovers performed by the UEs. mance is achieved by the fact that our proposal optimizes the
handover hysteresis margin directly related to the diminished
In Fig. 8, the handover failure ratio is depicted. It can be ratio of unnecessary handovers. In contrast, the competitive
seen that the HF ratio increases with the number of SCeNBs. solutions lead to a relatively high HPP ratio (1.5%, 2%, 4.4%,
This increase is caused by a stronger interference imposed and 7.5%, for STHA, FMCCS, LTE, and BC, respectively) for
in the scenario with more SCeNBs leading to rapid drops low to medium densities of the SCeNBs (up to 50 SCeNBs).
in SINR and consequent failures of the handover. The fig- The HPP of all four competitive solutions lowers for a higher
ure further shows that, for the low densities of the SCeNBs density of the SCeNBs. This decrease is due to the fact that
(roughly up to 100 SCeNBs), the proposed algorithm reaches with the higher density of the SCeNBs, the probability that
similar performance as our previous work – the STHA. Both the connection is handed over to a new SCeNBs is increased
the proposed algorithm and STHA outperform all three com- comparing to the probability of the handover back to the
petitive algorithms and lower the HF ratio to almost a half. former serving SCeNB. Nevertheless, even for 200 SCeNBs,
For higher numbers of the SCeNBs (more than 100), the the HPP ratio is still at 0.7%, 1.9%, 3.8%, and 6.2% for the
performance of STHA is getting worse and converges to the STHA, FMCCS, LTE, and BC algorithms, respectively, while
FMCCS (i.e., to HF ratio about 4.5%). Contrary, the proposed the proposed scheme keeps the HPP ratio at 0.4%.

17186 VOLUME 6, 2018

K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

FIGURE 10. Impact of tMTS on the ping pong ratio for 200 SCeNBs (solid FIGURE 12. Impact of inputs to the fuzzy logic system (RSRP, RSRQ,
lines) and 50 SCeNBs (dashed lines). velocity) on the average number of performed handovers.

all three inputs of the fuzzy system resulting in the hysteresis

As the ping pong effect is defined by the tMTS , we also margin that is adapted according to channel quality.
demonstrate the impact of tMTS on the HPP ratio in Fig. 10.
The HPP ratio increases with tMTS , because more handovers B. IMPACT OF FUZZY SYSTEM IMPUTS
are considered as ping pong with increasing tMTS . We can ON THE PERFORMANCE
see that the reduction in HPP ratio demonstrated in Fig. 9, In this subsection, we evaluate and discuss the impact of
is valid for a wide range of tMTS and the proposed algorithm individual inputs to the fuzzy system and their combinations
outperforms all competitive schemes disregarding the tMTS . in order to identify an importance of the inputs on the overall
performance. As in previous subsection, four performance
metrics, Average Number of Handovers, Handover Failure
ratio, Ping pong ratio and Average time in SCeNBs, are
considered and investigated.
In Fig. 12, the average number of performed handovers is
shown. We can see that the velocity and RSRQ are key inputs
to the fuzzy logic system. The combination of velocity and
RSRQ leads to the lowest number of handovers and inclusion
of RSRP do not change this performance metrics notably,
because, from the handover decision point of view, RSRQ
already includes information about RSRP.

FIGURE 11. Average time spent by the UEs connected to the SCeNBs.

Last, we analyze the average time spent by the UEs

connected to the SCeNBs. Many handover algorithms focus
on the mitigation of handovers while decreasing the time
when the UEs are connected to the SCeNBs. Nevertheless,
this mitigation leads to an underutilization of resources pro-
vided by the SCeNBs and consequently to a loss in the poten-
tial of the SCeNBs to improve network throughput. As we can
see in Fig. 11, our proposal even slightly improves the time
spent by the UEs in SCeNBs comparing to all competitive
algorithms. The most notable gain (from 1% to 4% depending FIGURE 13. Impact of inputs to the fuzzy logic system (RSRP, RSRQ,
of the number of SCeNBs) is introduced with respect to the velocity) on the handover failure ratio.
STHA, which reaches the closest performance in all other
investigated performance metrics. The prolongation of the The impact of various combinations of the inputs on HF
time spent in SCeNBs is introduced by the combination of ratio is depicted in Fig. 13. Like in previous figure, the

VOLUME 6, 2018 17187

 K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

VI. CONCLUSIONS
In this paper, we have introduced a novel handover decision
algorithm exploiting new fuzzy logic system for dynamic
determination of the hysteresis margin. The proposed algo-
rithm leads to a superior performance improving key han-
dover performance indicators comparing to state of the art
solutions. The proposed solution almost eliminates handover
ping pong effect and, besides, it also reduces the handover
failure ratio and the total number of handovers comparing
to the state of the art algorithms. With respect to the com-
petitive solutions, the achieved improvements are not at the
cost of reduced time spent by the UEs in the SCeNBs. The
proposed algorithm keeps the time spent in the SCeNBs at
the same level as the algorithms, which demonstrate worse
FIGURE 14. Impact of inputs to the fuzzy logic system (RSRP, RSRQ, performance in terms of the number of handover, ping pong
velocity) on the handover ping pong ratio. ratio and handover failure ratio. Thus, the proposed algorithm
allows to preserve the benefits of the SCeNBs while all
velocity and RSRQ are of the major importance for the HPP key handover performance indicators are notably improved.
ratio and improvement introduced by further inclusion of the This indicates suitability of the proposed algorithm for future
RSRP is negligible as the key information related to the HF mobile networks with a very high density of the SCeNBs.
(i.e., channel quality) is already covered by the RSRQ. In the future work, the research should focus on a pre-
In Fig. 14, we analyze the impact of the fuzzy logic sys- diction of a signal level for determination of the throughput
tem inputs on HPP ratio. In this case, we can see that the gain and investigate the potential improvements regarding
performance gain is generated again by a combination of handover performance indicators.
velocity and RSRQ; however, RSRP can further improve the
HPP ratio for higher densities of SCeNBs notably. The RSRP REFERENCES
provides the UE with essential information about the strength [1] P.-H. Huang, H. Kao, and W. Liao, ‘‘Cross-tier cooperation for optimal
of signal from the serving and neighboring cells. This helps to resource utilization in ultra-dense heterogeneous networks,’’ IEEE Trans.
Veh. Technol., vol. 66, no. 12, pp. 11193–11207, Dec. 2017.
determine the optimum time for handover and to avoid ping
[2] D. Xenakis, N. Passas, L. Merakos, and C. Verikoukis,
pong if the handover is not necessary. ‘‘Handover decision for small cells: Algorithms, lessons learned and
Last, the impact of various combinations of the inputs on simulation study,’’ Comput. Netw., vol. 100, pp. 64–74, May 2016.
time spent in small cells is depicted in Fig. 15. We can observe [3] R. Ahmad, E. A. Sundararajan, N. E. Othman, and M. Ismail, ‘‘Han-
dover in LTE-advanced wireless networks: State of art and survey of
minor variation of the time for all input combinations. decision algorithm,’’ Telecommun. Syst., vol. 66, no. 3, pp. 533–558,
2017.
[4] P. Munoz, R. Barco, and I. de la Bandera, ‘‘On the potential of handover
parameter optimization for self-organizing networks,’’ IEEE Trans. Veh.
Technol., vol. 62, no. 5, pp. 1895–1905, Jun. 2013.
[5] T. Levanen, J. Pirskanen, and M. Valkama, ‘‘Dense small-cell networks:
Rethinking the radio interface beyond LTE-advanced,’’ in Proc. Int. Conf.
5G Ubiquitous Connectivity, Nov. 2014, pp. 163–169.
[6] J. G. Andrews, X. Zhang, G. D. Durgin, and A. K. Gupta, ‘‘Are we
approaching the fundamental limits of wireless network densification?’’
IEEE Commun. Mag., vol. 54, no. 10, pp. 184–190, Oct. 2016.
[7] N. Panwar, S. Sharma, and A. K. Singh, ‘‘A survey on 5G: The next
generation of mobile communication,’’ Phys. Commun., vol. 18, pp. 64–84,
Mar. 2016.
[8] Y. Sun, Y. Chang, M. Hu, and T. Zeng, ‘‘A universal predictive
mobility management scheme for urban ultra-dense networks with
control/data plane separation,’’ IEEE Access, vol. 5, pp. 6015–6026,
2017.
[9] M. Vondra and Z. Becvar, ‘‘Distance-based neighborhood scanning for
handover purposes in network with small cells,’’ IEEE Trans. Veh.
Technol., vol. 65, no. 2, pp. 883–895, Feb. 2016.
FIGURE 15. Impact of inputs to the fuzzy logic system (RSRP, RSRQ, [10] M. Mehta, N. Akhtar, and A. Karandikar, ‘‘Impact of handover parame-
velocity) on the average time spent by the UEs connected to the SCeNBs. ters on mobility performance in LTE HetNets,’’ in Proc. 21st Nat. Conf.
Commun., Feb./Mar. 2015, pp. 1–6.
[11] K. Kitagawa, T. Komine, T. Yamamoto, and S. Konishi, ‘‘A handover
From the analysis of all three inputs, we can conclude optimization algorithm with mobility robustness for LTE systems,’’ in
that the velocity of the UE and RSRQ are the most impor- Proc. IEEE PIMRC, Sep. 2011, pp. 1647–1651.
tant inputs for the proposed fuzzy logic system. However, [12] Q. Kuang, J. Belschner, Z. Bleicher, H. Droste, and J. Speidel,
‘‘A measurement-based study of handover improvement through range
consideration of the third inputs, RSRP, is useful for reduction expansion and interference coordination,’’ Wireless Commun. Mobile
of the ping-pong ratio in scenarios with the dense SCeNBs. Comput., vol. 15, no. 14, pp. 1784–1798, 2015.

17188 VOLUME 6, 2018

K. da Costa Silva et al.: Adaptive Hysteresis Margin Based on Fuzzy Logic for Handover in Mobile Networks

[13] P. Muñoz, R. Barco, and I. de la Bandera, ‘‘Load balancing and handover [35] E. Hyytiä and J. Virtamo, ‘‘Random waypoint mobility model in cellular
joint optimization in LTE networks using fuzzy logic and reinforcement networks,’’ Wireless Netw., vol. 31, no. 2, pp. 177–188, 2007.
learning,’’ Comput. Netw., vol. 76, pp. 112–125, Jan. 2015. [36] Technical Specification Group Radio Acess Network; Mobility Enhance-
[14] C. F. Kwong, T. C. Chuah, S. W. Tan, and A. Akbari-Moghanjoughi, ments in Heterogeneous Networks (Release 11), document TR 36.839
‘‘An adaptive fuzzy handover triggering approach for long-term evolution V11.1.0, 3rd Generation Partnership Project, Dec. 2012.
network,’’ Expert Syst., vol. 33, no. 1, pp. 30–45, 2016. [37] Evolved Universal Terrestrial Radio Access (E-UTRA); Radio Resource
[15] K. C. Silva, Z. Becvar, E. Cardoso, C. R. L. Francês, ‘‘Self-tuning han- Control (RRC); Protocol Specification, document TS 36.331, 3GPP,
dover decision based on fuzzy logic in mobile networks with dense small Jul. 2011.
cells,’’ in Proc. IEEE WCNC, Apr. 2018, pp. 1–6. [Online]. Available:
http://5gmobile.fel.cvut.cz/Papers~/WCNC18_STHA_Fuzzy.pdf
[16] Z. Becvar and P. Mach, ‘‘Mitigation of redundant handovers to femto- KETYLLEN DA COSTA SILVA received the B.Sc.
cells by estimation of throughput gain,’’ Mobile Inf. Syst., vol. 9, no. 4, degree in computer science from Amazon Uni-
pp. 315–330, 2013. versity in 2010 and the M.Sc. degree in electrical
[17] M.-H. Song, S.-H. Moon, and S.-J. Han, ‘‘Self-optimization of handover engineering with emphasis in applied computing
parameters for dynamic small-cell networks,’’ Wireless Commun. Mobile from the Federal University of Pará in 2014, where
Comput., vol. 15, no. 11, pp. 1497–1517, Aug. 2015. she is currently pursuing the Ph.D. degree. She was
[18] P. Lin, L. F. G. Casanova, and B. K. S. Fatty, ‘‘Data-driven handover on internship with Czech Technical University in
optimization in next generation mobile communication networks,’’ Mobile Prague, Czech Republic, in 2017. She is currently
Inf. Syst., vol. 2016, 2016, Art. no. 2368427. a member of LPRAD (Laboratory of High Per-
[19] D.-W. Lee, G.-T. Gil, and D.-H. Kim, ‘‘A cost-based adaptive handover formance Network). Her research interests include
hysteresis scheme to minimize the handover failure rate in 3GPP LTE radio access technologies, performance evaluation, mobility management,
system,’’ EURASIP J. Wireless Commun. Netw., vol. 2010, Dec. 2010, and heterogeneous networks.
Art. no. 750173.
[20] C. Fischione, G. Athanasiou, and F. Santucci, ‘‘Dynamic optimization
of generalized least squares handover algorithms,’’ IEEE Trans. Wireless ZDENEK BECVAR (M’08–SM’17) received the
Commun., vol. 13, no. 3, pp. 1235–1249, Mar. 2014. M.Sc. and Ph.D. degrees in telecommunica-
[21] S. Sesia, I. Toufik, and M. Baker, LTE-the UMTS Long Term Evolution tion engineering from Czech Technical Univer-
From Theory to Practice. Hoboken, NJ, USA: Wiley, 2009. sity (CTU) in Prague, Czech Republic, in 2005 and
[22] B. Davaasambuu, K. Yu, and T. Sato, ‘‘Self-optimization of handover 2010, respectively. He is currently an Associate
parameters for long term evolution with dual wireless mobile relay nodes,’’ Professor with the Department of Telecommu-
Future Internet, vol. 7, no. 2, pp. 196–213, Jun. 2015. nication Engineering, CTU in Prague, Czech
[23] D. Xenakis, N. Passas, L. Merakos, and C. Verikoukis, ‘‘Energy-efficient
Republic.
and interference-aware handover decision for the LTE-Advanced fem-
tocell network,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Jun. 2013,
From 2006 to 2007, he joined Sitronics R&D
pp. 2464–2468. Center, Prague, focusing on speech quality in VoIP.
[24] M. Ahmad, A. Saleem, K. Hesham, and N. Javaid, ‘‘A survey on fuzzy logic Furthermore, he was involved in research activities of Vodafone R&D Center,
applications in wireless and mobile communication for LTE networks,’’ CTU in Prague, in 2009. He was on internships with Budapest Politechnic,
in Proc. 10th Int. Conf. Complex, Intell., Softw. Intensive Syst., Jul. 2016, Hungary, in 2007, CEA-Leti, France, in 2013, and EURECOM, France,
pp. 76–82. in 2016. In 2013, he become representative of the Czech Technical Uni-
[25] A. M. Aibinu, A. J. Onumanyi, A. P. Adedigba, M. Ipinyomi, versity in Prague in ETSI and 3GPP standardization organizations. In 2015,
T. A. Folorunso, and M. J. E. Salami, ‘‘Development of hybrid artificial he founded 5Gmobile research laboratory at CTU in Prague, focusing on
intelligent based handover decision algorithm,’’ Int. J. Eng. Sci. Technol., research towards 5G mobile networks and beyond.
vol. 20, no. 2, pp. 381–390, Apr. 2017. He has published four book chapters and over 60 conference or journal
[26] Y. S. Hussein, B. M. Ali, M. F. A. Rasid, A. Sali, and A. M. Mansoor, papers. He is involved in the development of solutions for future mobile
‘‘A novel cell-selection optimization handover for long-term evolution networks (5G and beyond) with special focus on optimization of mobil-
(LTE) macrocellusing fuzzy TOPSIS,’’ Comput. Commun., vol. 73, ity and radio resource management, device-to-device communication, self-
pp. 22–33, Jan. 2016. optimization, power control, architecture of radio access network, and small
[27] Technical Specifiction Group Radio Acess Network; Mobility Enhance-
cells. He is a member of more than 15 program committees at international
ments in Heterogeneous Networks (Release 11), document TR 36.839
V11.1.0, 3rd Generation Partnership Project, Dec. 2012.
conferences or workshops.
[28] L. A. Zadeh, ‘‘Fuzzy Sets,’’ Inf. Control, vol. 8, no. 3, pp. 338–353, 1965.
[29] J. Jantzen, Foundations of Fuzzy Control: A Practical Approach. Hoboken, CARLOS RENATO LISBOA FRANCES received
NJ, USA: Wiley, 2013. the B.S. degree in computer science from the
[30] W. Pedrycz, ‘‘Why triangular membership functions?’’ Fuzzy Sets Syst., Federal University of Pará (UFPA), in 1995,
vol. 64, no. 1, pp. 21–30, 1994. the M.S. and Ph.D. degrees from the University of
[31] T. Ross, Fuzzy Logic With Engineering Applications. Hoboken, NJ, USA:
São Paulo, Brazil, in 1998 and 2001, respec-
Wiley, 2010.
[32] Y. Bai, H. Zhuang, and D. Wang, Advanced Fuzzy Logic Technologies in
tively. He held a Postdoctoral position with the
Industrial Applications. London, U.K.: Springer-Verlag, 2006. Institute for Systems and Computer Engineering,
[33] Q. Shen, J. Liu, Z. Huang, X. Gan, Z. Zhang, and D. Chen, ‘‘Adaptive Technology and Science, Porto, Portugal, in 2012.
double thresholds handover mechanism in small cell LTE-A network,’’ in He is currently an Associate Professor with the
Proc. IEEE WCSP, Oct. 2014, pp. 1–6. Telecommunication and Computer Engineering
[34] Technical Specification Group Radio Access Network; Evolved Universal Department, UFPA. His research interests include performance evaluation,
Terrestrial Radio Access (E-UTRA); Requirements for Support of Radio optimization, Markov processes, and artificial intelligence models.
Resource Management, document TS 36.133, v.14.4.0, 3GPP, Aug. 2017.

VOLUME 6, 2018 17189