Autonomic Handover Management for Heterogeneous Networks in a Future Internet Context: A Survey

Abstract

Future Internet (FI) scenarios envisage ubiquitous broadband coverage and seamless mobility, enabling the availability of personalized network connectivity at all places and with a high quality of service (QoS). In such a heterogeneous and complex environment, Vertical Handover (VHO) management is key for achieving the connectivity objectives, but also presents several challenges. In surveying this topic, the paper takes a direction that extends the state-of-the-art: By incorporating concepts of Autonomic Network Management (ANM), in particular the self-management and cognitive functionalities therein, to VHO management, the paper sheds new light to VHO operations from an ANM point of view, encompassing FI environments and the emerging 5G networks. In doing so, the survey identifies the main concepts and provides a taxonomy of relevant architectural components. Based on this taxonomy, a number of important autonomic features are identified, each one promoting the system’s self-optimization along a certain direction towards the overall enhancement of the VHO operations. Another contribution of the paper is the consideration of robustness in VHO decision making (i.e., achieving stable decisions under uncertainty) and a discussion of the relation between robustness factors and the autonomic features previously introduced. The general concepts developed in the paper are applied to representative state-of-the-art handover management solutions with autonomic characteristics. These specific solutions are presented, analyzed and correlated according to the proposed taxonomy and criteria, culminating to conclusions that provide useful insights towards future, further enhanced solutions.

Full text

Autonomic Handover Management for Heterogeneous
Networks in a Future Internet Context: A Survey
Adamantia Stamou, Student Member, IEEE, Nikos Dimitriou, Senior Member, IEEE,
Kimon Kontovasilis, and Symeon Papavassiliou, Senior Member, IEEE
Abstract— Future Internet (FI) scenarios envisage ubiquitous
broadband coverage and seamless mobility, enabling the
availability of personalized network connectivity at all places and
with a high quality of service (QoS). In such a heterogeneous and
complex environment, Vertical Handover (VHO) management is
key for achieving the connectivity objectives, but also presents
several challenges. In surveying this topic, the paper takes a
direction that extends the state-of-the-art: By incorporating
concepts of Autonomic Network Management (ANM), in
particular the self-management and cognitive functionalities
therein, to VHO management, the paper sheds new light to VHO
operations from an ANM point of view, encompassing FI
environments and the emerging 5G networks. In doing so, the
survey identifies the main concepts and provides a taxonomy of
relevant architectural components. Based on this taxonomy, a
number of important autonomic features are identified, each one
promoting the system's self-optimization along a certain
direction towards the overall enhancement of the VHO
operations. Another contribution of the paper is the
consideration of robustness in VHO decision making (i.e.,
achieving stable decisions under uncertainty) and a discussion of
the relation between robustness factors and the autonomic
features previously introduced. The general concepts developed
in the paper are applied to representative state-of-the-art
handover management solutions with autonomic characteristics.
These specific solutions are presented, analyzed and correlated
according to the proposed taxonomy and criteria, culminating to
conclusions that provide useful insights towards future, further
enhanced solutions.
Index Terms—Vertical Handover, Mobility Management,
Future Internet, 5G, Context Awareness, Self-management, Self-
optimization, Autonomic Network Management, Cognition,
Robustness.
I. INTRODUCTION
The provision of ubiquitous broadband network access for
mobile users has been a key research issue for a number of
years. Many of the relevant technological advances align with
the notion of a FI environment, consisting of open, intelligent
and collaborative wireless and wire-line access networks [1].
Within the FI ecosystem, the Fifth Generation (5G) networks
are already underway. These exploit higher frequency bands
with wider available bandwidths and consider extreme base
station and device densities, forming heterogeneous cellular
topologies where macro cells coexist with small cells (such as
femto and pico cells) utilizing multiple radio access
technologies [2]. At the same time, these networks target the
lowest possible energy consumption and end-to-end latency
[3]. In parallel, the FI vision includes the Internet of Things
[4], regarding the management of information about real
world objects and their surroundings, provided by an
enormous number of sensors, wireless communications
devices and embedded systems operating in different
environments and providing a number of different services
[5], [6].
In such a heterogeneous and complex networking
ecosystem, users should be able to have contextualized,
proactive and personalized access to services everywhere,
under a seamless experience, extending the „always best
connected‟ (ABC) notion. Therefore, it becomes essential to
take a unified approach that integrates all diverse networking
technologies available [7], towards enabling seamless
roaming between networks, while accessing applications with
different service requirements, and towards providing
enhanced QoS and user satisfaction. Collection, modeling,
reasoning, and distribution of context in relation to sensor
data would play a critical role in this challenge [8]. Hence, a
context-aware vertical handover management framework is
needed, in order to choose optimally the appropriate time to
initiate the handover and the most suitable access network for
each specific service, to ensure service continuity and
robustness against link and network impairments.
In this direction, the ideal answer and at the same time, the
key, to ensure seamless mobility in a complex heterogeneous
FI environment is provided by the vision of ANM,
encompassing self-management and cognitive functionalities.
ANM addresses the ability of networks to be aware of
themselves and their environment and self-govern their
behavior to achieve specific goals [9], without compromising
the performance of the other coexisting networks or the global
network performance metrics. ANM shares motivation and
has confluent goals with other emerging technologies, such as
Software Defined Networks (SDN) and Network Function
Virtualization (NFV), as all three concepts seek to increase
the flexibility, reliability and efficiency of operations and
optimize network management and control. As it has been
recognized, the notions of ANM, SDN and NFV can coexist
[10], [11], [12]. In particular, ANM could be used to promote
the local optimum in balance with the global optimum and the
self-awareness of each distributed entity could be used to
build the global awareness, enabling the development of
appropriate global policies used to optimize the operation of
the whole network.
A. Stamou is with NETwork Management and Optimal DEsign Laboratory,
National Technical University of Athens, Greece, and N.C.S.R. “Demokritos”,
Institute of Informatics and Telecommunications, Agia Paraskevi, Athens,
Greece (e-mail: stamouad@mail.ntua.gr).
N. Dimitriou and K. Kontovasilis are with N.C.S.R. “Demokritos”, Institute
of Informatics and Telecommunications, Agia Paraskevi, Athens, Greece
(email: {nikodim, kkont}@iit.demokritos.gr).
S. Papavassiliou is with NETwork Management and Optimal DEsign
Laboratory, National Technical University of Athens, Greece (e-mail:
papavass@mail.ntua.gr).

TABLE I
LIST OF ACRONYMS
Symbol
Description
ANDSF
Access Network Discovery and Selection Function
APAV
Access Point Acceptance Value
APSV
Access Point Satisfaction Value
ABC
Always Best Connected
ABS
Always-Best-Satisfying
AHP
Analytic Hierarchy Process
AHM
Autonomic Handover Manager
ANM
Autonomic Network Management
BER
Bit Error Rate
CNAPT
Client Network Address and Port Translator
CPN
Conventional Parameter Normalization
DF
Decision Function
DM
Deterministic Markovian
EPC
Evolved Packet Core
5G
Fifth Generation
FSA
Finite State Automata
TFFST
Finite State Transducer with Tautness Functions and
Identities
FI
Future Internet
FL
Fuzzy Logic
GRA
Grey Relational Analysis
IS
Information Server
MDP
Markov Decision Process
MIIS
Media Independent Information Service
MAC
Medium Access Control
MT
Mobile Terminal
MAD
Multiple Attribute Decision
NFV
Network Function Virtualization
NN
Neural Networks
OSS
Operations and Support Systems
PB
Policy-Based
QoS
Quality of Service
RSS
Received Signal Strength
RTT
Round-Trip-Time
SNAPT
Server Network Address and Port Translator
SINR
Signal to Noise plus Interference Ratio
SAW
Simple Additive Weighting
SDN
Software Defined Networks
TOPSIS
Techniques for Order Preferences by Similarity to Ideal
Solution
VHO
Vertical Handover
A number of earlier studies (including [13], [14], [15], [9],
[16], [17] and [18], among others) have surveyed issues
related to ANM in general, but without specializing on VHO
management, while other studies (e.g., [19], [20], [21], [22],
[7], [23]) have focused only on general aspects of VHO
management. Additionally, publications [24] and [25] have
surveyed several purposed VHO management solutions
featuring some degree of intelligence or a cognition potential.
Despite such prior works, however, to the best of our
knowledge there is a lack of a survey that focuses particularly
on autonomic VHO management in the FI era, defining the
subject and providing a comprehensive analysis.
Towards filling this gap, here we combine the concepts of
ANM and the notion of VHO management, introducing the
concept of autonomic VHO management framework in the era
of FI. We start by reviewing in Section II basic concepts
regarding cognition and autonomicity. Subsequently, in
Section III we employ these concepts in an analysis of the
autonomic handover management procedures under the light
of the autonomic functions monitor, analyze, plan, and
execute, providing an overview of the involved sub-processes
and corresponding algorithms. We introduce a new taxonomy
of the relevant architectural components, considering the
scenario of context-aware mobile terminals (MTs) that
operate within a complex FI environment and self-manage
their mobility behavior.
Section IV builds upon the aforementioned taxonomy,
towards addressing another issue of paramount importance for
autonomic systems, namely self-optimization. A number of
important autonomic features related to autonomic handover
management are discussed, each one promoting the system's
self-optimization along a certain direction, towards the overall
enhancement of the VHO operations. In the same section, and
still in connection with the autonomic features mentioned, we
also investigate robustness issues associated with the VHO
parameters and metrics of the network selection decision
function. Such considerations relate to the ability of a system
to achieve stable decisions under conditions of partial and
possibly imprecise knowledge of contextual information, still
a largely open issue in the present state of the art on network
selection frameworks [25], [7].
The general concepts developed in the paper are applied to
representative state-of-the-art handover management solutions
with autonomic characteristics: Specifically, Section V
reviews key characteristics of selected such solutions, while
Section VI provides a comparison of these solutions
according to the framework, autonomic features and
robustness considerations established earlier. Additionally,
Section VII provides useful insights towards future, further
enhanced solutions. Finally, Section VIII culminates with
conclusions. Table I provides a list of acronyms used
throughout the paper.
II. BASIC CONCEPTS CONCERNING COGNITION AND
AUTONOMICITY
Cognition is related to intelligence and has been employed
to enhance the effectiveness in network management
solutions. The cognitive network concept is described in [26],
as encompassing networks that can perceive current network
conditions, plan, decide, act on those conditions, learn from
the consequences of these actions and follow end-to-end
goals. This feedback loop implements a learning model, in
which past interactions with the environment guide current
and future interactions, resulting in intelligence
enhancements. Furthermore, in [27] it is claimed that
cognition is mostly related to the inference plane, being
driven by sensors, related to network planning and
optimization and being differentiated from “involuntary
functions” related to the management plane and configuration
management, which is being driven by the “effectors”. In
other words, this second approach differentiates cognition
from network management execution.
With respect to ANM, the ultimate aim is to create self-
managed networks to overcome the rapidly growing
complexity of networks. In 2001, IBM presented the
autonomic computing framework, describing a system with
„self-x‟ properties, such as self-management, self-
configuration, self-optimization, and self-protection [28].
Essential characteristics of an autonomic computing system
include the capabilities to perceive its state and the state of its
environment, to react accordingly to specific stimuli and to
optimize its performance based on the reported status and
stimuli. It is noted that autonomicity is frequently discussed
by means of drawing analogies to biological entities, such as
the human autonomic nervous system, so the relevant
terminology can be metaphorically related to functional
and/or structural aspects of a living organism [29], [30].
Correspondingly, the vision towards autonomic networking
includes the following four closed control loops [31]: sensing

(or monitoring) changes in the network and its environment;
analyzing changes to achieve the goals; planning
reconfiguration if goals cannot be achieved; and executing
those changes and observing the results. The operation of the
control loops is enhanced by adding learning and reasoning
processes, as well as by employing a well-structured
knowledge base.
ANM enables the system to evolve and to adapt to changes,
in terms of either business objectives or users‟ requirements.
For this reason, ANM introduces rules to formalize the
description of operations of various network elements in
response to changes in the environment [15]. These rules are
generally implemented by policies, defined (initially) by
network administrators, guiding the behavior of network
components. A typical advantage of policy-based network
management systems is their ability to reconfigure and adapt
their behavior by modifying the applied policies at runtime,
without suspending system operation.
Policies at the lowest level are typically defined by the
Event-Condition-Action triplet, which specifies the actions
that have to be taken in response to predefined conditions,
triggered by events. At the next level, goal policies are
defined, which describe the goal-conditions that should be
met, e.g. „„Response time not greater than 2 sec” [32]. At an
even higher level, some efforts have been dedicated to model
system control using utility function based policies. Utility
functions provide a natural and advantageous framework for
achieving self-optimization in a dynamic, heterogeneous
environment [33], [34]. Given a utility function, the system
must use an appropriate optimization technique to determine
the most valuable feasible state by tuning system parameters
or reallocating resources, considering also aspects such as
cost [35], [36], [37]. Additionally, utility functions allow
degrees of flexibility in selecting different levels of QoS,
matching the needs of different applications or user classes
[38], [39].
By putting the various approaches just mentioned together,
policies can be organized according to their purpose, forming
a hierarchy, as depicted in Fig. 1. Significant autonomic
network management architectural frameworks are based on
policies, such as Autonomia [40], DRAMA [41], Unity [42],
ACCORD [43], CA-MANET [44], AutoI [45], ANA [46],
and FOCALE [31].
In recapitulation, it can be stated that the concepts of
cognitive networks and autonomic network management
respond to almost the same expectations. Two differentiating
factors on these definitions may be considered: the extent to
which intelligence can be considered an axiomatic property
for autonomicity; and the consideration of including network
management execution among the cognitive networking tasks
[13]. Based on IBM‟s definition for autonomic computing, the
basic self-x properties include awareness, adaptivity,
proactivity and optimization, thus it can be argued that a
system does not have to be intelligent to implement
autonomic features, although intelligence can advance its
overall degree of autonomicity. On the other hand, in recent
research papers concerning autonomic network management,
learning and intelligence are being considered as fundamental
dimensions of autonomic systems [15], [14], [16], [9]. In our
point of view, considering a holistic autonomic approach,
cognitive functions shall be considered as part of the
autonomic network management framework of a FI system.
Fig. 1. Policy-based network management hierarchy
III. VHO MANAGEMENT IN LIGHT OF ANM
VHO management is a considerably more complex process
than the management of horizontal handovers enabling user
mobility in a single radio access network. Furthermore, VHO
management in the era of 5G concerns user mobility among
multiple radio access technologies, multi-layer and even
multi-operator dense network scenarios, where the user may
have to perform multiple vertical handovers during the
connection-time to switch among different cellular layers (e.g.
macro-small cell) and/or radio interfaces (e.g. 4G, 5G, WiFi).
Due to this high degree of heterogeneity, interoperability
issues are posed [7]. Also, there are other aspects contributing
to the increased complexity, including the need for
accommodating application demands and user preferences
and for exploiting the capability of handling multiple active
network interfaces concurrently.
One way to address these challenges is by introducing
context aware MTs that self-manage their mobility patterns,
towards meeting QoS requirements and maximizing user
satisfaction. In particular, the support of connectivity
management between macrocells and femtocells dictates
migration from network-controlled to autonomous, self- and
environment-aware MTs, which can be founded on the use of
cooperative and cognitive radio strategies [47]. Such
functionality may assist in the neighbor cell list discovery and
the cell reselection. For such operations the serving cell
configures the MT to perform signal quality measurements to
acquire the system information of the new cell [47]. In
general, the context-aware MTs just mentioned, may be
assisted by further components of the handover management
architecture, higher up in the network hierarchy that
provide/enforce appropriate policies to the MTs, in order to
achieve global optimization goals (e.g., load balancing).
It is noted that current research directions considering the
increasingly denser and unplanned network layout, promote
context-aware strategies that are not necessarily autonomic.
For example, [48] proposes a strategy to minimize
unnecessary handovers, aimed at multi-tier cellular networks,
combining both user-location awareness and cell-size
awareness, while [49] develops a velocity-aware solution via
stochastic geometry, which resolves handover rate problem in
dense cellular networks. Further optimizations can be
utility function based policies:
Optimizing tradeoffs subject to
constraints expressed in high-level
business terms
goal oriented policies:
Describing goal-conditions to be met
e.g. "Response time <= 2 sec
low level policies:
Event -Condition - Action triplet
specifying actions in response to
predefined conditions triggered by
events

achieved by splitting the control plane and user plane, using
phantom cells. This has been proposed as a potential solution
to minimize network control overhead in 5G networks [50].
Such solutions may be enhanced by incorporating elements
from the autonomic networks that can enable distributed
context awareness, processing and decision making.
In line with the trends just discussed, this section discusses
handover management frameworks in light of ANM,
assuming context aware MTs self-managing their mobility
behavior (at least to a degree) according to policy-based
management principles. As part of this discussion, we
introduce a new taxonomy of the relevant architectural
components.
Fig. 2. Phases of the autonomic VHO management and associated
architectural components
In principle, the structure of the media independent
handover mechanisms can be taken as a basis for organizing
the discussion of relevant autonomic management features.
With respect to this structure, and according to established
VHO frameworks (such as the IEEE 802.21 [51] and 3GPP
Access Network Discovery and Selection Function (ANDSF)
[52]), the handover management procedure can be separated
into three phases: handover initiation, which contains network
discovery, network selection and handover negotiation,
followed by handover preparation which contains layer 2
connectivity and IP connectivity, and then complemented by
handover execution, which includes handover signaling,
context transfer and packet reception. However, here we
organize the relevant operations in a slightly different manner
that enables us to highlight the autonomic character/elements
of the handover management procedure. A similar
organization has been followed in [19].
Specifically, the operations are grouped into the phases of
information collection, being linked to a knowledge base,
followed by the handover decision making, which includes
handover initiation and network selection processes and its
corresponding algorithms, itself followed by handover
execution1, which includes handover preparation and
signaling. In the context of these redefined phases, the
handover management complies with the autonomic
1 The term 'handover execution' here refers to the Monitor-Analyze-Plan-
Execute loop in ANM, instead of the actual handover execution phase
within the standards' based handover management procedure mentioned
earlier.
management principles of monitor (information collection),
analyze & plan (handover decision making), and execute
(handover execution) functions.
Fig. 2 illustrates the interaction of these phases and their
connection with key architectural components. The alternative
grouping of phases just presented reflects better the
autonomic control loops involved, while remaining fully
aligned with the aforementioned, more conventional,
grouping (i.e. handover initiation, preparation and execution),
in the sense that all individual actions are included in both
groupings.
Each of the phases (and the knowledge base employed by
the information collection) will be further discussed in the
following subordinate Sections III-A to III-D. As a summary,
Fig. 3 depicts relevant attributes discussed in these sections.
For comprehensiveness, Section III-E reviews the standard
media independent handover management frameworks and
correlates them with the autonomic framework in the earlier
sections. Relevant amendments and/or extensions to the
standards from the literature are also reviewed therein,
together with works addressing efficiency, performance
evaluation and related modeling aspects.
A. Information Collection
The information collection process gathers the required
user, terminal and network context, in order to provide to the
MT self- and environment-awareness. This process is critical,
as it constantly provides appropriate information to the
„analyze & plan‟ functions of the handover decision making
phase, indicating the need for a handover initiation and
assisting in the network selection. According to [20] and [53],
contextual information encompasses user context, including
user-related information such as preferences, priorities and
profiles history. Another type of context is terminal-related
information, such as power status, physical mobility
parameters (e.g., distance, location), Received Signal Strength
(RSS) and Signal to Noise plus Interference Ratio (SINR)
measurements, as well as information related to running
applications (e.g., QoS requirements).
Furthermore, network context may be included, providing
indicators of the quality and the availability of resources of
neighboring networks (through metrics such as bandwidth or
throughput), or provider context (e.g., cost, security
management, etc). Finally, another important type of context
relates to handover performance, including parameters such as
handoff latency, decision latency, execution latency,
degradation rate, and improvement rate.
Concerning autonomic handover management solutions,
user-related context plays a significant role, permitting the
maximization of the user satisfaction by taking into account
the user preferences. More precisely, the handover decision
making module uses the collected information to evaluate the
available access networks and to select the most capable
network, satisfying at the same time the user‟s request at a
particular time (e.g., “maximize throughput, but also
minimize monetary cost”), referred as Always-Best-Satisfying
(ABS) network [54], [31].
This more elaborate consideration of user preferences
refines the concept of an Always Best Connected (ABC)
device. Therefore, it is important for the information
collection module to maintain user profiles, in order to be able
to accumulate the user-related context.

Fig. 3. Key attributes/properties of the autonomic VHO management phases
It is noted that the volume of collected data must be post
processed and/or converted in a form suitable for later use. In
particular, raw measurement data should be converted to a
common format, understandable by subsequent decision
making processes. Also, to avoid overloading the information
collection and knowledge base components with raw
measurement data, filtering is needed [55]. Consequently,
there is a trade-off between precision and measurement load
[56].
In autonomic handover management, information collection
may be characterized as active or passive. In the active case,
the MT itself can initiate data collection periodically,
including the issuance of testing messages. By contrast, in
passive information collection status capturing is initiated and
(more generally) coordinated by components at the network
side [57].
In order to make the information collection process even
more effective, global statistics of network-wide scope could
be collected and analyzed, about neighboring MTs and their
experiences with the different access networks available in the
area [24], [9]. The global statistics gathering/analysis may
potentially employ cloud computing services and/or big data
analysis [58]. Since the network-wide view is built by
sampling local views from various MTs within the network, a
particular MT can utilize the global view to compare against
its own status and to potentially self-adjust. For example, the
global knowledge could provide hints to MTs for generating
dynamically optimized policy parameters. More generally,
information about neighboring MTs can be exploited in the
decision-making process (e.g., to determine the right time to
initiate a VHO) and to identify the best course of action with
respect to, e.g., QoS or energy efficiency.
In the sense just mentioned, global statistics lead to an
advanced awareness that could promote the adaptation and
learning processes, leading to optimal handover decisions and
ultimately improving the QoS for the end users. However, it is
noted that collecting and maintaining network-wide global
statistics may require more computational and memory
resources and may lead to increased power consumption and
signaling overhead. Therefore, the information collection
should strive for balancing the trade-off between the extra
overhead and the more comprehensive network view.
B. Knowledge Base
The knowledge base stores user, terminal and network
context received from the information collection module,
making this information available and accessible to other
autonomic handover management entities that require it,
such as the handover decision making functions (as shown
in Fig. 2), contributing to the cognition loop. Based on [56],
we classify the components of the knowledge base, into
four logical groups, depicted at the top of Fig. 2 and further
discussed in the following.
To begin with, the Context Server stores current terminal,
application and network context, logically divided in two
parts: the Service Information Base and the Resource
Information Base. The first part contains information about
the service instances activated by customers, such as parties
involved (customers and service providers), rules regulating
the service delivery, types of resources needed, amount of
each resource type needed in each occasion, billing plan for
the service, and operation history. The second part
maintains an up-to-date account of the type and quantity of
currently available resources.
The User Profiles Repository contains information related
to the users of the mobile device, such as user preferences,
user history, list of the subscribed services, and updated
billing information. It is noted that, while the User Profiles
Repository could in principle be regarded as a part of the
Context Server, we keep it separate, in order to emphasize
its importance in autonomic handover management.
The third logical group of the knowledge base, namely
the Policy Repository, contains information related to
policies, for use in the handover decision making [52].
Complementarily, the History Repository logs information
about previous handover decisions, such as parameters
employed, cause that triggered the handover, time of
occurrence, parties involved, target network selection, and
effect of this selection. Using this log, the current situation
may be correlated with previous comparable ones, so that
decisions can be made faster, saving time and
computational power.
Depending on the implementation, a knowledge base may
be classified as centralized, when the entire knowledge base is
a single central entity residing at the network side, or
Autonomic
VHO
Management
Information
Collection
Active / Passive
Global / Local view
Knowledge
Base
Memory Strength
Centralized /
Distributed
Handover
Decision
Making
Parameter Selection
Levels of Abstraction
Static / Dynamic
Context Info
Parameter Processing Output Format
Handover
Execution
Management
Architecture
Handover Control

distributed, when the knowledge resides at various places,
mostly at the edge of the network, or even at individual MTs.
The cloud computing concept is conformal with the
centralized knowledge base paradigm, offering centralized
data storage and processing through remotely deployed server
farms and software networks [59]. However, the traditional
centralized cloud computing architecture may fall short in
meeting the strict latency requirements for mobility
management in a 5G network environment. Edge, mobile
edge, mobile cloud and fog computing concepts, which use
computing resources and storage at the edge of a network
[60], can be used as alternatives related to the distributed
knowledge base paradigm, potentially offering a higher delay
efficiency. Such distributed knowledge base paradigms could
facilitate the MTs to store individually essential information
about their mobility for later use [19], further promoting the
concept of self-management.
Furthermore, another attribute of the knowledge base,
related to the history repository, is memory strength [9],
referring to the ability of the system to remember past
behaviors, significant events, corresponding reactions and
results, towards assisting the system in its current and future
management decisions.
C. Handover Decision Making
The handover decision making can be considered as the
core phase of the VHO, since it is in charge of analyzing the
context collected by the information collection phase and
planning the actions to determine the best handover target
[20]. This phase includes handover initiation and network
selection processes and the corresponding algorithms. In the
autonomic context of interest here, the handover decision
making also includes cognitive self-learning mechanisms that
enable the system to meet the forthcoming needs, promoting
self-optimization and self-healing. In reflection of this fact,
the handover decision making phase can be organized into
two distinct steps: the parameter selection and the parameter
processing.
The parameter selection exploits the context gathered in the
information collection phase, towards selecting suitable
parameters from a given set/pool (determined by the user or
by a policy in effect). The selected parameters are fed as input
to the parameter processing algorithms, essentially
determining the criteria for the decision making therein. The
versatility of the parameter selection is characterized by two
attributes: context time variability and levels of abstraction
[24]. The context time variability expresses the potential for
including in the selected parameters set both static and
dynamic context. The levels of abstraction refer to the
capability of the autonomic system of jointly treating multi-
layer context in uniform, abstract terms and thus the
capability to make parameter selections spanning several
layers among the physical, link, network, transport and
application ones [20], [61].
Since, as already mentioned, the parameter processing
receives parameters selected on the basis of gathered context,
the decision making therein becomes context aware. In
particular, the context encapsulated in the selected parameters
drives the parameter processing algorithms, towards making
optimal decisions (with respect to multiple criteria). With
respect to the algorithms themselves, and considering the
current state-of-the-art of context-aware VHO decision
making solutions, parameter processing methods can be
classified into the following four distinct approaches: a) the
decision function (DF) approach (including simple DFs and
Multiple Attribute Decision strategies (ΜΑD)); b) the Markov
decision process (MDP) approach; c) the policy-based (PB)
approach (including Finite State Automata (FSA)); and d)
approaches based on fuzzy logic (FL) or neural networks
(NN). Each of these approaches is discussed further in the
following.
DF strategies use the selected parameters to calculate the
values of specific decision functions that assess the merit of
individual alternative actions. The decision simply selects the
action with optimal merit. In this sense, DFs can be regarded
also as award, cost or objective functions. For specific related
applications of the concept, see [62], [63], [64], [65]. The
prime advantage of this approach is simplicity. In particular,
for cases involving only a small number of parameters,
network selection may employ a simple DF evaluating the
weighted sum of values derived from the selected parameters
(repeatedly, for each network in the service area of a user).
A more sophisticated distinct sub-family of decision
function-based methods involves ΜΑD strategies. These
combine and evaluate multiple decision criteria
simultaneously, dealing efficiently with complex problems,
and providing high flexibility [19], [66]. MAD strategies can
be classified into several groups, including:
 the Simple Additive Weighting (SAW) [67], involving
a larger number of parameters than the simple DFs,
where the score of a particular network is determined
by the weighted sum of all the attribute values;
 the Techniques for Order Preferences by Similarity to
Ideal Solution (TOPSIS) [67], where the preferred
network is the one closest to the ideal solution and
farthest from the worst case solution;
 the Grey Relational Analysis (GRA) [68], which ranks
the candidate networks and selects the one with the
highest ranking; and
 the Analytic Hierarchy Process (AHP) [69], which
decomposes the network selection problem into several
sub-problems and assigns a weight value for each sub-
problem.
According to [70], the advantage of AHP solutions is their
strong robustness for solving problems with complex
hierarchical structure. On the other hand, considering
problems with relatively simple hierarchy, SAW is less
complex and thus preferred. In [71], AHP is used to determine
weights to the selected parameters (bandwidth, delay, jitter,
and Bit Error Rate (BER)), applied to several MAD
algorithms, including SAW, TOPSIS, and GRA, while a
performance comparison between them is performed. Results
show that SAW, and TOPSIS provide similar performance for
conversational streaming and interactive traffic classes,
whereas GRA provides a slightly higher bandwidth and lower
delay for the interactive traffic class.
Another parameter processing approach involves MDPs.
The handover problem under consideration is formulated with
the objective of determining the action that maximizes the
total expected reward per connection [72], [73]. To this end,
Deterministic Markovian (DM) decision rules are employed.
These are functions that specify the action choice when the
system occupies a particular state at a specified decision
epoch. Transitions from state to state are governed by a
Markov chain, which captures memory effects. The state
information includes the current network status plus
availability of other networks in the area. The time between

transitions corresponds to the time between successive
decisions. To specify the MDP, one should calculate the
probability of transition from one state to another. The
transition probabilities can be estimated by the network
operator based on gathered statistics.
Several particular applications have been based on this
general framework. For example, in [74] the transition
probabilities are assumed to depend on the suitability (rank)
of candidate networks in relation to each decision parameter
and on the weight of each such parameter. Analysis of this
model enables the determination of the optimal candidate
network [74] under a particular set of state conditions, while
the derived results can be exploited for future decisions. In
[75], the optimal decision rules are constructed by means of
AHP, combining the benefits of MDP and MAD approaches.
In [72] the calculation of the optimal decision is performed by
the operator offline and is periodically updated whenever
spare processing capacity is available at the network access
controller.
In general, the update frequency of the Markov chain
transition matrix and the flexibility and adaptability of the
decision parameters are crucial factors determining the
suitability of MDP for use in autonomic handover
management. Apart from the core handover management
functions, however, MDP techniques may also be used for
user (physical) mobility modeling with a Markov chain,
towards extracting the user‟s mobility patterns from a
historical mobility trace [75]. In this way the next possible
location of a user can be estimated, which will determine the
next possible network connection(s), optimizing handover
performance.
The third approach to parameter processing involves policy-
based decisions. In this case, network selection proceeds by
determining the most suitable network according to a specific
set of policies. For policy conflict resolution, FSA can be
employed, where policies can be represented as deterministic
transducers [55], used to resolve potential conflicts, both
static and dynamic, among the different policy rules. At a next
step, a decision function (Tautness Function (TF) [76]) is
formed, to indicate how tautly a condition fits to an event.
Subsequently, priorities are assigned to the conditions,
depending on their probability to occur.
The common drawback of all three parameter processing
approaches already reviewed is their inefficiency to handle a
decision problem that involves ambiguous decision criteria.
To remedy this deficiency, in specific scenarios FL or NN
could be used as an intermediate step. FL-based strategies
convert parameters into fuzzy sets. A set of fuzzy rules are
applied utilizing a series of branches roughly analogous to
ordinary IF-THEN clauses, producing a decision set (growing
or shrinking as successive rules are applied) that is
subsequently mapped into a single-valued quantity. Related
applications can be found in [19], [54], [77], [78], [79]. On
the other hand, NN are usually employed with only one
parameter and one type of handover policy (i.e., "keep WLAN
connection when it is available"). However, NN architectures
require training delay and prior knowledge of the radio
environment [19]. Related applications can be found in [80],
[81].
Finally, another important aspect of parameter processing
relates to the output format of the network selection,
indicating the target network candidate(s). For example, the
output may be a list of the candidate networks in prioritized
order, where the top of the list represents the one with the
highest significance/weighting factor according to the
predefined criteria [19]. Alternatively, the output format could
specify only one candidate network, selected by a policy-
based framework [55]. Moreover, when multiple active
interfaces are supported, the output could specify the
appropriate network interface for each application [54].
D. Handover Execution
The handover execution process implements the VHO
management and control [20]. In this survey, we are
concerned with the control methods and management
architectures considering state-of-the-art autonomic VHO
management solutions, assuming these are distributed enough
to enable the MT to make (at least some) decisions on its
own, promoting self-management. Accordingly, fully
centralized and/or fully network controlled management
approaches are not considered in the following, being out of
scope.
In particular, self-management is regarded as an essential
property of autonomic VHO management, giving to the MT
the ability to control its own context and enabling it to
determine the appropriate time to execute handovers [82].
Furthermore, self-management promotes adaptivity,
flexibility and self-optimization to the decisions of the MT
[9]. According to the most recent trends, distributed handover
management may provide a paradigm most congruous to the
need for handling effectively the complexity of the FI
environment and the emerging 5G networks, avoiding at the
same time a single point of failure (characteristic of the
classical centralized approaches, frequently together with high
latencies and signaling overhead) [58], [83].
In general, the autonomic handover process may be
characterized by the entity that is responsible of controlling it.
It is characterized as mobile controlled [20], [19], [84] when
the VHO initiation and decision is fully controlled by the
mobile device. This is a flexible solution that enhances user
satisfaction. The disadvantages are that the MT must possess
advanced computational capabilities, which also lead to
increased power consumption. Alternatively, the VHO may
be characterized as network assisted, if the handover
initiation is done by the mobile device, but the network
selection, or a part of it, is implemented by the network,
making use of the information services and undertaking the
heavy programming tasks [55], [54]. Finally, the VHO is
characterized as mobile assisted when it is initiated by the
network, but assisted by the mobile device [85].
Beyond the control-related characterization just discussed,
the structure of the management architecture is important,
since it affects the scalability, performance, intelligence and
overall autonomicity of the system [9], [57]. This structure
can be classified into three basic categories: flat, hierarchical
and hybrid.
The flat approach refers to fully self-managed MTs, where
autonomic handover managers (AHMs) are assumed to reside
only in the intelligent MTs (as depicted in Figure 4). This
distributed type of management architecture addresses the
limitations of centralized management with respect to fault-
tolerance and scalability, advancing autonomicity. However,
this approach raises challenges in the domain of distributed
information management, system-wide coordination, security,
and resource provider‟s policy heterogeneity. It may also put
on MTs excessive requirements in terms of computational
capabilities and power consumption. Examples of flat

autonomic handover management approaches are found in
[19], [84].
In the hierarchical category, a main AHM supervises a set
of multiple lower-level AHMs. Thus, a coordination
management overlay should be defined, to arrange the
operation of lower-level autonomic managers. Considering
hierarchical architectures, intelligence resides in both the
terminal and network sides, avoiding excessive complexity at
the MT. In general, hierarchical architectures can be
centralized hierarchical or distributed hierarchical [9],
depending on whether the main AHM resides in the network
side or in the MT side. Hierarchical autonomic architectures
consider a distributed manager level (see Figure 4) [9]. A
significant advantage of using distributed hierarchical
architecture is that MTs present a higher degree of self-
management and thus they can support autonomic handovers
more efficiently. Also, more personalized management
policies can be deployed at the autonomic manager of each
MT. Examples of such approaches are [55], [62] and [54].
More specifically, according to [54] and [62], the main
AHM resides in the MT. However, some functionalities are
placed also to the operations and support system (OSS),
where network monitoring is performed. Furthermore, a
context server that resides in the core network collects the
relevant contextual information from the various repositories
and assists in the handover decision, in response to requests
from the MT. Similarly, according to [55], those components
that involve operator‟s management or high computational
cost are located in the core network to minimize the
complexity of the MT. Such tasks include policy definition,
storage, and conflict resolution. These network-side
components assist in handover decision, always with the
coordination of the MT.
Fig. 4. A general architectural model for VHOs with multiple levels of
AHMs
In addition to the two categories already discussed, there are
various hybrid architectures combining the previously
mentioned concepts to a varying degree. In autonomic hybrid
architectures, some self-organization and self-optimization
algorithms, mostly those related to tasks with local scope, are
running locally on the MT, while the tasks with wider scope
(global network view) are being managed by a central
managing authority on the network side (usually at the base
station or in a cluster of base stations, as shown in Figure 4).
It is noted that while the distribution of functionality just
discussed is at present considered to be fixed, future
autonomic VHO management architectures may have
dynamically adjusted structure, towards an increased potential
for customization [16]. In general, hybrid architectures
achieve load balancing and traffic management, hiding the
complexity from the MT.
Examples of hybrid architecture can be found in [86], where
autonomic MTs are assumed to cooperate with autonomic
base stations and access points in order to make optimal
handover decisions and QoS-aware resource management.
Also, [82] proposes a scheme for vertical handover decision
making that leverages the cooperation between the MTs and
a controller, which manages a cluster of different access
networks locally available. This controller is also responsible
for resource control and load balancing among the MTs.
E. Major Architectural Frameworks and Recent Trends
Working groups within major standardization bodies have
provided specifications enabling the unified management of
handovers in heterogeneous networking environments. One
major such specification is by the IEEE 802.21 standard [51],
(recently updated in IEEE 802.21.2017 [87] and IEEE
802.21.1.2017 [88]), which introduces the Media Independent
Information Service (MIIS) to provide a framework and the
related mechanisms for discovering and obtaining information
about networks within a geographical area. Contextual
information is collected from both the client (MT) and
network sides. Information provided by the MIIS includes
policies and static link layer parameters, such as channel
information, a list of available networks and their associated
operators, roaming agreements between different operators,
costs for using the network, etc. According to the standard,
the Information Server (IS – referred as the Context Server, in
the terminology of Section III-B) that stores the context and
policies is defined as a centralized entity.
In another major specification, the Access Network
Discovery and Selection Function (ANDSF), an optional
network element in the 3GPP Evolved Packet Core (EPC),
has the purpose to provide MTs with useful information and
operator-defined policies to guide network selection
decisions, enabling mobility between 3GPP and non-3GPP
systems (e.g., 802.11, 802.16, etc) ) [89]. Specifically, the
ANDSF provides a list of access networks accessible to the
MTs and exchanges discovery information and policies
according to operator requirements. With respect to the
consideration of dynamic information, however, the support
provided by the ANDSF server is also limited.
Following the taxonomy discussed earlier in Section III,
both aforementioned standards focus on the phases of
information collection (especially with respect to network-
related information) and of handover execution. The handover
decision making is not specified in the standards, thus any
VHO decision algorithm can be used. Considering handover
execution, IEEE 802.21 and ANDSF can support all VHO
control methods and kinds of management architecture
discussed in Section III-D, depending on the location of the
main managing entity (in the network or the MTs). In both
standards the MT receives a list of RANs reflecting operator
policies, but it is responsible to choose the appropriate RAN,
based on application demands and user preferences. Thus,
intelligence can potentially be found in both the network and
the MTs. Consequently, both standards can accommodate
aspects of autonomic VHO management.

To remedy the limited support for dynamic context in IEEE
802.21 and ANDSF, several works in the literature have
proposed amendments introducing more elaborate
information collection. For example, [90] proposes an
enhanced IS for this purpose. This IS, still a centralized
component, like its "conventional" counterpart, receives
regularly dynamic context updates (such as available
bandwidth) from all RANs in its area of interest and makes
the updated information available to querying MTs. New
signaling is required, to support the update messages from the
various RANs. The main shortcoming of the proposal,
however, is the centralized character of the enhanced IS that
may create scalability issues, especially when monitoring a
big number of cells, e.g., dense deployments of small cells.
Accordingly, other proposals extend the standards by
introducing distributed entities that collect dynamic context
proactively. Such distributed solutions are closer in spirit to
5G network scenarios and provide a better basis for
autonomic mobility management. Along this direction, [91]
proposes several extensions to the IEEE 802.21 framework,
introducing new update messages and state machines to
handle these messages. The proposed architecture employs a
centralized network information server for static network-side
context in conjunction with distributed RAN-associated
information repositories handling dynamic contextual
information (such as available bandwidth, network latency,
packet loss, etc.). In a similar spirit, the work in [92] proposes
a framework employing local instances of ANDSF (as
proposed in, e.g., [93], [94]) in combination with Hotspot 2.0
protocol [95], to enable status evaluation of WiFi access
points. A more disruptive approach is taken by [96], where
the centralized IS, as defined in the standards, is omitted
altogether. This work proposes a hierarchical knowledge base
with three layers of hierarchy, employing hash tree-based
information servers that do not store all data, but use
references instead to data located in other servers.
The performance of architectural amendments like those
just mentioned must be assessed, to evaluate the efficiency of
each proposal with respect to factors such as handover latency
or signaling overhead. Relevant quantitative assessments in
the literature can be organized in two groups: The first one
adopts a more elementary approach, determining a nominal
execution time for each step in the handover process and
summing the individual step times to calculate the overall
handover latency. This simple methodology is useful for
identifying which steps are more expensive than others, but is
somewhat oversimplified. Assessments in the second group
involve more elaborate modeling, to capture queueing,
congestion and other dynamic phenomena that may introduce
variations to the individual step times.
The simple methodology of the first group is used in [97] to
evaluate a framework for VHOs among collaborative wireless
networks, based on the principles of the IEEE 802.21 and
IEEE 1900.4 standards. The evaluations in [91] and [96] are
of a similar flavor, but make use of additional simulation-
based refinements to take into account MT (physical) mobility
and a more realistic representation of pertinent physical layer
characteristics. The evaluation methodologies in [90] and [98]
belong to the second group and employ queueing models to
track congestion when signaling messages are exchanged
between elements of the architectural frameworks under
evaluation.
IV. AUTONOMIC VHO FEATURES TOWARDS SELF-
OPTIMIZATION AND ROBUSTNESS ISSUES
We now turn to a number of important features, which
characterize autonomic handover management and jointly
lead to performance optimization. Section IV-A discusses the
nature and effect of each of these features, in correlation with
the taxonomy of Section III. Section IV-B deals with the issue
of robustness, also discussing how the autonomic features
may be exploited towards more robust handover decision
making.
A. Autonomic Features
Autonomic VHO management aims at performance
optimization, related to seamless mobility and user
satisfaction. The deployment of autonomic features to
automatically manage, optimize, and adapt the management
of operations can significantly improve the resulting
performance [35], [99]. More specifically, the combination of
awareness, adaptivity, flexibility and proactivity drive the
system to performance improvements and enable the system
to select the best choice among a set of available alternatives,
advancing the system‟s overall self-optimization, which can
be described as the objective of autonomicity. The
functionality of individual autonomic feature and the inter-
relations between them towards the optimization of VHO
management performance are further described in the
following. A summary is depicted in Fig 5.
1) Awareness
This is a fundamental property, present in most autonomic
functionalities. Awareness is primarily related to the monitor
function of the information collection phase and the
associated knowledge base and is most directly exploited in
the parameter selection step of handover decision making.
The term refers to self and environment awareness,
addressing information collection from the MT, the networks
and the user [35]. Awareness is expected to trigger a `prompt
reaction' associated with the handover execution, thus closing
the autonomic loop in Fig. 2.
Specifically in connection with VHO management, an
enhanced level of awareness is positively correlated with the
ability of the system to extract contextual information from
multiple layers (a notion linked to the levels of abstraction
discussed in Section III-C), enabling the consideration of the
QoS requirements of running applications. The level of
awareness is also related to the frequency of parameters
monitoring, which affects the precision of the selected
parameters used in the VHO decision making. Finally,
awareness also affects the capability of the system to support
an adjustable monitoring process. This is further discussed in
the following, in connection with adaptivity & flexibility.
As a concrete example, [54] demonstrated that a high
degree of awareness resulted in enhanced (by more than
20%), end user satisfaction metrics when compared against
other algorithms not considering user preferences.
2) Adaptivity & Flexibility
In the more general context of autonomic network
management, adaptivity deals with the ability of the network
to analyze changes indicated by current events (perceived due
to the system's awareness) and to decide why, when, where
and how a reaction should take place [9]. Thus, adaptivity

involves the „analyze‟ and „plan‟ components of the
autonomic loop. For example, adaptivity may trigger changes
to the frequency of measurements during the information
collection phase, and may promote adjustments to the
parameter selection and parameter processing methods,
according to environment changes and system needs [14],
[35]. Towards this direction it is noted that adaptivity could
be further enhanced by the use of biologically inspired
solutions. For example, swarm intelligence has been
employed in autonomic network management, addressing
load balancing and route construction and maintenance [14].
The achievable degree of adaptivity depends on the level of
flexibility [24] (characterized as limited or advanced), which
is related to the capability of modifying at runtime parameter
selection and processing methods for use in handover decision
making. The term „advanced flexibility‟ refers to the
capability of dynamically adjusting the set of said parameters,
potentially including newly identified parameters at run-time,
without requiring modifications to the implementation of
either the support system or the application logic. The term
„limited flexibility‟ characterizes approaches that are narrower
in scope and involve a predefined parameters‟ domain,
determined during system design.
Obviously, VHO solutions featuring advanced flexibility
equip the system with a greater ability to evolve, and thus
improve awareness and adaptivity, making the system capable
to adjust to a changing environment [100], [57], [9]. For
example, the adaptive approach used in [55], reduced
handover latency, resulting in close to seamless connectivity
on the move.
3) Learning
The learning functionality is part of cognition and equips
the system with the ability to remember past behaviors, or
problems and their solutions. This ability, in turn, helps the
system to gain experience (to an extent determined by the
memory strength of the knowledge base, as mentioned in
Section III-B) that may be utilized in the decision making, in
combination with adaptivity. The knowledge of behavioral
trends and occurrence patterns of conditions/scenarios is
valuable, especially in highly dynamic environments, as it can
dramatically enhance the system performance, by identifying
frequently repetitive patterns of actions and behaviors.
Towards this end, artificial intelligence techniques may be
employed, such as neural networks [15]. It has been shown in
[84] that the exploitation of historically available information
led to an improvement of about 50% in the mobile handset‟s
battery autonomy and to about 25% lower content
downloading times and network usage costs.
4) Proactivity
Proactivity signifies the use of preventive measures to
maintain a target level of system performance (by means of an
appropriately and timely initiated handover procedure), based
on the analysis of the current state and on the anticipation of
events and their effect on the system. Anticipation is a
cornerstone of proactive computing, promoting actions in the
direction of future prediction. Proactivity involves the
„analyze‟ and „plan‟ components of the autonomic loop,
utilizing data from the information collection process, which
is equipped with awareness. Proactive techniques focus on
context aware operation, statistical reasoning, and intelligent
data-handling [101]. By proactively collecting and analyzing
predicted information about e.g., the link status or the battery
status, the resulting VHO decisions can be optimized [21],
[102].
Therefore, proactive systems exploit context for responding
faster and more efficiently to specific stimuli, providing
further benefit if used in conjunction with learning
techniques. Statistical reasoning techniques such as Hidden
Markov Models, genetic algorithms, and Bayesian techniques,
can be used instead of traditional deterministic methods. For
example, [75] computed the user‟s mobility regularity from
the historical trace of the user using an MDP process, toward
providing estimations for the next possible location of the
user, subsequently exploited for making more robust VHO
decisions. Evaluation results in [75] showed that this
proactive strategy, used in conjunction with a multi-attribute
decision algorithm, achieved around 50% better performance
gains (in terms of throughput and latency) compared to a
baseline greedy strategy. Other proactive user location
estimation algorithms [48] and [49], resulted in minimization
of unnecessary handovers, providing throughput gains up to
47% and 70%, respectively. In general, proactive features in
autonomic network management promote network and
resource availability, service level agreement compliance, and
enhance user satisfaction.
Fig. 5. Autonomic VHO management features towards overall self-
optimization.
B. Robustness Issues
As already mentioned, VHO management in a FI
environment must cope with the heterogeneous, diverse and
dynamic character of the target setting and the need to jointly
consider many different sources of context. In such an
environment, robustness (generally defined as the ability to
achieve stable and efficient decisions [103]) becomes an
important attribute of the VHO decision making process. The
following subordinate Sections IV-B1 to IV-B4 identify a
number of robustness-related issues and review mechanisms
to overcome them. Subsequently, Section IV-B5 discusses
how the autonomic features of Section IV-A can contribute
towards enhancing robustness. A synopsis of the relevant
discussion appears in Table I, at the end of the section.
1) Diversity of Parameters (Context Diversity)
The joint consideration of multiple sources of context
creates the need for dealing concurrently with a diverse set of
parameters. This, in turn, requires a methodology to enforce a
uniform representation, so that different parameters-
characteristics (naturally involving different units) are
expressed through comparable values. The way to address this
issue depends on the type of the method used for the

parameter processing step of the handover decision making
(Section III-C). For parameter processing using DF (including
MAD) or MDP approaches, conventional parameter
normalization (CPN) techniques [104] are appropriate, while
FL-based parameter processing naturally resorts to techniques
employing fuzzification. Note that PB approaches do not
require a uniform representation methodology, as each
parameter is processed individually, through a relevant policy.
Accordingly, CPN techniques can be organized in two
categories [104]. The first one employs absolute
normalization, where each parameter‟s value is individually
scaled between 0 and 1, with respect to a given minimum and
maximum value [104]. For example, delay could be
normalized with respect to a given minimum value of 0, and a
maximum value of 100 ms. Examples of multi-criteria
applications, which incorporate scales that conform to
absolute normalization, can be found in [104]. The second
category employs relative normalization [105], [106], where
the scores corresponding to parameter values associated with
different options (i.e., the various candidate networks scores)
are summed up and scaled to 1. For example, if network‟s A
delay is 45 ms and network‟s B is 25 ms, then the network‟s
A normalized delay results to 0.36 and network‟s B to 0.64,
accordingly, as the bigger normalized score corresponds to
the better network. This is a more complex process, as all
parameters have to be rescaled whenever there is a change to
any candidate network‟s score. However, under relative
normalization methods, the final result is more distinctive
[105], [106].
With FL-based parameter processing, conversion of
absolute parameter values to relative ones comes as part of
FL's inherent capability for handling a decision problem that
involves ambiguous decision criteria [54], [107]. The
approach of FL is comprised of four steps [108]. The first step
is the fuzzification. For example, if the delay of a voice call is
25ms, through the membership function the delay is identified
as low or high [54]. The second step is the rule evaluation,
e.g., “if delay is low and jitter is low, then quality of voice
call is high”. The third step is the rule aggregation, where
every result is aggregated into one fuzzy set for each output
variable. The last step is the defuzzification, where the fuzzy
sets are converted into appropriate output values. For example
the output values can vary between “strong accept” and
“strong reject”, acquiring numerical values between 1 and 0,
respectively.
A special form of the parameter representation issue
emerges when considering the QoS requirements of different
applications. The QoS parameters should be treated
differently by each application, as each one has its own QoS
constraints [62]. For example, the jitter-related requirements
for a voice call differ from those of a streaming application.
Therefore, a different treatment for the jitter scores should be
used in each case [54]. When absolute normalization is used,
the appropriate upper and lower values should be identified
for each case. For relative normalization, application-specific
thresholds for the relative scores are needed, to ensure that
that the decision yields acceptable values for each criterion,
for each application. Similarly, for parameter processing
involving PB approaches, different policies should be
specified for each application. Finally, if FL is used, different
membership functions should be used for handling the same
QoS criterion in connection to different applications [54],
[107].
2) Diversity of Criteria/Rules
An effective handover decision making should be capable
of jointly employing multiple criteria/rules, and assigning
different importance to each of these criteria, towards
optimized decisions tailored to the environment. Specifically,
[109] demonstrated that that properly assigning importance to
criteria has a direct impact on the handover failure
probability.
FL-based decision making inherently lends itself to the joint
consideration of multiple criteria, through the definition of
parallel rules that may be applied simultaneously, to obtain
the desirable outcome [54], [110]. For example, the fuzzy rule
"If bit error rate is low AND burst error rate is low AND
packet loss ratio is low, then quality is Strong Accept" [54],
ensures that a candidate network would be strongly preferred
as the handover target when all three conditions are satisfied.
PB approaches can also handle groups of parameters
according to different criteria, through relevant policies.
However, in this case, the occurring conflicts have to be
resolved, as mentioned in Section III-C.
For other parameter processing methods, employing CPN
techniques, a different importance can be defined for each
individual criterion [109], while, AHP [69] may be used in
order to assign a different level of importance to each group
of criteria. Indeed, as mentioned in Section III-C, AHP
decomposes the decision problem into several sub-problems,
making use of hierarchy. Different groups of criteria may be
associated with different AHP sub-problems and their relative
importance may be tuned through the assignment of
corresponding weights [107]. For example, according to AHP,
the first tier of parameters could include cost and QoS,
associated with respective weights. The QoS could be further
analyzed into a set of second-tier parameters, such as
bandwidth, delay and jitter. The weights for the parameters in
the second tier could be adapted according to the demands of
each application and to user preferences.
It is worth mentioning that a number of VHO management
proposals use initially FL followed by CPN techniques to
employ AHP, (such as [54], [107]), combining the benefits of
both approaches.
3) Context Uncertainties & Incompleteness
Another set of robustness-related challenges arises in
connection with the ability of the decision system to cope
with uncertainties and incomplete information. Uncertainties
refer to the imprecise knowledge of context, particularly when
successive measurements for the values of some parameters
fluctuate beyond a level of tolerance. Incompleteness is
associated with missing information, including the lack of
information due to failures encountered during the
information collection phase of the VHO management
process.
One way to rectify the effects of uncertainties, particularly
considering performance-related measurements, is by
verifying the measured data against related data referring to
other layers [61], [111]. This general concept is consistent
with all methods for the uniform representation of parameters,
as discussed in Section IV-B1, including CPN and the
methods appropriate for PB- or FL-based parameter
processing algorithms.
The way to address incompleteness varies slightly,
depending on the uniform representation method in use. For
CPN or PB-relevant methods, an "average" value may be

substituted for the missing one. For FL-based parameter
processing, substituting a "neutral" value is the suitable course
of action [110]. These general principles for handling
uncertainties and incompleteness can be further enhanced by
making use of the memory strength available at the VHO
framework and of any available learning techniques, towards
exploiting historically available relevant data.
4) Marginal/Borderline Cases
Robustness is important also for coping with cases where
there are marginal differences among candidate networks that
may lead to unnecessary VHO decisions. This phenomenon is
frequently described as the „ping-pong‟ effect [19], referring
to repeated successive VHOs between the same two networks,
which eventually leads to QoS degradation. A related
phenomenon is the „corner effect‟ problem [112], where the
MT cannot assess correctly if a neighboring network is a
suitable VHO candidate, due to poor line-of-sight
communication. To remedy those marginal/borderline cases,
following either FL or CPN/PB techniques, a score margin
may be introduced marking a minimum difference on the
candidate networks‟ scores and a hysteresis (i.e., time) margin
to discourage very frequent VHO initiations [113]. The extent
to which the score and hysteresis margins are changed to
encourage or discourage a handoff depends on trends
indicated through the values of relevant parameters. For
example in [114], the authors used criteria such as the RSS-
based link quality and the distance between MT and base
station and made use of training algorithms, proving that
minimization of unnecessary handovers (approx. by 20%) can
be achieved, optimizing the resulting performance by 10-
20%, considering throughput, delay and packet loss.
5) Autonomic Features Addressing Robustness
Autonomic features could be exploited in various ways,
towards enhancing the robustness of the VHO decision
making. To begin with, awareness by definition aims at
untangling uncertainties [17], resulting in enhanced
robustness. Moreover, as already mentioned enhanced
awareness considers also the frequency of parameters
monitoring, which affects the precision of the parameter
values that provide the basis for the uniform representation
process discussed in Section IV-B1.
Along a similar line of reasoning, adaptivity and flexibility
are essential for allowing the dynamic modification of the
membership functions of FL systems, or the upper and lower
values used by CPN approaches for the uniform
representation of parameters, as well as, the score and
hysteresis margins used to avoid marginal/borderline cases.
The aforementioned features can be beneficially combined
with learning mechanisms (e.g., those based on neural
networks) enabling the exploitation of historically available
data and making use of memory strength, to optimize the
tuning of upper and lower values, membership functions,
and/or score and hysteresis margins and to help in combating
more effectively context uncertainties or incompleteness.
Moreover, proactive measurements enable the analysis of
the current state and the anticipation of events and their effect
on the system, which would assist in addressing
marginal/borderline cases. For example, estimations for the
next possible location of the user, would fine-tune the
hysteresis margin, preventing unnecessary VHOs resulting
from the ping-pong effect.
TABLE I
ROBUSTNESS ISSUES IN VHO DECISION MAKING AND THE RELATION WITH AUTONOMIC FEATURES
Robustness Issues
Awareness
Adaptivity &
Flexibility
Learning
Proactivity
Diversity of parameters
(Context diversity)
CPN: Different upper and lower values or
thresholds for each parameter for each
application.
Enables multi-
layer
parameter
selection and
defines the
precision of
the parameter
values.
Enable the
dynamic
modification of
upper and lower
values or
thresholds /
membership
functions /
score and
hysteresis
margins.
Optimizes the
tuning of upper
and lower values
or thresholds /
membership
functions / score
and hysteresis
margins and deals
with uncertainties
and
incompleteness.
Anticipates
upcoming events
confronting
marginal/borderline
cases.
PB: Different policy for each parameter
for each application.
FL: Different membership functions for
each parameter for each application.
Diversity of criteria/
rules:
CPN: Different importance for each
individual parameter, AHP.
PB: Policies‟ conflict resolution.
FL: Parallel fuzzy rules.
Context uncertainties
and incompleteness:
CPN/PB/FL: Verify the measured data
through comparison with related data
from other layer(s) to combat
uncertainties.
Substitute an average (CPN/PB) or
neutral (FL) value (or, a value derived
from historically available data) for the
missing criteria.
Marginal/borderline
cases:
CPN/PB/FL: adjust score and hysteresis
margins.

V. SELECTED VHO MANAGEMENT SOLUTIONS WITH
AUTONOMIC ORIENTATION
To demonstrate the applicability of the general concepts
previously discussed, we now review six representative VHO
management solutions with an autonomic orientation, taken
from the literature. All reviewed proposals possess some
context-awareness and cognitive characteristics, but differ in
terms of the management architecture, the scope of
information collection, the computational methods employed
and/or possibly other aspects. Three cases (those treated in the
subordinate Sections V-A, V-C and V-D), have been
reviewed again in a prior study [25] addressing always best
connectivity. Here, however, they are considered from a
different perspective, in line with the focus of this survey.
In the rest of this section we individually examine each
solution in turn, identifying relevant characteristics and
associating them with the classification and taxonomy of
Section III; a summary of the results appears in Table II at the
end of the section. Subsequently, in Section VI we compare
the six VHO management solutions with respect to a number
of criteria. For succinctness, future references to the reviewed
management solutions are through corresponding acronyms.
Two of the acronyms (for the solutions in Sections V-D and
V-F) have been borrowed from the publications proposing the
respective solutions.
A. Simple Terminal-Controlled Autonomic VHO Management
Approach (TCAM) [84]
TCAM enables a simple and light-weight VHO
management approach that does not require changes in the
network infrastructure. All the intelligence lies in the MT and
the handover is mobile-controlled, thus the type of the
management architecture (see Section III-D) is flat.
Information collection (Section III-A) is addressed by the
MT, which monitors the RSS and SINR over the available
radio interfaces, remaining energy level on the device's
battery, and the velocity of the MT's motion. Velocity
information is directly deduced from the Doppler spread in
the received signal envelope. User preferences are also
included, considering QoS, monetary cost and energy
efficiency, where the user asserts priority for each one.
The knowledge base (Section III-B) includes a user profiles
repository that contains the identities with which the user
accesses different radio networks and the respective
subscriptions to services, the user preferences and mobility
policies. Additionally, the MT maintains a mobility policy
database that contains a black list of access network operators
with whom the user has had a bad experience. This feature
enhances memory strength and enables learning. Both the
user profiles repository, and the mobility policy database
reside in the MT.
The handover decision process (Section III-C) employs a
set of parameters including both static and dynamic
contextual information, and thus presents context-time
variability. However, the parameters‟ set does not possess a
high level of abstraction, as only physical-layer QoS
parameters (SINR, RSS) are used. The parameters processing
is based on a decision function, whose weights are
dynamically adjusted according to user preferences. The
system relies on users‟ criteria scoring for conflict resolution.
Regarding the output format, the scheme produces one
selected access network, having the highest score according to
the user preferences.
B. An Autonomic VHO Scheme with a Client/Server
Application Module (CSAP) [111]
This has been claimed to be one of the first solutions that
can function under diverse real-world scenarios involving a
multitude of network technologies, network providers and
applications. To achieve this versatility, the solution adopts a
client/server scheme operating at OSI Layer 7 through a pair
of applications: the CNAPT (Client Network Address and Port
Translator), which resides at the MT, and the SNAPT (Server
Network Address and Port Translator) at the network side.
These applications abstract technology-dependent details and
introduce a form of virtualization.
The management architecture is described as having an
adjustable structure, a characterization stemming from the
versatile form of cooperation between the CNAPT at the MT
and the SNAPT at the network side. In view of this fact and in
accordance with the discussion in Section III-D, the VHO
management architecture of this solution can be classified as
hybrid.
Considering information collection, user, terminal and
network context is gathered by the CNAPT, with assistance
from the SNAPT. It is stated that the system periodically
searches for available network connections (search activity)
and at the same time, periodically verifies reliability and
performance of the current connection (check activity). The
check activity is related to sampling of the RSS at the
physical layer and to application-layer parameters, inferring
the experienced Round-Trip-Time (RTT) with the help of
ping messages. The scheme does not consider monitoring of
variables at the link-layer, since some NICs do not support
reading such values through standard APIs. With respect to
the knowledge base, the CNAPT includes a history repository,
providing a high memory strength.
The parameters selection step of the handover decision
making presents context-time variability, as it includes not
only static, but also dynamic contextual information (RSS,
RTT). A higher level of abstraction is supported in
comparison to TCAM, as both physical and application layer
QoS parameters are considered. These give some indication
on the effective status of the connection (i.e., being active or
not) and of the effective load. Still, the considered parameters
do not span all layers. Concerning parameters processing,
handover initiation and network selection processes are based
on a generic framework based on thresholds, which can be
classified as a form of PB processing. Specifically, if the
reliability or performance index goes below the specified
critical thresholds or the current network connection is
experiencing an interruption, the „check‟ activity triggers the
handover initiation procedure. The ensuing network selection
relies on the results provided by the search activity.
C. An Intelligent Cross-Layer Terminal-Controlled VHO
Management Scheme (CLTC) [107]
This is another mobile-controlled handover scheme with a
flat VHO management architecture, placing all intelligence on
the mobile devices. The information collection is
implemented by the MT through monitoring and
measurements, to identify the need for handover. The context
information can be relative to the network, the terminal, the
service and the user. QoS parameters are included, such as
bandwidth, delay, jitter, packet loss, RSS and BER of the
current access network and the neighboring available
networks. Furthermore, context information related to user

preferences, service capabilities (real-time and non real-time),
MT status (battery and network interfaces), priority given to
interfaces, location and velocity is collected. The knowledge
base includes a policy repository maintained in the MT, but
this repository does not provide support for the assessment of
past policies and VHO decisions.
The parameter selection step of the handover decision
making is dynamically adjustable, determined by multiple
criteria. The selected parameters present context time
variability, including both static and dynamic contextual
information (such as access network availability, MT‟s
velocity, etc.). Moreover, a high level of abstraction is
supported, as QoS-related parameters are extracted from all
network layers. With respect to the parameter processing,
VHO initiation employs FL. The information gathered is fed
into a fuzzifier converting the aforementioned elements into
fuzzy sets. A fuzzy set contains a varying degree of
membership in a set. For instance, RSS can be weak, medium
or strong. After fuzzification, fuzzy sets are fed to an
inference engine, where a set of fuzzy rules are applied to
determine whether the handover is necessary. Fuzzy rules
utilize a series of IF-THEN rules and the result is YES,
Probably YES, Probably NO or NO. At the final step, the
resultant decision sets have to be "defuzzified". For that, the
centroid method is used to obtain a handover initiation factor
(YES or NO) based on membership values and decision sets.
If a handover is necessary, the network selection stage is
based on an AHP method that allows the decomposition of the
network selection problem into several sub-problems,
corresponding to the decision criteria. The method assigns a
weight to each sub-problem and calculates for each network
the weighted sum characterizing the cumulative impact of all
criteria. The output format of the parameter processing
process is a ranked list of handover targets, with networks
featuring higher weighted sums placed closer to the top of the
list.
D. PROTON: An Autonomic VHO Framework with Finite
State Transducers [55]
A primary characteristic of PROTON is a metric called TF,
related to a Finite State Transducer with Tautness Functions
and Identities (TFFST), which enables policy modeling and
resolves potential conflicts. The relevant management
architecture includes components at both of the network and
MT sides. Those components that involve heavy
computations are placed on the network side, to minimize
complexities at the MT. The VHO is initiated by the MT, but
uses assistance from the network side, which provides
information and computational services (the TFFST models
creation). Thus, the overall handover process can be classified
as network-assisted. Since the main managing entity
controlling the handover process is on the MT, the
management architecture is distributed hierarchical.
The information collection activity is implemented by the
‘sentinels‟ and „retrievers‟, located at the terminal. The
sentinels are responsible for collecting dynamic elements,
whereas the retrievers manage static elements (e.g., user
preferences or application profiles). The knowledge base
includes a policy repository on the network side and a TFFST
Repository on the MT. During the handover decision making,
the parameter selection process is driven by policies and it is
divided into three steps, executed on the MT. Specifically,
the collected information is filtered according to simple local
rules, and then it is grouped into sets. The parameters used
include both static and dynamic context originating from the
physical, network and application layers.
The parameter processing occurs on the network side,
where the conflict resolution module builds a deterministic
Finite State Machine modeling every active policy, and
subsequently generates the set of TFFST profiles, which is
flexible and can be updated according to the MT
requirements. During the TFFST profiles generation, all
possible static and dynamic conflicts are foreseen. Therefore,
the algorithms that are executed have a high computational
cost. Subsequently, the mobile device stores and uses the
TFFST profiles, to be able to react quickly to incoming
events. In order to prioritize TFFST profiles, the tautness
function is formed, to indicate how tautly a condition fits to
an event. In order to quantitatively represent the tautness, a
real number in the interval [−1, 1] is used, so that the stronger
a condition is, the closer its TF is to zero. The corresponding
output format is the most fitting candidate network.
E. An Autonomic VHO Approach with a Context Evaluation
Matrix at the Network Side (COEVAL) [62]
COEVAL implements VHOs by introducing a context
evaluation matrix and a respective context evaluation
function. The management architecture may be classified as
distributed hierarchical, as it includes cooperation between
terminal and network side components. More precisely, the
context server located in the network collects information,
compiling it into a matrix, in response to the handover
initiation request from the MT. During the subsequent
network selection the MT processes the matrix and makes the
handover decision, also considering current dynamic
information. In view of these facts, the handover control
method can be characterized as network-assisted.
Considering information collection, the scheme provides
the mechanisms for the collection, aggregation and filtering of
contextual information, utilizing context from the MT and the
context server (located in the network. More precisely, the
context server collects the relevant context information from
the various context repositories. Then, the MT collects
dynamic context such as the received signal strength, the CPU
usage and the remaining charge on the battery and combines
the collected information with the data from the context
server.
With respect to the knowledge base, the main entity is the
context server (that has no memory of past events), in
addition to various other context repositories, including the
respective Operations and Support Systems (OSS), the
location information database and the user profile database,
all of them residing at the network side.
During the handover decision making, the parameter
selection is based on the received information from the
context server and the current dynamic information from the
MT, derived from all the layers. The parameter processing
method uses a context evaluation (decision) function that
manipulates the matrix context using dynamically adjustable
weights and chooses the appropriate network interface for
each application, taking into account both user and network
preferences. The output of the context evaluation function is
the appropriate network interface for each running
application.

F. AUHO: An Autonomic Personalized Handover Decision
Scheme [54]
AUHO, employs the same architecture, information
collection process and knowledge base components with
COEVAL, in conjunction with a different decision making
process, which employs FL and MAD. Specifically, the
parameter selection step of the handover decision making
employs contextual information provided from all the layers,
including dynamic context. Additionally, user preferences are
perceived, as a set of attributes ordered from most to least
desired, considering RSS, Cost, Quality and Lifetime (i.e.,
remaining battery charge). Considering parameter processing,
the handover initiation stage is performed by means of a FL-
based method, employing fuzzification and defuzzification
mechanisms for the calculation of APAV (Access Point
Acceptance Value) for all available networks. The network
selection employs an additive aggregate utility function (i.e., a
MAD function), which computes the APSV (Access Point
Satisfaction Value) for all candidate networks and chooses the
most satisfying network. The output format is formed by
choosing among the best access points (based on RSS,
Quality, Cost and Lifetime) the one being most important to
the user (prioritized set of candidates), for each application.
VI. COMPARISON AND DISCUSSION
We now compare the VHO solutions of Section V
according to the extent these solutions incorporate and exploit
the autonomic features of Section IV-A, towards enhancing
the effectiveness and efficiency of VHOs. The comparison
also addresses the robustness issues identified in Section IV-
B.
Additionally, we consider issues related with the
operational complexity. This is another important aspect,
which determines the achievable degree of self-management
for the MT. Operational complexity can be generically
characterized as the “degree of complexity of memory and
time” [20] and can be linked to the computational overhead
and the signaling overhead. Accordingly, the comparison of
the solutions also considers the tradeoff between intelligence/
sophistication and operational complexity.
The following subordinate Sections VI-A to VI-E address
individual dimensions of the comparison; a summary appears
in Table III. Finally, Section VI-F culminates with an overall
view.
A. Awareness
Awareness, the basis of all other autonomic criteria, is
related with the information collection and parameter
selection processes. All aforementioned VHO solutions
present awareness, though in a varying degree: the two
simpler and more lightweight approaches, namely TCAM and
CSAP, provide a basic form of awareness, while the other
solutions exhibit more enhanced awareness, but at the cost of
higher complexity. Specifically, TCAM limits information
collection to just physical layer parameters used to measure
the signal quality of the candidate network (SINR, RSS).
Thus, TCAM does not have potential for multi-QoS
consideration. CSAP takes a simple approach too, but
supplements the physical layer monitoring of RSS with the
application layer monitoring of RTT (Round Trip Time),
which gives some indication on the effective status of the
connection, the effective load and the available throughput.
Still, the level of abstraction is not high enough to provide
potential for explicit multi-QoS consideration.
TABLE II
CLASSIFICATION OF THE SELECTED VHO MANAGEMENT SOLUTIONS, ACCORDING TO THEIR CHARACTERISTICS
Solution
Terminal Side
Functionalities
Network Side
Functionalities
Type of Management
Architecture/
Handover Control
Parameters Processing Method
TCAM [84]
All the intelligence at
the MT, VHO
Decision & Execution
None
Flat / Mobile Controlled
DF, with dynamically adjustable weights (based
on user preferences)
CSAP [111]
VHO Initiation,
Network Selection &
Execution
Assists the MT during the
information collection and
the network selection
(search & check
activities)
Hybrid / Network
Assisted
Generic PB framework based on thresholds
CLTC [107]
All the intelligence at
the MT, VHO
Decision & Execution
None
Flat / Mobile Controlled
FL for VHO initiation
& AHP (i.e., MAD) for network selection
PROTON [55]
VHO Initiation &
Network Selection
(TF computation),
VHO Execution
Computationally
demanding tasks, TFFST
profiles computation
Distributed Hierarchical
/ Network Assisted
Policy-based: TFFST model creation,
implemented with Finite State Automata &
Tautness Function
COEVAL [62]
VHO Initiation,
Network Selection &
Execution
Context Server and
various context
repositories assist in
information collection
Distributed Hierarchical
/ Network Assisted
Context Evaluation Function (i.e. DF)
AUHO [54]
VHO Initiation,
Network Selection &
Execution
Context Server and
various context
repositories assist in
information collection
Distributed Hierarchical
/ Network Assisted
FL for VHO initiation
& Additive Aggregate Utility Function (i.e.,
MAD) for network selection

Turning to the more sophisticated approaches, CLTC
implements active monitoring, with parameters extracted
from all layers. The solution considers QoS parameters
(bandwidth, delay, jitter, packet loss, traffic load), coverage,
monetary cost, link quality (RSS and BER) of the current
access network and its neighbors, as well as location
information. However, all these parameters may not be
needed in every scenario, thus, the tradeoff between enhanced
awareness and the resulting signaling and computational
overheads should be taken into account, especially
considering that according to this approach all the intelligence
is placed at the MTs.
PROTON provides active monitoring and a high degree of
awareness, through monitoring parameters at different layers
and organizing the collected data according to a three-level
hierarchy, which reduces the volume of data processing.
While PROTON‟s framework could in principle enable multi-
QoS consideration, it lacks information necessary for
explicitly considering the demands of different running
applications. This shortcoming might be due to the fact that
PROTON is one of the first approaches on autonomic VHO
management.
AUHO and COEVAL support a high degree of awareness
through active monitoring. More precisely, the mobile device
performs measurements to retrieve updated dynamic and
static information from its sensors, the user and the context
server (in the network side). Emphasis is given to information
related to the running application requirements and the
available network interfaces, considering parameters such as
bandwidth, packet error rate, delay, jitter and packet loss
ratio, which assist in proposing the best network interface for
each running application. Also, location information is
included in the handover management parameters that add a
spatial dimension in the handover initiation criteria.
B. Adaptivity and Flexibility
Considering adaptivity and flexibility, the information
collection mechanisms and handover decision making
processes are compared in view of the presented solutions. In
general, there is an inherent trade-off between flexibility and
computational overhead. TCAM, being the simplest and most
lightweight solution, is characterized by a rather limited
adaptivity and flexibility, as it deals with a predefined set of
parameters and does not provide adaptation mechanisms in
information collection. The other approaches present more
enhanced adaptivity and flexibility characteristics, but are also
more computationally demanding. In CSAP, for example, the
monitoring activity is still non-adaptive, as it uses a constant
rate of parameters sampling. However, adaptive thresholds
are used in parameter processing (“check activity”).
In CLTC, the Analytic Hierarchy Process offers advanced
flexibility and also adaptivity (through the possibility of
dynamically adapting the various weighting factors). On the
other hand, adaptive monitoring mechanisms are not
considered. By contrast, PROTON offers sophisticated
monitoring adaptivity, as each parameter is collected
according to a specific polling frequency, depending on
connectivity resources and mobility profiles. Specifically, the
system adapts the frequency of active monitoring
proportionally to the MT‟s velocity, matching thus the
information collection rate to variations of the user's physical
mobility. Considering the decision phase, PROTON, provides
advanced flexibility and accordingly provides enhanced
adaptivity mechanisms through a dynamic set of TFFSTs and
the use of suitable tautness functions for each case.
In COEVAL, the MT fills in the dynamic contextual
information and calculates the evaluation matrix when a
decision is needed, applying policies that may include rules to
set the upper or lower bounds. The matrix mechanism
provides advanced flexibility, being dynamically filled with
the available parameters. Also, the dynamic upper/lower
bounds offer adaptivity. Finally, AUHO features advanced
flexibility through the Multiple Attribute Decision method,
where the output is calculated as a linear function of context
input and dynamically changing weights, with respect to
different criteria.
C. Learning
This autonomic criterion is related to the ability of the
system to learn, enabled by the memory strength provided by
historically available data. Interestingly, only the simpler
approaches, TCAM and CSAP, provide a form of memory
strength, which can be exploited to include learning
mechanisms. In TCAM a black list of access network
operators is included, containing the networks where the user
has had a bad experience. Additionally, the description of the
solution [84] mentions that users can specify and alter their
preferences dynamically, through a learning process. CSAP
employs a repository of the most significant information,
containing trends, failures, trajectories, user choices, etc.,
about past experience, and providing the ability to adjust
internal parameters and derive statistical measures of trend.
For instance, if an on-board GPS is available, the system can
decide to store and learn maps identifying good coverage
areas together with the characteristics of the network access
that can provide the coverage. This might prove quite useful
in the case of users constantly traveling along the same routes,
as it is the case for people daily commuting between their
homes and work places.
D. Proactivity
Proactivity is based on preventive measurements promoting
actions in the direction of system anticipation. The solutions
under investigation that involve proactive mechanisms are
CSAP, CLTC and PROTON. Specifically, the description of
CSAP [111] mentions that the system is able to efficiently
smooth the sampled values of measures RSS, through simple
weighted moving averages, and at the same time calculate a
simple trend indicator to be used in cross-validation with the
moving average, enhancing proactivity. Considering CLTC,
in a new and enhanced version of the approach [61],
predictive Link layer information is taken into account
extending the proactivity of the solution. More specifically,
the system detects the quality of the current link (concerning
physical and MAC layers) and can issue periodically a polling
command to check the status of the link, expressing the
likelihood of future changes in the link properties (e.g., link
going down, link going up, etc) based on present conditions.
Finally, PROTON implements a conflict resolution module to
resolve conflicts among the policy rules. During this task, all
possible static and dynamic conflicts are foreseen, enhancing
proactivity.

TABLE III
COMPARISON CRITERIA FOR AUTONOMIC VHO MANAGEMENT SCHEMES
Autonomic
Criteria
Awareness
Adaptivity &
Flexibility
Learning
Proactivity
Robustness
TCAM
low level of
abstraction: only
physical layer
parameters to
measure the
signal quality
limited flexibility,
non-adaptive
monitoring, limited
adaptivity in the
decision making
(dynamic weights
applied for user
preferences only)
black list of
access network
operators
no relevant information
about the frequency of
monitoring measurements,
relies only on users‟
criteria scoring for
conflict resolution
no multi-QoS consideration, no provision
for managing uncertainties or
incompleteness, no marginal/borderline
cases consideration
CSAP
medium level of
abstraction:
physical (RSS) &
application
(RTT) parameter
selection
advanced flexibility,
adaptive thresholds
in decision making,
non-adaptive
monitoring
repository of the
most significant
information
(trends, failures,
trajectories, user
choices) about
past experience
"check activity" follows a
periodic activation
scheme, trend indicator
no full specifications provided about the
network selection process; temporary
fluctuations of physical parameters are
verified by application layer ping messages,
managing uncertainties and avoiding ping-
pong effect
CLTC
high level of
abstraction:
multi-layer
parameter
selection
advanced flexibility,
adaptive decision
making (through
adaptive weights),
non-adaptive
monitoring
no memory
strength
predictive link layer
information (in the new
version [61])
enhanced robustness: context and criteria
diversity consideration through parallel
fuzzy rules and AHP, multi-QoS
consideration, provision for managing
incompleteness, but no marginal/borderline
cases consideration
PROTON
high level of
abstraction:
multi-layer
parameter
selection, three-
level
organizational
hierarchy of
collected data
advanced flexibility,
adaptive decision
making, well
structured adaptive
monitoring
framework
no memory
strength
all possible static and
dynamic conflicts are
foreseen
no multi-QoS consideration, criteria
diversity consideration through policies‟
conflict resolution, policies related to
hysteresis that could manage
marginal/borderline cases
COEVAL
high level of
abstraction:
multi-layer
parameter
selection
advanced flexibility,
adaptive decision
making (adaptive
thresholds), non-
adaptive monitoring
no memory
strength
no
multi-QoS consideration, no provision for
managing uncertainties/incompleteness, no
marginal/borderline cases consideration
AUHO
high level of
abstraction:
multi-layer
parameter
selection
advanced flexibility,
adaptive decision
making, non-
adaptive monitoring
no memory
strength
no
enhanced robustness: context and criteria
diversity consideration through parallel
fuzzy rules, multi-QoS consideration,
provision for managing incompleteness, but
no marginal/borderline cases consideration
E. Robustness
We now focus on the robustness issues considering the
decision making procedure. With respect to the uniform
representation enabling context diversity consideration,
TCAM and COEVAL use CPN, while, CLTC and AUHO use
a combination of FL and CPN techniques. Lastly, CSAP and
PROTON use PB techniques, where each parameter is
processed individually, through a relevant policy. Adjustable
tailoring of the normalization parameters, according to QoS
demands for each running application is considered in
COEVAL, but not in TCAM. CLTC and AUHO follow the
framework proposed by [115] and consider different
membership functions for each application, while they use
CPN to simpler criteria. CSAP and PROTON lack
information about multi-QoS considerations according to
running applications requirements (as already mentioned in
Section VI-A).
Considering the diversity of criteria/rules, including the
assignment of a different importance to each group of criteria,
the most comprehensive approach is taken by CLTC and
AUHO, which use FL with parallel fuzzy rules. Additionally,
in CLTC parameters are grouped into a hierarchical model, in
order to be handled more efficiently through AHP. However,
this approach uses a rather complex weighting method, so the
complexity of CLTC is higher than that of AUHO. COEVAL
deals with the matter in simpler terms: while it allows the
assignment of a different importance to each individual
parameter, it does not provide support for handling an entire
group of criteria. PROTON employs its policies conflict
resolution module to combat diversity of criteria/rules.
Finally, TCAM and CSAP inherently lack capabilities for
dealing with complex decisions.
We now turn to the management of uncertainties and
incomplete information during the decision making process.
To combat incompleteness, the FL-based AUHO and CLTC
solutions substitute a neutral value in place of missing
parameters. PROTON, COEVAL and TCAM do not provide
any explicit support for managing incompleteness. Finally,
CSAP provides some means to guard against uncertainties
arising from excessive parameter value fluctuations.
Specifically, the solution tries to avoid improper reaction to
temporary fluctuations of physical layer parameters, by cross-
checking a bad link status against application layer
information (obtained through ping messages).
Concerning marginal/borderline cases and the ping-pong
effect, predictive link layer information is included in the new

version of CLTC [61], which could possibly assist in the
confrontation of this problem, while, PROTON presents
policies related to hysteresis margin. CSAP may deal with the
ping-pong effect through its previously mentioned mechanism
for managing uncertainties. The rest of the solutions do not
include support for managing marginal/borderline cases.
As a whole, the solutions with the most comprehensive
provision for robustness are CLTC and AUHO.
F. Overall Comments
While the simpler solutions TCAM and CSAP provide only
moderate potential for overall self-optimization, due to their
incomplete awareness and limited adaptivity and flexibility,
the overall complexity of the corresponding VHO decision
making procedures is low, signifying a high degree of
achievable self-management for the MT. On the contrary, the
performance potential of CLTC, PROTON, COEVAL and
AUHO is greater, in view of their enhanced awareness and
adaptivity & flexibility features, but this comes at the cost of a
higher complexity. The complexity of CLTC, in particular,
may be characterized as quite high, so the flat and mobile
controlled architecture of this solution might prove
impractical, as the heavy programming tasks could
overwhelm the MTs. PROTON, COEVAL and AUHO are
better positioned in this respect, as their distributed
hierarchical architecture and network assisted control foresee
centralized entities to take up the heavy programming tasks,
reducing the burden put on the MTs.
Another noteworthy aspect, particularly in a FI context,
relates to the concurrent exploitation of different network
interfaces on the MT for serving different running
applications. In this direction, the parameters selection set
should allow information from multiple layers to be included
and matched with the running application requirements, so
that the system can select the most appropriate access network
for each running application. CLTC, AUHO and COEVAL
provide the most comprehensive support for this.
As already mentioned in Section VI-C, only TCAM and
CSAP provide some form of memory strength that may be
exploited towards cognition and learning. This existence of
memory strength might be seen as a supplement to the
moderate degrees of awareness, flexibility and adaptivity
present in these simpler solutions. However, the other four
more sophisticated solutions could also benefit from memory
strength and additional learning mechanisms. Although the
incorporation of such mechanisms may involve initially
increased computational overheads, the more effective
prevention of unnecessary VHOs could counter-balance these
overheads and eventually lead to enhanced performance.
The additional mechanisms could be hosted by higher level
entities, particularly for hybrid or hierarchical architectures,
such as those in AUHO and COEVAL, avoiding an extra
burden on the MTs. In AUHO, for example, pre-calculated
APSV values (see Section V-F) characterizing the network
interfaces under typical patterns of context could be stored in
the knowledge base, towards faster and less computationally
demanding decisions. Along further directions, learning
mechanisms can be employed to optimize the formulation of
membership functions for FL-based solutions (AUHO and
CLTC), to tune the upper/lower values and thresholds used
for CPN (in COEVAL and TCAM), to optimize the
formulation of policies in PB systems (CSAP and PROTON),
or to formulate and optimize the score and hysteresis margins
used when dealing with marginal/borderline cases.
VII. FUTURE RESEARCH DIRECTIONS
Based on the aforementioned analysis, connectivity
management in 5G networks is still an open issue, considering
the augmenting densification of networks, the multitude of
radio access technologies and the unplanned network layout,
posing several challenges. Current research directions,
promote context-aware strategies that demand feedback from
the MTs. Related operations can be further optimized
implementing self-x capabilities, which can be founded on the
use of cooperative and cognitive radio strategies to reduce the
mobile operator‟s maintenance and administration overhead,
towards performance enhancement and minimization of the
required energy consumption and delay overhead. Towards
these goals, ANM provides context-aware MTs that are able
to self-manage their mobility behavior according to policy-
based management principles.
Future autonomic VHO management architectures are
likely to have a more flexible and customizable structure,
towards adjusting dynamically the degree of self-management
for each entity in the distributed architecture. Meanwhile,
network elements higher up in the hierarchy should be able to
enforce the appropriate policies on the MTs, in order to
achieve global optimization goals. The relevant processes
may combine ANM with SDN and network virtualization
concepts, enabling a shift from device-driven management
models to context-aware and QoS-aware management
models, covering the market demand for more flexible and
extensible network designs.
In such hybrid future autonomic architectures, some self-
organization and self-optimization algorithms, mostly those
related to tasks with local scope, would run locally on the
MTs, while the tasks with wider scope (global network view)
could be managed by a central managing authority (i.e. a
network controller) on the network side. In particular, the
global knowledge could provide enhanced information to
MTs enabling dynamically optimized policies, towards
achieving a local optimum in balance with the global
optimum, according to an evolutionary process. In this way,
the integration of autonomic MTs and autonomic network
entities higher up in the hierarchy may lead to a VHO
management solution featuring increased reliability and
efficiency.
Accordingly, global awareness can be built by combining
the self-awareness of each distributed entity, while MTs could
utilize the global view to evaluate/verify and complement
their own status. Towards the implementation of such global
knowledge base, the strict latency requirements for mobility
management in a 5G network environment, demand
computing resources and storage at the edge of a network,
introducing edge, mobile edge, mobile cloud and fog
computing concepts. Such distributed knowledge base
paradigms could facilitate the MTs to store individually
essential information about their mobility for later use, further
promoting the concept of self-management.
VIII. CONCLUSIONS
This survey explored the field of autonomic VHO
management. By employing concepts of ANM to VHO
management, it became possible to shed new light to VHO
operations from an ANM point of view, investigating the role

of context-awareness and self-x capabilities, towards
encompassing FI environments and the emerging 5G
networks.
As a first step, the survey reviewed basic concepts
regarding cognition and autonomicity. In the point of view
taken, cognitive functions are considered as parts of ANM,
characterizing a system aware of itself and its environment,
self-governing its behavior to achieve specific goals. This
view includes the notion of self-management. Subsequently,
these concepts were employed in a classification and analysis
of the components, processes and algorithms involved in
autonomic handover management.
Ultimately, a new taxonomy of the relevant architectural
components was introduced, considering the scenario of
context-aware MTs that operate within a complex FI
environment and self-manage their handover behavior.
According to this taxonomy, the autonomic handover
management procedure was organized into the phases of
information collection, being linked to a knowledge base,
followed by the handover decision making, which includes
handover initiation and network selection processes and its
corresponding algorithms, itself followed by the handover
execution, which includes handover preparation and the
related signaling. In this way, the VHO management complies
with the autonomic management principles of monitor,
analyze & plan, and execute functions.
Following the survey‟s point of view, the standard media
independent handover management frameworks were
reviewed and correlated with the new autonomic framework.
Relevant amendments and/or extensions to the standards from
the literature were also reviewed, together with works
addressing efficiency, performance evaluation and related
modeling aspects.
As an additional contribution, the survey highlighted a
number of important autonomic features that may be
leveraged to automatically make the autonomic handover
management adaptive and to optimize its performance,
towards the overall enhancement of the VHO operations. It
was demonstrated that the combination of awareness,
adaptivity & flexibility, learning and proactivity drive the
system to performance improvements and enable the system
to select the best choice among a set of available alternatives,
advancing the system‟s overall self-optimization.
Robustness issues, related to the ability to achieve stable
and efficient VHO decisions in the diverse and dynamic
context of a FI environment, were also considered, filling a
gap in the literature. The survey identified a number of
robustness-related issues and reviewed mechanisms to cope
with them, presenting also how robustness can be enhanced
through the exploitation of autonomic features.
To demonstrate the applicability of the general concepts,
the survey reviewed a number of representative VHO
management solutions with an autonomic orientation taken
from the literature. These solutions were analyzed and
relevant characteristics were associated with the classification
and taxonomy contributed by the survey. Furthermore, the
solutions were compared in terms of the extent they
incorporate and exploit the autonomic features identified
earlier in the survey, towards enhancing the effectiveness and
efficiency of VHOs and achieving robustness. The principles
employed in the analysis and comparison of these particular
solutions can be useful also for the future treatment of other
VHO management solutions with an autonomic orientation.
In the course of this survey, it was seen that, in order to
provide seamless VHOs and enhanced decisions in a FI
environment, there is need for information collection targeting
parameters over multiple layers. This requires advanced
awareness. A flexible and adaptive set of parameters is also
key to a more effective support for the QoS requirements of
running applications. Once the sophistication of the
parameters set increases, it becomes important to ensure the
robust operation of the system, even in unpredictable
situations, by appropriately handling the diversity of
parameters and criteria/rules, by providing resilience under
context uncertainties and incompleteness and by including
mechanisms to withstand marginal/borderline cases.
Also, the survey highlighted the tradeoff between
intelligence/sophistication and operational complexity, which
determines the achievable degree of self-management for the
MT. While autonomic VHO management architectures should
remain distributed enough to enable the MTs to make at least
some decisions on their own, it may prove infeasible to
implement sophisticated techniques solely on end-devices. In
order for the system to keep its prompt reaction and maintain
its potential for self-optimization, it may be advisable to
introduce hierarchy in the VHO architecture and have specific
time- and resource-demanding procedures linked with the
cognition cycle (such as conflict resolution correlated to
network-wide statistics and big data analysis, or learning
algorithms) be coordinated by network entities higher up in
the hierarchy, removing the burden from the MTs. As a final
step, the lessons learnt by the survey led to a consideration of
future research directions, considering customizable ANM
architectures, and the combination of ANM and emerging
technologies in the 5G network environment.
REFERENCES
[1]
A. Galis and A. Gavras, Eds., The Future Internet - Future
Internet Assembly 2013: Validated Results and New Horizons.
Heidelberg: Springer, 2013.
[2]
J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C.
Soong, and J. C. Zhang, "What Will 5G Be?," IEEE Journal On
Selected Areas In Communications, vol. 32, no. 6, pp. 1065-1082,
2014.
[3]
T. Chen, M. Matinmikko, X. Chen, X. Zhou, and P. Ahokangas,
"Software defined mobile networks: concept, survey, and research
directions," IEEE Communications Magazine, vol. 53, no. 11, pp.
126-133, 2015.
[4]
J. F. Monserrat, G. Mange, V. Braun, H. Tullberg, G.
Zimmermann, and O. Bulakci, "METIS research advances
towards the 5G mobile and wireless system definition," EURASIP
Journal on Wireless Communications and Networking, vol. 53,
no. 1, pp. 1-14, 2015.
[5]
A. Kousaridas, C. Polychronopoulos, N. Alonistioti, A. Marikar,
J. Mödeker, A. Mihailovic, G. Agapiou, I. Chochliouros, and G.
Heliotis, "Future Internet Elements: Cognition and Self-
management Design Issues," in Autonomics '08 Proceedings of
the 2nd International Conference on Autonomic Computing and
Communication Systems, 2008, pp. 13-20.
[6]
J. Gubbi, R. Buyya, S. Marusic, and M. Palaniswami, "Internet of
Things (IoT): A vision, architectural elements, and future
directions.," Future Generation Computer Systems, vol. 29, no. 7,
pp. 1645-1660, 2013.
[7]
A. Ahmed, L. M. Boulahia, and D. Gaiti, "Enabling Vertical
Handover Decisions in Heterogeneous Wireless Networks: A
State-of-the-Art and A Classification," IEEE Communications
Surveys & Tutorials, vol. 16, no. 2, pp. 776-811, 2014.

[8]
C. Perera, A. Zaslavsky, P. Christen, and D. Georgakopoulos,
"Context aware computing for the internet of things: A survey,"
IEEE Communications Surveys & Tutorials, vol. 16, no. 1, pp.
414-454, 2014.
[9]
Z. Movahedi, M. Ayari, R. Langar, and G. Pujolle, "A Survey of
Autonomic Network Architectures and Evaluation Criteria," IEEE
Communications Surveys & Tutorials, vol. 14, no. 2, pp. 464-490,
2012.
[10]
K. Tsagkaris, M. Logothetis, V. Foteinos, G. Poulios, M.
Michaloliakos, and P. Demestichas, "Customizable autonomic
network management: integrating autonomic network
management and software-defined networking.," IEEE Vehicular
Technology Magazine, vol. 10, no. 1, pp. 61-68, 2015.
[11]
P. Neves, R. Calé, M. Costa, G. Gaspar, J. Alcaraz-Calero, Q.
Wang, J. Nightingale, G. Bernini, G. Carrozzo, A. Valdivieso, L.
J. García Villalba, M. Barrose, A. Gravas, J. Santos, R. Maia, and
R. Pretog, "Future mode of operations for 5G–The SELFNET
approach enabled by SDN/NFV," Computer Standards &
Interfaces, vol. 54, 2017.
[12]
J. A. Wickboldt, W. P. De Jesus, P. H. Isolani, C. B. Both, J.
Rochol, and L. Z. Granville, "Software-defined networking:
management requirements and challenges," IEEE
Communications Magazine, vol. 53, no. 1, pp. 278-285, 2015.
[13]
S. Schmid, M. Sifalakis, and D. Hutchison, "Towards Autonomic
Networks," Autonomic Networking, Lecture Notes in Computer
Science, vol. 4195, pp. 1-11, 2006.
[14]
B. Jennings, S. Van Der Meer, S. Balasubramaniam, D. Botvich,
M. O. Foghlú, W. Donnelly, and J. Strassner, "Towards
Autonomic Management of Communications Networks," IEEE
Communications Magazine, vol. 45, no. 10, pp. 112-121, 2007.
[15]
N. Samaan and A. Karmouch, "Towards Autonomic Network
Management: an Analysis of Current and Future Research
Directions," IEEE Communications Surveys & Tutorials, vol. 11,
no. 3, pp. 22 - 36, 2009.
[16]
K. Tsagkaris, P. Vlacheas, G. Athanasiou, V. Stavroulaki, S.
Filin, H. Harada, J. Gebert, and M. Mueck, "Autonomics in
wireless network management: Advances in Standards and
Further Challenges," IEEE Network, vol. 25, no. 6, pp. 41-49,
2011.
[17]
P. Makris, D. N. Skoutas, and C. Skianis, "A survey on context-
aware mobile and wireless networking: On networking and
computing environments' integration," IEEE communications
surveys & tutorials, vol. 15, no. 1, pp. 362-386, 2013.
[18]
M. Z. Asghar, P. Nieminen, S. Hämäläinen, T. Ristaniemi, M. A.
Imran, and T. Hämäläinen, "Towards proactive context-aware
self-healing for 5G networks," Computer Networks, vol. 128, pp.
5-13, 2017.
[19]
M. Kassar, B. Kervella, and G. Pujolle, "An overview of vertical
handover decision strategies in heterogeneous wireless networks,"
Computer Communications, vol. 31, no. 10, pp. 2607-2620, 2008.
[20]
J. Marquez-Barja, C. T. Calafate, J. C. Cano, and P. Manzoni,
"An overview of vertical handover techniques: Algorithms,
protocols and tools," Computer Communications, vol. 34, no. 8,
pp. 985-997, 2010.
[21]
X. Yan, Y. A. Şekercioğlu, and S. Narayanan, "A survey of
vertical handover decision algorithms in Fourth Generation
heterogeneous wireless networks," Computer Networks, vol. 54,
no. 11, pp. 1848-1863, 2010.
[22]
S. Fernandes and A. Karmouch, "Vertical mobility management
architectures in wireless networks: A comprehensive survey and
future directions," IEEE Communications Surveys & Tutorials,
vol. 14, no. 1, pp. 45-63, 2012.
[23]
M. F. Tuysuz and R. Trestian, "Energy-efficient vertical handover
parameters, classification and solutions over wireless
heterogeneous networks: a comprehensive survey," Wireless
Personal Communications, vol. 97, no. 1, pp. 1155-1184, 2017.
[24]
P. Bellavista, A. Corradi, and C. Giannelli, "A Unifying
Perspective on Context-Aware Evaluation and Management of
Heterogeneous Wireless Connectivity," IEEE Communications
Surveys & Tutorials, vol. 13, no. 3, pp. 337-357, 2011.
[25]
M. Louta and P. Bellavista, "Bringing Always Best Connectivity
Vision a Step Closer: Challenges and Perspectives," IEEE
Communications Magazine, vol. 51, no. 2, pp. 158-166, 2013.
[26]
R.W. Thomas, L.A. DaSilva, and A.B. MacKenzie, "Cognitive
networks," in Proceedings of the First IEEE International
Symposium on New Frontiers in Dynamic Spectrum Access
Networks, DySPAN, 2005, pp. 352-360.
[27]
J. Strassner, J. N de Souza, D. Raymer, S. Samudrala, S. Davy,
and K. Barrett, "The design of a novel context-aware policy
model to support machine-based learning and reasoning," Cluster
Computing, vol. 12, no. 1, pp. 17-43, 2009.
[28]
"Autonomic Computing: IBM‟s Perspective on the State of
Information Technology," IBM Research Headquarters
(manifesto), 2001.
[29]
J. O. Kephart and D. M. Chess, "The Vision of Autonomic
Computing," IEEE Computer, vol. 36, no. 1, pp. 41-50, 2003.
[30]
W. F. Truszkowski, M. G. Hinchey, J. L. Rash, and C. A. Rouff,
"Autonomous and Autonomic Systems: A Paradigm for Future
Space Exploration Missions," IEEE Transactions on Systems,
Man, and Cybernetics, Part C (Applications and Reviews), vol.
36, no. 3, pp. 279-291, 2004.
[31]
J. Strassner, D. Raymer, and , S. Samudrala, "Providing Seamless
Mobility Using the FOCALE Autonomic Architecture," Next
Generation Teletraffic and Wired/Wireless Advanced Networking,
NEW2AN 2007, Lecture Notes in Computer Science, vol. 4712,
pp. 330-341, 2007.
[32]
R. Boutaba and I. Aib, "Policy-based Management: A Historical
Perspective," Journal of Networks System Management, vol. 15,
no. 4, pp. 447-480, 2007.
[33]
W. E. Walsh, G. Tesauro, J. O. Kephart, and R. Das, "Utility
Functions in Autonomic Systems," in Proceedings of the IEEE
International Conference on Autonomic Computing, 2004, pp. 70-
77.
[34]
E. E. Tsiropoulou, A. Kapoukakis, and S. Papavassiliou, "Uplink
resource allocation in SC-FDMA wireless networks: A survey
and taxonomy.," Computer Networks, vol. 1-28, pp. 1-28, 2016.
[35]
M. Salehie and L. Tahvildari, "Self-adaptive software: Landscape
and research challenges.," ACM Trans. Autonom. Adapt. Syst.,
vol. 4, no. 2, pp. 1-14, 2009.
[36]
E. E. Tsiropoulou, P. Vamvakas, and , S. Papavassiliou, "Joint
Customized Price and Power Control for Energy-Efficient Multi-
Service Wireless Networks via S-Modular Theory.," IEEE
Transactions on Green Communications and Networking, vol. 1,
no. 1, pp. 17-28, 2017.
[37]
P. Vamvakas, E. E. Tsiropoulou, and S. Papavassiliou, "A user-
centric economic-driven paradigm for rate allocation in non-
orthogonal multiple access wireless systems.," EURASIP Journal
on Wireless Communications and Networking, vol. 129, no. 1,
2018.
[38]
E.E. Tsiropoulou, T. Kastrinogiannis, and S. Papavassiliou,
"Realization of QoS provisioning in autonomic CDMA networks
under common utility-based framework," in World of Wireless,
Mobile and Multimedia Networks & Workshops, WoWMoM 2009,
2009, pp. 1-7.
[39]
E. E. Tsiropoulou, P. Vamvakas, G. K. Katsinis, and S.
Papavassiliou, "Combined power and rate allocation in self-
optimized multi-service two-tier femtocell networks.," Computer
Communications, vol. 72, pp. 38-48, 2015.
[40]
X. Dong, S. Hariri, L. Xue, H. Chen, M. Zhang, S. Pavuluri, and

S. Rao, "Autonomia: an autonomic computing environment," in
IEEE International Conference Proc. in Performance,
Computing, and Communications Conference, 2003, pp. 61-68.
[41]
R. Chadha, Y.-H. Cheng, J. Chiang, G. Levin, S. W. Li, and A.
Poylisher, "Policy-based mobile ad hoc network management for
drama," in Military Communications Conference - MILCOM
2004, vol. 3, 2004, pp. 1317-1323.
[42]
D. M. Chess, A. Segal, I. Whalley, and S. R. White, "Unity:
Experiences with a prototype autonomic computing system," in
ICAC ’04: Proc. First International Conference on Autonomic
Computing, IEEE Computer Society, 2004, pp. 140-147.
[43]
L. Hua and M. Parashar, "Accord: a programming framework for
autonomic applications," in IEEE Transactions on Systems, Man,
and Cybernetics, Part C (Applications and Reviews), 2006, pp.
341-352.
[44]
A. Malatras, A. M. Hadjiantonis, and G. Pavlou, "Exploiting
context-awareness for the autonomic management of mobile ad
hoc networks," Journal of Network and Systems Management,
vol. 15, no. 1, pp. 29–55, 2007.
[45]
A. Bassi, S. Denazis, A. Galis, C. Fahy, M. Serrano, and M.
Serrat, "Autonomic internet: a perspective for future internet
services based on autonomic principles," in Proc. 2nd IEEE
International Workshop on Modelling Autonomic
Communications (MACE), 2007.
[46]
G. Bouabene, C. Jelger, G. Tschudin, S. Schmid, A. Keller, and
M. May, "The autonomic network architecture (ANA)," IEEE
Journal on Selected Areas in Communications, vol. 28, no. 1,
2010.
[47]
D. Xenakis, N. Passas, L. Merakos, and C. Verikoukis, "Mobility
management for femtocells in LTE-advanced: key aspects and
survey of handover decision algorithms.," IEEE Communications
Surveys & Tutorials, vol. 16, no. 1, pp. 64-91, 2014.
[48]
R. Arshad, H. ElSawy, S. Sorour, T. Y. Al-Naffouri, and M. S.
Alouini, "Velocity-aware handover management in two-tier
cellular networks.," IEEE Transactions on Wireless
Communications, vol. 16, no. 3, pp. 1851-1867, 2017.
[49]
R. Arshad, H. ElSawy, S. Sorour, T. Y. Al-Naffouri, and M. S.
Alouini, "Handover management in 5G and beyond: A topology
aware skipping approach.," IEEE Access, vol. 4, pp. 9073-9081,
2016.
[50]
H. Ibrahim, H. ElSawy, U. T. Nguyen, and M. S. Alouini,
"Mobility-aware modeling and analysis of dense cellular networks
with C-plane/ U-plane split architecture.," IEEE Transactions on
Communications, vol. 64, no. 11, pp. 4879-4894, 2016.
[51]
"IEEE Standard for Local and Metropolitan Area Networks- Part
21: Media Independent Handover, IEEE Std 802.21-2008, pp.c1-
301," Jan. 2009.
[52]
"Third-Generation Partnership Project, Access to the 3GPP
evolved packet core (EPC) via non-3GPP access networks," 3GPP
TS 24.302, 2014.
[53]
A. Francisco, R. A. Enriquez-Caldera, J. M. Ramirez-Cortes, J.
Martinez-Carballido, and E. Buenfil-Alpuche, "Towards a
Cognitive Handoff for the Future Internet: A Holistic Vision," in
COGNITIVE 2010 : The Second International Conference on
Advanced Cognitive Technologies and Applications, 2010.
[54]
J. M. Kang, J. Strassner, S. S. Seo, and J. W. K. Hong,
"Autonomic personalized handover decisions for mobile services
in heterogeneous wireless networks," Computer Networks, vol.
55, no. 7, pp. 1520-1532, 2011.
[55]
P. Vidales, J. Baliosian, J. Serrat, and G. Mapp, "Autonomic
System for Mobility Support in 4G Networks," IEEE Journal on
Selected Areas in Communications, vol. 23, no. 12, pp. 2288-
2304, 2005.
[56]
Y. Cheng, R. Farha, M. S. Kim, A. Leon-Garcia, and J. W. K.
Hong, "A generic architecture for autonomic service and network
management," Computer Communications, vol. 29, no. 18, pp.
3691-3709, 2006.
[57]
M. Rahman, R. Ranjan, R. Buyya, and B. Benatallah, "A
taxonomy and survey on autonomic management of applications
in grid computing environments," Concurrency and computation:
practice and experience, vol. 23, no. 16, pp. 1990-2019, 2011.
[58]
Hucheng Wang, Shanzhi Chen, Hui Xu, Ming Ai, and Yan Shi,
"SoftNet: A Software Defined Decentralized Mobile Network
Architecture toward 5G," IEEE Network, vol. 29, no. 2, pp. 16-22,
2015.
[59]
S. Azodolmolky, P. Wieder, and R. Yahyapour, "Cloud
computing networking: challenges and opportunities for
innovations," IEEE Communications Magazine, vol. 51, no. 7, pp.
54-62, 2013.
[60]
A. V. Dastjerdi and R. Buyya, "Fog Computing: Helping the
Internet of Things Realize Its Potential," IEEE Computer, vol. 49,
no. 8, pp. 112-116, 2016.
[61]
T. A. Yahiya, G. Pujolle, and A. Beylot, "An Autonomic-Oriented
Framework Based IEEE 802.21 for Mobility Management in 4G
Networks," ACM SIGMOBILE Mobile Computing and
Communications Review, vol. 15, no. 1, pp. 37-51, 2011.
[62]
J. M. Kang, H.T. Ju, and J.W.K. Hong, "Towards autonomic
handover decision management in 4G networks," in IFIP/IEEE
International Conference on Management of Multimedia and
Mobile Networks and Services, 2006, pp. 145-157.
[63]
H. J. Wang, R. H. Katz, and J. Giese, "Policy-Enabled Handoffs
Across Heterogeneous Wireless Networks," in Second IEEE
Workshop on Mobile Computing Systems and Applications,
WMCSA, 1999, pp. 51-60.
[64]
W. Shen and Q. Zeng, "Cost-Function-Based Network Selection
Strategy in Integrated Wireless and Mobile Network," IEEE
Transactions on Vehicular Technology, vol. 57, no. 6, pp. 3778-
3788, 2008.
[65]
S. Lee, K. Sriram, K. Kim, Y. H. Kim, and N. Golmie, "Vertical
Handoff Decision Algorithms for Providing Optimized
Performance in Heterogeneous Wireless Networks," IEEE
Transactions on Vehicular Technology, vol. 58, no. 2, pp. 865-
881, 2009.
[66]
F. Tansu and M. Salamah, "Vertical Handoff Decision Schemes
for Heteregeneous Wireless Networks: An Overview," Recent
Trends in Wireless and Mobile Networks, vol. 84, pp. 338-348,
2010.
[67]
K. P. Yoon and C. L. Hwang, Multiple attribute decision making:
an introduction.: Sage publications, 1995.
[68]
Y. Kuo, T. Yang, and G. W. Huang, "The use of grey relational
analysis in solving multiple attribute decision-making problems,"
Computers & Industrial Engineering, vol. 55, no. 1, pp. 80-93,
2008.
[69]
T. L. Saaty, The Analytic Hierarchy Process: Planning, Priority
Setting, Resource Allocation. New York: McGraw-Hill, 1980.
[70]
R. Chai, W. G. Zhou, Q. B. Chen, and L. Tang, "A survey on
vertical handoff decision for heterogeneous wireless networks," in
Information, Computing and Telecommunication, YC-ICT'09,
IEEE Youth Conference, 2009, pp. 279-282.
[71]
E. Stevens-Navarro and V. Wong, "Comparison between Vertical
Handoff Decision Algorithms for Heterogeneous Wireless
Networks," in IEEE Vehicular Technology Conference, 2006, pp.
947-951.
[72]
E. Stevens-Navarro, Y. Lin, and V. W. Wong, "An MDP-Based
Vertical Handoff Decision Algorithm for Heterogeneous Wireless
Networks," IEEE Transactions on Vehicular Technology, vol. 57,
no. 2, pp. 1243-1254, 2008.
[73]
S. A. Sharna, M. R. Amin, and M. Murshed, "An Enhanced-MDP
Based Vertical Handoff Algorithm for QoS Support over

Heterogeneous Wireless Networks," in IEEE International
Symposium on Network Computing and Applications, 2011, pp.
289-293.
[74]
Y. Wang, P. Zhang, Y. Zhou, J. Yuan, F. Liu, and G. Li,
"Handover Management in Enhanced MIH Framework for
Heterogeneous Wireless Networks Environment," Wireless
Personal Communications, vol. 52, no. 3, pp. 615-636, 2008.
[75]
L. Ni, Y. Zhu, N. Li, and Q. Deng, "Optimal Mobility-aware
Handoff in Mobile Environments," in IEEE 17th International
Conference on Parallel and Distributed Systems, 2011, pp. 534-
540.
[76]
J. Baliosian and J. Serrat, "Finite state transducers for policy
evaluation and conflict resolution," in Proceedings of the Fifth
IEEE International Workshop on Policies for Distributed Systems
and Networks (POLICY 2004), 2004, pp. 250-259.
[77]
M. Rafiq, S. Kumar, N. Kammar, G. Prasad, G. K. S. Garge, S. V.
R. Anand, and M. Hegde, "A Vertical Handoff Decision Scheme
for End-to-End QoS in Heterogeneous Networks: An
Implementation on a Mobile IP Testbed," in IEEE National
Conference on Communications (NCC), 2011, pp. 1-5.
[78]
X. Haibo, T. Hui, and Z. Ping, "A novel terminal-controlled
handover scheme in heterogeneous wireless networks,"
Computers & Electrical Engineering, vol. 36, no. 2, pp. 269-279,
2010.
[79]
K. Vasu, S. Mahapatra, and C. S. Kumar, "QoS Aware Fuzzy
Rule Based Vertical Handoff Decision Algorithm for Wireless
Heterogeneous Networks," in IEEE National Conference on
Communications (NCC), 2011, pp. 1-5.
[80]
K. Pahlavan, P. Krishnamurthy, A. Hatami, M. Ylianttila, J. P.
Makela, R. Pichna, and J. Vallstron, "Handoff in hybrid mobile
data networks," IEEE Personal Communications Mag., vol. 7, no.
2, pp. 34-47, 2000.
[81]
N. Nasser, S. Guizani, and E. Al-Masri, "Middleware vertical
handoff manager: a neural network-based solution," in
Proceedings of the 2007 IEEE International Conference on
Communications (ICC’07), 2007, pp. 5671-5676.
[82]
I. Lassoued, J. M. Bonnin, and A. Belghith, "Towards an
Architecture for Mobility Management and Resource Control," in
Wireless Communications and Networking Conference IEEE
Proceedings, 2008, pp. 2846-2851.
[83]
F. Giust, L. Cominardi, and C. J. Bernardos, "Distributed mobility
management for future 5G networks: overview and analysis of
existing approaches.," IEEE Communications Magazine, vol. 53,
no. 1, pp. 142-149, 2015.
[84]
Q. T. Nguyen-Vuong, N. Agoulmine, and Y. Ghamri-Douda,
"Terminal-Controlled Mobility Management in Heterogeneous
Wireless Networks," IEEE Communications Magazine, vol. 45,
no. 4, 2007.
[85]
Y. Chen, J. Hsia, and Y. Liao, "Advanced seamless vertical
handoff architecture for WiMAX and WiFi heterogeneous
networks with QoS guarantees," Computer Communications, vol.
32, no. 2, pp. 281-293, 2009.
[86]
G. Aristomenopoulos, T. Kastrinogiannis, Z. Li, M. Wilson, J. M.
González, J. A. Lozano-López, Y. Li, V. Kaldanis, and S.
Papavassiliou, "Autonomic Mobility and Resource Management
Over an Integrated Wireless Environment – A GANA Oriented
Architecture," in IEEE International Workshop on Management
of Emerging Networks and Services, 2010, pp. pp. 545-550.
[87]
"IEEE Standard for Local and metropolitan area networks-Part
21: Media Independent Services Framework," IEEE Std. 802.21-
2017, April 2017.
[88]
"IEEE Standard for Local and metropolitan area networks-Part
21.1: Media Independent Services," IEEE Std. 802.21.1 -2017,
April 2017.
[89]
"Access Network Discovery and Selection Function (ANDSF)
Management Objects (MO)," 3GPP TS 24.312, V14.1.0 Release
14, June 2017.
[90]
B. S. Ghahfarokhi and N. Movahhedinia, "Context-Aware
Handover Decision in an Enhanced Media Independent Handover
Framework," Wireless personal communications, vol. 68, no. 4,
pp. 1633-1671, 2013.
[91]
P. Neves, J. Soares, S. Sargento, H. Pires, and F. Fontes,
"Context-aware media independent information server for
optimized seamless handover procedures," Computer Networks,
vol. 55, no. 7, pp. 1498-1519, 2011.
[92]
S. Barmpounakis, A. Kaloxylos, P. Spapis, and N. Alonistioti,
"Context-aware, user-driven, network-controlled RAT selection
for 5G networks," Computer Networks, vol. 113, pp. 124-147,
2017.
[93]
4G Americas, "Integration of Cellular and Wi-Fi Networks ,"
White Paper 2013.
[94]
B. Orlandi and F. Scahill, "Wi-Fi Roaming–Building on ANDSF
and Hotspot 2.0," Alcatel-Lucent, Boulogne-Billancourt, France,
Tech. Rep. 2012.
[95]
"Hotspot 2.0 (Release 2) Technical Specification," Wi-Fi Alliance
Technical Committee, Hotspot 2.0 Technical Task Group,.
[96]
Y. Xu, J. Song, J. Li, G.X. Kok, and K. Utsu, "Load-Aware Based
Independent Information Services For Heterogeneous Handover,"
International Journal of Innovative Computing, Information and
Control, vol. 12, no. 1, pp. 243-262, 2016.
[97]
N. Dimitriou, L. Sarakis, D. Loukatos, G. Kormentzas, and C.
Skianis, "Vertical handover (VHO) framework for future
collaborative wireless networks," International Journal of
Network Management, vol. 21, no. 6, pp. 548-564, 2011.
[98]
A. Stamou, N. Dimitriou, K. Kontovasilis, and S. Papavassiliou,
"Delay Analysis of Context Aware Mobility Management
Systems Addressing Multiple Connectivity Opportunities," Ad-
hoc, Mobile, and Wireless Networks, ADHOC-NOW 2015,
Lecture Notes in Computer Science, vol. 9143, pp. 121-133, 2015.
[99]
S. Abolfazli, Z. Sanaei, E. Ahmed, A. Gani, and R. Buyya,
"Cloud-based augmentation for mobile devices: motivation,
taxonomies, and open challenges," IEEE Communications
Surveys & Tutorials, vol. 16, no. 1, pp. 337-368, 2014.
[100]
A. G. Ganek and T. H. Corbi, "The dawning of the autonomic
computing era," IBM Systems Journal, vol. 42, no. 1, pp. 5-18,
2003.
[101]
R. Want, T. Pering, and D. Tennenhouse, "Comparing autonomic
and proactive computing," IBM SYSTEMS JOURNAL, vol. 42, no.
1, pp. 129-135, 2003.
[102]
J. Puttonen, G. Fekete, J. Makela, T. Hamalainen, and J. Narikka,
"Using Link Layer Information for Improving Vertical
Handovers," in IEEE 16th International Symposium on Personal,
Indoor and Mobile Radio Communications, 2005, pp. 1747-1752.
[103]
T. J. Stewart, "Dealing with Uncertainties in MCDA," in Multiple
criteria decision analysis: state of the art surveys.: Springer New
York, 2005, vol. 78, ch. V, pp. 445-466.
[104]
K. Steele, Y. Carmel, J. Cross, and C. Wilcox, "Uses and Misuses
of Multi-Criteria Decision Analysis (MCDA) in Environmental
Decision-Making," Risk analysis, vol. 29, no. 1, pp. 26-33, 2008.
[105]
A. Anselin and P. M. Meire, "Multicriteria Techniques in
Ecological Evaluation: an Example Using the Analytic Hierarchy
Process," Biological Conservation, vol. 49, pp. 215–229, 1989.
[106]
G. Herath, "Incorporating community objectives in improved
wetland management: the use of analytic hierarchy process,"
Journal of Environmental Management, vol. 70, no. 3, pp. 263-
273, 2004.
[107]
M. Kassar, B. Kervella, and G. Pujolle, "Autonomic-oriented
Architecture for an Intelligent Handover Management Scheme,"
in IEEE Communication Networks and Services Research

Conference, 2008, pp. 139-146.
[108]
E. H. Mamdani, "Application of fuzzy algorithms for control of
simple dynamic plant," Proceedings of IEEE, vol. 121, 1974.
[109]
S. Maaloul, M. Afif, and S. Tabbane, "New method of handling
weights for a reliability handover decision," in International
Conference on Multimedia Computing and Systems (ICMCS),
2014.
[110]
R. A. Ribeiro, "Fuzzy multiple attribute decision making: A
review and new preference elicitation techniques," Fuzzy Sets and
Systems, vol. 78, no. 2, pp. 155-181, 1996.
[111]
G.A. Di Caro, S. Giordano, M. Kulig, D. Lenzarini, A. Puiatti,
and F. Schwitter, "A cross-layering and autonomic approach to
optimized seamless handover," in Proc. of the 3rd Annual
Conference on Wireless On Demand Network Systems and
Services, 2006, pp. 104-113.
[112]
G. Edwards, A. Kandel, and R. Sankar, "Fuzzy handoff
algorithms for wireless communication," Fuzzy Sets and Systems,
vol. 110, 2000.
[113]
N. D. Tripathi, J. H. Reed, and H. VanLandingham, "An adaptive
direction biased fuzzy handoff algorithm with unified handoff
candidate selection criterion," in in Proc. 48th IEEE Veh.
Technology Conf., 1998.
[114]
S. Barmpounakis, A. Kaloxylos, P. Spapis, and N. Alonistioti,
"COmpAsS: A Context-Aware, User-Oriented Radio Access
Technology Selection Mechanism in Heterogeneous Wireless
Networks," in MOBILITY 2014 : The Fourth International
Conference on Mobile Services, Resources, and Users, 2014.
[115]
T. Ahmed, K. Kyamakya, M. Ludwig, K. R. Anne, J. Schroeder,
S. Galler, and K. Jobmann, "A Context-Aware Vertical Handover
Decision Algorithm for Multimode Mobile Terminals and its
Performance," in Proceedings of the IEEE/ACM Euro American
Conference on Telematics and Information Systems (EATIS
2006), 2006, pp. 19-28.
Adamantia Stamou is a PhD Candidate at NETwork Management and
Optimal DEsign Laboratory (NETMODE) Lab, NTUA Greece, in
collaboration with Networks Laboratory (NeL), NCSR “Demokritos”
Greece, since 2012. She received her diploma (Master equivalent) from
the Department of Informatics and Telecommunications Engineering at
the University of Western Macedonia, Greece, in 2011. Also, she
received an MBA under full honorary scholarship from Glyndwr
University, UK, in 2014. Adamantia gained the Best Paper Award for
the paper based on her diploma thesis “Energy-Efficiency Evaluation of
a Medium Access Control Protocol for Cooperative ARQ” at IEEE ICC
2011. Her research interests include mobile and wireless
communications; context-aware and autonomic network management;
cross-layer optimization and energy efficiency. Adamantia Stamou is a
student member of the IEEE and of the Technical Chamber of Greece.
Nikos Dimitriou (SM.IEEE, 2011) holds a Diploma in Electrical &
Computer Engineering from the National Technical University of
Athens, Greece (1996), an M.Sc. with distinction in Mobile and Satellite
communications (1997) from the University of Surrey, UK and a Ph.D.
in Mobile Communications from the same university (2001). Since then
he has been actively involved in numerous research projects in the area
of Wireless Communications via his affiliations with the Institute of
Accelerating Systems and Applications at the National Kapodistrian
University of Athens, with the Institute of Informatics and
Telecommunications at the National Center of Scientific Research
"Demokritos" in Greece, with European Dynamics S.A. in Belgium and
with the Computer Science Department at the King Abdulaziz
University in Saudi Arabia. His current research interests include Radio
Resource Management for multi-tier HetNets, Energy Efficient Mobility
Management in Composite Wireless Networks and Robust Routing for
Mobile-Airborne Ad-Hoc Networks.
Kimon Kontovasilis holds a Diploma (1987) and a Ph.D. (1993) in
Electrical Engineering from the National Technical University of
Athens and the M.Sc. in Computer Science (1990) from North Carolina
State University. He is with the Institute of Informatics and
Telecommunications at the National Center for Scientific Research
“Demokritos” as a member of the research staff,, currently ranking as a
Research Director and serving as the Head of the Networks Laboratory
(NeL). His research interests span several areas of networking,
including: modeling, performance evaluation and resource management
of wired and wireless networks; management and optimization of
heterogeneous wireless networks; mobile ad hoc and delay-tolerant
networks; and energy efficient networks. He has taken part in a
considerable number of research projects at a national and international
scale (many times in a leading/organizational role) and he serves
regularly in the organizing and/or technical program committees of
international conferences in the field of networking. He is a member of
IFIP WG6.3.
Symeon Papavassiliou is a Professor with the Faculty of Electrical and
Computer Engineering Department, NTUA. He received the Diploma in
Electrical Engineering from NTUA in 1990 and the M.Sc. and Ph.D.
degrees in Electrical Engineering from Polytechnic University,
Brooklyn, New York in 1992 and 1995 respectively. From 1999 till 2004
he was with the Electrical and Computer Engineering Department at the
New Jersey Institute of Technology (NJIT), USA, where he was an
Associate Professor. From 1995 until 1999 he was a Senior Technical
Staff Member at AT&T Laboratories, New Jersey-USA. Dr.
Papavassiliou was the Director of the Broadband, Mobile and Wireless
Networking Laboratory (2000-2004) at the New Jersey Institute of
Technology, USA and a founding member and Associate Director of the
New Jersey Center for Wireless Networking and Internet security (2002-
2004, New Jersey, USA). He has an established record of publications in
his field of expertise, with more than 300 technical journal and
conference published papers. He received the Best Paper Award in IEEE
INFOCOM‟94, the AT&T Division Recognition and Achievement
Award in 1997, the US National Science Foundation Career Award in
2003, the Best Paper Award in IEEE WCNC 2012, the Excellence in
Research Grant in Greece in 2012, the Best Paper Awards in
ADHOCNETS 2015 and in ICT 2016. Dr. Papavassiliou also served on
the board of the Greek National Regulatory Authority on
Telecommunications and Posts (2006–2009). His main research interests
lie in the area of communication networks, with emphasis on the
analysis, optimization and performance evaluation of mobile and
distributed systems, wireless networks and complex systems. He has
been Associate Editor for IEEE Transactions on Parallel and Distributed
Systems (2010-2014) and Technical Editor for IEEE Wireless
Communications Magazine (2011-2015), Editor of IEEE
Communications Letters (2016-now), Member of the Editorial Board of
MDPI Sensor Journal (2016-now), Member of the Editorial Board of
MDPI Future Internet Journal and Section Editor in Chief for Internet of
Things Section (2017-now).