![Relationship diagram for mobility related KPIs and their interplay with the associated network parameters (grouped in different colors) Source: [17], [18].](https://www.researchgate.net/profile/Marvin-Manalastas/publication/346919085/figure/fig3/AS:1002227791564801@1615961134628/Relationship-diagram-for-mobility-related-KPIs-and-their-interplay-with-the-associated_Q320.jpg)
Content uploaded by Marvin Manalastas
Author content
All content in this area was uploaded by Marvin Manalastas on Mar 24, 2021
Content may be subject to copyright.
Available via license: CC BY 4.0
Content may be subject to copyright.
Received August 26, 2020, accepted September 9, 2020, date of publication September 28, 2020, date of current version October 19, 2020.
Digital Object Identifier 10.1109/ACCESS.2020.3027258
Mobility Management in Emerging Ultra-Dense
Cellular Networks: A Survey, Outlook,
and Future Research Directions
SYED MUHAMMAD ASAD ZAIDI 1, (Graduate Student Member, IEEE),
MARVIN MANALASTAS1, (Member, IEEE), HASAN FAROOQ 2, (Member, IEEE),
AND ALI IMRAN 1, (Senior Member, IEEE)
1Department of Electrical and Computer Engineering, The University of Oklahoma-Tulsa, Tulsa, OK 74135, USA
2Ericsson Research, Silicon Valley, CA, USA
Corresponding author: Syed Muhammad Asad Zaidi (asad@ou.edu)
This work was supported in part by the National Science Foundation under Grant 1718956 and Grant 1559483, and in part by the Qatar
National Research Fund (QNRF) under Grant NPRP12-S 0311-190302.
ABSTRACT The exponential rise in mobile traffic originating from mobile devices highlights the need
for making mobility management in future networks even more efficient and seamless than ever before.
Ultra-Dense Cellular Network vision consisting of cells of varying sizes with conventional and mmWave
bands is being perceived as the panacea for the eminent capacity crunch. However, mobility challenges
in an ultra-dense heterogeneous network with motley of high frequency and mmWave band cells will
be unprecedented due to plurality of handover instances, and the resulting signaling overhead and data
interruptions for miscellany of devices. Similarly, issues like user tracking and cell discovery for mmWave
with narrow beams need to be addressed before the ambitious gains of emerging mobile networks can be
realized. Mobility challenges are further highlighted when considering the 5G deliverables of multi-Gbps
wireless connectivity, <1ms latency and support for devices moving at maximum speed of 500km/h, to name
a few. Despite its significance, few mobility surveys exist with the majority focused on adhoc networks.
This paper is the first to provide a comprehensive survey on the panorama of mobility challenges in the
emerging ultra-dense mobile networks. We not only present a detailed tutorial on 5G mobility approaches
and highlight key mobility risks of legacy networks, but also review key findings from recent studies and
highlight the technical challenges and potential opportunities related to mobility from the perspective of
emerging ultra-dense cellular networks.
INDEX TERMS 5G cellular networks, network densification, mobility prediction, mmWave band, reliabil-
ity, latency, multi-connectivity, user tracking, cell discovery, energy efficiency.
I. INTRODUCTION
The unprecedented rise in the Internet traffic volume seen
in recent years is attributed to high speed internet, and the
advent of smart phone technology. It is anticipated that
the number of 5G subscriptions will be 2.8 billion by the
year 2025 [1]. Furthermore, the insatiable demand for new
bandwidth-hungry applications will lead to an avalanche
of traffic volume growth. Mobile data traffic will increase
from 10.7 exabytes/month in 2016 to 83.6 exabytes/month
by 2021 [2], and that number will further increase exponen-
tially in the years to follow.
The associate editor coordinating the review of this manuscript and
approving it for publication was Fan-Hsun Tseng .
The emerging cellular networks including 5G mobile net-
work standard as the next revolution of mobile cellular tech-
nology needs to support the ever-increasing mobile users,
provide adequate data rate for the bandwidth hungry appli-
cations, address the QoS issues of delay tolerant applications
and realize the concept of Internet-of-Things (IoT) [3], [4].
5G promises to deliver ‘‘more’’ of everything [5]: a) top
speeds of up to 1 Gbps, b) 100 Mbps data rate per end
user even at the cell edge, c) RTT (Round-Trip-Time) laten-
cies in the millisecond range, d) higher connection densities
(1 million connections per km2[6]), and e) support for mobile
devices at the speed of up to 500 km/h.
Currently, Signal to Interference and Noise Ratio (SINR) is
considered as the primary metric for planning, dimensioning
VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ 183505
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
and optimization of the existing cellular networks [3]. How-
ever, for a few exceptions like fixed IoT services, an addi-
tional network planning/design criterion in the future may be
the mobility related QoE. This is likely the outlook in the
backdrop of the following observations:
1) Coverage and SINR provisioning will become a rela-
tively easy challenge given the anticipated higher Base
Station BS) density in emerging cellular networks,
along with the sophisticated interference management
schemes and massive MIMO assisted beamforming.
2) However, the very same advances in the network design
i.e. densification, beamforming, massive MIMO make
the mobility management a more challenging problem.
The challenges stem not only from the increased num-
ber of handovers HOs) but also, beam management to
maintain the expected QoE. Challenges related to beam
management includes focusing narrow beams on the
mobile users, cell discovery in narrow beam cells, and
large signaling overheads when the user moves from one
massive MIMO cell to another cell.
3) With the advent of mmWave, narrow beams of mmWave
bands will have limited overlap with each other, mak-
ing HO a challenging problem (see Fig.4 for observing
the difference in HO scenarios in low frequencies and
mmWave frequencies).
The growing demand for mobile services in public trans-
port, highways, open-air gatherings etc. [7] will be critical
to customer experience. Providing a satisfactory Quality of
Experience (QoE) to a relatively large number of mobile users
and a miscellany of the devices including phones, tablets,
sensors etc. at the speed up to 500km/h imposes extreme
challenges to the future mobile networks. Mobility require-
ments in emerging cellular networks require high efficiency
of the HO mechanism, which makes the cell-change seamless
for the users. Unlike the legacy technologies (i.e. 3G and
4G) that do not give primary importance to high mobility,
future mobile networks will treat mobility as an integral
part of the communication standard. Moreover, the mobility
management schemes in Long Term Evolution (LTE) systems
(also known as 4G system) and to a certain extent, even in the
latest 5G New Radio (NR) standard are not well adapted to
the typical deployment of the futuristic mobile networks due
to multiple factors, few of which are highlighted below:
•The legacy LTE architecture makes use of a centralized
network control entity called MME (Mobility Manage-
ment Entity) located in the core network. The emerging
cellular networks are expected to have 10-folds higher
density [8], with a larger fraction of mobile users. Thus,
without a mobility centric redesign of the architecture,
future networks should have 10 times more MME’s just
to achieve a similar QoS as in LTE.
•To achieve the logistic feasibility for high density
deployment, BS placement in future mobile networks
are likely to be impromptu or much less planned [8].
This will increase mobility related signaling load that is
bound to complicate the core network management and
planning.
•HO decision in existing networks is made by partici-
pating BSs without considering the deployment of the
BSs and backhaul limitations. In futuristic mobile net-
works with flexible BS deployment, the chances of User
Equipment (UE) in selecting the optimal target BS may
become smaller.
•While the capacity crunch will be addressed by
small-cells (SC), a large number of inter-SC HOs will
take place leading to frequent session interruptions
during HO.
•With smaller inter-site-distance as in SCs, the perfor-
mance of the existing mobile network reduces sharply
owing to the risk of HO failures due to high radio link
variability as shown in [9].
•In existing mobile networks, UE context has to move
from one BS to another for every HO. This will
impose unprecedented signaling overhead in the future
ultra-dense network architecture. While signaling is
already growing 50% faster than data traffic [10], net-
work efficiency will drop by many folds using the cur-
rent HO approaches.
•HOs in 4G networks are based on the broadcast sig-
nal called Reference Signal (RS). The mmWaves with
narrow beams cannot have RS broadcast to the whole
coverage area within the cell range. Hence, cell discov-
ery, especially for mobile UEs is another key mobility
challenge in emerging cellular networks not faced by the
traditional mobile networks.
•With SON stepping up the automatization of network
configuration and optimization in LTE, myriad of mobil-
ity management parameters associated with the large
number of closely deployed 5G BSs need to be well
managed. For that, the existing SON solutions will not
be sufficient.
•5G applications with Ultra Reliable Low Latency Com-
munications (URLLC) e.g. self-driven cars demand very
low latency requirements as shown in Table 1 [11].
•When UE perform HO to a better cell, it experiences a
latency and data interruption period. HO management in
the future mobile networks should ensure a seamless and
latency-free transition from the source to the target cell.
•With mobile phone traffic on the rise, and with the
advent of self-driven cars and drones needing robust
connectivity, seamless and reliable mobility manage-
ment has become more significant than ever. The adapta-
tion of ultra-dense cellular networks and mmWave BSs
makes the mobility management even more complex
challenge requiring significant research effort.
In light of the above discussion, we can conclude that mobil-
ity management will have much stronger impact on the design
and architecture of upcoming cellular networks, than it had on
the legacy networks. The futuristic networks will incorporate
high mobility requirements as an integral part, and apprecia-
ble efforts are required to attain ubiquitous top-notch QoE.
183506 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
TABLE 1. Comparison of LTE latency with 5G expected goals.
Majority of mobility oriented surveys in the literature tar-
get adhoc networks [12]–[14]. Mobility surveys on cellular
networks do exists e.g. Xenakis et al. [15] presented survey on
HO decision algorithms for the femtocells in LTE-Advance.
Another survey on high mobility wireless communication has
recently been presented in [16], however, the attributes and
intricacies of the 5G architecture have not been addressed.
To the best of the authors’ knowledge, this survey is the first
to address the novel contributions by research community
targeting mobility in emerging ultra-dense mobile networks.
The contributions in this paper and its organization are as
follows:
•To the best of the authors’ knowledge, this paper
gives the first comprehensive tutorial on 3GPP based
5G mobility management procedures for both a)
idle/inactive mode, and b) connected mode mobile users.
•Mobility related surveys do exist in the literature
(e.g. [12]–[14] on adhoc networks), but none of the
aforementioned surveys addresses the futuristic mobile
networks. This paper presents a single go-to manuscript
where future researchers not only understand the 3GPP
mobility procedure and the existing mobility related
literature but also assist them in finding the research
directions they might undertake.
•It presents a first of its kind framework to correlate
all mobility management related parameters with all
mobility management related KPIs. To facilitate easy
understanding, this framework is presented in the form
of a flow chart shown in Fig. 8.
•It presents a comprehensive and taxonomized review of
the literature on mobility management.
•It identifies the need for a new paradigm for mobility
management deemed essential to meet the quality of
experience (QoE) requirements of the emerging appli-
cations and use-cases.
•It proposes a novel proactive mobility management
framework to meet the requirements of the emerging
mobile networks. Since the challenges of 5G networks
(e.g. low latency, less overhead and high quality of
experience) cannot be addressed by the current reac-
tive mobility management techniques, we discussed the
proactive mobility management in section IV.
•It highlights the need to come up with Mobility oriented
Network planning and dimensioning
•It provides a collection of the latest AI-based techniques
to smartly address mobility related challenges.
•It identifies the future research direction and few open
research problems to achieve this paradigm shift.
Fig. 1 outlines the structure of the paper. It also provides a
taxonomy of the literature on mobility.
II. UNDERSTANDING MOBILITY IN CELLULAR
NETWORKS
Mobility in cellular networks plays a pivotal role ensuring
an optimal experience to the subscribers. It guarantees that
mobile users won’t just be able to maintain connectivity
but attain the best available connection to the network as
they move towards the destination. Seamless and timely HO
and cell reselection has always been a major challenge in
any wireless communication systems including 5G. Mobility
has been categorized as Idle and Connected Mode Mobility
in 5G. Note that the mobility procedure in LTE (4G) is very
similar in 5G New Radio (NR) using events A1, A2, A3, A4,
A5 and A6 to trigger HOs. Event A2 and A1 are triggered
when RF condition of the UE falls below and exceeds the
configured threshold respectively and are used to start and
stop inter-frequency neighbor search. Intra-frequency HO
is initiated by event A3 where the neighbor RF condition
becomes higher than serving RF condition by a configured
threshold. Event A4 and A5 are typically used for inter-
frequency HO where target inter-frequency cell has to be
higher than an absolute threshold for the event A4 to be
triggered. On the contrary, event A5 in addition to event
A4 condition, requires serving cell RF condition to be below a
certain threshold. Finally, event A6 is similar to event A3 but
is used for intra-frequency HO of the secondary frequency
the UE is camped onto. Event A4 and A5 can also be used
for conditional HO management for e.g. for load balancing.
In addition to the events described above, event B1 and B2
(A4 and A5 alike) are also used for inter-technology HO, and
for dual-connectivity, but they are not discussed here to keep
the focus of this paper confined to basic mobility procedures
and the associated challenges.
The only difference between 5G and 4G mobility crite-
ria is in the idle mode where respective idle mode rese-
lection parameters in 5G NR are present in different SIB#
than in LTE. Moreover, the idle mode parameter names and
functionalities in 5G are similar as in 4G. Comprehensive
explanation of 5G mobility procedure while keeping in view
the 5G network architecture and interfaces is presented in the
following subsections.
A. IDLE MODE MOBILITY
UE is in idle mode when it is neither running any active
communication service nor is connected to any particular cell.
UE in idle mode is constantly trying to search and maintain
services such as Public Land Mobile Network selection, cell
selection and reselection, location registration, and reception
of system information. By maintaining an idle mode connec-
tion, UE can readily establish a Radio Resource Connection
(RRC) for signaling or data transfer as well as be able to
receive any possible incoming connections.
UE always powers ON in idle mode and selects the cell
with the maximum signal strength through a process known
VOLUME 8, 2020 183507
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
FIGURE 1. Layout of the contents and paper contributions.
as cell selection. However, this initially selected cell will not
always be the best to serve especially when UE moves from
one place to another. Therefore, to maintain the quality of
signal, UE has to camp on another optimal cell, a process
known as cell reselection.
1) CELL RESELECTION CRITERIA
In 5G, BS broadcasts nine System Information Block (SIB)
messages for the UE as defined in 3GPP [17]. Out of those
messages, SIB 1, 2, 3 and 4 contain critical parameters to
execute idle mode cell reselection to the optimal 5G cell.
SIB1 has the serving cell parameters as well as the cell
selection parameters, while SIB2 has the common parame-
ters used for intra-frequency and inter-frequency reselection.
SIB3 is dedicated to intra-frequency reselection parameters,
however, operators can broadcast the related parameters in
SIB2 instead, and thus SIB3 is not broadcasted. SIB4 contains
inter-frequency reselection through target frequency priority
and the associated parameters. Fig. 3 illustrates a pictorial
demonstration of the reselection conditions and evaluation
in 5G as described by 3GPP. Description of the related res-
election parameter, and the respective location (SIB#) can be
found in Table 3. LTE uses the same reselection procedure
with the only difference that the contents of SIB2, SIB3 and
SIB4 in 5G are found in SIB3, SIB4 and SIB5 of LTE instead.
2) USER TRACKING
The idle mode mobility of the UE is the responsibility
of Access and Mobility Function (AMF) at the Tracking
Area (TA) level for RRC idle mode users and at the RAN
Notification Area (RNA) for RRC inactive mode users. Here
we only talk about the idle mode users as the mobility pro-
cedure in 5G is similar for RRC idle mode and RRC inactive
mode users. Note that unlike the connected mode, network is
unaware of cell-level UE location in idle mode. After power-
ing ON, UE acquires the Tracking Area List (TAL) composed
of a list of TA codes through the periodic SIB1 broadcast
from the cell. As UE traverses through the network while
183508 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
TABLE 2. List of acronyms.
performing cell reselection procedure, it compares the TA
code of the new cell with its own TAL. If the TA code of a
newly visited cell does not match with its own TAL, it initiates
TA Update (TAU) process to request AMF for location update
as seen in the Fig. 3(a). TAU helps to track the UE in case of
any incoming call. Bigger TA size reduces signaling overhead
of TAU at the expense of larger paging domain, ultimately
resulting in higher paging-based downlink signaling load at
network level.
TABLE 3. 3GPP [17] intra/inter-frequency reselection parameters.
3) COMMON IDLE MODE MOBILITY RISKS
In this subsection, we discuss about the common idle mode
mobility risks in the existing LTE network. But since the
mobility process is similar in 5G networks, 5G capable UEs
are expected to face similar challenges.
In idle mode, data transmission does not take place, there-
fore reliability and QoS are not the issues of concern. How-
ever, reselection procedure can incur accessibility and user
tracking issues in rare occasions.
During the network attach procedure, idle mode UE first
sends connection request and awaits connection setup mes-
sage from the BS. If UE does not receive any message from
the BS within a predefined time (t300 timer known to UE
VOLUME 8, 2020 183509
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
FIGURE 2. 3GPP [17] cell reselection criteria based on SIB3 and
SIB5 parameter for intra-frequency and inter-frequency reselection
respectively.
FIGURE 3. (a) Tracking Area Update (TAU) procedure in LTE networks,
(b) Common Tracking Area (TA) planning approaches.
via SIB2 ‘SIB1 in 5G [18]’), it restarts the accessibility
procedure. Under special circumstances, if UE sends a con-
nection request to the serving cell followed by reselection to a
neighboring cell, it cannot receive the connection grant simul-
taneously. The new serving cell in this case does not become
aware that the UE which just moved under its coverage needs
to access the network. Thus, UE has to wait for a time defined
in t300 before re-initiating the access procedure in the new
serving cell. During this time, UE experiences latency and can
have serious impact on the applications requiring ultra-low
latency. The delay can be suppressed by having smaller t300
timer, but at the cost of increased signaling load due to the
increase in redundant connection requests and replies. More-
over, smaller t300 also negatively impact UE energy con-
sumption (due to recurrent Random-Access Channel ‘RACH’
attempts). Repeated RACH attempts might result in higher
Central Processing Unit (CPU) load of serving cell, especially
at busy hour.
Similar accessibility delay at TA border can result in pag-
ing failure, since the network can be unaware of the accurate
UE location unless TAU followed by a successful accessibil-
ity is performed.
TA planning is a crucial task and two approaches are
used in existing networks: a) horizontal approach, b) vertical
approach, as shown in Fig. 3(b). TAU procedure initiates for
every inter-frequency reselection in horizontal approach, thus
it is deployed where radio condition is good, and user is
least expected to make recurrent inter-frequency reselection.
On the contrary, poor radio condition area should have verti-
cal approach to minimize TAU for inter-frequency reselection
instances. Horizontal approach is favorable for high speed
traffic like train lines or highways. One approach to address
this issue in the existing cellular network is the use of adaptive
TA codes, where users are configured with a list of TA
codes to prevent ping-pong TAUs. However, determining the
optimal number of TA codes in a list and the cumulative TA
size still remain an open research problem.
B. CONNECTED MODE MOBILITY
UE is said to be in connected mode when it has established a
connection with its peer Radio Resource Control (RRC) layer
at the serving BS and the network can transmit and/or receive
data to/from the UE. As there is an exchange of data between
the UE and the BS, uninterrupted data transfer needs to take
place for a seamless continuity of service when a UE moves
from one BS to another BS. This ideally seamless mobility in
connected mode is termed as handover (HO).
1) UE SIDE MOBILITY TRIGGER
UE triggers an intra-frequency HO request to the next optimal
cell by sending A3-Measurement Report (MR) to its serving
cell as shown in Fig. 4. The serving cell then decides whether
to entertain the request and perform the HO, by communicat-
ing with the target cell and serving AMF. An intra-frequency
HO is the first preference in cellular networks; however, there
are instances in which an inter-frequency HO is the preferred
choice. For example: a) when there is a coverage hole in
the serving frequency, b) when the current serving cell does
not support the requested service e.g. Voice over NR, and
c) when load balancing is needed to avoid congestion in
the serving frequency. In Fig. 5 we illustrate the 3GPP [18]
defined inter-frequency HO criteria. For a description of each
HO parameter, refer to Table 4.
2) NETWORK SIDE MOBILITY TRIGGER
HOs are undoubtedly more complicated than cell reselection.
Aside from the source and target cell, core entities which
183510 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
FIGURE 4. General HO procedure. (a) UE performs HO from cell A to cell
B at cell-edge as it moves closer to the cell B. Scenario 1 and 2 represents
HF coverage and mmWave narrow beams, (b) 3GPP [18] based
intra-frequency HO process.
FIGURE 5. 3GPP [18] intra-frequency and inter-frequency handover
criteria in LTE networks.
include Access and Mobility Function (AMF), Session Man-
agement Function (SMF) and User Plane Function (UPF)
need to be updated as well. Depending on the scenario, data
transfer and handling could pose several challenges. In nor-
mal cases, when AMF, SMF and UPF do not change during
the HO, signaling is reasonable and it is termed Xn based
HO. Here, the Xn interface is used for the preparation phase
of the HO. However, when the Xn interface does not exist
between the participating cells, an N2 based HO is performed
where cells use a longer path for communication. Signaling
flow for the Xn based HO is illustrated in Fig. 6. 3GPP [18]
named Xn as the interface used to connect 5G BSs directly,
and N2 interface is the logical interface between two 5G BSs
connected through the core network (AMF). N2 interface is
FIGURE 6. Xn based handover without UPF re-allocation in 5G networks.
TABLE 4. 3GPP [18] handover parameters conveyed to UE in Rrc
reconfiguration layer 3 message.
used if the direct Xn interface between the neighboring BSs
do not exists.
3) COMMON CONNECTED MODE MOBILITY RISKS
Apart from the fast fading effect due to Doppler shift in
physical layer, the mobile UE has to cope with several
Layer 3 issues as well, which can be eluded primarily by
a timely HO and an optimal selection of the target bs.
Some of the issues mobile UE experiences during inter-site
mobility are presented in Fig. 7, with possible solution(s)
in Table 5.
VOLUME 8, 2020 183511
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
FIGURE 7. Common mobility related risks in 4G/5G networks.
TABLE 5. Common HO issues and their solutions.
C. INTERPLAY BETWEEN MOBILITY KPI AND KEY
PARAMETERS
Network operators optimize their network by tuning a
set of mobility related parameters, and then by observ-
ing the HO attempt, HO success and few other QoE
KPIs affected by those modified network parameters.
Few of the vital mobility related KPIs are outlined
below:
•User tracking KPI indicates the paging hit rate when
users served under the TA are notified by an incoming
call. The idle mode mobile user must update its loca-
tion (via TAU) to the core network when it moves into
the neighboring TA. By doing so, the respective TA is
broadcasted with paging attempt messages in case of
any incoming call. A delay in TAU can result in paging
failure and reattempts.
•Mobility oriented HO process or TAU trigger results in
the control plane messages being sent in the air interface
and in the core network. The percentage of network
resources used by control plane are measured by signal-
ing data KPI.
•User terminal energy consumption e.g. during data
delivery and location update, can be measured by the UE
battery KPI.
•Reliability (or retainability) KPI indicates the percent-
age of users that dropped the connection with their
participating cells during the HO procedure. Majority
of the HO failure instances are observed due to late
HO attempts.
•Ping-pong HO KPI point out the early HO occasions
in a cell. UE undergoing ping-pong HOs leads to back-
and-forth HOs between the participating cells and can
lead to higher signaling load and sometimes even low
retainability KPI.
•Cell discovery KPI measure the small cell camping rate
each time a UE is configured with a cell search pro-
cess. Timely cell discovery can result in more offload-
ing opportunities, and hence, efficient utilization of the
available resources.
•Latency or data interruption KPI represents the delay UE
observe during HO execution, paging attempt to success
duration, accessibility etc.
•Accessibility KPI for a given time interval repre-
sents the percentage of idle mode UEs that were
able to successfully acquire network access. Acces-
sibility KPI indirectly impacts latency and user
tracking KPI under rare circumstances for mobile
users.
183512 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
FIGURE 8. Relationship diagram for mobility related KPIs and their interplay with the associated network parameters (grouped in different colors)
Source: [17], [18].
In most cases the KPI-parameter dependency is multi-pronged
and leads to complex and often conflicting interplay between
the KPIs and parameters. This interplay in the mobility KPI
and the associated key parameters is summarized in Fig 8.
The key challenges that arise from the convolved association
between the mobility KPI and parameter [17], [18] are briefly
described below:
1: Smaller qHyst value accelerates reselection, as soon as
the target cell RSRP becomes greater than serving cell
RSRP. As a result, accessibility issues related to idle
mode mobility (as discussed earlier in the section) can
be addressed. However, too low of a qHyst can result in
unnecessary reselection (for instance, to an over-shooting
cell).
2: Shorter Treselection will improve the accessibility KPI
at the cell boundary due to timely reselection. However,
too short Treselection will result in ping-pong reselection
especially for stationary users (i.e. due to shadowing).
3: Idle mode Cell Individual Offset (CIO) to accelerate or
decelerate reselection towards a neighboring cell. (con-
figuring a positive CIO towards a particular neighbor can
accelerate reselection, and vice versa)
4: Time window to evaluate mobility State [17] of a UE.
Number of reselections made within this time window
will dictate mobility state (normal, medium or high) of a
UE. Reselection criteria is typically eased as mobility state
changes from normal to medium or high.
5: Specify additional time period before UE can enter back
to its normal mobility state with default reselection param-
eters. Recurrent mobility state change can be avoided by
tuning this parameter.
6: Number of cell change needed (ignoring similar cells)
within ‘parameter #4’ before UE changes mobility state
from normal to medium or high respectively.
7: Scaling factor by which the default qHyst (parameter #1)
is decreased when the mobility state is changed to medium
or high.
8: Scaling factor by which the default treselection (parame-
ter #2) is decreased when the mobility state is changed to
medium or high.
9: Amount and location of RACH resources to ensure RACH
success (providing adequate RACH resources, and avoid-
ing RACH resource conflict between neighboring cells).
10: Higher target power can increase chances of RACH suc-
cess at first attempt (better accessibility KPI) at the cost of
a) higher battery consumption and b) chances of increased
uplink interference for neighboring cells. An optimal tar-
get receive power is vital for better network operations.
11: Increase in the transmission power every time a RACH
attempt fails. Higher step size can increase RACH success
but with more battery consumption and vice versa.
12: Maximum allowable UE RACH power - Increasing maxi-
mum allowable UE transmission improves RACH success
probability but with high energy consumption.
13: Improved accessibility to achieve a faster TAU can ensure
accurate user tracking and prevent paging failure instances
for mobile users.
14: Reduce latency through faster accessibility for mobile
users (e.g. fast reselection to best signal cell and appro-
priate power for RACH success).
15: Smaller TA size will improve UE location estimate and
will decrease the core network signaling due to smaller
VOLUME 8, 2020 183513
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
paging area. However, frequent TAU by mobile users will
add radio access side signaling.
16: Suitable TA design (horizontal/vertical assignment) based
on coverage conditions and type of traffic (e.g. high speed
UEs) to ensure accurate user tracking and minimize TAU
and hence, conserve UE battery and network signaling
load.
17: Reducing TAU attempts for mobile users to conserve UE
battery.
18: Reducing TAU attempts for mobile users to lessen signal-
ing load.
19: Fast and efficient user tracking to reduce latency in access-
ing the network.
20: Minimizing signaling helps avoid unnecessary transmis-
sion and the UE battery can be conserved.
21: Higher cell search frequency will be beneficial to offload
users to other cells. However, more battery will be con-
sumed while searching. In addition, signaling load will
increase every time a UE is configured with cell search
procedure.
22: Periodic search mode will reduce signaling data genera-
tion as search configuration will be transferred to UE just
once. However, small periodicity will waste the UE bat-
tery, and a large periodicity might miss a suitable offload-
ing opportunity. On the contrary, a smart aperiodic search
mode (e.g. location triggered) will be efficient and will
save battery but signaling will be generated with each
search configuration.
23: Signaling data generated for cell discovery purposes
should be minimized.
24: UE consumes battery during cell search, hence,
cell discovery should be minimized with high hit
rate.
25: Timely cell discovery (intra-frequency) will prevent out-
of-service (unreachable UE) occasions and Radio Link
Failure (RLF) can be prevented.
26: Timely cell discovery (intra-frequency) will prevent recur-
rent re-transmissions and ultimately lead to Radio Link
Failure at the cell edge.
27: Timely cell discovery (intra-frequency) will ensure HO
success especially for mmWaves and the UE will not
observe Radio Link Failure.
28: Smaller report interval (HO requests) will have more sig-
naling data and battery utilization. However, the reliability
KPI will improve as there will be more chances of BS
being able to successfully receive and decode the HO
request.
29: HO offset/threshold can be tuned to achieve timely HO.
30: Suitable hysteresis parameter will minimize chances of
ping-pong HOs.
31: Small timeToTrigger can result in ping-pong HOs (e.g. for
non-mobile users), while long timeToTrigger can avoid
the HO resulting in low reliability/retainability KPI (e.g.
to overshooting cells). Similarly, high speed users should
be configured with lower timeToTrigger to accelerate HO
to cell with best RSRP.
32: Frequency based CIO to accelerate or decelerate inter-
frequency HOs to all neighboring cell(s). Optimal CIO can
prevent late and/or early HO.
33: Relation based CIO to accelerate or decelerate intra/inter-
frequency HOs toward the configured neighboring cell(s).
Optimal CIO can prevent late and/or early HO.
34: Configuring a large CIO range can avoid the chances
MRO assigns a large CIO (a large CIO is not recom-
mended as it can have negative consequences especially
for static users)
35: Shorter MRO cycle can recommend suitable CIO config-
uration based on changing traffic conditions. However,
too short of a cycle should be prevented as it can have
sub-optimal recommendations due to inadequate statisti-
cal data required to configure optimal CIO.
36: Similar to ‘parameter #4’ but for connected mode.
37: Similar to ‘parameter #5’ but for connected mode.
38: Similar to ‘parameter #6’ but for connected mode.
39: Similar to ‘parameter #8’ but for connected mode.
40: HO failure results in higher latency and more data inter-
ruption occasions.
41: Frequent HOs increases the risk of HO failure both for
static and mobile users.
42: Latency and data interruption are intrinsic to break-before-
make HOs, hence ping-pong HOs should be avoided.
Fig 8 illustrates the simplest representation of the complex
interaction between various KPIs and mobility related net-
work parameters. It can act as a foundation, with the help of
which, researchers can devise an ideal mobility management
scheme that aims to minimize the negative impact on KPIs
indirectly affected by tuning mobility related network param-
eters. Now, we present a detailed survey of the state-of-the-art
literature available on mobility challenges and corresponding
research proposals. Insights from this tutorial section will
be leveraged to evaluate the research papers in terms of
conflicting KPI(s).
III. MOBILITY CHALLENGES AND RESEARCH
PROPOSALS
Seamless mobility experience at a very high-speed is con-
sidered as one of the major use cases for 5G networks,
particularly in wake of advent of autonomous cars, low alti-
tude drones, and emerging high-speed commute systems.
The mobility characteristics of the emerging networks, such
as densification and adaptation of mmWave narrow beam
cells (discussed in section I), combined with the intrinsic
complexity of the mobility management process (discussed
in section II) means that the mobility management in 5G
and beyond requires significant research efforts by wider
community. In this section, we review the recent contributions
made by the research community to address 5G and beyond
mobility challenges, by categorizing them in six sections as
shown earlier in Fig. 1. Studies focused on reliability goals
that involve achieving seamless and timely HO while pre-
venting HO failures and ping-pong HOs are discussed in the
183514 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
TABLE 6. Reliability enhancement approaches.
first sub-section. Studies focused on achieving mobility while
maintaining small delay are discussed in the Latency Require-
ments sub-section. Signaling Minimization approaches are
presented in the next sub-section, followed by User Tracking
in futuristic ultra-dense networks. Subsequent sub-section
covers studies on cell discovery including the goal to perform
timely offloading from macro-cells to small-cells in order to
prevent network congestion and efficiently utilize network
resources. Finally, research work focused on lessening energy
consumption are presented in the last sub-section.
A. RELIABILITY GOALS
Mobility casts a serious threat to reliability especially when
HO is being performed from one cell to another. Now we
will discuss different research work on different HO types
and the respective reliability goals. Comparison of reliability
enhancement approaches has been presented in Table 6.
1) BREAK-BEFORE-MAKE AND RELIABILITY
5G NR employs break-before-make (hard) HO approach [18]
where UE breaks the connection with the serving BS before
resuming the new connection with the target BS, and this
process makes the mobile UE prone to undesirable service
interruption. Repetition of this type of HO under ping-pong
scenario makes it even more susceptible to call drops. An
effort to deal with the frequent HO case has been presented
in [19]. This paper focuses on the multi-objective learning-
based mobility management strategy where a learning model
is described to obtain a comprehensive network information.
Then a multi-objective mobility management method is pro-
posed taking into consideration user QoE and number of HOs.
Results are compared with 3GPP based HO scheme, and the
authors show that number of HOs are reduced by more than
5 times. As a future step, simulations can be presented by
using a stochastic network model.
Much of the reliability concerns are studied while keep-
ing in view the UE downlink performance only. Authors
in [20] studied reliability for uplink channel of multi-user
MIMO channel. Authors employed Quadrature Spatial Mod-
ulation (QSM) to lower the uplink Bit Error Rate (BER) from
10−1(when using spatial multiplex) to the order of 10−3. As a
future work, BER results can be shown with different user
velocity to evaluate the efficacy of the proposed approach for
a realistic scenario of mobile users.
2) MAKE-BEFORE-BREAK AND RELIABILITY
Unlike 5G NR and LTE, 3G uses an alternative of break-
before-make HO, i.e. make-before-break vis-a-vis soft HO.
3G UE apply macro diversity where it can establish simul-
taneous connection to more than one cell, and the set of
participating cells are referred to as Active Set (AS). Authors
in [21] propose a 3G like soft HO approach where multiple
serving cells are represented by AS. The results show that
fixed AS window can prevent RLF to a great extent. However,
throughput degradation is observed as radio resources of the
weaker cells are unnecessarily wasted by the user. To counter
this problem, the authors propose a dynamic AS window
where add/remove parameters are adapted based on the slope
of the linear curve that creates the dependency between the
add/remove offset and the size of AS. AS based approach
will result in more signaling, computation and energy require-
ments in maintaining and updating the connectivity to differ-
ent cells in the AS.
One drawback of make-before-break HO scheme is the
complexity at UE side to process multiple RF chains. Note
that the advent of narrow mmWave beams in 5G that is likely
to lower the source link reliability for the mobile users, further
undermines the perceived advantages of make-before-break
HO. Authors in [22] analyzed the pros and cons of make-
before-break HO in more detail and concluded that they are
VOLUME 8, 2020 183515
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
unsuitable for 5G networks. For similar reasons, 3GPP RAN
WG2 during its meeting #94 decided to discard make-before-
break like procedures from the scope.
For the above-mentioned reasons and to achieve higher
reliability and retainability goals, the 5G networks have
employed hard HO process requiring successful break-
before-make procedures. Reliability goals in literature are
usually addressed through multi-connectivity approaches.
3) RELIABILITY THROUGH MULTI-CONNECTIVITY
Multi-Connectivity (MC) can be employed in conjunction
with break-before-make HO approach to mitigate interfer-
ence through coordination. MC can attain ultra-reliability,
low latency, and interruption-free communication by prepar-
ing the target cell before the transmission is broken. Further-
more, it tackles connection failures by using a coordinated
transmission among the serving cells. As a result, HO fail-
ures and RLFs are drastically suppressed. However, draw-
back of MC includes added complexity in adding/removing
MC participant cells. A study by Tesema et al. [23] on
intra-frequency MC shows that the RLFs can be avoided
while enhancing throughput through joint transmission of
BSs. The authors in [23] then extended their idea in [24]
to inter-frequency MC and prove availability benefits in that
scenario. However, stationary users were considered with
focus on modeling of the best server association. Their study
did not incorporate reliability for mobile users.
In a separate study [25], the same group of authors deal
with mobility concerns and evaluated reliability performance
through different intra/inter frequency cells. For intra fre-
quency, Dynamic Single Frequency Network (DSFN) is pro-
posed to dynamically add BSs to the coordination set. This in
turn helps to achieve reliability and low latency of less than
1ms. For inter-frequency on the other hand, redundant trans-
missions are performed on the different frequency layers,
such that the UE selects the best transmission, i.e., selection
combining is applied. The proposed approach can avoid poor
SINR of <−6dB (marked as RLF) and achieve higher
reliability of 99.999% or greater.
Tesema et al. further enhanced their work in [26] by
proposing a novel multi-connectivity scheme that uses fast
selection of serving cell from a set of prepared cells
similar to Co-ordinated Multi-Point Transmission (CoMP).
Fig. 9 shows different types of CoMP. Control plane in CoMP
is served by a primary cell only, and if radio condition of the
respective control channel degrades, then user plane data may
not be guaranteed even if radio condition of user plane cell is
better. On the contrary, Fast Cell Select (FCS) is proposed in
which the selected cell from the set of pre-arranged cells is
used for transmission of both data and control signals. The
presented work provides gain in the quality of the control
and data signals, which ultimately solves RLF problem and
improve throughput of cell-edge user.
CoMP, although beneficial, has an intrinsic conflict with
the hard-HO methods used in 5G networks, as connection
with source cell terminates before setting up a connection
FIGURE 9. Types of downlink CoMP.
to the target cell. In [27], authors addressed this conflict by
introducing a new HO mechanism based on CoMP joint trans-
mission scheme in order to minimize inter-cell-interference
(ICI) level between the adjacent cells during the HO execu-
tion. Their algorithm consists of Coordination set (CS) and
Transmission set (TS) of BSs. CS selection is assisted by
the UE through sending periodic measurement report which
contains UE velocity and RF condition. Velocity metric is
used to avoid small-cells for high velocity UEs, and RF
condition is used to determine TS. Performance evaluation
results show that ICI is reduced considerably leading to a
better average throughput per user during the HO procedure.
Benefits are achieved at the cost of higher complexity and
increase in signaling data. A study on optimal TS size to
improve reliability, and throughput, taking into consideration
the processing complexity and the magnitude of the control
data would be a good research contribution.
B. LATENCY REQUIREMENTS
Besides reliability, another mobility management objective
of paramount importance is to minimize the length of
transmission disruption during the HO process. In this sub-
section we review the studies and research efforts aimed to
minimize HO delay.
1) RACHLESS HANDOVER
Authors in [22] identified that RACH takes about 8.5ms out
of 50ms interval required to accomplish HO task in LTE.
Based on this assumption, they proposed a RACHless HO
technique to improve the latency by 17%. Authors suggest
alternate means to perform the same functionalities as of
RACH. For instance, RACH helps target BS to compute Tim-
ing Advance, though with lower accuracy. In the proposed
RACHless HO, UE can estimate timing advance from the
time difference between the source and target cell signals.
183516 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
Accuracy evaluation of the proposed approach will help gain
confidence to the researchers. Such timing advance estima-
tion method has been further evaluated in [28]. Alternatively,
target BS can also compute timing advance through Sounding
Reference Signals (SRS) which is used in LTE for uplink
channel estimation as shown in [29]. However, this process
might result in the timing advance estimation delay as it
requires UE to be configured with SRS first. Initial uplink
power, Physical Uplink Shared Channel (PUSCH) in LTE,
normally known after successful RACH procedure, can be
determined through source BS prior to HO initiation. Elimi-
nating RACH is a novel proposal. However, UE in turn has to
do more processing to compute timing advance that may lead
to decreased battery life in a dense network.
While RACHless HO has its merits, the aforementioned
challenges call for alternative approaches to reduce HO
latency. One example of such approach is mobility aware
caching.
2) MOBILITY AWARE CACHING
From the mobile users’ perspective, more data rate alone is
not enough to ensure better user experience. Any bottleneck
in the distribution network between RAN and content servers
can result in a prolonged Round-Trip-Time (RTT). During a
HO, the chances of such bottleneck increase as momentarily
the UE’s QoE becomes dependent on two cells instead of one.
This makes caching in the BS a useful tool to help accelerate
the data delivery to the intended user. However, mobility
degrades cache efficiency when UE moves to another BS.
A study in [30] proposes to incorporate caching and comput-
ing ability deep into the base stations. The authors in [30] pro-
posed a seamless RAN-cache HO framework based on mobil-
ity prediction algorithm (MPA). In the proposed scheme,
the target BS is predicted for a UE with unfinished transmis-
sion during HO. This prediction is then used to pre-trigger
the source RAN cache. This notifies the target RAN cache
associated with the target BS to prepare for serving the UE
and ultimately reducing latency. As a result, false probability
of RAN-cache HO pre-trigger through MPA though recorded
to be less than 1.36% show an 8% increase in the maximal
RAN-cache HO processing time. Researchers should benefit
from the history of user mobility to come up with an improved
algorithm.
Mobility aware caching has been investigated in [31] to
maximize the cache hit ratio that is defined as the number
of requests delivered by the cache server, divided by the
total number of requests. Compared to [30], authors in [31]
considered both macro-cells and small-cells. The first priority
is given to the local cache followed by small-cell. However,
if data is not received within the set deadline, macro-cell is
then accessed to acquire data. Results assert that the proposed
caching strategy outperforms prior caching strategies. The
proposed cache scheme has a better cache hit ratio and low
latency requirement for 5G networks.
3) PAGINGLESS APPROACH
Authors in [32] presented a novel frame structure with
sub-millisecond subframe duration operating in Time Divi-
sion Duplex (TDD) mode aimed for 5G networks. The
frame structure carries UL beacon resources to enable
a pagingless system for idle mode users. For connected
mode users, UL beacons provide channel state informa-
tion (CSI) for improved frequency selective scheduling. How-
ever, a caveat of this approach is that it can lead to an
excessive amount of uplink messages. This in turn, may
cause accelerated UE battery drainage and thus smaller
battery life which is contradictory to one of the major
5G requirements.
C. SIGNALING MINIMIZATION
In both LTE and 5G NR, the processing unit is shifted to
the edge, i.e., BS, primarily to reduce latency. However, this
comes at the expense of increased signaling generated as the
UE context is shifted from one cell to another during the
HO procedure. This issue aggravates with the ultra-dense
BS deployment. High signaling not only chokes the CPU of
BSs, but also results in lower effective spectrum efficiency
by consuming a substantial amount of resources in the air
interface. Too much signaling between neighboring BSs and
BS-Core can result in potential congestion in the backhaul
for the 5G networks with ultra-dense BS deployment. Reason
being the expected myriad of mobile UEs, ultra-dense BS
deployment, and added features that require high coordina-
tion e.g. multi-connectivity, carrier aggregation, and inter-
ference mitigation techniques. Thus, there is a possibility
of network being paralyzed especially in busy hours due to
the avalanche of signaling traffic. Signaling avalanche is an
eminent threat in future ultra-dense networks. The research
efforts by the research community to minimize the mobility
signaling load can be loosely categorized in the following
four sub-categories.
1) HO SIGNALING REDUCTION THROUGH MINING HO
PATTERNS
One basic but effective way to reduce HO signaling is to
characterize HO behavior among cells to identify cells with
an unusually large number of HOs or otherwise abnormal
HO pattern e.g. ping-pong. Authors in [33] study the HO
behavior of cells and propose a clustering model using
K-means, to group cells with similar HO behavior. Further
evaluation was done using actual HO attempt and HO suc-
cess KPI of nearly two thousand WCDMA cells. The idea
is to forecast the number of HOs and detect abnormal HO
behavior among cell pairs using linear regression and neural
network techniques. The detection is then used to perform
targeted optimization of HO parameters in respective cells
to minimize HO signaling. Adding a temporal component
to training data can further increase the accuracy of the
prediction.
VOLUME 8, 2020 183517
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
2) MOBILITY SIGNALING REDUCTION THROUGH RAN
CENTRALIZATION
Another method to reduce mobility signaling is to leverage
the centralization of RAN e.g. using Cloud-RAN (C-RAN).
Karneyena et al. [34] recently proposed mobility aware hier-
archical clustering approach (HIER) to group Virtual Base
Stations (VBSs). Clustering based on the location of Radio
Resource Heads (RRH) aims to reduce costly HOs and thus,
minimize signaling data. They also proposed location aware
packing algorithm (LA) where inter-cluster mobility statistics
are obtained by keeping track of UE movement, UE history
to predict the traffic intensity between BSs. In addition,
the history of inter-RRH HOs is considered as well. The
proposed scheme when compared with affinity propagation
clustering [35] can reduce up to 34.8% HOs, but at the cost
of much higher requirement of RRHs. The approach can
be beneficial for urban areas, but for less dense sub urban
and rural areas, network deployment at this scale won’t be
feasible.
3) MOBILITY SIGNALING REDUCTION THROUGH CELL
EXTENSION
An Extended Cell (EC) concept is proposed in [36] to dynam-
ically form groups of several adjacent cells. HO performance
improvement is rendered by increasing the overlapping area
between two adjacent cells in the Radio over Fiber (RoF)
indoor networks. The proposed approach reduces the num-
ber of HOs and the call drop probability during the HO
by 70%. Although proven effective, it lacks the dynamic
procedures to define ECs to optimize network resources.
Shortcomings were addressed by authors in [37] by extend-
ing the idea and coming up with a proposal on the Mov-
ing Extended Cell (MEC). Here, each mobile UE is cov-
ered by 7-cell EC where each EC transmits the same user
data at every instance. This in turn, reduces HO latency
through early preparation. Evaluation results show the pro-
posed architecture can totally avoid call drop and packet
loss for UE’s with a velocity of up to 40 m/s. The authors
in [37] suggested that MEC is very efficient in tackling HO
for mmWave cells but is vulnerable to throughput ineffi-
ciency as all seven cells in the cluster transmit for a single
user.
4) MOBILITY SIGNALING REDUCTION THROUGH
VIRTUALIZATION
Virtual Cell (VC) has been proposed as a solution by
Hossain et al. in [38] to reduce mobility signaling while
increasing the throughput efficiency of 60 GHz RoF network.
VC is a central part of an actual cell, and the remaining
boundary area is divided into numbered tiles. Wireless Sensor
Network keeps track of the UE location and periodically
sends report to a centralized controller. Multiple Antenna
Terminals (AT) cover a single cell, and only a single AT
is activated at an instant. When the UE steps on one of
the boundary-located tiles, the controller activates respective
neighbor AT to transmit similar data. In the VC scheme
proposed in [38], maximum of only two ATs can be activated
for HO preparation in contrast to 6 in MEC [37]. End results
of using VC concept show an increase of 33% throughput
efficiency in comparison to MEC. Drawback of the proposal
involves management of a wireless sensor network to track
and report UE location. And if the UE velocity is high, the
low powered sensors may not be able to timely report or even
identify the presence of a high-speed user.
D. USER TRACKING
Location management, sometimes referred to as mobility
tracking or user tracking, is defined as the set of procedures
that determines UE location at any instance. User tracking is
inevitable in cellular networks, so that incoming data from
the core network can be delivered to the user. Densifica-
tion of both cells and users, as well as increased mobility
focused use cases such as Intelligent Transportation Sys-
tems (ITS)/Unmanned Aerial Vehicles (UAV) etc. bring new
challenges to user tracking in 5G environment. The recent
attempts to address these challenges can be loosely catego-
rized into following three subcategories:
1) DISTRIBUTED TRACKING AREA UPDATE
A framework to minimize conflicting metrics, Tracking Area
Update (TAU) and paging, is presented in [39] by distribution
of Tracking Area (TA) into Tracking Area Lists (TAL) in two
phases. First phase is offline, which is responsible to assign
TAs to TALs using three different approaches. The first two
favors paging overhead and TAU respectively, while the third
one uses nash bargaining game to ensure fairness between
paging overhead and TAU. Second phase is online which
controls the probabilistic distribution of TALs on UEs by
taking into account their behavior, incoming transmission fre-
quency and mobility patterns. Numerical results were shown
for the three approaches of the first phase, where the third
solution provides a fair tradeoff between paging overhead and
TAU. As a future step, results should be compared with prior
schemes.
No research work focusing on the horizontal or vertical
deployment of TAs is present, therefore researchers can come
up with smarter and more effective ways for operators to
define Tracking Areas.
2) HYBRID TRACKING AREA UPDATE AND PAGING
5G network will have large range of UEs and dense network
deployment as discussed earlier. Hence, a huge amount of
paging especially for millions of IoT devices is expected. As a
result, signaling associated with paging may become enor-
mous if currently available approach is used. To address this
problem, authors in [40] propose a hybrid scheme in which
either RAN or core network can initiate paging. RAN based
paging with Tracking Area (TA) of just one BS is proposed
for the RRC inactive [41] UEs to have low latency at the
expense of high buffering capacity to transfer the content to
the neighboring BS in case of user mobility. Meanwhile, core
183518 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
network-based paging is recommended to be used for idle
UEs. Authors also proposed a hierarchical paging and loca-
tion tracking scheme to minimize signaling load by assigning
an anchor BS for location management. They conclude that
RAN based paging is not efficient for high mobility UEs as
TA is limited to a single BS. For hierarchical approach on the
other hand, there should be more data management and pro-
cessing for every user at anchor BS which becomes another
single point of failure. Processor overload or X2 (inter-cell
communication link in LTE) congestion, as a result, can
disrupt the paging process.
3) DYNAMIC/ADAPTIVE TRACKING AREA UPDATE
Authors in [42] proposed an adaptive method that employs
smart TAs to reduce the frequencies of TAUs and the sizes of
paging areas. The proposed scheme uses the interacting mul-
tiple model (IMM) algorithm [42] to determine the estimated
location of a UE at the time of the latest registration and pro-
vide a predicted location after a certain time frame. An exper-
imental evaluation with an artificial trajectory showed that
this approach cuts half of the extra location registrations
compared with non-adaptive methods. Aside from that, this
method also determines TA adaptively to significantly reduce
the average paging sizes resulting in to lesser signaling for
each paging attempts. As a future step, comparison results
can be added for different types of mobile users at different
speeds and trajectories to prove the effectiveness of their
approach.
Authors in [43] employed Apriori algorithm [44] for
dynamic Location Area planning using call logs of sev-
eral mobile users. Apriori algorithm finds frequent itemset
using an iterative level-wise search procedure. By taking
minimum support of 100%, Apriori algorithm can highlight
those cells which serve mobile users every day. Based on
this approach, authors in [43] suggested to create a dynamic
TA based on more than 80% minimum support. Authors
in [43] categorized mobile users into predictable, expected
and random groups based on the minimum support value.
For each category, the authors propose to minimize loca-
tion management cost by employing a suitable algorithm.
However, the exact algorithms needed to minimize location
updates, in this scheme, remain to be investigated as future
work.
E. CELL DISCOVERY
Traditional networks with High Frequency (HF) bands broad-
cast the reference signals (pilot symbols) for cell discovery
as mandated by 3GPP. Majority solutions proposed in lit-
erature for cell discovery involve periodic scanning by the
UE of these broadcast signals. The higher frequency of this
periodic scanning ensures timely cell discovery but results in
increased battery consumption leading to trade-off between
energy efficiency on UE side, network side, QoE, overall
capacity and load distribution. In the following we discuss
studies that have investigated these trade-offs and proposal
solution to optimize one KPI or other.
1) CELL DISCOVERY WITH UE ENERGY CONSTRAINT
5G networks will have heterogeneity of BSs with a motely
of macro-cells and small-cells. A mobile UE connected to
a macro-cell must scan for potential small-cells to benefit
from the high data rate and traffic offloading opportunity.
If a mobile UE uses high scanning periodicity, it is likely to
discover small-cells in a more timely fashion. Thus, it may
avail better offloading opportunities, but at the cost of reduced
battery life due to increased amount of energy consumed
by the scanning process, and vice versa. The investigation
of this tradeoff is interesting and yet a challenging research
problem as the optimal scanning periodicity, if exists, might
be dependent on the cell density and user speed among several
other factors.
Authors in [45] use a rigorous approach that leverages
stochastic geometry-based modelling of the network and
empirical modeling of UE mobility. Analytical expressions
have been derived to characterize and quantify the depen-
dency of the UE energy efficiency on the cell density, cell
discovery periodicity and the user velocity. Through ana-
lytical as well as Monte Carlo simulation results, it’s been
shown in [45] that UE battery life reduces significantly
with increased cell discovery rate, while the UE throughput
increases and vice versa. The key finding of this analysis
is that, there exists an optimal cell discovery frequency for
a given cell density and user speed statistics. This optimal
cell discovery frequency maximizes the UE energy effi-
ciency (EE) by achieving a Pareto optimal point between
the capacity lost by missing cells with low cell discovery
frequency and energy saved at UE in doing so and vice versa.
Calabuig et al. [46] proposed an energy efficient small-
cell discovery technique using radio fingerprints. In this pro-
posed solution, network configures UE with several radio
fingerprints which are lists of cell-IDs and RSRP strength
at different intervals. As a normal procedure, users served
by the macro-cell performs the neighbor cell measurement
as it moves around and compares those to the configured
radio fingerprints. Upon a successful match, macro-cell is
reported back which in return configures the corresponding
small-cell. Authors show that energy efficiency of 70-80%
is achieved on UE side by avoiding unnecessary small-cell
discovery measurements, and up to 45% on network side
by small-cell activation/deactivation. Practical use of this
approach will be limited to shadowing since RSRP at a given
point changes with time and the effect of environmental
changes like rain/snow also affects the standard deviation of
shadowing. Moreover, MDT will reveal better results as the
location of the UE with respect to the small-cell location can
be known, followed by the successful small-cell association.
2) CELL SELECTION WITH NETWORK ENERGY EFFICIENCY
PERSPECTIVE
The Information and Communications Technology (ICT) sec-
tor contributes around 2-3% to world’s carbon emissions
and is doubling every four years [47]. Since mobility is
VOLUME 8, 2020 183519
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
closely coupled with uneven and dynamic user distribution,
the mobility patterns can be exploited to turn OFF/ON cells
for enhancing energy efficiency. A solution to conserve net-
work energy using such mobility leveraging approach is pro-
posed in [47]. Decision of powering OFF the BSs is made
using the UE velocity, receive power, BS load and energy
consumption. In addition, HO to the small-cell can be made
only if the UE velocity and the cell load is lower than the
respective thresholds. As a result, the low load cells can be
powered OFF. However, the paper does not address when
and how to turn ON the cell, as the powered OFF cell in the
presence of the candidate UEs can have negative impacts on
the capacity, efficiency and user satisfaction.
Random way point mobility models and the stochas-
tic geometry theory are utilized in [48] to evaluate the
energy efficiency of 5G networks. The network capacity
and energy efficiency are evaluated for Ultra-Dense Cellular
Networks (UDN) considering the user mobility. Results were
demonstrated using Monte Carlo scheme where a user will
keep stationary for a certain time, and then start moving
to a random direction with variable but bounded velocity
range. Results indicate that the energy efficiency decreases
exponentially with increase in the small-cell density. Energy
efficiency decreases from 160bits/J to 155bits/J and 144bits/J
when small-cell density was increased from 10 cell/km2to
15 cell/km2and 20 cell/km2respectively.
3) mmWave BEAM ALIGNMENT AND USER TRACKING
The studies discussed in the last two subsections do not
consider the several idiosyncrasies arising from the advent
of mmWaves cells, as discussed in the following. mmWave
band cell discovery becomes far more complex compared to
the high frequency (HF) cells because of the high penetration
loss and narrow beams [49].
Directional path in mmWave can deteriorate sharply due to
rapid changes in the environment which calls for an intense
tracking and alignment. The situation can be aggravated
when considering mobile users. To address these issues,
authors in [50] proposed two innovative schemes by which
UE can alternately scan the whole angular space exhaustively
and select the beam with the best SINR. They propose the
mmWave BS to send pilots in the configured finite directions
at regular intervals, one at a time. The UE then scans for
the mmWave-cell beam using two mechanisms: a) periodic
refresh (PR) – The UE scans in all directions one at a time
and the direction with the maximum SINR is selected; b)
periodic refinement and refresh (PRaR) – The first optimal
beam with the maximum SINR is selected as per the PR, and
then the UE performs a refinement procedure by scanning
the neighboring direction to adapt according to the changing
condition or due to the UE mobility. This mmWave tracking
approach is depicted in Fig. 10. Comparison between both
schemes were done using the real-world measurement data
collected in New York city on carrier frequency of 28GHz.
As expected, PRaR is less energy efficient than PR because of
the much frequent refinement procedure. However, they did
FIGURE 10. mmWave tracking. (a) Refresh procedure through
12 directions, (b) Refinement procedure through 2 directions.
not compare their schemes with the broadcasting approach
or direct alignment schemes. Also, the scenario might arise
where both the mmWave BS (in sending pilots) and the UE (in
scanning pilots) are not synchronized with each other in terms
of direction. Such a scenario is likely to lead to the tracking
and alignment delay. Alignment process is done by scanning
the adjacent beams only and can give sub-optimal results for
the high-speed users.
Esmaiel et al. [51] proposed a novel mmWave multi-
level beamforming approach. mmWave link is established
after multi-level beam search is conducted using a com-
pressive sensing-based channel estimation. The estimated
UE location is used to determine the number of beams and
the bandwidth required for constructing the sensing matrix
used in each beam searching level. Results show an increase
in the spectral efficiency by 40% under good radio condi-
tions. Authors in [51] also proposed a novel concept [52]
of two-level control and user data (2CU/U) planes splitting,
where the LTE BS and the WiFi access point provides con-
trol over the distributed sub-clouds and distributed mmWave
BSs respectively. With the proposed approach, mmWave
miss-detection probability as low as 10% can be obtained
compared to 90% with the conventional approach when
mmWave BS are deployed in a sparse manner. The result
can be further improved by incorporating the user move-
ment historical data, and to observe the result for different
UE speed.
a: HO IN mmWAVE BAND
Traditional HO is based on the Received Signal Strength
(RSS) wherein pilot signal strength measured by the UE
determines the cell-edge and thus lends assistance in per-
forming HO to the target cell. This approach is ineffective
for addressing the unique challenges associated with the
mmWaves. In mmWave cells, the RF reception changes dras-
tically with UE speed and direction. Hence relying on the RSS
to anticipate a cell edge may not suffice.
Authors in [53] suggest a novel Inter-Beam HO
Class (IBHC) concept combined with the HO control and
radio resource management functionalities. Initially, the user
is assigned to a mobility classes depending on its estimate
183520 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
speed. The corresponding HO frequency is defined such
that the high velocity UEs are expected to observe more
HOs than the pedestrians. The mobile user is assigned a
group of beams as per mobility class, load conditions and
the expected path of UE. Each beam in the group contains
similar resource allocation to improve the reception quality.
HO is thus performed only at the edge of the beam-group.
The underlying assumption in the proposed scheme is that the
individual signals of each beam are perfectly synchronized.
This can be true for low speed users; however, it may not
hold for the high-speed users. Another strong assumption is
the perfect estimation of UE velocity. UE velocity estimation
is a big challenge even in the existing mobile networks,
where the number of HOs in a moving time window are
used to estimate UE velocity. Emerging networks with dense
deployment of multi-frequency networks will make the pre-
diction of UE velocity even a bigger challenge. Concept
presented in the [53] can be extended by considering the
relationship between the maximum user velocity and the
mmWave footprint where its beneficial for the mobile user
to camp to the mmWave cell. The study should include the
signaling cost and energy consumption in scanning for the
mmWave cells.
In [54], authors leverage the concept of moving cell for
train communication using 60 GHz band. To avoid the large
number of HOs in high speed train, authors propose to employ
the Radio over Fiber (RoF) technique. The key idea is to
make the serving cells move together with the train and thus
provide smooth uninterrupted transmission to the passengers.
However, for this scheme to be practical, the train’s velocity
and the direction needs to be pre-known to achieve synchro-
nization. Furthermore, due to the inability to cope up with
randomness of user mobility, this concept is not appropriate
for mobility management in indoor environments.
The state-of-the-art literature work reviewed in this section
is focused on managing mobility in a reactive way. Two
of the key challenges in mobility management in emerging
networks that are not addressed by the current reactive mobil-
ity management paradigm in the industry and the associated
literature in academia are high latency of the HO process and
the large signaling overhead. These challenges become more
important with the increasing fraction of mobile UEs, more
bandwidth hungry applications and the advent of delay sen-
sitive use-cases like self-driven vehicles. Proactive mobility
management is an emerging paradigm that has the potential
to address these challenges. It’s a vital component by which
the network operators can guarantee the success of the futur-
istic mobile networks. Key concept of the proactive mobility
management and the recent studies that have presented few
novel ideas to achieve the proactive mobility management are
discussed in the next section.
IV. PROACTIVE MOBILITY MANAGEMENT
It is a well-researched fact that people tend to visit the
same places repeatedly in their daily life, e.g. workplace,
school, gym, parks, shopping venues, etc. This makes their
movement to feature a high degree of repetition and hence
predictability. According to some large-scale studies, this
perceptibility can be as high as 93% [55]. This intrinsic pre-
dictability in human mobility can be leveraged to build mod-
els to predict the UE mobility patterns. In cellular networks,
these models can be built by harnessing the large volumes of
UE mobility related data such as call detail records (CDRs),
GPS traces, and data traffic from existing networks. Follow-
ing is the list of some of the potential use cases of mobility
prediction in the current and emerging cellular networks:
•Enhancing the overall QoS and QoE by reserving and
managing radio resources a priori for users expected to
arrive in a cell [56].
•Prevent failures and minimize HO delay e.g. by proac-
tively triggering HO [57], [58].
•Prevent ping-pong HOs.
•Efficient load balancing e.g. by predicting cell loads and
emergence of hot spots.
•Assist in cell activation/deactivation, and hence, con-
serve energy consumption.
Mobility prediction models in literature can be classified into
three broad groups:
1) History based prediction models: In this type of predic-
tion models, UEs next target cell is predicted based on
the statistical analysis of historical records such as HO
records or CDR records.
2) Measurement based prediction models: Such prediction
schemes derive probability of user transition to next cell
based on the real time measurements e.g. RSSI, SINR,
distance, etc.
3) Location based prediction models: Current user loca-
tion and in some cases urban transportation infrastruc-
ture is used to predict the future user location in the
location-based prediction models.
In the following, we discuss the recent studies in literature
that have made use of the two types of prediction approaches
for various use cases.
A. HISTORY BASED PREDICTION
History based mobility prediction approaches can be further
divided into the following categories:
1) CELL TRACE BASED PREDICTION
Location prediction based on cellular network traces has
recently attracted a lot of attention. Zhang et al. propose
NextCell scheme [59] that utilizes social interplay factor to
enhance mobility prediction. Social interplay is characterized
by the convolution between entropy of the average call dura-
tion between two users, and the probability distribution of
these two users to be co-located in the same cell. NextCell
predicts the user location at cell tower level in the forthcoming
one to six hours. It shows that inclusion of the social interplay
improves prediction accuracy by 20% when compared to
behavior periodicity-based predictor. However, results were
not compared with the existing prediction schemes.
VOLUME 8, 2020 183521
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
Authors in [60] presented a HO prediction scheme that
combines signal strength/quality to physical proximity along
with the UE context in terms of speed, direction, and HO
history. The presented scheme achieves 33.6% reduction in
HO latency when compared with conventional HO approach.
2) MACHINE LEARNING BASED PREDICTION
Complex interaction between different components of a net-
work can be well captured by Machine Learning approaches.
For the same reason, much of the history-based prediction
works revolve around machine learning based approaches.
Authors in [33] argue that most of the research involving
behavior prediction of a single UE is an infeasible and
impractical approach. The argument is backed by the fact
that some HOs are coverage based, while some are network
initiated (e.g. load balancing). They propose to address these
challenges by employing the K-means algorithm to group the
cells with the most similar HO behavior into a cluster. Next,
the future HOs were forecasted, and abnormal HOs were
identified. The main target of the proposal is to minimize the
signaling load by avoiding the abnormal HOs.
Now we will present some of the research work done on
specific machine learning algorithms:
a: SUPPORT VECTOR MACHINE
Authors in [61] capitalize on Support Vector Machine (SVM)
to predict the user location in the next 5 seconds. A framework
to minimize HO delay using mobility prediction is proposed.
However, they did not validate the framework, neither did
they compare their work with the existing proposals. In [62],
SVM predicts the next cell in a real-time manner, by combin-
ing GPS data, short-term Channel State Information (CSI),
and long-term HO history. The presented model was applied
on a synthetic Manhattan grid scenario. Results show that CSI
results in almost 100% better prediction accuracy compared
to using HO history alone. Using different shadowing values
to represent different terrain and environment can further
strengthen the idea practicality.
b: NEURAL NETWORKS
Few works in [63], [64] have leveraged neural networks for
mobility prediction. The basic idea is to utilize the neural
network to learn mobility-based model for every user and
then make prediction about the future serving cell. Authors
in [63] performed clustering of the input RSS samples
through k-means. The clusters and input RSS samples were
then fed to a classifying model, where neural network was
used to predict the user position. Results show that the pre-
diction accuracy increase by just 5% when compared to the
prediction using neural networks alone.
3) MARKOV CHAIN BASED PREDICTION
A large number of research studies have used Markov chain
based approaches for mobility prediction for their ability to
yield better accuracy than most other predictors with lower
complexity [65]. In the following, we review recent studies
for commonly used Markov Chain (MC) variants:
a: STANDARD MARKOV CHAIN
Standard Markov Chain is a memory-less algorithm as the
next state depends only on the current state and not on the
sequence of the events that preceded it.
Authors in [66] extracted trajectories of 4,914 individ-
uals using 27-day log of the mobile network traffic data.
They compared the original Markov algorithm with the
Lempel-Ziv (LZ) family algorithm [67]. The core opera-
tion of the LZ predictor is by maintaining a prediction tree
which adds more complexity compared to Markov. It was
concluded that although slightly more accurate, LZ fam-
ily algorithm consumes a lot more resources and time
than Markov algorithm. Most of the mobility prediction
algorithms only consider spatial factors to predict future
movements. Authors in [67] improved Markov Chain based
model by adding a temporal factor and achieved 6% higher
accuracy.
Humans usually follow regular paths as discussed earlier,
however, they may deviate from their accustomed routine at
some instances. Authors in [68] proposed a practical model
based on State Based Prediction (SBP) method to predict
the place to be visited when the user’s trajectory exhibits
unexpected irregularities. When user diverts from the routine,
SBP is employed to conduct the prediction. Experiments
reveal that the accuracy of proposed model can reach more
than 83%, which is higher than the accuracy of 60% achieved
by LZ predictor used in [67].
Authors in [69] proposed an implementation architecture
for the MOBaaS (Mobility and Bandwidth prediction as a
Service). The MOBaaS can be readily integrated with any
other virtualized LTE component to provide the prediction
information. Spatial information (location history) and tem-
poral information (time and day data) are collected and ana-
lyzed. The results show a 33% reduction in access time for the
requested content using the MOBaaS prediction information
can be achieved. Due to its appeal, several extensions of
MOBaaS were proposed later. For example, in [70], authors
stressed that MOBaaS can be implemented in a cloud based
mobile network architecture and can be used as a support
service by any other virtualized mobile network service.
Authors also evaluated the feasibility and effectiveness of the
proposed architecture.
Fazio et al. [71] propose Distributed Prediction with Band-
width Management Algorithm (DPBMA). The algorithm
uses Markov Chains to predict the user movement at each BS
in a distributed way. This makes the proposed solution differ-
ent from many other studies [66], [68], [69] where Markov
chains are used to improve system utilization by reserving
resources prior to the HO. This helps in preventing the call
drop occurrences. However, distributed algorithm means BS
needs to do a lot of processing making this solution not an
attractive option for low cost BS or small-cells.
183522 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
b: ENHANCED-MARKOV CHAIN
In [72], subscriber’s mobility is predicted using the enhanced
Markov chain algorithm. The core idea is to add the behavior
pattern and temporal data of the users from CDR into the
Local Prediction Algorithm (LPA) and the Global Prediction
Algorithm (GPA). LPA and GPA are based on first and second
order Markov processes where transition probability to next
cell depends only on the present cell, and both present and
previous cell respectively. Results show that the proposed
prediction methodology achieves prediction accuracy of 96%
compared to GPA with prediction accuracy of 81.5%. How-
ever, users without any historical record in the training pro-
cess showed poor prediction accuracy. Techniques such as
particle filter or Kalman filter can be employed to increase
accuracy for new users.
c: SEMI-MARKOV MODEL
Authors in [73] argue that both discrete and spatial Markov
Chain assume human mobility as memory less. By using
these approaches, we can achieve spatial prediction of future
cell, but time factor cannot be incorporated. To address
this concern, authors predicted HO to the neighboring BS
using Semi-Markov Model. Semi Markov process allows for
arbitrarily distributed sojourn times. Experimental evaluation
leveraging on the real network traces generated by the smart-
phone application showed prediction accuracy of 50% to
90%. An extension of this approach can be to have ping-pong
HO predictions.
d: HIDDEN-MARKOV MODEL (HMM)
Cheikh et al. [74] proposed HO decision algorithm (OHMP)
using HMM predictor to accurately estimate the next
femto-cell using a) the current and historical movement infor-
mation, and b) the strength of the received signals of the
nearby BSs. The performance of OHMP is validated by com-
parison with the nearest-neighbor and random BS selection
strategies. Results show that the number of ping-pong HOs
reduce by 7 times when considering dense deployment of
femto cells. Results in [74] are demonstrated for a single user
scenario only and does not portray futuristic cellular networks
with large number of users. To address this concern, same
set of authors extended their idea in [75] by incorporating
multiple UEs. They take into consideration the available BS
resources of serving femto-cell and interference level from
the target femto-cell. The presented OHMP-CAC algorithm
introduced a proactive HO scenario where HO is triggered
when SINR of the serving cell reaches a predefined threshold.
OHMP-CAC minimized the number of HOs by 64% and
reduced the average HO decision delay by up to 75% when
compared with the traditional RSSI based scheme.
As discussed earlier, mobility prediction using Markov
chain is a memory-less system as future state can only be
determined by the current state. On the other hand, enhanced
Markov Chains are based on historical data, but their appli-
cation is very complex. Moreover, mobile operators may not
be allowed by the customers to use their historical data due
to privacy concerns. Even if historical records are accessible,
HO delay might still be observed due to the extraction and
processing complexity of historical records. Due to these
factors, history-based prediction algorithms might render
impractical.
B. MEASUREMENT BASED PREDICTION
Measurement based mobility prediction approaches are more
accurate than history-based mobility prediction schemes.
However, the processing complexity due to the measurement
procedure cannot be ignored.
1) RSSI BASED PREDICTION
Soh and Kim [76] introduced RSSI based mobility predic-
tion while keeping in view different UE velocities. They
incorporated UE trajectory and road topology information
to yield better prediction accuracy. The prediction goal is
to achieve timely HO and limit the probability of forced
termination during HOs. In addition, bandwidth reserva-
tion scheme was proposed that dynamically reserves radio
resources at both participating BSs during the HO procedure.
Results show that proposed mobility prediction scheme helps
achieve almost similar forced termination probability as the
benchmark scheme with perfect knowledge of the mobile
UE’s next cell and HO time.
Authors in [77] proposed an RSSI-based prediction
scheme to reduce VoLTE end-to-end delay and HO delay
under different UE velocities in mixed femto-cell and macro-
cell environments. The core idea is to send the measurement
reports based on user velocity and predict when and where
to trigger HO procedure. As a result, HO delay is reduced
by 28%. For ultra-dense BS deployment, mobile UE may
not perform HO to each BS on its trajectory. Future work
can include the consideration of load condition, so that both
low latency and adequate resources can be guaranteed for
improved QoE.
The decision to skip the HO to a better radio condition
cell can be based on dwell time or cell load condition. Next
femto-cell prediction based on radio connection quality and
cell load status is presented in [78]. Authors proposed two cell
selection methods; a) BS prediction after analyzing the col-
lected data of average RSSI from nearby femtocells, b) using
cognitive radio to sense neighboring femtocells load before
triggering HO. Results show that appreciable number of HOs
can be avoided when compared with only RSS based HO
approach. Thus, data interruption during HO and chances
of Radio Link Failure e.g., due to ping-pong HOs can be
avoided.
Authors in [79] argue that RSS alone should not be
considered when performing inter-RAT HO. Instead cur-
rent RSS predicted RSS and available bandwidth should be
considered. They proposed Fuzzy logic based Normalized
Quantitative Decision (FNQD) scheme which aids in elim-
inating ping-pong effects in HetNets. This work can help
realize improved mobility management for LTE-Unlicensed
VOLUME 8, 2020 183523
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
(LTE-U). However, the key performance metrics such as
throughput and HO delay should be added for validation
purposes.
2) MEASUREMENT REPORT BASED PREDICTION
Song et al. used Grey system theory in [80] to predict the
(N+1)th measurement report (MR) from Nth MR for high
speed railways. The key idea is to utilize the predicted MR to
make proactive HO trigger decisions. Their findings showed
that the difference between predicted MR and actual MR is
within 1%. Thus, the proposed scheme is capable of proac-
tively triggering HO in advance and HO success probability
is enhanced from 5% to 10%.
3) USER DIRECTION BASED PREDICTION
Authors in [81] present a user mobility prediction method
for ultra-dense networks using Lagrange’s interpolation.
They predicted user’s arrival into their neighboring fem-
tocells based on users moving direction and the distance
between users and neighboring cells. The presented approach
increases the prediction accuracy when compared with only
distance based and direction based mobility prediction. How-
ever, the performance of their proposed prediction scheme
is not compared with other existing schemes to quantify the
performance gains.
4) USER VELOCITY BASED PREDICTION
Higher UE velocity imposes additional threat to reliability
making prediction of UE velocity extremely important to help
tune the parameters more effectively. 3GPP based solution
assigns mobility states (high, medium, low) depending on
certain number of HOs in a moving time window. How-
ever, this technique will be inefficient in 5G networks with
unplanned and highly dense deployment of heterogeneous BS
having variable cell radius. UE velocity was estimated in [82]
based on the sojourn time sample and accuracy was analyzed
via Cramer Rao Lower bound. Numerical results show that
the velocity prediction error decreases with the increase in BS
density. The authors in [82] further extended their idea in [83].
The predicted UE velocity was used to assign the appropriate
mobility state. Validation was done by gathering statistics
of the number of HOs as a function of UE velocity, small-
cell density, and HO count measurement time window. The
results show similar conclusion as in [82] that the accuracy of
a suitable mobility state detection (known from UE velocity)
increases with increasing small-cell density.
Authors in [84] observed that mobility in urban areas
depends on the traffic laws and is affected by the behavior
of other people (red signal, other driver brakes etc.). They
predicted user mobility based on the observation that a UE
with constant velocity will probably go straight, while a UE
decreasing in velocity might indicate stoppage on red light or
a turn to a different direction. User location in their model is
estimated from uplink time difference of arrival or provided
by the UE via AGPS while velocity estimation is achieved
by increasing sampling rate of location or by Doppler shift.
Results showed that overall throughput can be enhanced by
39%, 31%, and 19% for UE velocities ranging from 25, 50,
75 km/h respectively.
C. LOCATION BASED PREDICTION
The knowledge of UE location can assist in an improved
mobility prediction. Effective localization when combined
with the mobility prediction algorithms can yield more effi-
cient HO related QoE results.
Soh and Kim in [85] presented a decentralized Road Topol-
ogy Based mobility prediction technique where the GPS
equipped UEs shall perform mobility prediction based on
approximated cell boundary data that was shared by the serv-
ing BS. Cell boundary data is represented by a set of points
at the cell edge and is populated based on historical measure-
ment reports sent by UEs. UE at the cell edge will thus report
the corresponding location ID back to the BS, and proactive
resource reservation at potential BS can be achieved. Results
show considerable reduction in forced termination compared
to a reactive HO approach without mobility prediction. This
approach can be applied to the macro-cells but is not reason-
able to small-cells as mobile UEs will have to send a lot of
high-powered uplink messages at cell edge (high path loss
condition). This can lead to an increase in HO failure due to
high uplink RSSI. Moreover, UE battery consumption will be
high.
Authors in [85] proposed mobility prediction scheme
based on road topology information. The main idea is based
on the approximated cell boundary based on prior HO
instances, being configured by the serving cell. The authors
in [85] extended their idea in [86] to add the temporal
component to mobility prediction. The scheme uses linear
extrapolation from a UE positioning data to predict its HO
cell and time. 70% mobility prediction accuracy was achieved
compared to 60% in their prior work [85].
Location based mobility prediction approaches assume all
cell phones to have an accurate position information, which
cannot always be guaranteed. Moreover, security concerns of
the subscribers may hinder the collection of necessary data to
realize accurate cell boundaries.
While proactive mobility management seems to be a great
fit to address the stringent QoE requirements in the emerg-
ing cellular networks, the trivial network dimensioning tasks
should be planned while keeping in view the effect of mobil-
ity on the deployed network.
V. MOBILITY ORIENTED NETWORK PLANNING AND
OPTIMIZATION
Realizing massive potential of network densification to
address the capacity crunch has introduced additional net-
work planning challenges as discussed by Azar et al. in [87].
One such challenge will be faced due to larger fraction of
the mobile users in the network; hence, the network must
be planned while considering mobility management in mind.
Suitable network architecture can help achieve QoS goals
183524 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
FIGURE 11. Directional network deployment using RRHs [88].
while keeping the cost (e.g. signaling) to a minimum, and
ultimately help attain higher network efficiency.
A. SIGNALING MINIMIZATION BY REDUCTION IN
HANDOVERS IN HIGH SPEED TRAINS
Since considerable signaling overhead is being generated due
to a single HO, network planning and architecture aimed to
reduce the number of HOs can certainly be very effective.
High speed train users are subjected to frequent HO as they
move along the track. Apart from a huge amount of signal-
ing data generation, they can also encounter severe issues
like RACH failure, late HO, Radio Link Failure (RLF), and
Release with Redirect (RwR). Futuristic mobile networks
with smaller footprint small-cells will cast an even bigger
risk.
To address this problem, authors in [88] presented a HO
minimization technique where they propose to install an
antenna on top of the train that will perform connectivity
and trigger HO with covering BSs. Network deployment
approach has been demonstrated in Fig. 11. This elevated
antenna interfaces with an inner-train network to serve the
passengers.
Thus, instead of several users performing HOs simulta-
neously, only one HO will be performed by the elevated
antenna. This not only reduces signaling load, but also min-
imizes the risk of HO failure as UEs will not experience
penetration loss of 20-30 dB inside the train. Field trial
conducted on a 2.4km run showed downlink throughput
of 1.25Gbps.
The concept of elevated antenna seems practical and is
studied even by 3GPP [89]. However, single point of failure
lies on its very foundation; if elevated antenna fails and
observes HO failure then the multiple users being served
under that antenna will have disrupted data transmission.
Intelligent switching of the elevated antennas based on prox-
imity to the BS can not only avoid HO failure but also
deliver high throughput due to better SINR, but at the cost of
complexity and cost. Another drawback will be the latency
due to the addition hop between the top-mounted antenna
and the inside-train UEs. As a result, self-driven trains in
the near future might not achieve the required latency QoE
goal.
FIGURE 12. Frame structure for legacy LTE vs C/U plane split architecture.
B. CHANGING CORE NETWORK (CN) TO ACHIEVE
LATENCY GOALS
Authors in [11] studied the latency, HO execution time, and
coverage of four live LTE networks based on 19,000 km of
drive tests. The test was conducted in a mixture of rural, sub-
urban, and urban environments. Their measurements reveal
that the lion’s share of latency comes from the core network
rather than the air interface. Based on the study in [11],
Johanna et al. [90] proposed a new entity called the edge
node that integrates MME and control plane part of SGW
and PGW. Each edge node covers several BS, and when
UE moves to coverage of another edge node, the application
server and gateway is also shifted to minimize the latency.
This approach helps to reduce latency for every HO done
within BSs connected to the same edge node. However,
HO associated with inter-edge node is followed by IP address
reassignment and application-server transfer, which adds to
delay and data interruption.
Keeping in view that the number of 5G subscriptions will
be 2.6 billion by the year 2025 [1], authors in [5] suggested
a simplified 5G core network which will be connectionless,
and will incorporate the best effort without the support for
node mobility. The core idea is to have a legacy internet-like
core network that will not be QoS centric, and the majority of
the traffic will flow through default bearers only. Experiments
were conducted on a smartphone to show that video stream-
ing, web browsing, and messaging will work well, thus,
the future core network can be radically simplified, resulting
in a cost-effective solution. The authors in [5] mainly focused
on a simplified core network with low complexity. Over-
simplification of core network is not a practical approach
as major functionalities of billing and access control cannot
proceed. Similarly, IP re-allocation at every single HO is not
feasible and may result in high latency or even packet loss.
C. C/U PLANE SPLIT
With improvement and advancement in the hardware technol-
ogy, telecom operators can benefit from decoupling control
and user plane (see Fig. 12). By doing so, future mobile
networks with the composite of macro-cells and small-cells
can be used intelligently for efficient resource utilization.
Moreover, signaling overhead from large number of HOs can
be minimized by assigning control plane and user plane to
macro-cells and small-cells respectively.
Authors in [91] address mobility support for high den-
sity, flexible deployment of small-cell architecture with
VOLUME 8, 2020 183525
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
TABLE 7. Mobility prediction approaches and their key-goals.
flexible backhaul using Localized Mobility Manage-
ment (LMM) technique. The first step centralizes control-plane
from small-cells to a Local Access Server (LAS). The second
step allows individual small-cells to handle the mobility
events, but still requires the LAS to act as a mobility anchor.
Analytical model based on discrete time Markov chain is used
to evaluate the average HO signaling cost, average packet
delivery cost, average HO latency and average signaling load
to the core network. Results show that average HO latency
decrease by 10ms compared to the 3GPP scheme [11].
Authors in [92] minimized signaling overhead in a 5G net-
work with a high density of mmWave BSs serving users under
183526 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
TABLE 8. AI-assisted mobility management approaches.
the umbrella of macro BSs. C/U split was employed where
macro BS provides the control plane and several mmWave
cells were group into clusters. Inter-cell HO signaling was
curtailed by using a gateway cluster controller, resulting in
signaling reduction in the core network as well. Results show
that normalized X2 signaling overhead reduces from 100% to
10% as the density of the deployed mmWave cells increases.
Authors in [93] targeted latency minimization in their pro-
posed novel mobility management scheme for intro-domain
handover (HO within the same SDN domain) and
inter-domain handover (HO across different SDN domains).
Layer 2 information and buffering approach was used to
achieve HO latency of just 400ms compared to the legacy
DMM with 100ms of HO latency.
While proactive mobility management and mobility-oriented
network planning seem to deliver promising results, the con-
stant temporal variations in a live network and the importance
of key landmarks can be addressed by introducing Artificial
Intelligence (AI) to the cellular network domain.
VI. AI-ASSISTED MOBILITY MANAGEMENT
In recent years, AI has gained much popularity for proactively
managing mobility in future cellular networks. This is primar-
ily because of an increasing number of configuration param-
eters and due to the complex interaction between network
parameters and associated KPIs (as illustrated in Fig. 8). Once
the research community is able to overcome those complex
challenges, AI-assisted solutions will have a revolutionary
effect on the telecom industry. The tutorial section (Section II)
of this paper can help researchers understand the convoluted
interplay between the network parameters and affected KPIs.
Now we will present some of the AI enabled mobility man-
agement solutions present in the literature. The comparison
of the presented algorithms can be found in Table 8.
The mobility prediction algorithm is presented in [94].
Authors use realistic mobility patterns to capture the human
movement and a 3GPP compliant 5G simulator was used to
represent the HetNets scenario. Results show that mobility
prediction accuracy of almost 87% can be achieved for 2dB
shadowing with XGBoost compared to 78% with Deep Neu-
ral Network (DNN). The work can be extended by using time
series predictors such as recurrent neural network or LSTM.
Authors in [95] employed XGBoost supervised machine
learning algorithm to perform partially blind HOs from sub-
6GHz to co-located mmWave cell. Authors show that this
machine learning-based algorithm to achieve partially blind
HOs can improve the HO success rate in a realistic network
setup of co-located cells. The proposed algorithm should be
compared with the existing HO approach in terms of energy
efficiency and RLF to further validate the efficacy of the
algorithm.
The idea of inter-frequency HO from a macro-cell to a non-
co-located high frequency cell with a much lower footprint
is presented in [96]. The authors use the Random Forest
classification approach and also presented a use case of load
balancing by which an efficient resource utilization for the
static users can be achieved. The shortcoming in the presented
approach is that for high-speed users, the load balancing
based HO to smaller footprint cell may be inefficient due
to large HO rate and the resultant signaling overhead and
chances of HO failure.
Authors in [97] develop a Reinforcement Learning (RL)
based HO decision algorithm for the mmWave cells by
taking into account the user experience as a weighted sum of
throughput and HO cost. Based on the user’s mobility infor-
mation, the optimal beamwidth is selected by considering the
trade-off between the a) directivity gain and b) beamforming
misalignment. The algorithm approves the HO trigger for
VOLUME 8, 2020 183527
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
mobile users depending on UE velocity and BS density. The
work can be extended by evaluating the signaling overhead
reduction and throughput gain achieved when compared with
other existing algorithms in the literature.
Authors in [98] predicted the RSRP of the serving and
the HO target cell using Long Short-Term Memory (LSTM)
and Recurrent Neural Network (RNN). The algorithm also
predicts RLF instances with an accuracy of 84% using only
RSRP as an input feature. An extension to [98] has been made
in [99] where other features like SINR, out-of-sync identifier,
RACH issues, and max RLC retransmission have been used
for RLF prediction.
A wrong HO avoidance algorithm has been proposed
in [100]. It uses neural networks to prevent the HO to BSs
which are affected by the undesirable radio propagation
scenarios in the network, e.g., coverage hole caused by an
obstacle. The proposed algorithm enables a UE to learn from
past experiences (coverage unavailability) to select the best
cell for HO in terms of QoE. The authors show that their
algorithm helps achieve users to successfully complete the
downlink transmissions more than 93% of the time. However,
the simulation environment is quite simplistic where the UE
traverses a straight line with only three BSs along the way.
Hence, the movement of UE is almost deterministic, and the
Neural Network can easily learn its pattern and can identify
the optimal BS to perform HO. Furthermore, a single test UE
gives a limited evaluation of the proposed algorithm. Elabo-
rated results with a HetNet scenario and arbitrary movement
of multiple users will have more realistic results.
Based on HO attempts per hour, authors in [33] cluster
cells into different groups with similar HO profiles using the
K-means algorithm. For each cluster, hourly HO attempts
were forecasted using linear regression, polynomial regres-
sion, neural networks and gaussian processes. the highest
R2 value of 0.99 was obtained when using the gaussian
process. The proposed model then checks for abnormal HO
behavior e.g. ping-pong. Future work can be to proactively
predict abnormal HO behavior ahead of time and to rec-
ommend suitable proposed parameters to prevent HO KPI
degradation.
VII. FUTURE RESEARCH DIRECTIONS AND CONCLUDING
REMARKS
Ultra-Dense Cellular Networks (UDN) containing mmWave
based small-cells are being considered an essential part of
the future vision of cellular systems vis-à-vis 5G and beyond.
Harnessing mmWave spectrum has a strong potential to solve
the two long-standing problems in cellular networks: spec-
trum scarcity and interference. Remarkably, most research
towards UDN remains focused on channel modelling and
hardware design aspects of the mmWave based UDN, and
mobility management in UDN so far remains a Terra incog-
nita. The panorama of mobility challenges arising in emerg-
ing mobile networks implies that if no drastic and timely
measures are taken to rethink mobility management for future
UDN, user mobility management can become the bottleneck
in practical deployments of UDN despite advances in the
hardware design of mmWave and conventional spectrum
based small-cells. Enabling seamless mobility in futuristic
mobile networks require much complex network design and
planning in order to achieve the QoE goals and to address
the intricacies of the network architecture needed to realize
the promised user experience. The high throughput require-
ment, heterogeneity of UEs and BSs, and security awareness
of 5G environments appeal for a fast, distributed and privacy
preserved mobility management. This article provides an
extensive survey of mobility management for future cellular
networks. As studied in the prior section, researchers have
added healthy contributions in an attempt to realize an opti-
mal and satisfactory network. However still, some research
domains are untouched or haven’t been given the attention
they deserve. Now we will discuss a few of the key points
related to future research directions:
A. HO DELAY BASED SINR DISTRIBUTION
Current SINR modelling is based on best-server-association,
however, the UE always camp on the second-best cell
prior to HO. This is the result of the HO evaluation pro-
cess [18] which ensures that the target cell is the best
candidate cell for HO. A mobility oriented SINR distri-
bution which captures the temporal negative SINR [101]
before HO needs to be studied for more realistic throughput
estimation.
B. HO DELAY BASED UPLINK INTERFERENCE
Current researchers do not consider the practical situation
where due to intra-frequency HO delay, high mobility users
are closer to the target cell while still being served by the
comparatively farther located serving cell. Under those cir-
cumstances, high uplink power to achieve target SINR in
the serving cell can cause strong temporal interference in
the target cell. The issue can be aggravated under highly
dense BSs deployed in an impromptu fashion. However, this
problem can be tackled by utilizing an eICIC ABS (Almost
Blank Subframe) scheme for highly mobile users. Proactive
HO trigger can also eliminate the possibility of high uplink
RSSI by performing timely HO.
C. LATENCY GOALS
Another challenging aspect of the small cell deployment is
that the small-cells are typically not directly connected to
the core network and lack Xn or N2 interfaces (for inter-cell
communication) which are the real means of coordinating
mobility procedures in the macro-cells. The lack of a low
latency connection to the core network can contribute to
significant HO signaling delays.
D. ENERGY-EFFICIENCY
Achieving both UE and network-level energy efficiency is
a big challenge for futuristic cellular networks, especially
when considering ultra-dense BS deployment and the addi-
tion of a wide variety of user devices. Most of the existing
183528 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
FIGURE 13. Load balance (LB) opportunities (i, ii, iii, iv) in different
stages of 5G UE connection.
energy-saving schemes have a common tenancy; cells are
switched ON/OFF reactively in response to changing cell
loads. A meritorious effort has been made by Hasan et al.
in [56], where authors proposed the AURORA framework
in which the past HO traces are utilized to determine future
cell loads. The prediction is then used to proactively sched-
ule small-cell sleep cycles. Load balancing is also achieved
through the use of appropriate Cell Individual Offset (CIO).
E. SMART INTRA-FREQUENCY SEARCH
Dense deployment poses challenges for small-cell discovery
as conventional cellular networks broadcast a neighbor list
for the user to learn where to search for potential HO cells.
However, such a HO protocol does not scale to the large num-
bers of neighboring small-cells and the underlying network
equipment is not designed to rapidly change the neighbor cell
lists as small-cells come and go.
F. SMART INTER-FREQUENCY SEARCH
Inter-Frequency (IF) mobility is a vital component of cellular
networks but has not got the attention it deserved in the
research community. IF-mobility requires event A2 to be trig-
gered, which is followed by the BS to configure measurement
gap periodicity to the UE. However, this process interrupts
data transmission and reception. This is because UE shifts
the radio to measure appropriate IF-cell(s). Futuristic mobile
networks with a variety of frequencies ranging from HF to
mmWave band may require the UE to undergo an extensive
search of available frequencies before initiating a mobility
decision. This issue can be aggravated when considering the
latency goal of <1ms.
G. IMPROVING MOBILITY LOAD BALANCING
Mobility Load Balance (MLB) is a vital component of het-
erogeneous multi-layer cellular networks and are open to the
following challenges:
•LB can be achieved at four different instances as shown
in Fig. 13. It can be triggered through i) idle mode
SIB4 configuration, ii) after network access using A4 or
A5 measurement report, iii) in connected mode using
A4 or A5 measurement report (as configured), iv) when
UE is released from connected to idle mode using 3GPP
proposed IMMCI (Idle Mode Mobility Control Info).
In IMMCI, traffic steering is achieved by varying the
idle mode SIB5 priority of the serving or target layer.
LB in idle mode is the most optimal as signaling and
data interruption associated with connected mode LB
can be avoided. Moreover, complexity in parameter con-
figuration and management by IMMCI can be mini-
mized. Research contributions are currently lacking for
idle mode load balancing. Similarly, a new variant of
IMMCI (SON based) is needed which can adaptively
steer traffic to achieve load balancing under varying load
conditions.
•LB detail procedure has not been provided by 3GPP and
is left intentionally to vendors for innovation purposes.
LB requires the exchange of load information between
participating BSs via the Xn interface. However, differ-
ent vendors have their own proprietary version of LB
implementation, thus, inter-vendor BS cannot perform
LB due to mismatch in LB metrics. The existing LTE
networks deploy offloading feature, where high load cell
offload users to another vendor cell without considering
its load condition. This can cause service rejection and
ping-pong HO conditions. The frequent IF-search will
disrupt continuous reception and will result in higher
latency. 5G heterogeneous network can assume numer-
ous vendors, and to benefit from the load balancing
feature, a standard inter-vendor LB mechanism need to
be devised.
•Cells with smaller footprints will have few serving UEs,
and mobility-based ingress and egress of even a single
user can have drastic load imbalance among available
frequency bands. Hence, ways to achieve proactive LB
is mandatory to have fairness and efficient resource
utilization.
H. MOBILITY IN mmWave NETWORKS
mmWave with bandwidth as large as 500MHz is the remedy
to the spectrum saturation in the HF band, however, an intrin-
sic feature of narrow beams can pose serious challenges in
supporting mobility in the emerging cellular networks. Few
of the main challenges are presented here:
•Simic et al. [102] practically demonstrates mmWave to
prove multi-Gbps connectivity but conclude that sup-
porting mobility is a very challenging task due to the out-
age area of as high as 40% with 90BS/km2deployment.
The reason for the coverage hole is the high diffraction
phenomena in mmWaves, and absence of Non- Line of
Sight (NLoS) paths.
•Corner Effect: Indoor areas have cell edge near doors,
where the user is more likely to make a sharp turn
and hence, time available for HO would be very less
especially in the 60GHz mmWave scenario. This issue
suggests that some sophisticated techniques, other than
conventional methods are required for the HO trigger.
•Current mmWave standards such as IEEE 802.11ad fol-
lows the max-RSSI based approach for UE-BS associ-
ation, however, this solution appears rudimentary and
ineffective for emerging network with an ultra-dense
BS density. There will be chances of an unbalanced
VOLUME 8, 2020 183529
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
number of users per BS, and ping-pong HOs will be
highly likely.
•In addition, cell discovery for mobile users is a major
challenge due to the absence of Reference Signal (RS)
broadcast as in HF bands.
Presently, an overwhelming understanding of the research cir-
cle is to use mmWave-cells for static users only. Intricacies of
mobility between the beams (of both intra-frequency cells and
inter-frequency cells) need to be addressed to support mobil-
ity. One possible solution is to come up with a hybrid solution
where HF macro-cells with much accurate UE location guide
the UEs how, when and to which small-cell they need to
connect to. This is similar to control-data split architecture
with mmWave providing data support while UE is under the
coverage of macro-cell providing control signals.
I. LOW-COST MULTI-CONNECTIVITY
Dual connectivity architecture has been proposed to mitigate
mobility management problems in HetNets by allowing UE
to connect with the macro-cell for control connectivity as well
as simultaneous data connectivity with small-cells.
The effect of the user association on dual connectivity
performance is an interesting research problem that needs
to be investigated in detail. Researchers need to study the
gain dual connectivity can yield in terms of HO overhead
reduction, synchronization complexity, and radio resource
efficiency.
Most of the research work address reliability and latency
goals through multi-connectivity, however, signaling load
increment is not addressed. More efficient proposals with
special consideration of signaling load need to be devised.
J. ACCURATE AND EFFICIENT MOBILITY PREDICTION
The mobility prediction schemes are seen as a driving force
for context aware cellular network as they are used to proac-
tively reserve resources, trigger LB, and activate/deactivate
small-cells. Few challenges associated with mobility predic-
tion are:
•Users not willing to share location information due to
the privacy reasons.
•GPS data acquisition consume user battery and inter-
mittent accessibility requests resulting in signaling or
RACH issues (some RACH failure issues cannot be seen
in the KPI data).
•Accuracy and reliability of 3GPP proposed Minimiza-
tion of Drive Test (MDT) feature is needed to be evalu-
ated since multitude of factors like the GPS error [103],
quantization resolution etc. affect the accuracy of the
measurements reported by the UE.
•Although human trajectory exhibits high predictable
component [55], however, mobility prediction is always
bound to have some inaccuracy as can be understood
through an example: an office employee may have lunch
in a canteen, in a conference room, with colleagues in
an outside restaurant etc. These random variations are
almost impossible to predict.
A possible solution can be resource reservation to be done
in the multiple neighbors, however, the cost of signaling and
available resource for other UEs especially during busy hour
needs to be considered.
ACKNOWLEDGMENT
The statements made herein are solely the responsibility of
the authors. For more details about these projects please visit:
http://www.ai4networks.com.
REFERENCES
[1] Ericsson. (Jun. 2020). Ericsson Mobility Report. [Online]. Available:
https://www.ericsson.com/49da93/assets/local/mobility-report/
documents/2020/june2020-ericsson-mobility-report.pdf.
[2] VNI Global Mobile Data Traffic Forecast Update 2017-2021, Cisco Syst.,
San Jose, CA, USA, 2017.
[3] O. G. Aliu, A. Imran, M. A. Imran, and B. Evans, ‘‘A survey of self
organisation in future cellular networks,’’ IEEE Commun. Surveys Tuts.,
vol. 15, no. 1, pp. 336–361, 1st Quart., 2013.
[4] S. A. Busari, K. M. S. Huq, S. Mumtaz, L. Dai, and J. Rodriguez,
‘‘Millimeter-wave massive MIMO communication for future wireless
systems: A survey,’’ IEEE Commun. Surveys Tuts., vol. 20, no. 2,
pp. 836–869, 2nd Quart., 2018, doi: 10.1109/COMST.2017.2787460.
[5] W. Kiess, Y. Xun Gu, S. Thakolsri, M. R. Sama, and S. Beker, ‘‘Sim-
pleCore: A connectionless, best effort, no-mobility-supporting 5G core
architecture,’’ in Proc. IEEE Int. Conf. Commun. Workshops (ICC),
May 2016, pp. 367–372.
[6] G. Liu and D. Jiang, ‘‘5G: Vision and requirements for mobile commu-
nication system towards year 2020,’’ Hindawi Publishing Corp., London,
U.K., White Paper, 2016, vol. 2016, Art. no. 5974586.
[7] P. Fan, ‘‘Advances in broadband wireless communications under high-
mobility scenarios,’’ Chin. Sci. Bull., vol. 59, no. 35, pp. 4974–4975,
Dec. 2014.
[8] M. Khanfouci, ‘‘Distributed mobility management based on centrality
for dense 5G networks,’’ in Proc. Eur. Conf. Netw. Commun. (EuCNC),
Jun. 2017, pp. 1–6.
[9] X. Gelabert, G. Zhou, and P. Legg, ‘‘Mobility performance and suitability
of macro cell power-off in LTE dense small cell HetNets,’’ in Proc. IEEE
18th Int. Workshop Comput. Aided Model. Design Commun. Links Netw.
(CAMAD), Sep. 2013, pp. 99–103.
[10] Cisco. (2016). MME Administration Guide, StarOS Release 20 (Chap-
ter: Mobility Management Entity Overview). [Online]. Available:
https://www.cisco.com
[11] M. Lauridsen, L. C. Gimenez, I. Rodriguez, T. B. Sorensen, and
P. Mogensen, ‘‘From LTE to 5G for connected mobility,’’ IEEE Commun.
Mag., vol. 55, no. 3, pp. 156–162, Mar. 2017.
[12] S. Batabyal and P. Bhaumik, ‘‘Mobility models, traces and impact of
mobility on opportunistic routing algorithms: A survey,’’ IEEE Commun.
Surveys Tuts., vol. 17, no. 3, pp. 1679–1707, 3rd Quart., 2015.
[13] S. Thangam and E. Kirubakaran, ‘‘A survey on cross-layer based
approach for improving TCP performance in multi hop mobile adhoc
networks,’’ in Proc. Int. Conf. Edu. Technol. Comput., Apr. 2009,
pp. 294–298.
[14] E. Spaho, L. Barolli, G. Mino, F. Xhafa, and V. Kolici, ‘‘VANET simu-
lators: A survey on mobility and routing protocols,’’ in Proc. Int. Conf.
Broadband Wireless Comput., Commun. Appl., Oct. 2011.
[15] D. Xenakis, N. Passas, L. Merakos, and C. Verikoukis, ‘‘Mobility man-
agement for femtocells in LTE-advanced: Key aspects and survey of
handover decision algorithms,’’ IEEE Commun. Surveys Tuts., vol. 16,
no. 1, pp. 64–91, 1st Quart., 2014.
[16] J. Wu and P. Fan, ‘‘A survey on high mobility wireless communica-
tions: Challenges, opportunities and solutions,’’ IEEE Access, vol. 4,
pp. 450–476, 2016.
[17] 3GPP. (2020). NR; User Equipment (UE) Procedures in Idle Mode and
in RRC Inactive State. [Online]. Available: https://portal.3gpp.org/
desktopmodules/Specifications/SpecificationDetails.aspx?
specificationId=3192.
183530 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
[18] 3GPP. (2020). NR; Radio Resource Control (RRC); Protocol Spec-
ification. [Online]. Available: https://portal.3gpp.org/desktopmodules/
Specifications/SpecificationDetails.aspx?specificationId=3197.
[19] S. Zhao and Q. Wang, ‘‘A contextual awareness-learning approach to
multi-objective mobility management,’’ in Proc. 12th Int. Conf. Comput.
Sci. Educ. (ICCSE), Aug. 2017, pp. 277–282.
[20] S. Adnan, Y. Fu, N. U. R. Junejo, Z. Chen, and H. Esmaiel, ‘‘Sparse
detection with orthogonal matching pursuit in multiuser uplink quadra-
ture spatial modulation MIMO system,’’ IET Commun., vol. 13, no. 20,
pp. 3472–3478, Dec. 2019.
[21] F. B. Tesema, A. Awada, I. Viering, M. Simsek, and G. P. Fettweis,
‘‘Evaluation of adaptive active set management for multi-connectivity in
intra-frequency 5G networks,’’ in Proc. IEEE Wireless Commun. Netw.
Conf., Apr. 2016, pp. 1–6.
[22] J. Stanczak, ‘‘Mobility enhancements to reduce service interruption time
for LTEand 5G,’’ inProc. IEEE Conf. Standards Commun. Netw.(CSCN),
Oct. 2016, pp. 1–5.
[23] F. B. Tesema, A. Awada, I. Viering, M. Simsek, and G. P. Fettweis,
‘‘Mobility modeling and performance evaluation of multi-connectivity in
5G intra-frequency networks,’’ in Proc. IEEE Globecom Workshops (GC
Wkshps), Dec. 2015, pp. 1–6.
[24] D. Ohmann, A. Awada, I. Viering, M. Simsek, and G. P. Fettweis,
‘‘Achieving high availability in wireless networks by inter-frequency
multi-connectivity,’’ in Proc. IEEE Int. Conf. Commun. (ICC), May 2016,
pp. 1–7.
[25] D. Ohmann, A. Awada, I. Viering, M. Simsek, and G. P. Fettweis, ‘‘Impact
of mobility on the reliability performance of 5G multi-connectivity
architectures,’’ in Proc. IEEE Wireless Commun. Netw. Conf. (WCNC),
Mar. 2017, pp. 1–6.
[26] F. B. Tesema, A. Awada, I. Viering, M. Simsek, and G. P. Fettweis,
‘‘Fast cell select for mobility robustness in intra-frequency 5G ultra dense
networks,’’ in Proc. IEEE 27th Annu. Int. Symp. Pers., Indoor, Mobile
Radio Commun. (PIMRC), Sep. 2016, pp. 1–7.
[27] M. Boujelben, S. B. Rejeb, and S. Tabbane, ‘‘A novel mobility-based
COMP handover algorithm for LTE—A/5G HetNets,’’ in Proc. 23rd
Int. Conf. Softw., Telecommun. Comput. Netw. (SoftCOM), Sep. 2015,
pp. 143–147.
[28] S. Barbera, K. I. Pedersen, C. Rosa, P. H. Michaelsen,
F. Frederiksen, E. Shah, and A. Baumgartner, ‘‘Synchronized RACH-less
handover solution for LTE heterogeneous networks,’’ in Proc. Int. Symp.
Wireless Commun. Syst. (ISWCS), Aug. 2015, pp. 755–759.
[29] Further Details of RACH-Less Handover, document R2-164239, Qual-
comm 3Gpp TSG RAN2 Meeting #94, 2016.
[30] H. Li and D. Hu, ‘‘Mobility prediction based seamless RAN-cache
handover in HetNet,’’ in Proc. IEEE Wireless Commun. Netw. Conf.,
Apr. 2016, pp. 1–7.
[31] M. Chen, Y. Hao, L. Hu, K. Huang, and V. K. N. Lau, ‘‘Green and
mobility-aware caching in 5G networks,’’ IEEE Trans. Wireless Com-
mun., vol. 16, no. 12, pp. 8347–8361, Dec. 2017.
[32] P. Kela, X. Gelabert, J. Turkka, M. Costa, K. Heiska, K. Leppanen,
and C. Qvarfordt, ‘‘Supporting mobility in 5G: A comparison between
massive MIMO and continuous ultra dense networks,’’ in Proc. IEEE Int.
Conf. Commun. (ICC), May 2016, pp. 1–6.
[33] L. L. Vy, L.-P. Tung, and B.-S.-P. Lin, ‘‘Big data and machine learning
driven handover management and forecasting,’’ in Proc. IEEE Conf.
Standards Commun. Netw. (CSCN), Sep. 2017, pp. 214–219.
[34] U. Karneyenka, K. Mohta, and M. Moh, ‘‘Location and mobility
aware resource management for 5G cloud radio access networks,’’ in
Proc. Int. Conf. High Perform. Comput. Simulation (HPCS), Jul. 2017,
pp. 168–175.
[35] B. J. Frey and D. Dueck, ‘‘Clustering by passing messages between data
points,’’ Amer. Assoc. Advancement Sci., vol. 315, no. 5814, pp. 972–976,
2007.
[36] B. Dang, R. Prasad, I. Niemegeers, M. Larrode, and A. J. Koonen,
‘‘Toward a seamless communication architecture for in-building networks
at the 60 GHz band,’’ in Proc. 31st IEEE Conf. Local Comput. Netw.,
Nov. 2006, pp. 300–307.
[37] N. Pleros, K. Tsagkaris, and N. Tselikas, ‘‘A moving extended cell con-
cept for seamless communication in 60 GHz radio-over-fiber networks,’’
IEEE Commun. Lett., vol. 12, no. 11, pp. 852–854, Nov. 2008.
[38] F. A. Hossain and A. M. Chowdhury, ‘‘User mobility prediction based
handoff scheme for 60 GHz radio over fiber network,’’ in Proc. IEEE
11th Consum. Commun. Netw. Conf. (CCNC), Jan. 2014, pp. 557–562.
[39] M. Bagaa, T. Taleb, and A. Ksentini, ‘‘Efficient tracking area management
framework for 5G networks,’’ IEEE Trans. Wireless Commun., vol. 15,
no. 6, pp. 4117–4131, Jun. 2016.
[40] S. Hailu and M. Saily, ‘‘Hybrid paging and location tracking scheme
for inactive 5G UEs,’’ in Proc. Eur. Conf. Netw. Commun. (EuCNC),
Jun. 2017, pp. 1–6.
[41] Discussion of RRC States in NR, Nokia, Espoo, Finland, 2016.
[42] S. Ikeda, N. Kami, and T. Yoshikawa, ‘‘Adaptive mobility management
in cellular networks with multiple model-based prediction,’’ in Proc.
9th Int. Wireless Commun. Mobile Comput. Conf. (IWCMC), Jul. 2013,
pp. 473–478.
[43] N. B. Prajapati and D. R. Kathiriya, ‘‘Dynamic location area planning
in cellular network using Apriori algorithm,’’ in Proc. Int. Conf. Ind.
Instrum. Control (ICIC), May 2015, pp. 660–662.
[44] J. Han and M. Kamber, Data Mining: Concepts and Techniques.
San Francisco, CA, USA: Morgan Kaufmann, 2006.
[45] O. Onireti, A. Imran, M. A. Imran, and R. Tafazolli, ‘‘Energy efficient
inter-frequency small cell discovery in heterogeneous networks,’’ IEEE
Trans. Veh. Technol., vol. 65, no. 9, pp. 7122–7135, Sep. 2016.
[46] D. Calabuig, E. Ternon, V. Venkatasubramanian, M. Maternia,
S. Barmpounakis, S. Gimenez, A. Kousaridas, T. R. Lakshmana,
J. Lorca, P. Lunden, Z. Ren, and P. Sroka, ‘‘Resource and mobility
management in the network layer of 5G cellular ultra-dense networks,’’
IEEE Commun. Mag., vol. 55, no. 6, pp. 162–169, Jun. 2017.
[47] M. Boujelben, S. Ben Rejeb, and S. Tabbane, ‘‘A novel green handover
self-optimization algorithm for LTE—A/5G HetNets,’’ in Proc. Int. Wire-
less Commun. Mobile Comput. Conf. (IWCMC), Aug. 2015, pp. 413–418.
[48] J. Ye, Y. He, X. Ge, and M. Chen, ‘‘Energy efficiency analysis of 5G ultra-
dense networks based on random way point mobility models,’’ in Proc.
19th Int. Symp. WirelessPers. Multimedia Commun. (WPMC), Nov.2016,
pp. 177–182.
[49] A. S. Mubarak, O. A. Omer, H. Esmaiel, and U. S. Mohamed, ‘‘Geometry
aware scheme for initial access and control of mmWave communications
in dynamic environments,’’ in Proc. Int. Conf. Adv. Intell. Syst. Inform.,
2019, pp. 760–769.
[50] M. Giordani and M. Zorzi, ‘‘Improved user tracking in 5G millime-
ter wave mobile networks via refinement operations,’’ in Proc. 16th
Annu. Medit. Ad Hoc Netw. Workshop (Med-Hoc-Net), Jun. 2017,
pp. 1–8.
[51] A. Abdelreheem, E. M. Mohamed, and H. Esmaiel, ‘‘Location-based
millimeter wave multi-level beamforming using compressive sensing,’’
IEEE Commun. Lett., vol. 22, no. 1, pp. 185–188, Jan. 2018.
[52] A. S. Mubarak, H. Esmaiel, and E. M. Mohamed, ‘‘LTE/Wi-Fi/mmWave
RAN-level interworking using 2C/U plane splitting for future 5G net-
works,’’ IEEE Access, vol. 6, pp. 53473–53488, 2018.
[53] J. S. Kim, W. J. Lee, and M. Y. Chung, ‘‘A multiple beam management
scheme on 5G mobile communication systems for supporting high mobil-
ity,’’ in Proc. Int. Conf. Inf. Netw. (ICOIN), Jan. 2016, pp. 260–264.
[54] B. Lamnoo, D. Colle, M. Pickavet, and P. Demeester, ‘‘Radio-over-fiber-
based solution to provide broadband Internet access to train passengers
[topics in optical communications],’’ IEEE Commun. Mag., vol. 45, no. 2,
pp. 56–62, Feb. 2007.
[55] C. Song, Z. Qu, N. Blumm, and A.-L. Barabasi, ‘‘Limits of predictabil-
ity in human mobility,’’ Science, vol. 327, no. 5968, pp. 1018–1021,
Feb. 2010.
[56] H. Farooq, A. Asghar, and A. Imran, ‘‘Mobility prediction-based
autonomous proactive energy saving (AURORA) framework for emerg-
ing ultra-dense networks,’’ IEEE Trans. Green Commun. Netw., vol. 2,
no. 4, pp. 958–971, Dec. 2018.
[57] A. Mohamed, O. Onireti, S. A. Hoseinitabatabaei, M. Imran, A. Imran,
and R. Tafazolli, ‘‘Mobility prediction for handover management in cel-
lular networks with control/data separation,’’ in Proc. IEEE Int. Conf.
Commun. (ICC), Jun. 2015, pp. 3939–3944.
[58] A. Mohamed, O. Onireti, M. A. Imran, A. Imran, and R. Tafazolli,
‘‘Predictive and core-network efficient RRC signalling for active state
handover in RANs with Control/Data separation,’’ IEEE Trans. Wireless
Commun., vol. 16, no. 3, pp. 1423–1436, Mar. 2017.
[59] D. Zhang, D. Zhang, H. Xiong, L. T. Yang, and V. Gauthier, ‘‘NextCell:
Predicting location using social interplay from cell phone traces,’’ IEEE
Trans. Comput., vol. 64, no. 2, pp. 452–463, Feb. 2015.
[60] A. Mohamed, M. A. Imran, P. Xiao, and R. Tafazolli, ‘‘Memory-full
context-aware predictive mobility management in dual connectivity 5G
networks,’’ IEEE Access, vol. 6, pp. 9655–9666, 2018.
VOLUME 8, 2020 183531
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
[61] J. Yang, C. Dai, and Z. Ding, ‘‘A scheme of terminal mobility prediction
of ultra dense network based on SVM,’’ in Proc. IEEE 2nd Int. Conf. Big
Data Anal. (ICBDA), Mar. 2017, pp. 837–842.
[62] X. Chen, F. Meriaux, and S. Valentin, ‘‘Predicting a user’s next cell
with supervised learning based on channel states,’’ in Proc. IEEE 14th
Workshop Signal Process. Adv. Wireless Commun. (SPAWC), Jun. 2013,
pp. 36–40.
[63] S. Premchaisawatt and N. Ruangchaijatupon, ‘‘Enhancing indoor posi-
tioning based on partitioning cascade machine learning models,’’ in Proc.
11th Int. Conf. Electr. Eng./Electron., Comput., Telecommun. Inf. Technol.
(ECTI-CON), May 2014, pp. 1–9.
[64] N. Sinclair, D. Harle, I. A. Glover, J. Irvine, and R. C. Atkinson,
‘‘An advanced SOM algorithm applied to handover management within
LTE,’’ IEEE Trans. Veh. Technol., vol. 62, no. 5, pp. 1883–1894,
Jun. 2013.
[65] L. Song, D. Kotz, R. Jain, and X. He, ‘‘Evaluating next-cell predictors
with extensive Wi-Fimobility data,’’ IEEE Trans. Mobile Comput., vol. 5,
no. 12, pp. 1633–1649, Dec. 2006.
[66] Y. Cheng, Y. Qiao, and J. Yang, ‘‘An improved Markov method for pre-
diction of user mobility,’’ in Proc. 12th Int. Conf. Netw. Service Manage.
(CNSM), Oct. 2016, pp. 394–399.
[67] Y. Qiao, J. Yang, H. He, Y. Cheng, and Z. Ma, ‘‘User location prediction
with energy efficiency model in the long term-evolution network,’’
Int. J. Commun. Syst., vol. 29, no. 14, pp. 2169–2187,
Sep. 2016.
[68] A. Li, Q. Lv, Y. Qiao, and J. Yang, ‘‘Improving mobility prediction per-
formance with state based prediction method when the user departs from
routine,’’ in Proc. IEEE Int. Conf. Big Data Anal. (ICBDA), Mar. 2016,
pp. 1–7.
[69] G. Karagiannis, A. Jamakovic, K. Briggs, M. Karimzadeh, C. Parada,
M. I. Corici, T. Taleb, A. Edmonds, and T. M. Bohnert, ‘‘Mobility
and bandwidth prediction in virtualized LTE systems: Architecture and
challenges,’’ in Proc. Eur. Conf. Netw. Commun. (EuCNC), Jun. 2014,
pp. 1–5.
[70] M. Karimzadeh, Z. Zhao, L. Hendriks, R. de O. Schmidt, S. la Fleur,
H. van den Berg, A. Pras, T. Braun, and M. J. Corici, ‘‘Mobility
and bandwidth prediction as a service in virtualized LTE systems,’’
in Proc. IEEE 4th Int. Conf. Cloud Netw. (CloudNet), Oct. 2015,
pp. 132–138.
[71] P. Fazio, M. Tropea, F. D. Rango, and M. Voznak, ‘‘Pattern prediction
and passive bandwidth management for hand-over optimization in QoS
cellular networks with vehicular mobility,’’ IEEE Trans. Mobile Comput.,
vol. 15, no. 11, pp. 2809–2824, Nov. 2016.
[72] A. Hadachi, O. Batrashev, A. Lind, G. Singer, and E. Vainikko,
‘‘Cell phone subscribers mobility prediction using enhanced Markov
chain algorithm,’’ in Proc. IEEE Intell. Vehicles Symp., Jun. 2014,
pp. 1049–1054.
[73] H. Farooq and A. Imran, ‘‘Spatiotemporal mobility prediction in proactive
self-organizing cellular networks,’’ IEEE Commun. Lett., vol. 21, no. 2,
pp. 370–373, Feb. 2017.
[74] A. B. Cheikh, M. Ayari, R. Langar, G. Pujolle, and L. A. Saidane,
‘‘Optimized handoff with mobility prediction scheme using HMM for
femtocell networks,’’ in Proc. IEEE Int. Conf. Commun. (ICC), Jun. 2015,
pp. 3448–3453.
[75] A. Ben Cheikh, M. Ayari, R. Langar, and L. A. Saidane, ‘‘OHMP-CAC:
Optimized handoff scheme based on mobility prediction and QoS con-
straints for femtocell networks,’’ in Proc. Int. Wireless Commun. Mobile
Comput. Conf. (IWCMC), Sep. 2016, pp. 936–941.
[76] W.-S. Soh and H. S. Kim, ‘‘Dynamic bandwidth reservation in cellular
networks using road topology based mobility predictions,’’ in Proc. IEEE
INFOCOM, Mar. 2004, pp. 2766–2777.
[77] M. R. Tabany and C. G. Guy, ‘‘A mobility prediction scheme of
LTE/LTE—A femtocells under different velocity scenarios,’’ in Proc.
IEEE 20th Int. Workshop Comput. Aided Model. Design Commun. Links
Netw. (CAMAD), Sep. 2015, pp. 318–323.
[78] N.-D. Hoang, N.-H. Nguyen, and K. Sripimanwat, ‘‘Cell selection
schemes for femtocell-to-femtocell handover deploying mobility pre-
diction and downlink capacity monitoring in cognitive femtocell net-
works,’’ in Proc. TENCON-IEEE Region 10th Conf., Oct. 2014,
pp. 1–5.
[79] L. Xia, L.-G. Jiang, and C. He, ‘‘A novel fuzzy logic vertical handoff
algorithm with aid of differential prediction and pre-decision method,’’
in Proc. IEEE Int. Conf. Commun., Jun. 2007, pp. 5665–5670.
[80] H. Song, X. Fang, and L. Yan, ‘‘Handover scheme for 5G C/U plane
split heterogeneous network in high-speed railway,’’ IEEE Trans. Veh.
Technol., vol. 63, no. 9, pp. 4633–4646, Nov. 2014.
[81] B. Li, H. Zhang, and H. Lu, ‘‘User mobility prediction based on
Lagrange’s interpolation in ultra-dense networks,’’ in Proc. IEEE 27th
Annu. Int. Symp. Pers., Indoor, Mobile Radio Commun. (PIMRC),
Sep. 2016, pp. 1–6.
[82] A. Merwaday, I. Guvenc, W. Saad, A. Mehbodniya, and F. Adachi,
‘‘Sojourn time-based velocity estimation in small cell Poisson networks,’’
IEEE Commun. Lett., vol. 20, no. 2, pp. 340–343, Feb. 2016.
[83] A. Merwaday and I. Güvenç, ‘‘Handover count based velocity estimation
and mobility state detection in dense HetNets,’’ IEEE Trans. Wireless
Commun., vol. 15, no. 7, pp. 4673–4688, Jul. 2016.
[84] A. Klein, A. Rauch, R. R. Sattiraju, and H. D. Schotten, ‘‘Achievable
performance gains using movement prediction and advanced 3D sys-
tem modeling,’’ in Proc. IEEE 79th Veh. Technol. Conf. (VTC Spring),
May 2014, pp. 1–5.
[85] W.-S. Soh and H. S. Kim, ‘‘QoS provisioning in cellular networks based
on mobility prediction techniques,’’ IEEE Commun. Mag., vol. 41, no. 1,
pp. 86–92, Jan. 2003.
[86] W.-S. Soh and H. S. Kim, ‘‘A predictive bandwidth reservation scheme
using mobile positioning and road topology information,’’ IEEE/ACM
Trans. Netw., vol. 14, no. 5, pp. 1078–1091, Oct. 2006.
[87] A. Taufique, M. Jaber, A. Imran, Z. Dawy, and E. Yacoub, ‘‘Planning
wireless cellular networks of future: Outlook, challenges and opportuni-
ties,’’ IEEE Access, vol. 5, pp. 4821–4845, 2017.
[88] H. Chung, J. Kim, G. Noh, B. Hui, I. Kim, Y. Choi, C. Choi, M. Lee, and
D. K. Kim, ‘‘From architecture to field trial: A millimeter wave based
MHN system for HST communications toward 5G,’’ in Proc. Eur. Conf.
Netw. Commun. (EuCNC), Jun. 2017, pp. 1–5.
[89] Evolved Universal Terrestrial Radio Access (E-UTRA); Study on Mobile
Relay, document 36.936, 3GPP, 2017.
[90] J. Heinonen, P. Korja, T. Partti, H. Flinck, and P. Pöyhönen, ‘‘Mobility
management enhancements for 5G low latency services,’’ in Proc. IEEE
ICC, 3rd Workshop 5G Archit., May 2016, pp. 68–73.
[91] H. Wang, S. Chen, M. Ai, and H. Xu, ‘‘Localized mobility management
for 5G ultra dense network,’’ IEEE Trans. Veh. Technol., vol. 66, no. 9,
pp. 8535–8552, Sep. 2017.
[92] A. S. Mubarak, O. A. Omer, H. Esmaiel, and U. S. Mohamed, ‘‘Backhaul
overhead traffic reduction in dense mmWave heterogeneous networks
towards 5G cellular systems,’’in Proc. 36th Nat. Radio Sci. Conf. (NRSC),
Apr. 2019, pp. 234–241.
[93] Y. Bi, G. Han, C. Lin, M. Guizani, and X. Wang, ‘‘Mobility management
for Intro/Inter domain handover in software-defined networks,’’ IEEE
J. Sel. Areas Commun., vol. 37, no. 8, pp. 1739–1754,
Aug. 2019.
[94] M. Manalastas and A. Imran, ‘‘Where to go next? A realistic evaluation of
AI-assisted mobility prediction for HetNets,’’ submitted for publication,
IEEE Trans. Veh. Technol., 2020.
[95] F. B. Mismar and B. L. Evans, ‘‘Partially blind handovers for
mmWave new radio aided by Sub-6 GHz LTE signaling,’’ in Proc.
IEEE Int. Conf. Commun. Workshops (ICC Workshops), May 2018,
pp. 1–5.
[96] H. Ryden, J. Berglund, M. Isaksson, R. Coster, and F. Gunnarsson, ‘‘Pre-
dicting strongest cell on secondary carrier using primary carrier data,’’
in Proc. IEEE Wireless Commun. Netw. Conf. Workshops (WCNCW),
Apr. 2018, pp. 137–142.
[97] S. Zang, W. Bao, P. L. Yeoh, B. Vucetic, and Y. Li, ‘‘Managing vertical
handovers in millimeter wave heterogeneous networks,’’ IEEE Trans.
Commun., vol. 67, no. 2, pp. 1629–1644, Feb. 2019.
[98] S. Khunteta and A. K. R. Chavva, ‘‘Deep learning based link failure
mitigation,’’ in Proc. 16th IEEE Int. Conf. Mach. Learn. Appl. (ICMLA),
Dec. 2017, pp. 806–811.
[99] S. M. A. Zaidi, M. Manalastas, A. Abu-Dayyay, and A. Imran, ‘‘AI-
assisted RLF avoidance for smart EN-DC activation,’’ in Proc. IEEE
GlobeComm, 2020.
[100] Z. Ali, N. Baldo, J. Mangues-Bafalluy, and L. Giupponi, ‘‘Machine
learning based handover management for improved QoE in LTE,’’
in Proc. NOMS-IEEE/IFIP Netw. Oper. Manage. Symp., Apr. 2016,
pp. 794–798.
[101] Z. Ali, N. Baldo, J. Mangues-Bafalluy, and L. Giupponi, ‘‘Machine
learning based handover management for improved QoE in LTE,’’ in
Proc. IEEE/IFIP Netw. Oper. Manage. Symp. (NOMS), Istanbul, Turkey,
2016, pp. 794–798, doi: 10.1109/NOMS.2016.7502901.
183532 VOLUME 8, 2020
S. M. A. Zaidi et al.: Mobility Management in Emerging UDNs: A Survey, Outlook, and Future Research Directions
[102] L. Simic, S. Panda, J. Riihijarvi, and P. Mahonen, ‘‘Coverage and robust-
ness of mm-Wave urban cellular networks: Multi-frequency HetNets are
the 5G future,’’ in Proc. 14th Annu. IEEE Int. Conf. Sens., Commun.,
Netw. (SECON), San Diego, CA, USA, Jun. 2017, pp. 1–9.
[103] I. Akbari, O. Onireti, A. Imran, M. A. Imran, and R. Tafazolli, ‘‘How
reliable is MDT-based autonomous coverage estimation in the presence
of user and BS positioning error?’’ IEEE Wireless Commun. Lett., vol. 5,
no. 2, pp. 196–199, Apr. 2016.
[104] Make-Before-Break HO in LTE and its Implications, document R2-
163981, Nokia, Alcatel-Lucent Shanghai Bell, 3GPP TSG RAN2 Meet-
ing #94, Nanjing, China, 2016.
[105] 3GPP. (May 23, 2016). 3gpp.com. [Online]. Available: http://www.
3gpp.org/DynaReport/TDocExMtg–R2-94–31668.htm
[106] D. S. Wickramasuriya, C. A. Perumalla, K. Davaslioglu, and R. D. Gitlin,
‘‘Base station prediction and proactive mobility management in virtual
cells using recurrent neural networks,’’ in Proc. IEEE 18th Wireless
Microw. Technol. Conf. (WAMICON), Apr. 2017, pp. 1–6.
[107] N. P. Kuruvatti and H. D. Schotten, ‘‘Framework to support mobility
context awareness in cellular networks,’’in Proc. IEEE 83rd Veh.Technol.
Conf. (VTC Spring), May 2016, pp. 1–5.
SYED MUHAMMAD ASAD ZAIDI (Graduate
Student Member, IEEE) received the B.Sc. degree
in information and communication engineering
from the National University of Science and
Technology (NUST), Pakistan, in 2008, and the
M.S. degree from Ajou University, South Korea,
in 2013. He is currently pursuing the Ph.D. degree
with the AI4Networks Lab, The University of
Oklahoma-Tulsa, USA, where his researches focus
on mobility robustness and optimization of futuris-
tic ultradense base station deployment. With almost ten years’ experience in
telecom industry, he has worked with Jazz, Pakistan, Korea Electronics and
Technology Institute (KETI), South Korea, MOTiV Research, Japan, AT&T,
USA, and Sprint, USA.
MARVIN MANALASTAS (Member, IEEE)
received the degree in electronics and communica-
tions engineering from the Polytechnic University
of the Philippines, in 2011, and the M.S. degree
in electrical and computer engineering from The
University of Oklahoma-Tulsa, in 2020, where he
is currently pursuing the Ph.D. degree with the
AI4Networks Lab, The University of Oklahoma-
Tulsa, with research interest in machine learning
applied in 5G self-organizing networks. He has
more than seven years of industry experience in the field of telecommu-
nications. He worked as a Microwave Transmission Engineer with Huawei
Technologies Philippines, from 2011 to 2015. Then, he moved to work as a
Radio Network Performance Engineer in Tokyo, Japan, from 2015 to 2018,
focused on LTE improvement and optimization of key locations all over
Japan. He also did an internship as a RF Optimization Engineer for clients,
like AT&T at Dallas, TX, USA, in 2019.
HASAN FAROOQ (Member, IEEE) received
the B.Sc. degree in electrical engineering from
the University of Engineering and Technology,
Lahore, Pakistan, in 2009, the M.Sc. degree in
information technology from Universiti Teknologi
Petronas, Malaysia, in 2014, where his research
focused on developing ad hoc routing protocols
for smart grids, and the Ph.D. degree in electrical
and computer engineering with The University of
Oklahoma-Tulsa, USA. He is currently working
with Ericsson, USA. His research interests include big data empowered
proactive self-organizing cellular networks focusing on intelligent proactive
self-optimization and self-healing in HetNets utilizing dexterous combina-
tion of machine learning tools, classical optimization techniques, stochastic
analysis, and data analytics. He was a recipient of the Internet Society
First Time Fellowship Award toward Internet Engineering Task Force 86th
Meeting, USA, 2013.
ALI IMRAN (Senior Member, IEEE) received
the B.Sc. degree in electrical engineering from
the University of Engineering and Technology,
Lahore, Pakistan, in 2005, and the M.Sc. degree
(Hons.) in mobile and satellite communications
and the Ph.D. degree from the University of Surrey,
Guildford, U.K., in 2007 and 2011, respectively.
He is currently a Presidential Associate Professor
in ECE and the Founding Director of the Artifi-
cial Intelligence (AI) for Networks (AI4Networks)
Research Center and the TurboRAN Testbed for 5G and Beyond, The
University of Oklahoma-Tulsa. His research interests include AI and its
applications in wireless networks and healthcare. His work on these topics
has resulted in several patents and over 100 peer-reviewed articles, including
some of the most influential articles in domain of wireless network automa-
tion. On these topics, he has led numerous multinational projects, given
invited talks/keynotes and tutorials at international forums and advised major
public and private stakeholders and co-founded multiple start-ups. He is
an Associate Fellow of the Higher Education Academy, U.K. He is also a
member of the Advisory Board to the Special Technical Community on Big
Data in the IEEE Computer Society.
VOLUME 8, 2020 183533
