Content uploaded by Andrea Zanella
Author content
All content in this area was uploaded by Andrea Zanella on Sep 04, 2015
Content may be subject to copyright.
Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
HOSTED BY
journal homepage: www.elsevier.com/locate/dcan
Available online at www.sciencedirect.com
The challenges of M2M massive access
in wireless cellular networks
Andrea Biral, Marco Centenaro, Andrea Zanella
n
,
Lorenzo Vangelista, Michele Zorzi
Department of Information Engineering of the University of Padova, Via Gradenigo 6/B, 35131, Padova, Italy
Received 21 January 2015; received in revised form 16 February 2015; accepted 28 February 2015
Available online 27 March 2015
KEYWORDS
M2M;
MTD;
MTC;
Massive access;
5G
Abstract
The next generation of communication systems, which is commonly referred to as 5G, is expected
to support, besides the traditional voice and data services, new communication paradigms, such as
Internet of Things (IoT) and Machine-to-Machine (M2M) services, which involve communication
between Machine-Type Devices (MTDs) in a fully automated fashion, thus, without or with minimal
human intervention. Although the general requirements of 5G systems are progressively taking
shape, the technological issues raised by such a vision are still partially unclear. Nonetheless,
general consensus has been reached upon some specific challenges, such as the need for 5G
wireless access networks to support massive access by MTDs, as a consequence of the proliferation
of M2M services. In this paper, we describe the main challenges raised by the M2M vision, focusing
in particular on the problems related to the support of massive MTD access in current cellular
communication systems. Then we analyze the most common approaches proposed in the literature
to enable the coexistence of conventional and M2M services in the current and next generation of
cellular wireless systems. We finally conclude by pointing out the research challenges that require
further investigation in order to provide full support to the M2M paradigm.
&2015 Chongqing University of Posts and Communications. Production and Hosting by Elsevier B.V.
All rights reserved.
1. Introduction
As telecommunication technologies continue to evolve
rapidly, fueling the growth of service coverage and capacity,
new use cases and applications are being identified. Many of
these new business areas (e.g., smart metering, in-car
satellite navigation, e-health monitoring, smart cities)
involve fully-automated communication between devices,
without human intervention. This new form of communica-
tion is generally referred to as Machine-to-Machine (M2M)
Communication, or Machine-Type Communication (MTC),
while the involved devices are called Machine-Type Devices
(MTD). Examples of common MTDs are environmental and
biomedical sensors, actuators, meters, radio frequency tags
http://dx.doi.org/10.1016/j.dcan.2015.02.001
2352-8648/&2015 Chongqing University of Posts and Communications. Production and Hosting by Elsevier B.V. All rights reserved.
n
Corresponding author.
E-mail address: zanella@dei.unipd.it (A. Zanella).
Peer review under responsibility of Chongqing University of Posts
and Telecommunications.
Digital Communications and Networks (2015) 1,1–19
(RFtags), but also smartphones and tablets, vehicles, cam-
eras, and so on.
The number of MTDs is continuously growing, together
with the set of M2M applications and services that they
enable. As a matter of fact, MTDs are key elements in the
emerging “Internet of Things”and “Smart City”paradigms
[1,2], which are expected to provide solutions to current
and future social-economical demands for sensing and
monitoring services, as well as for new applications and
business models in areas such as building and industrial
automation, remote and mobile healthcare, elderly assis-
tance, intelligent energy management and smart grids,
automotive, smart agriculture, traffic management, and
many others [3].
In the last years, the potential of the M2M paradigm has
been recognized by both academia and industry, which have
generated a number of studies, protocols and products
oriented to the support of M2M services. Worth mentioning
for their popularity and market penetration are the IEEE
802.15.4 standard for low-bitrate short-range transmissions
[4], the 6LowPAN protocol suite for low power devices [5],
the ZigBee solution for MTD interconnection in small wire-
less sensor networks [6], and other communication systems
like EIB/KNX, LON and BACnet for home automation [7].
Another interesting paradigm that can be adapted to M2M
services is the 3GPP Proximity-based Service (ProSe) proto-
col [8]. The basic idea is to offload the network by
exploiting the physical proximity of the terminals, e.g.,
enabling direct communication between user devices, or
limiting the signaling to the local area, without involving
core network elements.
Despite the appeal of such solutions, the potential of the
M2M vision can be fully unleashed only when MTDs con-
nectivity will be possible everywhere, with no (or minimal)
configuration, and without the need for deploying addi-
tional devices, such as gateways or concentrators. As a
matter of fact, the ideal scenario is such that MTDs need
just to be placed in the desired locations to get connected
to the rest of the world. Taking inspiration from the well-
known plug-&-play notion, we refer to this new connectivity
paradigm as place-&-play.
Unfortunately, the technologies that can support some
form of MTC are currently incapable of fulfilling the demand
for ubiquitous access of MTDs to the communication sys-
tems. Local network solutions, such as ZigBee/6LoWPAN or
IEEE 802.11ah extension for MTC, are suitable to intercon-
nect MTDs in the same local area, but are able neither to
offer coverage everywhere, nor to guarantee highly reliable
coordinated control of the network. On the other hand, the
ubiquitous coverage offered by satellite connections has
prohibitively high cost, both economically and in terms of
energy consumption, and pose significant challenges when
used in indoor environments.
The place-&-play concept, hence, calls for terrestrial
radio technologies that are capable of providing widespread
(ideally ubiquitous) coverage, with extremely low energy
consumption, low complexity at the end device, possibly
low latency, and minimal cost per bit. A few proprietary
solutions that satisfy some of these requirements have been
recently commercialized, though the deployment of a new
infrastructure network at a global scale is economically
challenging. Therefore, the most natural and appealing
solution is to add MTC to the services provided by the
existing cellular networks. Indeed, cellular networks have a
world-wide established footprint and are able to deal with
the challenge of ubiquitous and transparent coverage.
Furthermore, the wide-area mobile network access para-
digm offers a number of other advantages over local-area
distributed approaches, such as higher efficiency, robust-
ness and security, thanks to locally coordinated control,
coordinated infrastructure deployment, ease of planning,
performance predictability and the possibility of deploying
advanced MTC-tailored PHY/MAC schemes that shift com-
plexity from MTDs to base stations (BSs).
The MTC paradigm is hence expected to play a significant
role in current and future cellular networks, both as the
enabler of potentially disruptive markets (e.g., Smart Cities
[9]) and for the new challenges that M2M services will pose
to the communication systems. Unfortunately, current
cellular network technologies will likely be unable to cope
with the expected growth of M2M services. Indeed, today's
standards are designed to provide access to a relatively
small number of devices that need to transfer a significant
amount of data, so that the signaling traffic generated by
the management and control plane is basically negligible.
M2M services, instead, are generally expected to involve a
huge number of devices that generate sporadic transmis-
sions of short packets. The risk is then a collapse of current
wireless cellular systems under the weight of the signaling
traffic generated by MTDs [10]. In addition, although
transmissions from MTDs are, in many cases, delay tolerant
(smart metering, telemetry), there is also an important
class of M2M applications that require ultra-low latency
(e-health, vehicular communications). Furthermore, most
MTDs are expected to be severely constrained in terms
of computational and storage capabilities, and energy
capacity.
For all these reasons, the M2M scenario is considered as a
major challenge for next generation wireless cellular sys-
tems, commonly referred to as 5G [11]. In addition to
increased bit rate and energy efficiency of the terminals and
of the whole system, 5G will hence be required to provide
minimal latency to critical services and seamless integration
of Internet of Things (IoT) nodes, and to support massive
M2M communication services, all without degrading the
quality of services like voice, audio/video streaming, and
web browsing, which are referred to as “conventional”
services in the following.
In this paper we survey the main challenges offered to
the current and next generations of wireless cellular
standards by the expected massive diffusion of M2M ser-
vices. To begin with, Section 2 discusses how M2M services
are supported by current cellular standards and highlights
the limits of such solutions. Section 3 addresses in greater
detail the pivotal challenge of massive MTD access in LTE
systems, describing the methods that have been proposed
by 3GPP to counteract the service degradation in case of
overload of the Random Access Channel (RACH) due to
massive MTD access requests. Section 4, instead, discusses
the schemes that have been studied to improve the energy
efficiency and the quality of service (QoS) of MTC. After
that, in Section 5 the issue of cell coverage extension
is addressed. We then consider in Section 6 the studies
that address the massive access problem with a more
A. Biral et al.2
fundamental approach. Section 7 wraps up the paper by
discussing the current state-of-the-art in massive M2M
cellular access and the gaps that need to be filled in the
near future in order to fully support the M2M paradigm, with
specific reference to the advanced features that are envi-
sioned in 5G systems. Finally, Section 8 concludes the paper
with some final considerations.
2. M2M support in current cellular networks
Considering the relatively low Average Revenue Per User
that is expected from MTD-based services, the costs to
provide ubiquitous coverage at global level will become
sustainable only if the volume of the M2M market will
enable economies of scale, which requires the utilization
of the same resources for multiple services and, hence,
the interoperability of the different enabling technologies.
This necessity has been recognized by the “oneM2M”
Partnership Project (PP)
1
that was created in 2012 with
the goal of developing global, access-technology agnostic
Service Layer specifications for M2M. While the oneM2M
initiative can effectively contribute to enable the afore-
mentioned place-&-play paradigm for what concerns the
“play”aspect, i.e., the standard and zero-configuration
access to M2M data and the interoperability of M2M
services, the “place”aspect, which requires ubiquitous
and seamless connectivity of MTDs, still lacks a satisfactory
solution. In this respect, the current cellular systems
are natural and appealing candidates for the support of
widespread MTD wireless access, because of their capillary
geographical coverage, their technological maturity, and
the cost-effectiveness provided by conventional higher-
revenue services, such as voice, video and wideband
data connectivity. Unfortunately, current systems may not
be able to support the expected growth of MTC, so that
proper countermeasures need to be taken, as better
explained in the remainder of this section.
2.1. Revamping GSM for M2M support
The second generation of standards for cellular systems,
i.e., GSM, GPRS, and EDGE, is progressively being replaced
by the third (UMTS/HDSPA) and the fourth (LTE) generations
for what concerns Human-to-Human (H2H) services, which
require higher data-rates, lower power consumption, and
better support to user mobility. In this framework, GSM
becomes an attractive candidate to provide ubiquitous
connectivity to MTDs, which may exploit the pervasive
presence of GSM coverage and the empty space left by
the migration of the conventional services towards 3G and
4G networks. However, considering the scarcity of available
radio frequencies and the always growing demand for new
wireless services, the refarming of GSM frequency bands is
being debated by governments, so that the remaining
operational time for GSM is uncertain.
In addition, several studies show that the GSM radio
access network faces serious capacity issues in the presence
of the synchronized access of a massive number of MTDs
[12–14]. As a consequence, there has been an effort in 3GPP
to further enhance the GSM related standards to facilitate
the support of MTC. The attention has been mostly focused
on improvements in the core network and at the higher
layers of the radio access network, with reference to the
following aspects [15,14]:
identification of MTDs within the network;
Short-Messaging-Service in the Packet Switched domain;
load control through the introduction of extended class
barring;
reduction of the signaling overhead by minimizing the
occurrence of signaling procedures not essential to the
MTD profile.
While these efforts can indeed make GSM a valuable
solution for MTC support in the mid-term, there are a
number of practical considerations and technical limits that
will likely prevent GSM from becoming the ultimate access
technology for MTDs in the long run. To begin with, the
number of MTDs that can be connected to a single GSM base
station (BS) is quite limited. This limit can be alleviated by
tightening the granularity of the transmission resources,
while keeping the backward compatibility with the original
system, as proposed in [14]. In this way, the MTDs that can
be served by a single BS can range between 10
4
and 10
5
,
depending on the requirements in terms of delay tolerance.
However, the energy consumption and the delay of GSM
access may still be prohibitively high for most M2M services.
In addition, operators are reluctant to guarantee that GSM
networks will be operational for the quite long lifetime of
M2M applications. Furthermore, the always growing demand
for mobile wideband services (e.g., video streaming) may
push the operators to re-farm the spectrum currently
allocated to GSM to further increase the capacity of LTE
and/or 5G RANs.
2.2. M2M services <E
Acknowledging the shortcomings of GSM as a long-term
access network for M2M services, some operators are slowly
pushing M2M applications towards UMTS. Compared to GSM,
however, UMTS has a number of disadvantages: because it is
typically used in a higher frequency band, it is more difficult
to get good (esp. indoor) coverage; furthermore, UMTS
modules are more expensive than GSM modules.
A rather natural option is to resort to the latest cellular
standard, i.e., LTE. However, LTE design was mainly
intended to increase the data rate offered to mobile users,
without considering the requirements and traffic character-
istics of typical M2M services. Supporting MTC in the LTE
architecture, hence, raises a number of challenges, includ-
ing control overhead, energy efficiency, coverage extension,
robustness to malfunctioning devices, security, and scal-
ability. However, the most compelling problem regards the
coexistence of M2M and conventional services and, in
particular, the support of massive MTD access without
hampering the quality of conventional services [16].
The problem originates from the Random Access Channel
(RACH), i.e., the transport-layer channel defined by LTE
1
http://www.onem2m.org.
3The challenges of M2M massive access in wireless cellular networks
standard to manage the channel access requests by end-
devices, which is better explained below [17].
2.3. RACH: the random access procedure of LTE
Hereafter, we will use User Equipment (UE) and eNodeB to
refer to end-device and BS, as per LTE terminology.
According to LTE specifications, the random access proce-
dure is triggered by a UE any time it needs to establish (or
re-establish) a data connection with the eNodeB, e.g., for
the association to the network or the synchronization with
the eNodeB after a long idle period; after radio-link failure;
or when changing the serving eNodeB because of a
handover.
Depending on the purpose, the random access procedure
can either be contention-free or contention-based. The first
procedure is not much impacted by M2M traffic, being used
for managing delayed-constrained access requests with high
success requirements, such as those related to handover.
The process is indeed under full control of the eNodeB that
coordinates the access requests of the end-devices in order
to avoid conflicts and to minimize the access delay.
Conversely, the contention-based random access procedure
is much more sensitive to M2M traffic. To better understand
the origin of the problem, it is convenient to describe in
more detail the random access procedure.
The RACH is formed by a sequence of time–frequency
resources, called Random Access (RA) slots. UEs are allowed
to transmit their access requests only in RA slots by using
specific orthogonal preambles, which are called “signa-
tures.”The time–frequency resource on which the RA
preamble is transmitted is known as the Physical Random
Access Channel (PRACH), which is time and frequency
multiplexed with the Physical Uplink Shared Channel
(PUSCH), as illustrated in Fig. 2.
In each LTE cell, there are 64 preambles, created by the
so-called Zadoff–Chu sequence. Some preambles are
reserved for contention-free (or coordinated) RA, while
the remaining ones are used for contention-based (or
uncoordinated) RA. This second set of preambles is further
divided into two groups: Group A preambles are intended
for sending small packets and Group B preambles are
intended for sending large packets.
In the frequency domain, the PRACH resource has a
bandwidth corresponding to 6 resource blocks (1.08 MHz).
Instead, the periodicity of RA slots, the total number of
random access preambles available for contention-based
random access, the total number of random access pre-
amble sequences available within Group A, the maximum
message size allowed for such preambles, and other para-
meters related to the RACH are broadcast to the UE in the
system information block 2 (SIB2).
Fig. 1 illustrates the logical sequence associated to a
contention-based RA procedure, which develops along the
following steps .
Fig. 1 Contention-based Random Access procedure in LTE.
Fig. 2 Illustration of PRACH transmission resources in LTE.
A. Biral et al.4
Message 1[Preamble Transmission]: The UE randomly
chooses one of the preambles (P) reserved for
contention-based RA and transmits its request in the
first available RA slot. Because of the orthogonality of
the different preambles, multiple UEs can transmit their
access requests in the same RA slot, using different
preambles. In this case, the eNodeB is able to decode
the requests and estimate the transmission timing of the
terminals. Conversely, if two or more devices transmit
the same preamble, a collision occurs and the corre-
sponding requests will not be detected by the eNodeB
(see Fig. 3). However, it is also possible that multiple UEs
choose the same preamble and the eNodeB correctly
detects it (e.g., because one is much stronger than the
others or the different signals appear as a single trans-
missions going through multiple fading paths). In this
case, the acknowledgement sent by the eNodeB will
trigger a transmission by multiple UEs and a collision will
occur at the third step of the handshake.
Message 2[Random Access Response]: For each success-
fully decoded request, the eNodeB transmits a RA
response on the Physical Downlink Shared Channel
(PDSCH), which includes a Timing Alignment (TA) com-
mand to adjust the terminal transmit timing, and the
resources on the uplink channel (PUSCH resources) that
have been assigned to UEs for the third step of the RA
procedure. This message, furthermore, carries an optional
backoff indicator (BCK) that is used to reduce the
probability of further collisions in successive attempts by
UEs that collided in the previous RA slot. Indeed, UEs
whose Message 1 is not acknowledged by Message 2 within
aspecific time window wait for a random backoff interval
before starting again the RACH procedure in the next
available RA slot. Moreover, if the counter of consecutive
unsuccessful preamble transmission attempts exceeds a
certain threshold, a Random Access problem message is
indicated to upper layers.
Message 3[Connection Request]: Upon receiving the ran-
dom access response, UEs will transmit a Radio Resource
Control(RRC)messageonthereserved resources in the
PUSCH. Note that UEs involved in an undetected preamble
collision will transmit over the same PUSCH resource
blocks, thus generating a new collision (see Fig. 4).
Message 4[Contention Resolution]: If the connection
request of the UE is successfully detected by eNodeB, it
replies with a contention-resolution message on the
PDSCH, echoing the mobile terminal identity (mobile
ID) in order to acknowledge the correct reception of its
request. Conversely, UEs that do not receive a contention
resolution message from the eNodeB assume that their
access requests failed and, after waiting a random
backoff time, perform a new preamble transmission
attempt in the next RA slot. Again, when the number
of unsuccessful attempts reaches a certain value, the
network is declared unavailable by the UE and an access
problem exception is raised to the upper layers.
Fig. 3 Collision event in Message 1.
5The challenges of M2M massive access in wireless cellular networks
2.4. The threat of PRACH overload
The RACH procedure has been identified by 3GPP as a key
challenging task [16] for M2M communications because of
the signaling and traffic load spikes caused by a sudden
surge of the number of M2M devices trying to access the
same base station simultaneously (e.g., a huge number of
smart meters becoming active almost at the same time
after a period of power outage).
The risk, hence, is that massive access requests by MTDs
can overload the PRACH, yielding an increase of the
contention probability and, in turn, of the access delay
and failure rate. A possibility to reduce the load of the
PRACH is to increase the number of access opportunities
scheduled per frame, but this determines a reduction of the
amount of resources available for data transmission and,
hence, a contraction of the data transport capacity of the
uplink channel. Furthermore, the total amount of RA slots
that can be allocated in an LTE frame is limited. Finally, the
processing of the Zadoff–Chu sequence, which is employed
in the LTE RACH preambles, is computationally demanding
and can be a further issue for resource-constrained devices,
such as MTDs.
Summing up, the standard LTE procedure for managing
channel access requests by end-devices will not properly
scale in the presence of massive access attempts by a large
number of MTDs, which may result in a sharp degradation of
the quality offered to conventional services because of long
access delay and high access failure rate. Of course, M2M
services are also affected by these impairments, though the
impact may be less significant with respect to conventional
services. Nonetheless, the overload of the PRACH impacts
MTDs on other aspects, such as the energy consumption and
computational effort, which are generally critical for MTD
applications.
3. Standard schemes to alleviate the PRACH
overload problem
Coexistence and management of human-triggered and
machine-triggered traffic is a major challenge for next
generation cellular networks.
2
In particular, the contention
to access PRACH resources may lead to dramatic degrada-
tion of H2H services. For this reason, the PRACH overload
problem has been attracting the attention of the scientific
community, and many possible methods to improve the
RACH procedure have been proposed [10]. Most of these
Fig. 4 Collision event in Message 3.
2
Since the access requirements are mainly determined by the
class of the service initiator, with a slight abuse of terminology in
the following we use H2H and M2M to refer to human-triggered and
machine-triggered services, respectively, whatever the actual
nature of the destination.
A. Biral et al.6
methods provide some form of separation between access
requests originated by H2H and M2M services, with the aim
of shielding the former from the PRACH overload issues that
can be generated by the latter. The various approaches
differ in the way this separation is enforced. We hence
distinguish between “strict”schemes, in which the pool of
access resources is deterministically split between H2H and
M2M, thus achieving perfect isolation between the two
types of access requests; and “soft”schemes, where H2H
and M2M share the same resources, but with different
access probabilities. These two approaches can also be
combined, giving rise to “hybrid”schemes.
The remainder of this section briefly discusses the
different approaches that have been proposed to counter-
act PRACH overload, most of which appeared in 3GPP
technical documents such as [18]. The interested reader is
also referred to [10] for a different classification of (most
of) these schemes.
3.1. Strict-separation schemes
As mentioned, strict-separation schemes achieve perfect
isolation between H2H and M2M access requests by allocat-
ing different physical resources to UEs and MTDs. In this
category, we can list the following schemes.
(1) Resource separation: The simplest and most immedi-
ate way to shield H2H from the risk of access request
collisions due to massive MTC requests is to assign orthogo-
nal PRACH resources to H2H and M2M devices. The separa-
tion of resources can be done by either splitting the
preambles into H2H and MTC groups, or by allocating
different RA slots in time and/or frequency to the two
categories of terminals [18]. This solution, however, can
yield suboptimal performance when the number of
resources assigned to each class of devices does not reflect
the actual demand.
To be effective, this scheme needs to be coupled with
mechanisms to dynamically shift resources among the two
classes, according to the respective access request rates. In
some scenarios, the network can predict sharp increments
of the access load due to MTDs, e.g., using the Self-
Optimizing Overload Control (SOOC) scheme proposed in
[16] and described later under the “hybrid”category.
(2) Slotted access: This scheme was proposed by 3GPP in
[18]. It consists in defining access cycles (similar to paging
cycles), which contain RA slots dedicated to MTD access
requests. Each MTD can only access its dedicated RA slots,
in a collision-free manner. The RA slots reserved to each
MTD in a cycle are determined from the unique identifier of
the devices (namely, the International Mobile Subscriber
Identity –IMSI) and the RA cycle parameter broadcast by the
eNodeB. While this scheme protects H2H devices from MTC,
the allocation of dedicated RA slots to each MTD may yield
very long RA cycles and, hence, long access latency, which
may not be compatible with the service requirements of
delay-constrained M2M applications (e.g., alarms).
(3) Pull-based scheme: This is a centralized mechanism
that allows MTDs to access the PRACH only upon being
paged by the eNodeB [18]. Paging is triggered by the MTC
server that is assumed to know in advance when MTDs need
to establish a radio link connection, to either send or
receive data. The eNodeB can control the paging taking
into account the network load condition, thus preventing
PRACH overload. This is already supported by the current
specification. The paging message may also include a back-
off time for the MTDs, which indicates the time of access
from the reception of the paging message. This approach is
suitable to manage channel access of MTDs with regular
traffic patterns. However, its centralized nature limits the
number of MTDs and M2M services that can be managed by a
single M2M server. Furthermore, the scheme cannot deal
with an unexpected surge of MTD access requests.
3.2. Soft-separation schemes
In soft-separation schemes there is no neat separation of
access resources between M2M and H2H, rather all devices
can use the same resources but with different probabilities.
Therefore, the separation between MTDs and classic UEs is
achieved in a statistical sense. The main schemes based on
this approach are described below.
(1) Backoff tuning: A way to smoothly decrease the rate
of channel access requests by MTDs in case of congestion is
to assign longer backoff intervals to MTDs that fail the
transmission of Message 1 in the RACH procedure [18].
Although this method can alleviate the contention between
H2H and M2M devices in case of peaks of MTD requests, it is
not very effective when dealing with stationary MTDs
massive access, due to the instability issue that charac-
terizes ALOHA-like access mechanisms.
(2) Access Class Barring: The backoff tuning scheme is
generalized by the Access Class Barring (ACB) method,
which is actually part of LTE and LTE-A specifications. ACB
makes it possible to define multiple access classes with
different access probabilities [18]. Each class is assigned an
access probability factor and a barring timer. The devices
belonging to a certain access class are allowed to transmit
Message 1 in a RA slot only if they draw a random number
that is lower than the access probability factor. Otherwise,
the access is barred and the devices have to wait for a
random backoff time, which is determined according to the
barring timer of that class, before attempting a new access.
The ACB parameters are broadcast by the eNodeB as part of
the system information. Furthermore, 3GPP proposed the
Extended Access Barring (EAB) scheme, which is a method
for the network to selectively control access attempts from
UEs that can tolerate longer access delays or higher failure
probability [18]. These devices will hence be barred in case
of overload of the access and/or the core network, without
the need to introduce any new Access Classes. These
mechanisms can be used to alleviate the MTD massive
access issue by defining a special class for MTDs, with lower
access probability factor and/or longer barring timer,
or labelling MTDs as EAB devices. However, MTDs with
delay-constrained access requirements can be associated
to classes with higher access probability and lower
barring timer.
ACB mechanisms are quite effective in preventing PRACH
overload, but at the cost of longer access delay for MTDs.
Furthermore, ACB does not solve the access contention
problem when many delay-constrained MTDs need to access
the channel in a short time interval, as the result of certain
7The challenges of M2M massive access in wireless cellular networks
events (e.g., alarms triggered by unexpected events, such
as failures of the power grid, earthquakes, and flooding).
Nonetheless, ACB mechanisms can be combined with other
techniques to counteract the PRACH overload due to
massive MTD access.
3.3. Hybrid schemes and other solutions
Here we discuss the solutions that cannot be classified as
either strict- or soft-separation schemes, since they include
aspects from both families or are based on totally different
approaches.
(1) Self-Optimizing Overload Control: SOOC is a compo-
site scheme presented in [16] to counteract PRACH overload
by combining many of the schemes described above,
including PRACH resource separation, ACB, and slotted-
access schemes. The fundamental feature of the SOOC
scheme is the execution of a control loop to collect
information for overload monitoring at each RA cycle. Then,
based on such data, the eNodeB adapts the number of RA
slots in the random access cycles.
More specifically, when a device is not able to get an
access grant at the first attempt, it enters the overloaded
control mode. In this status, the classical p-persistent
mechanism is applied in order to regulate RA retries for
collided terminals. Besides, in order to distinguish between
time-tolerant MTDs and time-sensitive MTDs, two access
classes are added to the LTE-A ACB scheme for M2M devices
(namely, low access priority and high access priority) and
different pparameters are set according to the access class
of the terminal.
In order to monitor the congestion level of the system,
when a terminal receives Message 2 in the contention-based
RA procedure (see Section 2.3), it includes a PRACH over-
load indicator, which contains the number of RA retries
attempted by the device, within Message 3. Based on this
information, the eNodeB reacts by dynamically increasing or
decreasing the number of PRACH RA slots in the successive
cycle in order to maintain a target maximum collision
probability for the system. Moreover, in the borderline case
when the number of RA slots can not be further increased
due to insufficient uplink radio resources, the eNodeB can
deny access to low priority MTDs until the overload condi-
tion improves.
Unfortunately, although the goal of handling high traffic
loads is clear and the proposed scheme surely goes in this
direction, [16] only describes SOOC theoretically and no
performance results have been presented.
(2) Random access scheme for fixed-location M2M com-
munication: When MTDs are static, the fixed uplink Timing
Alignment (TA) between the MTDs and the BS can be
exploited in the resource allocation procedure, as proposed
in [19], where the focus is on a large class of motionless
MTDs (e.g., smart meters). In this context, authors propose
an energy-efficient RA scheme to reduce collision probabil-
ity and average access delay. The procedure consists of
5 steps:
1. The device randomly chooses a preamble out of M
orthogonal ones and transmits it in a certain slot
(Message 1).
2. The BS detects which preamble is transmitted, deter-
mines the proper TA value, and broadcasts a resource
allocation response that contains several parameters,
including TA information (Message 2).
3. The device compares the TA value carried by the
response with its own TA value: in case of matching the
handshake goes on with step 4, otherwise, the MTD
performs a retransmission after a random backoff time.
4. The MTD synchronizes the uplink transmission time to the
received TA information and sends a Radio Resource
Control message on the allocated uplink resource
(Message 3).
5. If the message is successfully decoded, the BS sends an
acknowledgement message (Message 4). Otherwise, the
MTD needs to repeat the access request procedure.
Note that the procedure resembles the conventional LTE
resource allocation process described in Section 2.3 except
for step 3, which takes advantages of the TA information to
reduce the collision probability in the transmission of
Message 3. Indeed, the TA value of a fixed-location MTD is
expected to remain constant over time. If the TA received
from the eNodeB in Message 2 does not match with the
expected TA of the MTD, then there is a high probability
that Message 2 is actually intended for a different MTD that
transmitted on the same PRACH. In this case, according to
step 3, the MTD will avoid transmission of Message 3,
thereby reducing the probability of collision at step 4 and,
in turn, the access delay.
(3) Bulk MTC signaling scheme: Another possible solution
to congestion/overload events may be to enable bulk MTC
signaling handling, as stated in [20], where the authors
remark the lack of mechanisms to simultaneously handle the
overhead generated from a group of MTDs. Therefore, under
the assumption that signaling messages from MTDs are
moderately delay tolerant, it may be convenient to mini-
mize the overhead at the BS by exploiting bulk processing,
i.e., aggregating signaling data coming from MTDs before
forwarding them to the core network. For example, con-
sider the case in which a group of MTDs are triggered to
send a Tracking Area Update (TAU) message: the BS could
wait a default timeout interval or until it gathers a
sufficient number of signaling messages to forward a single
aggregate message towards the Mobility Management Entity
(MME). Indeed, since the MTDs are associated to the same
MME, the TAU messages will differ only on the MME
Temporary Mobile Subscriber Identity (M-TMSI). Considering
an average of 30 TAU messages per second, and a message
aggregation period of 10 s, 300 TAU messages can be
aggregated in a single 1211 bytes message. Individual
messages would instead require 4500 bytes. This approach
can alleviate the traffic produced by massive channel access
towards the MME, but it does not address the issue of batch
MTD transmissions on the access network side.
(4) Q-learning solution: The standard RACH access is
basically derived from the classical slotted ALOHA protocol,
of which it inherits simplicity and limitations. In particular,
the system may drift to an unstable region in the presence
of massive M2M traffic. In this context, [21] suggests a
solution based on Q-learning to enhance the throughput of
RACH and shield H2H traffic from the performance loss that
A. Biral et al.8
can be caused by massive M2M access requests. According to
the authors, users should be divided into two groups: a
learning group containing all MTDs, and a non-learning
group composed of H2H devices. M2M communication uses
a virtual frame of RA slots called M2M-frame, whose length
(in time slots) should be equal to the number of MTDs in the
network. Every node keeps a Q-value for each slot in the
M2M-frame, which records the transmission history on that
slot in consecutive frames. Such value is updated after
every transmission attempt as follows:
Q’ð1αÞQþαr
where αis the learning rate, and ris the reward, which
equals 1 if the transmission is successful or kotherwise,
where kis known as penalty factor and is introduced to
mitigate the effect of collisions with H2H devices [21].
Each MTD will transmit in the slot with the highest Q-
value. The performance evaluation shows that, in case of
high load from H2H traffic, Q-learning access stabilizes the
total RACH throughput at 35% (approximately the maximum
efficiency of Slotted ALOHA), as the M2M traffic increases.
When the H2H traffic load is low, instead, the proposed
solution provides a significant improvement, raising the total
RACH normalized throughput to 55%. Moreover, delay is
reduced and the learning convergence is quickly achieved.
(5) Game theory scheme:In[22], the problem of H2H and
M2M coexistence is formulated by adopting a game theore-
tic approach. As mentioned, the standard LTE RACH proce-
dure is characterized by a unique pool of preambles, from
which each device randomly picks a preamble to be used for
its channel access request to the BS. In the proposed
solution, instead, different preamble pools for M2M and
H2H usage are reserved: in particular there are R
H
pre-
ambles for H2H, R
M
for M2M, and R
B
available for both.
Then, MTDs are allowed to extract the preamble either in
the M2M-dedicated pool (action ai¼M), in the shared one
(ai¼B), or to stay silent (ai¼S) with a probability distribu-
tion that is determined according to the outcome of a game.
The game formulation consists of a constant number H
B
of
H2H users that transmit preambles taken from the shared
pool, and NMTDs, which are the players of the game. They
play a mixed strategy σiðaiÞ, choosing actions M,Bor Swith
probability pi;M,pi;Bor 1pi;Mpi;B, respectively. All MTDs
have a cost λA½0;1for preamble transmission (e.g., in
terms of energy consumption), yielding the following gains:
gi¼
1λif transmission is successful;
λif transmission fails;
0 if transmission is not attempted:
8
>
<
>
:
Denoting by PSðaiÞthe RACH success probability if action a
i
is taken by player i, the expected payoff is given by
E½gi¼ X
aiAfM;Bg
σiðaiÞðPSðaiÞλÞ:ð1Þ
Simulations show that, following a mixed strategy Nash
Equilibrium, every MTD has non-negative utility, and the
throughput of both M2M and H2H users is improved with
respect to the baseline scheme in the case of overloaded
RACH. Moreover, the authors provide a procedure to esti-
mate the actual number of H2H and M2M devices in real
systems, which in practice may have imperfect knowledge
of the exact values of H
B
and N. The proposed approach is
proved to converge quickly and to provide small estimation
errors for N.
4. Tackling energy efficiency and QoS
challenges
As already pointed out in Section 1, PRACH overload is just
one of the issues that are raised by M2M massive access.
Other relevant challenges regard energy efficiency, hetero-
geneous quality of service (QoS) support, coverage exten-
sion, robustness to malfunctioning devices, security, and
scalability [23–26].
In this section, we will discuss some proposed approaches
that explicitly address the aspects related to energy effi-
ciency and QoS support for M2M services. We classify the
different techniques according to the methodological nat-
ure of the proposed solutions, rather than their specific
objectives, which can involve one or more performance
indices, as explicitly indicated in Tab. 1, and discussed in
Section 7.
Based on the adopted methodology, the different solutions
are hence divided into Clustering techniques, Game Theoretic
Approaches, and Machine Learning algorithm. The remainder
of this section describes in greater detail such solutions.
4.1. Clustering techniques
One possible way to handle massive access to the BS is to
appoint a few nodes, called coordinators or cluster-heads,
as relays for the remaining terminals. In this way, the
number of access requests to the BS is limited to the
number of coordinators. Furthermore, a proper selection
of the coordinators can also contribute to decrease the
energy consumption of the system by exploiting multi-hop
transmissions over high-gain links in place of direct trans-
missions over poor quality links. The problem now becomes
the design of suitable policies for electing the relays and
assigning the terminals to the different clusters. In the
following we describe some solutions that have been
recently proposed.
(1) Energy efficient clustering of MTDs: The massive
access management and energy efficiency aspects are
jointly addressed in [27], where authors proposed a cluster-
ing approach to limit the number of simultaneous accesses
to the BS and the energy consumption of MTDs.
Specifically, the authors consider a scenario with NMTDs,
which are randomly deployed in a single cell centered at the
BS and have each one packet to transmit. The BS is assumed
to know the channel conditions to each terminal. The idea is
to group the MTDs in Gclusters and, for each group, select a
coordinator that is the only device allowed to communicate
with the BS, relaying the communications of the other
terminals in its cluster (see Fig. 5).
The total energy consumption of the system can hence be
expressed as
EC ¼X
G
i¼1X
jAGi\fcig
PtLs
Rðdj;ciÞþPtLs
Rðci;BSÞ
ð2Þ
9The challenges of M2M massive access in wireless cellular networks
where G
i
is the set of nodes in the ith cluster,
3
c
i
and d
j
denote the concentrator of cluster iand the jth MTD,
respectively, Rðx;yÞis the transmit bit rate from node xto
node ywith transmit power P
t
, while L
s
is the length of the
packet to be transmitted.
Now, the objective is to minimize (2) while keeping the
number of groups Gbelow a certain threshold M, hence
limiting the maximum number of access requests to the BS
and reducing the redundant signaling of M2M services. To
this end, the authors of [27] study a number of algorithms
that combine different grouping and coordinator selection
techniques, which are briefly described below.
Grouping: The clusterization of MTDs is obtained by the
K-means algorithm: initially, Gnodes are randomly selected
as coordinators, and the other MTDs join the cluster based
on their channel conditions to the respective coordinator.
Then, a new coordinator is selected for each group,
according to certain coordinator selection policies which
will be presented in the next paragraph. Then, the remain-
ing MTDs are clustered again with respect to the new
Fig. 5 Proposed grouping model for M2M system.
Tab. 1 Comparison of the proposed approaches.
Solution Main challenge 3GPP H2H &
M2M
Performance indices
Delay Energy
efficiency
Access
probability
Resource separation [18] PRACH overload ✓✓
Slotted access [18] PRACH overload ✓✓
Pull-based scheme [18] PRACH overload ✓✓ ✓
Backoff tuning[18] PRACH overload ✓✓ ✓
Access Class Barring [18] PRACH overload ✓✓
a
✓
SOOC [16] PRACH overload ✓✓
RA for fixed-location [19] PRACH overload ✓✓
Bulk signalling [20] PRACH overload ✓✓
Q-learning [21] PRACH overload ✓✓ ✓
Game theory scheme [22] PRACH overload ✓✓✓
Energy-efficient clustering [27,28] Energy consumption ✓
QoS-based clustering [29–31] QoS support for M2M ✓
M2M-aware scheduling [32] QoS support for H2H &
M2M
✓✓ ✓
Matching theory scheme [33] QoS support for H2H &
M2M
✓✓
Reinforcement learning [34] BS selection ✓✓
Physical layer design [35–38] Coverage extension ✓✓✓
Cooperative coverage extension
[42]
Coverage extension ✓
Clean slate approaches [47,48,51] Massive access ✓
Note:Delay is intended as the time from the first transmission attempt until the successful conclusion of the access procedure.
a
The scheme can support H2H and M2M separation, though it is not specifically designed for this purpose.
3
The notation Gi\fcigindicates the set G
i
without the element c
i
.
A. Biral et al.10
coordinators. The whole procedure is iterated until the
coordinators do not change between consecutive cycles.
Coordinator selection: The core of the algorithm is the
policy adopted to elect the coordinator of each group. The
authors of [27] consider various schemes for coordinator
selection, based on different criteria to minimize the
energy consumption of MTDs. Some policies are indepen-
dent of the quality of the link between coordinator and BS,
while others keep this parameter into account when select-
ing the coordinator of each group. The policies that emerge
as more interesting from the simulation study are the K
Maximal Channel Gain (K-Max-CG), which simply selects as
cluster-heads the MTDs with highest channel gain towards
the BS, and the Optimum Energy Consumption (Opt-EC),
which implements an exhaustive search for the cluster-head
that minimizes the energy consumption within each of the
groups returned by the K-means algorithm.
Actually, the simulation results presented in the paper show
that clustering techniques are effective in reducing the massive
access issue and improving the energy efficiency of the MTDs.
Indeed, almost all proposed schemes perform better than direct
transmission between MTDs and BS in terms of both energy
consumption and channel contention. More specifically, in a
scenario with randomly distributed MTDs around the BS, the
energy consumption tends to decrease with the number Gof
groups, until it becomes almost constant for GZ10. In this
case, the best performance is attained by K-Max-CG. On the
other hand, when the MTDs distribution over the cell is not
uniform (e.g., nodes are concentrated in two or three smaller
areas around the BS), the energy consumption is minimized by a
lower number of groups, and the optimum coordinator selection
algorithm turns out to be Opt-EC, while the performance of K-
Max-CG dramatically deteriorates.
Another clustering technique to maximize the energy effi-
ciency of MTC has been proposed in [28] with reference to an
OFDMA-based cellular network. Once again, the authors pro-
pose to appoint some nodes as coordinators of a certain group
of MTDs and use two-hop communication to connect the
peripheral nodes to the BS. Differently from [27], here authors
consider some details of the transmission channel. More
specifically, they assume that communications between MTDs
and coordinators are managed by means of a Time Division
Multiple Access (TDMA) scheme, while the coordinators com-
municate with the BS by using an Orthogonal Frequency Division
Multiple Access (OFDMA) channel (which is the basis for the
Single Carrier Frequency Division Multiple Access adopted in LTE
uplink channels). Due to the short delay spread and the narrow
bandwidth, the first link is subject to flat fading, whereas the
second link can be affected by either flat or frequency selective
fading, i.e., the channel gain is either the same for all
subchannels or varies across different subcarriers.
Clustering and coordinator selection are based on an
energy-consumption model that accounts for both the energy
spent by each terminal to transmit data and some additional
energy expenditure due to the circuitry, i.e.,
EC ¼X
G
i¼1
ðPciþPcirÞDci
Rðci;BSÞþX
jAGi\fcig
PdjLs
Rðdj;ciÞ
!
ð3Þ
where the notation is as in (2),exceptforP
x
that denotes the
transmit power of node x,Pcir is the fixed circuitry power
consumption, and Dciis the aggregate data received by the
coordinator c
i
from all the MTDs in its cluster. Therefore, the
optimization problem can be formulated as follows:
min
G;Gi;ci;Pci
fECg
subject to GrN
Pdj¼Pt;iAf1;…;Gg;jAGi\fcig;
where Nis the total number of MTDs and P
t
is a fixed power
value. Although this formulation holds under the assumption
that all links are subject to flat fading, a similar problem can
be defined in the case of frequency selective fading. The
problem, however, is very complex; therefore, the authors of
[28] propose a suboptimal solution that consists in an
iterative algorithm that first clusters the MTDs into groups,
and then selects the coordinator for each group. The
algorithm actually starts from a random selection of the
coordinators. Then, each MTD joins the group whose coordi-
nator has minimum energy cost for delivering the packet
from the MTD to the eNodeB. Successively, a new coordinator
is selected among all nodes in the same cluster in order to
minimize the group average energy cost. The procedure is
iterated until the composition of the groups and the set of
coordinators remain unchanged. After that, the transmit
power of cluster-heads is also optimized by using another
iterative algorithm still with the aim of minimizing the
overall energy consumption.
Notably, neither [27] nor [28] account for the energy spent
in reception. Furthermore, as for all cluster-based scheme,
coordinators are subject to higher power consumption and
may fail because of energy depletion before the other nodes.
Hence, mechanisms for the rotation of the cluster-head role
shall be considered. On the other hand, these counter-
measures would require higher costs in terms of signaling
and control traffic, which shall also be accounted for.
(2) QoS-based clustering:In[29–31], clustering is used as
an effective solution to manage radio resource assignment
to a large population of MTDs with small data transmissions
and very disparate QoS requirements. Specifically, MTDs are
grouped into Gclusters based on their packet arrival rate
(ρ) and maximum tolerable jitter (δ), in such a way that
devices in the same group have very similar traffic char-
acteristics and QoS requirements. By this grouping opera-
tion, the BS can manage radio resources at the cluster level
rather than per single MTD, in the following way.
With reference to an LTE-Advanced network, eNodeB
allocates an Access Grant Time Interval (AGTI) to the ith
cluster every 1=ρims. Clusters are served in order of
priority, dictated by their traffic rate ρ, so that the cluster
with highest traffic rate is served first and if AGTIs for
different clusters are arranged in the same subframe, the
AGTI for the cluster with lower priority (i.e., lower ρ)is
postponed to the subsequent subframe. Each AGTI has a
fixed duration τfor all clusters. Based on such radio
resource management, the jitter of packets in the ith
cluster is upper bounded by
δn
i¼τþX
i1
k¼1
ρk
ρi
;i¼2;…;Mð4Þ
and δn
i¼τfor i=1 (see [30] for details). Then, if δn
irδifor
all clusters, packets in all clusters can meet the jitter
11The challenges of M2M massive access in wireless cellular networks
constraints. Hence, this bound can serve as a call admission
control scheme to guarantee that QoS constraints of all
admitted MTDs transmissions can be satisfied.Asaproofof
concepts, a simulation-based evaluation is developed in [29]
considering the system parameters of LTE-Advanced and
various QoS characteristics of the MTDs that cover a rather
general set of M2M applications. The results show that the
grouping scheme proposed, and based on the worst case
analysis given by (4), can achieve the desired QoS guarantees.
Moreover, in [30], the authors distinguish between deter-
ministic (hard) and statistical (soft) QoS guarantees. The
flexibility of the QoS requirements for cluster iis captured by
means of the “QoS outage”probability ϵ
i
,i.e.,theprob-
ability that the jitter exceeds the threshold δ.Clearly,hard
QoS constraints correspond to ϵi¼0. In case the number n
i
of
MTDs in cluster iis less than a certain threshold, groups of
resource blocks, called Allocation Units (AU), originally
dedicated to MTC can be opportunistically reallocated by
the BS, with probability q, to serve non-MTD users. The
results show that the (soft) QoS requirement of cluster iis
satisfied if, besides the condition δn
irδi,qis such that
qrmin ϵi
1Cni1
x1
Cni
x
0
B
B
B
@
1
C
C
C
A
1=x
;1
8
>
>
>
<
>
>
>
:
9
>
>
>
=
>
>
>
;
ð5Þ
where 0rxrniis the number of AUs still allocated to MTC
and Ca
b¼a
b
¼a!=abðÞ!b!
½
. These two sufficient conditions
ensure that the QoS constraints of MTDs are satisfied,
providing at the same time a flexible solution that enables
the BS to opportunistically schedule UE devices to achieve
efficient resource allocation.
The management of the QoS requirements when M2M and
H2H applications coexist is addressed in [32]. Here, the
authors apply clustering to divide the devices into two
classes: the high-priority class collects all classical H2H
users and some delay-sensitive M2M service users, while all
the other MTDs are grouped in the low-priority class. Then,
the authors define the M2M Aware Scheduling Algorithm
(M2MA-SA), which aims at preserving the performance
experienced by high-priority services in case of massive
presence of low-priority devices. The scheduler implements
a compound 2-phase procedure. In the first phase, resources
in the LTE uplink channel are allocated to the devices in the
high-priority class, according to a rather sophisticated
priority metric that accounts for multiple factors, including
the rate achievable by the user in each available resource
block, the remaining time before the delay threshold, the
gap between the current rate and the rate required to
satisfy the delay constraint, and the time spent waiting in
the queue. When all the H2H devices are served, the
system starts assigning the residual resources to the MTDs
in the low-priority queue. In this case, the algorithm
works with a timeline divided in periods of Tseconds.
4
Within every interval T, resources are allocated according
to two scheduling policies, namely max-utility and round
robin. The max-utility policy assigns resources to MTDs
with better channel quality. This approach, however, can
lead to starvation of users with long-lasting poor channel
conditions. To avoid this drawback, users that have not been
served for a certain period of time are put in a “timeout”
queue and scheduled in a round robin fashion before
performing the max-utility scheduling.
The simulation study presented in [32] confirms that
M2MA-SA can yield smaller delay, lower outage probability,
and higher throughput for H2H users than a simple schedul-
ing mechanism that does not discriminate between H2H and
M2M queues. Of course, this result is obtained by penalizing
the performance of M2M users.
4.2. Game theoretic approaches
An alternative to QoS-based clustering approaches is given by
the distributed matching algorithm introduced in [33].Once
again, the authors refer to a scenario in which both traditional
UEs and MTDs are randomly deployed in the coverage area of a
BSandsharethesameresources.However,inthiswork,the
radio resources are originally assigned to the Cellular Users
(CUs) only, while MTDs communication is subordinated to the
availability of idle portions of such resources. More specifi-
cally, each MTD is coupled with a CU and exploits parts of its
transmission resources. The coupling between MTDs and CUs
results from a distributed algorithm, based on the matching
theory, which negotiates the associations between CUs and
MTDs, according to a specific metric. In particular, MTDs trade
subchannel access in exchange for a reward that is propor-
tionaltothefractionoftheCUresourcesutilizedbytheMTDs.
At this stage, utility functions can be defined in order to
optimallymatchtheMTDstotheCUsinthenetwork.In
particular, the utility of the CU can be defined as the
(weighted) sum of its own data-rate and the reward earned
from the set of the MTDs it is coupled with. On the other hand,
the utility for the MTD is given by the data-rate obtained by
using the resources offered by the CU that is suitably scaled by
the so-called “urgency factor,”which is a decreasing function
of the maximum tolerable delay, and then subtracted from the
cost paid to the CU in return for such resources.
Now, the goal is to solve an optimization problem that finds
the optimal matching between CUs and MTDs, such that the
total sum of the utilities of all MTDs and CUs is maximized.
The proposed solution, which makes use of a matching
algorithm, can be described as follows: after initialization,
each CU that has not yet been matched offers a price for the
allocation of its own resources to the unmatched MTDs. Each
MTD then bids for the CU that provides the highest positive
utility (demand set). The CUs that get a bid from one MTD
only will be matched with the corresponding device, while
those that collect multiple bids increase their allocation price
by a certain amount for the next iteration. Results show that
the proposed algorithm achieves a stable matching state and
an average aggregate utility comparable with a classical
centralized algorithm, but with lower overhead for the BS.
4.3. Machine learning algorithm
Machine learning techniques are widely applied to protocol
design thanks to their ability to deal with very complex
systems in a relatively simple and efficient manner. The
management of M2M massive access is no exception, and
4
The length Tis chosen according to the delay requirements of
M2M services.
A. Biral et al.12
several studies in this domain have adopted machine
learning techniques to address different problems.
One such problem is the selection of the best serving BS
by a MTD. Indeed, one of the expected characteristics of 5G
systems is the proliferation of pico and femto cells, which
will lead to the densification of the radio coverage. As a
consequence, each MTD will likely be in the coverage range
of multiple BSs that, however, may offer quite disparate
QoS, depending on their distance, signal propagation con-
ditions, traffic load, and so on. Therefore, the selection of
the most suitable serving BS becomes an interesting
problem.
In [34], focusing on an LTE-Advanced network, the
authors propose a Q-learning algorithm to enable MTDs to
choose the best serving BS. The algorithm consists of the
following 5 steps:
(i) Every device initializes the Qvalue Qðs;aÞ, where sis
the state (i.e., the current reference BS), and ais the
action (i.e., the next BS to select), for all states and
actions.
(ii) With probability p51 the device performs an explora-
tion phase, i.e., randomly chooses a, while with
probability 1pit performs an exploitation phase, i.
e., chooses a¼arg maxa0fQðs;a0Þ}.
(iii) The MTD transmits its packet to the BS determined by
the selected action aand it measures the current QoS
performance Ds;a.
(iv) Q-value is updated as follows:
Qðs;aÞ’ð1αÞQðs;aÞþαðDs;aþψmax
a0Qðs0;a0ÞÞ
where αis the learning rate (typically α¼0:5), ψis the
discount rate (typically ψ¼0:1), and s0is the next state
if action ais taken when in state s.
(v) Repeat from step ii.
A scenario with two BSs is considered for the performance
evaluation. The simulation results show that the proposed
reinforcement learning method yields a balanced distribu-
tion of the MTDs between the two BSs. Indeed, when the
number of nodes that choose one BS increases, the packet
delivery delay also increases due to higher congestion at
that BS, so that the algorithm will progressively move some
devices to the other BS. Moreover, the higher the number of
allocated resource blocks by a certain BS, the more the
MTDs that select that BS. Finally, it is shown that random BS
selection is outperformed by the proposed reinforcement
learning algorithm in terms of average packet delivery
delay. On the downside, the algorithm is not capable of
promptly reacting to sudden changes of the M2M offered
traffic, thus possibly yielding a suboptimal behavior for
relatively long periods.
5. Cell coverage extension
The cell coverage of current cellular networks has been
designed considering devices that, in general, are much more
powerful (and expensive) than MTDs. The last generations of
smartphones and tablets, for instance, are capable of con-
siderable uplink power levels and are equipped with complex
RF components and extremely powerful processing units.
Moreover, users are generally satisfied when the battery
charge of these devices lasts enough to cover a working
day, so that daily recharging is acceptable and, actually,
common practice. When dealing with MTDs, instead, the
situation is dramatically different. MTDs are generally
expected to be very cheap devices, with limited computa-
tional and storage capability and extremely long battery life
(tens of years). Reducing the power in uplink, the transmit
bandwidth, and the complexity of the baseband components
usually decreases the cost and power consumption of the RF
chain, but at the price of a lower coverage range. Therefore,
the maximization of the cell coverage for M2M services is an
important objective that calls for innovative solutions both at
the physical and access layers. Some techniques that can be
considered to reach this objective are discussed in the
following.
5.1. Physical layer improvements
From the physical layer perspective, since the room for
increase in power and bandwidth is limited due to the strict
hardware and battery life constraints of MTDs, one remain-
ing option consists in extending the time duration of the
transmission. This can be achieved by time spreading, which
can also be done using quasi-orthogonal codes, so that
multiuser capabilities are exploited as well. Additionally,
retransmissions at the MAC layer can further enhance the
coverage for power- and bandwidth-limited transmission.
These approaches are investigated in [35–37]. One
approach is based on repetition codes and consists in
transmitting multiple copies of the same packet in different
time instants, in order to increase the decoding probability
of the BS and, in turn, the coverage range. Another
technique consists in increasing the power spectral density
of the resource blocks that are assigned to MTDs, reducing
the power of those allocated to other devices. In this way,
the coverage of the MTDs is clearly extended, to the
detriment of the service offered to the other terminals.
Note that, in [35], these techniques are actually proposed
to boost the coverage in downlink, i.e., from the BS to the
MTDs. Nonetheless, repetition coding can also be applied in
uplink, though at the cost of higher energy consumption of
the MTD node. The latter technique, instead, cannot be
used in uplink, because of the limited transmit power of
the MTDs.
A physical layer approach to increase both uplink and
downlink coverage is proposed in [38]. The proposed
mechanism consists in enabling MTDs with low uplink signal
strength to transmit over multiple transmission time inter-
vals, which are bundled together, in order to increase the
energy received by the BS. The transmission mode of each
MTD is stored by the BS, and then used to provide adequate
uplink resources to the MTDs during their wake-up time
windows.
In summary, physical layer techniques are valuable can-
didates to increase both downlink and uplink cell coverage
for MTDs. However, the solutions proposed in the literature
will generally consist in increasing the resources (power,
13The challenges of M2M massive access in wireless cellular networks
time, or frequency) allocated to peripheral MTDs, thus
exacerbating the massive access issues we discussed in the
previous sections. A possible way around these problems is
offered by some new technologies that offer long-range,
low-bitrate connectivity for MTDs, such as LoRa™[39],
Sigfox [40], and Weightless™[41], just to mention a few
that have been gaining momentum in the past few years.
These technologies, however, are generally based on pro-
prietary communication standards, which are not natively
compatible with the IP world, so that suitable gateways
need to be provided to enable global connectivity.
5.2. Cooperative approach
Another technique to increase the cellular coverage for
MTDs in downlink is presented in [42]. This solution
leverages on dedicated devices, named Cooperative Gap
Fillers (CGFs) [43], which act as relay nodes to MTDs and are
equipped with additional functionalities, such as multiple
communication interfaces (both long and short range) and
larger energy resources. The basic idea is that CGFs forward
BS messages through the short-range interface by using
network coding techniques [44]. Each message is broken
into several packets, which are linearly combined by the
intermediate CGF nodes in a random fashion. A global
encoding vector is inserted in packet headers to enable
message decoding when a sufficient number of linearly
independent combinations of packets received.
Although this technique can also be applied in uplink, its
effectiveness is limited because messages generated by
MTDs are typically short, so that the network coding over-
head will become dominant.
6. Clean slate approaches
While the previous studies generally consider as a starting
point the current cellular standards, though with different
degrees of abstraction of the system components, a few
works have investigated the massive access issues in a
standard-agnostic manner, with the aim of finding more
fundamental results and shading light on the intrinsic
performance bounds of these types of systems. Some
important results obtained from such a “clean slate”
approach are described below.
6.1. Schemes based on Slotted ALOHA
The performance of coordinated and uncoordinated trans-
mission strategies for multiple access is analyzed in [45],
where it is shown that, for payloads shorter than 1000 bits
(which are typical values in the M2M context), uncoordi-
nated access schemes support more devices than coordi-
nated access mechanisms, because of the lower signaling
overhead. A well-known protocol for uncoordinated access
is Slotted ALOHA. An enhanced version of this protocol,
called Fast Adaptive Slotted Aloha (FASA), is proposed in
[46]: taking into account the burstiness of M2M traffic, the
knowledge of the idle, successful, or collided state of the
previous slots is exploited in order to improve the perfor-
mance of the access control protocol. In particular, the
number of consecutive idle or collision slots is used to
estimate the number of active MTDs in the network (the so-
called “network status”), enabling a fast update of the
transmission probability of the MTDs and, hence, reducing
access delays. By means of drift analysis techniques, the
authors prove the stability of the FASA protocol when the
normalized arrival rate is lower than e1.
Another improved version of Slotted ALOHA exploiting
Successive Interference Cancellation (SIC), called Frameless
ALOHA, is presented in [47]. A simple example illustrating
the principle of such SIC-enabled Slotted ALOHA is shown in
Fig. 6. It depicts the situation in which N=3 users contend
to transmit within the same frame, composed by M=4 slots.
The nodes on the left represent users, the nodes on the
right stand for time slots, and the edges connect the users
with the slots in which their respective transmissions take
place. All transmissions made by a user in the frame are
replicas of the same packet; moreover, every transmission
includes pointers to all its replicas. With this in mind, SIC
can be effectively exploited as follows.
First, slots containing a single transmission (singleton
slots) are identified and the corresponding transmission
resolved. Referring to Fig. 6,s
4
is recognized as a singleton
slot and the associated packet of user u
2
is hence correctly
decoded (Fig. 6.1). In the next step, using the pointers
carried by the decoded packets, all their replicas are
removed from the associated slots, i.e., the interference
caused by such transmissions is canceled from the aggregate
received signal, thus potentially leading to new singleton
slots.
5
In the example of Fig. 6, the replica sent by u
2
in s
1
is
deleted and, as a result, s
1
becomes singleton slot
(Fig. 6.2). Such procedure iterates identically until either
there are no new singleton slots, or all transmissions have
been recovered (Fig. 6.3).
Generalizing this procedure and applying it to the M2M
communication context, we can consider a scenario in
which NMTDs contend to access the same BS. The protocol
assumes that users are slot-synchronized and aware of the
start of the contention period, which will be broadcast by
the BS. For each slot in the contention period, each active
device randomly decides whether or not to transmit a
replica of the pending packet, according to a predefined
probability:
pa¼β
Nð6Þ
where βis a suitably chosen parameter, subject to optimi-
zation. After each slot, the BS collects the received
compound signal and tries to decode the transmitted
packets using the above described SIC procedure. The key
feature of the frameless ALOHA proposed in [47] is that the
Fig. 6 Sample graph representation of SIC procedure.
5
For the sake of simplicity, signal cancellation is assumed to be
perfect, i.e., to completely remove the power of the cancelled
signal without leaving any residual interference.
A. Biral et al.14
end of the contention period is dynamically determined in
order to maximize the throughput. Users that have not
successfully delivered their packet by the end of a conten-
tion period will keep performing the algorithm in the
subsequent contention round. Note that, if the contention
period is terminated at the Mth slot and the number of
resolved MTCs is N
R
, then the instantaneous throughput can
be computed as TI¼NR=M. The results presented in [47]
show that the proposed algorithm can achieve extremely
high throughput and very low loss rate, thus proving the
effectiveness and efficiency of the described model in a
M2M scenario.
The performance of the SIC-based Frameless ALOHA
scheme can be further improved by considering the capture
phenomenon, as done in [48]. Indeed, the BS can actually be
able to decode colliding signals with a signal-to-interfer-
ence-plus-noise-ratio (SINR) γlarger than a certain thresh-
old, which is called capture ratio and denoted by b.
Mathematically, signal jis captured if
γj¼Pj
PhajPhþη0
4bð7Þ
where P
i
denotes the power of the ith signal at the receiver,
and η
0
represents the background noise power. Concerning
the capture effect, the performance of the iterative
decoding algorithm is further enhanced, since packet
decoding can also occur in non-singleton slots, where
capture phenomena take place. In [48] it is shown that, at
least when the impact of noise is low, the capture effect
may result in substantially higher throughput compared to
the scenario without capture.
All in all, frameless ALOHA can ideally guarantee high
performance in a M2M scenario in terms of throughput.
Nonetheless, energy efficiency and complexity aspects have
not been considered yet. In particular, the SIC mechanism
sets quite high requirements to the BS in terms of storage
and processing capabilities. The BS indeed has to store the
raw samples of the compound received signal in all the slots
of a contention period, and carry out SIC-based signal
decoding on many slots in real time. In addition, the
frameless ALOHA protocol has a strong impact on MTDs’
energy consumption because, for each frame, the devices
must transmit a possibly large set of replicas of the same
packet to the BS. This aspect is a major issue in M2M
communication since, as we already pointed out, many
MTDs are constrained by the need to operate for years
without any battery replacement/recharge.
6.2. Asymptotic analysis of massive access
capacity
In the recent literature [49,50], it was observed that using
SIC in combination with multi-packet reception capabilities
makes it possible to dramatically increase the system
throughput even when transmitters are not centrally coor-
dinated. Let us then consider a cellular system where the
base station (BS) is capable of decoding all signals that
satisfy (7), and can also perform perfect SIC, completely
removing the interference caused by the decoded signals to
the remaining signals. Suppose that Nnodes transmit
simultaneously, and that the received signal powers at the
BS, fPj;j¼1;…;Ng, are iid random variables with Cumula-
tive Distribution Function FðÞ. A simplified expression of the
maximum achievable throughput of such a system is derived
in [51], elaborating on the results provided in [50]. For the
sake of completeness, we report here the derivation.
Denoting by N0¼Nthe initial number of overlapping
transmissions, the average number of still undecoded
signals after hinterference cancellation cycles can be
recursively estimated as
Nh¼NFðIh1Þ;Ih¼ðNh1ÞbEPjPoIh1
½
;h¼1;2;…ð8Þ
where I
h
denotes the (approximate) residual mean aggre-
gate interference at the hth cycle, with I0¼1, while
EPjPoIh1
½
¼RIh1
0ð1FðxÞ=FðIh1ÞÞ dx is the mean power
of each still undecoded signal. Since N
h
cannot increase in
h, the throughput achieves it maximum when hgrows to
infinity, for which we get
S1ðNÞ¼Nlim
h-1Nh¼Nð1FðI1ðNÞÞÞ ð9Þ
when I1ðNÞis equal to the fixed-point of (8), i.e.,
I1ðNÞ¼ NFðI1ðNÞÞ1ðÞbEPjPoI1ðNÞ
½ ð10Þ
provided it exists, and I1ðNÞ¼0 otherwise. Finally, given b,
we can determine
SnðbÞ¼max
NS1ðNÞ
Now, supposing that MTDs work in the low SNR regime, the
per-user transmit rate will linearly scale with b. We hence
define the aggregate system capacity as
Cn¼bSn
which is an approximate measure of the maximum spectral
efficiency that can be achieved in the cell, assuming perfect
SIC capabilities and capture threshold b. This value is shown
in Fig. 7, for three of the scenarios defined in [50], namely
pure path loss (PL2), Rayleigh fading channels (RF), and
Shadowing channels with standard deviation of 3 dB (SH3)
and 6 dB (SH6). From these results we observe that a BS
capable of performing perfect SIC and MPR can theoretically
decode an arbitrary large number of simultaneous transmis-
sions by proportionally reducing the per-user data rate.
Doing this, the aggregate system capacity remains almost
constant. Furthermore, from this analysis it appears that
the capacity of the cell depends on the statistical distribu-
tion of the signal powers, and the higher the variance, the
more effective the SIC. Therefore, combining SIC, multi-
packet reception, and coded random access techniques, it is
possible to support massive access of sporadic transmitters
to a common BS, provided that the receiver is capable of
performing multi-slot SIC decoding. Once again, however,
the analysis has not yet considered aspects related to the
energy consumption of the MTDs.
7. Discussion
This paper provides a survey of the main challenges raised
by massive M2M access in wireless cellular systems, and of
the methodologies and approaches that have been proposed
to improve the coexistence of human-initiated and
machine-initiated communications and counteract the pro-
blems generated by signaling overload and channel access
15The challenges of M2M massive access in wireless cellular networks
contention. In this section, we sum up our study by
providing a comparative analysis of the solutions described
in the previous sections. Then, we discuss the potential of
the most promising approaches that are being proposed for
the next generation of wireless cellular systems to solve the
challenges of M2M massive access.
7.1. Comparative analysis of current solutions for
massive access
Tab. 1 offers a compound view of the solutions described in
the paper, with an indication of the characterizing features
and the main targeted performance indices. More specifi-
cally, we consider the following aspects:
Main challenge: primary issue addressed by the scheme;
3GPP: the scheme has been proposed in 3GPP technical
report and is designed with explicit reference to 3GPP
standards (LTE in particular);
H2H &M2M: the scenario assumes the coexistence of
H2H and M2M services;
Performance indices: the scheme is designed to improve
the following figures of merit
○minimization of access delay;
○minimization of energy consumption of MTDs;
○maximization of access probability/throughput of UEs
and/or MTDs.
A quick look at the table shows that much attention has
been devoted to the PRACH overload problem, which does
not only impact the performance of M2M services but, more
critically, can severely degrade the quality of conventional
H2H services, which have high Average Revenue Per User
(ARPU). This risk is much feared by operators because the
quality degradation of conventional services would impact
the customer satisfaction and loyalty, thus jeopardizing the
operators business. For this reason, great effort has been
devoted to the study and test of solutions for enabling M2M
services in the current and next cellular system architec-
tures or, at least, for mitigating the possible impact of M2M
traffic on conventional services. Besides this main trend of
research, there have been several studies regarding other
aspects of MTC and, in particular, energy efficiency, QoS,
and coverage extension. Although the literature contains a
fair number of works that apply different approaches and
methodologies to explore the challenges posed by M2M
massive access scenarios, the full realization of such
scenarios requires much more work. Most of the proposed
schemes, indeed, refer to specific use cases and are
designed to optimize only few aspects of the system, such
as the energy efficiency of MTC, the delay, or the coex-
istence with H2H traffic. However, M2M applications have
extremely different characteristics in terms of generated
traffic and service requirements, so that the full support of
the M2M paradigm calls for flexible solutions, capable of
discriminating among different types of MTDs and of provid-
ing differentiated services according to the specific require-
ments of the applications and the current conditions of the
system.
Finally, we observe that, while the fundamental limits for
broadband systems are well understood in the literature and
considered by the standardization initiatives, a similar level
of understanding for MTC-oriented systems is still lacking.
Finding such limits will serve as a basis to understand what
is possible and to design efficient and flexible MTC systems.
Therefore, there is a need for more studies that, adopting a
clean-slate and standard-agnostic approach, can provide
insights on the fundamental aspects of MTC, thus contribut-
ing to the definition of radio protocols and architectural
principles able to serve a large number of MTC connections.
7.2. Massive access in 5G
The support of M2M services has been identified as one of
the most challenging objectives of 5G [11]. While the
general requirements of 5G systems are progressively taking
shape [52,53], the technological issues raised by the M2M
paradigm are still partially unclear. General consensus has
been reached on the importance of few, key approaches and
technologies, including massive MIMO, small cells, milli-
meter wave (mmWave) communication, and virtualization
of system elements and network functions. Nonetheless,
whether these technologies will be able to provide efficient
support to M2M services in 5G systems and to coexist with
traditional broadband services is still an open question, as
argued in the following.
(a) Massive MIMO: This technique consists in equipping
the BS with much more antennas than the number of
devices, so that the channels to the different devices will
be quasi-orthogonal, thus making it possible to increase the
spectral efficiency by using simple spatial multiplexing/
demultiplexing procedures [54]. Therefore, massive MIMO
can dramatically enlarge the number of simultaneous
transmissions that can be successfully received by a (power-
ful) BS, without burdening the peripheral nodes. These
characteristics, in principle, make massive MIMO extremely
attractive for supporting massive M2M access. The limit of
such an approach is that enabling massive MIMO for a
massive number of MTDs may require an exceedingly large
number of antennas at the BS, which can be infeasible due
to logistic and technical problems. Furthermore, despite
the huge interest in massive MIMO, there is still much to be
learned, in particular regarding the propagation and cost
10−3 10−2 10−1 100
2.5
3
3.5
4
4.5
5
5.5
Capture threshold (b)
Aggregate cell capacity (C*)
PL2
RF
SH3
SH6
Fig. 7 Asymptotic optimal capacity of a cell with SIC and MPR
capabilities.
A. Biral et al.16
effectiveness, so that the actual performance and feasi-
bility of this technique are still open to investigation.
(b) Small cells: One possible solution for dealing with the
increase of the number of devices in hot spot areas are the
densification of the network by employing small cells [55],a
paradigm that is also known as Heterogeneous Networks
(HetNets). Small cells are indeed deployed to reduce the
distance between devices and access points, thus enabling
higher bit rates (or lower transmit power and interference),
while also improving the spatial reuse. In an M2M setting,
however, the focus is not on high transmit rate, but rather
on reliable and ubiquitous connectivity. MTDs are in fact
expected to be spread across wide areas, also where
human-generated broadband traffic may be light, e.g.,
along highways/road/railroads or in agricultural areas.
Hence, providing access to MTDs will require uniform and
ubiquitous coverage that is not economically sustainable by
using microcells. Even in case of a high concentration of
MTDs in relatively small areas, the Average Revenue Per
User of MTD-based services is likely lower compared to
conventional services, thus not justifying the deployment of
small cells for the sake of MTD-coverage only. Finally, the
densification of the network does not impact the signaling
overhead at the PHY layer, which is inefficient due to the
MTC transmission characteristics.
(c) mmWave communication: After years of striving to
squeeze more spectral efficiency from the crowded band-
width used by current microwave cellular systems, the huge
bandwidth available at mmWave frequencies, from 3 to
300 GHz, represents an irresistible attraction for 5G sys-
tems. Although the signal propagation at these frequencies
is not yet thoroughly understood, the measurements
reported in [56] indicate that transmission can occur even
in the absence of line of sight, though with a much higher
path loss exponent. In combination with large antenna
arrays, mmWave communication can make it possible to
reach huge bitrates over short distances. However, the
sensitivity to blockage, the rapid power decay with dis-
tance, and the higher power requirements of mmWave
communications make this technology less attractive for
MTDs that, instead, need long-range, low-power, and low
bitrate connections.
(d) Virtualization: Software Defined Networking (SDN)
and Network Function Virtualization (NFV) are two emerging
paradigms that basically consist in abstracting low-level
network functionalities to enable a much more flexible
management of the network resources and a better and
adaptive support of different types of services [57]. The
accomplishment of these concepts would make it possible to
differentiate the services offered to the different traffic
flows and to dynamically instantiate network elements
where and when needed. Ideally, these mechanisms shall
deliver the illusion of “infinite capacity,”giving to each
application exactly the resources it needs to achieve the
desired Quality of Experience (QoE). This vision is extremely
appealing for what regards the support of massive M2M
traffic, in that SDN can naturally provide separation
between M2M and H2H traffic, while guaranteeing the
desired QoS levels to each type of flow (both at the access
network and across the core network). Moreover, the fine-
grained and per-flow resource allocation paradigm enabled
by SDN will result in a better utilization of the network
resources, thus contributing to alleviate the massive access
problem. NFV, on the other hand, can be used to dynami-
cally shape the network architecture according to the
traffic requirements. For example, NFV can instruct net-
work elements in a certain area to act as concentrators to
collect MTDs data, or as relays to extend the coverage
range, or even as additional BSs to satisfy temporary peaks
of access requests. While this virtualization principle can
bring a disruptive change in the architectural design of next
generation communication systems, major research efforts
are still required to turn this vision into reality.
8. Conclusions
Today's cellular systems have not been designed to support
M2M services, yet they are able to supply the present-day
demand for M2M services, realizing the “place-&-play”
concept that was described in the introduction. However,
if the M2M market will fulfil the big expectations of the
stakeholders, the limits of these technologies will soon
become apparent. These considerations have driven acade-
mia, research institutes, industries, and standardization
bodies to devise improvements of current standards and to
design novel solutions to face the challenges posed by these
new types of services. Meanwhile, the growing demand from
the M2M market has fueled the proliferation of proprietary
solutions, which are not natively compatible with the IP
world and necessitate suitable gateways to interact with
the rest of the world.
For the moment, then, the picture of M2M support
appears quite composite and variegate, with no clearly
emerging solution. The evolution of the M2M services in the
coming years will likely rely upon a mix of proprietary
technologies, explicitly designed for MTD connectivity, and
legacy cellular standards, suitably enhanced to better scale
with massive access requests from MTDs. Considering that
many M2M systems are expected to remain operational for a
long time with minimal intervention and maintenance, such
hybrid architectures will probably endure for many years to
come, and will be eventually absorbed by 5G that will offer
native support to M2M services.
References
[1] A. Zanella, N. Bui, A. Castellani, L. Vangelista, M. Zorzi,
Internet of Things for Smart Cities, IEEE Internet Things J. 1
(February (1)) (2014) 22–32.
[2] L. Atzori, A. Iera, G. Morabito, The internet of things: A
survey, Comput. Netw. 54 (October (15)) (2010) 2787–2805.
[3] P. Bellavista, G. Cardone, A. Corradi, L. Foschini, Convergence
of MANET and WSN in IoT urban scenarios, IEEE Sens. J. 13
(October (10)) (2013) 3558–3567.
[4] J. Zheng, M.J. Lee, A comprehensive performance study of
IEEE 802.15.4, in: Sensor Network Operations, IEEE Press Book,
Wiley Interscience, 2006, pp. 218–237 (Chapter 4).
[5] G. Montenegro, N. Kushalnagar, J. Hui, D. Culler, Transmission
of IPv6 Packets over IEEE 802.15.4 Networks, Technical Report
IETF Request for Comments 4944, September 2007. [Online].
Available 〈http://tools.ietf.org/pdf/rfc4944.pdf〉.
17The challenges of M2M massive access in wireless cellular networks
[6] P. Kinney, et al., Zigbee technology: Wireless control that
simply works, in: Communications Design Conference, October
2003.
[7] H. Merz, J. Backer, V. Moser, T. Hansemann, L. Greefe, C.
Hübner, Building Automation: Communication Systems with
EIB/KNX, LON and BACnet, Springer-Verlag, Berlin, Heidel-
berg, June 2009.
[8] 3GPP, Feasibility Study for Proximity Services (ProSe), Techni-
cal Report TR 22.803 V12.2.0, June 2013.
[9] H. Schaffers, N. Komninos, M. Pallot, B. Trousse, M. Nilsson, A.
Oliveira, Smart cities and the future Internet: towards coop-
eration frameworks for open innovation, in: The Future
Internet, Springer, Berlin, Heidelberg, vol. 6656, 2011,
pp. 431–446.
[10] A. Laya, L. Alonso, J. Alonso-Zarate, Is the Random Access
Channel of LTE and LTE-A Suitable for M2M Communications? A
Survey of Alternatives, IEEE Commun. Surv. Tutor. 16 (First
Quarter(1)) (2014) 4–16.
[11] F. Boccardi, R. Heath, A. Lozano, T. Marzetta, P. Popovski, Five
disruptive technology directions for 5G, IEEE Commun. Mag. 52
(February (2)) (2014) 74–80.
[12] 3GPP, Bottleneck Capacity Comparison for MTC, Technical
Report TSG GERAN 46 GP-100895, May 2010.
[13] R. Paiva, R.D. Vieira, M. Saily, Random access capacity
evaluation with synchronized MTC users over wireless net-
works, in: IEEE Vehicular Technology Conference, May 2011,
pp. 1–5.
[14] G. Madueno, C. Stefanovic, P. Popovski, How many smart
meters can be deployed in a GSM cell? in: IEEE International
Conference on Communications (ICC) Workshops, June 2013,
pp. 1263–1268.
[15] P. Jain, P. Hedman, H. Zisimopoulos, Machine type commu-
nications in 3GPP systems, IEEE Commun. Mag. 50 (November
(11)) (2012) 28–35.
[16] A. Lo, Y. Law, M. Jacobsson, M. Kucharzak, Enhanced LTE-
advanced random-access mechanism for massive Machine-to-
Machine (M2M) communications, in: 27th World Wireless
Research Forum Meeting, October 2011.
[17] E. Dahlman, S. Parkvall, J. Sköld, 4G LTE/LTE-Advanced for
Mobile Broadband, Elsevier, October 2013.
[18] 3GPP, Study on RAN Improvements for Machine-type Commu-
nications, Technical Report TR 37.868 V11.0.0, September
2011.
[19] K.S. Ko, M.J. Kim, K.Y. Bae, D.K. Sung, J.H. Kim, J.Y. Ahn,
A Novel Random Access for Fixed-Location Machine-to-Machine
Communications in OFDMA Based Systems, IEEE Commun. Lett.
16 (September (9)) (2012) 1428–1431.
[20] T. Taleb, A. Kunz, Machine type communications in 3GPP
networks: potential, challenges, and solutions, IEEE Commun.
Mag. 50 (March (3)) (2012) 178–184.
[21] L.M. Bello, P. Mitchell, D. Grace, Application of Q-learning
for RACH access to support M2M traffic over a cellular
network,in: European Wireless Conference, May 2014,
pp. 1–6.
[22] Y.-C. Pang, S.-L. Chao, G.-Y. Lin, H.-Y. Wei, Network access for
M2M/H2H hybrid systems: a game theoretic approach, IEEE
Commun. Lett. 18 (June (5)) (2014) 845–848.
[23] W.H. Chin, Z. Fan, R. Haines, Emerging technologies and
research challenges for 5G wireless networks, IEEE Wirel.
Commun. 21 (April (2)) (2014) 106–112.
[24] H. Dhillon, H. Huang, H. Viswanathan, R. Valenzuela, On
resource allocation for machine-to-machine (M2M) communi-
cations in cellular networks, in: IEEE Globecom Workshops,
December 2012, pp. 1638–1643.
[25] S. Jayashree, B. Manoj, C. Murthy, A battery aware medium
access control (BAMAC) protocol for ad hoc wireless networks,
in: IEEE International Symposium on Personal, Indoor and
Mobile Radio Communications (PIMRC), vol. 2, September
2004, pp. 995–999.
[26] R. Kim, Snoop based group communication scheme in cellular
Machine-to-Machine communications, in: International Con-
ference on Information and Communication Technology Con-
vergence (ICTC), November 2010, pp. 380–381.
[27] C.-Y. Tu, C.-Y. Ho, C.-Y. Huang, Energy-efficient algorithms and
evaluations for massive access management in cellular based
machine to machine communications, in: IEEE Vehicular
Technology Conference (VTC Fall), 2011, September 2011,
pp. 1–5.
[28] C.Y. Ho, C.-Y. Huang, Energy-Saving Massive Access Control and
Resource Allocation Schemes for M2M Communications in
OFDMA Cellular Networks, IEEE Wirel. Commun. Lett. 1 (June
(3)) (2012) 209–212.
[29] S.-Y. Lien, K.-C. Chen, Y. Lin, Toward ubiquitous massive
accesses in 3GPP machine-to-machine communications, IEEE
Commun. Mag. 49 (April (4)) (2011) 66–74.
[30] S.-Y. Lien, K.-C. Chen, Massive Access Management for QoS
Guarantees in 3GPP Machine-to-Machine Communications,
IEEE Commun. Lett. 15 (March (3)) (2011) 311–313.
[31] P. Si, J. Yang, S. Chen, H. Xi, Adaptive Massive
Access Management for QoS Guarantees in M2M Commu-
nications, IEEE Trans. Veh. Technol., Available: 〈http://dx.
doi.org/10.1109/TVT.2014.2349732〉, in press.
[32] S. Zhenqi, Y. Haifeng, C. Xuefen, L. Hongxia, Research on
uplink scheduling algorithm of massive M2M and H2H services
in LTE, in: IET International Conference on Information
and Communications Technologies (IETICT), April 2013,
pp. 365–369.
[33] S. Bayat, Y. Li, Z. Han, M. Dohler, and B. Vucetic, Distributed
massive wireless access for cellular machine-to-machine com-
munication, in: IEEE International Conference on Communica-
tions (ICC), 2014, June 2014, pp. 2767–2772.
[34] M. Hasan, E. Hossain, D. Niyato, Random access for machine-to-
machine communication in LTE-advanced networks: issues and
approaches, IEEE Commun. Mag. 51 (June (6)) (2013) 86–93.
[35] Y. Qi, A. Ijaz, A. Quddus, M. Imran, P. Navaratnam, Y. Ma, R.
Tafazolli, M. Webb, Y. Morioka, On the physical layer design for
low cost machine type communication in 3GPP LTE, in: IEEE
Vehicular Technology Conference (VTC Fall), September 2014,
pp. 1–5.
[36] 3GPP, LTE Coverage Enhancements, Technical Report TR
36.824 V11.0.0, June 2012.
[37] General Dynamics Broadband UK, Coverage Extension for
MTC UEs, Technical Report TSG RAN171 R1-125204, November
2012.
[38] Y. Chan, J. Yang, P. Zong, Machine-to-machine Communica-
tions over Fixed Wireless Networks, April 2013, US Patent
8,416,741. [Online]. Available 〈http://www.google.com/
patents/US8416741〉.
[39] 〈http://www.semtech.com/wireless-rf/lora.html〉.
[40] 〈http://www.sigfox.com〉.
[41] 〈http://www.neul.com〉.
[42] G. Cocco, C. Ibars, N. Alagha, Cooperative coverage extension
in heterogeneous machine-to-machine networks, in: IEEE
Globecom Workshops, December 2012, pp. 1693–1699.
[43] EXALTED Project, First Report on LTE-M Algorithms and Proce-
dures, Technical Report, August 2011. [Online]. Available:
〈http://www.ict-exalted.eu/fileadmin/documents/EXAL
TED_WP3_D3.1_v2.0.pdf〉.
[44] R. Ahlswede, N. Cai, S.-Y. Li, R. Yeung, Network information
flow, IEEE Trans. Inf. Theory 46 (July (4)) (2000) 1204–1216.
[45] H. Dhillon, H. Huang, H. Viswanathan, R. Valenzuela, Funda-
mentals of Throughput Maximization With Random Arrivals for
M2MCommunications, IEEE Trans. Commun. 62 (November
(11)) (2014) 4094–4109.
A. Biral et al.18
[46] H. Wu, C. Zhu, R. La, X. Liu, Y. Zhang, FASA: Accelerated
S-ALOHA Using Access History for Event-Driven M2MCommuni-
cations, IEEE/ACM Trans. Netw. 21 (December (6)) (2013)
1904–1917.
[47] C. Stefanovic, P. Popovski, ALOHA Random Access that Oper-
ates as a Rateless Code, IEEE Trans. Commun. 61 (November
(11)) (2013) 4653–4662.
[48] C. Stefanovic, M. Momoda, P. Popovski, Exploiting capture
effect in frameless ALOHA for massive wireless random access,
in: IEEE Wireless Communications and Networking Conference
(WCNC), April 2014, pp. 1762–1767.
[49] X. Wang, J. Garcia-Luna-Aceves, Embracing interference in ad
hoc networks using joint routing and scheduling with multiple
packetreception, Ad Hoc Netw. 7 (March (2)) (2009) 460–471.
[50] A. Zanella, M. Zorzi, M2M massive wireless access: Challenges,
research issues, and ways forward, IEEE Trans. Commun. 60
(April (4)) (2012) 1058–1071.
[51] A. Zanella, M. Zorzi, A. dos Santos, P. Popovski, N. Pratas, C.
Stefanovic, A. Dekorsy, C. Bockelmann, B. Busropan, T. Norp,
M2M massive wireless access: challenges, research issues, and
ways forward, in: IEEE Globecom Workshops, December 2013,
pp. 151–156.
[52] NetWorld2020 ETP, 5G: Challenges, Research Priorities, and
Recommendations, Technical Report, August 2014. [Online].
Available 〈http://networld2020.eu/wp-content/uploads/2014/
02/NetWorld2020_Joint-Whitepaper-V8_public-consultation.
pdf〉.
[53] A. Osseiran, F. Boccardi, V. Braun, K. Kusume, P. Marsch,
M. Maternia, O. Queseth, M. Schellmann, H. Schotten,
H. Taoka, et al., Scenarios for 5G mobile and wireless
communications: the vision of the METIS project, IEEE Com-
mun. Mag. 52 (May (5)) (2014) 26–35.
[54] Y.-H. Nam, B.L. Ng, K. Sayana, Y. Li, J. Zhang, Y. Kim, J. Lee,
Full-dimension MIMO (FD-MIMO) for next generation cellular
technology, IEEE Commun. Mag. 51 (June (6)) (2013) 172–179.
[55] N. Bhushan, J. Li, D. Malladi, R. Gilmore, D. Brenner,
A. Damnjanovic, R. Sukhavasi, C. Patel, S. Geirhofer, Network
densification:the dominant theme for wireless evolution into
5G, IEEE Commun. Mag. 52 (February (2)) (2014) 82–89.
[56] T.S. Rappaport, S. Sun, R. Mayzus, H. Zhao, Y. Azar, K. Wang,
G.N. Wong, J.K. Schulz, M. Samimi, F. Gutierrez, Millimeter-
wave mobile communications for 5G cellular: It will work!,
IEEE Access 1 (May) (2013) 335–349.
[57] P. Demestichas, A. Georgakopoulos, D. Karvounas, K. Tsagkaris,
V. Stavroulaki, J. Lu, C. Xiong, J. Yao, 5G on the Horizon: Key
Challenges for the Radio-Access Network, IEEE Veh. Technol.
Mag. 8 (September (3)) (2013) 47–53.
19The challenges of M2M massive access in wireless cellular networks
