ARTICLE
Modelling IoT devices communication employing representative
operation modes to reveal traffic generation characteristics
Basel Barakat
a
, Simeon Keates
b
, Ian J. Wassell
c
and Kamran Arshad
d
a
School of Engineering, University of Greenwich, Chatham, UK;
b
School of Engineering and
the Built Environment, Edinburgh Napier University, Edinburgh, UK;
c
Computer
Laboratory, University of Cambridge, Cambridge, United Kingdom;
d
Electrical Engineering
Department, Ajman University, Ajman, UAE
ARTICLE HISTORY
Compiled August 28, 2019
ABSTRACT
Several traffic models for the Internet of Things (IoT) have been proposed in the
literature. However, they can be considered as heuristic models since they only re-
flect the stochastic characteristic of the generated traffic. In this paper, we propose a
model to represent the communication of IoT devices. The model was used to obtain
the traffic generated by the devices. Therefore, the proposed model is able to cap-
ture a wider understanding of device behaviour than existing, state-of-the-art traffic
models. The proposed model illustrates the behaviour of Machine-to-Machine up-
link communication in a network with multiple-access limited information capacity
shared channels. In this paper, we analysed the number of transmitted packets using
the traffic model extracted from our proposed communication model and compared
it with the state-of-the-art traffic models using simulations. The simulation results
show that the proposed model has significantly higher accuracy in estimating the
number of transmitted packets compared with the current models in the literature.
KEYWORDS
Internet of Things Communication; Communication System Traffic; Traffic Model;
Stochastic Process.
1. Introduction
The amount of data carried through wireless networks has increased by more than 100
fold in the past decade [1]. Several market research studies have predicted that the
amount of data will continue to grow exponentially [2]. Furthermore, the number of
connected devices is also expected to grow exponentially. The increase in the number
of connected devices is occurring due to the variety of new applications coming on to
the market, such as smart homes and wearable devices. Handling this extraordinary
increase in the amount of communication data and number of connected devices is
the driving force for researchers around the world investigating the next generation of
wireless communication, i.e., the fifth generation (5G).
For the previous two generations of wireless communications, the typical challenges
were energy efficiency [3], data throughput [4,20], coverage [5] and end-to-end latency.
For 5G, these issues are still considerably challenging; however, serving the expected
CONTACT Basel Barakat. Email: bb141@gre.ac.uk
number of connected devices might be overwhelming. The Internet of Things (IoT) is
one of the leading forces in increasing the number of connected devices. The IoT can
be defined as the network connecting billions of Machine-to-Machine communication
(M2M) devices. M2M, also known as Machine-Type-Communications (MTC), is de-
fined as the communication between machines or from machine to the network with
little or no human intervention [6]. IoT is expected to play a crucial role in several
sectors, including smart grids [7], environmental monitoring, surveillance, healthcare
[8], and intelligent transport systems [9]. Several market studies have predicted that
there will be more than 50 billion M2M devices in operation by 2020 [10]. Providing a
ubiquitous service for this extraordinary number of connected devices and the conse-
quent volume of data generated by those devices is the biggest challenges for network
operators [14,18].
To design a network that can serve a large number of IoT devices, it is critical
to have a comprehensive understanding of IoT communication and the traffic gener-
ated by its devices. It is known that the characteristics and the traffic patterns of
M2M differ significantly from the conventional Human-to-Human (H2H) communica-
tion (mobile phone calls and computer video calls)[11,13,33]. For instance, commonly
M2M applications generate short bursts of periodic data, and the cellular network is
not well adapted for such short messages [14–17].
In this paper, a model for IoT communication is proposed. The model is used to
represent the traffic generated by IoT devices extending the work done in [37]. To bet-
ter understand the communication model, let us consider a conference as an analogy.
If it is intended to model the noise that will be produced generated by the audience,
we can model it as a random process (an Analytical approach shown in section 2.1),
or alternatively, use a sensor to record the noise level at several conferences and then
generalise the measured noise level (an empirical approach that will be presented in
section 2.2). However, it would be much more comprehensive to perceive the confer-
ence program and use it to estimate the noise level. The conference program here is
analogous to the communication model.
Consequently, the traffic extracted from the IoT communication model considers
several related factors (as shown in Fig.1). The first factor is the channel information
capacity. The channel information capacity plays a significant role in the time required
to transmit data. Most traffic models available in the literature do not consider the
information capacity as they are mainly based on the Erlang model [19] (such as
[6]). The Erlang model was proposed for telephone networks (i.e., circuit switched
networks) and are arguably not valid for M2M traffic.
The second factor not accounted for in the existing M2M traffic models is the block-
ing incidence in which the user requires access to the shared channels, but the chan-
nels are fully occupied [21–23]. Additionally, the multiple-access technique is missing
in the existing M2M traffic models [21–23]. For a shared channel, there are two main
multiple-access techniques (i) Centralised Scheduled Access in which a centralised de-
vice determines what part of the channel is allocated to each user, and (ii) Distributed
Access in which each user locally decides the channel to access.
Modelling the communication can be insightful to better understand the behaviour
devices in networks. For instance, it can help the researchers to model the traffic
generated by the devices. Another example application can be the modelling of the
energy consumption of the devices. One application that the authors believe that the
contribution made in this paper can be very insightful; is the modeling for real-time
systems. In particular, the work done on the Age of Information, in which several
researchers assumed that the traffic is generated according to Poisson distribution
2
[24,25].
This paper is organised as follows, section 2 briefly present the state of the art traffic
models. Section 3, presents the proposed Machine Communication Model; section 4
shows the simulation results in which we present the number of transmitted packets
in a predefined time period. This paper is concluded in section 5.
M2M
Traffic
Finite State
Machine
Behavior
Number of
Channels
Channels
Information
Capacity
Blocking
Incident
Number of
Transmitting
devices
Multi-access
Mechanism
Figure 1. Factors affecting M2M communication traffic.
2. Traffic Models proposed in literature
In the literature, two main approaches have been taken to model the traffic generated
by the M2M devices (M2MDs). The first approach was to propose a stochastic model
to evaluate the traffic (analytical approach) and the second approach was to measure
the traffic generated by the M2MDs (an empirical approach) as shown in Fig. 2.
Fixed and event
scheduling traffic
model
M2M Traffic
Models
M2M traffic model
framework
Parameterised
Markovian model
Empirical
models
Analytical
models
Figure 2. Traffic models proposed in literature.
2.1. Analytical Approach
2.1.1. Fixed Scheduling and Event-driven M2MDS Traffic Model
The authors in [21,23,29] proposed splitting the M2MDs’ traffic modelling into two
distinct models according to the transmission periodicity. The first model considers the
traffic generated by the periodic updates referred to as Fixed Scheduling (FS) nodes,
e.g., sending a sensor measurement. The traffic generated by an FS node was assumed
to follow a deterministic process. The second modelling problem was focusing on the
3
non-periodic data traffic referred to as Events-Driven (ED) nodes, e.g., the report of
an emergency alarm. The traffic packets generated by the ED notes are modelled as a
Poisson Process with rate λ
D
(number of packets sent in an explicitly defined time).
Table 1 summarises the modelling classification:
Table 1. Machine-to-Machine communication devices classification proposed in [23].
M2MD node group Traffic transmission periodicity Transmission statistical distribution
Fixed Scheduling Periodic Deterministic
Events-Driven Non-periodic Poisson
Although the authors of [23] remarked on the inaccuracy of conventional traffic
models, they made some inaccurate simplifying assumptions in their modelling. The
first assumption made was to assume that the M2MDs can be either FS nodes or ED
nodes. This assumption makes the model only applicable to specific devices. These
devices can do only a particular job (such as periodically report the temperature, but
where it cannot report an event such as when the temperature is higher than a set
threshold), while most of the M2MDs at the moment in the market can be both types.
Assuming that the Fixed Scheduling nodes are synchronised is another one of the in-
accurate assumptions. Hence, the authors in [30] investigated the synchronisation of
machine-generated traffic such as router states update messages (a message reporting
the current link states). It was demonstrated (analytically and empirically) that be-
haviour transition from asynchronous to synchronous is practically abrupt even if it
was affected by an external influence (such as turning the devices On simultaneously).
The synchronisation in the case of M2MDs would be an even more significant chal-
lenge. Hence most of the M2MDs will be connected to the network through a wireless
connection; the propagation delays and multipath will play a vital role in preventing
synchronisation.
2.1.2. M2M Traffic Model Framework [21]
The authors in [21], made a remarkable contribution in demonstrating the differences
between human to human communication (H2H) and M2M traffic. They proposed an
M2M traffic model similar to the Engset Traffic model (also known as On-Off model
[27]). The only difference between the two models was that in the model proposed
they assumed a Semi-Markov chain while in the Engset model, it is a Markov chain.
The principal difference between a Markov chain and a Semi-Markov chain is the time
between successful states transitions. In particular, in the Semi-Markov process, the
states transitions times are random variables [31].
The M2M traffic model proposed in [21] is shown in Fig. 3. It assumes that the
transmission of data occurs in one the following instances: (1) Periodic Update data
referred to as PU; (2) Event-Driven data referred to as ED; or, (3) Payload Exchange
which refers to the data traffic following the PU and ED traffic. A Timer or an Event
drive the transition from the OFF state to ON state. On the other hand, the transition
between the ON state and OFF state occurs when data transmission finishes.
They also proposed a model for the Sensor-Based Alarm and Event Detection device
shown in Fig. 4. In this model, they used the sub-states of the ON state in Fig. 3 as
main states. However, they did not use the PE exchange sub-state as they assumed
that PU and ED are implicitly included in the PE state.
4
OFF
PU
ED
PE
Timer or Event
End of process
ON
Figure 3. M2M traffic model proposed in [21]. PU refers to Periodic Update, ED refers to Event Driven and
PE refers to Payload Exchange.
OFF
PU ED
Timer
End
of
P
r
o
cess
End of Proces
s
E
ve
nt
Event
Figure 4. Sensor based alarm and event detection model used in [21]. PU refers to Periodic Update, ED
refers to Event Driven.
The inter-departure times between the states and the size of the packets are assumed
to be identical and independent random variables. However, in practical cases, this
does not reflect the situation of M2MD traffic unless it is an exceptional case in which
the device transmits a very short burst of data traffic. Additionally, the researchers
did not take into consideration the channel characteristics and the number of devices.
2.1.3. Coupled Markov Modulated Poisson Process Model [22]
The authors in [22] proposed a traffic model for M2MDs relying on the Markov Modu-
lated Poisson Process (shown in Fig. 5). However, they used a Coupled Markov Mod-
ulated Poisson Process (CMMPP) to illustrate the M2MDs’ synchronisation effect.
The CMMPP refers to multiple Markov chains that influence each other’s transition
probabilities P
n
[t]. The transition probability is defined as the probability of changing
from one state into the next state in a defined unit of time. The CMMPP were ini-
tially proposed in the context of pattern recognition. They assumed that the arrival
is a Poisson process. The arrival rate in the proposed model depends on the current
state of the MMPP, e.g., λ
1
represents the rate of arrival of the first state.
𝑠
𝑛
= 1, &𝜆
1
𝑠
𝑛
= 3, &𝜆
3
𝑝
1,1
𝑝
1,2
𝑝
2,1
𝑝
1,3
𝑝
3,1
𝑠
𝑛
= 2, &𝜆
2
Figure 5. Markov Modulated Poisson Process model used in [22]. s
n
represent the number of M2MD trans-
mitting data and λ represent the state arrival rate.
The model proposed was compared with those models proposed by 3GPP. That
was developed to model the aggregated traffic of several M2MDs. The focus of the
5
![Figure 6. The parameters used in the traffic model in [14]. The model proposed used the data rate (i.e., the data throughput achievable in terms of the number of bits that can be communicated using a communication channel) from a lab measurement. The lab measurement relies on the Signal to Noise Ratio (SNR) and the number of Resources Blocks (RB) to measure the Data Rate. The simulations were used to obtain SNR statistical properties. In particular, the Cumulative Distribution Function (CDF).](/figures/figure-6-the-parameters-used-in-the-traffic-model-in-14-the-30yvystq.webp)
