Received April 26, 2019, accepted May 26, 2019, date of publication June 13, 2019, date of current

version July 17, 2019.

Digital Object Identifier 10.1109/ACCESS.2019.2922766

Vehicular Network Simulation Environment

via Discrete Event System Modeling
LE WANG , RENATO IIDA, AND ALEXANDER M. WYGLINSKI , (Senior Member, IEEE)
Wireless Innovation Laboratory, Department of Electrical and Computer Engineering, Worcester Polytechnic Institute, Worcester, MA 01609, USA
Corresponding author: Le Wang ([email protected])
This work was supported by the MathWorks Inc.

ABSTRACT A vehicular ad-hoc network (VANET) enables vehicles to communicate with each other
directly or via roadside infrastructure in order to improve road safety and efficiency. Within a VANET,
communications can potentially alter vehicular mobility and conversely, that the mobility could potentially
influence vehicular communications. Therefore, a VANET simulator environment is needed that can accu-
rately model interactions between vehicular mobility and network protocols. In this paper, we present a novel
vehicular network simulation environment designed using the MATLAB discrete event system (DES) in the
SimEvents toolbox. The proposed simulation environment is a bit-accurate, discrete event simulator that
integrates vehicular mobility operations with wireless network communication. This paper provides details
on the design of the proposed simulator. Its computational costs are evaluated in terms of events quantities and
execution time. The physical (PHY) layer of the proposed simulator shows a more realistic packet success
rate (PSR) using bit-level processing techniques when compared with the packet-based NS-3 simulator. The
performance of the priority-based media access control (MAC) layer proves the data with different priorities
that can coexist in the same channel.

INDEX TERMS Vehicular network simulation, vehicular mobility models, discrete-event system.

I. INTRODUCTION accidents within the U.S. with respect to frequency, cost,

The vehicular inter-communication concept is motivated and casualties [5]. To address these crash scenarios,
by the opportunity to improve road safety and efficiency. VSC-A proposed several safety applications such as
Several safety applications benefit by supporting direct Emergency Electronic Brake Lights (EEBL) [6], Blind
vehicle-to-vehicle (V2V) communications [1], including Spot Warning (BSW)/Lane Changing Warning (LCW) [7],
accident prevention applications and lane changing applica- and Intersection Movement Assist (IMA) [8]. These safety
tions. In vehicle-to-infrastructure (V2I) communications [2], applications are often evaluated using vehicular mobility
the roadside unit (RSU) can gather and analyze traffic status models [9].
information, as well as guide vehicles within an area to A vehicular network is considered to be a complicated
improve traffic efficiency [3]. At the 15th Intelligent Trans- operating environment since it must account for both vehic-
portation System (ITS) World Congress [4], it was mentioned ular mobility and communication network simultaneously.
that vehicular networks have the potential to save time and The position and speed of vehicles could potentially impact
save lives. Therefore, vehicular network applications can be the quality of wireless communications, and the informa-
classified into two types: safety applications, and efficiency tion shared over the vehicular communication network could
applications. influence vehicular path and mobility decisions. This inter-
In 2011, the U.S. National Highway Traffic Safety Admin- action requires having the traffic mobility simulators to work
istration (NHTSA) released the final report of Vehicle Safety closely with vehicular network simulators [10].
Communications - Applications (VSC-A), which summa- In general, these types of simulators have often been
rized eight crash scenarios based on the statistics of vehicle created and controlled separately from one another, and
thus their interactions have rarely been considered. Over
The associate editor coordinating the review of this manuscript and the past several years, vehicle networking researchers have
approving it for publication was Zhiwu Li. worked on creating an interface between these two simulation

2169-3536 2019 IEEE. Translations and content mining are permitted for academic research only.
87246 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 7, 2019
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

environments and several approaches have been

proposed [11]–[14]. Based on the level of interaction,
we classify the simulators as joint simulators and integrated
simulators.
For the joint simulator approach, an interface is created
to associate the existing traffic mobility simulators with the
network simulators. An example is the iTetris [11] project,
which associates the traffic simulator SUMO [15] with the
network simulator NS-3 [16], or OMNet++ [17] simula-
tion environment. Another example is using TraCI [18] to
connect SUMO with another simulator such as OMNet++
or MATLAB. The interface here performs the role of relaying
messages between the simulators. Traffic flows are extracted
from SUMO and sent to the network simulator through the
interface, and conversely the instructions from the network
simulators are sent to SUMO in order to alter the traffic
behavior. The advantage of this cross-layer joint approach is
that one is able to enjoy the benefits of both well-developed
simulators. However, one limitation of this approach is the
design complexity of the interface since it needs to let both
simulators operate simultaneously. Another limitation of this
approach is the configuration complexity, since the users
often need to tweak a large number of parameters for both
simulators in order to make the overall simulation experiment FIGURE 1. VANET Toolbox consists of a Simulink library containing
work correctly. vehicular network blocks. One can create simulation models by dragging
the needed blocks from the library to an empty Simulink model.
The alternative approach to combine the network and
traffic simulators is to use them into one single simulator
in order to achieve full interaction. This type of simulator together. It supports a hybrid of time-driven and event-
is called an integrated simulator, which has the capabil- driven simulation environment.
ity having both simulators directly work and interact with • A PHY layer that precisely models bit-level processing
each other. Several examples include MoVes [12], NCTUns techniques, which is essential to emulate precise and
simulator [14], and VISSIM [13]. The limitations for this realistic wireless channels.
approach mainly come from an over-simplified communi- • Performance evaluation of vehicular networks across the
cation network. For example, several simulators only have PHY layer and the MAC layer. The results proves the
a basic radio propagation model with Carrier-sense mul- effectiveness of the proposed simulation environment.
tiple access with collision avoidance (CSMA/CA) as the The rest of paper is organized as follows: Section II pro-
MAC layer [19]. vides an overview to the Discrete Event System (DES) theory
In this paper, we present an integrated vehicular network and MATLAB DES framework. Section III describes the
simulation environment, which we refer to as VANET Tool- design architecture of the proposed simulation environment
box, that functions in the MATLAB/Simulink environment including PHY, MAC and APP layers. Section IV shows
as shown in Figure 1. The proposed simulator consists of a several simulation scenarios with different mobility models.
Simulink library with custom-built blocks covering the main Testing results and performance evaluations of the proposed
stack of vehicular network protocols including the application simulator are shown in Section V. Section VI concludes the
(APP) layer, the medium access control (MAC) layer, and the paper.
physical (PHY) layer as shown in the figure. Several mobil-
ity operations including car following model, lane changing II. OVERVIEW TO DISCRETE EVENT SYSTEM
model, as well as intersection management that are embed- A. DISCRETE EVENT SYSTEM
ded in the APP layer. The design objective of the proposed A system is a set of interacting components which behave
simulator is to provide a vehicular simulation environment together to perform a function and this function cannot be
to enable research and development in this expanding field. performed by any of the individual parts [20]. At a specific
Table 1 provides a summary of the acronyms used in this time, a system’s behavior can be described in a measurable
paper. way, i.e., state. The state of a system at time t0 is defined as the
The contributions of this paper include the followings: output y(t) of a system for all t >= t0 is uniquely determined
• The proposed novel vehicular network simulation envi- by the system status at t0 and system input u(t), t >= t0 .
ronment is an integrated type simulator combining Given the initial condition xE(t0 ) = x0 and the input uE(t) for all
both vehicle traffic simulation and network simulation t >= t0 , the state xE(t) is presented by state equations shown

VOLUME 7, 2019 87247

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

TABLE 1. Acronyms used in this paper arranged in alphabetical order (A-Z).

FIGURE 2. Systems overview and Discrete Event Systems. Discrete Event System is classified as a discrete-state event-trigger
system. A hybrid system including both time-driven and event-driven DES is suitable for network PHY layer simulation.

in Eq. (1): The state space of a system is a set of all possible values
a state may take. Based on the type of states in a model,
Ė = f (Ex (t), uE(t), t),
x(t) (1) a system can be classified into either a continuous-state sys-
Then the output Ey(t) is determined by state equations, tem or a discrete-state system, as shown in Figure 2 [21].
input, and time shown by: In a continuous-state system, the time variable, t, enables
the system to transit from one state to another state con-
Ey(t) = gE(Ex (t), uE(t), t), (2) tinuously. A system with such property is referred to as

87248 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

time-driven system. A continuous-state system is considered At time tk = tk−1 + vk , the timer reaches 0, ek has to
to be time-driven. occur, a state transition is caused from xk−1 to xk . Then,
In a discrete-state system, the states are only allowed to the same process repeats with the (k + 1)th event, ek+1 . Thus,
change from one discrete state value to another. The state a DES can be specified by the clock sequence of events, that
transitions are either synchronized by a clock or occurred is, Ev = {v1 , v2 , . . . , vk }.
asynchronously at some special time point due to events. A state x, including the initial state x0 , has clock
An event can occur instantaneously and cause state values yi where i ∈ 0(x). The triggering event e0 is the next
transitions. event which will occur at that state x, i.e., the event chosen
If state transitions in a discrete-state system are synchro- with the smallest clock value defined by:
nized by clock ticks, it is a discrete-state, time-driven system.
e0 = arg min {yi }, (5)
Otherwise, if state transitions occur asynchronously at vari- i∈0(x)
ous random time instants due to events, this system is referred
When event e0 occurs at state x, a new state x 0 is generated
as a discrete-state, event-driven system or, shortly, a discrete-
from the state transition function f (x, e0 ). The inter-event
event system (DES) as shown in Figure 2 . The set of events
time, y∗ , represents the amount of time spent at state x and
serves the purpose of driving a DES since each event may
is calculated by:
cause a state transition.
Even though a DES is defined as a discrete-state event- y∗ = min {yi }, (6)
i∈0(x)
driven system, it may be modeled as event-driven and/or
time-driven. In fact, a hybrid DES when both time-driven The updated time is obtained by t 0 = t + y∗. Once the new
and event-driven are present is more general, as shown state x 0 is generated, the clock values for all feasible events
in Figure 2. For example, the operating system (OS) in a are updated. If an event i ∈ 0(x 0 ) and i 6= e0 remains feasible
computer is designed to not only respond to asynchronous in the new state x 0 , the new clock value is y0i = yi − y∗. For
events that have occurred at any time but also process func- all events which are not feasible in x but become feasible
tions synchronized by the computer clock. in x 0 , i.e., e0 ∈ 0(x 0 ) but e0 ∈
/ 0(x), a set of new lifetimes
A continuous state system can be modeled either by dif- are supplied by the DES.
ferential equations (continuous time) or difference equations From a computer implementation standpoint, a DES simu-
(discrete time), while a discrete-state event driven can be lator should have an event scheduling scheme, i.e., a variation
analyzed by automata. For instance, suppose we have an of timed state automaton DES model, such that whenever an
event set E = {a, b, g}. The state space of the automation event i is enabled at time tn , its next occurrence is scheduled at
is X = {x, y, z}. The transition function of the automation time tn +vi , where vi is a lifetime sample supplied by the DES.
is denoted as f : X × E → X . For example, f (x, g) = y Thus, a Scheduled Event List (SEL) replaces maintaining the
means that the automation is in state x, after event g hap- clock values yi , i ∈ 0(x) defined by:
pened, the automation transits the state to y, i.e., x → y.
L = {(ek , tk )}, k = 1, 2, . . . , mL , (7)
The trigger event g may be either an external input to the
system, or an event generated by the system itself. The initial where m is the number of events in the events set E,
state is denoted by x0 . A deterministic automation, denoted mL is the number of feasible events for the current state,
by G, is defined in Eq. (3) [21]: i.e., m = |E|, mL = |0(x)|, mL ≤ m. The SEL is always
ordered on a smallest-scheduled-time-first basis. Additional
G = (X , E, f , 0, x0 ), (3)
research efforts on DES simulation concepts can be found in
where 0 is the active event set of G at x. Each state has a references [21]–[26].
feasible events set 0(x), the events from 0(x) are the only
events which may occur at this state. C. MATLAB DES SIMULATION FRAMEWORK
In order to simulate a specific DES of interest, there are
B. DISCRETE EVENT SIMULATION two options: either build a DES simulator or use an existing
In order to perform DES simulation, the events sequences DES simulator. Based on the event scheduling scheme and/or
need to be associated with clocks. Suppose we have a process-oriented scheme, it is possible to create a DES model
DES with a single event E = {α}. The feasible event set using standard computer languages such as C++. However,
0(x) = {α} for all x ∈ X . The event sequence in this DES building a DES simulator is outside of the scope of this
is Ee = {e1 , e2 , . . . , ek } and e1 = e2 = . . . = ek = α. paper. As our goal is to simulate vehicular network behaviors,
Event Lifetime denoted by vk is defined as the length of the choosing an existing DES simulator is preferred.
time interval of the two successive events. For the single event In this paper, the vehicular network simulator is developed
DES, the kth lifetime of the event is defined as: using SimEvents, which is a MATLAB toolbox from The
MathWorks, Inc. SimEvents needs to work with Simulink in
vk = tk − tk−1 , k = 1, 2, . . . , k, vk ∈ R+ , (4)
a time-driven simulation environment. The MATLAB DES
At time tk−1 , the kth event, ek , is enabled with a life- system object inside the SimEvents library allows users
time vk . A timer attached to ek starts to count down from vk . to create an event-driven system using MATLAB object

VOLUME 7, 2019 87249

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

oriented programming (OOP) languages. Thus, a system

developed by SimEvents with MATLAB DES supports
both time-driven and event-driven components via a hybrid
approach. The DES model created using SimEvents is a
timed DES model with a event scheduler, which maintains
a scheduled event list (SEL) as we have discussed in the
previous section. A DES model might contain more than one
DES object. A DES object can customize the event
type and its life cycle by overloading the methods, i.e.,
member functions, inherited from the base class mat-
lab.DiscreteEventSystem.
A MATLAB DES has three elements: Entity, Event and
Action.

1) ENTITY
The elements flow in a DES are called entities. The entities
contain static information, which can be MATLAB built-in
data type or structured/bus data types. A MATLAB DES
may have one or more entity storages, with each storage
containing entities in certain order, such as a first-in, first-out
(FIFO) queue or a priority queue. A MATLAB DES can take
entities as input or output and entities can leave a MATLAB
DES and enter another MATLAB DES.

2) EVENT FIGURE 3. Example code on the implementation of the PHY links. The
PHY link DES receives a waveform entity, delays it for a predefined
Multiple types of events can be scheduled and executed period, and forward out to the receiver. The operations are achieved by
to an entity. These events model activities such as entity alternately activated events and actions.
creation, destroy, forward (send/receive), delay and search.
As of MATLAB/Simulink R2018a, MATLAB DES supports
five types of events: performed and, according to the type of the waveform entity,
• Generate: obj.eventGenerate( ) can generate an entity in i.e., data or Acknowledgment (ACK), two separate Timer
the target storage. events are created to delay the entity for a period. Once the
• Destroy: obj.eventDestroy( ) can destroy an entity in the delay is done, the Timer action is activated, in which the
target storage. waveform collision is simulated and the waveform entity is
• Timer: obj.eventTimer( ) delays an entity to a period of forwarded out of the DES via Forward event. After the wave-
time. form entity left the DES, the Exit action is triggered to reset
• Iterate: obj.eventIterate( ) iterates entities in the target the channel status. More details about the implementation of
storage with conditions. the PHY layer are provided in Section III.
• Forward:obj.eventForward( ) forwards entities to a stor- A MATLAB DES environment is suitable for simulating
age or an output port. network behaviors since the data flowing through differ-
ent network layers can be treated as entities. For example,
3) ACTION the data units of the PHY layer waveforms can be mod-
When an event is due for execution, actions are invoked. eled as waveform entities. The data generation, movement,
These actions are conducted by user-defined methods, which and destroy functions within a network can be implemented
may contain the algorithms. The flexibility characteris- by events. Users can further define more dynamic behav-
tics of actions make the developers to create varieties of iors, such as media access for the MAC layer and channel
DES modules. model for the PHY layer, using actions. In the next section,
Partial code of the PHY layer implementation is shown the implementation of each component in the proposed sim-
in Figure 3. In a communication environment, the transmitter ulation environment is presented with details.
sends out a waveform, which flows through the wireless
channel link and arrives at the receiver. The wireless chan- III. PROPOSED DES V2x SIMULATOR: VANET TOOLBOX
nel link can be modeled as a discrete-event system. In the Generally, a vehicular network simulation is a combination
figure, the Entry action is triggered when a waveform entity of a time-driven system and an event-driven system. Figure 4
enters the DES and stays inside the storage. In the Entry illustrates the design structure of two vehicular nodes com-
action method, a series of channel sensing related actions are municating over a wireless channel. The framework consists

87250 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

FIGURE 4. Design Structure of VANET Toolbox. The APP layer integrating network model and vehicle mobility model is a hybrid of event-driven and time
driven. The MAC layer focuses on EDCA and is purely event-driven. The wireless channel link in PHY layer is a time-driven DES module.

of three DES modules: APP Layer DES Module (APP DES), Furthermore, the MAC layer is responsible for generating
MAC Layer DES Module (MAC DES) and PHY Layer DES frames, waiting for ACKs, and initiated retransmissions when
Module (PHY Link DES), among which APP DES inte- timeouts occur. All of these MAC layer behaviors are event-
grates vehicle mobility models with APP message generation driven. Additionally, the transmitter (Tx) and receiver (Rx)
operations, MAC DES includes MAC layer activities and of the PHY layer are integrated with the MAC Layer DES
PHY transmitter (Tx) / receiver (Rx) on bit-level processing Module. The PHY Tx is responsible for converting the binary
and PHY Link DES only simulates wireless propagation message information into wireless waveform symbols, while
channels. the PHY Rx is used to reverse the process. Both operations
The mobility models are integrated in the APP DES, which from the PHY Tx and Rx are based on bit-level processing,
makes our proposed simulator an integrated type simulator. i.e., users can manipulated every single bit of then data when
This design facilitates the information exchange between the necessary. The integration design of the MAC DES has two
vehicle mobility activities and the network communication purposes. First, the PHY operations of the bit-level process-
operations. The movements of vehicles are controlled by vari- ing is time-continuous instead of event-driven, thus the PHY
eties of mobility models such as car-following model (CFM) Tx/Rx cannot be implemented using DES. In our proposed
and lane-changing model (LCM). The mobility models are simulator, a series of functions are created to perform the bit-
implemented by different safety-related or non-safety appli- level processing operations. The second reason is to constrain
cations. According to the vehicular traffic scenario, the appli- the total number of DES units in the simulation model in
cations may generate messages and share with other vehicle order to enable simulation efficiency. The creation of a DES
nodes. This situation is event-driven pattern. Additionally, involves overhead computational costs, including assigning
the applications may create beacon messages such as Basic input/output ports and allocating queue memories. This over-
safety messages (BSMs) at 10 Hz, which is time-driven head may potentially lower the simulation speed. Thus, in our
pattern. Thus the APP DES is a hybrid of event driven and design process one of the most basic requirements is to use
time driven. These generated messages are disseminated via as few DES units as possible.
wireless network communication and reciprocally the perfor- The PHY link DES module only simulates the wireless
mance of network communication could affect the vehicle channel links since both PHY Tx and Rx are integrated
operations. The section only focuses on the design of network with the MAC DES module. The PHY link is a relatively
communication simulation environment, the discuss of the straightforward DES, as it is only responsible for accepting
mobility models is presented in Section IV. the incoming waveforms from the PHY transmitters and for-
The MAC layer of a vehicular network is different com- warding them to the PHY receivers after an air propagation
pared to other WLAN devices since it grants priorities to delay. This process is an event-driven pattern. During the
various messages such that the messages with higher priority air propagation delay, channel models such as AWGN and
have shorter deference in channel contentions. This mecha- two-ray ground reflection model can be applied to the wave-
nism is referred to as Enhanced Distributed Channel Access forms and this process is implemented by functions instead
(EDCA) and it is defined in the IEEE 802.11 standard [27]. of DES.

VOLUME 7, 2019 87251

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

FIGURE 5. VANET PHY Link Modeling. The PHY Tx and Rx are for the data
encoding and decoding on bit-level processing. Wireless channel link is a
DES module with two-ray ground reflection model and AWGN as default.

A. PHY LAYER IMPLEMENTATION

In vehicular network simulations, a precise representation of
the PHY layer is necessary in order to obtain reliable results
for comparison with real hardware performance. Popular
vehicular network simulation tools including VEINS [28] and
iTETRIS [29], [30], which usually have a simplified PHY
layer. The network simulators they adopted, NS-3 and/or
OMNet++, employ abstracted PHY layer [31], where the
smallest indivisible data unit used is the packet, i.e., the
packet is either received entirely or not at all. Several details
of the wireless communication implementation, such as chan-
nel estimation, frequency offset estimation and correction,
waveform modulation and demodulation, are omitted due to FIGURE 6. Modeling the Tx and Rx of the PHY layer on bit level. The data
this abstraction. However, individual bits inside a waveform unit flows in the channel is wireless symbols exactly the same as the real
are necessary to perform accurate simulations of the PHY radio transmission. (a) Bit-level processing on transmitting a waveform.
The frame is converted into bits and based on the configuration on
layer and channel models. In this section, we will introduce Non-HT transmission, the bits are converted into symbols and sent to the
the proposed PHY layer with bit-level processing techniques. wireless channel. (b) Bit-level processing on receiving a waveform. The
waveform goes through packet detection, frequency offset detection and
The performance of the PHY layer in terms of packet success correction, channel estimation, decoding, CRC check and finally being
rate (PSR) is evaluated in Section V. converted to a frame to the MAC layer.

1) DESIGN THE PHY LAYER ON BIT LEVEL transmission specifications. In MATLAB, wlanNonHTCon-
Figure 5 illustrates a basic wireless PHY link model that con- fig creates a Non-HT object in order to configure the
verts the received frame from the MAC layer into a wireless transmission parameters. It is configured for a 10 MHz chan-
waveform and lets the waveform pass through the wireless nel bandwidth with a single transmit antenna according to
channel. The interaction between the PHY layer and the MAC the IEEE 802.11p standard [27]. A Non-HT Orthogonal
layer is handled by the Physical Layer Convergence Protocol Frequency-Division Multiplexing (OFDM) symbol consists
(PLCP). A PLCP Service Data Unit (PSDU) is generated by of 64 sub-carriers with a 10 MHz bandwidth and symbol
serializing the MAC layer frame into a binary bit stream. period of 6.4µs. A 1.6µs guard interval (GI) is inserted
The PSDU bits along with the PLCP preamble and header between each symbols in order to prevent inter-symbol inter-
according to the IEEE 802.11 [27] are grouped into sym- ference (ISI). A PLCP header, including information of data
bols and finally becomes a waveform. The wireless chan- rate and PSDU length, is prepended to the PSDU. From the
nel consists of a two-ray Ground Reflection Channel model perspective of a waveform, the PLCP header is the Legacy
and an Additive White Gaussian Noise (AWGN) model by Signal (L-SIG) field and the PSDU along with a tail and
default. This design is based on the research of line-of-sight padding becomes the data field. Ahead of the L-SIG field,
(LOS) conditions specified in [32]. Additional channel mod- a PLCP amble is attached, which includes a Legacy Short
els including Rural_LOS and Urban_NLOS can be selected Training Field (L-STF) and a Legacy Long Training Field
during the simulation. The received waveform is decoded and (L-LTF). L-STF is used for the packet detection, initial fre-
verified using the bit-level receiver design in [33]. Only the quency offset estimation, and coarse timing synchroniza-
PHY channel link is designed in DES, both PHY Tx and Rx tion. The L-LTF is used for the fine time synchronization,
are implemented using functions from WLAN Toolbox and channel estimation, and fine frequency offset estimation.
are integrated with the MAC DES module. Thus, a complete waveform consists of a L-STF, a L-LTF,
Figure 6(a) illustrates the process of generating a waveform a L-SIG, as well as a data field. These fields are gener-
at the transmitter (Tx) at the bit level. The IEEE 802.11p ated separately and concatenated to form a complete Non-
PHY layer is derived from the IEEE 802.11a Non-HT HT transmit waveform. The Non-HT configuration object

87252 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

DES module, it stays in storage. An action called wavefor-

mEntry() is activated due to this ‘entering’ activity. In the
body code of the waveformEntry(), a timer event is associated
with the waveform entity creating a delay for a predefined
period. This delay is to simulate the air propagation delay.
Once the delay is completed, the corresponding timer action,
waveformTimer(), is triggered in which the waveform entity
is processed by customized actions, such as simulating chan-
nel collision and passing through multipath channel model.
When the timer action has been completed, the waveform
entity is forwarded to the output port of the PHY DES
module. After the waveform entity has left the DES module,
FIGURE 7. Modeling the PHY link using MATLAB DES. The waveform the waveformExit() action is called to reset the channel status.
entity enters the DES module, stays for a period and then left the module.
Only one waveform type storage is created. waveformEntry(),
waveformTimer() and waveformExist() are actions.
B. MAC LAYER IMPLEMENTATION
Vehicular networking architecture supports vehicle-to-
vehicle (V2V) and vehicle-to-infrastructure (V2I) commu-
specifies the parameters for generating the data fields of nications. The implementation of a MAC layer should be
a waveform. The cfgnonHT.PSDULength property indi- able to cope with all communication modes. For example,
cates the length of bytes to be sent in the Non-HT data after receiving a frame from the other nodes, the MAC
field. A Non-HT waveform is then generated by function layer should find out whether it comes from another peer
wlanWaveformGenerator according to the configuration of vehicle or from infrastructure. Furthermore, the MAC layer
wlanNonHTConfig. should also check the type of the received frame, i.e., broad-
Figure 6(b) shows the process of payload extraction when cast, multicast, or unicast, and prepare for an ACK response
a waveform arrives at the receiver (Rx). The first field that to unicast type frames.
needs to be processed is the L-STF. In the vehicular network,
L-STF has a length of 16µs with 10 repetitions. Due to its 1) BRIEF INTRODUCTION TO V2x AND EDCA
correlation properties, the first seven repetitions are used for One fundamental difference of IEEE 802.11p when com-
time synchronization purposes by performing self-correlation pared to other types of IEEE 802.11 networks is the usage of
calculations. The rest of the sequence is used for packet detec- EDCA for the purpose of Quality-of-Service (QoS). Different
tion, coarse frequency offset (CFO) detection, and correction frames are granted with different priorities. Eight priorities
and setting the automatic gain control (AGC). The second are defined and can be placed in four possible Access Cate-
field that needs to be examined is the L-LTF, which is gories (ACs): AC0, AC1, AC2 AC3. Each frame is assigned
composed of a cyclic prefix (CP) equaling to the period of one of the AC descriptions by the application that created
two GIs, i.e., 3.2µs, followed by two identical long training the message depending on the importance and urgency of
symbols, i.e, 2 × 6.4µs. Channel estimation, fine frequency the content. Specifically, AC0 denotes regular access, AC1 is
offset estimation, and fine symbol offset estimation all rely on for non-prior background traffic, while AC2 and AC3 are for
the L-LTF. With all estimation and correction stages executed, prioritized messages, e.g., critical safety messages.
the L-LTF demodulator and channel estimator operations are IEEE 802.11 channels are all contention-based, where all
performed based on the demodulated L-LTF. Note that the nodes need to compete with each other for channel access.
demodulated L-LTF is also used for noise power estimation. During the contention process, the data is required to wait
Finally, the Non-HT data field is extracted and recovered into for a random period of time prior to transmitting, which
the PSDU. The integrity of received PSDU, rxPSDU, is veri- is referred to as defer access. The defer access process
fied by the Cyclic Redundancy Check (CRC). Consequently, includes an Arbitration InterFrame Spacing (AIFS), which
the rxPSDU is sent to the MAC layer if it passes the CRC. is a replacement for Distributed coordination function(DCF)
The above operations of both PHY Tx and Rx are based on InterFrame Space (DIFS), and a backoff period, which is
the bit-level processing features. This is exactly the same calculated based on a contention window (CW) value. After
process when an actual waveform is transmitted among radio sensing a busy medium, a node will wait for an AIFS period
hardwares. Thus the PHY layer in our proposed simulator is before sensing the channel again. If the channel is idle,
more realistic and accurate. the node will start to backoff, otherwise the node has to wait
for another AIFS period. During the backoff period, the node
2) MODELING THE PHY LINK IN MATLAB DES keeps monitoring the channel status. In the event that a busy
The discrete-event implementation of the PHY layer link channel is detected, the node will immediately pause the
using MATLAB DES is shown in Figure 7. In the PHY DES backoff and restart the AIFS channel sensing step. In short,
module, only one storage resource is created to contain the both AIFS and backoff define the waiting period for a node
waveform type entities. When a waveform entity enters the before accessing the channel.

VOLUME 7, 2019 87253

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

FIGURE 9. Modeling the MAC layer using MATLAB DES. The MAC DES
module involves threes types of entities: Payload entity, frame entity and
waveform entity. It is responsible for the data streams sending to the PHY
layer and receiving from the PHY layer.

AC queues has frames the are ready to send after the

deference period, thus a contention is created. Since this
situation happens inside the same node, this type of con-
tention is called an internal contention, which is unique for
nodes using EDCA. Whenever an internal contention occurs,
the frame with the highest priority will be the first one to be
sent out, while the other frames have to redo the defer access.
If a frame is of the unicast type and requires an ACK
from the receiver, i.e., Reliable Data Transmission (RDT),
FIGURE 8. Flow chart of EDCA based MAC layer design. MAC outbound a replica is created inside the buffer and it waits for the ACK.
indicates the process of receiving a payload from the APP layer and If the ACK is not received within a predefined time period,
sending a waveform to the PHY layer. MAC inbound is vice versa. (a) MAC
layer outbound: Data flow from the APP layer to the PHY layer. A payload the replica will be sent again until an ACK is received or the
is received from the APP layer, converted into a frame, experience maximum retransmission limit is reached. If the ACK is still
channel access backoff and finally converted into a waveform. (b) MAC
layer inbound: Data flow from the PHY layer to the APP layer. Drop the not received by then, this frame will be dropped.
corrupt frame entity, or extract the payload from intact frame and send to For the inbound flow, a waveform is received from the
the APP layer. Reply an ACK frame if necessary.
PHY layer. The MAC layer will first check if the frame is
intact. The CRC is performed by the PHY Rx, but the action
In EDCA, the ACs decide different (AIFS, backoff) pairs. of discarding a corrupted waveform is performed by the MAC
Therefore, the frames with different priorities own different layer since the MAC layer is implemented by the MATLAB
defer access periods. In general, the higher priority the shorter DES, which is the only option to destroy a data entity. If the
the defer period and vice versa. The design purpose of EDCA waveform is intact, the MAC layer needs to check if it is being
is to enable the frames the with higher priority to gain channel sent to the correct node by checking the fromDS/toDS and
access more frequently. srcAddress/dstAddress fields.
The type of waveform that could be sent is either a
2) FLOW CHART OF MAC LAYER DESIGN IN DES data or an ACK. If it is an ACK, the MAC layer needs to make
In order to depict more clearly the implementation of the sure if it is a valid ACK since replicated ACKs in response
MAC layer, we define the data flow from the PHY layer to to the same data may be received due to the congestion of
the APP layer as the inbound flow, as shown in Figure 8(b), the channel. If it is a data type waveform, the MAC layer
and the flow from the APP to the PHY layer as the outbound extracts the payload and sends it to the APP layer. If it is a
flow, as shown in Figure 8(a). RDT waveform, i.e., an ACK is required, the MAC layer will
For the outbound flow, a payload from the APP layer is generate the corresponding ACK and send it to the PHY layer.
converted into a frame by adding the necessary MAC layer
headings, then forwarded to the AC0-3 queues. Frames from 3) MODELING THE MAC LAYER USING MATLAB DES
the four AC queues will perform deferred access simulta- Figure 9 illustrates the design of a MAC DES module,
neously. It might be possible that more than one of the in which one payload type storage, six frame type storages,

87254 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

and one waveform type storage are defined to contain pay-

load, frame, and waveform entities, respectively. The payload
type entity enters into the MAC layer from the APP layer and
stays in the payload storage. In the corresponding payload-
Entry() action, a new event frameGenerate() is called, which
converts the payload entity to a frame entity by adding the
necessary header and trailer.
Depending on its priority, the frame entity is forwarded to
different AC queues. Here, the frame entity experiences the
channel access deference, backoff, internal contention, and
finally is forwarded to a Hybrid Coordination Function (HCF)
FIGURE 10. APP Layer Design using MATLAB DES. The messages from
storage. The waveformGenerate() action is triggered in the mobility model applications are converted into payloads and sent to the
HCF storage, the frame entity is converted into a waveform MAC layer. When receiving payloads from the MAC layer, the messages
are extracted and dispatched to different mobility models.
entity using the bit level processing technique we introduced
in the above section, i.e., PHY (Tx) activities, and finally sent
to the PHY layer DES module. such as retransmission timer events. This indicates the whole
In a reliable data transmission (RDT) environment, a uni- reliable data transmission (RDT) process has been completed.
cast message is required to have an ACK returned. Unlike
the broadcast scenario, the frame entity is directly converted C. APP LAYER IMPLEMENTATION
to a waveform entity, the unicast frame creates a replica of One significant challenge of implementing this proposed sim-
itself, which is converted into a waveform entity and sent to ulation model is that the behavior of the APP layer depends on
the wireless channel. The original unicast frame entity stays the application itself and there is no comprehensive standard
in the HCF storage with a timer attached to it. If the ACK is defining all the application requirements since it is almost
not received within the timer period, the unicast frame entity impossible to anticipate all possible future needs. Therefore,
creates another replica, converts it to a waveform entity, and within the scope of this paper, we only focus on the critical
retransmits. If the ACK is still not received after the maxi- safety applications, which will be discussed in Section IV
mum retransmission limit is reached, the unicast frame entity with respect to mobility models. In this section, we only
is destroyed in order to prevent further retransmission. If the introduce the basic message dissemination functions.
needed ACK arrives in time, an iteration event is called and
its corresponding action frameIterate() destroys the original 1) MESSAGE DISSEMINATION
unicast frame in the HCF storage. A Dedicated Short Range Communication (DSRC) device is
The purpose of above design is because in a DES, an event required to transmit at least 300 meters [2], [34], and it is
is associated with an entity. If this entity is no longer existing assumed that the surrounding vehicle positions are changing
in the DES object, all the associated events become invalid frequently in the highly dynamic environment. Consequently,
and will never be triggered. Two causes will result in this we can assume the safety messages are physically broadcast
‘not existing’ situation. First, an entity is destroyed by event using a single hop. Therefore, packet collisions and packet
eventDestroy(). Second, the entity has left the DES object, loss are major challenges for communication system per-
i.e., it is forwarded to the output port via event eventFor- formance. One solution is to decrease the channel load by
ward(’output’). The replica case mentioned above is the sec- grouping similar messages together, as shown in Figure 10.
ond situation. A timer is required for retransmissions, and this A mobility model may involve several safety applica-
timer is a event associated to the unicast frame entity. If this tions, such as lane changing, braking, and collision warning.
frame entity is converted to a wavefrom entity and forwarded These applications share different types of messages with
to the PHY Link DES module, the timer will become invalid other peers differentiated by application IDs (AppIDs). For
along with the retransmission activity. Thereby, a replica is example, a lane changing application creates messages that
necessary to be sent meanwhile the original frame entity stays include driving direction information, while braking applica-
in the HCF storage with its associated timer activated. tions generate messages containing brake status. While both
When receiving a waveform type entity from the PHY types of messages may contain the same information, such as
layer DES module, MAC DES module stores it to the wave- currently location and speed, if these two messages are sent
form type storage, where the intact waveform is converted to separately the overlapping information will cause a waste of
a rxPSDU, the payload is extracted, and the frame sent to the transmission resources. The APP layer DES maintains a mes-
APP layer module. The corrupted waveform entity will be sage list created by map containers. Whenever an application
destroyed. If the waveform is an ACK, an iteration event is is activated, it has to register its AppID as well as its message
triggered inside the HCF storage. The original frame entities requirements to the message list. The AppID serves as keys
inside the HCF storage will be iteratively check the sequence while the requirements are values.
number (SN) field until the target frame is found. Then, this Once an empty payload is generated, the APP layer assim-
frame entity is destroyed along with all its associated events ilates the data requirements from these applications and

VOLUME 7, 2019 87255

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

stored in Storage 3. This driving entity stays inside Storage 3

forever and will never be sent out. A timer event is recursively
triggered on the driving entity at an interval of 0.1 seconds.
This is the refresh rate of the vehicles moving across the
map. The corresponding timer action payloadTimer(driving)
is called every 0.1 seconds to update the driving information
including vehicle position, speed, and direction. This refresh
rate can be increased at the cost of execution speed.

IV. MOBILITY MODELS AND SCENARIOS

An integrated vehicular network simulator usually consists of
FIGURE 11. Modeling the APP layer using MATLAB DES. The APP DES two sections: the vehicular mobility model and the vehicular
module generates payload entities containing messages from the
mobility models and sends to the MAC DES module. The received
network model. An accurate vehicular mobility model is
messages may update the traffic information in mobility models. The necessary for representing the real vehicular traffic behaviors
mobility models are integrated and make the vehicles move by the rules. since vehicular mobility significantly impacts the vehicular
network performance. A vehicular network application is
designed to make use of shared traffic information across
compiles them into one single message using a dictionary the vehicles in order to change traffic patterns, either for the
of standardized message construction guidelines. In the last purpose of road safety or for road efficiency improvements.
example, an assembled message consisting of position, speed, Therefore, a vehicular network simulator should describe
driving direction, and brake status is created instead of the interactions between the network protocols and vehicular
two separate messages. Society of Automotive Engineering mobility. In the proposed simulation environment, the vehic-
(SAE) standards J2735 [35] and SAE J2945 [36] define a ular mobility models are integrated with the APP layer, where
dictionary with over 150 data elements. Each data element a variety of vehicular mobility models have been proposed for
can be indexed using the AppIDs. different purposes including random models, flow models,
When a payload is received, the APP layer is responsible traffic models, behavioral models, and trace-based models.
for separating the data elements according to its appID and Traffic safety applications usually requires traffic flow mod-
dispatches them to the corresponding applications so as to eling, in which the detailed interactions between vehicles are
finally affect the mobility models. modeled as flows, as shown in Figure 12.
Vehicular network applications can be classified into
2) MODELING THE APP LAYER USING MATLAB DES V2V applications and V2I applications depending on which
Figure 11 shows the design of APP DES module. Since the V2x mode is used. According to [10], V2V applications are
APP DES is the top layer, it is responsible for generating the generally safety applications while V2I are usually dedicated
needed entities in order to trigger the whole simulation and to traffic efficiency improvements. In this paper, we only
no external input ports are needed. All entities are generated focus on V2V communications and only two V2V-based
inside the APP DES module by setupEvents() events. mobility models: (i) car following models, and (ii) lane
The vehicular mobility models are integrated with the changing models.
APP layer. This means that the APP DES module has two
responsibilities. First, the APP DES module should generate A. CASE STUDY: V2V COMMUNICATION
payload entities containing traffic information and sent to the Vehicle i is the target vehicle that will perform either car
MAC layer. As shown in the figure, two payload storages following or lane changing operations. At time t, the x posi-
resource (storage 1 and storage 3) are involved, while the tion and velocity of vehicle i are represented as xi (t) and
setupEvents() generates Basic Safety Messages (BSMs). The vi (t), respectfully. Vehicle i − 1 and i + 1 are the vehicles
0.1 in payloadGenerate(‘BSM’,0.1) event indicates the pay- immediately behind and in front of vehicle i with x positions
load generation intervals since BSMs are generated at a rate xi−1 (t) and xi+1 (t), and with speed vi−1 (t) and vi+1 (t). The
of 10 Hz. Once generated, the payload entities are forwarded variable 1di (t) indicates the distance from vehicle i to vehicle
to Storage 3 and finally sent to the MAC DES Module. i + 1 at time t. For the adjacent lane, the vehicles that are
When receiving payloads from the MAC DES module, immediately in back and in front are denoted as vehicle j − 1
the APP DES module extracts the message from the payload and j + 1. Similarly, their positions and speeds at time t are
entities and forwards to the mobility models. The traffic infor- denoted as xj−1 (t) and xj+1 (t), as well as vj−1 (t) and vj+1 (t).
mation will be updated according to the received message. These notations are summarized in Table 2.
A Emergency (EMG) type message can be generated upon
request by the vehicles. 1) CAR FOLLOWING MODELS (CFMs)
The APP DES module is also responsible for making the The CFMs control the individual vehicle’s driving dynamics
vehicles to move on the road according to mobility models, in order to maintain a safe distance to the vehicle that is imme-
thus a new type of entity with a driving tag is created and diately ahead. The objective of CFMs is to model vehicular

87256 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

FIGURE 12. Notations for Highway Mobility Model based on V2V Communications. The green car is the target. The green car is surrounded by the yellow
cars, who may involve in the coordinated car-following and lane-changing operations. The grey cars are the predicted positions according to the
prediction algorithm.

TABLE 2. Notations used in the Highway mobility model. The deceleration, dec, is determined by the current vehicle
speed, road surface friction coefficient µ, as well as the fric-
tion type, i.e., static friction and kinetic friction. If a vehicle
is driving on a dry concrete road surface, then according
to [37] the static friction coefficient is µs = 1 and the kinetic
friction coefficient is µk = 0.8. When the vehicle brakes free
and slides, the kinetic friction deceleration deck only depends
on µk and the acceleration due to gravity g = 9.80 m/s2 ,
as indicated in Eq. (11).
For static friction, NHTSA was able to show a mapping
between speed and braking distance, based on which we
set the maximum static friction deceleration to decs =
−6.50 m/s2 [38]. For dec ∈ [decs , 0], we define this type of
braking as regular brake. For dec < decs , the brake action
is called as emergency (EMG) brake. During EMG brake
traffic flows without car collisions using the help of vehicular process, the wheels are drifting on road surface, the friction
communications. between them is kinetic friction, thus we set dec = deck .
In CFMs, the vehicle keeps monitoring the distance to
di,safe = Li + treact ∗ vxi (t) + dbrake , (8) the vehicle immediately ahead of it based on the received
vxi (t + 1t)2 − vxi (t)2 BSMs and adjusts its speed adaptively in order to maintain
dbrake = , (9) a safe distance. When the front vehicle is braking, the vehicle
2 · µs · dec
behind it will be aware of it using the BSM information and
−vxi (t)2
dbrake = , (10) it will start to brake.
2 · dec µs =1,vxi (t+1t)=0
deck = −µk ∗ g = −0.8 ∗ 9.8 = −7.84m/s2 , (11) 2) LANE CHANGING MODELS (LCMs)
LCMs are based on multi-lane traffic scenarios. A general
The safe distance of vehicle i is generally calculated by LCM should include three parts: the trigger for a lane chang-
Eq. (8), where Li is the length of vehicle i, treact is the reaction ing event, the feasibility of a lane changing event, and the
time either of the driver or from the autonomous vehicular scheme used during the lane changing process.
dynamics, dbrake is the braking distance and is calculated As an example, suppose we consider the situations and
in Eq. (9), vxi (t) is the instantaneous speed when brake is notations shown in Figure 12, which describes a basic lane
performed, and vxi (t +1t) is the velocity when braking action changing scenario involving four vehicles. The trigger for
is finished (for a complete stop, vxi (t + 1t) = 0). Therefore, vehicle i to change its lane is the distance to vehicle i+1 being
we have the full stop brake distance defined in Eq. (10). shorter than the safe distance, i.e., when 1di (t) < di,safe .

VOLUME 7, 2019 87257

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

The lane changing feasibility of vehicle i is determined by for BPSK signal with different channel environment. Further-
the distance to vehicle j − 1, i.e., 1dj−1 (t + 1t), and the more, based on the case study of V2V communication in the
distance to vehicle j + 1, i.e., 1dj+1 (t + 1t) during the lane above section, we perform two sets of simulations focusing
changing process. Both distances should not violate the safe on the MAC layer behaviors and compare the performance
distance rule in order to avoid potential car collisions in the of EDCA with distributed coordination function (DCF).
adjacent lane when changing lanes. The speed boundary for
vehicle i is calculated as follows: A. PERFORMANCE OF VANET TOOLBOX
02 − vx,i(t)2 02
− vx,j+1(t)2 In this section, the performance of VANET Toolbox is ana-
| |<| | + di,j+1 (t), (12) lyzed and evaluated in terms of events numbers and the
2 · decs 2 · decs
execution time with different number of vehicles.
02 − vx,i (t)2
| | + di,j−1 (t)
2 · decs 1) COMPUTATIONAL COSTS IN TERMS OF EVENTS
02 − vx,j−1 (t)2 A detailed simulation of the entire vehicular network stack is
>| |, (13)
2 · decs time-consuming, especially with a large number of vehicles
q
max(vx,i (t)) = vx,j+1 (t)2 + 2 · |decs | · di,j+1 (t), (14) to simulate and taking into consideration large-scale vehic-
q ular communication effects. In this section, we show the
min(vx,i (t)) = vx,j−1 (t)2 − 2 · |decs | · di,j−1 (t), (15) computational costs in term of the number of events E in a
discrete event-based simulation model.
The lane changing prediction algorithm assumes the vehi- Suppose we have n vehicles in a vehicular communication
cle shifts to the adjacent lane at a y-speed of vy,i (t) while scenario, with each vehicle transmitting data at rate r in Hertz,
adjusting the x-speed xi (t) during the process. Eq. (12) cal- and the simulation time is t in seconds. When a vehicle is
culates the upper speed boundary of vehicle i. During the willing to send a message, events are scheduled in order to
lane changing process, vehicle j + 1 can perform a braking generate the message in the APP layer and forwarded to
operation and the shortest braking distance is determined the MAC layer, where the message is converted to a frame,
by decs , with di,j+1 being the distance on the x axis before experiences channel sensing, backoff, and finally sent to the
lane changing. The lane changing algorithm should predict PHY layer, all of which are conducted by different events.
whether di,j+1 is a safe distance for a lane change, i.e., In the PHY layer, the frame is transformed into a waveform
di,j+1 (t) should be greater than the difference of the brake and sent into the wireless channel by events. After receiving
trails from both vehicles. Similarly, the lower speed boundary a waveform, events are called in order to extract the informa-
is calculated by Eq. (13). The maximum and minimum speeds tion from the waveforms and send it way up to the APP layer
for vehicle i during a lane changing operation are defined by and process it in the mobility models. We assume the number
Eq. (14) and (15), respectively. of events E per transmission is e. The number of events E per
If the lane changing feasibility is not fulfilled, then vehicle simulation can be calculated by :
i will need to reduce its speed vx,i (t), i.e., brake, in order to
meet the safety constraint. If the feasibility check does allow E(n, r, t) = (nr) · e · (n − 1) · t, (16)
for a lane change, vehicle i will start to change lanes. The
trajectory model includes a lane changing period, target lane It is observed that the number of events E is linearly
chosen, etc. The validation and performance evaluation of the proportioned with the simulation time t and the data rate r, but
above mobility models have been presented in [39]. nonlinear with respect to the number of vehicles n. Figure 13
More sophisticated CFMs and LCMs have been proposed, shows the number of simulated events for each layer of the
including the Intelligent Driver Model (IDM) [40], Wiede- vehicular network and for the overall communication set.
mann Model [41], the Nagel-Schreckenberg model [42], and We choose the car following model (CFM) as the scenario
the Krauss (1998) Model [43]. These models can be imple- since for this CFM each vehicle broadcasts only BSMs at a
mented and used in VANET Toolbox if necessary. rate of 10 Hz and no other transmissions are involved. For a
600-second simulation when 4 vehicles are involved, the total
V. PERFORMANCE ANALYSIS number of events is around 2 × 106 . When the number of
In this section, the performance of the proposed simulation vehicles increases to 36, the total number of events is around
environment is evaluated. We first discuss the computational 1.5 × 108 , which represents an increase by a factor of 75.
costs of a full-stack vehicular network simulator in terms of In addition to the computational cost in terms of num-
the number of events scheduled during the simulation. Then, ber of events, we also profile the execution time for
we compare the packet success rate (PSR) of BPSK in a a 30-second simulation with 30 vehicles. The overall execu-
AWGN channel between MATLAB PHY layer implementa- tion time costs 8535 seconds, among which the PHY Tx func-
tion and NS-3 error rate model. Due to the bit level processing tion phy_psdu2waveform consumes 511 seconds (5.98%)
of the PHY layer, the channel tracking (CT) techniques can and the PHY Rx function phy_waveform2psdu consumes
be enabled on the L-LTF field in order to cope with the 4337 seconds (50.7%). These two functions include all bit-
high Doppler spread. The performance of CT is evaluated level processing operations of the PHY layer. As we have

87258 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

FIGURE 14. Execution speed improvement due to MATLAB code

generation. The simulation time is 30 seconds, the number of vehicles
increases from 5 to 30.

FIGURE 13. Computational cost in terms of events for each layers. The
mono-color surfs indicate the number of events happened for the PHY 2) SimEvents CODE GENERATION
layer, the APP layer and the MAC layer (from bottom to top). The colorful MATLAB is a high-level interpreted type language and pro-
surf is the combination of events happened in all three layers.
vides an interactive programming environment. Compare
with lower-level languages such as C/C++, the execution
introduced in Section III-A, the PHY Tx function is straight- time is longer. Furthermore, the PHY layer includes bit-level
forward, i.e., serializing the PSDU to binary bits. While processing, which uses more numerical computations so that
most of the bit-level processing operations such as packet the execution speed is further slowed down.
detection, channel estimation and CRC verification are in In order to enhance the execution speed, MATLAB/
the PHY Rx function. Thus, the PHY Rx function costs Simulink supports converting MATLAB code into C/C++
more execution time than the PHY Tx function. Additionally, code using code generation techniques. Figure 14 presents a
the result also shows that even though the MAC layer creates comparison of the execution time with the code generation
the largest portion of events, the bit-level PHY layer costs (codegen) feature enabled and disabled. The computational
more computational cost in terms of execution time. testbed possesses the following characteristics: i7-6700k at
Among all the layers, the APP layer costs the fewest num- 4.0GHz (CPU), 32G DDR4 2133MHz (memory),Microsoft
ber of events since it is the top layer of the network stack Windows 10 (OS).
and it mainly deals with message generation and reception. In this figure, the execution time increases as the number
In our case, the BSM is the only application data generated in of vehicles increases. The red line indicates the simulations
the APP DES. If additionally applications are involved or the running in interpreted execution mode, i.e., without code
data generation rate is increased, the number of events will be generation. The blue line is the execution time with code
increased accordingly. The PHY link DES involves slightly generation. Based on this figure, the code generation shows
larger number of events relative to the APP layer because its advantage with respect to accelerating the execution speed.
whenever the APP DES generates one message and when For the simulation with 30 vehicles, interpreted execution
this message enters the PHY link DES module in the format costs 7945 seconds, while code generation execution costs
of a waveform entity, it triggers a series of events to deal 5185 seconds. The execution speed of 30 vehicles simulation
with activities such as delay, buffer, and waveform check. increases 53.23% because of code generation.
Therefore, the number of events in the PHY link DES cor-
relates to the number of messages generated in the APP B. PERFORMANCE OF THE PHY LAYER
DES module. The number of events increases dramatically In this section, the performance of the MATLAB PHY layer
in the MAC DES module. This is because the number of is evaluated and compared with the NIST error rate model of
events in the MAC DES module is not only correlated to NS-3, which is broadly adopted by iTETRIS and VSimRTI
the number of APP layer messages but also affected by the projects.
channel status. Suppose if the wireless channel is congested,
the channel sensing operation would be performed more 1) PRECISE PHY LAYER MODELING
frequently in order to monitor the channel status and seek NS-3 is a packet-based, discrete-event network simulator
a transmission opportunity. Consequently, the timer event equipped with several wireless models. When a waveform is
related to the channel sensing operation is called more fre- received, NS-3 calculates the signal to noise ratio (SNR) and
quently, which might cause a burst amount of events in the invokes its error rate model to decide the packet successful
MAC DES module. Additionally, since the PHY Tx and Rx reception rate. Two error rate models are integrated with
functions are integrated with the MAC/PHY DES Module, NS-3: the YANS [44] model and the NIST [45] model.
the events caused by PHY Tx/Rx are shown in the MAC/PHY The YANS model, which is based on an analytical bound
DES module surf instead of PHY Link DES module was replaced with NIST error model in 2010. In this paper,
surf. we only focus on the currently used NIST error rate model.

VOLUME 7, 2019 87259

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

In order to estimate the Packet Success Rate (PSR) for

orthogonal frequency division multiplexing (OFDM) sym-
bols, NIST calculates NEbo (SNR per bit Eb to the one-side noise
spectral density No ) based on SNR in dB using :
Eb
= SNR − 10log10(k), (17)
No
where k = log2 (M ), M is the modulation level, and k is the
number of bits per symbol. However, the NIST error model
has two limitations. FIGURE 15. The packet success rate (PSR) comparison between MATLAB
First, the NIST error model does not consider the over- and NS-3 (NIST model). The simulation is conducted with BPSK
sampling situation and equates SNR to NEos , i.e., ratio of sym- modulation in AWGN channel.

bol energy to noise power spectral density. The relationship

between NEos and SNR in dB for complex signals is defined
by [46]: from NIST_Error_Model, NS-3 randomly corrupts received
Es Tsym packets in order to emulate the packet corruption process.
= SNR + 10log10 , (18) However a realistic PHY layer of a communication system
No Tsamp
consists of more functions including frequency offset correc-
where Tsym is the symbol period of the signal and Tsamp is the tion, channel estimation, modulation, and demodulation. The
sampling period of the signal. For a complex baseband signal, proposed simulator implements all the PHY layer features at
if it is oversampled by a factor of n, then NE0s does not equal to the bit level.
SNR but exceed by 10log10 (n). In a vehicular network environment, the V2x channels
Second, the NIST error model does not account for the have different characteristics compared with other station-
energy in nulls. Take IEEE 802.11p for instance, an OFDM ary indoor channels [47]. First, V2x channels are affected
signal consists of 64 subcarriers, among which 48 subcarriers by longer multipath fading, which increases the possibility
are for data, 4 subcarriers for pilot information and 12 sub- of intersymbol interference (ISI). Second, the transmission
carriers are NULL. Thus, the SNR for occupied subcarriers environment is highly dynamic, which causes significant
in dB, SNRo is calculated using : Doppler effects resulting in more channel fading. When pass-
NFFT ing through the V2x channels, the waveforms are impacted
SNRo = SNR − 10log10 , (19) more than just passing through an AWGN channel. The
Ndata + NPilot
performance in terms of PSR will be degraded, thus chan-
where NFFT is the number of FFT sampling points, i.e., nel tracking techniques are needed in order to enhance the
the total number of subcarriers for a OFDM signal. Ndata performance.
is the quantity of subcarriers used for data and Npilot is for In the proposed simulator, we integrate a time and fre-
pilot. quency selective multipath Rayleigh fading channel as spec-
Figure 15 shows the comparison of PSR on AWGN chan- ified by [48] with an AWGN channel. The conventional
nels between NIST error model and proposed MATLAB error WLAN channel estimation from L-LTF is used for the entire
model. The PSR of the NIST error model is over optimistic packet duration. In order to compensate the high Doppler
while the proposed MATLAB error model is more realistic. spread of the V2x channel, channel tracking is enabled.
As a packet-based network simulator, the NIST model is With channel tracking, the channel estimation obtained from
the most comprehensive NS-3 model can implement. The L-LTF is updated per symbol using decision directed chan-
oversampling situation, as well as the energy in the null nel tracking as presented in [47]. We compare the perfor-
subcarriers, requires processing of the bit level, thus it is mance in terms of PSR on BPSK across different scenarios
potentially challenging for NS-3 to implement these features. including Highway LOS and Urban NLOS with channel
The AWGN channel shown in Figure 15 is a simple sce- tracking (CT) on and off. The results are presented
nario that can be compensated by incorporating an offset to in Figure 16. In this figure, the receiver with channel track-
NS-3 simulator. However, in a more complicated envi- ing (CT) enabled possesses a better PSR in V2x channels.
ronment, such as Non-line-of-sight (NLOS) conditions Due to the restrictions of NS-3 on packet-level process-
in an Urban scenario, a constant offset could poten- ing, implementing channel tracking for OFDM symbols is
tially be insufficient to cope with the different channel relatively difficult to perform. It is worth mentioning the
models. focus of this paper is the presentation of the ability and
accuracy of bit-level processing, and using simple chan-
2) PHY LAYER ON BIT-LEVEL PROCESSING nel models can help in providing a clearer evaluation for
Another limitation of NS-3 is the oversimplified PHY layer. straightforward comparison. Nevertheless, the authors will
The packets are forwarded among objects of Packet class explore more complicated channel models in future research
via methods. Based on the packet-error rate (PER) obtained activities.

87260 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

FIGURE 16. The performance of channel tracking (CT) for BPSK

modulation in multi-path fading channel. The channel models involves
highway line-of-sight (LOS) and urban non-line-of-sight (NLOS).

TABLE 3. Parameters of CSMA and EDCA in the simulations.

FIGURE 17. Performance Evaluation of DCF (CSMA) and EDCA (AC2-3) on

packet delivery latency (PDL) and packet delivery rate (PDR) as the
density of vehicle increases. (a) Packet Delay of EDCA (AC2), EDCA (AC3)
and DCF (CSMA). Delay mainly comes from the backoff process during the
channel access deference stage since no retransmissions are required for
broadcast packets. (b) Packet Delivery Rate (PDR) of EDCA (AC2), EDCA
(AC3) and DCF (CSMA). Packet corruption comes from either packet
C. PERFORMANCE COMPARISON BETWEEN
collision in the channel or severe noise due to the low SNR.
EDCA AND DCF
One significant feature of a vehicular network is adopting
EDCA in the MAC layer. In this section, we will compare the vehicular network. Only a minimum Contention Window
the performance between EDCA and distributed coordination (CW) is used for broadcasting purpose.
function (DCF), i.e., carrier sensing media access (CSMA), In Figure 17(a), the latency of CSMA is smaller than
where the latter is generally used in other IEEE 802.11 prod- EDCA (AC2) but greater than EDCA (AC3). This is due to
ucts. By analyzing the simulation results, we can evaluate the the fact that data possessing different priorities are queued
effectiveness of the proposed simulator. into different ACs, while in CSMA all data are buffered in
The simulation scenario is a unidirectional highway con- the same queue. Whenever an internal collision happens,
sisting of two lanes. The car following model (CFM) is AC2 always gives way to AC3. This is why AC3 shows a
chosen to be the mobility model. The vehicles are broad- steady and better performance in the figure. According to
casting BSMs (AC2) at 10 Hz and AC3 messages using SAE J2735 [35], the maximum latency for safety messages
a Poisson distribution modeled by λ = 2. We performed is 10 ms. When the latency of AC2 is below the threshold,
two sets of simulations for EDCA and CSMA with the key EDCA is shown to be the better option than CSMA.
MAC parameters listed in Table 3. As the PHY layer of Fig. 17(b) compares the Packet Delivery Rate (PDR) of
IEEE 802.11p is derived from IEEE 802.11a, we choose the CSMA with AC2 and AC3 transmissions of EDCA. When
IEEE 802.11a version of CSMA to minimize the difference the number of vehicles increases, the PDR of AC3 maintains
with other layers. at nearly 100 percent. This proves the AC2 and AC3 can
Figure 17 shows the simulation results of the Packet coexist in the same channel, and AC2 traffic affects little
Deliver Latency (PDL) and Packet Deliver Rate (PDR). As all on AC3 traffic. On the other hand, the PDR of AC2 starts
the messages in the simulation are broadcast, there are no to decrease when number of vehicles approaches to 15 due
retransmissions involved, with the latency mainly coming to the packet collisions. When less than 15 vehicles, EDCA
from the channel access deference process, i.e., IFS+backoff. (AC2) still performs better than CSMA. However, when more
In Table 3, the Inter-Frame-Space (IFS) of CSMA, i.e., DIFS, than 15 vehicles are present, CSMA acts better than EDCA
equals to the highest priority IFS (AC3), i.e., AIFSN(AC3), (AC2). This is due to the coexistence of AC2 and AC3, which
and slightly smaller than AIFSN (AC2). The Shortest-IFS increases the packet collision rate. However, for 30 vehicles,
(SIFS) and slot time of IEEE 802.11p is greater than IEEE the PDR of EDCA (AC2) is till above 85 percent, which
802.11a in order to cope with the mobility characteristics of performs well enough on broadcasting BSMs.

VOLUME 7, 2019 87261

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

TABLE 4. Comparisons of features and limitations between proposed simulator and NS-3.

D. COMPARISONS BETWEEN PROPOSED SIMULATOR implementation is significantly slower than NS-3. Further-
AND NS-3 more, VANET Toolbox does not support parallel computing,
NS-3 is a well known discrete-event network simulator that i.e., the model created by VANET Toolbox cannot be oper-
supports full stack standards across a variety of networking ated across multiple cores. Even though SimEvents supports
applications. The NS-3 modules are robust and can operate C/C++ code generation since MATLAB R2017b, it still has
at a faster speed due to its C++ implementation. However, several limitations. For instance, C/C++ code generation
NS-3 potentially possessed several limitations. First, its PHY does not support hash map, persistent variables, or changing
layer is packet-based, which means the minimum data ele- the values of properties of an object inside another object.
ment is at the packet level instead of the bit level. Therefore, Those functions have to be declared as extrinsic functions
the bit-related operations such as channel tracking, channel and cannot enjoy the benefit of C/C++ code generation.
estimation, and frequency offset correction cannot be applied. Thus, the relatively slow execution speed is the major limi-
Additionally, NS-3 lacks supports for real radio hardware as it tation of the proposed simulator. Second, the movements of
is unable to convert information into bits or symbols. Second, entities among different DES modules require the support of
NS-3 was originally designed to operate in a pure network Simulink. Therefore, the VANET Toolbox inherits the limita-
simulation environment. In order to simulate vehicular traf- tions of Simulink. For example, parameters such as the total
fic, NS-3 either uses predefined route information or inter- number of vehicles cannot be changed during the simula-
acts with mobility simulators asynchronously with the help tion since Simulink locks all the parameters. Nevertheless,
of interfaces. The randomness of vehicular traffic scenario all of these limitations may be solved in future releases of
might not be able to be simulated in real-time. MATLAB/Simulink. Table 4 summarized the major features
Our proposed simulation environment compensates for and limitations for both simulators.
the limitations of NS-3. First, it provides a more accurate
PHY layer representation at the bit level. The simulator is VI. CONCLUSION
able to model the channel impairments such as noise, path In this paper, we presented an integrated vehicular net-
loss, or shadow fading on the bits. Moreover, the bits are work simulator called the VANET Toolbox that is based
converted into symbols, which are exactly the same format on MATLAB Discrete Event System (DES). This is the
in the real wireless communications. Thus, it is reasonable to first vehicular network simulator in the MATLAB/Simulink
infer the simulated wireless channel can be replaced by actual environment that supports a full stack of network proto-
software defined radios (SDR) such as the USRP. Real radio cols. The design structure of the main components namely,
transmissions will be evaluated in future research activities the APP layer, MAC layer, PHY layer, and basic mobility
of the authors. Furthermore, as the mobility models are inte- models, were proposed and were demonstrated to accu-
grated with the APP layer, the reciprocal interactions between rately simulate the inter-communications between vehi-
the traffic application and the network communication are cles in different scenarios but not without limitations. The
sufficiently supported. This feature makes the proposed sim- VANET Toolbox requires only MATLAB/Simulink to use it
ulator to simulate vehicular driving operations and network in addition to SimEvents, Communication and WLAN System
communications sufficiently synchronized in real time. Toolbox. The design purpose of the VANET Toolbox is to
On the other hand, the proposed simulator has several provide a framework and an opportunity to attract more
limitations. First, MATLAB is an interpreted programming researchers to improve it and finally benefit the vehicu-
language that aims for precision modeling. Compared with lar network development. The vehicular network simulator,
other compiled programming languages such as C++, which VANET Toolbox, is open source and can be down-
is used in NS-3, fast execution speeds or low computa- loaded from https://github.com/lewangwpi/vanet_toolbox
tional costs are often difficult to achieve in MATLAB, espe- or https://www.mathworks.com/matlabcentral/fileexchange/
cially with the PHY layer of VANET Toolbox, which is 684 37-vanet-toolbox-a-vehicular-network-simulator-based-
designed on bit level processing. It turns out that the proposed on-des.

87262 VOLUME 7, 2019

L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

ACKNOWLEDGMENT [22] T. K. Som and R. G. Sargent, ‘‘A probabilistic event scheduling policy
The authors would like to thank the generous support of for optimistic parallel discrete event simulation,’’ in Proc. 12th Workshop
Parallel Distrib. Simulation, May 1998, pp. 56–63.
the MathWorks and the Brazilian Government Science With- [23] K. Akesson, M. Fabian, H. Flordal, and R. Malik, ‘‘Supremica-an inte-
out Borders program. Furthermore, they would also like to grated environment for verification, synthesis and simulation of discrete
acknowledge the guidance and input for Mike McLernon of event systems,’’ in Proc. 8th Int. Workshop Discrete Event Syst., Jul. 2006,
pp. 384–385.
the MathWorks throughout these research activities. [24] R. Davidrajuh, ‘‘Developing a new Petri net tool for simulation of discrete
event systems,’’ in Proc. 2nd Asia Int. Conf. Modelling Simulation (AMS),
May 2008, pp. 861–866.
REFERENCES
[25] C.-H. Chen, ‘‘A hybrid approach of the standard clock method and event
[1] M. Müller, ‘‘WLAN 802.11 p measurements for vehicle to vehicle (V2V) scheduling approach for general discrete event simulation, ’’ in Proc. 27th
DSRC,’’ in Application Note Rohde Schwarz, vol. 1, pp. 1–25, 2009. Conf. Winter simulation, Dec. 1995, pp. 786–790.
[2] J. B. Kenney, ‘‘Dedicated short-range communications (DSRC) standards [26] S. Zhongyuea and G. Zhongliang, ‘‘Vehicle routing problem based on
in the United States,’’ Proc. IEEE, vol. 99, no. 7, pp. 1162–1182, Jul. 2011. object-oriented discrete event simulation,’’ in Proc. 2nd Int. Conf. Adv.
[3] H. Hartenstein and L. P. Laberteaux, ‘‘A tutorial survey on vehicular ad hoc Comput. Control, vol. 5, Mar. 2010, pp. 638–643.
networks,’’ IEEE Commun. Mag., vol. 46, no. 6, pp. 164–171, Jun. 2008. [27] IEEE Standard for Information Technology–Telecommunications and
[4] M. Chowdhury, ‘‘15th intelligent transportation systems world Information Exchange Between Systems Local and Metropolitan Area
congress,’’ J. Intell. Transp. Syst., vol. 14, no. 2, pp. 51–53, 2010. Networks–Specific Requirements - Part 11: Wireless LAN Medium Access
doi: 10.1080/15472451003738194. Control (MAC) and Physical Layer (PHY) Specifications, Standard,
[5] F. Ahmed-Zaid, H. Krishnan, M. Maile, L. Caminiti, S. Bai, and Dec. 2016.
S. VanSickle, ‘‘Vehicle safety communications—Applications: System [28] Vehicles in Network Simulation (VeinS). [Online]. Available: http://veins.
design & amp, objective testing results,’’ SAE Int. J. Passenger Cars Mech. car2x.org
Syst., vol. 4, pp. 417–434, Jun. 2011. [29] i TETRIS: The Open Simulation Paltform for Intelligent Transport
[6] G. Howe, G. Xu, D. Hoover, D. Elsasser, and F. Barickman, ‘‘Commercial System (ITS) Services. [Online]. Available: http://www.ictitetris.eu/
connected vehicle test procedure development and test results–emergency itetris_platform.html
electronic brake light,’’ Tech. Rep., 2016. [30] VSimRTI—Smart Mobility Simulation. [Online]. Available:
[7] G. Howe, G. Xu, D. Hoover, D. Elsasser, and F. Barickman, ‘‘Commercial https://www.dcaiti.tuberlin.de/research/simulation/
connected vehicle test procedure development and test results–blind spot [31] S. Papanastasiou, J. Mittag, E. G. Strom, and H. Hartenstein, ‘‘Bridg-
warning/lane change warning,’’ Tech. Rep., 2016. ing the gap between physical layer emulation and network simu-
[8] G. Howe, G. Xu, D. Hoover, D. Elsasser, and F. Barickman, ‘‘Commercial lation,’’ in Proc. IEEE Wireless Commun. Netw. Conf., Apr. 2010,
connected-vehicle test procedure development and test results–intersection pp. 1–6.
movement assist,’’ Tech. Rep., 2016. [32] B. Aygun, ‘‘Distributed adaptation techniques for connected vehicles,’’
[9] J. Harri, F. Filali, and C. Bonnet, ‘‘Mobility models for vehicular ad hoc Ph.D. dissertation, WPI, Worcester, MA, USA, 2016.
networks: A survey and taxonomy,’’ IEEE Commun. Surveys Tuts., vol. 11, [33] Mathworks. (May 2017). 802.11n Packet Error Rate Simulation for
no. 4, pp. 19–41, Dec. 2009. 2x2 TGn Channel. [Online]. Available: https://www.mathworks.com
/help/wlan/examples/802-11n-packet-error-rate-simulation-for-2x2-tgn-
[10] H. Hartenstein and K. Laberteaux, VANET: Vehicular Applications and
channel.html
Inter-Networking Technologies, vol. 1. Hoboken, NJ, USA: Wiley, 2009.
[34] C. Michaels, ‘‘DSRC implementatio guide–a guide to users of SAE
[11] D. O. Lazaro, E. Robert, L. Lan, J. Gozalvez, S. Turksma, F. Filali,
J2735 message sets over DSRC,’’ SAE, Pennsylvania, PA, USA, Tech.
F. Cartolano, M. A. Urrutia and Krajzewicz, ‘‘An integrated wireless
Rep., 2010.
and traffic platform for real-time road traffic management solutions,’’
[35] J2735 Dedicated Short Range Communications (DSRC) Message Set Dic-
in COMeSafety Newsletter, 2010.
tionary, Standard, SAE.
[12] L. Bononi, M. Di Felice, G. D’Angelo, M. Bracuto, and L. Donatiello,
[36] J2945 On-Board System Requirements for V2V Safety Communications,
‘‘MoVES: A framework for parallel and distributed simulation of wireless
Standard, SAE.
vehicular ad hoc networks,’’ Comput. Netw., vol. 52, no. 1, pp. 155–179,
[37] Static and Kineticfriction. [Online]. Available: http://ffden-
2008.
2.phys.uaf.edu/211_fall2002.web.dir/ben_townsend/staticandkineticfriction.
[13] M. Killat, F. Schmidt-Eisenlohr, H. Hartenstein, and C. Rössel, P. Vortisch, htm
S. Assenmacher, and F. Busch, ‘‘Enabling efficient and accurate large-scale
[38] NHTSA. (Sep. 1999). Federal Motor Vehicle Safety Standards; Stopping
simulations of VANETs for vehicular traffic management,’’ in Proc. 4th
Distance Table. [Online]. Available: https://www.gpo.gov/fdsys/pkg/FR-
ACM Int. workshop Vehiculr Ad Hoc Netw., Sep. 2007, pp. 29–38.
1999-09-07/pdf/99-23226.pdf
[14] S.-Y. Wang, C. L. Chou, C. H. Huang, C. C. Hwang, Z. M. Yang, [39] L. Wang, R. F. Iida, and A. M. Wyglinski, ‘‘Coordinated lane changing
C. C. Chiou, and C. C. Lin, ‘‘The design and implementation of using v2v communications,’’ in Proc. IEEE 88th Vehicular Technol. Conf.
the NCTUns 1.0 network simulator,’’ Comput. Netw., vol. 42, no. 2, (VTC-Fall), Aug. 2018, pp. 1–5.
pp. 175–197, 2003.
[40] A. Kesting, M. Treiber, and D. Helbing, ‘‘Enhanced intelligent driver
[15] D. Krajzewicz, J. Erdmann, M. Behrisch, and L. Bieker, ‘‘Recent devel- model to access the impact of driving strategies on traffic capacity,’’
opment and applications of SUMO-Simulation of Urban MObility,’’ Int. Philos. Trans. Roy. Soc. A, Math., Phys. Eng. Sci., vol. 368, no. 1928,
J. Adv. Syst. Meas., vol. 5, nos. 3–4, pp. 128–138, Dec. 2012. pp. 4585–4605, 2010.
[16] Nsnam. (Feb. 2017). NS-3 Project-Introduction to NS-3. [Online]. Avail- [41] B. Higgs, M. M. Abbas, and A. Medina, ‘‘Analysis of the wiedemann car
able: https://www.nsnam.org/docs/tutorial/html/introduction.html following model over different speeds using naturalistic data,’’ in Proc.
[17] G. Pongor, ‘‘Omnet: Objective modular network testbed,’’ in Proc. Int. Procedia RSS Conf., 2011, pp. 1–22.
Workshop Modeling, Anal., Simulation Comput. Telecommun. Syst., 1993, [42] A. Schadschneider, ‘‘The nagel-schreckenberg model revisited,’’ Eur.
pp. 323–326. Phys. J. B, Condens. Matter Complex Syst., vol. 10, no. 3, pp. 573–582,
[18] A. F. G. Gil, Traci4matlab: User’s Manual. Universidad Nacional De 1999.
Colombia, 2014. [43] S. Krauß, ‘‘Towards a unified view of microscopic traffic flow theories,’’
[19] C. Gorgorin, V. Gradinescu, R. Diaconescu, V. Cristea, and L. Ifode, IFAC Proc. Volumes, vol. 30, no. 8, pp. 901–905, 1997.
‘‘An integrated vehicular and network simulator for vehicular ad-hoc net- [44] M. Lacage and T. R. Henderson, ‘‘Yet another network simulator,’’ in Proc.
works,’’ in Proc. 20th Eur. Simulation Modelling Conf., vol. 59, May 2006, workshop ns-2, IP Netw. simulator, Oct. 2006, p. 12.
pp. 1–8. [45] G. Pei and T. R. Henderson, ‘‘Validation of OFDM error
[20] M. R. Stiglitz and C. Blanchard, IEEE Standard Dictionary of Electrical rate model in ns-3,’’ in Boeing Research Technology, 2010,
and Electronics Terms. 1992, Sec. 4. pp. 1–15.
[21] C. G. Cassandras and S. Lafortune, Introduction to Discrete Event Systems. [46] MathWorks. AWGN Channel. [Online]. Available: https://www.
Springer, 2009. mathworks.com/help/comm/ug/awgnchannelḣtml

VOLUME 7, 2019 87263

 L. Wang et al.: Vehicular Network Simulation Environment via DES Modeling

[47] J. A. Fernandez, D. D. Stancil, and F. Bai, ‘‘Dynamic channel equal- RENATO IIDA received the B.Eng. degree in
ization for IEEE 802.11p waveforms in the vehicle-to-vehicle channel,’’ telecommunications engineering from the Univer-
in Proc. 48th Annu. Allerton Conf. Commun., Control, Comput. (Allerton), sity of Brasilia, Brazil, in 2002, and the M.Sc.
Sep./Oct. 2010, pp. 542–551. degree in electrical engineering from the Univer-
[48] P. Alexander, D. Haley, and A. Grant, ‘‘Cooperative intelligent transport sity of Brasilia, in 2006. He is currently pursuing
systems: 5.9-GHz field trials,’’ Proc. IEEE, vol. 99, no. 7, pp. 1213–1235, the Ph.D. degree in electrical engineering with the
Jul. 2011. Worcester Polytechnic Institute, Worcester, MA,
USA. From 2008 to 2014, he worked in system
level simulations with Nokia, and was involved in
the standardization of MUROS, VAMOS feature in
GSM network in 3GPP. His current research interests include V2V networks
and to improve the radio resource management in the type of networks. He
received an award from CAPES with science without borders program to
support his Ph.D.

ALEXANDER M. WYGLINSKI received the

B.Eng. and Ph.D. degrees from McGill University,
Montreal, QC, Canada, in 1999 and 2005, respec-
tively, and the M.Sc. (Eng.) degree from Queen’s
University, Kingston, ON, Canada, in 2000, all in
LE WANG received the M.Sc. and Ph.D. degrees electrical engineering. He is currently a Profes-
in electrical and computer engineering from sor of electrical and computer engineering and a
Worcester Polytechnic Institute, Worcester, MA, Professor of robotics engineering with Worcester
USA, in 2013 and 2019, respectively. For the past Polytechnic Institute, Worcester, MA, USA, where
four years, he has been collaborating with Math- he is currently the Director of the Wireless Inno-
Works, Inc., on designing vehicular network simu- vation Laboratory. Throughout his academic career, he has published over
lation environment using MATLAB discrete-event 35 journal papers, over 80 conference papers, nine book chapters, and two
system. His current research interests include textbooks. His current research interests include wireless communications,
wireless communication, vehicular network simu- cognitive radio, software-defined radio, dynamic spectrum access, spectrum
lation, wireless network security, and applications measurement and characterization, electromagnetic security, wireless system
development based on V2x communication. He has been a member of the optimization and adaptation, and cyber-physical systems.
Wireless Innovation Laboratory, since 2011.

87264 VOLUME 7, 2019