Simulation Platform for Connected Heterogeneous Vehicles

Tobias Meuser, Daniel Bischoff, Ralf Steinmetz and Bj

orn Richerzhagen

Multimedia Communications Lab (KOM), Technische Universit

at Darmstadt, Darmstadt, Germany

Keywords:

Vehicular Networks, Simulation, Rapid Prototyping.

Abstract:

The increasing number of connectivity features in current vehicles poses additional challenges for large-scale

vehicular communication systems. Already deployed systems rely on the cellular network infrastructure,

while the Wiﬁ-based 802.11p standard will likely be implemented on a large scale in the next years. As real-

world tests are costly, simulations are used to develop mechanisms for efﬁcient short-range communication via

802.11p. However, efﬁcient long-range communication between vehicles is pivotal for non-safety related in-

formation sharing. Current simulators often focus on short-range communication exchange, while approaches

for efﬁcient long-range communication are barely considered in automotive scenarios.

To enable rapid development of new approaches, we propose a scalable simulation environment for automotive

applications. Our contributions are (i) the realistic modeling of heterogeneous vehicles including sensors and

network interfaces, (ii) the automated generation of road properties like accidents and jams, and (iii) a conﬁg-

urable back-end infrastructure distributing events to the vehicles. All of the above contributions enable rapid

prototyping and evaluation of automotive applications in various environments. We showcase two exemplary

use cases to demonstrate the versatility of our simulation framework: an efﬁcient road-based dissemination

approach for long-range information exchange and a distributed information validation approach.

1 INTRODUCTION

In recent years, increasingly complex Advanced

Driver Assistance Systems (ADASs) were developed,

which require vast amounts of sensor information

(Tigadi et al., 2016). Vehicles exchange information

with each other to improve their perception range and

quality. As practical tests are often unsuitable for

the vehicular scenario, network simulators are used

to model this information exchange. These simula-

tors commonly model the information exchange via

802.11p, the cellular network, or a combination.

Local Inter-Vehicle Communication (IVC) via

802.11p is required for research performed on col-

laborative perception using Cooperative Awareness

Messages (CAMs) (ETSI 102 637-2 V1.2.1, 2011)

or Decentralized Environment Notiﬁcation Messages

(DENs) (ETSI 102 637-3 V1.1.1, 2010), and in-

formation exchange in Vehicular Ad-hoc Networks

(VANETs) in general (Riebl et al., 2015; G

unther

et al., 2015). This type of communication is used for

both safety-related applications like accident preven-

tion and non-safety-related applications like ofﬂoad-

ing of the cellular connection.

The cellular connection is mainly used for the ex-

change of non-safety related information like trafﬁc

information and road properties. If a vehicle receives

information, it can detour to prevent a jam or decel-

erate in-time to prevent emergency braking. The ef-

ﬁcient distribution of these messages is a challenging

topic in research, as the context of the vehicle and

the message need to be considered in the dissemina-

tion. For many non-safety related applications, an ex-

act model of the channel like in (Virdis et al., 2015)

is often not required, as a message drop can often be

compensated on the higher layers.

However, the heterogeneity of the vehicles in

terms of their available network interfaces, computa-

tional resources, and sensors is an essential aspect for

all vehicular applications. Today’s network simula-

tors often focus only on one dimension of heterogene-

ity, which limits the possible simulation scenarios.

Based on this gap, we design our simulation frame-

work for the vehicular scenario which simulates het-

erogeneous vehicles. Additionally, we enable rapid

prototyping of vehicular applications, as we provide

networking, processing and sensing components to

limit the additional implementation effort when eval-

uating new approaches. Our simulation framework

is based on the Simonstrator framework (Richerzha-

gen et al., 2015) and the vehicular trafﬁc simulator

Simulation of Urban Mobility (SUMO) (Lopez et al.,

412

Meuser, T., Bischoff, D., Steinmetz, R. and Richerzhagen, B.

Simulation Platform for Connected Heterogeneous Vehicles.

DOI: 10.5220/0007713004120419

In Proceedings of the 5th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2019), pages 412-419

ISBN: 978-989-758-374-2

2018). SUMO provides realistic movement of the ve-

hicles, while the Simonstrator focuses on rapid proto-

typing of networking applications of various domains.

Our contributions for the simulation of vehicular

scenarios are (i) the setup of the road network in-

cluding the realistic movement of the vehicles, (ii)

the generation events on the roads which can be de-

tected by the vehicles, (iii) the conﬁguration of each

vehicle in terms of available sensors and network in-

terfaces and (iv) the connectivity services utilized by

the vehicles. For the movement of the vehicles, we

couple the Simonstrator and SUMO bidirectionally

using the Trafﬁc Control Interface (TraCI) (Wegener

et al., 2008), i. e., the Simonstrator can inﬂuence the

vehicle movement. Based on the vehicle information

gathered from SUMO, we conﬁgure each vehicle in-

dividually. This conﬁguration includes different stor-

age and computation capabilities, sensors, and net-

work interfaces. Additionally, we provide different

connectivity solutions, including a central back-end

and a pure-forwarding based broker.

All of the above contributions enable rapid proto-

typing and evaluation of automotive applications un-

der varying environmental conditions. The variety of

available communication paradigms and methods in

the Simonstrator like the Publish/Subscribe (Pub/Sub)

paradigm (Richerzhagen et al., 2015), and different

ad-hoc communication schemes and mechanisms for

cellular ofﬂoading (Richerzhagen et al., 2016) further

support rapid prototyping of automotive applications.

Especially the Pub/Sub pattern is pivotal for the au-

tomotive scenario, as inter-vehicle communication is

often information-centered (Amadeo et al., 2016).

The rest of this work structures as follows: In

Section 2, we provide an overview of the existing

simulators for the vehicular scenario. Next, we de-

scribe our ﬁrst contribution, the modeling of the en-

vironment based on the Simonstrator in Section 3.

This modeling includes the road network, the vehi-

cles, and the generation of road events. In Section 4,

we describe the different connectivity conﬁgurations

we provide in our automotive simulation framework

in detail. Next, we show two exemplary use cases in

the vehicular scenario in Section 5. Finally, we con-

clude our paper in Section 6.

2 RELATED WORK

Several network simulators like NS-3 (Riley and Hen-

derson, 2010) and OMNeT++ (Varga and Hornig,

2008) aim to simulate vehicular communication. To

simulate inter-vehicle communication, these simu-

lators model the vehicles’ movement and network

communication. For the vehicle movement, SUMO

(Lopez et al., 2018) simulates the vehicle movement

for all these simulators. For the network simula-

tion, the above simulators differ signiﬁcantly regard-

ing available network models and vehicle modeling.

NS-3 with the RACE extension (Jomrich et al., 2017)

can precisely model the cellular channel including

X2-handover but focuses on the pure networking as-

pects of the vehicular scenario. Thus, the develop-

ment of new applications is complex.

In that regard, OMNeT++ is much more versatile

through the high number of available extensions. The

extension Veins (Sommer et al., 2011) couples OM-

NeT++ with SUMO bidirectionally, such that OM-

NeT++ applications can inﬂuence the movement of

the vehicles. While Veins initially provided only ad-

hoc communication between vehicles, it was later

extended to support the exchange of messages via

Long Term Evolution (LTE) (Hagenauer et al., 2014).

To follow current trends of the automotive industry,

Artery (Riebl et al., 2015) added the ITS-G5 stack to

Veins, which enable the test of novel VANET appli-

cations. In this approach, the vehicle’s onboard sen-

sors are modeled in an extendable manner. However,

Artery did not consider the heterogeneity of sensors

regarding sensor quality. Artery was further extended

by local perception sensors and the LTE communica-

tion stack (G

unther et al., 2017), which increased the

possible vehicular applications further.

While all of the above approaches are suitable for

their intended use cases, they lack at least one of

the following properties which are essential for long-

range communication in vehicular networks: the het-

erogeneity of the vehicles in case of the NS-3 exten-

sion or the lack of different cellular communication

paradigms in case of OMNeT++. This lack increases

the implementation time of vehicular applications and

slows down the development of new features.

Thus, we base our automotive evaluation setup for

non-safety-related applications on the event-based Si-

monstrator (Richerzhagen et al., 2015), which sup-

ports rapid prototyping in different scenarios. While

the Simonstrator only supports a limited amount of

channel models, it provides the possibility to use dif-

ferent communication paradigms and node sensors.

Consequently, we focus more on applications and the

improvement of message dissemination and ﬁltering.

To this end, we model the vehicle including vehicle

sensors, network interfaces, computational resources,

and different back-end communication strategies like

pull-based and push-based communication.

Different back-end communication strategies are

especially valuable for the evaluation of vehicular ap-

plications. Currently, these applications rely mostly

Simulation Platform for Connected Heterogeneous Vehicles

413

Figure 1: Overview of the required components for the

modeling of vehicles.

on a central server entity, which is continuously

queried for new information (Turcanu et al., 2016;

Rayeni et al., 2018). However, the monetary costs for

server maintenance might hinder their practical appli-

cability. These approaches might be enhanced fur-

ther by utilizing highly-scalable low-resource com-

munication paradigms like Pub/Sub (Gascon-Samson

et al., 2015; Majumder et al., 2009), but due to the re-

strictions of the existing simulators, these approaches

cannot be evaluated. Our automotive evaluation setup

resolves this issue and enables the variation of differ-

ent back-end communication strategies.

3 ENVIRONMENTAL MODELING

Figure 1 displays the required components for a net-

work simulator in the vehicular scenario. This sce-

nario consists of vehicles, connectivity features, and

the environment. Each vehicle is equipped with sev-

eral sensors and networking capabilities. A central

back-end may coordinate the message dissemination

and offer additional back-end services. As the sensors

measure the environment, we need to model the envi-

ronment ﬁrst. Section 3.1 describes the modeling of

the environment including the efﬁcient accessibility

of the road network in the network simulator and the

extensible generation of measurable road properties.

Subsequent, Section 3.2 describes the modeling of

the vehicle including the vehicle movement, available

sensors, network interfaces, and storage capabilities.

Using the components from these two sections, Sec-

tion 4 describes the connectivity, which includes dif-

ferent communication paradigms, ad-hoc and cellular

dissemination approaches, and ofﬂoading schemes.

3.1 Environment

For the vehicular scenario, information is commonly

geocasted to a particular area. Geocasting often relies

solely on the location of each vehicle. Thus, no infor-

mation about the road network is required. In recent

years, pure location-based geocasting has proven to

be inefﬁcient, as vehicle follow predictable routes. If

information about the road network is used in the dis-

semination process, the cellular trafﬁc decreases dras-

tically (Meuser et al., 2018b). To this end, the road

network needs to be accessible to develop approaches

that improve communication quality. We retrieve the

road network from SUMO using TraCI.

The road network class is accessible from every

class in the Simonstrator and manages all roads in the

scenario. A road consists of a unique id, the max-

imum allowed speed, its length, the corresponding

lanes, incoming and outgoing roads, and active prop-

erties. The incoming and outgoing roads are required

to enable rapid path search. Additionally, the active

properties describe the current state of the road.

3.1.1 Road Property Modeling

We model a road property if it supports the driver in

his driving decisions and, thus, is shared between ve-

hicles. It provides the creation date of the property,

the value of the property (if necessary), and the de-

fault value of the property, e. g., no jam in case of a

jam property. Currently available properties are fog,

bump, hazard, jam, rain, and trafﬁc signs. However,

new properties can be added easily due to the exten-

sible design of our environment properties.

Some properties like jams inﬂuence the behavior

of the vehicles directly. Thus, these properties change

the representation of the road in SUMO. In that case,

the road is manipulated using TraCI to set the cor-

responding properties like lowering the allowed road

speed. If only start and destination are provided, this

leads to issues in SUMO if shortest-path routing is

utilized. Due to the changes of the road speed, the ve-

hicle’s route might be adjusted. However, this route

change is based on global knowledge, which would

not be available in real-world scenarios. Thus, only

scenarios like the TAPAS Cologne scenario shall be

used currently, as the routes are predeﬁned there.

In our vehicular scenario, the location of road

events has a signiﬁcant inﬂuence on the system per-

formance. As the properties are generated artiﬁcially,

the generation is essential for the simulation results.

3.1.2 Road Property Generation

At the start of the simulation, no road has an active

property, i. e., no roads are jammed or otherwise af-

fected. To this end, the vehicles do not need to com-

municate. After a conﬁgurable time, the generation of

road properties starts. Once a property is created, the

location, duration, type, and the value of this prop-

erty need to be speciﬁed. As these parameters inﬂu-

ence the simulation results, our property generator is

VEHITS 2019 - 5th International Conference on Vehicle Technology and Intelligent Transport Systems

414

Figure 2: Connection of the Simonstrator and SUMO.

extensible by multiple plugins. Each plugin is respon-

sible for the generation of a particular road property.

These plugins choose the location of the property ran-

domly. This randomness can be conﬁgured to either

induce a speciﬁed load to the network or reproduce

real-world behavior based on statistical observations.

Regarding the lifetime of a property, we offer dif-

ferent possible lifetime distributions to simulate dif-

ferent environmental conditions. We designed three

lifetime distributions for our scenario, (i) an exponen-

tial distribution, (ii) a Gaussian distribution, and (iii)

a custom distribution to reﬂect real-world measure-

ments. Each distribution can be conﬁgured according

to the type of the property that is simulated. Once the

event lifetime is reached, the property value changes.

Depending on the chosen generator and the event

type, the behavior of the property value might be ei-

ther continuous or discrete. An example of a con-

tinuous variable is the driving speed in a jam, while

a hazard is a discrete variable with the states haz-

ard and no hazard. For discrete variables, we use a

Markov chain to determine the transitions between

the states of the variable. For continuous variables,

we also divide them into states of similar size, with

a transition probability between the states. Between

the state changes, the variable value is linearly inter-

polated between the two states. To simulate different

types of events, we have one property generator plu-

gin per event type. Thus, it is possible to plugin cus-

tom property generators. That way, new road proper-

ties can easily be added to our simulation.

3.2 Vehicle

Vehicles can only sense road properties using their

equipped sensors. As the range of these sensors is

limited, the movement of the vehicles impacts the

measurable road properties. Thus, realistic vehicle

movement is required for reliable simulation results.

3.2.1 Vehicle Movement

Map-based movement is a common characteristic of

vehicle movement. For map-based movement, the Si-

monstrator targets only human mobility (Richerzha-

gen et al., 2017), which does not satisfy the require-

ments of realistic vehicle movement. Thus, we con-

nect the Simonstrator to the trafﬁc simulator SUMO

to achieve realistic vehicle movement. Figure 2 dis-

plays the technical implementation of the connection.

The vehicle movement class utilizes TraCI to access

the information available at SUMO. Based on this in-

formation, the vehicle movement moves the vehicles

in the scenario. An issue regarding the connection of

the Simonstrator and SUMO is the ﬂuctuating num-

ber of active vehicles in SUMO. Due to simulator

restraints, we cannot generate new hosts at runtime.

To resolve this issue, we initialize a number of of-

ﬂine hosts which are initialized at the start of the sim-

ulation. This number depends on the scenario cho-

sen in SUMO. Once a new vehicle spawns in SUMO,

the vehicle movement selects one host from the set of

available hosts and turns its network interfaces online.

Similarly, the network interfaces are turned ofﬂine af-

ter the vehicle went inactive in SUMO. The vehicle

movement binds the selected host to the respective

vehicle in SUMO. In our simulation framework, ev-

ery application can read information from SUMO and

manipulate the vehicle movement in SUMO. Exam-

ples for information are the future route of the vehi-

cle and the current vehicle speed. To dynamically ex-

change the scenario through different simulation runs,

we mirrored the essential parts of the conﬁguration of

SUMO in the Simonstrator. As simulating the whole

network might not be necessary for huge scenarios,

we provide the possibility to reduce the size of the

simulated area independently of SUMO. Only if a ve-

hicle enters the described area of the scenario, it is

modeled in the Simonstrator, and its network inter-

faces and sensors are activated.

3.2.2 Sensors

For sensing the environment, we model multiple sen-

sors for the vehicular use case. Compared to previous

sensor modeling approaches like in (G

unther et al.,

2017), we explicitly model sensor inaccuracy in our

system. This inaccuracy is important for the vehicu-

lar use case, as perfect information will not be avail-

able. Thus, every sensor has an accuracy value, which

states the probability that the sensor measures the cor-

rect value. We added a conﬁgurable random deviation

to the sensor quality to reﬂect the sensor quality dif-

ference of vehicles in real-world conditions.

In this work, we can simulate both continuous and

discrete events. As each continuous variable can be

approximated with a discrete variable, we perform

most of our evaluations on discrete variables to re-

duce complexity. To this end, we divide the values

of the variable into n possible states. For continuous

variables, n should be chosen such that a required ac-

curacy level is reached.

Simulation Platform for Connected Heterogeneous Vehicles

415

There are different sensor results available: (i) a

single value that can be measured for both continu-

ous and discrete variables, and (ii) a probability vec-

tor that is only available for discrete variables. This

probability vector states the probability that the vari-

able is in a particular state. The advantage of the prob-

ability vector is the provided meta-information about

the inaccuracy of the sensor. This meta-information

is important in the aggregation process, as the impact

of low-quality sensors is reduced.

When a sensor measures the environment, it needs

to determine the measured state. This state depends

on the ﬁnal accuracy of the sensor, i. e., the accu-

racy after the random deviation was added. To reﬂect

different types of environmental variables, we mod-

eled the sensing process using different distributions.

However, for most use cases, the Gaussian distribu-

tion is considered to be most appropriate. Compared

to the work of (Meuser et al., 2018c), we revised the

design of the sensors, which simpliﬁes the conﬁgura-

tion of sensors for different use-cases. Additionally,

new sensors can easily be added through our modular

sensor design. Similarly, new road properties can be

added easily through our extensible design. To pro-

vide a basic set of properties, we implemented sensors

for fog, jam, accident, bump, and trafﬁc signs.

Every simulated automotive application can query

the current observations from the available sensors.

The observations of all plugins are bundled into an

Environment object. Depending on the application

requirements, only some sensor plugins may be re-

quired. Thus, the deployed plugins can be conﬁgured

either on startup or during the request.

At the simulation start, each vehicle is provided

with a conﬁgurable set of sensors with conﬁgurable

quality. This conﬁgurability is pivotal for vehicu-

lar networks, as the heterogeneity of the sensor net-

work is a common topic for vehicular applications.

While the heterogeneity of vehicular sensors is one

important aspect, vehicles are additionally heteroge-

neous regarding their available computation capabili-

ties, storage, and network resources.

3.2.3 Network Resources and Computational

Resources

Vehicles have the possibility to exchange informa-

tion via two network types, decentralized and central-

ized communication. In our simulation, we assume

that the decentralized communication is performed

Wiﬁ-based and the centralized communication is per-

formed based on the mobile network. While Wiﬁ-

based communication is commonly used for CAMs,

it can also ofﬂoad the cellular connection or carry

messages over large distances. However, mobile com-

munication is often more suitable for information ex-

change over large distances. As both communication

types are viable in the vehicular scenario, we provide

the possibility to use both decentralized and central-

ized communication in the vehicles.

We wrapped the road properties from above into

the transmittable RoadInformation class which con-

tains a RoadProperty and additional attributes like ex-

pected event lifetime. The RoadProperty contains the

respective road id, the measurement location, the date

of the measurement, and the value of the property.

With the help of this class, communication is inde-

pendent of the concrete road property implementa-

tion. Dependent on the road property, often one of the

two communication technologies is preferable. For

jam information, LTE is preferable, as the jam in-

formation required early to potentially detour. Con-

trary, information about a bump can be disseminated

via 802.11p, as the driver only needs to slow down.

However, not all vehicles can communicate via both

of these technologies. Thus, a single technology is

often not sufﬁcient for vehicular communication. To

design applications solving this issue, we conﬁgure

the network interfaces for each vehicle individually.

Depending on the available communication technolo-

gies, the Simonstrator provides different communica-

tion patterns. These include the Pub/Sub communica-

tion paradigm (Richerzhagen et al., 2015), ad hoc dis-

semination mechanisms and gateway selection strate-

gies (Richerzhagen et al., 2016). With the availability

of these patterns, new applications can easily be eval-

uated without consideration of the underlying com-

munication paradigm. The possibilities for communi-

cation are discussed in Section 4.

Every time vehicles exchange measurements, they

either use the received measurements immediately for

decision-making or store them locally. While the im-

mediate use of measurements requires only a few re-

sources, the storing of information requires storage

capabilities of the vehicles. While the limitation of

storage seems to be negligible in this scenario, the

huge number of roads with possibly important infor-

mation might lead to storage issues. Thus, we mod-

eled the cache of the vehicle with different sizes and

invalidation strategies. As measurements might be in-

consistent, the vehicles need to validate the received

information to use it in their decision-making. We en-

capsulate the validation of information such that it is

independent of the road property and uses only com-

monly available meta-information like location, de-

tection date, and Time to Live (TTL). Based on this

information validation, further investigation of opti-

mal vehicular decision-making is possible.

VEHITS 2019 - 5th International Conference on Vehicle Technology and Intelligent Transport Systems

416

4 COMMUNICATION

COMPONENTS

Connectivity is an essential feature for future vehic-

ular applications. While today’s network simulators

focus on the exact modeling of the data transmis-

sion, the overlay network is mostly ignored. An over-

lay network is on top of the underlay network (LTE,

802.11p) and provides different addressing schemes,

e. g., content-based addressing. Especially, content-

based information dissemination matches the require-

ments of vehicular applications well (Amadeo et al.,

2016). To this end, our goal is to provide different

types of overlay networks to kick-start the develop-

ment and evaluation of vehicular applications.

4.1 Addressing Schemes

For the inter-vehicle communication, overlay net-

works with different addressing schemes are avail-

able. All these schemes consider the location as a

pivotal property for the dissemination process. We

implement existing schemes for addressing based on

the exact vehicle location, geohash

, street segments,

and an information-dependent relevance score.

In the addressing based on the exact vehicle lo-

cation, each vehicle receives all events in a particular

area around its current location. While this addressing

scheme is commonly used for Mobile Ad-hoc Net-

works (MANETs), it is unsuitable for vehicular net-

works due to the high node mobility. This forces

frequent location updates and, thus, induces unnec-

essary load to the network. Based on this considera-

tion, we implemented the geohash concept in the Si-

monstrator. The geohash differs from the addressing

scheme based on exact location, as each vehicle only

subscribes to the grid cell it is currently in. The size

of the subscription area can be adjusted with increas-

ing vehicle speed, which further reduces the number

of location updates. However, even this addressing

scheme induces a considerable overhead: Informa-

tion is often not relevant in the whole grid cell, but

only for a small set of roads in that cell. To alleviate

for this issue, we implemented a road-based address-

ing scheme. In this addressing scheme, vehicles sub-

scribe to street segments and receive the information

of these segments. The subscriptions to street seg-

ments reduce the overhead compared to the address-

ing based on exact location. However, the overhead

is higher compared to the geohash approach, which

is justiﬁed by the more efﬁcient distribution of pay-

load messages. For road-based addressing, each ap-

plication must specify the set of roads, for which the

http://geohash.org/

information is relevant. In addition to this, we pro-

vide a relevance-aware addressing scheme which dis-

seminates measurements depending on the context of

both the measurement and the vehicle. The relevance-

aware addressing scheme simpliﬁes the conﬁguration

of the dissemination strategy drastically, as the appli-

cation only needs to specify a threshold for the rel-

evance between 0 and 1. The availability of these

addressing schemes decreases the development time

of new vehicular applications and the evaluation time

of new addressing schemes drastically. Moreover,

these strategies are applicable to every communica-

tion strategy, which are presented in the following.

4.2 Communication Strategies

There are two communication strategies, pull-based

communication, and push-based communication.

Both communication strategies are suitable for local

and cellular communication and have their speciﬁc

challenges and advantages. Pull-based communica-

tion relies on periodic pulling of information, which

induces additional network load for the requests and,

thus, reduces the available bandwidth for payload

messages. Push-based communication omits these

requests, but might share information that is not re-

quired by the receiver.

The normally used strategy differs for the under-

lying communication technology. Local communica-

tion via 802.11p is commonly push-based, as the cur-

rently available messages like CAM and Decentral-

ized Environment Notiﬁcation (DEN) are sent peri-

odically or event-based. Thus, pull-based communi-

cation would increase bandwidth usage without bene-

ﬁting the network performance. Cellular communica-

tion is often performed pull-based, as current vehicu-

lar applications require only a few data, which leads to

few pull requests. A central server manages all avail-

able information and distributes them to the vehicles.

Future vehicular applications will require an increas-

ing amount of information. Thus, the load on this

central server and the cellular network will increase.

Push-based communication generally decreases the

load on the server, as less server storage and compu-

tational power is required. This load decrease is due

to the local processing of information in the vehicles.

Additionally, no requests like in pull-based communi-

cation are required, which might decrease the overall

consumed bandwidth. To model both current and fu-

ture communication scenarios, we implemented both

communication strategies, including strategies com-

bining push-based and pull-based communication. To

further reduce the load on the cellular network and

compensate for dead spots, we implement strategies

Simulation Platform for Connected Heterogeneous Vehicles

417

for ofﬂoading the cellular connection based on Wiﬁ.

The required server conﬁguration is described in the

following section.

4.3 Backend Conﬁguration

The required amount of server resources depends

on the used communication strategy. While pull-

based communication generally requires computa-

tional power and storage, push-based communication

only requires information about the vehicle’s posi-

tions. Thus, the required server resources for push-

based communication are generally lower compared

to pull-based communication, as vehicles manage and

aggregate provided information.

For pull-based communication, the server often

aggregates information centrally and distributes the

aggregates to the vehicles. However, relying only

on a central server limits the scalability and contra-

dicts the edge computing trend for mobile networks.

If the server resources are not sufﬁcient or the server

costs should be minimal, push-based communication

is generally preferable. A push-based server with-

out storage provides information of lower quality to

the vehicles, which the vehicles aggregate locally us-

ing their computational resources. Additionally, a hy-

brid strategy is available, in which information is dis-

tributed pull-based or push-based dependent on dif-

ferent information properties. As the performance of

these approaches depend on the available server re-

sources, we provide the possibility to conﬁgure the

server regarding available storage.

5 EXEMPLARY USE CASES

In this section, we provide an overview of our pre-

vious work, which used our simulation platform

(Meuser et al., 2018b; Meuser et al., 2018a). Due

to space constraints, the reader is referred to the re-

spective papers for an in-depth description of the eval-

uation results. The difference in application scenar-

ios shows the versatility of our simulation framework

which is freely available

. To support the evaluation

of developed approaches, we designed different met-

rics assessing the behavior of developed approaches

from different perspectives. Examples of available

metrics are the quality of information at the vehicles,

the produced network trafﬁc, and the computational

effort of the approaches.

In (Meuser et al., 2018b), we evaluated our ap-

proach with a focus on information validation for ve-

hicular applications. Thus, we used vehicles with a

dev.kom.e-technik.tu-darmstadt.de/simonstrator/

2000 m

0.0 0.2 0.4 0.6 0.8

Information Dissemination

(a) Exemplary informa-

tion dissemination range

for a threshold of 1% in

(Meuser et al., 2018b).

(b) Required process-

ing for the information

validation approaches in

(Meuser et al., 2018a).

Figure 3: Additional insights to our approaches of our pre-

vious work.

default communication interface to speed the devel-

opment process of their information validation ap-

proach and focused on information quality metrics.

Figure 3(b) shows the computational overhead of the

different approaches, which can be useful to evaluate

the practical applicability of the developed informa-

tion validation approaches.

In (Meuser et al., 2018a), we focused on a dis-

semination strategy, which aims to distribute informa-

tion efﬁciently while preserving communication qual-

ity. Thus, we used the available information valida-

tion modules to concentrate on the networking as-

pect of the investigated problem and used network

trafﬁc metrics. Figure 3(a) displays the communi-

cation range of one speciﬁc message. This visual-

ization can be very useful to analyze the impact of

different environmental conditions on the dissemina-

tion and support the development of new dissemina-

tion approaches.

6 CONCLUSION

In this work, we present a platform for the rapid de-

velopment and evaluation of automotive networking

applications. Compared to existing simulators, our

platform focuses on the heterogeneity of the vehicles

under varying environmental conditions. For the ve-

hicles, this includes different sensors, storage capa-

bilities, and networking components. For the envi-

ronment, we simulate different types of measurable

events. Additionally, we provide different server con-

ﬁgurations to evaluate varying networking conditions.

We demonstrate the versatility of our platform by

evaluating different types of automotive applications.

While the focus of both applications differs, our plat-

form can be used to evaluate both applications. Addi-

tionally, our platform is used by many researchers and

continuously extended. Currently, we are extending

VEHITS 2019 - 5th International Conference on Vehicle Technology and Intelligent Transport Systems

418

our platform with strategies for ofﬂoading the cellular

connection and efﬁcient event dissemination.

ACKNOWLEDGEMENTS

This work has been funded by the DFG within the

CRC 1053 - MAKI (B1, C2).

REFERENCES

Amadeo, M., Campolo, C., and Molinaro, A. (2016).

Information-centric networking for connected vehi-

cles: a survey and future perspectives. IEEE Commun.

Mag., 54(2):98–104.

ETSI 102 637-2 V1.2.1 (2011). Part 2: Speciﬁcation of Co-

operative Awareness Basic Service. Standard, Euro-

pean Telecommunications Standards Institute, Sophia

Antipolis Cedex, FR.

ETSI 102 637-3 V1.1.1 (2010). Part 3: Speciﬁcations of

Decentralized Environmental Notiﬁcation Basic Ser-

vice. Standard, European Telecommunications Stan-

dards Institute, Sophia Antipolis Cedex, FR.

Gascon-Samson, J., Garcia, F.-P., Kemme, B., and Kien-

zle, J. (2015). Dynamoth: A Scalable Pub/Sub Mid-

dleware for Latency-Constrained Applications in the

Cloud. In Proc. of IEEE ICDCS. IEEE.

unther, H. J., Timpner, J., Wegner, M., Riebl, R., and

Wolf, L. (2017). Extending a holistic microscopic

IVC simulation environment with local perception

sensors and LTE capabilities. Vehicular Communica-

tions, 9.

unther, H.-J., Trauer, O., and Wolf, L. (2015). The poten-

tial of collective perception in vehicular ad-hoc net-

works. In Proc. of IEEE ITST. IEEE.

Hagenauer, F., Dressler, F., and Sommer, C. (2014). Poster:

A simulator for heterogeneous vehicular networks. In

Proc. of IEEE VNC. IEEE.

Jomrich, F., Wankhede, T., Ruckelt, T., Burgstahler, D.,

Bohnstedt, D., and Steinmetz, R. (2017). Demo: rapid

cellular network simulation framework for automotive

scenarios (RACE framework). In Proc. of IEEE Net-

Sys. IEEE.

Lopez, P. A., Behrisch, M., Bieker-Walz, L., Erdmann, J.,

otter

od, Y.-P., Hilbrich, R., L

ucken, L., Rummel, J.,

Wagner, P., and Wießner, E. (2018). Microscopic traf-

ﬁc simulation using sumo. In Proc. of IEEE ITSC.

IEEE.

Majumder, A., Shrivastava, N., Rastogi, R., and Srinivasan,

A. (2009). Scalable Content-Based Routing in Pub/-

Sub Systems. In Proc. of IEEE INFOCOM. IEEE.

Meuser, T., Lieser, P., Nguyen, T. A. B., B

ohnstedt, D., and

Steinmetz, R. (2018a). Lifetime-Aware Information

Aggregation under Uncertainty for Advanced Driver

Assistance Systems. Ad Hoc Networks.

Meuser, T., Richerzhagen, B., Stavrakakis, I., Nguyen, T.,

and Steinmetz, R. (2018b). Relevance-Aware Infor-

mation Dissemination in Vehicular Networks. In Proc.

of IEEE WoWMoM 2018. IEEE.

Meuser, T., Wende, M., Lieser, P., Richerzhagen, B., and

Steinmetz, R. (2018c). Adaptive Decision Making

based on Temporal Information Dynamics. In Proc.

of VEHITS. SciTePress.

Rayeni, M. S., Haﬁd, A., and Sahu, P. K. (2018). Quality of

service aware multicasting in heterogeneous vehicular

networks. Vehicular Communications, 13.

Richerzhagen, B., Stingl, D., Rueckert, J., and Steinmetz,

R. (2015). Simonstrator: Simulation and Prototyp-

ing Platform for Distributed Mobile Applications. In

Proc. of EAI SIMUTOOLS. ACM.

Richerzhagen, N., Richerzhagen, B., Stingl, D., and Stein-

metz, R. (2017). The human factor: A simulation en-

vironment for networked mobile social applications.

In Proc. of IEEE NetSys. IEEE.

Richerzhagen, N., Richerzhagen, B., Walter, M., Stingl, D.,

and Steinmetz, R. (2016). Buddies, not enemies: Fair-

ness and performance in cellular ofﬂoading. In Proc.

of IEEE WoWMoM. IEEE.

Riebl, R., G

unther, H.-J., Facchi, C., and Wolf, L. (2015).

Artery: Extending Veins for VANET applications. In

Proc. of IEEE MT-ITS. IEEE.

Riley, G. F. and Henderson, T. R. (2010). The ns-3 Net-

work Simulator. In Modeling and Tools for Network

Simulation. Springer Berlin Heidelberg, Berlin, Hei-

delberg.

Sommer, C., German, R., and Dressler, F. (2011). Bidirec-

tionally Coupled Network and Road Trafﬁc Simula-

tion for Improved IVC Analysis. IEEE TMC, 10(1).

Tigadi, A., Gujanatti, R., and Gonchi, A. (2016). Advanced

Driver Assistance Systems. iJERGS, 4(3).

Turcanu, I., Sommer, C., Baiocchi, A., and Dressler, F.

(2016). Pick the right guy: CQI-based LTE forwarder

selection in VANETs. In Proc. of IEEE VNC. IEEE.

Varga, A. and Hornig, R. (2008). An overview of the

OMNeT++ simulation environment. Proc. of SIMU-

TOOLS.

Virdis, A., Stea, G., and Nardini, G. (2015). Simulat-

ing LTE/LTE-Advanced Networks with SimuLTE. In

Simulation and Modeling Methodologies, Technolo-

gies and Applications. Springer.

Wegener, A., Piorkowski, M., Raya, M., Hellbr

uck, H., Fis-

cher, S., and Hubaux, J.-P. (2008). Traci: An interface

for coupling road trafﬁc and network simulators. Proc.

of CNS.

Simulation Platform for Connected Heterogeneous Vehicles

419