7 CONVAS: Connected Vehicle Assessment System for Realistic Co-simulation of Traffic and Communications – Networking Simulation for Intelligent Transportation Systems

7
CONVAS: Connected Vehicle Assessment System for Realistic Co-simulation of Traffic and Communications

7.1. Introduction

Connected vehicle technology enables vehicles to communicate with each other and the infrastructure wirelessly. Automated vehicle technology senses the driving environment and operates a vehicle with limited or even without human input. These technologies together provide a platform for creating a wide array of applications to address real-world problems of how to assist and improve mobility, safety and the environment through next generation Intelligent Transportation Systems (ITS). With a limited initial market penetration, emerging technology components and unknown human behavioral responses, we view realistic simulation as a very powerful and costeffective method for testing, developing and evaluating various components of these new technologies.

Traditional traffic simulation focuses on microscopic road behaviors, simple interactions between vehicles and interactions with the transportation infrastructure. Communications are often assumed to be ideal or are crudely simplified for modeling purposes. In many cases, communication effects are post-processed using detailed vehicle trajectories and do not affect the traffic simulation. The major limitation of such approaches is that simulated vehicles make no adjustments based on the packets received or wireless reception characteristics. Integrating both traffic and communication simulations at a fine time scale for large simulations becomes a challenge. Furthermore, realistic wireless communication models for traffic simulation become increasingly necessary for evaluating the impacts of new technologies especially for time-critical applications. For example, there is no widely agreed upon physical model for Dedicated Short Range Communications (DSRC) over the 75 MHz of spectrum (5850 to 5925 MHz) using the IEEE 802.11p standard. This spectrum was allocated by the FCC, and standards including IEEE 802.11p (Physical and MAC layers), IEEE 1609.1-4 (upper protocols) and SAEJ2735/2945.1 (application layer, targeting vehicular applications) have been developed to support ITS [KEN 11]. The number of variables to take into account is countless: communication frequency, vehicle movement, antenna mounting and type, speed of transmission, traffic density, environment type (from urban to highway), weather, to name just a few. There exist several advanced wireless network simulators capable of realistic modeling of wireless networks, but these simulators are not designed for simulating transportation applications.

While there is agreement among experts on the lack of practical measurements and studies to characterize the 5.8–5.9 GHz DSRC channels, recent large-scale efforts bring about the possibility to simplify models while accounting for the large number of parameters discussed above. The Research Data Exchange (RDE) from US DOT has been created as a transportation data sharing system for archived and real-time data from multiple vehicle probes to support the development, testing, and demonstration of mobility applications and connected vehicle research. In this chapter, we will be particularly interested in the Safety Pilot Model Deployment (SPMD) dataset that provides kinematic, geospatial and connectivity data of approximately 3000 equipped vehicles in Ann Arbor, Michigan. Can such data simplify the problem of realistic modeling of the wireless communication for traffic simulation?

In this chapter we describe the Connected Vehicle Assessment System (CONVAS), which flexibly integrates a traffic simulator with a communication simulator, providing an ideal platform for co-simulating transportation system applications. Its communication models can be tuned based on real-world measurements (e.g. from the Michigan SPMD test bed) in scenarios such as urban, residential and highway traffic. We advocate for representations of real-world wireless channels that capture uncertainty in the existing data and calibrations procedures. The platform can be used to test, validate and assess vehicular communications under a variety of operating traffic and communication conditions and settings.

The layout of the chapter is as follows: section 2 reviews research related to co-simulation of traffic and wireless communications, realistic channel models and current challenges. Section 3 overviews the CONVAS co-simulation platform and its two main components, traffic and communication simulation modules. Section 4 examines existing channel model options and our approach to implementing a parametric stochastic simulation model, called lognormal-Nakagami. Section 5 discusses the real data used and the procedure to tune the aforementioned channel model’s parameters for different scenarios, in agreement with the Michigan Safety Pilot Model Deployment data capture. Section 6 presents the CONVAS implementation of a connected vehicle application, Intelligent Dilemma Zone Avoidance that tackles the yellow traffic light dilemma by using the Signal Phase and Timing (SPaT) messages. Section 7 discusses overall CONVAS results with the application for both the communication and the application components. Finally, section 8 summarizes the main results of this work and future research ideas.

7.2. Related work

As connected vehicle technology advances and new standards are defined by regulatory and standardization organizations, many efforts have been seen in the literature to allow testing of applications using this technology. A common factor in these efforts is the integration between a road traffic simulator and a network simulator, thus attempting to provide realistic models in both scopes. However, the manner in which the interlinking between simulators is achieved can greatly affect the realism with which Connected Vehicle (CV) applications are evaluated.

One approach in interlinking the two simulators is executing the road traffic simulation and exporting its results, i.e. the mobility traces, to the network simulator. Examples of such an approach can be found in [BLU 04, KAI 11]. However, it has been shown in [SOM 08b] that this approach does not model the effects of inter-vehicle communication on the driving behavior. For example, it cannot show how vehicles can change their paths or lanes to avoid congestion based on the traffic information received in the vehicular network. Therefore, it is not suitable for providing an in-depth evaluation on the performance of Connected Vehicle Applications.

Examples of co-simulators which present a finely grained interaction between traffic and network simulators are Veins [SOM 11b], TRANS [PIO 08], NCTUns [WAN 09], iTETRIS [RON 13], MOVES [BON 08] and MSIE [LOC 05]. In particular, Veins supports a coupled network and road traffic simulation using well-established simulators from both communities, namely OMNeT++ and SUMO respectively. The co-simulation is achieved by extending each simulator with a communication module, enabling exchange of commands and mobility traces via TCP connections. From the communication simulation perspective, the OMNeT++ default channel models were extended with propagation models for two-ray interference [SOM 11c] and signal attenuation by buildings [SOM 11a], which were validated on experiments with a few vehicles on different scenarios.

Another co-simulator worth considering is iTETRIS. Its architecture comprises a unique module, the iCS Facilities, interfacing three well-defined environments: (1) iAPP or Applications; (2) ns-3 as a network simulator; (3) SUMO as a traffic simulator. iTETRIS implements four different access technologies in ns-3: ETSI IST G5A, WiMAX, UMTS and DVB-H. In addition, ns-3’s default propagation models have been extended to include models fit for urban and highway scenarios taken from the open literature [CHE 07, WIN 07].

However, in spite of allowing for tight interlinking between simulators, the validation of the realism of the presented models against real-world data, if present, was carried out only on a small set of vehicles and road-side equipment. Even in the cases when the authors used widely accepted simulation tools, such as ns-3 or OPNET, there is a need to calibrate and extend the off-the-shelf network and propagation models in order to adapt them to the standards of interest and to the peculiarities of the vehicular environment. In addition to this, the modeling of a particular application is only explicitly visible to the end user in the cases of iTETRIS and MSIE, in which the authors, however, used off-the-shelf versions of the network simulators.

Related areas of research are the evaluation of appropriate mobility models for vehicle-to-vehicle and vehicle-to-infrastructure communication and measurement studies to confirm theoretical characterizations of the DSRC channels. Many of the studies regarding channel models for vehicular communication converge on the use of the Nakagami model, which has been well developed theoretically. In spite of this, there is no agreed upon overall channel model for V2V, and researchers typically adopt combinations of models satisfying various constraints (e.g. short versus large distances, line-of-sight (LOS) versus no line-of-sight (NLOS)). In contrast to the wealth of literature on channel models, there is a lack of data from practical measurements. The many interference opportunities for the 5.9GHz band for DSRC make it even harder to converge on one standard model. Some of these are radar, fixed satellite services, amateur use for the same band, and use of the adjacent bands below 5850 MHz and above 5925 MHz. Industrial, scientific and medical operations of the same band are additional sources of interference in the 5850–5875 MHz portion of the band. This section of the spectrum is presently being evaluated for sharing with WiFi devices based on the IEEE 802.11ac standard [CHA 15].

Table 7.1 summarizes the most relevant state-of-the-art models and parameters applied in vehicular networks. In order to derive the models and parameters, the authors carried out field trials on a small set of vehicles and Road Side Units in different environments. The models account for the signal attenuation over distance by means of the path loss exponent α; occasionally represented as a dual slope model, i.e. different values of α as a function of distance. Large-scale fading and small-scale fading are represented by σ, the standard deviation used in lognormal shadowing, and m, the shape of the gamma distribution in the Nakagami model. In addition, P L0 represents the additional constant loss added to the total, which is measured at a close distance d0 to the transmitter.

At the same time, other authors heuristically define or select parameters for their propagation models. This is the case of [HAF 13], in which the authors present an analytic mobility model and its performance analysis for broadcasting via DSRC while taking into account the distance between the transmitter and receiver, speed and vehicle densities using the Nakagami propagation model.

Table 7.1. Parameter selections for common propagation models

EnvironmentModeld0(m)PL0(dB)ασmReference
UrbanFree space--2.2--[SOM 11c]
“Free Space”
Highway
Free space--2--[EEN 09]
Urban LOSFree space--2.7–5--[EEN 09]
Urban NLOSFree space--3–5--[EEN 09]
“Outdoor” HighwayLognormal shadowing--24–12-[EEN 09]
UrbanLog-distance30802.02–2.13--[ROI 14]
SuburbanLognormal shadowing--2.564.0-[KAR 07]
HighwayLognormal shadowing1063.31.773.1-[KAR 11]
UrbanLognormal shadowing10621.681.7-[KAR 11]
SuburbanLognormal shadowing1064.61.592.1-[KAR 11]
SuburbanNakagami--2.1–3.8-0.16–5.8[ISL 13] Dual Slope model
SuburbanNakagami--2.2–2.4-1[BAG 12]

As stated, the wide variety of propagation models and parameters in the literature is partially caused by a lack of data from extensive field trials covering real-world scenarios. In [DRE 14], the authors list three relevant efforts for real-world data collection: (1) the simTD German project [STU 10], conducted by professional instructed drivers in a controlled environment; (2) the ongoing CAMP (Crash Avoidance Metrics Partnership) project [LUK 12], whose data has not yet been made publicly available; (3) the SPMD (Safety Pilot Model Deployment) [MIC 12], with data readily available from experiments carried out in Michigan. The SPMD stands out in this list as it appears to capture real-world driver behavior with greater accuracy given that the experiments were conducted by common drivers who were allowed to move freely in the city and surroundings of Ann Arbor, Michigan.

In contrast to the multitude of attempts to describe the propagation models for vehicular environments, ITS applications have not been explored in the same measure by the research community. iTETRIS, however, presented two applications in an effort to illustrate the capabilities and potential of the platform. The applications are Cooperative Traffic Congestion Detection (CoTEC) that enables detection of congestion without any fixed infrastructure sensors, and Cooperative Bus Lane Management for allowing private vehicles to use bus lanes when high traffic density is detected.

7.3. CONVAS co-simulation platform

CONVAS is a platform used to test and evaluate connected vehicle applications, automated control, and autonomous driving technologies using simulations of both traffic and communications, which emphasizes the use of commercially available tools, existing communication protocols and standards. Two types of simulators form the skeleton of CONVAS: traffic simulation, such as the Vissim microscopic traffic simulation system, and communication simulation, such as ns-3 or OPNET models and tools. The key feature of CONVAS is the tight integration of the two simulators, in such a way that future events in the traffic simulator are influenced by previous events in the communication simulator and vice versa. The resolution of the co-simulation is determined by the configurable parameter Δt, namely, the period of time each simulator runs individually (e.g. 100 msec).

Vissim traffic simulation creates and populates a given traffic scenario consisting of vehicles and infrastructure elements, defines the vehicle control logic and driver behavior parameters for all vehicles within the simulation environment, exchanges traffic control events and status, and ensures vehicle movement in the simulated world that replicates the physics grounded movement of real vehicles. Vissim is a time-based simulation system, which will advance simulation time by the constant time step Δt.

Communication simulation is driven by a discrete event network simulator that provides high-fidelity packet transmission modeling and detailed analysis capabilities for very large wired and wireless networks. It should allow CV application developers to assess the impact of real-world communication issues such as received power, signal-to-noise ratio, path loss, channel utilization, packet errors and packet delays on the performance of the connected vehicles and infrastructure-based applications. For V2X applications, we are interested in modeling networks using up-to-date wireless protocols such as WiFi-based DSRC, LTE and WiMAX. We have considered two options: OPNET Modeler from Riverbed Communications and ns-3. In the case of OPNET, the wireless models have been extensively tested by consortia. In addition, the open-source alternative ns-3 includes a sufficiently detailed DSRC model that implements the 802.11p and WAVE 1609 protocols.

The details of the simulation, including the application to be simulated, the communication parameters (such as propagation model or transmission power), the traffic density, driver profiles, and other traffic parameters such as Connected Vehicle penetration, are configured in what we call the Application Testing Environment (ATE). Presently, CONVAS has interfaces supporting Vissim for traffic simulation and both OPNET Modeler and ns-3 for network simulation. Details of our implementation are given below.

Initialization of a simulation defines all the static and mobile nodes that could transmit information throughout the entire duration of the simulation. Each node is uniquely identifiable and will be known in both traffic and communication simulators. For each node, information such as the type of node (static or mobile), initial position, antenna characteristics (power, orientation, pattern, etc.) is specified. The traffic simulator passes to the communication simulator: (1) the set of nodes that transmit new information over the most recent Δt period and, for each such node, also the number and type of packet sent and the size of payload (i.e. the amount of information to be transmitted); (2) the present position and the heading over the duration Δt for each mobile node. The trajectory for each mobile node is assumed to be linear and at uniform velocity between two points. The communication simulator passes to the traffic simulator the set of the nodes that will have received packets over the most recent Δt execution period and, for each such node, the type of packet received and the time when the packet was received relative to the present interval.

Their integration follows a client–server model, where the Traffic Simulation Environment (TSE), namely Vissim, is the server and the Communication Simulation Environment (CSE), namely ns-3 or OPNET, acts as a client, and is described in more detail in [SON 17]. Specifically, we implemented TCP sockets directly into both Vissim API and each network simulator. The server will be in listening mode waiting for the socket connection from the CSE upon the start of the simulation. Once the connection is established, each simulation time step is advanced with the exchange of the data through the socket.

7.4. Realistic DSRC channel models

The wireless medium in vehicular communications has characteristics that make it unique. Both transmitters and receivers are mobile and their relative movement creates Doppler shifts, there exist large metal objects constantly moving (other cars), antennas are placed at low elevations and the channel is statistically non-stationary or even random. The strong dependence on the environment and the dynamism of the state of the vehicular environment make it necessary to differentiate among several types of scenarios of interest, such as urban, residential and highway scenarios.

Specifically, we are interested in simulating communication in the following prototypical scenarios in an intersection or on the road: (1) Urban scenario represents streets with two to four lanes guarded by large buildings from the side, or most sides of an intersection, and sidewalks; (2) Residential/Suburban scenario represents residential two lane streets and intersections in residential areas, characterized by smaller dimensions, possible winding shape, with trees and parceled houses along the street; (3) Highway scenario represents broader arteries with two to three lanes per direction, which are flanked by large open spaces and forests.

In simulations of vehicular wireless communications, the propagation model used plays an important role since the received power is crucial when determining whether a packet is received or not. In order to simulate the wireless medium realistically, we distinguish large-scale and small-scale effects in radio wave propagation phenomena. Large-scale effects include reflection, diffraction and scattering. Reflection occurs when a wave encounters a large medium with a different refractive index to air. In models, reflection is often translated to a path loss exponent. Diffraction is a phenomenon explained by Huygens’ principle, which states that every point on a wave front acts as a seed for a subsequent wave front to enable waves to propagate around edges or holes. This effect can be modeled with the knife-edge diffraction model, which can be used for site-specific modeling of propagation over hills and large buildings, for example. Scattering of a radio wave occurs when the wave encounters an object whose size is comparable to the wavelength, of the order of tens of centimeters. The effect of this phenomenon is the spreading of the wave in all directions. This can account for a received signal that is stronger than what would have been predicted by reflection or diffraction alone. Scenarios (1) and (3) above have strong large-scale effects that need to be captured by our model.

Small-scale effects include fading. At the receiver, multiple versions of the original signal superimpose. They may be reflected and diffracted, and arrive with time and phase differences. These multi-path waves interfere with each other, which can cause large fluctuations in signal quality with apparently small changes in time or receiver location. This relative motion causes frequency modulation because each multi-path will have a different Doppler shift (variation in the perceived frequency of the signal as a result of the relative speed between the receiver and the transmitter). V2V channels tend to show higher Doppler spreads than conventional mobile radio channels. Scenario (2) exhibits small-scale effects that will be captured by our model.

The OPNET Modeler is equipped with several propagation models: Free space, Longley-Rice, forest, CCIR, HATA and Walfisch–Ikegami [RIV 16]. ns-3 similarly offers the following channel models: Friis, Two-ray ground propagation, Log-distance, Nakagami, Range, etc. [NS 15]. Many of these models are restricted to certain frequencies and distance ranges or are simply not suitable for vehicular environment.

The next section describes how we built on the existing channel models to create and tune our channel model based on real-world data. We first formally describe the channel model used, a combination of lognormal and Nakagami models. We then show how real data is used to tune the parameters of the resulting lognormal-Nakagami channel model.

7.4.1. CONVAS propagation models

We have created and tuned specific channel models to extend the standard options available and take into account the special characteristics of the transportation environment. A deterministic model, taking into account signal attenuation by buildings or ray tracing, results in very accurate estimations of path loss and other signal degradation effects. However, its implementation would introduce significant overhead and increase simulation time. The path loss algorithm needs to be executed for every pair of TX and RX antennas on a per packet basis. In the case of Ray tracing, it is necessary to model all rays emanating from the source towards every receiver due to the broadcast nature of the communication. Thus, the number of potential receivers scales up with the square of the number of vehicles times the number of paths. On the other hand, stochastic models offer less accurate but reliable enough results in exchange for a rapid execution and easy implementation. The most widely accepted stochastic model for the simulation of vehicular communications is the Nakagami model. Rician distributions model fading with a single stronger line-of-sight in the presence of scatterers, while Rayleigh distributions are used to model dense scatterers when no line-of-sight is present. The Nakagami distribution is the more general model that can represent Rician, Rayleigh or fading that is more severe than Rayleigh, depending on model parameters, and thus is capable of describing a wide range of fading situations.

We extended the ns-3 WAVE component for vehicular communications with a lognormal-Nakagami propagation model that can account for real-world shadowing and multi-path fading of the wireless signal, known to be predominant effects in the attenuation of the signal in a vehicular environment (see [EEN 09]). The new propagation model is a combination of two already existing models: lognormal shadowing and Nakagami. It calculates the received power in two steps. First (see [MEC 11]) we take into account the effect of shadowing, where the reception power at a distance x to the transmitter is calculated with a log-distance rule as follows:

where α is the path loss exponent, P L(d0) is the path loss measured at a close distance (x = d0) to the transmitter and σ is the standard deviation of the normal and zero mean random variable X representing the Gaussian noise. The power is expressed in dBm and the path loss is expressed in dB. Second, we apply the effects of multi-path fading, modeled by the Nakagami distribution, to obtain the final received signal power:

where m controls the shape of the gamma distribution. This formula shows the power expressed in Watts. Depending on the parameter values used in both equations, we can simulate different types of scenarios, such as urban, suburban and highway. Parameter sets for these scenarios have been tuned separately from the real-world data (see the following subsection).

In summary, the set Θ of parameters for this model is given by: (1) the path loss measured at a close distance to the transmitter P L0; (2) d0 the reference distance close to the transmitter for P L0; (3) the path loss exponent α, which varies significantly for different environments; (4) the standard deviation σ of a normal and zero mean random variable X, which represents shadowing fluctuations; (5) the gamma shape of the Nakagami model m, which represents the intensity of the small scale fading effects.

Our lognormal-Nakagami model is therefore parameterized by:

7.4.2. Model tuning based on real-world data

The specific combination of parameters Θ significantly affects the resulting reception profile and therefore the CV application performance indicators. The related literature provides many examples of parameter settings for different environments; however, the resulting settings do not always match real-world data. Furthermore, there is a wide variability in the choices for these parameters and also in the specifics of the models. Section 7.2 of this chapter presented a selection of models, scenarios and parameter settings used in the literature. Our approach has been to optimize the model parameters in order to fit real-world measurements (see section 7.5) for the scenarios outlined. By varying the set of parameters, we are able to simulate all these scenarios using one formal model.

Now, we turn our attention to the criterion for tuning the parameters of the lognormal-Nakagami model based on real measurements. The paramount concern in connected vehicle applications is the guarantee for message delivery. A successful application implementation guarantees the communication of sufficient information for closing the vehicle control loop (with the human in or out of the loop) and affecting how vehicles drive. Typical measures used to capture the reliability of DSRC wireless channel include average packet delay (APD), consecutive packet drops (CPD) and packet delivery ratio (PDR) [BAI 06]. PDR is of interest here and is denoted as pr(x), being defined as the probability of successfully receiving a packet at a receiver located at a distance x from the sender after broadcast. We will use PDR curves estimated from real-world data for the scenarios of interest over various transmitter–receiver distances x. Our optimization criterion minimizes the integral of the absolute difference between the PDR of the lognormal-Nakagami model M, , and the PDR measured from data as functions of distance:

where R is the transmission range.

We perform global optimization of the parameter set Θ using, for example, simulated annealing (SA), a simple randomized technique for iterative improvement [KIR 83]. SA repeatedly traverses a Markov chain of iteratively improved suboptimal parameter settings by sampling the search space of acceptable settings.

7.5. Channel model tuning

7.5.1. Michigan safety pilot model deployment data

CONVAS channel models were tuned based on data collected during the Michigan Safety Pilot Model Deployment (SPMD) [MIC 12], a research initiative that featured real-world implementation of CV safety technologies, applications and systems using everyday drivers. The SPMD data was collected at a test site under real conditions with multi-modal traffic around Ann Arbor, Michigan, from approximately 3000 vehicles equipped with V2V communication devices in an area that spans more than 4000 square miles from Medina in the Southwest corner to Auburn in the Northeast corner.

In the field test, Basic Safety Messages (BSMs) were transmitted at a frequency of 10 Hz. Around 75 percent of the vehicles had transmit-only capabilities, while the rest could transmit, receive and log information at a 10 Hz rate. Vehicles with logging capabilities record two different databases. In short, these had the following information that was used in our analysis (while other available details were not considered in this study): (1) DAS1-DataRV recorded entries for every received BSM, its vehicle ID, trip ID, time of reception and position coordinates of the transmitter; (2) DAS1-DataWSU recorded the vehicle ID, trip ID, time and position coordinates at a 10 Hz rate.

We selected data corresponding to the three different scenarios: urban, highway and residential/suburban. The selection was done by mapping position coordinates of the vehicles to recognizable geographic areas whose characteristics match those of the scenarios of interest (see Figure 7.1).

Figure 7.1. SPMD data collection region in Ann Arbor, Michigan (left), and zoom in for the downtown area (right). Data for the highway, residential/suburban and urban areas selected is marked in red, blue and green, respectively. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

7.5.2. Estimation of PDR

Our main goal is to obtain a reference Packet Delivery Ratio (PDR) curve over Tx-Rx distance for each of these scenarios. The data recorded in both databases allows us to compute all distances at which a successful reception occurred. For each distance, we also record the time of reception, the receiver and transmitter IDs, the trip ID and the position coordinates of the receiver.

Instances of distances at which reception failed can only be estimated. For this, we consider each interaction between any two vehicles separately. An interaction is defined by the two vehicle IDs and the trip ID of the receiver, where the maximum contiguous gap of packets not received during the interaction cannot be longer than 60 seconds. This time is equivalent to the time it takes to ride along a 200 m street block at a low urban speed average of 10–15 mph and a stop at one traffic light, or a longer residential stretch of road of 500 m at a higher average residential speed of 30–40 mph and a stop at one traffic light. For each interaction, we identify the first and the last packets received and assess lost packets based on the time gaps when no packets are received. Packet receptions are expected at every 100 ms. Whenever the time gap exceeds this value, we estimate distances for lost packets by interpolation based on the distances from the last received packet to the next received packet in periods of 100 ms within the interval of reception. For example, Figure 7.2 depicts eight examples out of the 536 interactions found in the urban region. Note that in many of these cases the two vehicles begin interacting at larger distances, only to converge and further depart away such as when two vehicles travel in opposite directions.

Figure 7.2. Examples of Tx-Rx interactions found in the urban region for the SPMD dataset, where received packets are blue and estimated lost packets are red. This allows the estimation of interaction distances for lost packets. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

As a result of applying the previously described procedure, we obtain a distribution of received packets and a lower bound of the number of lost packets for each one of the regions of interest. We consider only the distances in the range from 0 to 500 meters, since the typical transmission range for DSRC radios is R = 250...300 meters1 [QIN 04, SHO 09]. Given this and the definition of a wireless communication interaction between two vehicles given earlier, the estimation of channel models will be also considered over lengths of 250 m for urban and 500 m for residential scenarios. Figure 7.3 shows the distributions obtained for urban and residential/suburban environments with 536 and 1123 interactions, for a total number of 128,162 and 237,395 packets received respectively. The selected highway regions comprised only several interactions, and therefore do not offer sufficient statistics for further analysis of the real highway environment.

Note that the number of lost packets for distances approaching the transmission range decreases with distance. Intuitively this number should increase; however, the number of packets lost at larger distances is likely to be heavily underestimated due to the computational assumption that there exist no (lost) packets before and after the first received packet for any interaction. This typically happens for large distances rather than for short distances. The confidence in the estimates is high for short distances, where there exists more than 300,000 packets received at such distances, and decreases when the distance approaches the transmission range, where the number of packets received is less than 0.5% of the corresponding number for short distances. These statistics from the real world also suggest that the estimated PDR, will be accurate only for short distances given the amount of data in SPMD.

Finally, we obtain the PDR as a function of distance, given by:

where Nr(x) and are the number of packets received and the estimate of the number of packets lost respectively, at a distance x from the sender. represents an optimistic packet reception rate in the real world that is close to the true rate at the receiver for short distances. Figure 7.4 represents PDR for the urban and residential/suburban scenarios where a reasonable amount of logged data was available in SPMD.

7.5.3. Model tuning

We match specific real-world scenarios as defined by the Safety Pilot test bed data [MIC 12]. The optimal parameters defining each scenario were initialized based on other studies from the literature and further optimized by simulated annealing. Note that we used a fixed value of 10 meters for the reference distance d0 and we assumed that the SPMD transmission power was 23 dBm and the reception power threshold −88 dBm. Figure 7.4 shows the best solutions obtained after global optimization of the parameters as above.

Figure 7.3. Received and lost packet distributions as functions of distance for: a) urban environment; b) residential/suburban environment in the SPMD data. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

Figure 7.4. Estimated Packet Delivery Rate (PDR): (a) urban environment; (b) residential/suburban environment. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

The corresponding values of the parameters Θ presented in Table 7.2 for the scenarios of interest have been used further in the ns-3 lognormal-Nakagami propagation model for co-simulation experiments with various applications in the urban, residential/suburban or highway scenarios. When compared to the parameters commonly used in the literature in Table 7.1, we observe that our values for P L0 and σ are much higher, thus showing that, according to the SPMD dataset, propagation conditions are much worse than previously studied. Also, m > 1, therefore the shape of the gamma distribution indicates Rician fading for both scenarios, which suggest dominant line-of-sight settings. Overall, this means that in the residential/suburban scenario the signal presents a mix of propagation along a strong line of sight and non LOS between transmitter and receiver.

Table 7.2. Parameters for Lognormal-Nakagami model tuned to match urban and residential/suburban scenarios from the SPMD dataset, and [KAR 11] for highway scenarios

Environment d0 [m] P L0 [dB] α σ m
Urban 10 89.96 1.50 15.26 2.06
Residential/Suburban 10 94.97 1.08 16.05 2.14
Highway 10 63.3 1.77 3.1 -

Figure 7.4 also presents, for each of the urban and residential/suburban scenarios, the lognormal-Nakagami model learned from SPMD data, and a stochastic model given by a collection of probability mass functions, one for each distance interval, called the distribution model. The latter has been obtained from Monte Carlo simulations using the parameters of the learned lognormal-Nakagami model. The distribution model can be updated from new real world measurements and plugged into the communication simulation. It better captures the uncertainty in the real-world data used for training.

7.6. Connected vehicle applications

The quality of the wireless communications will affect the performance of large-scale connected and automated vehicle (CV/AV) applications. This can be studied in CONVAS without the effort and costs of real-world deployments. Examples of applications being modeled using CONVAS include Forward Collision Warning, Cooperative Adaptive Cruise Control and Intelligent Signalized Intersection Control. In this chapter, we describe and present results with an application called Intelligent Dilemma Zone Avoidance (IDZA) [SON 17] to demonstrate the use of the CONVAS platform for testing applications.

7.6.1. Intelligent dilemma zone avoidance

The goal of this application is to provide automated warnings to the driver or even control the vehicle automatically in order to solve the dilemma of whether to slam on the brakes and risk being rear-ended, or speed through the light and risk a collision or a traffic ticket while driving towards an intersection when the traffic light changes from green to yellow. The distance around the traffic light where this decision must be made is called dilemma zone (DZ). Certainly, the dilemma zone depends on the driving speed and the geometrical configuration of the intersection.

Technically, IDZA aims to provide automated longitudinal control to the connected and automated (equipped) vehicles to avoid getting trapped in the dilemma zone by using current vehicle state and Signal Phase and Timing (SPaT) messages from signalized intersections. The application resides within the vehicle on-board unit (OBU) and will engage the vehicle’s throttle if it is predicted to be in the DZ based on its instantaneous speed upon the reception of the SPaT message.

7.6.2. IDZA implementation in CONVAS

Details of the modeling and implementation of IDZA are presented in [SON 17]. The IDZA application is modeled within a Vissim environment and relies on a road side equipment unit located in a corner of the signalized intersection and capable of broadcasting SPaT messages. The wireless communication model simulates the message delay and reception every time step for all equipped vehicles in the network. Upon the reception of the message, vehicles determine if they are predicted to be in the DZ and dynamically adjust acceleration/deceleration rates as necessary to prevent themselves from getting trapped in the DZ, while the lead vehicle continues at its current speed. The application can take control of the vehicle’s throttle within 100 ms. The model includes a delay for switching from manual to automated driving upon the detection of potential DZ traps.

The application implements the following conditions for the activation (manual to automated) of IDZA: (1) the vehicle speed has to be higher than the minimum speed threshold; (2) the vehicle must be approaching and no more than the minimum time to stop bar threshold; (3) the vehicle has to approach the green indication of the phases designated for DZ avoidance. The conditions for deactivation (automated to manual) are: (1) acceleration required is outside the applicable range or (2) the vehicle loses the reception of SPaT messages.

Furthermore, during the auto-to-manual transition, the vehicle control uses the acceleration of the previous time step. If a vehicle has been in accelerating mode, the control reverts to the cruise mode, with zero acceleration. If a vehicle has been in decelerating mode, the control continues to decelerate at the same rate during the transition. The automatic to manual transition is canceled if the automated longitudinal control starts within the configured period. If all the conditions are met and both acceleration and deceleration options are viable, the algorithm chooses the closer edge of the dilemma zone based on its current estimate of time to stop bar at the onset of yellow.

7.6.3. IDZA performance criteria

When the vehicle is predicted to be in the DZ, IDZA uses the vehicle kinematics to derive the acceleration required to reach the first or the second edge of the DZ, and thus computes the acceleration rate needed to clear the dilemma zone trap as a function of the phase time remaining to the onset of yellow (from its SPaT message), the vehicle speed and the distance to stop bar (from the radar sensor). A vehicle can then either accelerate to reach 2.5 seconds or decelerate to reach 5.5 seconds as the time to stop bar at the onset of yellow.

Possible performance criteria are: (1) the rate of vehicles trapped in the dilemma zone; (2) the distribution of time to stop; (3) average reception loss from the moment of the first reception. These measures will be tracked in the next section.

7.7. Experimental results

7.7.1. CONVAS setup

The objective of the co-simulation experiment is to demonstrate the use of CONVAS and to evaluate the effect of wireless communications on the application performance. The CONVAS initialization sets up CSE (i.e. OPNET or ns-3), TSE (Vissim) and the application parameters. Table 7.3 presents the setup parameters for the wireless simulation (either of OPNET or ns-3).

Table 7.3. Wireless communication simulation parameters

Parameter Example Units Description
Packet Generation Window 10 ms Maximum time uniformly distributed [0, window] to wait before generating a new application layer packet
Rx/Tx Additional Delay 15 ms Processing delay added to the packet in Rx/Tx application layers
Data Rate 12e6 Mbps MAC layer data rate
Tx Power 0.199 W PHY transmit power
Rx Power Threshold –88 dBm PHY power sensitivity
Min Frequency 5855 MHz Channel’s minimum frequency
Bandwidth 10 MHz Channel bandwidth
Modulation OFDM Modulation scheme
Packet Length 40 Bytes Standard length of BSM Part I

The following parameters define the IDZA application setup:

  1. – delay in transitioning from DZ automation to manual driving is 1 second (when deactivation condition is satisfied);
  2. – minimum speed threshold is 35 mph;
  3. – minimum time to stop bar for tracking is 10 seconds;
  4. – dilemma zone time thresholds are 2.5 or 5.5 seconds.

Accordingly, the application takes control from the driver only when the following conditions are met:

  1. – the vehicle is approaching a green phase designated for DZ avoidance;
  2. – the vehicle has received the SPaT message;
  3. – the vehicle speed is greater than 35 mph;
  4. – estimated time to stop bar is less than 10 seconds;
  5. – the computed acceleration rate is within comfortable acceleration and deceleration limits of ±2m/s2;
  6. – the computed acceleration rate is less than the speed-dependent acceleration rate. The acceleration performance of the vehicle is also known to decrease with vehicle speed. We adopted a linearly decreasing acceleration profile to calculate maximum allowable speed-dependent acceleration rate using the equations reported from [LON 00];
  7. – the front gap corresponds to a time headway currently of minimum 2 seconds.

We set up an isolated signalized intersection test bed in Vissim with an 8-phase fixed time operation. The operating speed is set at 55 mph on all approaches. The IDZA is set to be active on all through phases. Figure 7.5 shows the geometry and layout of the intersection.

7.7.2. Co-simulation results

We performed experiments in scenarios with an intersection volume of 2500 vehicles per hour, 75% of the traffic volume being on the major street, and with either 25% or 75% of DSRC equipped vehicles, under various communication conditions ranging from ideal communication to several cases of transmission range, power and precision of the communication simulation, and without or with IDZA control. During a co-simulation run, we log both network statistics and the trajectory of the equipped vehicle from the moment at which it meets the criteria for activating the control until the onset of yellow. This allows the visualization of a dashboard of communications and application statistics. We present sample results next.

Figure 7.5. Signalized intersection for IDZA. Black vehicles are not DSRC equipped. Yellow vehicles have no reception and are manually driven. Red vehicles have reception and are automatically decelerated. Blue vehicles have reception but are manually driven. Light blue vehicles have reception and are controlled to accelerate. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

Figure 7.6 shows the received power scatter plot for both urban and residential/suburban scenarios. It can be appreciated how the residential/suburban received power is much more compact than in the urban case; this is the outcome of the combined effects of each of the parameters in our lognormal-Nakagami model. For instance, higher values of σ produce more spread received power.

Figure 7.7 shows the number of vehicles present in one instance of the simulation over time and the number of packets received in the network over time. Note an obvious correlation between the two plots.

Figure 7.6. Received power for the lognormal-Nakagami model and the receiver sensitivity (marked in red): a) urban environment;
b) residential/suburban environment. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

Figure 7.7. IDZA co-simulation: a) number of vehicles over simulation time; b) number of packets received over simulation time. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

At the application layer, the actual time to the stop bar at the onset of yellow represents the true benchmark of whether a vehicle is trapped in the DZ. Several data attributes are also logged during the approach including vehicle speed, approaching phase, phase indication, actual acceleration, planned acceleration, front gap and SPaT reception status. The log data from CONVAS was used in order to assess the overall performance of IDZA control, given by the trap rate in the danger zone. We experimented with a variety of conditions, representing various penetration rates of DSRC (e.g. 25–30% or 70–75%), volumes of traffic (e.g. 5000 vehicles per hour entering the intersection), communication models (described later in this section), and environments (urban, residential, etc.). The results are consistent across models and show a critical dependence of the performance of the application on reception rates for wireless communication. The results aggregating many hours of co-simulation over tens of experiments are presented in Figure 7.8.

Figure 7.8. Average Trap Rate vs. Average Reception Rate for IDZA autonomous control

Figures 7.9 and 7.10 show a selected DZ-projected vehicle’s statistics as it approaches the stop bar in one successful and one unsuccessful approach to clear the DZ. Each figure contains three plots representing the vehicle time to stop bar (top), speed (middle), and acceleration (bottom) against time to yellow on the x-axis. The traffic light changes from green to yellow when time to yellow is zero, while prior times have a negative value leading to the yellow transition. The family of curves represent the sample vehicles when using different communication models and an automated control strategy, as follows: (a) Driver Model is the mode where the vehicle has its DZ control algorithm turned off; (b, c, d) ns-3 suburban/urban/ highway represent three Lognormal-Nakagami scenarios with channel models tuned based on real measurements. If the projected Time to stop bar (top) at T ime to Y ellow = 0 is outside the [2.5, 5.5] seconds interval, the vehicle will have successfully avoided DZ.

Figure 7.9. Example of successful automated DZ avoidance under various communication models. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

For example, Figure 7.9 (middle) shows that the vehicle maintains its 65 mph speed in the driver model case to eventually get locked in the DZ, while in all the other cases automated control based on DSRC and radar inputs eventually drive the vehicles faster or slower (with the acceleration as function of time showed in the bottom plot) in order to avoid the DZ (as seen in the top plot). In particular, the vehicle running the Residential/Suburban model breaks and has at least 5.5 seconds projected time to the stop bar when the light turns from green to yellow at time zero. In the example from Figure 7.10 the vehicle control fails to clear the vehicle from the DZ in all cases because the timing of DZ prediction for control activation and the intermittent reception loss during the approach do not allow sufficient time necessary for the algorithm to control the vehicle. The variation in vehicle trajectories under various communications conditions emphasizes the importance of the effect of wireless communications on the automated control outcome.

Figure 7.10. Example of DZ avoidance failure. For a color version of this figure, see www.iste.co.uk/hilt/transportation.zip

Table 7.4 summarizes some key performance measures collected for each simulation scenario, while a full statistical analysis is undergoing. Note that the default driver model used by VISSIM to control vehicles is the Wiedermann model. In contrast, our IDZA autonomous driving model relies on wireless communication of SPaT messages from the infrastructure. The Nakagami model is a simplified version of wireless communication model where the reception probability depends only on two key parameters, distance and transmission range, for a fixed gamma shape m = 3. Nakagami 500 and 1000 represent the case of the plain Nakagami channel model with a transmission range of either 500 or 1000 ft respectively [KIL 09]. Our addition to the ns-3 model is more complex and realistic. It captures factors such as shadowing and multi-path fading of the wireless signal. In addition, it implements the full WAVE protocol stack and offers realistic information about reception time of the packets, which is important for implementing safety applications. In contrast, the Nakagami does not model the WAVE protocol stack, does not indicate reception time and only confirms reception of the packets.

Table 7.4. Summary of IDZA simulation scenarios and performance

DSRC Vehicles [%] Control Comm. Model IDZA Reception IDZA Trap Rate [%]
Rate [%]
25 – 30 Driver Model 88.2
25 – 30 IDZA Nakagami 1000ft 86.3 23.5
25 – 30 IDZA Nakagami 500ft 37.2 33.3
25 – 30 IDZA ns-3 Residential 41.0 11.0
25 – 30 IDZA ns-3 Urban 48.2 5.6
70 – 75 Driver Model 85.6
70 – 75 IDZA Nakagami 1000ft 85.6 42.5
70 – 75 IDZA Nakagami 500ft 42.5 42.5
70 – 75 IDZA ns-3 Suburban 30.1 12.2
70 – 75 IDZA ns-3 Urban 41.4 5.3

7.8. Conclusions

ITS have at their core connected vehicles talking to each other using wireless communications. Also, non-vehicle devices such as smart phones, backpacks and bicycles could incorporate the talking technology to communicate with vehicles. These technologies are converging towards real-world use under the big promise of increasing safety for people and vehicles, and improving traffic flow by aiding modern traffic management and autonomous driving. Large-scale simulations are of tremendous value for understanding the critical factors in outstanding safety and mobility applications.

In this work we guide V2V and I2V communication simulation based on received packet rates, which is the most critical factor in evaluating a CV application’s success. Our goal was to use the largest available dataset showing reception in real environments, from the Michigan test bed. We evaluated applications using propagation settings grounded in the real-world macro-level measurements, i.e. at the level of actually received packets rather than signal power, interference or other low level measures. The results offer a view surprising to the present understanding (from both theoretical and pragmatic perspective) of vehicular channel models. For example, the 50–60% reception rates at short distances in urban areas are factual, and the information then leads to the tuning of the propagation model parameters that are far different from settings seen in previous research. Similarly, our models do not explicitly include a congestion component in packet loss, but rather implicitly account for the effect of high densities of vehicles in the urban scenario. Our results are not as optimistic as other tests under real-world conditions. For example, in [AHM 14] the NHTSA CAMP partnership illustrated an effective average PER below 10% for distances where vehicles spent most of their time. This translated into a PDR of 90%, much higher than the PDRs obtained from the SPMD data. While the CAMP researchers used the same 10 Hz transmission scheme of BSMs, the experiments were conducted in a reduced set of only eight vehicles, which kept the same convoy formation at all times, and the presented PER was a result of the superposition of results obtained in all the tested scenarios: mountainous, deep urban, freeway and major/local roads.

Communication characteristics and quality will have significant impacts on the performance of connected and automated vehicle applications, particularly when implementing safety features as shown in our CV application. For instance, reception of packets affects the timeliness of context information, i.e. awareness of the presence of other vehicles, traffic light status, etc., being made available to the running application. Therefore, realistic communication simulation using models that capture stochasticity as presented in this chapter plays a paramount role in the accurate modeling of connected and automated vehicle technologies. Multiple co-simulators for VANETs can be found in the literature; however, not all of them present an active exchange of information and commands among their components as simulation time advances. CONVAS is a deeply interlinked co-simulator that offers a high level of realism to assist researchers in testing of CV applications. Realism is achieved first by integrating up-to-date communication simulation tools with accurately parameterized propagation models based on real-world measurements, and second by employing a traffic simulation tool with the ability to implement specific vehicle behaviors in order to enable modeling of varied applications.

At the physical level, knowledge has advanced to better understand interference, fading and congestion. However, lack of extensive real-world measurements and usage of various hardware and software settings for the existing measurements lead to inconclusive data sets. The divergence of results from various real-world based studies as shown here highlights the need for rigorous validation of the data and measurements acquired in the community. This research can be extended to consider actuated or adaptive signal control, where the dilemma zone avoidance may be in tighter correlation with safety. An additional direction of future work is the validation of a large number of CV applications in CONVAS, while channel models can be learned and updated as new real-world measurements become available. It will be desirable to model congestion of the wireless medium as bandwidth utilization increases for large penetrations. Awareness of channel utilization can be arguably taken advantage of across layers; however, we do not presently have measurements for high utilization factors. Finally, it is of interest to optimize the CONVAS simulation performance for large-scale traffic scenarios.

7.9. Acknowledgments

This work was funded by the Exploratory Advanced Research Program (EARP) of the Federal Highway Administration (FHWA) – Grant Number DTFH6114C00003. We would like to thank Juan Aparicio for early work on propagation models and discussions in this area, and to Apoorba Bibek for co-simulation testing.

7.10. Bibliography

[AHM 14] AHMED-ZAID F., KRISHNAN H., VLADIMEROU V. et al., Vehicle-to-vehicle safety system and vehicle build for safety pilot (V2V-SP), Final Report Volume 2 of 2, Performance Testing – DRAFT, Crash Avoidance Metrics Partnership, NHTSA, 2014.

[BAG 12] BAGUENA M., CALAFATE C.T., CANO J. et al., “Towards realistic vehicular network simulation models”, Wireless Days (WD), pp. 1–3, 2012.

[BAI 06] BAI F., KRISHNAN H., “Reliability analysis of DSRC wireless communication for vehicle safety applications”, IEEE Intelligent Transportation Systems Conference, pp. 355–362, 2006.

[BLU 04] BLUM J., ESKANDARIAN A., HOFFMAN L., “Challenges of intervehicle ad hoc networks”, IEEE Transactions on Intelligent Transportation Systems, vol. 5, no. 4, pp. 347–351, 2004.

[BON 02] BONNESON J., MIDDLETON D., ZIMMERMAN K. et al., Intelligent detectioncontrol system for rural signalized intersections, Technical Report FHWA/TX-03/4022-2, Texas Transportation Institute, August 2002.

[BON 08] BONONI L., DI FELICE M., D’ANGELO G. et al., “MoVES: a framework for parallel and distributed simulation of wireless vehicular ad hoc networks”, Computer Networks, vol. 52, no. 1, pp. 155–179, 2008.

[CHA 15] CHACHICH A., FESSMANN V., ARNOLD J. et al., “DSRC-Unlicensed Device Test plan, USDOT Intelligent Transportation Systems – Joint Program Office”, available at: http://www.its.dot.gov/connected_vehicle/pdf/DSRC_TestPlanv3.5.3.pdf, 2015.

[CHE 07] CHENG L., HENTY B.E., STANCIL D.D. et al., “Mobile vehicle-to-vehicle narrow-band channel measurement and characterization of the 5.9 GHz dedicated short range communication (DSRC) frequency band”, IEEE Journal on Selected Areas in Communications, vol. 25, no. 8, pp. 1501–1516, 2007.

[DRE 14] DRESSLER F., HARTENSTEIN H., ALTINTAS O. et al., “Inter-vehicle communication: Quo vadis”, IEEE Communications Magazine, vol. 52, no. 6, pp. 170–177, 2014.

[EEN 09] EENENNAAM E.M.V., A Survey of Propagation Models Used in Vehicular Ad Hoc Network (VANET) Research, Faculty of EEMCS, University of Twente, 2009.

[HAF 13] HAFEEZ K.A., ZHAO L., MA B. et al., “Performance analysis and enhancement of the DSRC for VANET’s safety applications”, IEEE Transactions on Vehicular Technology, vol. 62, no. 7, pp. 3069–3083, 2013.

[HWU 88] HWUANG C.R., “Simulated annealing: theory and applications”, Acta Applicandae Mathematicae, vol. 12, no. 1, pp. 108–111, 1988.

[ISL 13] ISLAM T., HU Y., ONUR E. et al., “Realistic simulation of IEEE 802.11 p channel in mobile vehicle to vehicle communication”, Microwave Techniques (COMITE) Conference, pp. 156–161, 2013.

[KAI 11] KAISSER F., GRANSART C., KASSAB M. et al., A Framework to Simulate VANET Scenarios with SUMO, University Lille Nord de France, 2011.

[KAR 07] KARNADI F.K., MO Z.H., LAN K.C., “Rapid generation of realistic mobility models for VANET”, Wireless Communications and Networking Conference, pp. 2506–2511, 2007.

[KAR 11] KAREDAL J., CZINK N., PAIER A. et al., “Path loss modeling for vehicle-to-vehicle communications”, Vehicular Technology, IEEE Transactions, pp. 323–328, 2011.

[KEN 11] KENNEY J.B., “Dedicated Short-Range Communications (DSRC) standards in the United States”,Proceedings of the IEEE, vol. 99, no. 7, pp. 1162–1182, 2011.

[KIL 09] KILLAT M., HARTENSTEIN H., “An empirical model for probability of packet reception in vehicular ad hoc networks”, EURASIP Journal on Wireless Communications and Networking, vol. 2009, no. 1, p. 721301, 2009.

[KIR 83] KIRKPATRICK S., GELATT C.D., VECCHI M.P., “Optimization by simulated annealing”, Science, vol. 220, no. 4598, pp. 671–680, 1983.

[LEE 15] LEE J., PARK B.B., “Investigating communications performance for automated vehicle-based intersection control under connected vehicle environment”, IEEE Intelligent Vehicles Symposium (IV), Seoul, 2015.

[LOC 05] LOCHERT C., BARTHELS A., CERVANTES A. et al., “Multiple simulator interlinking environment for IVC”, 2nd ACM International Workshop on Vehicular Ad Hoc Networks (VANET 2005), Cologne, pp. 87–88, 2005.

[LON 00] LONG G., “Acceleration characteristics of starting vehicles”, Transportation Research Record, vol. 1737, pp. 58–70, 2000.

[LUK 12] LUKUC M., V2V Interoperability Project, US DOT ITS Connected Vehicle Workshop, Chicago, September 2012.

[MEC 11] MECKLENBRAUKER C.F., MOLISCH A.F., KAREDAL J. et al., “Vehicular channel characterization and its implications for wireless system design and performance”, Proceedings of the IEEE, vol. 99, pp. 1189–1212, 2011.

[MIC 12] MICHIGAN SAFETY PILOT MODEL DEPLOYMENT, available at: https://www.its-rde.net/data/showds?dataEnvironmentNumber=10018, 2012.

[NS 15] NS-3 MODEL LIBRARY, Release ns-3.24 (September 2015), available at: https://www.nsnam.org/docs/models/ns-3-model-library.pdf, 2015.

[NS 16] NS-3 NETWORK SIMULATOR, available at: https://www.nsnam.org/, accessed on 25 February 2016.

[PIO 08] PIORKOWSKI M., RAYA M., LUGO A. et al., “TraNS: realistic joint traffic and network simulator for VANETs”, ACM SIGMOBILE Mobile Computing and Communications Review, vol. 12, no. 1, pp. 31–33, 2008.

[QIN 04] QING X., MAK T., KO J. et al., “Vehicle-to-vehicle safety messaging in DSRC”, Proceedings of the 1st ACM International Workshop on Vehicular Ad Hoc Networks, pp. 19–28, 2004.

[RIV 16] RIVERBED (OPNET) MODELER, available at: http://www.riverbed.com/products/steelcentral/steelcentral-riverbed-modeler.html, 2016.

[ROI 14] ROIVAINEN A., JAYASINGHE P., MEINILA J. et al., “Vehicle-to-vehicle radio channel characterization in urban environment at 2.3 GHz and 5.25 GHz”, IEEE 25th Annual International Symposium on Personal, Indoor, and Mobile Radio Communication (PIMRC), pp. 63–67, 2014.

[RON 13] RONDINONE M., MANEROS J., KRAJZEWICZ D. et al., “iTETRIS: a modular simulation platform for the large scale evaluation of cooperative ITS applications”, Simulation Modelling Practice and Theory, vol. 34, 2013.

[SHO 09] SHOREY R., WEIMERSKIRCH A., JIANG D. et al., “Characterization of DSRC performance as a function of transmit power”, Proceedings of the Sixth International Workshop on Vehicular Ad Hoc Networks (VANET), Beijing, ACM, pp. 63–68, 2009.

[SOM 08a] SOMMER C., YAO Z., GERMAN R. et al., “Simulating the influence of IVC on road traffic using bidirectionally coupled simulators”, IEEE INFOCOM Workshops 2008, Phoenix, pp. 1–6, 2008.

[SOM 08b] SOMMER C., YAO Z., GERMAN R. et al., “On the need for bidirectional coupling of road traffic microsimulation and network simulation”, Proceedings of 9th ACM International Symposium on Mobile Ad Hoc Networking and Computing (Mobihoc 2008): 1st ACM International Workshop on Mobility Models for Networking Research, pp. 41–48, 2008.

[SOM 11a] SOMMER C., DRESSLER F., “Using the right two-ray model? A measurement based evaluation of PHY models in VANETs”, Proceedings of ACM MobiCom, pp. 1–3, 2011.

[SOM 11b] SOMMER C., GERMAN R., DRESSLER F., “Bidirectionally coupled network and road traffic simulation for improved IVC analysis”, IEEE Transactions on Mobile Computing, vol. 10, no. 1, pp. 3–15, 2011.

[SOM 11c] SOMMER C., ECKHOFF D., GERMAN R. et al., “A computationally inexpensive empirical model of IEEE 802.11 p radio shadowing in urban environments”, Wireless On-Demand Network Systems and Services (WONS), Eighth International Conference, pp. 84–90, 2011.

[SON 17] SONGCHITRUKSA P., SUNKARI S., UGALDE I. et al., “Interlinking Vissim and ns-for Connected-Vehicle Simulation: Case Study of Intelligent Dilemma Zone Avoidance” Journal of the Transportation Research Board, (in press) 2017.

[STU 10] STUBING H., BECHLER M., HEUSSNER D. et al., “simTD: A Car-to-X system architecture for field operational tests”, IEEE Communications Magazine, vol. 48, no. 5, pp. 148–154, 2010.

[WAN 09] WANG S.Y., CHOU C.L., “NCTUns tool for wireless vehicular communication network researches”, Simulation Modelling Practice and Theory, vol. 17, no. 7, pp. 1211–1226, 2009.

[WIN 07] WINNER consortium, D1.1.2, WINNER II channel models, WINNER European Research project Public Deliverable, 2007.

Chapter written by Justinian ROSCA, Ines UGALDE, Praprut SONGCHITRUKSA and Srinivasa SUNKARI.