Modern vehicles are making inroads in the market. These vehicles are not only equipped with global positioning system (GPS) and navigation systems, but also more advanced features such as environmental awareness to prevent vehicle collisions, multimedia systems, and integrated wireless access systems to improve vehicle performance and user experience. In addition, there is much interest in improving the efficiency of vehicular communications. For this purpose, ITS aim at improving safety, reliability, efficiency, and quality of transport infrastructure and vehicles through the use of information and communication technologies (ICT). Additionally, ITS focus on providing sustainable and affordable transportation by designing advanced applications and services to optimize transportation times and energy consumption.

ITS support different communication scenarios including all types of communications in vehicles, between vehicles, as well as between vehicles and roadside infrastructure. Figure 2 shows the taxonomy of vehicular communications. Vehicular communications can be classified into V2V and V2I. V2I further includes vehicle-to-roadside (V2R) communication and communication using cellular networks. In V2V communications, a vehicular ad hoc network (VANET) is formed among vehicles for exchanging information e.g., safety information. In V2R, information is exchanged between the roadside unit (RSU) and the onboard unit (OBU) of a vehicle. In V2I, information is exchanged between the RSU, or possibly a cellular network, and OBU.

V2V communications are relatively more complex and challenging as compared to V2R and V2I communications. The V2I communications are more to exchange information with the centralized servers somewhere in the Internet through various access networks and technologies. Currently, there is continuing effort by cellular network operators to enhance the network capacity that will improve the cellular network support for V2I communications. The V2R communications that are presently deployed are mainly short-distance one-to-one communications, for example, toll collection, whereas the V2V communications are expected to be mostly ad hoc communications with transmission range up to several 100 m. However, currently, the V2V communications are not yet ad hoc-based. There are mainly one-hop communications for unicast and multicast. Standardization bodies and researchers are working on multihop for geo-casting and V2V safety applications, but there are still strong scalability concerns. The transmission range commonly ensured is 300 m in urban area and about 1 km in free space.

Here we describe the unique characteristics of VANETs and identify some major issues. Deploying a vehicular networking system requires addressing several challenges posed by the unique characteristics and requirements of vehicular communications 43 .

With the increasing demand for spectral efficiency, cognitive radio has emerged as a very active research area for wireless communications and networks’ research community in recent years. CR 26 is an emerging technology that enhances the performance of existing radio by integrating artificial intelligence (AI) with software-defined radios (SDRs). Unlike conventional radios in which most of the components are implemented in the hardware, SDRs 24 are radios that use software implementations for some functionalities enabling flexible radio operation. These radios are reconfigurable, and hence the need to modify existing hardware isreduced.

Though several definitions have been provided to describe CR, the commonly used definitions in literature are given by Joseph Mitola and Simon Haykin. Joseph Mitola described CRs 22 23 24 as intelligent radios that can autonomously make decisions using gathered information about the radio frequency (RF) environment through model-based reasoning and can also learn and plan according to their past experience. Afterwards, Simon Haykin defined CR as an intelligent wireless communication system capable of being aware of its environment, learning, and adaptively changing its operating parameters in real time for providing reliable communication and efficient utilization of the radio spectrum 25 .

As mentioned previously, a cognitive radio can be built on software-defined radio (SDR) that makes it reconfigurable by converging the two key technologies: software-defined radio and artificial intelligence 52 53 54 55 56 57 . While the field of CR is independent from SDR, CR has its origin in SDR as an enabling technology to support its implementation.

In the context of VANETs, with increasing number of vehicles equipped with wireless communication systems, vehicular communication will require faster connectivity and thus, wireless spectrum will have to be adapted to the new requirements (of bandwidth). Using CR and SDR will prevent the need to implement hardware upgrades with emergence of new protocols in future. SDR enables ‘reconfigurable systems’ for wireless networks. CR allows SDR to determine which mode of operation and parameters to use. In general, CR and SDR will allow CR-enabled vehicles to search for the best frequency-based predetermined parameters.

QoS support is important for applications such as related to safety. From the point of view of radio technology, satisfying QoS guarantees is easier when there is enough bandwidth, which can be traded with QoS guarantees such as low delay or high reliability. Additionally, there should be mechanisms to protect important flows from lower priority flows. We discuss the issues related to bandwidth and flow priority in the following text.

The flexibility and agility offered by cognitive radio is very useful for resilient communications in vehicular scenarios. In case of emergency situations, CR can be reconfigured in real time to operate in emergency mode by focusing on minimizing bit error rate (BER) and avoiding interference 59 . This property is very useful for ITS safety applications, and with CR functionalities, some objectives like bandwidth maximization or power minimization can be flexibly traded for more resiliency.

In many cases, highways are open spaces and there is a high chance of finding a spectrum hole that can be accessed opportunistically. This is unlike downtown and urban areas where chances of finding spectrum holes can be low due to high population. In fact, some experiments have been done 29 which show free abundant spectrum available for opportunistic use on highways. Thus, cognitive radio technology is very attractive as it can exploit such spectrum availability using opportunistic spectrum access. This, in turn, can answer some of the bandwidth and congestion problems discussed above.

Some of the advanced cognitive radio capabilities come at a price in terms of bigger size of onboard units. Some functionalities can also consume energy. However, vehicles have sufficient space and power supply and are not limited by them unlike the case with smart phones and other highly portable devices. Cost can still be a factor, but performance vs. cost trade-off can be exploited. Cost can also be reduced with the help of mass production, by optimally designing an OBU with cognitive radio capabilities.

Every 2 to 3 years, a new communication standard is being proposed, such as DVB-H, DVB-T2, WiMaX, 802.11p, LTE, and HSDPA. Some of these standards face the risk of becoming a commercial failure such as was the case with DVB-H. Moreover, different countries have different regulations related to spectrum usages, transmit power limits, etc. Thus, global manufacturers of vehicles face a dilemma on which technology to deploy and how to deploy different versions of communication units for different countries. A vehicle has a life cycle of more than 15 years, and onboard communicating devices should not become obsolete during that period. Moreover, the drivers as well as passengers would like to use the newer technologies that come out in the future.

Cognitive radio combined with SDR offers a solution to cope with evolving and numerous technologies. Reconfigurable and reprogrammable capabilities of CR and SDR provide the possibilities to design ‘future proof’ onboard units that are upgradeable with software updates. This allows flexibility in deploying different versions of units for different countries and allows upgradeability when new technologies come out in future.

An OBU can integrate an SDR transceiver with navigational devices (GPS) and onboard computer to provide capabilities of context awareness and adaptation.

Deploying base stations to provide wireless services (e.g., Internet access) in vehicular networks is challenging due to the highly mobile environment. VANETs are considered as a special case of mobile ad hoc networks (MANETs) due to their specific characteristics such as special mobility pattern, varying vehicle density, and interference with other types of networks.

Several applications of VANETs have been proposed that take into account the above-mentioned constraints of vehicular communication. As described before, different communication technologies have been standardized (WAVE, IEEE 802.11p, etc.) for vehicular communications. However, presently, the deployment and performance evaluation of the proposed standards, especially in a large-scale vehicular environment, are works in progress. Moreover, different standards and protocols proposed for vehicular communication applications should be interoperable. Additionally, short-range communication protocols may not be able to provide good Internet connectivity for high-speed vehicles and that would require developing new efficient protocols suitable for them.

The network architecture of CR-VANETs consists of vehicles with onboard units (OBUs) and infrastructure units such as RSUs and base stations of network providers. Note that setting up of roadside units everywhere may be costly; thus, V2V communication, including multihop, should be able to work without any infrastructure support.

Several functionalities are needed for CR-VANETs paradigm, and an initial architecture related to different functionality blocks and protocols is proposed in 65 . The architecture proposed by its authors consists of several blocks related to policies such as those imposed by FCC and local authorities and onboard sensors, including radio sensors and GPS. It also includes a knowledge database that consists of previously learned cases, rules, spectrum information, and radio maps. Further, it consists of a cognitive engine that is the core of all functionalities. Additionally, there are software-defined radio or multiple radios, performance measurement block for feedback, and finally a user interface to allow users to configure the system offline. Figure 7 illustrates the architecture showing these different functional blocks for CR-VANETs.

As illustrated in the figure, the cognitive engine (CE) uses real-time data from onboard sensors and history and map information from knowledge database to learn from the present as well as the past. It then controls the operation of software-defined radio through optimization of transmission parameters and decision making related to spectrum use and management.

A taxonomy of machine learning approaches for CE is provided in Figure 8. The classification is done in terms of supervised learning, decision making and parameter tuning. Supervised learning can be used for prediction or classification, by learning the mapping between the given inputs and outputs. It is called supervised learning because the outputs for the given inputs are known beforehand. For example, in the case of signal identification, the target signal and the input signal features are known during the learning phase. Decision making is related to the problem of making optimal decisions. Some available tools for decision making use reinforcement learning, which can be used in unknown environments, or case-based reasoning based on history, etc. Parameter tuning is needed to set the parameters optimally. A system model is used, and some metaheuristics can be used to find the optimal values of the system parameters for a given optimization objective. Multiple conflicting optimization objectives, such as maximization of bandwidth vs. minimization of bit error rate, can be assigned with weights differentially according to the desired mode of communication, ranging from bandwidth maximization mode to emergency mode for high resiliency 59 , etc.

Several machine learning techniques have been studied in the literature to realize cognitive functionalities in the general context of cognitive radio. However, they need to be adapted for CR-VANETs scenario. Some interesting ideas for applying machine learning to CR-VANETs are discussed in 61 . We summarize those ideas in Table 2 and discuss them in the following text.

The authors of 61 argue that case-based reasoning is interesting for CR-VANETs scenario because vehicles have a high chance of passing through the same locations at approximately the same time of the day. This is especially the case for example, for public transports, such as buses and trains but also for cars considering that people commute everyday from their homes to offices, at approximately the same time. With case-based reasoning, the optimization process can be populated from the previously found optimal configurations, and then further search for optimization can start from there. This can highly improve the convergence time of optimization algorithms such as genetic algorithms and simulated annealing, which in turn can be used for optimizing transmission parameters.

Machine learning tools for classification such as support vector machines and neural networks are interesting for signal identification and discovery of spectrum holes. With cognitive radio, it is required that the opportunistic use of spectrum should not collide with primary user traffic. Thus, classification tools can be used to enhance the detection of primary user signals and can provide improved accuracy as compared to just using energy detection. This improved primary user detection accuracy is desired especially in VANETs scenarios, where high-mobility conditions leading to signal fading can cause difficulties in spectrum hole detection. Neural networks can also be used as function approximators to predict throughput at a given location on a particular time of the day. Different inputs such as PU and SU traffic statistics can be mapped to predict throughput through trained neural networks 61 . Some online learning tools like reinforcement learning can be used. Such tools are useful in unknown environments and are useful for channel selection and decision making.

Spectrum sensing is required by cognitive radio to detect the presence of spectrum holes and more importantly to detect the presence of PUs. The priority is to avoid interference with PUs and thus, spectrum sensing should be highly reliable.

High mobility in VANETs can deteriorate the performance of spectrum sensing due to signal fading and fast changing location. In the case of fast varying channels over time, some works focus on improving the performance of sensing 66 and signal identification 67 68 69 . Similarly improved algorithms have been proposed for channel estimation and equalization 70 in high-speed environments.

Additionally, cooperative spectrum sensing 71 can be used to tackle the problems due to high mobility. Several spectrum sensing vehicles combined with a regularly updated central spectrum database can improve the performance of spectrum sensing in VANETs scenarios. The work in 72 investigates the effect of mobility on the performance of spectrum sensing in CR networks. The study confirms that cooperation can improve the spectrum sensing performance as a result of increased spatial-temporal diversity in received primary signal strengths.

This forms the basis of their proposed sensing technique that finds an optimal combination of the number of sensors for cooperation and the number of times to sense the spectrum in order to reduce the sensing overhead. This work does not specifically target CR-VANETs scenario, but it is still interesting for that. Additional works such as 31 34 73 aim to achieve better spectrum sensing precision and efficiency through a coordinated spectrum sensing. The data sensed by the vehicles is sent to RSUs that forward the gathered data further to a central processing unit. The delay and reliability of spectrum sensing in vehicular environments is discussed in 74 . They argue that cooperative spectrum sensing can be used to tackle the sensing limitations of a single vehicle as it exploits spatial diversity. They show that sensing time can be reduced significantly with cooperative sensing as compared to sensing by only a single vehicle.

The authors of 75 point out that cooperative sensing can help a vehicle to detect the spectrum availability in future locations by exploiting sensing information from other cooperating vehicles. An experimental study on sensing performance in VANETs scenario is conducted in 30 . They found that sensing accuracy depends on vehicular speed as well as location. For example, in the areas with open space, the fast fading is not high and hence, sensing performance is good unlike at places such as downtown. They also propose a framework for cooperative spectrum sensing and sharing called Cognitive V2V (Cog-V2V). In Cog-V2V, each vehicle keeps a spectrum availability database that is regularly updated. Note that using such knowledge database, shared by different vehicles or kept distributed among them, is an interesting way of collaborating for spectrum sensing. The data from different cooperating vehicles is merged using weights that depend on distance between the locations at which sensing samples were collected. Intuitively, they calculate weights such that weights from nearby vehicles are weighed lower as they are likely to have similar radio conditions due to their similar location. After the spectrum opportunities are detected, Cog-V2V decides the channel to be used between the two vehicles.

Another work 76 uses spectrum database as well as support from infrastructure. They also provide guidelines for the uniform placement of RSUs that will provide information about the availability of TV channels and thereby support the mobile CR-enabled vehicles. Systems using centralized approaches and RSUs can be effective, specifically for safety applications as it improves their reliability by improving data delivery in vehicular networks.

However, the centralized approach is not apt to support multihop communication, especially in the absence of infrastructure support. Nonetheless, in the absence of central infrastructure, clustering techniques can be used to create cluster leaders among several vehicles. Cluster leaders can then assume the role of central entities responsible for coordinating spectrum sensing and data fusion as used in 35 .

In 77 , a different type of sensing is used that focuses on estimating the amount of congestion in a given channel. The authors address the problem of insufficient spectrum for reliable exchange of safety information over congested urban scenarios through the use of cooperative spectrum sensing provided by vehicles. The spectrum sensing provided by vehicles is used to provide a distributed measure of contention on the roads and locate white channels in the industrial, scientific, and medical (ISM) bands. If contention is detected, free channels from the ISM spectrum are allocated to the vehicles. Note that some of these works are also discussed in the next section in the context of spectrum management.

One of the main focuses of research in CR-VANETs is to design specific spectrum management techniques which take into account spectrum sensing and access as well as mobility for CR-enabled vehicles. Spectrum management should also aim at providing QoS support in CR-VANETs scenario. QoS guarantees are important for messages and signals related to vehicular safety. QoS guarantees can be in terms of maximum permitted delay and reliability in terms of permitted packet loss, etc. For example, in the event of an emergency brake, the vehicles that follow should be notified as quickly as possible in order to leave some time for other drivers to react.

However, providing such QoS guarantees can be difficult in case of congestion or when there are several vehicles. Moreover, it is noteworthy that the feasibility of using TV white space for vehicular communications is restricted due to the scarcity of TV white space in large urban cities, where the average number of vehicles is also higher as compared to the remote areas. Such factors can make it difficult to ensure QoS support for some important applications.

Taking the example of safety messages, authors in 77 argue that the bandwidth of a single DSRC control channel may not be enough for reliable delivery of safety messages during congestion. They propose to use cognitive radio to augment the spectrum of the control channel. They proposed to use TV white space (TVWS) spectrum to enhance the bandwidth of the control channel 15 and in a follow-up work 77 , 5.8 GHz ISM band is used instead of TVWS, as 5.8 GHz ISM band is the neighboring band to the DSRC 5.9 GHz band and it can simplify the transceiver design. They argue that Wi-Fi users, who use 5.8 GHz ISM band, are unlikely to be present on highways.

In order to counter congestion problems, authors in 77 first propose a contention metric C _r (t) to estimate the amount of congestion in different regions r, at time t:

C r ( t ) = α D r × B r S r + β U r .

where the delay D _r is the delay experienced by the transmitted packet and it increases during congestion. The parameter B _r is the channel’s bitrate, S _r is the average packet size, and U _r is the number of untransmitted packets due to congestion. The parameters α and β are tuning parameters, which are estimated from curve fitting of simulation data. Different scenarios were simulated to obtain such data by varying the number of vehicles, payload sizes, and bitrates.

Different vehicles perform sensing and report the measurements to their nearest RSU. RSUs communicate with other RSUs and vehicles to form an estimate of congestion in different zones. Based on congestion estimates, a fuzzy-logic-based controller is used to decide the amount of spectrum needed, and RSUs allocate more spectrum to the vehicles traveling in a zone with increased congestion.

The idea is that the spectrum is a scarce resource and additional spectrum is allocated only when needed. Additionally, the proposed contention metric looks interesting and it will be interesting to see how it can be used in other 802.11-like scenarios. Another interesting research direction is to explore improved contention metrics.

Several interesting ideas related to dynamic spectrum access in CR-VANETs are explored in 44 . They call the dynamic spectrum allocation problem for VANETs as vehicular dynamic spectrum access (VDSA). Several VDSA approaches covering spectrum measurements and machine learning are proposed; for example, reinforcement learning is used for intelligent channel selection (also presented in 65 ). Moreover, a testbed implementation with some experimental results is also provided. A feasibility analysis is done for VDSA using queuing theory. Different TVWS bands are considered as servers which are either occupied by PUs or not. Taking the case of V2I communication, the requests to access those channels from SU vehicles are modeled as clients arriving in a queue and these requests stay in the queue until they are served. Thus, with this formulation, the queuing theory results can be used to analyze VDSA scenario. However, it is pointed out in 78 that such model can be over optimistic because vehicles can move out of RSU range. Thus, vehicular dynamics need to be taken into account and an analysis considering such dynamics is proposed in 78 .

Additional works such as in 30 31 propose to first detect the available spectrum resources, then according to the QoS requirements of the vehicular applications, spectrum decision algorithms select the channel to be used at a specific location. Finally, note that even if vehicles have sufficient power supply, the energy consumption is still an important factor for environmental considerations 79 . An energy-efficient technique is proposed in 79 by formulating the problem of deciding whether to use direct transmission or relay-based transmission as an optimization problem, with the objective of minimizing energy subject to QoS constraints in terms of delay.

Most of the above approaches assume a centralized architecture where decisions can be taken in a central manner. However, this may not be always possible in the case of CR-VANETs. Thus, in 35 the authors focus on a clustering strategy as illustrated in Figure 9 where vehicles are grouped into clusters and one of the vehicles act as a cluster leader making central decisions for the cluster. It can be seen as a hybrid of distributed and centralized approaches. The authors study the problem of optimal channel access to provide QoS for data transmission in CR-VANETs. The data from cluster members is first transmitted to the cluster leader and then it is forwarded to its destination.

The problem is to opportunistically use available channels using CR, optimally reserve some bandwidth in DSRC channels, and control the size of the vehicular cluster. The authors use constrained Markov decision process (MDP) for decision making. The vehicles use two radios: one for opportunistic spectrum access and the other for DSRC channel. Vehicles send a request to join a given cluster using DSRC channel. The cluster head takes the decision of whether the new node should be added to the cluster or not. The cluster head also optimally reserves some bandwidth in the DSRC channel. For opportunistic spectrum access, the cluster head collects the sensing results from cluster members and then the scheduling decision is broadcasted to the cluster. The authors point out that there are two different time scales involved in decision making: opportunistic channel access (fast) and cluster size control with DSRC bandwidth reservation (relatively slow). Thus, a hierarchical MDP with two levels with each level corresponding to each problem is used. These two levels interact with each other using the following parameters: QoS performance measures, location, cluster size, and available bandwidth in DSRC channel. The optimization problem is solved using linear programming, and different statistics and variables are assumed to be available in advance. Thus, one interesting extension of this work could be to explore online learning.

Some distributed spectrum management schemes exist for ad hoc CR networks that focus on MAC design 63 as well as QoS 64 . Such proposals may be customized for CR-VANETs. The distributed opportunistic multiChannel MAC (OMC-MAC) proposal in 64 assumes a fixed CCC, transmits beacons at regular intervals for synchronization, and divides the beacon interval into three phases: channel sensing, contention, and data transmission. SUs sense the channel in the first phase and then using CCC contend for reservations of time slots belonging to data transmission phase. Data transmission phase is then used for contention-free transmission of data based on the reservations on different available channels. QoS provisioning is done by first letting only the prioritized users contend for the reservations and if some time slots are left, then other users can also make some reservations. This can be useful for prioritizing safety messages in CR-VANETs. However, the impact of vehicular mobility and short-lived links needs to be studied for such schemes. The common control channel may also get congested when there are many vehicles. The discovery process of available channels, during the sensing phase, can be improved by using a common spectrum database, as explained in Section 4.2. Adaptive bitrate as well as beacon interval can be explored to further improve the performance.

In this section, we discuss the works on CR-VANETs that are related to upper layers. We mainly focus on network layer but also talk about P2P approaches that work on top of an overlay network.

Cognitive Radio for Railway Through Dynamic and Opportunistic Spectrum Reuse (CORRIDOR) 92 is a French research project that targets opportunistic spectrum access for railways. Communication demands are increasing for modern railways from the point of view of railway operations as well as for providing Internet connectivity to the passengers. However, there is no single universal wireless technology that can answer the needs of heterogeneous services and railway applications. Modern railway applications and passenger’s Internet connectivity demands call for more bandwidth and spectrum.

One answer to such needs is cognitive radio. Thus, CORRIDOR plans to use CR technology to support multiple railway applications such as command and control, CCTV, maintenance, and Internet connectivity for passengers. The project aims to develop algorithms and technology for very high speed environments that are typical of railways. The focus will be on spectrum access and management, exploiting TVWS, cognitive engine, cross-layer optimization, mobility management, and handover optimization. Different techniques will be validated with some real on-site trials.

The PROTON-PLATA project 93 is a European project that aims at developing a reconfigurable prototype based on emerging software-defined technologies for telematics applications for cars-to-roadside and car-to-car communications. The project proposes a multitechnology cooperative Advanced Driver Assistance System (ADAS) that is based on the integration of SDR devices in vehicles.

The first part of the project deals with developing a prototype including SDR device and driver assistance applications to demonstrate the utility of a cooperative ADAS for improving communication performance. This involves designing a multitechnology communication infrastructure and equipping onboard units with SDR devices that support different communication scenarios i.e., V2V communications, V2I communications, and broadcasting (infrastructure-to-vehicle) at the same time. The performance evaluation using network simulation tools forms the second part of the project.

The Rail-CR project 94 is a US project that is a part of the efforts toward implementing the positive train control (PTC) technology to equip trains with wireless communication capabilities. This will allow trains to communicate with wayside wireless stations, while moving, and provide useful information related to their location, speed, direction, etc. The project adapts Virginia Tech cognitive radio technology 95 to improve safety and operations of the railways.

The complex radio characteristics faced during train transportation such as continuously changing situation of the train as well as continuously changing communications environment due to varying noise, multiple sources of interference, multiple users competing for scarce spectrum, etc. make wireless communication with trains very demanding. This has motivated the railway industry to use SDRs equipped with a reconfigurable platform for packet-data transmission. Though SDR provides interoperability and enables multiple configurations, it lacks the capability of integrating additional adaptation to alleviate crowded spectrum, intentional jamming, or learning from past experiences.

In this context, the project developed a railroad-specific cognitive radio namely, Rail-CR, capable of fulfilling the requirements of future wireless communication systems for trains. The Rail-CR combines AI-based decision-making and learning algorithms with SDR to address the railway transportations’ specific need for adaptable communications. The Rail-CR cognitive radio will enhance the performance of railway communications in terms of interoperability, robustness, reliability, and spectral efficiency, and will make it cost-effective to deploy and maintain.

The Rail-CR system includes a CE that relies on the tunable radio parameters (knobs) and observable performance indicators (meters) of the SDR platform. Based on situational awareness from the radio in the form of observable parameters (meters), a cognitive engine uses software-based decision-making and learning algorithms to decide if the radio parameters (knobs) should be adapted depending on sets of predefined objectives. The Rail-CR was evaluated under different interference conditions, and results show that the cognitive engine can manage to overcome the interference by adapting configurable parameters whereas a radio minus cognition capability was not able to overcome interference resulting in high errors or a loss of connectivity.

Further, a case study showing the use of metacognitive radio to improve wireless communication systems used for signaling and train control in railroads is presented in 96 . In a metacognitive engine 96 , a master process monitors and adapts the cognition process (i.e., cognitive engine). Metacognition enhances the performance of a cognitive radio with better learning and decision-making capabilities.

Research in VANETs has focused on simulation studies to evaluate the performance of communication protocols for VANETs. Testing and evaluation of VANETs communication protocols (e.g., VANETs-based multihop communication system) in real testbed scenarios for example, on the road, has been considered a difficult task as there are virtually no highly dense vehicular testbed that exists. Recently, a new cognitive approach to VANETs research was introduced in 97 that enables performing real vehicular experiments, which usually require several vehicles, using only a few vehicles equipped with wireless communication interfaces. This is done by setting up a virtual overlay network consisting of relaying and interfering vehicles on top of a group of only a few vehicles to conduct experiments. The communication packets travel over a random number of hops and experience interference from random number of vehicle transmitters with varying transmission characteristics.

The cognitive car testbed 98 is based on guidelines presented in 97 to conduct real experiments with VANET applications and communication protocols under constraints of vehicular environment and computing resources. The objective of the cognitive car testbed is to explore the possibility of using cognitive network technologies in advanced vehicular testbeds. It focuses on using cognitive radios to study the impact of different frequency bands and different frequency switching delays on the performance of VANET communication protocols. More specifically, cognitive car testbed enables evaluating a communication protocol vs. different variables including number of hops, density of interfering neighbors, wireless channel conditions, transmission frequency, and switching delay at the node.

In this context, it introduces the architecture guidelines for implementing a VANET testbed and, in addition, provides preliminary results for the experiments conducted with an accident warning system developed for highways and a cognitive network based on the software radio (SORA) technology provided by Microsoft. Preliminary results show that cognitive interfaces can be used as an additional tunable dimension in an experiment platform where highly dense vehicular testbeds can be set up with just few real vehicular resources (e.g., few cars and drivers). The cognitive interfaces enable testing novel techniques designed for addressing radio spectrum scarcity in a vehicular environment as well as performance of VANET communication protocols. The performance of VANET communication protocol can be evaluated by tuning two additional variables: frequency band and switching time. The first variable is the frequency band used for transmitting and receiving communication packets and the second variable is the switching time needed to change from one frequency to another.

VT-CORNET 95 is an open-source cognitive radio network testbed, developed by Virginia Tech (Blacksburg, VA, USA) for the evaluation of CR wireless communication protocols and applications. The testbed consists of 48 software-defined radio nodes (USRP2-based nodes) deployed over the four floors of a building and equipped with a custom-made daughterboard covering the frequency range of 100 MHz to 4 GHz. The CORNET nodes are remotely accessible to the registered users. A unique feature of the testbed is that it uses Software Communications Architecture (SCA) framework that assumes USRP nodes at the physical layer and provides support for generating and visualizing a range of radio configurations.

The testbed focuses on CE design, self-organizing networks, and topology research. VT-CORNET is a highly reconfigurable testbed that enables testing of independently developed CR engines, sensing techniques, applications, protocols, performance metrics, and algorithms. In addition to the static 48 nodes deployed in the ceiling, low-power mobile nodes will also be available to support vehicular environment. The rail-CR project has adapted the VR-CORNET cognitive radio technology for railways as described in the first part of this section.

The CORNET testbed is openly available for conducting research on advanced CR networks. It provides a collection of resources to researchers lacking the ability to perform advanced experiments because of limited exposure to software-defined radio and CR platforms. The resources are made available to the researchers through flexible RF front ends, open-source software, and FCC experimental licenses.

The ORBIT testbed 99 is an open-access research testbed for next generation wireless networking that allows large-scale reproducible experimentation on next generation protocols and applications. It consists of an indoor radio grid emulator for large-scale reproducible experiments and an outdoor field trial system for supporting real-world experimentation and evaluations.

A 400-node radio grid emulator has been deployed at the WINLAB Tech Center building and is open for general use by the research community to conduct experiments. The ORBIT testbed provides support for conducting a wide range of experiments including mobile ad hoc networks, dynamic spectrum coordination, network virtualization, wireless security, and vehicular networking. The radio grid testbed enables emulation of real-world network topologies via noise injection and packet filtering. The testbed has been recently upgraded to include programmable software-defined radios (GNU USRP/USRP2) for flexible MAC/PHY experiments.

The testbed users can access the radio grid through a web portal to specify their experiments and measurements. As mentioned previously, the testbed can be accessed worldwide for conducting experiments and evaluate protocols and applications related to ad hoc routing, P2P, spectrum sensing, and dynamic spectrum access. Recently, the testbed has started providing support for vehicular mobility as well. The ORBIT management software has been extended to enable mobile vehicular (vehicle-to-vehicle networks) experiments, and the radio grid is now supplemented by a number of outdoor and vehicular nodes.

Initial studies 30 72 on the effect of vehicular mobility on spectrum management show that the primary users may have to face adverse interference due to the incorrect detection of occupied frequencies as a result of the Doppler spread generated by vehicular mobility. At the same time, since a CR-enabled vehicle can gather samples of signal at different locations along its path while moving, this increases the spatio-temporal diversity of the samples and, in turn, decreases the chances of inaccurate decisions due to shadowing effects.

Open issues on how the performance of spectrum sensing is affected by vehicular mobility that need investigation include the following: techniques to achieve a balance between cooperation and spectrum scheduling for a CR-enabled vehicle on the move, how to use predictable mobility of vehicles to improve spectrum awareness, and how the mobility-related parameters such as speed and direction affect the performance of spectrum sensing 28 .

Mobile ad hoc networking is difficult in general. In case of CR-VANETs, V2V communication is more complex and difficult because of dynamically evolving topologies and frequent disconnections that are due to vehicles going out of range. This happens more often when the vehicular density is low. Such disconnections can create instability and performance issues for upper layer protocols such as routing and transport (TCP). Some cross-layer strategies to optimize transport protocols, for example TCP, can also be explored such as in 100 101 . For routing in vehicular networks, some approaches based on delay-tolerant networking (DTN) are becoming popular. DTN approaches can help in improving the stability and performance of routing protocols by providing some tolerance to disruptions. Thus, CR-VANETs routing protocols also need to integrate some DTN elements to counter frequent disconnections, apart from being spectrum aware.

Several machine learning approaches have been applied for cognitive radio. However, as discussed in this paper, CR-VANETs can have more stringent requirements in terms of requiring fast algorithms due to fast changing environment. Moreover, there can be different types of redundancies that can be exploited such as redundancy in network statistics at a given location at the same time of a day. This is especially pertinent in the case of public transport such as buses and trains. Thus, machine learning algorithms that are more specific to CR-VANETs should be explored. Moreover, it is also important to design the knowledge database, which is consultable by machine learning algorithms for faster convergence. Some design issues are the following: which statistics to measure and store, how to divide the region into different zones, how to characterize different environments from the point of view of radio propagation (open space vs. city landscape), how to fuse different measurements, etc.

Many of the works assume centralized decision making for spectrum access and sharing. This may be valid in cases with infrastructure support for example, when a RSU can take central decisions. However, setting up of infrastructure is difficult and using the services of cellular networks has some costs as well as bandwidth limitations. Thus, infrastructure support may not be available all the time and consequently, distributed algorithms are highly desired for CR-VANETs. Spectrum-aware distributed algorithms for CR-VANETs should be explored that only use local information, such as that available from neighboring vehicles. These algorithms should have low convergence time, incur low signaling overhead, and should be fault tolerant.

Directional antennas are an interesting way to reuse space through beamforming, by focusing the antenna pattern on a specific region. A recent work on analyzing multicast capacity with beamforming is 102 . Beamforming also improves throughput as signal quality is improved in the concentrated beam. Moreover, beamforming is more relevant in vehicles than as compared to, for example, smart phones. This is because smart phones rapidly change their orientation in our hands, which makes it difficult to exploit beamforming with smart phones. On the other hand, vehicles change their orientation only while changing the route and it can be used to enhance the communication. With adaptive beamforming, a vehicle can focus the antenna beam only on a particular vehicle and can thus avoid interference in another zones containing PUs or other SUs. Thus, adaptive beamforming combined with opportunistic spectrum access is an interesting direction to explore.

Most of the works focus on optimizing QoS metrics, such as throughput and delay. Some works optimize user utilities based on generic utility functions. However, it is ultimately the quality of experience (QoE) which matters to the users. For example, perceptual quality, an important QoE component for videos, is important for passengers streaming a video. For video streaming applications, it is now easier to adapt to changing network characteristics with some new standards such as scalable video coding (SVC) and dynamic HTTP streaming (DASH). One of the optimization approaches, in wireless networks, can be to use subjective testing to design perceptual video quality models for SVC and DASH 103 104 and apply such models for optimizing QoE, as done for wireless multicast in 105 . Thus, similar QoE aware approaches can be explored in the context of CR-VANETs by considering additional parameters related to spectrum characteristics and availabilities.

Typically, the hardware experiment platforms designed for VANETs and CR-VANETs consist of regular vehicles and software-defined radios 106 107 and it is complex to set up such platforms involving vehicle mechanics integrated with radios. Thus, currently, simulators are used. Using traffic simulators such as SUMO 108 and network simulators (NS-2/3 and OMNeT++) together enable designing simulators that combine vehicle mobility and communication networks. However, there is a lack of suitable simulators with features such as spectrum database, spectrum sensing and management, power spectral density, wireless interference level, and Doppler effects in order to be able to model and simulate varying topology of CR-VANETs and local wireless networks.

With the use of cognitive radio for vehicular networks, there are additional security-related issues which need to be addressed, such as protecting the privacy of cooperating vehicles and identifying malicious CRenabled vehicles that broadcast false spectrum sensing reports while on the move. Addressing security issues in CR-VANETs is challenging as neighboring vehicles can keep changing over time while moving. While moving on the road, it is difficult to detect a malicious car that may be transmitting fake spectrum sensing reports as this requires rapid detection and correction.

Moreover, it is important to consider the privacy concerns of end users while using collaborative spectrum sensing in CR-enabled vehicles. Security issues should focus on how to prevent a potential attacker from tracking a driver’s identity and movements by eavesdropping the periodic spectrum information broadcast by the vehicle 28 .