Abstract
In vehicular ad hoc networks (VANETs), some distinct characteristics, such as high node mobility, introduce new nontrivial challenges to qualityofservice (QoS) provisioning. Although some excellent works have been done on QoS issues in VANETs, security issues are largely ignored in these works. However, it is know that security always comes at a price in terms of QoS performance degradation. In this article, we consider security and QoS issues jointly for VANETs with cooperative communications. We take an integrated approach of optimizing both security and QoS parameters, and study the tradeoffs between them in VANETs. Specifically, we use recent advances in cooperative communication to enhance the QoS performance of VANETs. In addition, we present a preventionbased security technique that provides both hopbyhop and endtoend authentication and integrity protection. We derive the closedform effective secure throughput considering both security and QoS provisioning in VANETs with cooperative communications. The system is formulated as a partially observable Markov decision process. Simulation results are presented to show that security schemes have significant impacts on the throughput QoS of VANETs, and our proposed scheme can substantially improve the effective secure throughput of VANETs with cooperative communications.
Keywords
VANETs Cooperative communications Qualityofservice Security1 Introduction
Recently, there is a strong interest in vehicular ad hoc networks (VANETs), where vehicles can dynamically establish an ad hoc network without necessarily using a fixed infrastructure. VANETs can offer various applications and tremendous benefits to Intelligent Transportation Systems [1]. For example, safety information exchange using VANETs enables lifecritical applications, such as the alerting functionality during intersection traversing and lane merging. Valueadded services using VANETs can enhance drivers’ traveling experience by providing convenient Internet access, navigation, toll payment services, etc. [2].
Certainly, qualityofservice (QoS) issues in traditional mobile ad hoc networks in general are still of interest in VANETs. However, some distinct characteristics of VANETs, such as high node mobility, introduce new nontrivial challenges to QoS provisioning in VANETs [3, 4]. Particularly, in vehicletovehicle (V2V) communications, due to high vehicle mobility and relatively low elevation of the antennas on the communicating vehicles, other vehicles will act as obstacles to the signal, often affecting propagation even more than static obstacles (e.g., buildings or hills), especially in the case of an open road [5]. Indeed, nonlineofsight safetycritical conditions require careful attention in order to provide safety benefits in VANETs [6].
There are some studies on QoS issues in VANETs. Rawat et al. [7] propose a scheme to adapt transmission power at the physical layer and contention window size at the medium access control (MAC) layer based on the estimated local vehicle density to enhance VANET performance. Rate control, MAC, and routing problems in cooperative VANETs are studied in [8], where a crosslayer solution is developed. In [9], a contextual cooperative congestion control policy is proposed to exploit the traffic context information of each vehicle to reduce the channel load, while satisfying the vehicular applications requirements. Crosslayer routing is studied in [10] by applying cooperative transmission and a new strategy of path selection to achieve a better tradeoff between the transmission power consumption and endtoend reliability.
While some excellent studies have been done on QoS issues in VANETs, security issues are largely ignored in these works. This is not surprising, as these two important areas have traditionally been addressed separately in the literature. However, security is one of the main challenges for VANETs [11], and it is known that security always comes with a price in terms of QoS performance degradation, since securing communications against the adversary typically consumes network resources in terms of bandwidth and/or hardware capacities [12]. This price may be tolerable in wireline networks, but it may dominate the consumption of scarce network resources in VANETs. This situation makes the study of tradeoffs between QoS and security in VANETs an important open challenge [13].
In this article, we consider security and QoS issues jointly for VANETs with cooperative communications. We take an integrated approach of optimizing both security and QoS parameters, and study the tradeoffs between them in VANETs. To the best of the authors’ knowledge, combining security and QoS issues for VANETs with cooperative communications has not been considered in existing works. Some distinct features of this study are as follows.

We use recent advances in cooperative communication to enhance the QoS performance of VANETs. Cooperative communication in wireless networks takes advantage of the broadcast nature of the wireless medium to have nodes adjacent to the source transmit the message to the destination. As a result, nodes in the network act not only as end users but also as relays for others to create a spatial diversity that allows for increased throughput and reliability [14, 15]. Cooperative communication has been considered as a promising technique, and has been involved in the standards of WiMAX [16] and 3GPPLTE [17].

Preventionbased techniques, such as authentication, are crucial as the front line of defence for the integrity, confidentiality, and nonrepudiation of communications [18]. In this article, we propose a preventionbased security scheme for VANETs with cooperative communications. Specifically, we make use of an authentication protocol referred as adaptive and lightweight protocol for both hopbyhop and endtoend authentications (ALPHA) [19], which is based on hash chains and Merkle trees (MT), i.e., a tree of hashes [20].

Based on the proposed preventionbased security scheme for VANETs with cooperative communications, we study the relay selection problem in VANETs. In previous works on relay selection (e.g., [14]), it is generally assumed that the channel conditions are perfectly known and remain in the same state from the current frame to the next. However, these assumptions may not be realistic in VANETs due to high node mobility. Therefore, in this article, we consider channel estimation errors and Markov channel models to improve the performance in VANETs.

We formulate the system as a partially observable Markov decision process (POMDP) [21], which has successfully been used to solve the security scheduling problem [18] among others. The obtained policy for security and QoS parameters has an indexability property that dramatically reduces the computation and implementation complexity. Effective secure throughput is considered as the optimization objective in our formulation.

Simulation results are presented to show that security schemes have significant impacts on the throughput QoS of VANETs, and our proposed scheme can substantially improve the effective secure throughput of VANETs with cooperative communications.
The remainder of the article is structured as follows. Section 2 presents the system model. We derive the secure throughput in Section 3. Stochastic formulation of the joint design of security and QoS provisioning is presented in Section 4. Simulation results are presented and discussed in Section 5. Finally, we conclude this study in Section 6 with future work.
2 System model
In this section, we first describe a simple vehicle ad hoc network model. Then, the Markov channel model is introduced next. Finally, we describe the authentication model.
2.1 Network model
where P_{ w } is the average transmit signal power, W is the transmission bandwidth, and N_{0} is the noise.
We denote the channel gain between two nodes, x and y, as h_{ x y }. Therefore, the channel gain between the source vehicle and the destination vehicle is denoted as h_{ S D }. The channel gain between the source vehicle and a relay vehicle R_{ k } is denoted as ${h}_{S{R}_{k}}$, and the channel gain between a relay vehicle and the destination vehicle is denoted as ${h}_{{R}_{k}D}$. We further denote the average received SNR between the source vehicle and the destination vehicle as γ_{ S D }, the average received SNR between the source and the relay as ${\gamma}_{S{R}_{k}}$, and the average received SNR between the relay vehicle and the destination vehicle as ${\gamma}_{{R}_{k}D}$. Accordingly, we can get ${\gamma}_{\mathit{\text{SD}}}\phantom{\rule{0.3em}{0ex}}=\phantom{\rule{0.3em}{0ex}}\frac{\gamma}{{h}_{\mathit{\text{SD}}}}$, ${\gamma}_{S{R}_{k}}\phantom{\rule{0.3em}{0ex}}=\phantom{\rule{0.3em}{0ex}}\frac{\gamma}{{h}_{S{R}_{k}}}$, and ${\gamma}_{{R}_{k}D}\phantom{\rule{0.3em}{0ex}}=\phantom{\rule{0.3em}{0ex}}\frac{\gamma}{{h}_{{R}_{k}D}}$.
In this article, since our main focus is on the joint design of security and QoS issues, we assume that the problem of fighting for channel access among multiple nodes is handled by MAC layer, which will be responsible for resource sharing and contention resolution among multiple nodes. There are many articles studying MAC issues in cooperative communications in the literature (e.g., [4, 23]). The proposed design in this article can be used with these MAC schemes.
2.2 Channel model
In this article, we use finitesate Markov channel (FSMC) models. FSMC models have widely been accepted in the literature as an effective approach to characterize the correlation structure of wireless channels. These include the following channels: satellite channels [24], indoor channels [25], Rayleigh fading channels [26], Ricean fading channels [27], and Nakagami fading channels [28]. Considering FSMC models can enable substantial performance improvement over the schemes with memoryless channel models [29, 30].
In the FSMC, the range of the channel gain is partitioned (quantized) into L levels, and each level is associated with a state of a Markov chain. The channel varies over these states at each time slot according to a set of Markov transition probabilities. In VANETs, the different channel gains between source and relay (S2R) ${h}_{S{R}_{k}}$, relay and destination (R2D) ${h}_{{R}_{k}D}$, as well as source and destination (S2D) h_{ S D } can be modeled as a random variable according to an FSMC, which is characterized by a set of states Γ=γ_{0},γ_{1},…,γ_{L−1}. Due to high node mobility and channel estimation errors, the channel states may not be perfectly known.
Similarly, we can get the channel state transition probability matrix of relay k for relay to destination channel as Ψ_{ k }=[ψ_{ k }(i,j)]_{L×L}, and the channel state transition probability matrix for source to destination channel as Ξ_{ k }=[ξ_{ k }(i,j)]_{L×L}.
2.3 Authentication model
There are several ways to perform authentication in communication networks. Traditional public key infrastructure (PKI) approaches are gaining popularity in wireless networks. PKI scheme uses a public key validated by a trusted third party to encrypt a message that can only be decrypted by the corresponding private key. In general, PKIbased authentication mechanisms are relatively expensive in terms of generating and verifying digital signatures. Symmetric cryptography, where the communicating nodes share a secret, is more efficient due to its reduced computational complexity. However, when used in cooperative communication networks, distributing the shared keys in the first place becomes a problem.
where h a_{ i } is the anchor of the hash chain corresponding to the last hashed value for that hash chain.
Although hash chains are uncomplicated to calculate and easy to use, they are not sufficient to prevent insider attacks by relay nodes. However, the ALPHA can prevent insider attacks through integrity protection and also perform authentication making use of MT and interactionbased hash chains, which is based on delayed message disclosure [19]. When hash chains are combined in an MT in ALPHA [19], they allow for the authentication of identities while the MT provides integrity protection for individual messages, which is especially useful for onpath verification with the highvolume data in cooperative communication networks. We now begin to describe how the ALPHAMT scheme works in VANETs with cooperative communications.
3 Secure throughput in VANETs with cooperative communications
As we mentioned in Section 1, security always comes with a price in terms of QoS performance degradation. Throughput is one of the main QoS measures in VANETs. In this section, we derive the effective secure throughput in VANETs with cooperative communications, which will be used as the objective function in our optimization formulation in Section 4. We first derive the outage capacity. Then, bit error rate is derived. Finally, we obtain effective secure throughput considering both the authentication protocol and cooperative communications.
3.1 Outage capacity
where ∣h_{ S D }∣ is the channel between the source and the destination. To be sustainable, the data rate over this channel r should be less than the mutual information I_{noncoop}.
In the cooperative decodeandforward (DF) relaying mode, the transmission between the source and the destination makes use of the intermediate relay node. As stated, the relays operate in half duplex and cannot receive and transmit simultaneously. The relay that maximizes the mutual information between the source and destination is selected as the best relay. As indicated earlier, the transmission is divided into two time slots. In the first time slot, the source transmits the signal to both the selected relay and the destination. In the second time slot, the selected relay decodes the received signal, reencodes it, and forwards it to the destination node. The destination combines the received signal from the relay and source nodes using maximal ratio combining (MRC).
Given the halfduplex constraint, the factor $\frac{1}{2}$ reflects the two time slots for relaying.
In the DF opportunistic relay, the relay is selected from the entire set of available relays. The relay transmits only if both source–relay and relay–destination mutual information are above the required rate r. Thus, the source selects the relay that maximizes the minimum mutual information between the source–relay and the relay–destination channels.
This indicates that the channel cannot support the transmission rate and consequently the data transmission is unsuccessful. It is an important analytical metric that characterizes the probability of data loss providing a bound on the symbol error rate or equivalently of deep fading.
3.2 Bit error rate
where ${P}_{\text{out}}^{S{R}_{k}}$ is the outage probability of the link from source to relay.
3.3 Secure throughput
In this section, we discuss the throughput performance of the authentication protocol by considering the outage capacity and BER of the direct communication (DC, communication without the use of relay) and source–relay–destination communication paths. The error rate is also taken into consideration by applying ARQ retransmission schemes, which involves error detection and retransmission of lost or corrupted packets.
where S_{payload} is the amount of payload that can be transmitted with a single presignature, n is the number of messages/data blocks in the MT, S_{packet} is the size of the packet, and S_{ h } is the hash output.
where T_{1} is the time for the initial presignature process between the source and the destination. It works like a basic StopandWait ARQ model (explained below) with transmission of S_{1} packet by the source, processing at the destination, transmission of acknowledgment A_{1} packet by the destination and processing at the source. The message delivery is complete only after the source receives the confirmatory acknowledgment from the destination; T_{2} is the time taken for the actual message transmission and delivery, i.e., the actual transfer of messages from the source through the S_{2} packets and the transfer of acknowledgments from the destination through A_{2} packets.
Both T_{1} and T_{2} are dependent on the data transmission rate, which is equal to the outrage capacity described in Section 3.1.
Equation (21) shows the generic throughput for the authentication protocol. To improve system reliability, an ARQ scheme is needed. As selectiverepeat SRARQ has been proven to outperform other forms of basic ARQ schemes (stopandwait ARQ, gobackN ARQ) [32], we use SRARQ in this study.
For each packet size S_{packet}, the optimal value of the number of messages (n) in the MT, which corresponds to the number of S_{2} packets, is the value that results in the highest throughput, which is denoted by n^{∗}. There is a tradeoff as the throughput increases initially with the number of messages in the MT but then starts to decrease as a consequence of the larger signature size overheads from the increased number of messages in the MT. Therefore, one of the objectives in our research is to find the optimal number of messages in the MT for relay R_{ k }.
4 Stochastic formulation of the joint design of security and QoS provisioning
In this section, we formulate the effective secure throughput optimization problem in the system described above as a POMDP [21], which can determine the optimal policy for the number of messages/data blocks in the MT selection (for security) and relay selection (for QoS) to maximize the system effective secure throughput.
Markov decision process (MDP) provides a mathematical framework for modeling decision making in situations where outcomes are partly random and partly under the control of a decision maker. In VANETs with cooperative communications, the vehicles make decisions at specific time instances according to the current state s(t), and the system moves into a new state based on the current state s(t) as well as the chosen decision a(t).
As described in Section 2, we use FSMC. Given the current channel state s(t), the next channel state is conditionally independent of all previous states and actions. This Markov property of state transition process makes it possible to model the optimization problem as an MDP. Furthermore, in VANETs, due to channel sensing and channel state information errors, the system state cannot directly be observed. As a result, we formulate the optimization problem as a POMDP, in which it is assumed that the system dynamics are determined by an MDP, but the underlying state can only be observed inaccurately, or with some probabilities.
A POMDP can be defined by a hextuple <S,A,P,Θ,B,R>, where S stands for a finite set of states with state i denoted by s_{ i }, A stands for a finite set of actions with action i denoted by a_{ i }, P stands for transition probabilities for each action in each state, and ${p}_{\mathit{\text{ij}}}^{a}$ denotes the probability that system moves from state s_{ i } to state s_{ j } when action a is performed, Θ stands for a finite set of observations, and θ_{ i } denotes the observation of state i, B is the observation model, and ${b}_{\mathrm{j\theta}}^{a}$ denotes the probability that Θ was observed when the system state is s_{ j } and last action taken is a, and R stands for the immediate reward. ${r}_{\mathit{\text{ij}}}^{a}$ denotes the immediate reward received for performing action a and the system state moves from s_{ i } to state s_{ j }, with an observation Θ.
In our POMDP model, the vehicle node has to make a decision whenever a slot has elapsed. These instant times are called decision epochs. The optimal optimization policy can be obtained from value iteration algorithms in this formulation. Using the POMDPderived policy, a channel state is observed according to the information from last slot. Based on the observation, the system jointly considers the number of messages/data blocks selection and relay selection to maximize the system throughput.
In order to obtain the optimal solution, it is necessary to identify the states, actions, state transition probability, observation model, and reward functions in our POMDP model, which is described in the following sections.
4.1 Actions, states, and observations
where a_{ n }(t) is the action to decide the number of messages/data blocks in the MT, and a_{ n }(t)>0. a_{ R }(t) is the relay selection action, and a_{ R }(t)∈{1,2,…,K}, where K is the number of relays.
where ${h}_{S{R}_{k}}$ is the channel gain between source and relay R_{ k }, ${h}_{{R}_{k}D}$ is the channel gain between relay R_{ k } and destination, and h_{ S D } is the channel gain between source and destination.
where $\hat{{h}_{S{R}_{k}}}\left(t\right)$, $\hat{{h}_{{R}_{k}D}}\left(t\right)$, and $\hat{{h}_{\mathit{\text{SD}}}}\left(t\right)$ are the observation of ${h}_{S{R}_{k}}$, ${h}_{{R}_{k}D}$, and h_{ S D }, respectively, and they have the same space as the state space.
4.2 State transition model and observation model
where $\varphi \left({h}_{S{R}_{k}}\right(t),{h}_{S{R}_{k}}(t+1\left)\right)$, $\psi \left({h}_{{R}_{k}D}\right(t),{h}_{{R}_{k}D}(t+1\left)\right)$, ξ(h_{ S D }(t),h_{ S D }(t+1)) are the channel state transition probabilities for difference channels as described in Section 2.2.
where ${B}_{S{R}_{k}}$, ${B}_{{R}_{k}D}$, and B_{ S D } are channel observation probability matrices for S2R channel, R2D channel, and S2D channel, respectively. ⊗ denotes Kronecker product which is used here to expand the transition matrices. Note that all the channel observation probability is independent. That is why we can use ⊗ to expand it.
4.3 Information state
Information state is an important concept in POMDP. We refer to a probability distribution over states as the information state and the entire probability space (the set of all possible probability distributions) as the information space. Let ${\Pi}^{t+1}=\left\{{\Pi}_{0}^{t},{\Pi}_{1}^{t}\phantom{\rule{0.3em}{0ex}}\dots ,{\Pi}_{S}^{t}\right\}$ denote the information space, where ${\Pi}_{i}^{t}$ represents the probability that the current state is i at time t. As will be shown later, the knowledge of the system dynamics and the transition probabilities must be known in order to maintain an information state.
where ${p}_{\mathit{\text{ij}}}^{a}$ is the probability when the system state changes from i to j when action a is adopted. ${b}_{\mathrm{j\theta}}^{a}$ stands for the observation probability that we observe the system state j to Θ when action a is adopted. Both ${p}_{\mathit{\text{ij}}}^{a}$ and ${b}_{\mathrm{j\theta}}^{a}$ are described in Section 4.2.
The new information state will be a vector of probabilities computed according to the above formula. The information states capture all the history information at time t. Therefore, we can save all the past actions and observations by constantly updating the information state. Also, it is reasonable to make decisions according to the information state.
4.4 Reward function and objective
where T h r_{ S R } is the throughput for the authentication process with SRARQ, and it is derived in Section 3.3.
where β and (1−β) are importance weight factors to indicate the importance of throughput and communication delay. In (34), we combine throughput and delay into a single function. This is a common approach used in the optimization literature, which is called Aggregate Objective Function, to solve an optimization problem with multiple objectives [34, 35]. In reality, different VANETs have different throughput and packet delay requirements. By adjusting the parameters in (34), the proposed scheme is generic enough to accommodate different requirements in practical VANETs.
where μ_{ h } specifies the number of messages/data blocks selection policy, μ_{ R } is the relay selection policy, ${E}_{{\mu}_{n},{\mu}_{R}}$ is the expectation when the policies μ_{ h } and μ_{ R } are employed, and t_{0} is the initial time.
4.5 Separation principle for optimal policy
In this section, we solve the POMDP model to obtain the optimal policy for the number of messages/data blocks selection and relay selection. Specifically, we establish a separation principle that simplifies the calculation process.
where Π_{t+1} represents the updated knowledge of system state after incorporating the action a(t) and the observation θ(t) in the epoch t.
for some sets of vectors ${\alpha}_{i}^{k}\left(t\right)=\left\{{\alpha}_{i}^{0}\left(t\right),{\alpha}_{i}^{1}\left(t\right),\dots \right\}$. The sets of αvectors represents the coefficients of one of the linear pieces of a piecewise linear function. These piecewise linear functions can represent the value functions for each step in the finite horizon POMDP problem. We only need to find the vector that has the highest dot product with the information state to determine which action to take.
One of the main problem in our POMDP model is the action space. As shown in Section 4.1, the number of messages/data blocks selection action space is {a_{ n }(t):a_{ n }(t)>0}. The infiniteness of the action space makes it hard to solve the model with traditional value iteration algorithms. To this point, we establish a separation principle that leads to closedform optimal design of the number of messages/data blocks selection and relay selection strategy. The policy calculation is carried out in two steps without losing optimality.
5 Simulation results and discussions
In order to evaluate the performance of our proposed scheme, we have carried out a set of simulation experiments using NS2 simulator. We first illustrate our secure throughput model performance. The performance improvement of our POMDP optimization algorithm is given next. We then discuss the effects of the channel state transition matrix and observation model parameters on the optimal policy.
We took the processing time at each node as 10 μ s, hash size as 20 bytes, and a fixed outage probability of 0.01. In all figures, the values represent the average results of 20 different runs.
5.1 Throughput improvement
5.2 Effects of the state transition matrix
We can observe from this figure that the POMDP policy achieves a much greater performance improvement in comparison to the existing and random policies when the transition probability of staying in the same state is very small. The average throughput in the existing policy gradually approaches to the POMDP case with the increase of that probability. This is because when the transition probability of staying in the same state increases, the channel becomes more memoryless, and the advantage of POMDP policy is not obvious given a memoryless channel.
5.3 Effects of the observation model parameters
The observation matrix in (30) is derived from the channel estimation error δ. We evaluate how the channel estimation error affects the average throughput.
6 Conclusions and future work
The distinct characteristics of VANETs, such as high node mobility and relatively low elevation of the antennas on vehicles, make the QoS provisioning challenging. In this article, we proposed to use recent advances in cooperative communications to enhance the QoS performance of VANETs. In order to address the security problem caused by cooperative communications, we presented a joint design of security and QoS provisioning in VANETs. We proposed a preventionbased technique for secure relay selection taking into consideration authentication protocol, which is based on hash chains and MT, to provide both endtoend and hopbyhop authentication and integrity protection. Particularly, we considered channel estimation errors and the impacts of security on throughput QoS performance in VANETs. The dynamic wireless channel was modeled as a finitesate Markov process. With channel estimation errors, the channel state cannot accurately be observed. Therefore, we formulated the relay selection and the number of messages/data blocks selection problem as a POMDP. The optimal policy was obtained by a separated principle. Simulation results show that the number of messages/data blocks in the MT has significant impacts on the throughput QoS. The proposed scheme significantly improves the effective secure throughput. In addition, due to considering the channel errors in the formulation, the POMDP policy decreases the observation errors’ impacts on the throughput performance.
Future work is in progress to consider network topology control in VANETs using the proposed combined security and QoS provisioning framework.
Acknowledgements
We thank the reviewers for their detailed reviews and constructive comments, which have helped to improve the quality of this article. This study was in part supported by Beijing Laboratory For Mass Transit, he Key Projects in State Key Lab. of Rail Traffic control and Safety (RCS2012ZQ002, RCS2012K010), the China Education Ministry Funding Project (2013JBM124,2011JBZ014), the National Science Foundation of China (No. 61132003), the National High Technology Research and Development Program of China (863 Program) (2011AA110502), and by the Natural Sciences and Engineering Research Council (NSERC) of Canada and industrial and government partners, through NSERCDIVA Strategic Research Network.
Supplementary material
Copyright information
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.