Reliable Traffic Monitoring Mechanisms Based on Blockchain in Vehicular Networks

Consider a smart city, there is a traffic administration (TA) that is responsible for monitoring real-time traffic condition in this city. It can crowdsource its tasks of traffic information collection to vehicles on the road by paying them a certain amount of money. Therefore, the blockchain-based real-time traffic monitoring (BRTM) system consists of two important entities: TA and vehicle, which is shown in Fig. 1. In the BRTM system, the functionality of the entities in a smart city $S_{a}$, $a\in\mathbb{Z}^{+}$, can be shown as follows:

Traffic administration (TA): The TA in the city $S_{a}$ can be denoted by $TA_{a}$, which collects the real-time traffic information from those vehicles running in this city. For instance, during peak time, traffic accidents often occur on some heavily trafficked roads. At this time, the $TA_{a}$ needs to analyze road conditions instantly according to the responses provided by those vehicles that run on the designated location. If $TA_{a}$ adopts the information provided by some vehicles, it must pay them with data coins. Here, data coin is a kind of digital currency considered as the payment for traffic information.

Vehicles: The vehicles in this system are installed with camera sensors, wireless communication devices, and microcomputer systems, which are able to collect, process, and transmit data to other corresponding devices. The active vehicles in the city $S_{a}$ can be denoted by a set $V_{a}=\{v_{a1},v_{a2},\cdots,v_{ab},\cdots\}$, where active vehicles are those participating in information trading with $TA_{a}$ and idle vehicles do not want to share their traffic information. Once a transaction between $TA_{a}$ and $v_{ab}$ has been completed and validated, the vehicle $v_{ab}$ will receive a certain number of data coins from $TA_{a}$.

Based on that, a smart city in our BRTM system can be denoted by $S_{a}=(TA_{a},V_{a})$, and the whole BRTM system $\mathbb{S}$ can be denoted by $\mathbb{S}=\{S_{1},S_{2},\cdots,S_{a},\cdots\}$. It is very easy to understand, for example, this system can be constructed in a country that is composed of a number of smart cities. Here, all TAs that serve under the cities in $\mathbb{S}$ are interconnected with each other to make up a peer-to-peer (P2P) wide-area network. We take Fig. 1 as an instance to demonstrate it.

The P2P network among the TAs in the BRTM system is used for constructing the blockchain network. Here, blockchain is an effective technique to record the transactions and guarantee the security. In our blockchain network, there are two types of nodes which are full node and lightweight node. For the full nodes, they not only executes the consensus process to validate the candidate block but also manages stores the blockchain that contains the whole transactions between TA and vehicles. For the lightweight nodes, they merely store the block headers that is convenient for them to check and verify. In our system $\mathbb{S}$, each $TA_{a}$ of $S_{a}\in\mathbb{S}$ is defaulted as a full node and each vehicle $v_{ab}\in V_{a}$ of $S_{a}\in\mathbb{S}$ is defaulted as a lightweight node because of lacking enough storage and computational power. Details of the blockchain design will be described in Sec. IV.

Consider a smart city $S=(TA,V)$ such that $S\in\mathbb{S}$ at some point, we neglect the subscripts and denote by $V=\{v_{1},v_{2},\cdots,v_{n}\}$ for convenience. This traffic administration $TA$ releases a task set $T=\{t_{1},t_{2},\cdots,t_{m}\}$ about traffic information for the active vehicles to undertake. For the $TA$, there is an appraisement $a_{j}$ for each task $t_{j}\in T$. For each vehicle $v_{i}\in V$, it can attempt to finish a subset of tasks $T_{i}\subseteq T$ based on its ability and willingness. The cost of $T_{i}$ for the $v_{i}$ can be defined as $c_{i}$, which is private and not known to the $TA$. Then, the vehicle $v_{i}$ determines a bid $b_{i}$ and respones the $TA$ with its bid-task pair $(T_{i},b_{i})$, where $b_{i}$ is $v_{i}$’s reserve price that is the lowest price it want to undertake the tasks $T_{i}$. Generally, we have $T_{i}\cap T_{j}\neq\emptyset$ for any vehicle $v_{i}$ and $v_{j}$. Here, for each task $t_{i}\in T$, we can quantify it as the traffic information at some position in this city, thereby an active vehicle is able to submit a subset of tasks based on its running path. After receiving the responses from all active vehicles, the $TA$ needs to select a winner (acceptance) set $W\subseteq V$ and give a payment $p_{i}$ for each $v_{i}\in W$. The utility for a vehicle $v_{i}\in V$ is

The profit of the $TA$ is

where we denote by $A(W)=\sum_{t_{j}\in\cup_{v_{i}\in W}T_{i}}a_{j}$. Besides, in order to control cost, the $TA$ has a total budget $B$ such that $\sum_{v_{i}\in W}b_{i}\leq B$. Based on that, we want to design an incentive mechanism to execute this auction process.

Consider a smart city $S=(TA,V)$, each vehicle $v_{i}\in V$ has to register with the certificate authority (CA) through correlating its license plate so as to acquire a unique identification $ID_{i}$. Then, each legitimate vehicle will be assigned with a private/public key pair $(SK_{i},PK_{i})$, where the private key is reserved by itself and the public key is known by all the legitimate nodes in the blockchain as a pseudonym. The asymmetric encryption are adopted widely in the current blockchain technology for the sake of data integrity. The data encrypted by the private (resp. public) key can be decrypted by the public (resp. private) key. After registering, the vehicle $v_{i}$ will obtain a certificate $Cer_{i}$ signed by CA’s private key that can certify the authenticity of its identity. There is an account $A_{i}=\{ID_{i},Cer_{i},(SK_{i},PK_{i}),B_{i}\}$ associated with the $v_{i}\in V$ where $B_{i}$ is its data coin balance. Similarly, the $TA$ has an account denoted by $A_{*}=\{ID^{\prime},Cer^{\prime},(SK^{\prime},PK^{\prime}),B^{\prime},Rep^{\prime}\}$ as well, where $Rep^{\prime}$ is its reputation value which will be introduce later. If a round of trading is finished successfully, data coins will be transferred from $B_{*}$ to $B_{i}$.

The elaborate process of traffic information trading in our BRTM system are drawn in Fig. 2, whose specific operation steps are summarized as follows:

1) The $TA$ publishs its task set $T$ across the active vehicles in this city. The published message can be denoted by $PubMsg=\{SK^{\prime}(T),Cer^{\prime},STime\}$, where is the task set is encrypted by the $TA$’s private key for security and $STime$ is the time stamp of this message generation.

2) After receiving the $PubMsg$ from the $TA$, each active vehicle $v_{i}\in V$ in this city has to verify the $TA$’s certification $Cer^{\prime}$ by the CA’s public key and then read the task set by the $TA$’s public key. Once verified, each $v_{i}$ will determine its task-bid pair $(T_{i},b_{i})$ and send its request message $ReqMsg$, denoted by $ReqMsg=\{PK^{\prime}(T_{i},b_{i}),Cer_{i},STime\}$, back to the $TA$. Here, its task-bid pair is encrypted by the $TA$’s public key because it can only be read by the $TA$ for privacy.

3) The $TA$ waits to collect all the $ReqMsg$ from the active vehicles in this city, then determine the winner set $W$ and its corresponding payment $p_{i}$ for $v_{i}\in W$. This process is executed by the built-in smart contract, which adopts a budgeted auction mechanism explained in Sec. V.

4) The $TA$ sends the order message to each accepted vehicle $v_{i}\in W$, which can be denoted by $OrdMsg=\{PK_{i}(p_{i}),Cer^{\prime},STime\}$. Here, the payment is encrypted by the $v_{i}$’s public key which can be read by the $v_{i}$ merely.

5) If receiving the $OrdMsg$ from the TA, it means that the $v_{i}$ has been selected as an information provider. Then, the $v_{i}$ can read it by its private key and send the response message that includes the traffic information associated with task $T_{i}$, denoted by $ResMsg=\{PK^{\prime}(data),Cer_{i},STime\}$, back to the $TA$. The data is encrypted by the $TA$’s public key due to the same reason as (2).

6) After the data transmission, the $TA$ will check and confirm whether the traffic information about task $T_{i}$ from the $v_{i}$ meets its requirement. If yes, the $TA$ pays the data coins $p_{i}$ to the $v_{i}$’s public wallet address.

7) Once receiving the payment from the $TA$, the $v_{i}$ will sends the confirm message, denoted by $ConMsg=\{PK^{\prime}(Confirm),Cer_{i},STime\}$, back to the $TA$. The $TA$ will sign it with a confirm message as well, then a transaction record is formulated until now.

8) The built-in smart contract in the $TA$ will record a series of transactions between $TA$ and vehicles in this city within a period of time, and package them into a new block. Before this block is added into the blockchain, it has to be reached a consensus among all the TAs in the BRTM system. The consensus process we adopt here is called as reputation-based delegated proof-of-work (DPoS) mechanism explained in Sec. IV.C. After finishing the consensus process, this block will be appended into the blockchain in a linear and chronological order permanently.

A consensus process is essential to guarantee the consistency and security for the blockchain system. The classic PoW consensus mechanism is not suitable to our system because of lacking the necessary computational power in the vehicular networks. Therefore, we design a novel reputation-based DPoS mechanism that not only ensures reliability but also improves performance. Just as said before, all the TAs in the BRTM system $\mathbb{S}$ are considered as the full nodes, thereby the consensus process can be carried out among them. The operational steps are displayed as follows:

1) Witness election: Denote by the full node set $\mathbb{Z}=\{TA_{1},TA_{2},\cdots,TA_{a},\cdots\}$, there is a reputation value $Rep_{a}$ associated with each full node $TA_{a}\in\mathbb{Z}$. This reputation $Rep_{a}$ can be considered as its stake that determines its voting weight directly. It implies that the nodes with higher reputation values can determine the results more than those with lower reputation values. At each witness election epoch, the voting result $R_{a}$ for each $TA_{a}\in\mathbb{Z}$ is defined as

where the $\mathbb{I}(a,a^{\prime})$ is an indicator. Here, $\mathbb{I}(a,a^{\prime})=1$ if the $TA_{a^{\prime}}$ votes to support $TA_{a}$, otherwise $\mathbb{I}(a,a^{\prime})=0$. All these reputation values and voting results are stored in the blockchain, thereby each node can check them by itself. We select a witness committee $\mathbb{M}\subseteq\mathbb{Z}$ that have the top $|\mathbb{M}|$ highest voting results. Then, it can be divided into two part such that $\mathbb{M}=\mathbb{D}\cup\mathbb{E}$ and $\mathbb{D}\cap\mathbb{E}=\emptyset$, where the full node in $\mathbb{D}$ is an active witness and in $\mathbb{E}$ is a standby witness. The active witnesses have the top $|\mathbb{D}|$ highest voting results, which are able to generate a new block like a leader. However, the standby witnesses can only verify and broadcast the block generated by an active witness.

2) Block generation: First, the active witnesses in the set $\mathbb{D}$ are sorted in a random sequence. Then, each $TA_{a}\in\mathbb{D}$ takes turn to be a leader round by round based on this sorting that is responsible for generating a new block. The leader has to verify and check all the transactions occurred recently and package those valid transactions into a new block. If a node in its turn does not produce a block over a given period of time successfully, it will be skipped and have no chance to be leader again in this election epoch. After passing $|\mathbb{D}|$ consensus rounds, namely an election epoch, we re-execute the witness election to elect a new witness committee $\mathbb{M}$ according to the new reputation values of the full nodes in $\mathbb{Z}$.

3) Consensus process: After producing a new block, the leader should broadcast it to all witnesses in $\mathbb{M}$. All witnesses should verify the leader’s identity and check this block, then broadcast their verification results signed by their private key. Next, each witness $TA_{a}\in\mathbb{M}$ compares its verified outcome with those from other witnesses. If confirmed, it will send a confirmation message to the leader, which means that this witness agrees to accept the new block. For the leader, it will append the new block into the blockchain if the proportion of witnesses that agree to accept this block is more than two thirds. All the full nodes should synchronize their local blockchain storage according to the longest chain principle and all the lightweight node should update their blockchain headers based on the newest one.

4) Reputation update: When a consensus round is finished or interrupted, we need to update the reputation values of the full nodes in $\mathbb{Z}$ according to their behaviors during this consensus round. Generally, the positive behaviors contribute to the accumulation of reputation, but the negative behaviors had a bad effect. We denote by $Rep_{a}(i)$ the repuation value of the full node $TA_{a}\in\mathbb{Z}$ at the $i$-th consensus round. At the $i$-th consensus round, we define

for each full node $TA_{a}\in\mathbb{Z}$. Here, if the $TA_{a}\in\mathbb{Z}$ participates the voting in witness election, we give $\alpha_{a}(i)=1$, otherwise $\alpha_{a}(i)=-1$. When the $TA_{a}$ is the leader at this consensus round, we give $\beta_{a}(i)=1$ if it generates a new block that is accepted by other witnesses eventually, otherwise $\beta_{a}(i)=-1$. If the $TA_{a}\in\mathbb{Z}\backslash\{\text{leader}\}$ is not the leader, we give $\beta_{a}(i)=0$. When the $TA_{a}\in\mathbb{M}\backslash\{\text{leader}\}$ is a member of witness committee but not the leader, we give $\gamma_{a}(i)=1$ if it verifies the block produced by the leader correctly, otherwise $\gamma_{a}(i)=-1$. If the $TA_{a}\in\mathbb{Z}\backslash\mathbb{M}$ is not a witness, we give $\gamma_{a}(i)=0$. Then, the $A$, $B$, and $C$ are three adjustable parameters that represent the rewards or punishments of voting, leader, and verification behaviors. Based on its behaviors $\Delta_{a}(i)$ at the $i$-th consensus round, we define the reputation value of the full node $TA_{a}\in\mathbb{Z}$ as

where the $TA_{a}$’s voting is meaningless when its reputation down to zero according to the (3). Moveover, we initialize the reputation $Rep_{a}(0)=0.5$ for each $TA_{a}\in\mathbb{Z}$ and make $B>C>A>0$ because of their importance.

Our BRTM system inherits the advantages of blockchain technology which helps to ensure the reliability of traffic information trading and privacy protection. Its main characteristics are summarized as follows: (1) Decentralization: the traffic information trading can be performed in a distributed manner without relying on the third trusted intermediaries, which avoids single point failure; (2) Reliability: when there is a node failing because of malicious attacks, the blockchain network can be guaranteed to work as usual; (3) Privacy protection: shown as Sec. IV.B, asymmetric encryption is used to prevent the message from reading by others, which makes the private message remain confidential during the auction process; (4) Efficiency: our reputation-based DPos mechanism not only makes the consensus process more reliable but also improves efficiency by screening out the trustworthy nodes in advance; (5) Transaction authentication: all transaction must be checked and verified by the witness before appending into the blockchain, thereby it is extremely hard to dominate the majority of witnesses to create an unreal block. Based on that, our BRTM system provides us with a secure, reliable, and efficient traffic information trading platform.

Assume the $TA$ gives a payment $p_{i}$ with $p_{i}=b_{i}$ for each vehicle $v_{i}\in V$, the Problem 1 can be reduced to an optimization problem, called “budgeted vehicle allocation problem”, that selects a winner set $W\subseteq V$ such that $\bar{P}(W)$ is maximized. We denote

where $\sum_{v_{i}\in W}b_{i}\leq B$. To be meaningful, we assume $\bar{P}(\emptyset)=0$ and it exists at least one vehicle $v_{k}\in V$ such that $\bar{P}(\{v_{k}\})>0$ with $b_{k}\leq B$. However, the budgeted vehicle allocation problem can be reduced to the classic set cover problem in polynomial time, thereby it is NP-hard to find the optimal solution definitely. Based on that, we have to seek help from designing an approximation algorithm which will make use of submodularity of objective shown in (6).

As we known, Buchbinder et al. [27] designed a double greedy algorithm for the unconstrained submodular maximization problem with a $(1/3)$-approximation under the deterministic setting and a $(1/2)$-approximation under the randomized setting. Back to our budgeted vehicle allocation problem, there is a constraint $\sum_{v_{i}\in W}b_{i}\leq B$ that is a knapsack constraint. This constraint affects the resulting profit obtained by the $TA$ and the number of vehicles that are selected as winners in the auction. Thus, a valid reverse auction mechanism design are required to ensure the truthfulness carefully. Based on the aforementioned analysis, this optimization problem can be catagorized to a non-monotone submodular maximization with knapsack constraint. Lee et al. [28] proposed a $(1/5-\epsilon)$-approximation algorithm to maximize any non-negative submodular function with knapsack constraint by means of fractional relaxation and local search method. Even though this algorithm can give us a constant approximation ratio, its process is complex and its performance is worse than the greedy-heuristic algorithm actually. The greedy-heuristic algorithm is shown in Algorithm 1.

Here, we denote by $\bar{P}(v_{i}|W)=\bar{P}(W\cup\{v_{i}\})-\bar{P}(W)$ and $b^{*}$ is the bid offered by vehicle $v^{*}$. Shown as Algorithm 1, we select a vehicle $v^{*}$ from the current constrained set $F$ with maximum unit marginal gain until $\bar{P}(v^{*}|W)<0$ or $F=\emptyset$. Despite the winner set returned by Algorithm 1 has no any theoretical bounds because of non-monotonicity and knapsack constraint, it is intuitive and has a good performance in the practical applications. Now, we have to explore whether the greedy-heuristic satisfies the aforementioned four properties in the reverse auction mechanism.

Next, let us look at whether the greedy-heuristic satisfies profitability, individual rationality, truthfulness, and computational efficiency one by one as follows:

1) Individual rationality: The $TA$ pay each vehicle in the winner set its bid, thus it is individually rational.

2) Profitability: Form line 5 in Algorithm 1, the while loop is terminated when there is no vehicle having postive marginal gain, which guarantee $\bar{P}(W)>0$.

3) Truthfulness: Let us consider an example. There are five tasks issued by the $TA$, denoted by $T=\{t_{1},t_{2},t_{3},t_{4},t_{5}\}$, with appraisement $\{a_{1}=2,a_{2}=3,a_{3}=4,a_{4}=2,a_{5}=5\}$ and the budget $B=5$. There are three active vehicles denoted by $V=\{v_{1},v_{2},v_{3}\}$ in this city. They want to undertake the $TA$’s tasks as $T_{1}=\{t_{1},t_{3},t_{5}\}$, $T_{2}=\{t_{1},t_{2},t_{5}\}$, $T_{3}=\{t_{3},t_{4},t_{5}\}$, $c_{1}=2$, $c_{2}=2$, and $c_{3}=2$. When each vehicle offers a bid truthfully, we have $\bar{P}(\{v_{1}\}|\emptyset)/b_{1}=(A(\{v_{1}\})-b_{1})/b_{1}=(11-2)/2=4.5$. Similarly, we have $\bar{P}(\{v_{2}\}|\emptyset)/b_{2}=4$ and $\bar{P}(\{v_{3}\}|\emptyset)/b_{3}=3.5$. Thus, vehicle $v_{1}$ is selected at the first iteration. At the second iteration, we have $\bar{P}(v_{2}|\{v_{1}\})/b_{2}=0.5$ and $\bar{P}(v_{3}|\{v_{1}\})/b_{3}=0$. Thus, vehicle $v_{2}$ is selected at the second iteration. Then, the greedy-heuristic terminates because the budget is exhausted. However, when vehicle $v_{2}$ offer an untruthful bid $b_{2}=c_{2}+\lambda$, we have $\bar{P}(v_{2}|\{v_{1}\})/b_{2}=3/(2+\lambda)-1$. If $\lambda\in(0,1)$, it will be selected as the winner at the second iteration. The algorithm terminates here and vehicle $v_{2}$ can get more payment by offering an untruthful bid, thus is does not satisfy truthfulness.

4) Computational efficiency: The running time of the greedy-heuristic is $O(Bnm/\min_{v_{i}\in V}\{b_{i}\})$ because it takes $O(nm)$, $|T|=m$ and $|V|=n$, to compute $\bar{P}$ and iterates at most $B/\min_{v_{i}\in V}\{b_{i}\}$ times, so computationally efficient.

As mentioned above, the greedy-heuristic algorithm is not meaningful due to lacking truthfulness even though it simple to implement. Thus, we have to design a budgeted reverse auction mechanism that not only gets a good profit by encouraging vehicles to undertake the tasks, but also satisfies the four properties shown as before, especially for truthfulness, in order to protect from manipulating this system malignantly by offering a unreal bid. Therefore, a valid auction mechanism needs to be designed although it may lose some of its profits. We propose a truthful budgeted reverse auction mechanism based on Myerson’s introduction [29].

Based on the Theorem 1, we design our budgeted reverse auction mechanism consisted of the winner allocation stage and payment determination stage. From the bids $\{b_{i}\}_{v_{i}\in V}$, we denote by the unit marginal gain $\hat{P}$

The winner allocation stage selects vehicles from $V$ in a greedy approach. All active vehicles in $V$ are sorted in a non-increasing sequence as

where the $i$-th vehicle has the maximum unit marginal gain $\hat{P}(v_{i}|X_{i-1})$ over $V\backslash X_{i-1}$, $X_{i-1}=\{v_{1},v_{2},\cdots,v_{i-1}\}$, and $X_{0}=\emptyset$. From this sorting, we select the maximum $L$ with $L\leq n$ such that $\hat{P}(v_{L}|X_{L-1})\geq 0$ and $\sum_{v_{i}\in X_{L}}b_{i}\leq B$ where $X_{L}=\{v_{1},v_{2},\cdots,v_{L}\}$.