We put forward this approach for the sake of an improvement in the performance of routing protocol and to enhance its security. As an undebatable topic, relay nodes have the most role in this action. Thus, it is logical to consider an entity-centric trust model for hinted goals. On the other hand, researchers believe that data has a pivotal role in VANETs since it makes sense to focus on data-centric trust models. According to what was discussed before, we intend to adapt our model to what exists in the social sciences and seek help from it. In a rational society, we expect the right behavior from people with higher social and moral status. Although, it can be claimed that inherently many humans show the right behavior as long as their own interests are not at stake. Hence, to make way for trust determination based on the social sciences, the importance of exchanged data must be taken into account. That is, to ensure the acceptability of the received message, we say: “Who said and what he said”. Because, that person may not be in a position to be aware of such an important message, which from the point of view of computer sciences also indicates which kind of node and what kind of data are in the network. For this reason, node and data in the network are considered weighted, which implicitly states that in vehicular networks, the basis of reliable data transmission, is reliable nodes, and these two are necessary and bound to each other.
4.1 Network model based on weight
Like the majority of methods that focus on trust computing based on existing entities, we calculate this amount of trust in accordance with the first-handed exchanges between two nodes and the suggestion and opinions of other nodes, such as neighboring nodes. Here, we do focus on the weight of data and nodes. Events have distinct effects on transportation and avenue protection, and also dissimilar trustworthiness levels are requisite. We divide them into three groups as follows:
• Data that has safety applications and is effective in enhancing public safety and saving lives. Such as announcing car accidents, announcing blind spots on the roads, thick fog, etc. This type of data has the highest degree of sensitivity and we assign weight of 1 to them
• Data that can be used to improve efficiency. Such as traffic control, parking information, closed streets, gas station, etc. which are less important than the first part, although they are very valuable and we assign a weight of 0.8 to them
• Data use for infotainment. Such as data related to advertising or exchanging entertaining information such as music, restaurants, etc. We assign a weight of 0.5 to them, because of the least importance
Also, nodes in vehicular networks contain a huge range of vehicles that have different roles, so according to their authenticity in the network and experimental results, we divide them into three groups as follows:
• Nodes that have the highest level of authenticity (High Level). Such as police cars and roadside units controlled by centers. we assign a weight equal to 1 to them
• Nodes that have the second level of authenticity (Medium Level). Such as vehicles that perform public transportation or road maintenance vehicles, etc. which can be controlled by certain centers. We assign a weight equal to 0.7 to them
• Nodes that have the lowest level (Low Level). Such as personal cars and freight cars. We assign a weight equal to 0.5 to them
4.2 Trust computing algorithm
The amount of direct trust, recommendation trust from third parties, and also their combined values are used as three meaningful parameters to determine the trust level of each node. Afterwards, we can identify the malicious node based on a threshold value, and remove those nodes from the network. Most articles consider first-hand experience as the most important data. We should also mention that such information is not always available, so it is necessary to rely on second-hand information through testimonies from witness nodes.
In Table 1 some notations are listed to illustrate the proposed scheme (node A is sender and node B is receiver).
Table 1
Notations | Meaning |
\({N}_{A}^{B}\) | All messages that the sender pled the receiver to forward |
\({M}_{A}^{B}\) | All messages that the receiver has forwarded for the sender |
\({W}_{D}^{x}\) | The weight of data (x), (in detail, part 4.1) |
\({W}_{N}^{A}\) | The weight of the sender, (in detail, part 4.1) |
\({U}_{W}^{A.B}\) | Total weight of data that the sender pled the receiver to forward |
\({S}_{W}^{A.B}\) | Total weight of data that the receiver has successfully forwarded for the sender |
\({E}_{TW}^{A.B}\) | The average weight of all data that the sender pled the receiver to forward |
\({E}_{SW}^{A.B}\) | The average weight of all data that the receiver has successfully forwarded for the sender |
\({F}_{W}^{B}\) | The malicious tendency of the receiver, (part 4.2 & Eq. (1)) |
\(D{T}_{A}^{B}\) | The direct trust value of sender-receiver |
\(R{T}_{A}^{B}\) | The recommended trust value of sender to receiver |
\({T}_{A}^{B}\) | The comprehensive trust value of sender to receiver |
Direct Trust. This concept expresses the expectation that one node has of another node’s future behavior. Since, due to the dynamic behavior of vehicular networks, these nodes may not have communicated with each other in the past, initial direct trust values are usually considered 0.5 in most articles.
Here the value of direct trust is considered equal to the node’s weight to imply the importance of the vehicle’s position, also, received information from the center, is more reliable than normal vehicles, because of more control centers that monitor their actions. For this part, as an assumption, node A tries to compute node B’s direct trust. Thus, first and foremost, a malicious tendency is calculated by Eq. (1) which defines the nature of B as a member of society. Based on experimental results that elucidate the portion of miscellaneous data type in traffic models. Thus, 0.72 is considered as a threshold, in other words, having a malicious tendency more than this value clarifies the malevolent essence of the receiver node and vice versa.
$${F}_{W}^{B}=({U}_{W}^{A.B}-{S}_{W}^{A.B})({N}_{A}^{B}-{M}_{A}^{B})$$
1
Therefore, assuming that node A has requested to send message x from node B, taking into account the reaction of this receiver, direct trust can be calculated as Eq. (2) as below:
$$D{T}_{A}^{B}=\left\{\begin{array}{c}\frac{{W}_{D}^{x}.(\frac{(flag+1)}{2}-D{T}_{A}^{B})}{1+\frac{{E}_{TW}^{A.B}}{{E}_{SW}^{A.B}}}+D{T}_{A}^{B};{ F}_{W}^{B}<0.72\\ \frac{{W}_{D}^{x}.(\frac{(flag+1)}{2}-D{T}_{A}^{B})}{4}+D{T}_{A}^{B}; {F}_{W}^{B}\ge 0.72 \& (flag=1)\\ {W}_{D}^{x}.\left(\frac{\left(flag+1\right)}{2}-D{T}_{A}^{B}\right)+D{T}_{A}^{B}; {F}_{W}^{B}\ge 0.72 \& (flag=-1)\end{array}\right.$$
2
If the receiver sends the message successfully, the flag in Eq. (2) is equal to 1, otherwise, is equal to -1.
Recommendation Trust. The recommendation trust is taken from other nodes or even central units along the road so that the agent can express their opinion about the subject with others’ suggestions. Usually, this happens when in direct contact, the obtained value is not quite close to the upper or lower limit, and it is not possible to comment clearly on a node (subject). Therefore, here in Eq. (3), node A as the sender of message x to node B takes help from neighboring nodes to give its final vote.
$$R{T}_{A}^{B}=\frac{{\sum }_{i=1}^{n}D{T}_{A}^{{N}_{i}}.{T}_{{N}_{i}}^{B}.{W}_{N}^{{N}_{i}}}{{\sum }_{i=1}^{n}D{T}_{A}^{{N}_{i}}}; {N}_{i}\ne B;{N}_{i} \text{i}\text{s} \text{}\text{}\text{}\text{}\text{t}\text{h}\text{e} {i}^{{\prime }}\text{t}\text{h}\text{}\text{}\text{} \text{n}\text{e}\text{i}\text{g}\text{h}\text{b}\text{o}\text{r}\text{}\text{}\text{}\text{} \text{o}\text{f} \text{}\text{}\text{}\text{s}\text{e}\text{n}\text{d}\text{e}\text{r}.$$
3
Comprehensive trust & Fuzzy based trust assessment approach. The total amount of trust, in other words, how to combine the values of direct and recommendation trust in the previous sections is vital. How and in what pattern should combined these two values to achieve the best result? Most articles, such as [11] usually use one coefficient, or even several coefficients such as [19], which is a number in the range of zero to one, and try to create the best creation mode in the form of dynamic coefficients. For example, when one node directly has sufficient knowledge of another node, it tries to reduce the effect of the proposed trust value to a slight or even close to zero, and vice versa. But such a performance in the few articles discussed did not lead us to the desired results. So, it enforced us to move on to another model. We will make use of fuzzy logic to combine hinted values.
Here the proposed fuzzy-based paradigm is described. Indeed, the use of fuzzy logic, whose philosophy is to combine vague inputs to achieve crisp output [19–21]. By using three different fuzzy models for different groups of vehicles, we judge all these vehicles’ behavior, regardless of their weight, with the same threshold value for the desired results. Thus, each group uses the fuzzy logic model according to next stages: (1) a fuzzy set is created by a fuzzifier as a transmutational action; (2) layout fuzzy IF-THEN rules; (3) the credibility level is notched up for each node by an amalgamation of the fuzzy inference engine and IF-THEN rules; (4) the fuzzy trustworthiness output is transformed to a real value of trust by a defuzzifier [20, 24]. Therefore, for each group of vehicles, using fuzzy logic, according to Direct Trust Level (DTL) and Recommendation Trust Level (RTL), and considering Table 1, in VANETs, we can identify malicious nodes, in black-hole attack (in Table 1; L is Low, M is Medium, H is High, and TTL is Total Trust Level).
Table 2
Obtain total trust through the fuzzy inference engine.
Rule no. | DTL | RTL | TTL |
1 | L | L | L |
2 | L | M | L |
3 | L | H | L |
4 | M | L | L |
5 | M | M | M |
6 | M | H | H |
7 | H | L | H |
8 | H | M | H |
9 | H | H | H |
We use a Min-Max inference and also a defined fuzzy set to obtain the correlation among these columns. Finally, a fuzzy domain is converted to precise domain using a centroid method for the defuzzification aim. Thus, in three sections we will express it in a completely separate and understandable way:
Case 1
Input fuzzy set membership functions for the vehicles that receive the message and have the highest weight values equal to 1 (Fig. 1 and Fig. 2). Also, excerpts from the node trust evaluation process, taken from MATLAB software to better understand the differences between positions (Fig. 3)
Case 2
Input fuzzy set membership functions for the vehicles that receive the message and have the weight values equal to 0.7 (Fig. 4 and Fig. 5). Also, excerpts from the node trust evaluation process, taken from MATLAB software to better understand the differences between positions (Fig. 6)
Case 3
Input fuzzy set membership functions for the vehicles that receive the message and have the weight values equal to 0.5 (Fig. 7 and Fig. 8). Also, excerpts from the node trust evaluation process, taken from MATLAB software to better understand the differences between positions (Fig. 9)
In this algorithm, DTL and RTL, are two inputs in Table 1, and each of them includes three fuzzy sets, that are defined in the table. Therefore, according to these inputs, the table of rules is formed with 9 so-called IF-THEN rules, so that we can express the Total Trust level (TTL). It should be noted that in Table 1, the importance of direct communications between two nodes is quite clear. Because, when a node has a definite opinion about the behavior of the other node, there is no need to consult other nodes, and waste the time.