Building Trusted Routing with Subjective Logic in Wireless Sensor Networks

Since sensor nodes are usually deployed in a hostile environment, it is easy to be physically captured by an adversary. Thus, an adversary is capable of disturbing the entire network. To overcome it, we propose a trusted routing algorithm using subjective logic. In our solution, every node checks the similarity between itself and its neighbor nodes which are denoted as subjective opinion. On the opinion, the node choose its aggregation node in its neighbor nodes. Furthermore, if one aggregation node is compromised by an adversary, its neighbor nodes report the alarm accompanying with a subjective opinion and the routing is adjusted on the fusion reported opinions. Thus, the compromised nodes are circled. Our discussion and partly implementation show that our scheme provides trusted routing and improves the accuracy of sensing data without heavy energy overhead. Introduction Sensor nodes are not usually physically protected, and they are easy to be compromised by an adversary. With the compromised nodes, an adversary are capable of inserting the false data to subvert the sensor network. For example, wireless sensor network is deployed around the house, and the user are aware of changing stresses with the support of these nodes. However, if an adversary capture several nodes, and send false data to the user, the user fail to obtain the precise sensing data. Furthermore, to save the energy, data aggregating technology are used widely. So it is possible that an adversary tampers a collection of sensing data only capturing one node. Due to the limited cost, it is difficult to develop complex security mechanism for wireless sensor network to defeat this attack. There are some work to defeat the node compromised attack. Reputation-based framework are presented which evaluate every node's trust on their activity[1,2]. Our previous work[3] introduces subjective logic to check the false data sent by compromised node. However, the solutions do not eliminate the false data in wireless sensor network, and does not adjust the routing. In our opinion, due to the limited energy, the false data should not be transmitted because message transmission consume the main energy of the nodes. Moreover inserted false data disturb the user and reduce the accuracy of sensing data. In our opinion, the routing should circle the compromised node to avoid the interference of false data. To overcome the above limitation, in this paper, we present an trusted routing algorithm that is design to expose and circle compromised node, and rebuild the routing without heavy overhead. How to identify the false data? We identify them by cooperation of neighbor nodes with the support of the subjective logic. In other words, every neighbor node poses the opinion that the sensing data is a false data, and the conclusion is drawn on the fused opinion. To circle the compromised node, these neighbor nodes cooperate to choose a new aggregation node. Our discussion and partly implementation show that our scheme provides trusted routing and improves the accuracy of sensing data without heavy energy overhead. The rest of the paper is organized as follows. Section 2 presents the backgrounds of our solution, and section 3 describes our solution. Section 4 presents the detail of our discussion. At last, we concludes this paper. 2nd International Conference on Advances in Mechanical Engineering and Industrial Informatics (AMEII 2016) © 2016. The authors Published by Atlantis Press 50 Background In wireless sensor network, nearby nodes usually report similar data, and it is usually called spatial correlation. This is a solid ground for data aggregation and energy saving. For example, there are ten sensor nodes deployed in the building to obtain the temperature. Normally, their sensing data are similar. So it is not necessary that every node report its data every time. Thus, to save the energy, these nodes may construct a cluster, and only one report data to sink node. On spatial correlation, as a example, [5] develops a clustering-based aggregation algorithm that is still capable of providing accurate data. Spatial correlation can be also used to expose false data. On the above example, if nine node's data are similar, but one's data are sharply different from others, we can draw a conclusion that the node may be compromised and report a false data. Some work detect false data with spatial correlation[3,6]. Our solution also detect compromised node with the same features. In this paper, we use subjective logic to find compromised node. Subjective logic, which is a kind of probabilistic logic, is proposed by audun jØsang[7]. Its main feature is taking uncertainty into account, and it is suitable to model the situation without fully knowledge. For example, it can be used to analysis trust network[8]. Subjective opinion is a main concept of subjective logic. A binomial opinion about the proposition x is denoted as Wx={b,d,u,a} where b is belief that proposition x is true, and d is belief that proposition x is false, and u is the amount of uncertain, and a is the priori probability. For example, a proposition y is describe as follows: tomorrow may be fine. A forecaster hold an opinion Wy={0.5,0.3,0.2,0.5}. On the opinion, we know the forecaster believes that the chance of y is 60% with a formula exp=b+a*u. Fusion operator is a main operator of subjective logic, and we use it to fuse these opinions on the same proposition. For example, Wx={0.5,0.3,0.2,0.5} means user A hold an opinion on proposition x, Wx={0.6,0.1,0.3,0.5} is another opinion hold by user B on the same proposition. Fusing Wx and Wx, we can produce a fusion opinion Wx={0.61,0.25,0.14,0.5} according to the fusion rule of subjective logic. In this paper, we use fusion operator to fuse the neighbor's opinions, and present the expectation value of the fusion opinion. The expectation value is used to detect compromised node. Our solution In this paper, we suppose the sensor network include n sensor nodes and one sink node for simple discussion. However, we believe that it is easy to be extended to other situations. Our solution comprise of several parts: exploring location, selecting aggregation node, completing routing and circling compromised node. We discuss them as follows. Exploring location Every node has to locate the distance to sink node before building the routing. To this end, the sink node flood the location message m0. These nodes, which is deployed near the sink node, are received the message m0. We denoted these nodes as 1-nodes. Like sink node, 1-nodes flood the message m1, and some other nodes can be capture the message m1. Noticed that some 1-nodes also observe the message m1, but they discard the messages. These nodes that receive message m1 are denoted as 2-nodes. Similar to this, all nodes can be denoted as i-node. It is necessary to locate any node in sensor network for build the routing. An k-node has to find a road k-1-node→k-2-node→... →1-node to reach sink node. In some time, an k-node stop to work, for example it exhausts its energy, some k+1-nodes may find other node as their bridge to sink node. If these nodes fail to find any available k-node, it means that sink node receives no more than their sensing data, and we fail to build the routing for the network. On the other side, it is possible that the radio link may break. However, in the paper, we suppose that the links are reliable except that the node exhausts its energy. In this step, all nodes flood one location message. Selecting aggregation node In this paper, we use subjective opinion to score the spatial correlation between two nodes. As mentioned earlier, subjective opinion is a 4-tuple {b,d,u,a}. So the key point is to build the map from spatial correlation to {b,d,u,a}.