A trust enhanced secure clustering framework for wireless ad hoc networks

Abstract

Secure clustering in Wireless Ad Hoc Networks is a very important issue. Traditional cryptographic solution is useless against threats from internal compromised nodes. In light of this, we propose a novel distributed secure trust aware clustering protocol that provides secure solution for data delivery. A trust model is proposed that computes the trust of a node using self and recommendation evidences of its one-hop neighbors. Therefore, it is lightweight in terms of computational and communication requirements, yet powerful in terms of flexibility in managing trust. In addition, the proposed clustering protocol organizes the network into one-hop disjoint clusters and elects the most qualified, trustworthy node as a Clusterhead. This election is done by an authenticated voting scheme using parallel multiple signatures. Analysis of the protocol shows that it is more efficient and secure compared to similar existing schemes. Simulation results show that proposed protocol outperforms the popular ECS, CBRP and CBTRP in terms of throughput and packet delivery ratio with a reasonable communication overhead and latency in presence of malicious nodes.

Appendix: Proof of accuracy of the proposed trust model

In this section we present some important features as well as necessary and sufficient properties that a distributed trust model must possess. In addition to that we show our proposed trust model holds and satisfies all these conditions and performs well in different networking scenarios.

Claim 1

Each legitimate CH must eventually hold the replica of latest trust value of its MNs.

Proof

In our proposed scheme, each MN sends its public key to CH to join in the network at the time of initialization. Trust calculation and propagation is taken care by the CH itself and this trust information is sent periodically. Thus, the trust value of each node is periodically updated which in turn determines the reputation of the node. But due to node failure or mobility or reorganization of clusters, MNs may not get trust information of the CH. All living and legitimate CHs of the network hold the most recently updated trust information about its MNs to continue uninterrupted service. Within each cluster, CH maintains the trust table for all its MN and records all the updates accordingly. This eventually ensures propagation of correct trust information between the CHs. If any good node fails to communicate with its CH, and after some time it revives the connection, the trust value would be automatically assigned to that node from previous record.

Claim 2

Trust status of a node is uniform/same throughout the cluster and network at any particular time.

Proof

In our proposed scheme, at the time of trust generation phase each MN, GN and CH monitors the traffic of its neighbor nodes choosing some packets in a random interval of time and calculates the direct trust. CH accumulates all trust information from the neighbors of review node. Then CH combines all recommendation and finally determines the resultant trust. Therefore, CH of a cluster aggregates the trust information and finally decides and propagates the trust status throughout the network. Trust of a node is uniform throughout the network, i.e., all nodes have to trust a node by the degree decided by the CH.

Claim 3

The reputation of node can be directly acquired by the any other node without network wide flooding.

Proof

The direct trust of a node M is calculated using the local information and the local behavior monitoring of any node in the radio range. The periodic trust calculation and propagation procedure (by the CHs) ensure that the reputation of any member node can be easily detected by checking the trust table (stored in the CH). Any other member nodes while trying to communicate or get information about its neighbor (in the same cluster) or any other nodes in the network, it has to query to the associated CH only. This makes the distributed trust computation procedure flexible and efficient and thus completes the proof.

Claim 4

A node’s locally stored trust value is readable to the CH but confidential to any node including the node itself.

Proof

Our scheme holds this property. This is one of the important security measures. Any node cannot have access on its own trust value. After the final calculation of resultant trust CH securely propagates the Trust_Cert and Available-Neighbor list to its MNs digitally signing the message. The trust value is stored in CH only. So there is no way for any malicious node to advertise itself as trusted and disrupt the network functionality.

Claim 5

A node’s unauthorized modification to its local trust and reputation information can be identified and prevented.

Proof

This claim can be corroborated using Claim 4. A malicious node cannot join to two different clusters simultaneously. A node is permitted to join not more than one CH at a time. Suppose, a node tries to join in two clusters simultaneously, then it would get two different IP and ID from two different CHs. If this type of maliciousness cannot be detected, secret information may easily be leaked between two clusters. Our protocol guarantees detection of such malicious activity. If any GW finds a node initiates different communications with different ID or IP address, GW reports to the CH about the malicious node and subsequently the node is removed. If such node exists, GW detects, and block this node; restores previous local trust value. Thus the unauthorized modification to local trust and reputation are identified and can be prevented.

Claim 6

The model ensures better accuracy in trust calculation and its secure propagation.

Proof

The proposed model does not rely only on the local monitoring method or observation of any individual node. Firstly, direct trust of a node N, is calculated by all neighboring nodes depending on the behavior of N. Here, the behaviors (both good and bad) of N is parameterized by some metrics and corresponding weight. Once, direct trust are calculated by each neighboring node, these trust values are combined (based on appropriate weights) to formulate the resultant trust values using the weighted DS-theory presented in the Eq. 2. The uniqueness/major strength of this weighted trust calculation scheme is in combining the evidences from different nodes with uneven (unequal) priority distribution. Thus neighboring nodes of N modify the trust values depending on weights, which are again computed depending on environment parameters. This ensures more accuracy in the proposed trust calculation scheme. On the other hand, after final trust calculation for the member nodes of any cluster, only a digitally signed trust certificate is propagated. Thus, the proof completes.

Claim 7

Weight based trust calculation scheme ensures lower impact by malicious node on trust calculation.

Proof

In the proposed protocol, Clusterhead (CH) calculates direct trust of a node by directly monitoring the node. Subsequently it collects the trust evidences (i.e., Recommendation Trusts) from different neighbor nodes of the node under review. Finally, in order to calculate the resultant trust of the node under review, CH combines its direct trust and all the recommendation trusts using proposed DS-theory of combining evidences. In traditional DS-theory of combination, as all beliefs are given equal priority, the impact of recommendation from a malicious node which reports its belief about another node under review may be higher. As a result of this, a good node may be treated as bad one. However in our proposed proposal, we have eliminated such ambiguity. Here may be two cases: in the first case, a node is already detected as malicious (blacklisted) and in the second case, the node behaves maliciously but not marked as blacklisted. In the first case, the proposed protocol does not consider the recommendation from malicious nodes while CH combines the trust evidences. In the second case, the proposed weight-based DS-theory assigns weight to the evidences from different nodes before combining them. This weight is determined by the past behavior of the recommender node depending with environment parameters. The highest weight (say 50 %) is assigned to the belief generated by the CH itself to quarantine the impact of the malicious node. If the recommender node has higher trust value, it is less likely to report with wrong recommendation. So, our trust calculation scheme guarantees lower impact of evidence by any malicious node on the final trust of a node under review.