Multi Trust-based Secure Trust Model for WSNs

Introduction

Wireless sensor networks (WSNs) are “collections of several small sizes self-organized low-cost resource constraint (memory, power, processor, bandwidth and short communication range) sensor nodes (SNs) which mainly deployed in the hazardous (hostile) area. Sensor nodes (SNs) monitor events, collect continuous and discrete data, and sent (report) to the base station (sink node). WSNs nodes communicate via radio links with limited available bandwidth and form a temporary network, i.e., network without predefined infrastructure and centralized network administration (Boukerche et al., 2007) (Basan et al., 2016) (He et al., 2012). WSN uses a highly dynamic network topology where, any time, sensor nodes can leave and join a network and change their locations. Due to the deployment (distributed, dynamic and collaborative) nature of WSNs, sensor nodes are less reliable, failure-prone and prone to several attacks like blackhole attack, on-off attack, Sybil attack, etc. (Crosby et al., 2006) (Ganeriwal et al., 2008) (Jiang et al., 2015). Once a node is compromised by adversary force having access to cryptographic keys to access other nodes of WSNs, erroneous data routing by malicious (faulty, selfish, bad or spiteful) nodes will breakdown the entire network. In such cases, when the WSNs node itself becomes a malicious node and due to resource constraints or limitation of WSNs nodes, cryptographic techniques cannot prevent from internal attacks (Ishmanov et al., 2015). Thus we need a different kind of robust security mechanism to prevent WSNs from internal and external attacks known as Trust Estimations mechanisms in wireless sensor networks. Trust estimation (TE) methods are used to evaluate the dependability and reliability of sensor nodes for the survival of wireless sensor nodes (Jadidoleslamy et al., 2016). LEACH, EEHC, and EC provide good clustering scheme to enhance the network measurability and throughput. Within a cluster, a cluster with a sufficient number of resources (CPU, power, memory) is selected as a cluster head that monitors the cluster member nodes and states the degree of trustworthiness. Cluster head identifies the malicious nodes within the cluster and selects the reliable route to transmit the data. The altered information about the trustworthy (genuine) SNs of the whole network is maintained in a database. Size of Database and size of the network are directly proportionally to each other. It is unfeasible for a single node (Bin Ma et al., 2009) to store, compute, and monitor the trust values with alteration of the whole network. To reduce the communication and memory overheads, we have scheduled the task using a well-known algorithm (Riaz et al., 2009) and eliminate the unnecessary feedbacks from selfish nodes. In this paper, a specialized dataset for WSN is used (Tan et al., 2015) to detect and mitigate various internal attacks like bad mouthing attack, blackhole, and grey hole attacks. Rest of the work is divided into following sub-sections. Table 1 indicates a comparative analysis of these trust models in terms of various parameters. There are various methods and approaches to model trust, like probability, fuzzy, weighted, Bayesian, game theory, entropy-based, neural network”. Usually, Trust Establishment schemes have been considered as a useful and effective tool to enhance security and dependability (cooperation).

Table 1. Comparative analysis of the various existing state of the art trust models.

The primary objective of the research is to reduce the communication overhead and memory overhead issues in the clustered wireless sensor network. There are several challenges during communication among SNs in WSN, such as data security, congestion, packet loss, and resource management. The proposed trust model is supposed to resolve the issue regarding security using TMS and advanced cryptographic approach. The proposed system would be capable to prevent greyhole attack, bad mouthing attack with the help of analyzing the consistency of trust values. In this research work, we have considered Trust value range as [0, 4] to minimize memory overhead. Moreover, resource management in the proposed trust model has been performed using an optimal task-scheduling algorithm.  It has been observed that nodes usually estimate trust value for other nodes with the help of a well-known mechanism known as a timing-window concept (Khan et al., 2019).

Proposed System

This section discussed a clustered architecture of sensor nodes (SNs) deployed in the targeted area as shown in figure 1.

Figure 1.WSN Topology.

The proposed trust estimation approach is simple, effective and practical since it detects and eliminates the faulty data from the networks. It is attack resilient and trustworthy with less overhead due to task scheduling algorithm (Dai et al., 2011).  Figure 2 shows the flow chart of the proposed work.

Intracluster Trust Evaluation

Trust calculation within a cluster (intracluster) is discussed in the following two subsections, namely CM to CM and CH to CM trust evaluation scheme respectively, to obtain an accurate and robust trust values.

CM to CM direct trust evaluation scheme  

Within a cluster, communication trust and data trust  between CMs is computed using the Eq.(1) as follows.

                                                               (1)

Where   and  (Δt) denote the number of successful interactions and unsuccessful interactions respectively during the time Δt.

Figure 2.  Flow diagram

The first term  provide the ratio of successful interactions to the total interactions and second term   “provide a flexible punishment to the malicious nodes where can be tuned according to application requirement and network scenario. The third term  provides reward to good (genuine) nodes during successful interactions. In this model, punishment and reward can be adjusted according to application requirement takes makes it a novel, flexible and robust trust model. Successful data report between two SNs is estimated using (Khan et al., 2019) which is important to improve the accuracy of proposed” Trust model.

CH to CM feedback (indirect) trust estimation (  

When there are no interactions (successful and unsuccessful) among CMs then Cluster head collects direct trust values of (n-1) cluster members by periodically sending a request packet and store in an (n-1)* (n-1) matrix as follows:

Inspired from the beta distribution function (Li et al., 2013) (Singh et al., 2017), feedback trust can be estimated as follows   using Eq. (2).

                                                                                                             (2)

Where a and b are the amount of positive and negative feedbacks, respectively. A feedbacks is said to be positive if and  and negative if . Final trust value  is computed by simply aggregating eq(1) and eq(2) ( as simple averaging performs better than complex averaging (Ishmanov et al., 2015))  as follows using Eq. (3).

   =                                                                                                        (3)

To find the node status, component is used as follows

The value of (or parameter)  is application dependent i.e. ,its value can be tuned according to application requirements and network scenario.

Intercluster Trust Evaluation

Trust calculation at the intercluster level is also defined by CH-to-CH (direct) and BS to CH (indirect) trust calculation.  CH-to-CH “trust is calculated using Eq. (4) in the same way as at CM level, but during BS to CH trust calculation, BS discards the dishonest feedbacks and simply aggregate remaining feedbacks. Note that at inter-cluster level, we consider only communication trust (not data trust) because adjacent CHs aggregate the data coming from CMS, and it will be challenging to find the false (untrustworthy) data report from the aggregated” data. Figure 3 shows the flowchart of trust evaluation at the intercluster level.

Figure 3. Trust calculation at the intercluster level

CH to CH Direct trust estimation

The direct trust estimation between CH_i and CH_j is defined as follows using Eq. (4).

                             (4)

The first and third terms in Eq.(4) plays the same role as in Eq.(1). The second term is known as punishment factor (wrt. CHs) that give strict punishment with the increase in unsuccessful interactions. As we know that CHs are powerful and trusted node, unsuccessful interactions among CHs should decrease the trust value. We have intentionally  added 1 with  to keep the trust value within the specified range.  Consider the scenario when we have not added 1 with  and suppose the number of unsuccessful interactions between CH (i) and CH(j) is zero then the value of  that is non-realstic because trust value between any two entities never be infinite.

BS to CH feedback Trust calculation

To obtain CHs trust values, Base station periodically sends a request packet to cluster heads (suppose m) in the same fashion as cluster head sends to cluster members. In response to request packet, cluster heads forwards their direct trust values to the base station. To compute feedback trust value, the base station (BS) maintains these values into a matrix as follows

In the matrix B maintained at BS, the values  to   may be untrusthworthy recommeeedation which should be discared for correct trust decision at the base station.  So during trust calculation, BS discard self recommeeedation to reduce the self boosting trust value because a malicious CH may send false trust value about itself to boost self trust value. Inspired by the beta distribution function, feedback trust can be estimated as follows using Eq. (5).

                                                                                                        (5)

Where p is positive feedback and b is negative feedback. A global trust value  can be obtained at CHs as follows using Eq. (6).

.    =  (6)

Where are respective weight-age and depending upon the application requirement, will give more flexibility to select appropriate weight-age for the robust TMS. 

Note: In the proposed model, the term P can be used to represent the percentage of successful interactions as follows using Eq. (7).

.                                                                                                                                     (7)

Where s (or )  is the number of successful communication and f is the number of unsuccessful communication between sensor nodes x and y in one time unit of the time window.  Similarly, the percentage of unsuccessful interactions (U) can be defined as follows using Eq. (8).

                                                                                                                                (8)

Authentication based Data Trust (ADT)

 Let us suppose reward and penalty (in case of ADT) is r_ADT, and Penalty p_ADT respectively then the value of P and U using Eq.(7) and Eq.(8) in terms of reward and punishment can be defined using Eq. (9) and Eq. (10)  as follows

                                                                                                                    (9)

                                                                                                                  (10)

Scheduler based Node Trust (SNT)

Let us suppose reward and penalty (in case of SNT) is r_SNT, and Penalty p_SNT respectively then the value of P and U in terms of reward and punishment can be defined using Eq.(11) and Eq.(12) as follows

(11)

                                                                                                                             (12)

Trust Decision based on Either ADT or SNT

Using Eq. (1), we can write authentication or scheduler based trust (T_AOS) as follows.

T_{AOS     =}  (  + (                                                                                                        (13)

where   and   are the trust values for ADT and SNT computed using proposed eq (1) to obtain robust trust value.

Trust Decision based on ADT and SNT

Using Eq.(1), we can write authentication or scheduler based trust (T_AOS) as follows

T_{AAS   =} (  * (                                                                                                                                                              (14)

Overhead Analysis

This section discussed the comparative analysis of the various state of the art TMS.  Table 2 provides the communication overhead, and Table 3 presents a comparison of the different trust models.

Table 2. Communication Overhead Analysis

Table 3. Comparison of WSN Trust functions

Result

Since “it is already proved in (Khan et al., 2019) that ADCT (Talbi et al., 2017) is better than GTMS (Shaikh et al., 2008), TMA (Zhang et al., 2010), LDTS (Li et al., 2013) in terms of attack mitigation, communication overhead and memory overhead so we are comparing our proposed work with ADCT (Talbi et al., 2017) by considering parameter values in the same proportion for better comparative analysis.”In tradition work, reward and penalty are 0.5.  The value of P was calculated as P=reward / (reward + penalty). Table 4 had been generated according to the parameter (a orα) to analyze the severity of the trust model. Figure 4 is plotting in MATLAB with the corresponding data provided in table 5.

Figure 4. Graph representing trust value in ADCT (Talbi et al., 2017)

Table 4. Variation of trust values wrt. a  (or α).

Case 1. Comparing trust model with (strong penalty) and where (reward and penalty are equal)

In case 1 of the proposed work, the Trust values were calculated using Eq.(1) and Eq. (7). They used 0.4 as reward and 0.6 as a penalty.  The value of P was calculated as P=reward / (reward penalty). Table 4 had been generated according to the different parameter of severity and shows the variation in trust values with the change in α. Moreover, Table 5 indicates the decrease in trust value “which in turn indicate the robustness of trust model because of lesser the trust value with increased no. of interaction, more robust and attack resistant trust model. Note that trust value should not increase rapidly with increased no. of successful” interactions.                                 

Table 5. Trust values generated in case (1)

Similarly, table 6 and table 7 indicate the robustness, accuracy, and attack detection capability of the trust model.

Figure 5: Comparing trust model with strong penalty with existing model (ADCT (Talbi et al., 2017))

We have “compared proposed work with an existing approach ADCT (Talbi et al., 2017) using MATLAB, as shown in fig.5. In ADCT (Talbi et al., 2017), Alpha (α or a) has been considered as a parameter to provide the severity of trust function. The experimental result shows that our approach is far better than the existing method in terms of malicious node detection and persistence of malicious behavior. We have analyzed the consistency of trust values with the change in α value. The consistency line (in figure 5) shows that trust values are changing slowly that indicates the accuracy of the trust model because a good trust model should avoid rapid increase or decrease in trust values.”

Case 2.Integration of Trust Equation for ADT along with SNT

In case (2) of proposed work, the Trust values were calculated using Eq. (14) that has been derived from Eq. (1) and Eq. (9). We have used 0.4 as reward and 0.6 as a penalty. Here authentication mechanism and scheduling (Nasser et al., 2013) both simultaneously used to improve the performance of proposed TMS.

Table 6. Trust values generated in case 2

Fig. 6: Comparing trust model with Integration of Trust Equation for ADT and SNT

Table 6 had been generated in MATLAB according to the different parameter of severity (α). This “scenario represents the effect of Integration of trust equation for ADT along with SNT. Figure 6 provides a comparative analysis of change in trust values. This approach reduces suddenly increase in trust values by various selfish” nodes.

Case 3. Integration of either ADT or SNT.

In case (3) of proposed work, the Trust value was calculated using Eq. (13) that has been derived from Eq.( 1) and Eq.(11). We have used 0.4 as reward and 0.6 as a penalty. Table 7 had been generated in MATLAB according to the different parameter of severity.

Table 7. Trust values generated in case 3

This case represents the effect of using either Integration of Trust Equation for Authentication based Data or Trust along with Scheduler based Node Trust.  Figure 7 depicts the variation in trust values when authentication or scheduling is integrated with the trust model. This approach provides adaptability in the reduction of communication overhead (due to scheduling) or making it attack resilient (due to ADT). The word “or” plays a significant role as it covers either the first case or second case or both.

Fig.7. Comparing the trust model with Integration of Trust Equation for ADT or SNT

Conclusion

The proposed “hybrid trust estimation approach for large scale WSN employing clustering to improve cooperation, trustworthiness, and security by detecting selfish sensor nodes with reduced resource (memory, power) consumption. The proposed scheme consists of unique features like authentication based data trust, scheduler based node trust, and attack resistant by giving the high penalty and minimum reward during node misbehavior. A task scheduling mechanism is employed for scheduling the significant task to reduce computation overhead. The proposed trust model would be capable to provide security against blackhole attack, grey hole attack, and badmouthing attack. Moreover, the proposed trust model feasibility has been tested with MATLAB. Simulation results exhibit the great performance of our proposed approach in terms of trust evaluation cost, prevention, and detection of malicious nodes with the help of analyzing consistency in trust values and communication” overhead.

Conflict of interest

The authors declare no potential conflict of interest regarding the publication of this work. In addition, the ethical issues including plagiarism, informed consent, misconduct, data fabrication and, or falsification, double publication and, or submission, and redundancy have been completely witnessed by the authors.

Funding

 The author(s) received no financial support for the research, authorship, and/or publication of this article.