Type II Fuzzy Logic Based Cluster Head Selection for Wireless Sensor Network

: Wireless Sensor Network (WSN) forms an essential part of IoT. It is embedded in the target environment to observe the physical parameters based on the type of application. Sensor nodes in WSN are constrained by different features such as memory, bandwidth, energy, and its processing capabilities. In WSN, data transmission process consumes the maximum amount of energy than sensing and processing of the sensors. So, diverse clustering and data aggregation techniques are designed to achieve excellent energy efficiency in WSN. In this view, the current research article presents a novel Type II Fuzzy Logic-based Cluster Head selection with Low Complexity Data Aggregation (T2FLCH-LCDA) technique for WSN. The presented model involves a two-stage process such as clustering and data aggregation. Initially, three input parameters such as residual energy, distance to Base Station (BS), and node centrality are used in T2FLCH technique for CH selection and cluster construction. Besides, the LCDA technique which follows Dictionary Based Encoding (DBE) process is used to perform the data aggregation at CHs. Finally, the aggregated data is transmitted to the BS where it achieves energy efficiency. The experimental validation of the T2FLCH-LCDA technique was executed under three different scenarios based on the position of BS. The experimental results revealed that the T2FLCH-LCDA technique achieved maximum energy efficiency, lifetime, Compression Ratio (CR), and power saving than the compared methods.


Introduction
Wireless Sensor Networks (WSN), a kind of wireless network, are composed of several isolated and small sensors placed in a target field. These sensors interact with one another through wireless connections [1]. Sensor is nothing but an inexpensive device termed as mote/node which performs with restricted features based on energy assets and calculation ability. The recent developments in wireless communication technology and microelectronics expanded the application of this technique through several WSN applications. There is a drastic growth experienced in WSN applications in the recent years [2]. WSNs are widely applied in agriculture, industries, environment, defense, healthcare and smart city environment. Some of the instances where WSN is applied include smart grids [3] and health care [4]. The major goal of WSN sensors is to observe few physical parameters in their respective platforms and transmit the values to base station that collects the data for further processing. The last device in WSN is Base Station (BS) sink or gateway.
Base Station (BS) is generally linked to power grid and can perform complex calculations. To be accurate, the power assets in sensors are restricted to WSN due to which it ensures the proper functioning of the network. However, it collects all the necessary data and transmits the same to BS. Thus, the most common problem of WSN is its lifetime. So efforts are taken to raise the lifetime of network to avoid the battery of nodes from getting wasted, while attaining unwanted processes. Clustering is one of the most beneficial methods to increase the lifetime of networks, while conventional routing is a better candidate for huge networks. Among a bunch of nodes, a few of them are selected as Cluster Heads (CH) [5]. CHs collect the information from related nodes which are otherwise cited as 'contributing nodes'. The nodes that transmit their measurements to CH are generally present nearby the CH. Later, all the CHs aggregate the data from its sets and transmits the data to BS. This method prevents every node from getting exhausted. This is because the transmission does not occurs directly to BS as this process reduces their battery lifetime quickly, due to nonlinear dependency of the power loss with distance.
WSN consumes less energy for calculation than data transmission. Instead of transferring the sensed data to sink node separately every time, the data is initially gathered and combined by aggregate functions like avg( ), sum( ), and so on. Then, it is sent to sink node where several energies can be stored. The efficiency of the transmission between nodes is decided based on data aggregate method. Data aggregation is an essential process to reduce the consumption of energy and store the constrained assets. An efficient data aggregation method can improve energy effectiveness and lifetime of the network [6].
The current research article presents a novel Type-II fuzzy Logic-Based Cluster Head selection with Low Complexity Data Aggregation (T2FLCH-LCDA) technique for WSN. The presented model includes a two-stage process such as clustering and data aggregation. At first, three input parameters namely, Residual Energy (RE), distance to base station (DBS), and node centrality (NC) are used for CH selection and cluster construction in T2FLCH technique. In addition, the LCDA technique encompasses a Dictionary Based Encoding (DBE) process to achieve data aggregation at the CHs. At last, the aggregated data is transmitted to BS where it achieves the energy efficiency. The proposed T2FLCH-LCDA technique was simulated under three dissimilar scenarios based on the position of BS and the results are present in upcoming sections.

Literature Review
Clustering process can be executed through Artificial Intelligence (AI). Mainly, Fuzzy Rule Based Systems (FRBS) emphasize the clustering process as an appropriate method to decide the nodes that may act as the CH. Such system is applied in different fields to find the right candidate, whenever there is a need. For clustering, Moorthi et al. [7] defined a particle optimization method depending on LEACH protocol. Agrawal et al. [8] utilized an FL Type-1 distributed method with two outputs. In this method, the initial outcomes determine the sent radius of the transmitted message whereas the next one decides whether a node would become a CH or not. Cluster head Election using FL (CHEF) is one of the primary centralized techniques which utilizes the expert system [9]. The BS implements a FL Type-1 process to select the node that acts as CH in every event. In Zhang et al. [10], the BS chooses the optimum node depending upon fuzzy type-2 method.
In Heinzelman et al. [11], BS obtains the data about location and energy of the nodes. Later, the base station follows simulated annealing method to decide the node that can organize itself as a CH. Thangaramya et al. [12] defined a centralized technique which followed an extremely complicated method. Particularly, Convolutional Neural Network (CNN) was used to determine the optimum CH. Shivappa et al. [13] introduced a centralized method in which the BS utilizes AI as FCM to define the optimum position center for all the clusters. Merabtine et al. [14] demonstrated a centralized technique in which PSO was utilized for selecting the optimum CH. Abidoye et al. [15] emphasized the importance of IoT in WSN. This energy efficient method was proposed for allowing a service-oriented application in IoT-enabled WSN field under two phases. In initial phase, clustering-based method is utilized to serve the application whereas in the subsequent phase, an energy aware method was implemented. These methods were predominantly better though it remained insufficient for IoT networks. Further, their efficiency was poor in case of static networks. Li et al. [16] presented a comprehensive research on IoT networks in which the researchers introduced Analytical Hierarchy Procedure (AHP) and fuzzy-based energy management system for the organization of industrial tools. In Kasana et al. [17], fuzzy-based vehicular physical methods were investigated upon Internet of Vehicles (IoV) utilizing fuzzy architecture with Markov chain in order to optimize the location-oriented channel access delay.
Hu et al. [18] proposed an additional feature of IoV that allows transmission to occur at the edge using FL. Clustering Head or gateways were selected with the help of few parameters such as antenna height, fuzzy variable velocity, and vehicle neighboring density. Kaiwartya et al. [19] proposed a genetic virtualization method as a technique to overcome the torrent delay and minimize the energy consumption in IoT-allowed sensor networks. Kaiwartya et al. [20] presented a method which was utilized for appropriate placement of nodes to connect and the method was used for agricultural purposes. There are two techniques followed to place the nodes depending on seven metrics which calculate the quality measurement of sensor nodes. Delgado et al. [21] introduced a distributed method to create a dynamic cluster. However, it is rarely supported by the BS. Particularly, the BS sends three messages in network lifetime to reconfigure the skipped value of network.

The Proposed T2FLCH-LCDA Technique
The workflow involved in the proposed T2FLCH-LCDA technique is shown in Fig. 1. According to the figure, the nodes are randomly deployed in target region. Then, the initialization of nodes occurs after which CH selection process is executed by following T2FLCH technique. Once the CHs are selected, the nodes nearby the CHs join the CHs and form clusters. Then, the CM transmits the data to CH which then performs data aggregation using LCDA technique. Finally, the aggregated data is transmitted to the BS.

Figure 1:
The working process of T2FLCH-LCDA model

System Model
Assume a network of N nodes that are randomly deployed in the target area to sense the physical parameters in environment. Here, the clustering process occurs according to T2FLCH technique. Every cluster has only one CH that receives the data from its respective CMs. The nodes are static and homogeneous in nature i.e., possess identical initial energy and are able to sense the region, process the sensed data, and transmit it. The radio links present among the nodes are symmetric in nature. This denotes that the node necessitates an identical amount of energy to transmit data in all the directions [22]. BS is installed at three possible locations in the target field. The sensors are able to modify the transmission power based on the distance between the recipient nodes. Packet size is deemed to be m bits. The entire energy spent during the transmission of a packet of m bits over l meter distance between two nodes can be derived using the following Eq. (1): The energy spent on receiving a packet of m bits from the transmitted node is given herewith.
where E elect signifies the data with regards to energy dissipation that can be influenced by different factors such as bit rate, modulation, etc. In addition, ε fsp and ε mpf represent the energy utilization factors in free space path and multipath fading correspondingly. If two sensors are detached from the limit of threshold value l o l o = E fsp /E mp , it utilizes the free space model. In other terms, multipath fading channel is applied to determine the amount of energy utilized during transmission.

T2FLCH Based Clustering Technique
The presented T2FLCH-based clustering technique involves three input parameters to select the CHs in WSN. Residual Energy (RE), Node Centrality (NC), and DBS are the input variables in the proposed T2FLCH-LCDA model. Residual Energy is an essential feature to signify a node as either CH or not, since CH nodes use further energy to become a member node. CH nodes gather information from the members, collates the gathered information, and transmits it to BS. In Node Centrality (NC), the total number of one-hop neighboring nodes in R c of the node is named after Node Degree (ND). NC is a feature that defines the node when it is placed in the middle of its neighbors. A node has higher probabilities of getting chosen as a CH under less NC value.
In Eq. (3), ND relates to the counting of neighbors in transmission radius R c of node and Ntk-Dimension value is 'M' in M × M field region and dist 2 i implies the distance through ith adjacent nodes. The amount of energy, utilized in the transmission of data increases, when the distance between transmitter as well as receiver nodes increases [23]. From energy conservation perspective, the distance between CH as well as BS is to be minimized.
where d i refers to the distance between nodes i and the BS, d max represents the maximal distance between a node in the network and BS whereas α implies the network dimensional particular constant. T2FLCH-LCDA takes 1 output variable chance. It can be a crisp output value depends upon the fact that the CH ability of node gets resolved. The superior chance value signifies the superior chance of a node being chosen as a CH. Type-2 Mamdani system is employed in T2FLCH technique. It is composed of two components namely Fuzzifier, Rules, Inference Engine and Output (Type reducer and Defuzzifier). The general block diagram of T2FLCH technique is shown in Fig. 2 which has three input parameters that decides whether a node can become a CH or not.

Figure 2:
The structure of T2FLCH model

Fuzzification Process
The process of fuzzification is to map every individual crisp input value to a fuzzy set. In T2FLCH technique, the fuzzifier element performs the mapping of crisp input vector x = (x 1 , . . . , x p ) to Type 2 fuzzy setÃ x . To simplify the computation, singleton fuzzification model is employed, i.e., for ∀i = 1, . . . , p, the membership function of the fuzzified input setX i is represented herewith.
and μÃ . . . , x p ), and signify the juncture function of type-2 fuzzy sets. The three input values come under a uniform range of [0, 1]. To minimize the computation complexity, three inputs are partitioned into three levels namely, low, medium, and high. Every secondary Membership Function (MFs) is engaged to an interval set, whereas the primary membership remains triangular or trapezoidal [24]. The output variable denotes the chance of becoming CH which can be partitioned into six stages such as very weak, weak, little weak, lower medium, medium and higher medium [25].

Fuzzy Rules
The proposed T2FLCH technique generates a set of 27 IF-THEN rules. The rules are generally represented as given herewith.
whereF l i (i = 1, 2, 3) denotes the type-2 antecedent fuzzy set,G l j (j = 1, 2) represents the type-2 consequent fuzzy set. The node with maximum energy, minimum DBS, and high node degrees have the likelihood of becoming CH.

Inference and Output Processing
This technique utilizes the Center Of Sets (COS) type-reducer that integrates the task of typereducing and inference. Particularly, for jth (j = 1, 2) type-2 fuzzy output set, COS type-reducer substitutes all type-2 subsequent set,G l j , with its centroid, CGl The jth (j = 1, 2) extend output Y j cos (x) of T2MFLS is given as: where Y j cos (x) (j = 1, 2) represents the interval type-1 set defined by two endpoints, y jr (x), and y jl (x) and y l jl , y l jr (l = 1, . . . , 27) corresponding to the centroid of type-2 interval subsequent set G l j . Furthermore, it neglects x in firing interval f l (x), f −l (x) for simple indication. Lastly, it defuzzifies the jth extended output Y j cos (x) by an average of y jr (x) and y jl (x), the jth defuzzified output of this interval T2FLCH method is as given below.

Cluster Head Selection Algorithm
Next to T2FLCH process, all alive nodes have their probability, s i . Probability denotes that these nodes can be a candidate CH. The node with high probability is defined as the CH. When a node becomes a CH, then the nodes with competitive radius further cannot be chosen as CH. With the mortality of nodes, the number of nodes that are alive keeps on varying. While the optimum cluster number also varies accordingly. In this method, it selects k opt CH in sequence, where k opt denotes the optimum cluster number for all rounds which is estimated as given herewith.
where n represents the node counts whereas the 'dead' indicates the number of energy-exhausted nodes and P denotes the CH probability.

DBE Based Data Aggregation Technique
After the clusters are constructed, every CH executes the DBE technique to aggregate the data received from its CMs. The proposed DBE model is a bit-oriented, dictionary-based single character encoding model that utilizes Deterministic Code Allocation Dictionary (DBED) for codeword allocation to the input data. It allocates a predefined length codeword for all the characters in input data using DBED. A peculiar characteristic of DBE model is its utilization of 4-bit codeword to all possible characters that exist in the input data. When considering an input data of length N, the DBE technique desires the lowest number of C bits for storing the aggregated data which is determined as given below. (11) where N DBE denotes the bit count that exists in codewords. In most of the cases, the bit count required for storing a character is equivalent to 4. Then, the bit count which is required on an average for storing an individual character in DBE model is given below.
Primarily, the DBE model keeps DBED which holds a set of codewords for 12 possible characters (0-9 numerals, ',' and '*') that exist in WSN. It significantly assists in the reduction of complexity of the presented technique. Besides, astreik (*) is employed as a delimiting element and is employed in the detection of endpoint of every sample. DBED encompasses a static 4-bit codeword for 12 characters which eliminates the necessity of delimiting elements for characters. DBED is kept at fixed value at both transmitter and receiver sides apriori. Upon receiving the input data, the DBE model utilizes DBED and allots respective codewords. Next to the assignment of codewords, merging process occurs to generate a compressed file. At last, the resulting optimum codewords of the encoded characters undergo concatenation to generate the compressed file which includes only 50% of its actual size. This is also communicated to the receiving end. Fig. 3 illustrates the existing codewords in DBED.

Performance Validation
This section validates the efficiency of the proposed T2FLCH-LCDA model against existing methods. The results were examined under three distinct scenarios and on the basis of location of the BS. BS is located at the center of WSN in first scenario whereas it is exterior to WSN in the second scenario, though not very far from the monitored area. In third scenario, the BS is positioned far away from the WSN. In order to detect the aggregation performance of LCDA technique, a set of simulations was performed on benchmark WSN dataset. Fig. 4 depicts the results from energy consumption analysis of the proposed T2FLCH-LCDA model against existing methods under scenario 1. The figure demonstrates that the LEACH model consumed the maximum amount of energy than other methods. Followed by, the EEUC model consumed a slightly low energy over LEACH, but not lower than other methods. Next to that, PEGASIS model accomplished a moderate energy consumption per round which remained uniform over several rounds of execution. Then, CRT1FLACO and CRT2FLACO techniques demonstrated closer and competitive energy consumption outcomes. But the presented T2FLCH-LCDA technique consumed the least amount of energy, compared to existing techniques, over several rounds of execution. Followed by, EEUC model consumed somewhat lesser energy than LEACH, but not lower than other methods. Next to that, PEGASIS model consumed a moderate amount of energy per round which remained uniform over several rounds of execution. Afterward, CRT1FLACO and CRT2FLACO models demonstrated closer and competitive energy consumption outcomes. But the presented T2FLCH-LCDA approach consumed the least amount of energy, compared to existing methods, over several rounds of execution. that the LEACH model consumed the maximum amount of energy compared to all other methods. Then, EEUC model reduced its energy consumption over LEACH model, but not less than other methods. Next to that, the PEGASIS method accomplished a uniform and moderate energy consumption across several rounds of execution. Then, CRT1FLACO and CRT2FLACO approaches demonstrated closer and competitive energy consumption outcomes. But the proposed T2FLCH-LCDA technique exhibited the least energy consumption, compared to other models, over several rounds of execution.

Network Lifetime Analysis
Tab. 1 shows the network lifetime analysis results for the proposed T2FLCH-LCDA model in terms of FND, HND, and LND.

Compression Ratio Analysis
Tab.

Power Saving Analysis
The results of the detailed power saving analysis of LCDA model with existing data aggregation techniques are shown in Fig. 10.  Simultaneously, on test FN 101 RH dataset, a maximum power saving of 85.23% was attained by the proposed LCDA technique, whereas the LCE, S-LZW, ALDC, and FELACS techniques saved less power such as 49.46%, 15.04%, 55.11%, and 56.27% correspondingly. For test LG 20 RH dataset, a superior power saving of 83.82% was reached by LCDA approach, whereas the LCE, S-LZW, ALDC, and FELACS techniques accomplished low power saving such as 30.81%, 12.57%, 37.16%, and 38.47% correspondingly. The experimental values inferred that the T2FLCH-LCDA technique achieved maximum energy efficiency, lifetime, CR, and power saving over the compared methods. The proposed model improves energy efficiency than the methods compared due to the inclusion of effective clustering technique and data aggregation approach.

Conclusion
The current research article presented a novel T2FLCH-LCDA technique for WSN. Primarily, the nodes are randomly used in the target region. Then, the initialization of the nodes occurs after which CH selection process is executed using T2FLCH technique. T2FLCH technique has three input parameters namely, RE, DBS, and NC which are used in CH selection and cluster construction processes. Once the CHs are selected, the nearby nodes join the CHs and form clusters. Then, the CM transmits the data to CH which performs the data aggregation process using LCDA technique. This technique is loaded with DBE process to perform data aggregation at CHs. Finally, the aggregated data is passed on to BS. The T2FLCH-LCDA technique was simulated under three dissimilar scenarios based on the position of BS. The experimental values inferred that the T2FLCH-LCDA technique achieved maximum energy efficiency, lifetime, CR, and power saving over the compared methods.