Hybrid Handover Decision Using Neuro-Fuzzy Logic Approach for Heterogeneous Wireless Networks

This paper decides about the vertical handover which is a significant process in the fifth generation (5G), using neuro-fuzzy logic that works between artificial neural network and fuzzy logic of heterogeneous wireless networks has 3 categories as Wireless Local Area Network (WLAN), Long Term Evolution-Advanced (LTE-A), and Mobile Worldwide Interoperability for Microwave Access (Mobile WiMAX). In addition, determine parameters affecting the handover decision of wireless communication such as received signal strength (RSS), mobile speed, and bandwidth to enter the handover decision process of neuro-fuzzy and the simulation of a structure uses data measured from parameters. The conditions of handover decision are threshold value and dwell time that prevent connection quality and unnecessary handover from decreasing. The purpose of this paper is to make users satisfied with the quality of service (QoS) and reduce the number of handovers and blocked calls. The research result found that the neuro-fuzzy algorithm has better performance when compared with fuzzy logic and back-propagation neural network methods. Consequently, the neuro-fuzzy illustrates the number of handovers and the number of blocked calls on average decrease by 38% and 26% when compared with the back-propagation neural network and decrease by 59% and 36% when compared with the fuzzy logic method, respectively.


INTRODUCTION
Nowadays, telecommunications technology has been developed that is significant in our daily lives, it contributes to pushing various businesses in occurrence and affects the development of the nation.Wireless communication technology has developed continuously into the fifth generation (5G) that improves usability to be more efficient and stable.To reduce problems with system delays or internet crashes because it is used for work that requires high accuracy and fewer errors, this generation focuses on the speed of data transmission.However, the fast development helps to increase users, but the limited channels affect dropped calls or blocked calls.Thus, the procession of changing the communication channel from the active channel in a current cell to the channel of a neighboring cell within the wireless network is very important because this one ensures continuity and good quality for users.As the user moves away from the base station that is used, the received signal strength will gradually decrease.On the other hand, the received signal strength close to the base stations will increase.The system must forward timely communication services before a signal is lost, this processing is called the handover.Changing channels within the same network technology is called the horizontal handover (HH).Conversely, the vertical handover (VH) uses a different network technology.The process is significant for future communication, many researchers have proposed the vertical handover decision.Therefore, the hybrid handover decision using a neuro-fuzzy logic approach for heterogeneous wireless networks is another interesting method for the handover decision.
The author uses dwell time to reduce the burden of handover before entering the neuro-fuzzy process.Dwell time helps reduce the ping-pong effect, which refers to several handovers that happen back and forth between two Access Points (APs).This takes a severe disadvantage on the quality of service (QoS) and the network load.Hence, this paper brings various factors affecting the handover decision as proposed.The number of handovers and the number of blocked calls were reduced, to receive more service quality.

RELATED WORK
The vertical handover decision using the fuzzy logic (FL) method for the fourth generation (4G) that has WLAN and LTE.The input parameters such as RSS, network coverage, bandwidth, and quality of service (QoS) have latency, jitter, and packet loss.The selection of network uses the fuzzy multiple attribute decision making algorithm (FMADM) and finds results from the fuzzy Inference System (FIS) using Mamdani model.So, the experiments showed that using fuzzy logic reduces time and complexity in systems that need to consider many parameters.The algorithm can customize easily to support multiple networks and optimize parameters.However, the disadvantage of the fuzzy logic method is that it does not have a selflearning process for restructuring, which the system is determined by experts [1].
The handover decision process using fuzzy logic (FL) compared with fuzzy rule-based artificial neural network (ANN) in the design and development to support frequent handover in light-fidelity (Li-Fi) and wireless-fidelity (Wi-Fi) networks.Input parameters such as the instantaneous signal to interference noise ratio (SINR), received signal strength, average SINR and user velocity.From the experimental results, an accuracy test of 2 algorithms, the fuzzy logic was more successful in handover than the fuzzy rule-based ANN.However, the disadvantage is that decisions based on fuzzy logic can appropriate decisions based on rules.Meanwhile, the efficiency of fuzzy rule-based ANN was simulated using trained data from the fuzzy rules which have more recognition as a result this algorithm has little error when handover decisions [2].
Using a vertical handover mechanism to select the best position.Network selection affects to high quality of service.It provides a suitable network during handover, users could stay on networks for a long time and reduce the frequency of handovers such as the ping pong effect, that is multiple handoffs between two devices.This has an impact on user quality.The author solves a problem using network selection with QoS to define network rankings of significance for each algorithm [3]- [5].
Therefore, in this paper, the hybrid handover decision is using the neuro-fuzzy method, which is a combination of neural network and fuzzy logic to use the vertical handover decision for communication with the fifth generation of wireless networks more efficiently.

NEURO-FUZZY LOGIC ALGORITHM 3.1 Adaptive neuro-fuzzy inference system (ANFIS)
This model applies an artificial neural network system (ANN) which has advantages in learning but cannot explain the acquisition of decisions.It is working with Fuzzy Logic that qualifies like humanity but is unable to learn automatically.Before entering the process, input is converted by the fuzzy logic and entered the ANN for further processing.Using the rule of Sugeno fuzzy model or Fuzzy If-Then Rules to display data sets and relationship between input and output.The model consists of five layers.Within layer, layer 2, layer 3 and layer 5 are fixed while layer 1 and layer 4 are adaptive.Additionally, the adaptable parameters of the network are learned and improved to reduce the error between actual and desired output [6]- [7].
In Fuzzy If-Then Rules, IF is the premises parameter and THEN is the consequential parameter.For input x, y, fuzzy set A1, A2, B1, B2, defuzzification layer parameter p1, p2, q1, q2, r1, r2 and output f, the rules can be described as In Figure 1 represents the architecture of ANFIS model with two inputs.The procedure of each layer can be expressed as Layer 1: The first layer of the ANFIS model is called the fuzzification layer or input membership function.Nodes in this layer are adaptive nodes.For input , and premise parameter set {, , } that are changed by the triangular function as depicted in (9), so outputs of this layer are the premise parameters.
Layer 2: The second layer is called the fuzzy rule layer.The membership value calculated from the fuzzification layer is used to calculate the firing strength by the sum of membership.The result is called Normalized Firing Strength (W).Nodes in this layer are fixed nodes and each node follows the rule of Sugeno fuzzy model.For the value of membership ,  and the firing strength  , the output of the layer can be calculated as Layer 3: The third layer is called is the normalization layer.Nodes in this layer are fixed nodes, node  represents the proportion of fuzzy rule when compared   with the total fuzzy rule.So, the normalized firing strength is calculated against the sum of rules.For firing strength  1 ,  2 and normalized weight w , the information of this layer can be expressed mathematically as Layer 4: The fourth layer is called the defuzzification layer.In this layer, using a linear polynomial equation to define the weighted values of rules in each node.Nodes in this layer are adaptive nodes.For the normalized weight   , the parameters   ,   and   are adjusted using the recursive least square estimator (RLSE), the output can be determined as Layer 5: The last layer is called the summation layer which has a single node.Combine results from the defuzzification layers to get the output.Nodes in this layer are fixed nodes.The total output can be calculated as

Fuzzy Membership Function
The characteristic of membership function is significant in the design of the fuzzy inference system (FIS).The membership functions have many shapes, which have different shapes such as triangular function, trapezoidal function, membership function, etc.The choice of membership function depends on the data set used.For this project, we chose the triangular membership function.It is a commonly used function that is simple but efficient as it consists of two linear equations, so the membership value can be found from these linear equations.There are three variables {, , } where the lower limit is, the upper limit is , and the value  is  <  <  and the equation of triangular membership function can be represented as [8]  (; , , ) = The alternate equation is shown in the form of min and max. (; , , ) = max min

VERTICAL HANDOVER DECISION WITH NEURO-FUZZY LOGIC APPROACH
In Figure 2a, the handover decision process.Start by checking the user's position.Firstly, the users are within the WLAN area.In this case, check the following conditions.The received signal strength of WLAN (RSS WLAN ) is less than or equal to the received signal strength of WLAN threshold (RSS TH_WLAN ) and the dwell time of WLAN (DT WLAN ) is greater than the dwell time of threshold (DT TH ).If it is true as mentioned above, the handover decision by Neuro-Fuzzy of WLAN is operated and if it is false the users do not handover.
In Figure 3 b, the users under the Mobile WiMAX area check nearly above condition.The received signal strength of Mobile WiMAX (RSS WiMAX ) is less than or equal to the received signal strength of Mobile WiMAX threshold (RSS TH_WiMAX ) and the dwell time of Mobile WiMAX (DT WiMAX ) is greater than the dwell time of threshold (DT TH ).If true, the handover decision by Neuro-Fuzzy of Mobile WiMAX is operated and if it is false the users do not handover.Finally, the users are different from the above cases.LTE-A ) is greater than the dwell time of threshold (DT TH ).If it true, the handover decision by Neuro-Fuzzy of LTE-A is operated and if it false the users do not handover.Finally, when the handover decision by neuro-fuzzy of heterogeneous wireless networks is finished.Check the target of channel is empty and forward the data to handover.On the other hand, the data is blocked and completed.

SIMULATION ANALYSIS 5.1 Simulation environment
The received signal strength for networks can be determined as follow [9] where   is the transmit power in watt (w) and ()is the path loss at distance (d) in meter (m) between a mobile user and a base station.In this paper, the authors use   = 1 watt.In LTE-A, the path loss at distance (d) can be calculated as where  represents the path loss constant,  is the path loss exponent and   denotes the shadow fading effect which is the zeromean Gaussian distributed random variable with standard deviation (dB).In this paper, the authors used S = 19, n = 3.5 and   = 36 dB, respectively.In Mobile WiMAX, the path loss is used the Cost-231 Walfish-Ikagami model [11] that is defined as where  is the operating frequency (MHz) and  is the distance from a mobile user to a base station (m) Finally, received signal strength of WLAN can be calculated as where  intend the environmental factors of transmission which is set to 2.8.

Simulation parameter
The author has determined the range of 3 categories of wireless networks namely WLAN, Mobile WiMAX and LTE-A, that are used to estimate the received signal strength.Randomly selecting the user's position and determining the signal radius from the base station.The authors set the radius of WLAN, Mobile WiMAX and LTE-A networks to be 100 meters, 1,000 meters and 2,000 meters, respectively.The user's position was randomly from a point assigned in the plane of x-axis within 0-5,000 meters and The plane of y-axis within 400-5000 meters, providing a position covering all 3 networks as illustrated in Figure 4.

Simulation performance
The received signal strength in each network is compared.The user node can determine which direction to move within the range set

RESULTS
The number of handovers with neuro-fuzzy is less than using the back-propagation neural network and the fuzzy logic, respectively.The neuro-fuzzy has more recognition process than other methods which is the cause of less error value.This is consistent with result as shown in Figure 5, that the neuro-fuzzy method has a smaller number of handovers than other methods.On the contrary, there are increasing users.Moreover, the number of blocked calls with neuro-fuzzy is less than using the back-propagation neural network and the fuzzy logic, respectively.The neuro-fuzzy assists in reducing blocked calls allowing users to communicate more efficiently and continuously between heterogeneous wireless networks without being congested during handover decisions can be depicted as Figure 6.

CONCLUSION
In this paper, the author presented the vertical handover using neuro-fuzzy that combined the artificial neural network and the fuzzy logic with triangular function for recognition process and error performance is 0.001.Input parameters such as the received signal strength, bandwidth and mobile speed enter handover decision using neuro-fuzzy that improves user communication efficiency and service quality.From research results, it was found that neuro-fuzzy reduces the number of handovers and handover of blocked calls more than other methods.In conclusion, the neurofuzzy logic method is the best when compared with fuzzy logic and back-propagation neural network.Therefore, this proposed process makes users satisfied with the quality of service (QoS) and the best connection at anytime and anywhere.

Figure 5 :
Figure 5: Number of users vs number of handovers.

Figure 6 :
Figure 6: Number of users vs number of blocked calls.

Table 1 :
Simulation Network Parameters

Table 2 :
Received signal strength interval.