Journal of Communications and Information Networks

Heterogeneous networks provide ubiquitous services tomobile users. With the rapid growth in research and development of wireless technologies, acceleration of the in tegrated services of various technologies offers services anytime and anywhere. A 4G network provides a seamless system with high data rates, various service opportunities, and possible application to bandwidth availability. Mobile roaming with a wide range of access network technologies, including WLAN, WiMAX, satellite networks, and cellular networks, creates challenges in heterogeneous networks. The main objective of these networks is a seamless connection to other networks irrespective of time, location, and services. This process of change in the communication channel is performed by handover, which is an essential element in wireless networks for mobility management. Vertical handover is defined as the establishment of a seamless connection to a different network through interfaces. Handover may occur from a mobile terminal or can be initiated through a network, which can be classified into mobile-and network-controlled handover. Mobility management schemes^[1] initiate handover with various parameters, which helps in better system performance during the handover decision. Handover metrics are the measurable qualities that indicate whether a handover initiation is needed; these metrics are described as follows.

• Network-related: bandwidth, link quality(received signal strength(RSS), bit error rate(BER)), cost, security level, etc.

• Mobile terminal-related: velocity, battery power, location information, etc.

• User-related: user profile and preferences.

These metrics are classified into static and dynamic^[2]. Vertical handover metrics play a significant role in the process of handover decision, and have an impact on system performance. Fig.1 shows the various metrics that make the decision of vertical handover in a heterogeneous network. Handover frameworks also have drawbacks, and significant research has focused on designing potential handover polices. Fig.2 reviews a comprehensive survey^[3]of schemes pertaining to executing a handover to an interconnected network. This section presents an overview of the recently developed always best connected(ABC)algorithm using single to multiple parameters.

Various feasible mobility protocols are available in the literature. Mobile users have a deterministic option of choosing the available network and moving to it for service. An essential condition designed for a switch-over is based on single parameter measurement. RSS-based schemes consider signal strength as an evaluation metric to compare the measured network with the current and target networks. This simple study involves a feasible level of switch-over when the measured signal strength falls below the threshold value. The focus is laid on traditional VHD methods for choosing a network connected in an interoperable environment. A seamless support for the handoff management technique is available with respect to the link and network layers^[3]. In this support, the initiated handover proceeds when the threshold value of RSS connected to the old base station decreases. A fixed value of RSS threshold is used to observe the transition from one network to another. The simulation performance measures are executed between the WLAN and 3G network.

However, this study fails to consider the mobility of the mobile node in WLAN when the travelling time is less than the latency of handover, which degrades the system performance. Ref. [4]presents a vertical handover between light fidelity(Li-Fi)with a radio frequency(RF)and wireless fidelity (Wi-Fi)network. The proposed scheme considers hybrid RF small-cell networks, but these small cells create frequent handover. This is unnecessary as the target measurements are based on signal strength and not on cell distance. Moreover, local overload situations of user movement decrease the system throughput.

On the other hand, Chang et al. develop a handover decision method in two phases^[5]: one with polynomial regression RSS prediction and the other with Markov decision process. The scheme adopts the traditional measure of RSS and compares the RSS values of the serving point of attachment. The main criterion for network selection is based on the predicted RSS value. Due to the mobility of the node at the region closer to the other network, the RSS of the old network seems to be less than that of the new network. Therefore, the handover decision is carried to the next stage. This leads to many computations and unnecessary handovers. Although the scheme adopts the cross-layer method to compare the costs of the two candidate networks, the decision about node mobility is not accurate in the network region. A mathematical model to predict the traveling distance to avoid unnecessary handover to the WLAN network is developed^[6]. This scheme measures the RSS available for the mobile node to an access point in a circular cellular region. However, it has several shortcomings. The main aim is to avoid unnecessary handover, but the scheme fails when the mobile node moves away from the boundary of the WLAN network. Mobility of the mobile node in this scenario seems to be unobserved.

Similar work in Ref. [7]resolves the above mentioned issue by predicting the traveling distance of the mobile node. This scheme aims to reduce the handover failure with RSS signal measurement along with the distance prediction method in the WLAN network. The decision to execute handover is performed when the mobile node moves away from the WLAN region, using the distance prediction method. However, the scheme fails to consider the speed of the mobile node and rather takes a sample of RSS and performs the handover decision. As the node mobility increases, unnecessary handover execution increases. This results in considerable resource wastage. Moreover, the scheme processes the decision within the WLAN region only.

Tariq et al. consider the user velocity^[8], and based on the moving average of the RSSI samples, the VH algorithm connected the mobile terminal to a WLAN/cellular medium. Although the probability of deciding the handover process is more likely to be simple, the scheme does not consider the target network condition.

In brief, these schemes take RSS as a sufficient criterion to proceed with the vertical handover decision scheme. However, RSS-based handover sometimes, under signal interference and unfavorable conditions such as noise, also initiates unnecessary vertical handovers. This becomes the initial drawback affecting the overall network performance. The signal fl uctuations resulting from the fading effect cause the undesirable so-called ping-pong effect, which in turn increases the probability of call failures and drops during the handoff process.

The above mentioned traditional single parameter-based handover schemes are found to be inefficient due to the generation of several unnecessary handovers and the increase caused in decision delays. Moreover, such handover schemes are much more suitable for horizontal handover only^[9]. Therefore, an efficient handover to minimize blocking probability is still an open challenge.

Bandwidth-based schemes take the difference in bandwidth between the current point of attachment and target access network as the selection criterion. Lee et al. consider traffic handling for balanced load in an access network, and bandwidth along with RSS as the selection criterion^[10].

Although a proper selection of handover is carried out in this scheme, heavy packet loss is observed. To resolve this issue, a bandwidth-based scheme considering either SINR or bandwidth metric to select the access network is proposed^[11]. The design is created such that the knowledge of bandwidth makes way for network selection. The SINR measurement of one network is converted into equivalent SINR to be compared with the SINR of the target network. Although this scheme achieves good throughput, the variation in SINR causes a heavy ping-pong effect.

Ref. [12]shows the design of vertical handover procedures between WLAN and WiMAX. The scheme is proactive and based on the total radio resource, where a handover decision can be made. The available bandwidth is estimated to accommodate mobile terminals into the network. Although high throughput is achieved again, high handoff delay is predominant with the arrival of the number of mobile nodes. This delay leads to heavy chance of call dropping.

The cost based vertical handover approaches use the network parameter metric to derive the best benefit for a cost function that determines the handoff to the candidate network. Monetary cost, bandwidth, and power consumption are involved in forming the derived cost function to select the access network. The network cost is computed by the sum of the weighted function of all available parameters taken for evaluation. These approaches use the impact of quality of service (QoS)parameter on the user. A normalized function of cost is considered as the unit of parameter variation. Ref. [13]deals with the cost-based adaptive handover scheme. The parameters are multiples for deciding the handover process. The cost function is provided as a weighted sum of normalized functions, and is given as

where W_i represents the weight of the out parameter and N_i represents the cost factor for parameter i to offer service at the network. The function describes the cost spent by the user to access a particular network for different services. Through this computation, the handover is initiated by the user to establish connection with the new network.

A cost-based vertical handover to offer high-quality of service or WWAN/WLAN-integrated networks is designed^[14]. An analytical model designed to estimate the throughput along with user performance is presented in IEEE 802. 11. Every time the cost computation is less, the mobile nodes move into a coverage area, and the proposed scheme selects such network along with user preferences. Ref. [15]considers bandwidth and monetary cost as evaluation metrics to execute handover decisions.

The cost function is also computed based on data throughput, power consumption, and monetary cost. In addition, user preferences are considered to select an optimal network for each MN. The main advantage of the above scheme lies in simplicity; however, the scheme depends on user preference. Users generally tend to seek a low-cost network. In such a case, the network load or the available resources become a demand that leads to a very high number of unnecessary handovers and handover failures in reasonable scenarios.

These schemes calculate a decision function using RSS, bandwidth, number of users, and other parameters such as power consumption and network usage cost. M. Ramn et al. address the fundamental issue for mobility management analysis, where the network cost is added along with multiple parameters^[16]. However, the complexity increases with increase in network parameter analysis in a wide region of the network. In addition, the analysis for cost becomes too complex and unreliable due to the wide chance of change in cost factor for the network. The multi-criteria-based decision offers seamless support to mobile nodes^[17]. This multi-metric scheme provides fl exibility to select the best network. The prediction technique combines the signal levels between the mobile station and the neighboring base station at the time of handover initiation to decide handovers^[18]. Although a higher prediction efficiency can be obtained, this scheme is suitable for horizontal handover only. The distance prediction with thresholds is designed^[19], where a mobile node can decide whether it should perform handover to a WLAN cell and the threshold value considered for the system velocity is adjusted to keep the probabilities of handover failure and unnecessary handover to the minimum level. Ref. [20]proposes a system that is modeled into a semi-Markov process along with the occupied bandwidth, delay and BER of the current network. However, it is more complex to design and implement in a dynamic environment. An enhanced handover algorithm^[21,22] does not depend on the RSS alone, but also integrates additional criteria such as delay, available bandwidth, and network conditions to satisfy the mobile station user’s preferences. Although the user preferences are satisfied to a greater extent, the delay is higher. Moreover, users often choose a low-cost network, which leads to handover failure in many instances^[22]. A heuristic handover algorithm^[23], which estimates the user velocity(speed and direction)in the algorithm to improve the performance design^[24]in an ad hoc handover, does not require an internetworking architecture and signaling between two ad hoc nodes. The system enables an MN requiring a vertical handover to request an ad hoc node to undertake some of the handover procedures. However, the system requires a large latency, and whenever the ad hoc node fails, the entire system is intercepted.

A context-aware handover decision is made for customization by tracking the user movement, device capabilities, and other personal context information. The decision process is built on when and how to switch between heterogeneous networks that are defined into a context-aware mobility management system. The context information includes mobile station-related parameters such as capacity, remaining battery power, location, and mobile velocity. User-based information such as preferred network, usage cost, and application-based information, such as the type of service(conversational, background, streaming, etc. ), are considered to make an effective decision on vertical handover in order to maintain a high level of user satisfaction.

Refs. [25,26]claim that guarantee on efficient service continuity in a heterogeneous network needs visibility of the handoff-related context information. A clear knowledge of the deployment environment must clarify the main handoff challenges. The system exchanges considerable information, thereby increasing system traffic and load. A coverage areas or hotspot locations are considered in Ref. [26]to reduce exchange of information. A fl exible mobile-controlled handover is designed in Ref. [27], where a user selects preferred networks with network parameter evaluation. However, the designed solutions are centralized and need a long processing time as the decision is based on global knowledge. Ref. [28]introduces a protocol between two proxy servers to exchange information in an interoperable environment. Although the proxies help in QoS, huge data management results in system complexity. Ref. [29]designs a context prediction model incorporated with a hidden variable to infl uence the future network availability. The policy accommodates any available observed context information; however, the study is limited since predicting the presence of networks and an optimal decision to trigger handovers is still an open challenge.

A load balancing algorithm between WiMAX and a Wi-Fiintegrated network are presented^[30]. WiMAX has access to all user information, and thus, the system distributes the traffic load to a Wi-Fi hotspot. The handover policy that supports QoS guides the user in executing handover. The protocol defines association with interoperability and exchanges information over heterogeneous wireless networks. The system improves the capacity and quality of wireless services based on the information exchange between networks and mobile terminals. However, exchange of information leads to traffic in wireless networks. The study designed to give preference to users again leads to a bottleneck in certain networks, which further leads to handover failure.

Information gathering from client and server plays a significant role in cooperative handover decisions. Mediaindependent handover utilizes this scheme at the link layer of the network, and MIH services have been defined in the IEEE 802. 21 standard. MIH discovers and acquires the network information required for network selection. It provides the following three main services^[31]:

• Media-independent event service;

• Media-independent command service;

• Media-independent information service.

MIH enables a network device-distributed decision making in heterogeneous wireless access networks. MIH^[32]provides a QoS-based seamless vertical handover in heterogeneous network environments, which maintains a reasonable resource management in the network. MIH enables the mobility of end-user device to get connected from one network to another through an IP session during a handover. The network conditions and mobile device status required to switch-over from one network to another are carried through a connectivity selection algorithm^[28]. The efficiency of this algorithm depends on the accuracy of the network state information. The network parameters include mode, authentication, and cost, and the routing information includes gateway along with cost and parameters such as data rate, throughput, and delay under network conditions.

Ref. [33]shows a network selection algorithm in an LTE network. The method sends at least one handover required message and a relocation required message, along with the target cell capabilities, to a core network element. Ref. [34]proposes an information server-based vertical handover procedure to avoid a time-consuming scanning process for channels. However, signaling overheads are very high and the use of RSS as the only link quality metric creates unnecessary handover.

These existing studies take the location of the mobile node under the assumption^[22]. As the location information is sent to the link layer, the mobile node position may vary at that point. Therefore, there exists no proper direction of information collection. Designing a zone-based architecture provides reliable information of mobile nodes.

The prime objective of a multi-criteria-based algorithm is the selection of the target network with the multiple attributes measured from all interconnected networks in a heterogeneous network environment. The common characteristics of multi-attribute decision-making are:

• ALTERNATIVE—number of alternative options.

• MULTI-ATTRIBUTES—multiple criteria associated with each problem.

• INCOMPARABLE UNITS—unit of measurement for each attribute associated with the problem.

• ATTRIBUTE WEIGHTS—assigned relative weight for each attribute that gives importance to another attribute.

• DECISION MATRIX—problem representation with columns indicating attributes and rows indicating alternatives.

A candidate network is selected by the defined network ranking function^[35]. The candidate network obtains the score with parameter evaluation taken as the metric during the process. The network ranking is computed with multiple parameters and the top-ranked networks serve as the candidate network for switch-over in the vertical handover decision process, as shown in Fig.3. There are few specified available techniques:

• Simple additive weighting(SAW)method;

• Analytic hierarchy process(AHP);

• Technique for order preference by similarity to the ideal solution(TOPSIS);

• Multiplicative exponent weighting(MEW);

• Weighted product method(WPM);

• VIKOR.

AHP evaluates the alternative set A1, A2, A3, …, Ax for the network with a decisive factor to make a pair-wise comparison. Criteria weighting is determined for scoring the alternatives. AHP compares different alternatives and builds the decisive goal. Using a pair-wise comparison, procedure priority is assigned and the hierarchy is formed for decision making. After a pair-wise comparison is made, AHP follows the hierarchical application of SAW.

SAW, on the other hand, is the simplest and widely used method to obtain a score. The method obtains all feasible solutions with all multiple alternatives with selective attributes. The final score for various alternatives is computed by multiplying the rate factor for each attribute of the important weight factor, and finally, summing these products over all available attributes. MWE works very closely to SAW, allowing the ranking score to be compared against an ideal solution. This type of MADM is a scoring method. The decision problem is expressed in a matrix form. Each row represents an alternative and each column corresponds to an attribute. The rank score of the network is determined by the weighted product of the below-listed attributes:

where r represents the rank of the network and x_{i j} indicates the score of option i with respect to criterion j. The construction of a normalized decision matrix transforms the various attribute dimensions into non-dimensional attributes, thus allowing comparisons across criteria^[36]. TOPSIS gives an ideal solution close to an optimal value. SAW, WPM, and TOPSIS share a common number of steps, as listed below:

•Determine the relevant objective;

•Determine the alternatives for evaluation;

•Identify the appropriate attributes to evaluate the alternatives;

•Determine the individual attribute range;

•Obtain normalized weights for the attributes;

•Rank the range in order of relative importance.

This section discusses the various studies conducted using multi-attribute-based vertical handover decisions. The best candidate network for a vertical handover is selected through a combination of AHP. Three networks—GPRS, UMTS, and satellite system—are employed for selection. A clear demonstration of how AHP can be employed to obtain the optimum results in a satellite with certain criteria is provided. Thereafter, the AHP application becomes a popular choice for research on network selection. An effective comparison between all possible MADM methods is made^[37]. SAW and TOPSIS exhibit the same performance for different amounts of traffic, while GRA provides better interactive and background traffic class. The multi-criteria vertical handover decision based on TOPSIS is proposed^[38], which considers traffic class, speed, and network occupancy. The system gives average blocking probability and the paper states that the combination of multiple strategies improves network performance. This work is a reasonable effort, which shows that multiattribute algorithms can be simulated and analyzed. A comparative study of seven MAMD methods for vertical handoff is presented^[39]: SAW, MEW, TOPSIS, GRA, ELECTRE, VIKOR, and WMC. Among these, VIKOR exhibits the best performance for network selection. Each parameter in each network with a random variable defines a priori-known probability density function (PDF)^[40]. This algorithm can take decision about networks after evaluating more parameter pdfs, which in turn improves risk removal in multi-attribute decision-making. The paper designs a performance evaluation for network selection by using MADM-based methods, which aim to keep the mobile users always best connected. To choose a target network that offers the highest overall performance, the subjective and objective weights of all parameters are considered. The prevention of network congestion and ping-pong effect is improved by WLAN; however, this paper fails to mark better performance of WiMAX and UMTS networks.

Ref. [41]proposes network selection with the fuzzy-AHP method, where the weight for parameter selection is computed. The scheme uses TOPSIS to rank the network. The study provides a detailed analysis of FAHP and highlights few theoretical problems in the existing approach. This leads to wrong decisions, where the first problem lies in the independence among all attributes at the same hierarchy level. The next problem is rank reversal, which is common in MADM. Therefore, to overcome these problems, the integration of a fuzzy analytic network process (ANP) (parent of AHP), together with TOPSIS^[42], is integrated. The former method performs weight elicitation, while the latter method(TOPSIS)is used for rating candidate networks. However, this study fails to focus on embedding optimization techniques for vertical handover in network selection mechanisms.

In Ref. [43], a network selection scheme is proposed with energy efficiency and user preference as quality levels for multimedia streaming to choose the target network. The objective of the research is to reduce power consumption in nextgeneration wireless networks while meeting users’quality expectations during handover^[44]. It is stated in Ref. [45]that multiple parameter-based algorithms should be used to select the information, in order to make the most of the contextual environment.

However, most of the schemes are mobile-initiated and controlled while user information does not consider topological information and mobile device movements. Therefore, the algorithm needs further assistance in refining the choice of candidate network with the avoidance of unnecessary vertical handovers. Determining the most suitable weights for different criteria for each traffic class is one of the main problems in network selection decision. Most of the cases in which the decision process is involved, wrong assignment of weights degrades the network performance. Therefore, it is of utmost importance to have typical parameter values to inherent the QoS guarantee.

Fuzzy logic (FL)-based handover decision algorithms are extensively available in the literature. The dynamic conditions of wireless networks necessitate computational intelligence techniques^[9,46]. These techniques include FL, fuzzy multi-attribute decision making, and neural networks (NNs) for handover decision, as shown in Fig.4. Unlike the traditional vertical handoff decision, these approaches enable FL to deal with imprecise information. A vertical handover algorithm using FL theory in the overlay network for WLAN and WWAN is presented in Ref. [47]. In this approach, the received signal strength predicted along with bandwidth and cost of networks adds QoS support in reducing the call dropping probability. However, the authors have limited their work to WLAN and WWAN. Moreover, the various traffic classes are not considered for simulation, and the schemes are mobile initiated. The accuracy in such a case remains unanswered. An FL-based handover scheme reduces unnecessary handovers^[36]. RSSI, speed, and distance are input parameters. The scheme incorporates a sliding window to monitor the output score. Although the system shows good performance, the complex design of a mobile terminal lowers the success rate due to the limited memory available and insufficient handling of the mobile terminal.

In Refs. [47,48], Kantubukta et al. first present an initial check for network selection through FL stages. The second phase includes multi-criteria evaluation based on mobile device parameters such as velocity and RSSI level. The velocity estimation is performed using GPS and angle of arrival. Although the FL method provides various membership functions, RSSI is the main criterion for the selection of the access network. Therefore, an insufficient selection of decision parameters increases the number of operations. Only two traffic classes are considered for simulation. Therefore, the system remains challenging and unresolved under the dynamic condition.

A QoS-aware fuzzy-rule-based multi-criteria algorithm along with a new evaluation model for decision-making is proposed in Ref. [23]. The handoff scores with all feasible QoS-related parameters as inputs are computed. The weight computation by AHP and target network section with a top network score add design features to the system with various traffic classes.

A new evaluation model based on a non-birth Markov chain helps MS to switch from one state to another when found in a region with no network connectivity in a region or multiple connectivity with more than one network. Important parameters such as RSS, end-user mobility, and preferences are not considered, and furthermore, a major issue in design complexity arises with the large number of rules. The use of FL, together with the MADM method^[49] , calculates the membership values of each parameter. The parameter is measured from different networks. The weights associated with these parameters are computed with AHP.

The parameters including bandwidth, E2E delay, jitter, BER, and weight evaluation provide the important network metrics based on user satisfaction and network provider. The objective of this scheme is to select a wireless network (UMTS, WLAN, and GRPS)for a particular service^[50] . A reliable end-user preference triggers the best access network by adopting an application-specific fuzzy rule base. The system provides weighting factor calculations for each application. Then, a decision matrix is constructed to enable multiple triggering factors of different access networks. The available networks are separated into the best and worst networks based on the network score. The handover rate results and avoidance of the ping-pong effect show better achievement as compared to the classical handover schemes. However, the large number of decision criteria lead to complexity, and the author fails to report this impact. The design of a neural-fuzzy system does not rely on a comprehensive rule base for significant perfor mance improvement^[51]. The neural-fuzzy system reduces the number of rules with trained input and output data. However, a comprehensive amount of time to train the data and weights associated with the performance measure keep the system in an uncertain level in an interoperable environment.

A fuzzy-MADM-based network access selection algorithm is proposed in Ref. [52], which considers parameters such as user preferences, network conditions, QoS, and energy consumption requirements. The main aim of the algorithm is to select an appropriate candidate network for a suitable service class. The network selection algorithm with a combination of AHP and FL in a heterogeneous wireless environment is designed^[53]. The approach selects the best target network for a vertical handover from among the WLAN and WMAN network systems. The Grey Prediction Theory predicts the future values of RSS to minimize the call dropping probability; however, the system fails to mark load balancing. All the above mentioned techniques fail to consider a network load, which results in over-utilization of the network. That is, the network capacity becomes full, thereby resulting in call blocking and call dropping due to overloading situations. Tab.1 shows a comparative analysis of various schemes designed for vertical handover. The traditional schemes are the simplest ones as the handover relies on a particular parameter. Multiple parameters are adopted in context-aware, FL-based decision-making and other computation methods. However, the complexity still exists and the mobile terminal-based decision makes the handover policy unreliable in a heterogeneous environment.

According to the brief review presented in the above section, traditional vertical handovers using a single criterion are common. It is observed that handover failure probability increases when either velocity or handover signaling delay upsurges the fixed value of the RSS threshold. An exhaustive study on network selection with the availability of a rich variety of context information shows that user preference plays a major role in QoS. However, network coverage information updates still lack in the decision process. The richness and complexity of the parameters and decision measurements are challenging. The widely used multi-attribute scoring methods include SAW, WPM, and TOPSIS. However, the benefits of a multi-criteria algorithm exploit the performance of the selection to a greater extent.

The uncertainty and abnormality in network ranking appear when the parameter is not weighed properly. This affects the system performance. The overhead in gathering decision information occurs in fuzzy-based decisions, and feeding them back to the mobile terminal creates extra delay and makes the system unsuitable for a new connection. In addition, training the dataset for precise information is challenging. The fuzzyMADM schemes developed for network selection make a heterogeneous network stable. Linear scale normalization is applied for weight computation in the fuzzy method. This affects the creditability of the parameters. However, the complexity of the system still exists. A context-aware, multi-attribute algorithm also shows the direction for vertical handover decision. In addition, the user preference is highly satisfied. A detailed network information management is unreliable due to lack of location information. High signaling exists due to centralized control and the whole system is designed for mobile terminals. Handling a complex design in mobile-initiated control makes the decision process inefficient as the mobile device has limited memory storage and insufficient capacity. Therefore, an optimized decision on the network server side for vertical handover is still an open challenge.

An optimization technique for determining the best candidate network during the vertical handover process is much needed to sustain connection. The key idea is to provide efficient seamless vertical handover with good bandwidth alloca tion along with optimized QoS support in terms of delay measure, reduced handover failure, and zero-level ping-pong effect. The methodology needs an optimization approach to increase the accuracy of decisions obtained through the MADM method. In addition, vertical handover is an open challenge that needs energy efficiency as in other networks^[54]. The schemes should work on the network-controlled side, where the challenges due to mobile-initiated handovers such as limited power and limited information about the network can be eliminated. An efficient queuing model can further improve the call handling performance.