A Hybrid-Fuzzy Logic Guided Genetic Algorithm (H-FLGA) Approach for Resource Optimization in 5G VANETs

This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at 1 A Hybrid-Fuzzy Logic Guided Genetic Algorithm (H-FLGA) Approach for Resource Optimization in 5G VANETs Ammara Anjum Khan ∗ , Mehran Abolhasan† , Wei Ni‡ , Justin Lipman§ and Abbas Jamalipour¶ Faculty of Engineering and Information Technology (FEIT), University of Technology, Sydney, {ammaraanjum.khan, mehran.abolhasan, justin.lipman}@uts.edu.au ‡ wei.ni @data61.csiro.au ¶ a.jamalipour @ieee.org Abstract—To support diversified quality of service (QoS) demands and dynamic resource requirements of users in 5G driven VANETs, network resources need flexible and scalable resource allocation strategies. Current heterogeneous vehicular networks are designed and deployed with a connection-centric mindset with fixed resource allocation to a cell regardless of traffic conditions, static coverage and capacity. In this paper, we propose a Hybrid-Fuzzy Logic guided Genetic Algorithm (H-FLGA) approach for the SDN controller, to solve a multi-objective resource optimization problem for 5G driven VANETs. Realizing the service oriented view, the proposed approach formulates five different scenarios of network resource optimization in 5G VANETs. Furthermore, the proposed Fuzzy Inference system (FIS) is used to optimize weights of multi-objectives, depending on the Type of Service (ToS) requirements of customers. The proposed approach shows the minimized value of multi-objective cost function when compared with the GA. The simulation results show the minimized value of end-to-end delay as compared to other schemes. The proposed approach will help the network service providers to implement a customer-centric network infrastructure, depending on dynamic customer needs of users. Index Terms—5G VANETs, Resource optimization in 5G VANETs, Next generation VANETs, Fuzzy Logic, Genetic Algorithm I. INTRODUCTION In the near future, VANETs would expect to utilize 5G cellular networks to deliver broadband services and enhance traffic and road safety to the users. In the next few years, there will be a dramatic increase in Machine-to-Machine (M2M) communication due to the massive diffusion of the Internet of Things (IoT) traffic. This dramatic increase will boost innovation and generate economic growth across a wide range of verticals such as automotive, energy, media, food and agriculture, healthcare, management, manufacturing, public transportation [1]- [2]. On the other hand, Vehicular Social Networks (VSNs) [3] are emerging where passengers can share user centric information using mobile devices and can exchange data related to infotainment, utility and emergency services [4]- [5]. By 2020, smart and secure Intelligent Transportation Systems (ITS) are expected to be operational as a "Copyright (c) 2019 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending a request to pubs-permissions@ieee.org." ∗ VANET cloud [6]. With this view, the emerging scenario of VANET implementations is expected to be heterogeneous in terms of resources, network topology, contents [7] and traffic types (including legacy voice and data traffic, as well as those generated by emerging M2M connections), all with different quality-of-service (QoS) requirements [8]. Also, current heterogeneous VANET architectures using cellular systems such as 4G and recent LTE Advanced systems have been designed and deployed with a connection-centric mindset with fixed resource allocation to a cell regardless of traffic conditions, static coverage and capacity [9], [7], [10]. Furthermore, they lack flexibility to efficiently deal with the data off-loading over different access networks [11] and to provide reconfigurability of RAN equipment to adapt to varying traffic and QoS demands of users. In order to support the exponential growth of heterogeneous mobile data traffic of new ITS applications and to support a platform for IoT applications and social networking, a radical rethink of current VANET architecture is essentially required. According to our vision, this evolution can only be achieved by turning it into a more flexible and programmable fabric, through technological improvements enabled by the next generation emerging technologies like Cloud-RAN, Software Defined Networking (SDN) and Fog Computing, which can jointly be used to provide a multitude of diverse services and resource sharing over a common underlying physical infrastructure. In our previous study, we proposed a 5G driven VANET architecture in [12], which offers a flexible and programmable fabric, leveraging the concepts of SDN, C-RAN and Fog Computing. In this study, we propose a hybrid optimization approach over our 5G VANET architecture, to provide an efficient resource allocation using Fuzzy logic guided Genetic algorithm. Fuzzy logic is one of the most well-known tools used to solve problems in dynamic and constantly changing systems. To address decision making process in VANETs, fuzzy logic has been used in different scenarios such as a broadcast protocol in Vehicular Ad hoc Networks, where the the fuzzy logic system decides if the node is required to rebroadcast or not [13]. In [14], a fuzzy logic-based scheme is proposed in VANETs to select backbone nodes, which consider the velocity of vehicles, the number of neighboring vehicles moving in the same direction and the height of the antenna. The idea behind using Fuzzy logic is to make the protocol suitable for particularly implementing Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. The final version of record is available at http://dx.doi.org/10.1109/TVT.2019.2915194 to cell-edge user equipment (UE) from the neighboring cells. However using this scheme the Base stations cannot make effective use of resources of neighbor cells, when there are no cell-edge UEs in the neighboring cells. Hence, another challenge is to design efficient dynamic Radio Resource Management (RRM) in 5G networks which will adapt to distinct traffic and interference variations in small cells [16]. Different approaches namely auction based [17], optimization based [18], demand supply based [19], Evolutionary Game Theoretic (EGT) based [20] are applied to solve different optimization problems in heterogeneous wireless access networks and VANETs. Furthermore, current Heterogeneous VANET (HetVANETs) implementations allocate fixed resources to a cell regardless of traffic conditions in other cells. To achieve these goals in 5G VANETs, flexible and optimal resource allocation methodologies must be devised to enhance network capacity for highly mobile users, by keeping in view the different QoS requirements of users. Also, mobile operators are constrained by the inflexibility and reconfigurability of Radio Access Network (RAN) equipment with respect to distinct traffic and QoS demands of users. To meet these challenging requirements a revolution of technologies in both Radio access networks and the mobile core network is required. Cloud-RAN has recently been identified as a leading candidate for 5G mobile network architecture which enables the sharing of network resources in a centralized data center, being cost-effective to operators and enhances the spectrum efficiency of the next generation networks [10], [21]. In C-RANs, a large number of low-cost Remote Radio Heads (RRHs) are randomly deployed and connect to the Base Band Unit (BBU) pool through the fronthaul links. The operations of RRHs and the computing resources of the BBU pool can be dynamically controlled in order to adapt to the capacity demand fluctuations, which leads to significant reductions in the capital expenditures (CapEx) and the operating expenses (OpEx) with much higher growth of revenue [22]. Additionally, C-RAN also allows integration of Long-Term Evolution Advanced (LTE-A) technologies and evolutions of novel 5G radio access and WiFi [23]. On the other hand, Software Defined Networking (SDN) has emerged as one of the possible solutions for combining the management of base stations and access networks due to the separation of control and data plane [24]. In an SDN-enabled network, all devices are managed and controlled by a centralized controller and the network operators can dynamically assign network virtualization strategies and forwarding rules to the controller instead of defining rules at different devices [25]. In addition, the SDN allows operators to quickly configure and deploy new network services and provide fine-grained traffic engineering control for each user, using a policy-based management paradigm running on commodity hardware. For example, bandwidth allocation can be dynamically designed by operators on a per-flow basis instead of using generic origin-destination criteria [26] and operators can employ different policies for diversified service demands of users. In 5G C-RANs, resource allocation and the RRH-BBU mapping problem has been addressed in a number of research works in the literature [27]- [28], However, in [27] a dynamic RRH-BBU mapping algorithm is developed. However, the customer-centric network infrastructure with varying type of service requirements. Furthermore, given a large number of combinations of linking the FC-ZCs with the BBUCs and capacity demand fluctuations, in our proposed architecture [12], efficient resource allocation becomes increasingly difficult to tackle, using the conventional brute-force techniques. Since fuzzy logic is flexible and tolerant of handling imprecise data and contradicting inputs, a hybrid approach can provide us with a better solution. The following are the contributions; • We propose a hybrid approach using a Hybrid-Fuzzy Logic guided Genetic Algorithm (H-FLGA) for the SDN controller, which allows the network service providers to implement a customer-centric network infrastructure. To the best of our knowledge, this is the first work in the area of 5G driven VANETs, which uses a hybrid Fuzzy Logic guided GA approach for resource optimization. • We formulate five different scenarios of resource optimization which focus on different network aspects, such as capacity, minimizing the number of FC-BBUCs, minimizing delay, the number of FC-ZCs which one BBUC handles, the traffic load of each FC-ZC and consequently of each BBUC pool. The proposed approach optimizes connections between the FC-BBUCs and the FZ-ZCs using the proposed H-FLGA approach. • Realizing the service oriented view, we define input and rules of proposed Fuzzy Inference System (FIS), for optimizing weights of multi-objectives, depending on the Type of Service (ToS) requirements of customers. Using proposed FIS, different options are weighted and multiobjective weights are optimized, to provide the optimal solution. • We compare the results of the proposed hybrid H-FLGA approach with GA and 5G driven VANET architecture in [12]. • The proposed approach supports energy efficient optimization for service providers, as some idle BBUC’s may be switched off without any adverse effect on the overall system, thus reducing the OpEx The remainder of the paper is organized as follows: Section II provides some challenges and the key enabler technologies for 5G Driven VANETs. Section III describes the formulation of resource optimization scenarios in 5G driven VANETs. Section IV explains the details of the proposed H-FLGA approach. Section V provides results and discussions and finally, Section VI concludes the paper. II. C HALLENGES AND K EY ENABLER T ECHNOLOGIES FOR 5G D RIVEN VANET S One of the promising techniques to support 5G cellular networks is Ultra Dense Networks (UDNs) [15], in addition to macro cells which provides wide coverage [16], [10]. By deploying more small cells in a fixed region, the average distance between the users and the BS can be significantly reduced and hence system capacity can be increased by improving the spatial reuse of radio resources. Furthermore, to mitigate the drastic interference generated by the neighboring small cells, the inter-cell interference coordination (ICIC) scheme of current 4G cellular networks assigns different blocks of resources 2 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. The final version of record is available at http://dx.doi.org/10.1109/TVT.2019.2915194 service provider’s profit is not a focus of attention. Similarly a resource allocation problem with a bargaining solution is proposed in [29] by employing the news-vendor game model. However, this scheme requires additional time for resource reconfiguration, which can deteriorate QoS requirements. In SDN based VANETs, resource management and allocation are very important since they can significantly affect the QoS and resource utilization. However, the relationship between user satisfaction and resource limitation and the interaction among various types of resources have not yet been fully studied, due to the new hierarchical framework of SDN [30]. There is a need to develop flexible and scalable resource allocation strategies to support diversified QoS demands in VANETs. Furthermore, the Ultra-Dense Networks (UDN) are also envisioned to be 5G enabling and highly promising technology used to enhance network capacity and spatial multiplexing. Some state-of-the-art research works in UDN and Computation offloading in the field of VANETs can be found in [31], [32]. The authors study the MECO problem in UDN and propose a heuristic greedy offloading scheme [33]. Furthermore, collectively SDN and C-RAN will provide service providers with an opportunity to implement a customercentric network infrastructure, where the network can automatically adapt to dynamic customer needs and capacity demand fluctuations of users in VANETs. The success of virtualization and cloud technologies provides one of the possible solutions. However, there are many directions needed to investigate to support the SDN based VANETs. SDN based migration is inevitable and unless a network is built from scratch, there is a need to manage both legacy and the SDN based framework ensuring service delivery and performance across all domains. The key to all of this is going to be the availability of interoperable virtual network functions (VNFs). Furthermore, due to the exchange of security related data between the vehicles and the RSUs over a separate channel also impose different challenges such as identity protection and data integrity because of the expected heterogeneous network architecture in 5G networks [34]. Recently, Mobile Edge Computation offloading (MECO) is also emerging as a key technology toward 5G to achieve lower latency and higher reliability [35]. However, the existing MECO research only focus on the resource allocation between the Mobile devices (MDs’) and the MEC servers and ignored the huge computation resources in the centralized cloud computing centers. In [35] the problem of collaborative computation offloading with centralized cloud and multi-access edge computing is studied. Similarly in [36], they studied the collaborative task offloading problem in vehicular edge computing networks to fully utilize the computing resources of the remote cloud center and MEC servers. In [37], a distributed and adaptive resource management controller is designed and tested, which allows the optimal utilization of cognitive radio and soft-input/soft-output data fusion in VANETs. In our opinion, since fuzzy logic is based on natural language and is tolerant of handling imprecise data therefore, combining Fuzzy logic with GA can provide us with a better solution for optimal resource utilization in VANETs. Furthermore, in a highly dynamic VANET environment, an optimal solution is dependent on the network environment such as bandwidth, vehicle mobility and link status. The solutions based on any mathematical modelling are non-flexible and complex to derive for rapidly changing environments [38], [39]. Therefore, we use a hybrid approach using Fuzzy Logic guided Genetic Algorithm for an optimum resource allocation in 5G driven VANETs. Our proposed Fuzzy Inference system (FIS) is used to optimize weights of multi-objectives. These optimized weights are then used by the Genetic Algorithm to optimize the connections between the BBUCs and the FC-ZCs. III. R ESOURCE O PTIMIZATION S CENARIOS IN 5G D RIVEN VANETS We propose an extension of our previously proposed 5G Next generation VANET architecture [12]. In our proposed architecture, there are Fog Computing-Zone Controllers (FCZCs), Fog Computing BBU Controllers (FC-BBUCs), Fog Computing-Cluster-Heads (FC-CHs) and Fog ComputingVehicles (FC-Vehicles). The purpose of this study is to optimize the allowable connections between FC-ZCs and FCBBUCs and also to support cost and energy efficiency by switching off the idle FC-BBUCs. In this section, we formulate five different scenarios of network resource optimization in 5G driven VANETs. P ROBLEM F ORMULATION Let ZC = {ZC1 , ZC2 , ..., ZCn } with cardinality |ZC| = nZC represents the set of Fog Computing Zone Controllers (FC-ZCs) which are distributed in an area. nZC is the number of FC-ZCs and nBBU C represents number of Fog Computing BBU Controllers (FC-BBUCs). Let BBU C = {BBU C1 , BBU C2 , ..., BBU Cn } with cardinality |BBU C| = nBBU C represent the set of FC-BBUCs, such that nBBU C ≤ nZC . Let Links = {BBU Ci , ZCj } represents the set of possible link pairs between FC-BBUCs and FC-ZCs. Variables   1, if ZCj is served by BBU Ci 0, otherwise Zij =  where(i, j) ∈ Links  1, if BBU Ci is Chosen Yi = 0, otherwise (1) (2) A. Minimize the number of FC-BBUCs (Min-BBUC) The objective of this problem is to minimize the number of FC-BBUCs serving FC-ZCs which are requesting for resources as shown in Fig. 1. It is assumed that FC-ZCs can connect to any of the BBUC pools, which means that there are no restrictions concerning distance. Given as an input data to the problem includes: the capacity row vector for BBUCs, capacity demand row vector for FC-ZCs and a binary link matrix indicating allowable connections between the BBUCs and the FC-ZCs. The objective function is given by Minimize CnBBU C = nBBU XC CBBU Ci Yi i=1 3 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. (3) This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at where nBBU C is the number of FC-BBUCs in the pool, CBBU Ci is ith element with a value equal to the total available capacity (Aggregated Link Capacity) of FC-BBUCi in capacity row vector CBBU C . Minimize CDelay = nBBU ZC X C nX i=1 Costi,j Zij (4) j=1 where nBBU C is the number of BBUCs in the pool, Costi,j is the link cost for linking ZCs j and BBU Ci in the cost matrix Cost. SDN Controllers C. Capacity Load Balance(Cap-LB)) The Cap-LB algorithm aims to balance the traffic load in every BBUC pool. The information of traffic load of each BBUC pool is always available and updated by the SDN controller. Before evaluating the decision, the controller has the information of traffic load of the possible BBUC pool connections. Thus, the controller will check not only the maximum capacity limit of the BBUC pool, but also the possible load on the BBUC pool. With this approach, as illustrated in Fig. 3, it is guaranteed that the BBUC pools have a capacity balance in what concerns to traffic load. The objective function is given by v unBBU C X 1 u t (D − Di )2 (5) Minimize CcapL = nBBU C i=1 BBUC Pool 1 BBUC Pool 2 FC-ZCn FC-ZC1 FC-ZC3 FC-ZC4 FC-ZC2 Fig. 1: Minimize the number of BBUC pools PnZC th where Di = element indicating the j=1 Zij CZCj is i total load demand inP BBUCi in the total load demand vector nBBU C 1 D and D = nBBU Di is the average load demand i=1 C across all BBUCs. The idea behind the objective function is to reduce the standard deviations of the total load demand vector D. Under ideal conditions, if the load demand is the same in all BBUCs, the positive objective function value must be equal to zero. B. Minimize Delay (Min-Delay) The objective of this problem is to minimize the delay by connecting FC-ZCs closer to the possible BBUC pool location, as illustrated in Fig. 2 by using the Min-Delay algorithm. The SDN controller has all the possible locations of BBUC pools, thus knowing all the distances between possible link connections between FC-ZC and BBUC. Given as an input data to the problem is the available capacity row vector for BBUCs, capacity demand row vector for FC-ZCs, a binary link matrix indicating allowable connections between the FCZCs and the BBUCs and the cost associated with each link. Since the delay is considered to be directly proportional to the distance between FC-ZCs and BBUCs which in turn is related to the cost associated with linking BBUC and FC-ZCs. The objective function is given by SDN Controllers BBUC Pool 1 50Mbps BBUCPool2 40Mbps 10 Mbps FC-ZCn SDN Controllers FC-ZC7 FC-ZC6 FC-ZC1 FC-ZC3 FC-ZC5 FC-ZC2 FC-ZC4 BBUC Pool 1 BBUC Pool 2 Fig. 3: Capacity Load Balance 1 D. Number of FC-ZCs per BBUC Balance Algorithm (FCZC-per-BBUC-Bal) FC-ZCn FC-ZC1 FC-ZC3 FC-ZC4 The objective of this problem is to balance the number of connections to BBUCs serving FC-ZCs that are requesting for resources using the FC-ZC-per-BBUC-Bal algorithm. Under ideal conditions, the algorithm should produce a connection FC-ZC2 Fig. 2: Minimize Delay 4 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at 3) BBUC i should be serving if at least one FC-ZC is connected to it. Zij ≤ Yi (9) arrangement for BBUCs and FC-ZCs such that the number of FC-ZCs in every BBUC should be balanced as illustrated in Fig. 4. The information on the number of FC-ZCs that each BBUC pool has, is always available and updated by the SDN controller. The objective function is given by ∀jǫ{1, 2, ..., nZC } and ∀iǫ{1, 2, ..., nBBU C }. 4) BBUC i should not be serving if no FC-ZCs are connected to it. n ZC X Zij ≥ Yi (10) SDN Controllers j=1 ∀jǫ{1, 2, ..., nZC } and ∀iǫ{1, 2, ..., nBBU C }. E. Constant Traffic Load (CTL)) BBUC Pool 1 CTL algorithm aims to force a constant traffic profile in every BBUC pool, with the objective to avoid traffic peaks, taking into consideration the three types of FC-ZCs: Residential FC-ZCs, Commercial FC-ZCs and Mixed FCZCs as illustrated in Fig. 5. Each FC-ZC has different traffic behavior throughout the day. To obtain multiplexing gains and energy efficiency in a Cloud-RAN approach, as compared to traditional RAN, an ideal BBUC pool traffic profile should have a constant traffic load throughout the day. This algorithm takes the selected hours of the day as input parameters (i.e. a vector of integers ranging between 1 and 24). Since the information on the traffic profile of each BBUC pool is always available and updated by the SDN controller. Before evaluating the decisions, the controller evaluates the possible connections, by using a vector of hours. The controller establishes the connection with BBUC pool by considering the traffic load. This problem addresses a time-series type of problem where the capacity demand of all the FC-ZCs are given with respect to time (demand vs time (duration in hours)). Using CTL, it is very effective to turn off some idle BBUC’s without any adverse effect on the overall system thus consuming energy efficiently. Given as input data to the problem is the capacity row vector for BBUCs, time-series capacity demand matrix for FC-ZCs, a binary link matrix indicating allowable connections between the BBUCs and the FC-ZCs and the cost associated with each link. The objective function is given by BBUC Pool 2 FC-ZC6 FC-ZC1 FC-ZC3 FC-ZC4 FC-ZC5 FC-ZC2 Fig. 4: Number of FC-ZC per BBUC Balance Minimize CZCperBBU C = 1 nBBU C v unBBU C u X t (N − Ni )2 (6) i=1 PnZC th element indicating the where Ni = j=1 Zij is the i total number of FC-ZCs connected to BBU Ci in the total PnBBU C 1 Ni is the connections vector N and N = nBBU i=1 C average number of FC-ZCS connected across all BBUCs. The idea behind the objective function is to reduce the standard deviations of the total connections vector N . Under ideal conditions, if the number of connection is same in all BBUCs, the positive objective function value must be equal to zero. Constraints for Problem III-A, III-B, III-C, III-D : Problems III-A, III-B, III-C, III-D are subject to the following constraints: 1) ZC j should at least be connected to one FC-BBUC. nBBU XC Zij = 1 CCT L = (7) (11) where Pi = [Zi×nZC , Zi×nZC , ...]T for iǫ{1, 2, ..., BBU C }; PnnZC Qi = Pi × CZCH for iǫ{1, 2, ..., nBBU C }; Ri = j=1 qj for iǫ{1, 2, ..., nBBU C } and qj is the j th column vector in Qi ; and BT Di = standard deviation of Ri for iǫ{1, 2, ..., nBBU C }. Ri column vector essentially contains the total demand in the BBU Ci with each row corresponding to a duration interval. BT Di is the ith value indicating the standard deviation of the variation of total demand in BBU Ci during the course of its operation in BBUC traffic deviation vector BBU-Traffic-Dev BT D. 1) ZC j should at least be connected to one BBUC. ∀jǫ{1, 2, ..., nZC } and ∀(i, j)ǫLinks. Zij is an element of link matrix Z whose value is 1 or 0 if ZCj is connected to BBU Ci or otherwise, respectively. 2) Sum of link capacity demands of FC-ZCs connected to BBUC i must be less than or equal to the available link capacity of BBUC i. Zij CZCj ≤ CBBU Ci BT Di i=1 i=1 n ZC X nBBU XC (8) j=1 ∀jǫ{1, 2, ..., nZC } and ∀iǫ{1, 2, ..., nBBU C } . CZCj is j th element with a value equal to the capacity demand for ZCj in capacity demand row vector CZC . nBBU XC Zij = 1 i=1 5 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. (12) This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at F. Multi-Objective Optimization SDN Controllers The Multi-objective function is mathematically the algebraic sum of the post-operated individual objective function with the possible lower limit of value 0 and an upper limit of the value of 1. For the purpose of creating an impartial multiobjective function, each of the individual objective functions described in previous sections is normalized with respect to a factor (so as to obtain a minimum and maximum value of 0 and 1, respectively) and a weight of unity is assigned to each individual objective function to demonstrate equal importance to each of the objectives. The inputs of the problem include link capacity of BBUCs, time-series demands of FC-ZCs, a binary link matrix indicating allowable connections between the BBUCs and the FC-ZCs and the cost associated with each link. The Objective functions from the previous sections are given as BBUC Pool 1 BBUC Pool 2 FC-ZC1 FC-ZCn FC-ZC4 FC-ZC5 FC-ZC6 FC-ZC3 Costant Traffic profile per BBUC FC-ZC2 nBBU C C1 = Fig. 5: Constant traffic load per BBUC X CBBU C i Yi i=1 nBBU C nZC C2 = where jǫ{1, 2, ..., nZC } and (i, j)ǫLinks. Zij is an element of link matrix Z whose value is 1 or 0 if ZCj is connected to BBU Ci or otherwise, respectively. 2) Sum of capacity demands of FC-ZCs connected to BBUC i must be less than or equal to the available capacity of that BBUC i. n ZC X Zij CZC j (Z, CZCH ) ≤ CBBU C i C3 = C4 = • nBBU C 1 nBBU C v unBBU C u X t (D − Di )2 (16) i=1 v unBBU C u X t (N − Ni )2 i=1 nBBU C (13) C5 = X BT Di i=1 Hence, the multi-objective function can be formulated as the algebraic sum of normalized individual objectives functions with a weightage factor. It is given by min Cobj where where jǫ{1, 2, ..., nZC } and iǫ{1, 2, ..., nBBU C }. CZC j is j th element with a value equal to the capacity demand for ZCj in capacity demand row vector CZC of size 1 × nZC . CZCH is also a function of the link matrix Z and the capacity demand matrix for FC-ZCs for all the durations. The capacity demands corresponding to each FC-ZC in the CZC are selected by the SDN controller as follows; • 1 Costij Zij j=1 i=1 j=1 • X X C1 − C1,min C2 − C2,min + ω2 + C1,max − C1,min C2,max − C2,min C4 − C4,min C3 − C3,min + ω4 + ω3 C3,max − C3,min C4,max − C4,min C5 − C5,min ω5 C5,max − C5,min Cobj =ω1 Find the Ri for each BBUC i. Note the FC-ZC connected to that BBUC i. Find the position of the peak demand in Ri . Use the same position to choose the CZC value for the connected FC-Zcs to BBU Ci from the CZ CH matrix. Repeat for all BBUCs till all the elements in CZ C capacity demand vector are obtained. where C1 , C2 , C3 and C4 are the Objective functions for the objectives laid out in Sections III-A, III-B, III-C, III-D and III-E, respectively. In 17, the max and min subscripts indicate the maximum and minimum values, respectively, for the objective functions when optimized individually. It is necessary to normalize the cost function values for each individual objective function between 0 and 1. ω1 , ω2 , ω3 , ω4 and w5 are the weightage factors for each individual objective functions. Importance to any objectives can be increased or decreased by the service providers, by altering these values, depending on the dynamic customer needs and capacity demand fluctuations of users in VANETs. 3) BBUC i should be serving if at least one FC-ZC is connected to it. Zij ≤ Yi (17) (14) where jǫ{1, 2, ..., nZC } and iǫ{1, 2, ..., nBBU C }. 4) BBUC i should not be serving if no FC-ZCs are connected to it. nX RRH Zij ≥ Yi (15) IV. H YBRID -F UZZY L OGIC GUIDED G ENETIC A LGORITHM (H-FLGA) We propose a Hybrid-Fuzzy Logic guided Genetic Algorithm (H-FLGA) approach for the SDN controller that solves a multi-objective optimization problem. Different objectives j=1 where jǫ{1, 2, ..., nZc } and iǫ{1, 2, ..., nBBU C } 6 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at are combined to assign most accurate practical connection arrangement between the BBUCs and the FC-ZCs. Depending on the Type of Service (ToS) requirements of customers, different options are weighted and optimized using the Fuzzy Inference System (FIS) and then used by the GA to provide optimal solution. 1) Fuzzy Inference System (FIS): Fig. 6 illustrates the flow chart of the proposed algorithm. The fuzzy system has two inputs; Type of service (T oS) and value (V alue). ToS is the requirement of customers based on three parameters i.e., Throughout, Delay and Cost. The outputs of FIS is priorities coefficients ωi for the optimized weights of multi-objectives. Hence, ω = f (T os, V alue) where T oS = {Delay(D), T hroughput(T ), Cost(C)} and V alue = {0, 1}. The outputs are in the range in [0, 1]. We choose the Gaussian membership function for the inputs and outputs variables. Tables I and II show the possible ToS values and ToS vs. Priority of ωi . To define the rules for each output, we follow the information in tables I and II. Rules: The rules for ωi are as follows; r1 r2 r3 r4 r5 r6 =′ =′ =′ =′ =′ =′ if if if if if if T oS T oS T oS T oS T oS T oS is D and V alue is zero then ω1 is zero′ ; is D and V alue is not zero then ω1 is zero′ ; is T and V alue is zero then ω1 is medium′ ; is T and V alue is not zero then ω1 is high′ ; is C and V alue is zero then ω1 is medium′ ; is C and V alue is not zero then ω1 is high′ ; Algorithm 1 : H-FLGA Input: link capacity of BBUCs’, demands of FC-ZCs, threshold of critical demands, Type of Service (T oS) and priority V alue Output: Optimized weights ω1 , ω2 , ω3 , ω4 , ω5 , Optimized multi-objectives (C1 , C2 , C3 , C4 , C5 ) Methods: evalfis(), H-FLGA() H-FLGA: (1) choose a max generation number GM AX and the cycle size C, set t = 0, (2) Initialize randomly the population (3) Compute the fitness for each individual in the population, (4) Extract and save the current best individual in the population, (5) if t ≤ GM AX Stop, else set t = t + 1 and continue to step (5) (6) Apply selection, crossover and mutation, (7) if mod(t, C) 6= 0 back to step (2) else Call Fuzzy Inference System (FIS) to compute new weights (8) end Create Initial Population Evaluation (Compute the Fitness) In the same manner, we define the rules for ω2 , ω3 , ω4 and ω5 . Selection TABLE I: Possible Type of Service (ToS) Values Type of Service (ToS) Normal Low High Delay (D) 0 1 - Throughput (T) 0 1 Fuzzy Inference System for Optimizing weights Crossover, Mutation Cost (C) 0 1 - Fuzzy Inference System(Compute w) Re-evaluation Input Type of Service(ToS) Stopping Criterion Delay Throughput 0 Normal 0 Normal 1 Low 1 High Cost 0 Normal 1 Low Fuzzification TABLE II: Type of Service (ToS) vs. Priority ω for Fuzzy Inference System Delay D Throughput T Cost C Min-BBUC(ω1 ) 0/Zero 1/Zero 0/Low 1/Low 0/High 1/High 0/Low 1/Low 0/Low 1/Low 0/Medium 1/High 0/Medium 1/Medium 0/Low 1/Low 0/Medium 1/Medium 0/Medium 1/Medium 0/Medium 1/High 0/Low 1/Low 0/Low 1/Zero 0/Low 1/Low 0/Medium 1/High Cap-LB (ω2 ) Min-Delay (ω3 ) FC-ZC-per-BBUC-Bal (ω4 ) CTL (ω5 ) NO yes ToS / Priority ω If Satisfied Rule 1 Rule 2 Rule n Optimal Solution Aggregartion De-Fuzzification Output(Optimized Weights w ) Fig. 6: Flow Chart of Hybrid-Fuzzy Logic Guided Genetic Algorithm (H-FLGA) After the application of the Fuzzy Inference System, the reevaluation part of the GA is executed where the fitness of the population is computed using multi-objective function with the new weights. Details about different objectives are discussed in section III. V. S IMULATION R ESULTS AND D ISCUSSIONS The results for the proposed algorithm are simulated using MATLAB 2017b. We used evalfis() to implement the Fuzzy 7 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at TABLE III: Simulation Parameters Logic rules. The details of the FIS rules and simulation parameters are provided in Table II and III respectively. We perform a comparison of multi-objective optimization using the GA and the proposed H-FLGA approach. The main metric for assessing the proposed algorithm is the value of multiobjective cost function that should lie between 0 and 5 as seen from (17) using the proposed hybrid H-FLGA approach. We test our results by optimizing the weights of different objectives in (17). These objectives indirectly relate to five different resource optimization scenarios focusing on different network aspects discussed in detail in section III. Fig. 7 shows the Multiobjective Cost Function FLGA GA 8 6 4 2 0 10 20 30 40 50 Value Maximum capacity of BBUC pool Maximum Fronthaul distance Number of Vehicles Transmission range of vehicles Speed of Vehicles MAC protocol Mobility Model Packet size Population size Tolerance for objective function Crossover Operator 100M Hz 40KM 50 to 300 up to 300m between 10m/s and 30m/s IEEE 802.11(11M bps) Manhattan grid (2500m × 2500m) 512bytes 2000 1e − 8 single (or multi) point weights. Our results show when ω1 is set to 1 by keeping capacity constraints under consideration, the utilization of capacity is optimized and consequently the number of BBUCs are minimized. Fig. 8(d) to Fig. 8(f) show the results of Cap-LB, when ω3 is set to 1 considering the constraints defined for Cap-LB. Fig. 8(g) to Fig. 8(i) show the results of FC-ZC-per-BBUC-Bal, when ω4 is set to 1 considering the constraints defined for FC-ZC-per-BBUC-Bal. Similarly Fig. 8(j) to Fig. 8(l) show the results of CTL, when ω5 is set to 1 considering the constraints defined for CTL. The results in Figs. 9(a) to 9(c) show how different objectives are optimized using H-FLGA. Therefore, our results in Fig. 7 prove that our proposed H-FLGA approach performs better when compared with the GA. This is because, the value of multi-objective cost function is reduced and minimized from 10.32 to 2.2 when we applied our proposed FIS rules. After applying the FIS rules, the weights ωi for different objectives are optimized. It is worth mentioning that the value of the optimized weights may vary depending on the ToS values defined in the FIS rules. Hence, depending on ToS requirements of different customers, service providers can implement different FIS rules for different objectives and assign different priorities of Low, M edium and High to ToS parameters and get the optimized weights. Therefore, we conclude that our proposed hybrid HFLGA approach performs better than the GA and is flexible to set the weights of multi-objectives, depending on the QoS demands and requirements of different users. Fig. 10 shows the variation of the end-to-end delay for the vehicles within a maximum front-haul distance of 40km using three schemesH-FLGA, GA and [12]. The number of vehicle counts is increased from 50 to 300 with varying speeds. The delay is the highest using [12] while lowest using proposed H-FLGA. For each data point in the graph, vertical markers are used to indicate the confidence of an interval of 95%. However, [12] has the widest confidence interval compared to the GA and the H-FLGA, which indicates, these is likely to be more variation in the delay estimated using the [12] and the GA. Our results show that when ω2 is set to 1 and solved using the GA, the maximum value of delay is 0.113s and the maximum value of delay when calculated using [12], is 0.171s. However, the value of delay is lowered and improved to 0.062s when the delay is computed using our proposed H-FLGA approach. Complexity Analysis: We discuss the complexity of the proposed algorithm with respect to algorithm complexity of the proposed H-FLGA, including the signaling overhead on 10 0 Simulation Parameter 60 No. of Generations Fig. 7: Variation of multi-objective function value for different numbers of generations variation of the objective function for the optimized parameters using the H-FLGA approach. The population size of 1000 is used for each generation and the number of generations is increased from 10 to 50 with an increment of 10. It is observed that increasing the number of generations optimizes the value of the objective function however, beyond 10 generations it is observed that the value of objective function remains more or less the same. Also, a statistical study is carried out to determine the 95% confidence in the obtained solution with the algorithm. It is observed that with lower generations, there is likely to be more variation in the result however, beyond 10 generations the confidence of obtaining the optimum value is very high as there is no varying interval on those data points. The value of the cost function using the GA is optimized to 10.32 as the best score of the multi-objective function. Whereas, the best score of the multi-objective function should be between 0 and 5 as seen in (17). Therefore, the GA could not optimize the value of the multi-objective cost function. Hence, to improve the results of the multi-objective cost function, we run our proposed hybrid H-FLGA approach as a tool to optimize the weights of the multi-objective function. The results in Fig. 7 show the value of the multi-objective cost function is reduced to an optimized value of 2.2 using the proposed H-FLGA. We solve the multi-objective optimization problem in (17) with fixed values of weights ωi where we assume that all of them are equal to one, by considering all cost functions C1 to C5 as of equal importance. Figs. 8(a) to 8(l) show how different objectives are achieved with equal 8 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at 100 Total Capacity of BBUC Utilized Capacity of BBUC 0.5 0 5 1 80 Capcity Demands of FC-ZCs 1 Total Capacity of BBUC Active [1] / Inactive [0] Optimum Connections Graph 60 40 20 30 20 10 0 4 BBUCs 2 1 5 1 3 2 FC-ZCs 3 4 2 0 1 2 3 3 BBUC 3 2 BBUC 1 FC-ZCs (a) Optimum number of Connec-(b) Capacity Utilization of BBUC(c) Capacity Demands of FC-ZCs tions using Min-BBUC using Min-BBUC using Min-BBUC 100 Total Capacity of BBUC 90 Optimum Connections Graph Utilized Capacity of BBUC Total Capacity of BBUC 80 0.8 0.6 0.4 0.2 0 60 50 40 30 20 5 1 4 2 3 2 10 0 5 1 4 2 3 3 0 1 20 10 3 BBUC 30 70 Capacity Demands of FC-ZCs Active [1] / Inactive [0] 1 1 2 FC-ZCs 2 3 1 BBUCs BBUC FC-ZCs (d) Optimum number of Connec- (e) Capacity Utilization of BBUC (f) Capacity Demands of FC-ZCs tions using Cap-LB using Cap-LB using Cap-LB 100 Total Capacity of BBUC Utilized Capacity of BBUC 90 1 0.8 0.6 0.4 0.2 0 70 60 50 40 30 20 5 1 30 20 10 0 10 0 2 BBUCs 1 4 2 3 3 5 1 4 2 Capcity Demands of FC-ZCs 80 Total Capacity of BBUC Active [1] / Inactive [0] Optimum Connections Graph 1 2 FC-ZCs 3 3 3 2 BBUCs BBUC 1 FC-ZCs (g) Optimum number of Connec-(h) Capacity Utilization of BBUC(i) Capacity Demands of FC-ZCs tions using FC-ZC-per-BBUC-Bal using FC-ZC-per-BBUC-Bal using FC-ZC-per-BBUC-Bal 100 Total Capacity of BBUC Utilized Capacity of BBUC 90 1 0.5 0 6 5 1 4 2 Total Capacity of BBUC 80 Capcity Demands of FC-ZCs Active [1] / Inactive [0] Optimum Connections Graph 100 50 0 6 3 40 30 10 3 3 FC-ZCs 50 20 4 2 2 1 60 5 1 3 BBUCs 70 2 1 BBUCs 0 1 FC-ZCs 2 3 BBUC (j) Optimum number of Connec-(k) Capacity Demands of FC-ZCs(l) Capacity Utilization of BBUC tions using CTL using CTL using CTL Fig. 8: Multi-objective Optimization with equal weights ω1 = 1, ω2 = 1, ω3 = 1, ω4 = 1, ω5 = 1 Optimum Connection Graph 100 Active [1] / Inactive [0] Total Capacity of BBUC 25 1 80 70 60 50 40 20 0.8 15 0.6 10 0.4 0.2 0 30 20 Capacity Demands of FC-ZCs Total Capacity of BBUC Utilized Capacity of BBUC 90 5 1 5 0 5 1 4 4 10 2 3 0 1 2 BBUC 3 3 BBUC 2 2 1 3 3 FC-ZCs BBUC 2 1 FC-ZCs (a) Capacity Utilization of BBUC(b) Optimum number of connec-(c) Capacity Demands of FC-ZCs using H-FLGA tions using H-FLGA using H-FLGA Fig. 9: Multi-objective Optimization using Optimized weights 9 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at and the proposed FIS is used to optimize the weights of the multiple objectives. These optimized weights are then used by the Genetic Algorithm to optimize connections between the BBUCs and the FC-ZCs. This work will provide service providers with an opportunity to implement a flexible customer-centric network infrastructure, by improving the spectral efficiency. Moreover, the proposed approach can be used to support energy efficient optimization for service providers, as some idle BBUC’s may be switched off without any adverse effect on the overall system thus, reducing the OpEx. Future directions include the possibility of implementing the proposed method in future ultra-dense networks, especially its implementation in computation offloading and resource allocation in the ultra-dense networks. Furthermore, the proposed approach may be tested using OpenFlow and Mininet. 0.18 0.16 End-to-End (E2E) delay (s) 0.14 [12] GA FLGA 0.12 0.1 0.08 0.06 0.04 0.02 0 0 50 100 150 200 250 300 350 No. of Vehicles Fig. 10: End-to-End Delay R EFERENCES [1] C. V. N. Index, “Cisco visual networking index: global mobile data traffic forecast update, 2014–2019,” Tech. Rep, 2015. [2] K. M. Alam, M. Saini, and A. El Saddik, “Toward social internet of vehicles: Concept, architecture, and applications,” IEEE Access, vol. 3, pp. 343–357, 2015. [3] N. Kayastha, D. Niyato, P. Wang, and E. Hossain, “Applications, architectures, and protocol design issues for mobile social networks: A survey,” Proceedings of the IEEE, vol. 99, no. 12, pp. 2130–2158, 2011. [4] N. Abbani, M. Jomaa, T. Tarhini, H. Artail, and W. El-Hajj, “Managing social networks in vehicular networks using trust rules,” in Wireless Technology and Applications (ISWTA), 2011 IEEE Symposium on. IEEE, 2011, pp. 168–173. [5] S. Smaldone, L. Han, P. Shankar, and L. Iftode, “Roadspeak: enabling voice chat on roadways using vehicular social networks,” in Proceedings of the 1st Workshop on Social Network Systems. ACM, 2008, pp. 43–48. [6] S. Jia, S. Hao, X. Gu, and L. Zhang, “Analyzing and relieving the impact of FCD traffic in LTE-VANET heterogeneous network,” in Telecommunications (ICT), 2014 21st International Conference on. IEEE, 2014, pp. 88–92. [7] Z. He, D. Zhang, and J. Liang, “Cost-efficient heterogeneous data transmission in software defined vehicular networks,” in Proceedings of IEEE 17th International Conference on High Performance Computing and Communications (HPCC),New York, USA. IEEE, 2015, p. 666–671. [8] D. Sabella, A. Vaillant, P. Kuure, U. Rauschenbach, and F. Giust, “Mobile-edge computing architecture: The role of MEC in the Internet of Things,” IEEE Consumer Electronics Magazine, vol. 5, no. 4, pp. 84–91, 2016. [9] D. Sabella, P. Serrano, G. Stea, A. Virdis, I. Tinnirello, F. Giuliano, D. Garlisi, P. Vlacheas, P. Demestichas, V. Foteinos et al., “A flexible and reconfigurable 5G networking architecture based on context and content information,” in Networks and Communications (EuCNC), 2017 European Conference on. IEEE, 2017, pp. 1–6. [10] C. Pan, M. Elkashlan, J. Wang, J. Yuan, and L. Hanzo, “User-centric C-RAN Architecture for Ultra-dense 5G Networks: Challenges and Methodologies,” arXiv preprint arXiv:1710.00790, 2017. [11] K. Zheng, L. Hou, H. Meng, Q. Zheng, N. Lu, and L. Lei, “Soft-defined heterogeneous vehicular network: Architecture and challenges,” IEEE Network, vol. 30, no. 4, pp. 72–80, 2016. [12] A. A. Khan, M. Abolhasan, and W. Ni, “5G Next generation VANETs using SDN and Fog Computing Framework,” in Consumer Communications & Networking Conference (CCNC), 2018 15th IEEE Annual. IEEE, 2018, pp. 1–6. [13] E. Limouchi, I. Mahgoub, and A. Alwakeel, “Fuzzy logic-based broadcast in vehicular ad hoc networks,” in 2016 IEEE 84th Vehicular Technology Conference (VTC-Fall). IEEE, 2016, pp. 1–5. [14] C. Wu, S. Ohzahata, Y. Ji, and T. Kato, “Joint mac and network layer control for vanet broadcast communications considering end-toend latency,” in 2014 IEEE 28th International Conference on Advanced Information Networking and Applications. IEEE, 2014, pp. 689–696. [15] J. G. Andrews, S. Buzzi, W. Choi, S. V. Hanly, A. Lozano, A. C. Soong, and J. C. Zhang, “What will 5G be?” IEEE Journal on selected areas in communications, vol. 32, no. 6, pp. 1065–1082, 2014. the controller. The complexity of the proposed approach is a function of the GA and the FIS denoted by f (O(GA) O(F IS)). O(GA) depends on the operations: generate the first generation, selection, crossover, mutation and find the best individual. In addition, it also depends on the number of generations GM AX . In the case of integer optimization these operators are simpler to implement. Therefore, O(GA) mainly depends on the complexity of the multi-objective cost function O(Cobj ) and the number of maximum generations GM AX . Furthermore, O(F IS)) depends on the number of cycles nc of the FIS. Hence, the combined complexity of the proposed algorithm can be expressed as O(nc GM AX O(Cobj )). It is seen from the results in Fig.7 that the objective is optimized between first 20 generations. Hence, we conclude that the proposed algorithm is efficient. Signaling Overhead: To analyze the signaling overhead on the controller, we consider a graph G := (nBBU C , nZC ) comprising a set nZC of vertices together with a set nBBU C ⊂ ZC nZC × nZC of edges. There are a total of nnBBU C possible configurations of signaling. Let the average number of BBUCController control packets are represented by S. Then the total number of subsets of S is given by 2S . Equating the total possible configurations of signaling to the total number of subsets of S, we get ZC 2S = nnBBU C S ≈ nZC log2 (nBBU C ) (18) Hence, the signaling overhead on the controller is O(nZC log2 (nBBU C ). This shows that the control overhead on the controller will increase logarithmically, as nBBU C increases. Hence we conclude that, the complexity of signaling overhead on the controller will grow very slowly. VI. C ONCLUSION Realizing the service-oriented view, we propose a hybrid Fuzzy Logic guided Genetic Algorithm (H-FLGA) approach for the SDN controller, for an optimum resource allocation in 5G driven VANETs. This approach solves a multi-objective optimization problem, where different objectives are combined 10 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at [16] F. Teng, “Resource management in next generation wireless networks: Optimization and games,” Ph.D. dissertation, Northwestern University, 2016. [17] O. Sallent, J. Pérez-Romero, R. Agusti, L. Giupponi, C. Kloeck, I. Martoyo, S. Klett, and J. Luo, “Resource auctioning mechanisms in heterogeneous wireless access networks,” in Vehicular Technology Conference, 2006. VTC 2006-Spring. IEEE 63rd, vol. 1. IEEE, 2006, pp. 52–56. [18] W. Zhang, “Bearer service allocation and pricing in heterogeneous wireless networks,” in Communications, 2005. ICC 2005. 2005 IEEE International Conference on, vol. 2. IEEE, 2005, pp. 1367–1371. [19] H. Chan, P. Fan, and Z. Cao, “A utility-based network selection scheme for multiple services in heterogeneous networks,” in Wireless Networks, Communications and Mobile Computing, 2005 International Conference on, vol. 2. IEEE, 2005, pp. 1175–1180. [20] A. A. Khan, M. Abolhasan, and W. Ni, “An Evolutionary Game Theoretic Approach for Stable and Optimized Clustering in VANETs,” IEEE Transactions on Vehicular Technology, vol. 67, no. 5, pp. 4501– 4513, 2018. [21] V. W. Wong, R. Schober, D. W. K. Ng, and L.-C. Wang, Key Technologies for 5G Wireless Systems. Cambridge university press, 2017. [22] S. Shattil, “Cloud Radio Access Network,” May 12 2015, uS Patent App. 14/709,936. [23] J. S. Bedo, S. E. El Ayoubi, M. Filippou, A. Gavras, D. Giustiniano, P. Iovanna, A. Manzalini, O. Queseth, T. Rokkas, M. Surridge et al., “5G Innovations for New Business Opportunities,” 2017. [24] H. Li, M. Dong, and K. Ota, “Radio access network virtualization for the social Internet of Things,” IEEE Cloud Computing, vol. 2, no. 6, pp. 42–50, 2015. [25] C. Jiacheng, Z. Haibo, Z. Ning, Y. Peng, G. Lin, and S. Xuemin, “Software defined Internet of vehicles: architecture, challenges and solutions,” Journal of Communications and Information Networks, vol. 1, no. 1, pp. 14–26, 2016. [26] A. Argyriou, K. Poularakis, G. Iosifidis, and L. Tassiulas, “Video Delivery in Dense 5G Cellular Networks,” IEEE Network, vol. 31, no. 4, pp. 28–34, 2017. [27] M. Khan, R. Alhumaima, and H. Al-Raweshidy, “Quality of service aware dynamic bbu-rrh mapping in cloud radio access network,” in 2015 International Conference on Emerging Technologies (ICET). IEEE, 2015, pp. 1–5. [28] M. Qian, W. Hardjawana, J. Shi, and B. Vucetic, “Baseband processing units virtualization for cloud radio access networks,” IEEE Wireless Communications Letters, vol. 4, no. 2, pp. 189–192, 2015. [29] S. Kim, “News-vendor game-based resource allocation scheme for nextgeneration c-ran systems,” EURASIP Journal on Wireless Communications and Networking, vol. 2016, no. 1, p. 158, 2016. [30] B. Cao, Y. Li, C. Wang, G. Feng, S. Qin, and Y. Zhou, “Resource allocation in software defined wireless networks,” IEEE Network, vol. 31, no. 1, pp. 44–51, 2017. [31] H. Guo, J. Liu, J. Zhang, W. Sun, and N. Kato, “Mobile-edge computation offloading for ultradense iot networks,” IEEE Internet of Things Journal, vol. 5, no. 6, pp. 4977–4988, 2018. [32] N. Abbas, H. Hajj, S. Sharafeddine, and Z. Dawy, “Traffic offloading with channel allocation in cache-enabled ultra-dense wireless networks,” IEEE Transactions on Vehicular Technology, vol. 67, no. 9, pp. 8723– 8737, 2018. [33] H. Guo, J. Liu, and J. Zhang, “Computation offloading for multi-access mobile edge computing in ultra-dense networks,” IEEE Communications Magazine, vol. 56, no. 8, pp. 14–19, 2018. [34] M. H. Eiza, Q. Ni, and Q. Shi, “Secure and privacy-aware cloudassisted video reporting service in 5g-enabled vehicular networks,” IEEE Transactions on Vehicular Technology, vol. 65, no. 10, pp. 7868–7881, 2016. [35] H. Guo and J. Liu, “Collaborative computation offloading for multiaccess edge computing over fiber–wireless networks,” IEEE Transactions on Vehicular Technology, vol. 67, no. 5, pp. 4514–4526, 2018. [36] H. Guo, J. Zhang, and J. Liu, “Fiwi-enhanced vehicular edge computing networks: Collaborative task offloading,” IEEE Vehicular Technology Magazine, vol. 14, no. 1, pp. 45–53, 2019. [37] N. Cordeschi, D. Amendola, and E. Baccarelli, “Reliable adaptive resource management for cognitive cloud vehicular networks,” IEEE Transactions on Vehicular Technology, vol. 64, no. 6, pp. 2528–2537, 2015. [38] C. Wu, S. Ohzahata, and T. Kato, “Flexible, portable, and practicable solution for routing in VANETs: a fuzzy constraint Q-learning approach,” IEEE Transactions on Vehicular Technology, vol. 62, no. 9, pp. 4251– 4263, 2013. [39] G. El Mouna Zhioua, N. Tabbane, H. Labiod, and S. Tabbane, “A fuzzy multi-metric QoS-balancing gateway selection algorithm in a clustered VANET to LTE advanced hybrid cellular network,” IEEE Transactions on Vehicular Technology, vol. 64, no. 2, pp. 804–817, 2015. Ammara Khan is persuing her Ph.D. from School of Electrical and Data Engineering, University of Technology, Sydney. She has recently submitted PhD thesis for examination. Prior to this, she completed her M.E. in Telecommunications in 2015 from University of Technology, Sydney. She also received M.Sc. in Computer Science and M.Sc. in Telecommunication Systems degrees from Bahauddin Zakariya University, Pakistan. She has served as a lecturer in department of Telecommunication Systems, Bahauddin Zakariya University, Pakistan. Her current research interests include 5G Vehicular Ad-hoc Networks, game theory, Software Defined Networking, Network Function Virtualization, Mobile Edge Computing, as well as their applications in Vehicular Ad-hoc Networks. Mehran Abolhasan A/Prof. Mehran Abolhasan completed his B.E. in Computer Engineering and Ph.D. in Telecommunications in 1999 and 2003 respectively, at the University of Wollongong. He is currently an Associate Professor and Deputy Head of School at School of Electrical and Data Engineering at University of Technology, Sydney. A/Prof. Abolhasan has authored over 120 international publications and has won over 3 million dollars in research funding. His Current research Interests are in Software Defined Networking, IoT, Wireless Mesh, Wireless Body Area Networks, Cooperative Networks, 5G Networks and Beyond and Sensor networks. He is currently a Senior Member of IEEE. 11 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org. This is the author's version of an article that has been published in this journal. Changes were made to this version by the publisher prior to publication. http://dx.doi.org/10.1109/TVT.2019.2915194 The final version of record is available at Wei Ni (M’09-SM’15) received the B.E. and Ph.D. degrees in Electronic Engineering from Fudan University, Shanghai, China, in 2000 and 2005, respectively. Currently he is a Team Leader at CSIRO, Sydney, Australia, and an adjunct professor at the University of Technology Sydney (UTS). He also holds honorary positions at the University of New South Wales (UNSW) and Macquarie University (MQ). Prior to this, he was a postdoctoral research fellow at Shanghai Jiaotong University from 2005 – 2008; Deputy Project Manager at the Bell Labs R&I Center, Alcatel/Alcatel-Lucent from 2005 – 2008; and Senior Researcher at Devices R&D, Nokia from 2008 – 2009. His research interests include stochastic optimization, game theory, graph theory, as well as their applications to network and security. Dr Ni has been serving as the Vice Chair of IEEE NSW VTS Chapter and an Editor of IEEE Transactions on Wireless Communications since 2018, the secretary of IEEE NSW VTS Chapter from 2015 - 2018, Track Chair for VTC-Spring 2017, Track Co-chair for IEEE VTC-Spring 2016, and Publication Chair for BodyNet 2015. He also served as Student Travel Grant Chair for WPMC 2014, a Program Committee Member of CHINACOM 2014, a TPC member of IEEE ICC’14, ICCC’15, EICE’14, and WCNC’10. Dr Justin Lipman is an Industry Associate Professor at the University of Technology Sydney and Director of the RF Communications Technologies (RFCT) Lab, where he leads industry engagement in RF technologies, Internet of Things, Tactile Internet and Software Defined Communication. He serves as committee member in Standards Australia contributing to International IoT standards and is a Research Program Lead for the Food Agility Cooperative Research Centre. He received his PhD Telecommunications and BE Computer Engineering from the University of Wollongong, Australia in 2003 and 1999 respectively. From 2004 to 2017, Dr. Lipman was previously based in Shanghai, China and held a number of senior management and technical leadership roles at Intel and Alcatel leading research and innovation, product architecture and IP generation. He is an IEEE Senior Member. His research interests are in all “things” adaptive, connected, distributed and ubiquitous. Abbas Jamalipour (S’86–M’91–SM’00–F’07) is the Professor of Ubiquitous Mobile Networking at the University of Sydney, Australia, and holds a PhD in Electrical Engineering from Nagoya University, Japan. He is a Fellow of the Institute of Electrical, Information, and Communication Engineers (IEICE) and the Institution of Engineers Australia, an ACM Professional Member, and an IEEE Distinguished Lecturer. He has authored nine technical books, eleven book chapters, over 450 technical papers, and five patents, all in the area of wireless communications. Dr. Jamalipour is an elected member of the Board of Governors, Executive Vice-President, Chair of Fellow Evaluation Committee, and the Editor-inChief of the Mobile World, IEEE Vehicular Technology Society. He was the Editor-in-Chief IEEE Wireless Communications, Vice President-Conferences and a member of Board of Governors of the IEEE Communications Society, and has been an editor for several journals. He has been a General Chair or Technical Program Chair for a number of conferences, including IEEE ICC, GLOBECOM, WCNC and PIMRC. He is the recipient of a number of prestigious awards such as the 2016 IEEE ComSoc Distinguished Technical Achievement Award in Communications Switching and Routing, 2010 IEEE ComSoc Harold Sobol Award, the 2006 IEEE ComSoc Best Tutorial Paper Award, as well as 15 Best Paper Awards. 12 Copyright (c) 2019 IEEE. Personal use is permitted. For any other purposes, permission must be obtained from the IEEE by emailing pubs-permissions@ieee.org.