Optimization of the Spectrum Splitting and Auction for 5th Generation Mobile Networks to Enhance Quality of Services for IoT from the Perspective of Inclusive Sharing Economy

Abstract

Fifth generation (5G) mobile networks can accomplish enhanced communication capabilities and desired to connect things in addition to people. By means of optimally splitting the spectrum to integrate more efficient segments, mobile operators can deliver better Quality of Services (QoS) for Internet of Things (IoT), even the nowadays so-called metaverse need broadband mobile communication. Drawing on the Theory of Quality Value Transformation, we developed a 5G ecosystem as a sustainable organic coalition constituted of planners, providers, and users. Most importantly, we put forward the altruism as the ethics drive for the organic cooperative evolution to sustain the inclusive sharing economy to solve the problem of the Theory of Games and Economic Behavior. On the top of the collaboration framework for the coalition game for 5G, we adopted Pareto Optimality as the target situation for the optimization via cooperative evolution and further apply ISO 25000 to define the metrics for the value of 5G corresponding to Pareto Frontier. Based on the collaboration framework as above, we conducted a survey to gather the features and costs for the 5G spectrum in relation to IoT and the financial status of the mobile operators as the constraint for the optimization. Taking Simultaneous Multi-Round Auction (SMRA) as the standard rule for spectrum auction, we developed a novel optimization program of two hybrid metaheuristics with the combination of Simulated Annealing (SA), Genetic Algorithm (GA), and Random Optimization (RO) for the multiple objectives of quality, usability, and costs. The results of the simulation show that the coalition game for 5G spectrum auction is a dynamic group decision in which the government authority and mobile operators can achieve a synergy to maximize the profits, quality score, and usability, and minimize the costs. Last but not least, the hybrid metaheuristic with SA and RO is more efficient and effective than that with GA and BO, from the perspective of inclusive sharing economy. It is the first study of its kind as we know.

1. Introduction

The advent of ICT has been brought the world into Inclusive Sharing Economy in relation to New Economy, along which ICT resources, including knowledge and services, become the common wealth for this crowded world. On the stride of the sustainable development, Inclusive Sharing Economy has been take place and ICT ecosystems become the fundamental infrastructure for sustain evolution of the world.

5G mobile networks are expected to realize enhanced devices and communication capabilities with greater efficiency, throughput, and lower latency [1] and to connect things so that it is most important for mid- and long-range Interne of things (IoT) [2]. The release spectrum for 5G could allow mobile companies to increase their communication capacity and allow users to take advantage of more effective and trustworthy services. The spectrum is a scarce resource and hence relies on optimal spectrum splitting and auction to provide more efficient segments to integrate so that mobile operators can deliver better Quality of Service (QoS), particularly for IoTs. Therefore, mobile companies now engage in innovative IoT services, such as smart cities, e-healthcare, smart agriculture, etc. with 5G. Along with this trend, 5G IoT becomes the core techniques for the digital transformation into sustainable world.

Spectrum splitting and auctions are normal for the governmental authorities to allocate specific bands efficiently to the operators and maximize the release revenue. In Taiwan, the National Communications Commission (NCC) is the competent authority to hold spectrum auctions [3,4]. To ensure fairness, efficiency, and transparency, the NCC follows the regulation process and publishes the available frequencies for the mobile spectrum [4]. Operators need to prepare strategies for tender bids. The bidding spectrum is costly and competitive.

Numerous factors ought to be considered for designing an effective spectrum splitting and auction [5]. Conventional research has focused on finding different ways of auction to maximize the number of segments, the efficient allocation of segments to operators, and the revenue for government authorities. In a prior study, researchers developed an optimization model to balance multiple targets, such as the number of segments, time for auction, and costs among the authorities and operators from the perspective of revenue [5]. The results show that spectrum splitting and auctions are dynamic gaming processes that involve the government authorities and mobile operators so that spectrum splitting is not just the work of government authorities. However, quality was implied only implicitly within the number of segments related to throughput, and other quality features for communication (e.g., latency and penetration rate) were ignored. Second, the target of the optimization of the study was limited in terms of the number of segments and revenue, which is not enough for the 5G ecosystems. Therefore, we define the metrics of quality and value for the evaluation of benefits and costs and present the target for the optimization of the equilibrium of 5G ecosystems with respect to the symmetry of the positions of the participants in the coalition game of spectrum auction.

In Section 2, we present a survey on the theory of organic ecology, quality value transformation, metrics for quality and value, coalition game, optimization algorithms, and experiences of mobile spectrum. In Section 3, we demonstrate, for the spectrum planner, the prediction of the spectrum value and cost function for the optimization, market share, and Net Present Value (NPV) for the investment on spectrum and standard spectrum auction procedure in Taiwan. In Section 4, we introduce the hybrid metaheuristic optimization program to maximize the number and quality of segments and revenue and minimize the cost through a coalition game aiming at the equilibrium of the 5G ecosystem according to SMRA. For the hybrid metaheuristic optimization program, we align SA and GA to determine the optimal number of bidding rounds for improvement to obtain the optimal revenue with the anticipated number of segments in each band and quality; RO is used for bidding to determine the acceptable number of segments in each band and the quality score with respect to the cash flow and profit for each operator. Then, we analyze the results of both optimization strategies and compare the effects of the two strategies with respect to the equilibrium of participants in the auction. Last, we provide suggestions to the regions with different cultures and policies.

2. Related Works

2.1. Theory of Knowledge Value Transformation

Influenced by Taoism, Chiang developed the HyPloy Model for organic intelligent agents, which acts as a fusion function of intrinsic experiences as well as intention and extrinsic projection, including inter-promoting for cooperation and inter-inhibition for competition, to respond to the extinction of their environment [6]. With extrinsic projection, an organic agent like homo deus interacts with others to improve its behavior. Based on HyPloy, HyQVS stands for a framework for organic ecosystems that tends to improve quality and value through cooperation and competition among its organic constituents, which leads to the cooperation-competition strategy. An organic ecosystem, such as HyQVS, should have three kinds of capabilities, i.e., communication, cooperation, and coordination that altogether determine the capability for improvement [1].

The Theory of Knowledge Value Transformation is meant to concretize the improvement of features and values in HyQVS [6,7]. To conduct a value-based strategy, the quality features and values must be identified through a planning phase. Assume that the features of individuals in an ecosystem consists of two parts: a common part and a particular part. The common part is the indifference features within the ecosystem that sustains the common goal for the cooperation by orchestrated supported activities. The particular part is the distinct operation structure according to the individual domain to support its unique status as its core competence for competition. The common feature domain joins particular parts of each individual to develop a comparable feature domain for each individual. Now, the comparable feature domain can be seen as a unified feature space, preserving the landscape of the ecosystem and the detailed weights of the constituents. With the features of the ecosystem, we can prepare a transform function for the relations between the common features, individual features, and comparable feature space to generate the comparative value of each individual that builds the value proposition of the ecosystems. The function transforms the commonality and maintains the source generality. The optimization can be expressed as follows:

$$\arg mi{n}_{A,U}{\displaystyle \sum }_{c\in S,T}{\displaystyle \sum }_{i=1}^{j}L\left[y_{c_i},\left(A_c,U^T{x}_{c_i}^{\to }\right)\right]+\lambda ℒ$$

(1)

The optimization should obey a certain law or a in fact consensus. The optimal value that the optimization produces depends on the law for coordination, and the fitness of the law depends on the target for improvement, which typically compensate the internal intention, that is, the so-called optimization strategy. The performance of the optimization algorithm depends on the target for improvement.

2.2. Value Creation Ecosystem

For value creation within an organic ecosystem in sharing economy, we distinguish the roles of the members of an organic ecosystem as planners, providers, and users [6,7,8]. We define the metrics for the values of the planner, provider, and user as follows:

• The value for planners includes the planned outcome, planned benefits (incomes), costs, and product/service. These must be fixed in the planning phase, including the available resources, the expected income, costs, and outcomes of the planning phase.
• From the providers, the expected value comprises the causality, expected outcome, and expected attributes. The optimization starts from this phase.
• For the users, perceived value comprises benefits, costs, and meaningful attributes such as QoS and usage. Because it is “perceived,” it is typically comparative and often requires user experience for the checking activity. The results can trigger an improvement in the subsequent lifecycle of organic agents.

For spectrum splitting and auction in Taiwan, the spectrum planners is the Ministry of Transportation and Communication (MOTC) and the provider is NCC [3,4]. The users of the spectrum are the mobile operators.

2.3. Federation Game and Coalition Game

For the realization of the knowledge value transformation of HyQVS, we define the federation game as is a kind of cooperative evolution game with a gestalt and apply it for coordinating the organic agent of the HyQVS. The federation gestalt can unify sub-cooperative subgroups, viz. coalitions [9]. The federation should be in charge of global ICT resource planning and sharing, e.g., ITU, ICANN, and the coalitions work for regional ICT resource sharing, e.g., NCC, TWNIC (Taiwan Network Information Center). This suggests that in a union where members of higher positions or a strong subgroup collectively conclude the important decisions for the organization such as the governmental authorities, the core set would focus on stability rather than on the impartiality of all participants [6,10]. It is impossible to design a fair coordinated system, such as an auction that satisfies unanimity, non-dictatorship, and independence of irrelevant alternatives simultaneously. Therefore, the authorities should actively become involved in the planning and design of an open ecosystem. Improvements such as within HyQVS can be realized by optimizing the related collective behavior of the organic ecosystem in the cooperative game. Through such kind of cooperative game, we try determining appropriate strategies to acquire an equilibrium for better social welfare.

The organic ecosystem for inclusive sharing economy relates to a coalition between providers and users, risk prevention, and the division of rights and responsibilities among users, corresponding to the symmetry of the participants [6,9]. A successful auction must reside in a well-organized collaboration as HyQVS suggested among the authority, providers, and users. To achieve this goal, the cooperation must be fair and stable so that the participants will evolutionally benefit from it via continuous improvement.

Assume that every coalition has a payoff of a single value (which could be physical or potential) that is transferable and allowed to be redistributed to the participants like the theory of knowledge to value transformation indicates. Members of a game all look for the largest economic improvement from the coalition. The gestalt of the coalition should provide the property of synergy, that is, the value derived by the coalition should be at least as large as the sum of their respective payoffs without the coalition. The coalition should have a fair way to payoff, so the members should receive benefits in proportion to their marginal contributions. The gain from a coalition is a function of the number of customers the participants have. Because the coalition should aim at a stable equilibrium, it has to be founded on the egoism to redesign the mechanism by the core set in the game [10]. The core condition is an un-dominated imputation set that obtains a stable payoff in which no agent would suffer from the new criterion of state modification or resource reallocation, assuming that participants are rational wrt. behavior economics and the distribution of the payoff satisfies. The above reflects the strategy to keep the coalition sympathetic to help other non-dominant participants benefit from their involvement and contribute to the coalition.

The internal intention of the organic agents within an ecosystem of HyQVS is the most difficult part to deal with, because the intention of some agents may be irrational, and this is the problem that the subjective believes in the Theory of Games and Economic Behavior of von Neumann and Morgenstern could not deal with [11]. The Shapley value applies primarily in situations when the contributions of each actor are unequal, but each player works in cooperation with each other to obtain the gain or payoff. If participants want to manage stability and fairness in such games, it requires the dominant partners to make a concession [12]. An important implication of this is that the collective action in the cooperative must generate synergy [13], so it is no less than their union set. Both parties treat the other as a partner to achieve the same goals in the common domain as indicated in Equation (1). The benefits are on both sides and flow equally to each other.

Instead of mandatory norms with laws, the most valuable support or the concession and payoff comes from the moral and ethics [14]. For the coalition of providers and users, reciprocation originates from the altruism of the dominant party [15]. With respect to the inclusive share economy for sustainable ecosystems, altruism is the most incredible supports for cooperation within the coalition game due to reciprocation, and egoism on the contrary damages the cooperation [16]. This is the reason, we argue for that the sharing economy is inclusive. With an appropriate rule for the optimization of collective behaviors, the following pitfalls of egoism could be avoided:

• Spite: Providers and users treat each other as competitors in a zero-sum game [17], each developing their own protocol to try dominating the market without cooperative consensus. No one obtains direct benefit from the other side. Most importantly, spite is the root cause for security attacks and malware such as computer viruses.
• Selfishness: Users follow the existing protocol and business model of the provider and customize a specific environment for every provider. The provider receives benefits unilaterally from the users.

2.4. Pareto Optimality and Frontier

Pareto optimality is a situation where no individual or preference criterion can be better off without making at least one individual or preference criterion worse off or without any loss thereof and has been applied to the selection of alternatives in engineering and biology, in addition to economics [6,18]. The following three concepts are closely related:

• Given an initial situation, a Pareto improvement is a new situation in which some agents will gain, and no agents will lose.
• A situation is called Pareto-dominated if there exists a possible Pareto improvement.
• A situation is called Pareto-optimal or Pareto-efficient if no change could lead to improved satisfaction for some agents without losing some other agents, or if there is no scope for further Pareto improvement.

We adopt Pareto Optimality as the optimal situation for efficiency and income distribution in the coalition game grounded in Shapley value and altruism wrt. HyQVS. The Pareto Frontier is the set of all Pareto-efficient allocations [6,19]. Given a set of choices and a method for valuing them, the Pareto Frontier is the set of choices that are Pareto-efficient. By restricting attention to the set of choices that are Pareto-efficient, a designer can make tradeoffs within this set rather than considering the full range of every parameter. With respect to the inclusive sharing economy, if an innovation has surpassed the previous efficiency postulation or a new dimension is defined as the new target, the Pareto Frontier boundary comprises a quality indicator, access indicator, and usage indicator [5,16] that can be mapped to the features of 5G networks based on ISO 25000 [20] as follows:

• Quality: Efficiency that can enhance QoS.
• Access: Available spectrum and number of segments.
• Usage: Utilization ratio of resources accepted by the users against the planned resources by the providers.
• Income: Value of profits.

2.5. 5G Mobile Communication: IMT-2020 and IoT

Following the Radio-communication Assembly of the International Telecommunication Union (ITU) in 2015, 5G communication was entitled IMT-2020 (International Mobile Telecommunication-2020) [21]. The 5G mobile technique can accomplish enhanced devices and communication capabilities in terms of efficiency, throughput, and latency, as well as a high data rate, and ITU intends to use 5G to connect things, particularly in mid- and long-range [19]. Thus, 5G can enhance the QoS of the IoT.

Owing to the advent of Internet technology, the world has become more closely connected to information systems, as sensors and actuators are incorporated into a wider variety of objects and then connected to the Internet through the arrangement of wired and wireless networks, such as the IoTs [22]. Radio frequency identification (RFID) has been used for IoT, and short- and mid-range wireless networks, such as Zigbee, Bluetooth, and Wi-Fi, are suitable for situations where devices are close to the receiving gateway; however, they cannot cover long-range IoT applications, such as smart cities, e-health, and smart agriculture [2,4]. In addition to the limitation of distance, the race for bandwidth and latency may occur by these networks because of the narrow bandwidth and high frequency [2,4]. With the advent of Cloud and edge computing, mid- and short-range, very low-latency broadband is required for IoT services [23]. Mobile companies are competing to provide innovative services and better QoS for IoT as mentioned above. However, it depends on an optimal spectrum splitting and efficient auction to provide better QoS, including long range, low latency, and high penetration rate for 5G. Along with the development of 5G technology, NB-IoT has become a popular communication model for D2D-enhanced content uploading with social trustworthiness in 5G IoT [24], including in Taiwan [25].

Chiang et al. developed a RFID/NFC and mobile-computing based IoT service framework that classifies the IoT services into three categories, viz. human-enabled service, device-enabled service, and object-enabled service [26]. To realize the services, the software on mobile phones and RFID/NFC devices have to translate the unique ID (Identifier) of EPCglobal into the FQDN in the form of DNS (Domain Name System) URL (Unified Resource Locator) and then finds the specific ONS server according to the DNS URL of the web. After that, the ONS server feeds back the URL of the service corresponding to the FQDN to activate the service. Moreover, GPS was used to identify the locations of the users. In the meantime, Chiang et al. also developed a RFID/NFC based Augmented Reality (AR) system which uses RFID/NFC tags, mobile phone camera, and communication interface to realize the tracker and marker indicating the actual location of the object and can be adjusted with the user and the relative distance and angle between the subject matter when the user moves in the environment, and can change the virtual object marking for real-time location positioning [27,28]. Chiang and Yeh further developed a two-fold authentication technique with mobile phone number and e-mail address as the ID of a user, not only for the authorization but also for the account rescue [29]. The IoT service framework for solving problems in the field of O2O (Online to Offline) management of creative cultural derivatives by the National Palace Museum (NPM) in which Exhibition areas and special Exhibition (short-term social network) for antiquities and the stores selling creative cultural derivatives own their IoTs of physical things, and the online shops build the virtual world [30,31]. The users can move around the multiple IoTs with different behaviors in response to the purpose of the specific IoT environment. For the time being, the aforementioned development can be considered as the Multiple Social IoTs (MSIoTs).

The advances of mobile technologies lead to a ubiquitous computing environment. Tsai and Chiang put forward the framework in four broad areas: pervasive devices, networking, middleware, and applications in which the organic agents are seamlessly integrated into the environment and provides useful services to humans in their everyday lives on the case of the Ubiquitous Intellectual Properties Management System of NPM including multimedia Digital Archives, Digital Museum, and e-learning project [32]. Chiang and Huang have put the consensus forming as the starting point of the realization of IoT services so that the organic agents can have a meaningful social relationship with one another [33]. With regard to Hybrid Cloud with connection IoTs, Chiang has developed an approach for authentication and authorization based on OpenID and Oauth for social networks on the Cloud [34] and proposed a systematics collaboration framework with respect to interoperation [35]. On top of the overall framework, collaborative learning gamification with 3D AR courses were developed and used in practice [34]. What has been done above can be called Multiple Social IoTs and is one of the earliest pilot programs of the so-called Metaverse nowadays [36].

The study in [6] analyzed the ecosystems of different topologies with different time complexities. The costs for maintenance and participation in the ecosystems could be the extra social cost and inevitably the friction costs of consequent resource exchanges, including the trust and risk, which is especially important for SIoTs [37,38,39,40,41,42]. It is a matter of connectivity but not the bandwidth. With respect to the research results as above, the SIoTs and MIoTs [43,44,45] fall into the topology of partially centralized agents with some unified arrangements whose complexity is O(2m + ip + n + jp) or the topology of decentralized autonomous agents whose complexity is O(2m + n + i + j). The difference depends on whether the IoT ecosystem employs Client/Server architecture or serverless computing architecture. The latter can use Peer-to-Peer (P2P) communication to address the routing issue in delay-tolerance as described by [40]. Notably, the concept of the agent-oriented IoTs [46,47] belongs to decentralized autonomous systems like Blockchain as described in [6]. The results of the simulation of [47] is evidence for the complexity of O(2m + n + i + j) and the Open API will be the most important mechanism for the services of this kind of IoTs. Since the spectrum splitting and auction involve only the resource planners and providers but not the developers and users of the IoT services, the consideration for SIoT and MIoT is out of the scope of this study. However, the optimal sharing of 5G mobile spectrum can lead to the enhancement of QoS of almost all kinds of IoTs.

2.6. The Value of Spectrum and Auction

From the perspective of economics, the value of the spectrum reflects the number of users that will pay for the usage of the spectrum. Spectrum policy is highly influenced by governmental authority and socio-economic factors such as population, technique, market share, and geographic distribution. Thus, there is considerable variability in the spectrum values for different situations [48]. The prices of spectrum licenses are usually charged in $/MHz-pop [49]. According to HyPloy, intelligent organic agents can use experience to respond to external challenges. Hence, we start with a survey of the experiences of the USA, EU, and 4G in Taiwan.

2.6.1. The Value of Spectrum at 600 MHz Band and Auction

In 2016, the Federal Communications Commission (FCC) of the United States (USA) commenced the first-ever “incentive auction” to ease congestion on wireless networks and lay the groundwork for the 600 MHz spectrum [50]. The incentive auction comprised a reverse auction, where broadcasters voluntarily released some or all of their usage rights of the spectrum [51]. In addition, the incentive auction also included a forward auction for forthcoming licenses that were appropriate for tele-communication [52]. The reverse auction was used to retrieve some of the 600 MHz band from the television broadcasts, where the broadcasters announced the spectrum they submitted and the price they would like to sell with. Typically, the scale and price gradually decline while the auction proceeds. Thereafter, the forward spectrum auction can be conducted with most mobile phone carriers as the bidders, who carry out the scale and fee they are likely to take. The scale and price of the auction increases while the auction proceeds in progress, as shown in Table 1 [52]. The final result of the total bid in the USA was 19.6 billion USD for the spectrum of 70 MHz, where the average cost was USD 0.88/MHz-pop. Through the forward and reverse auctions, the spectrum providers and users interacted with each other in the manner of competition, and the final price was determined by the compromise of both parties. In this situation, users had a symmetric position to the providers.

Table 1. 600 MHz incentive auction in USA.

		Stage 1	Stage 2	Stage 3	Stage 4
Provider	$	88.4	54.6	40.3	10.5
	MHz	126	114	108	84
buyer	$	23.0	20.5	19.7	19.6
	MHz	100	90	80	70

2.6.2. 800 MHz Band

Aiming at the transformation of the digital dividend to economic growth as well as public profits, the European Commission Decision unified the technical standard and highlighted the significance of the 800 MHz spectrum as a fragment for digital dividends [53]. Therefore, rational investors must adjust nominal returns to changes in inflation in the dividend process and in the discount factor. This revealed that the EU cared more regarding fairness and social welfare. Since 2012, the EU has initiated auctions for a spectrum of 800 MHz [52].

2.6.3. L-Band

ITU presented the fresh bandwidths assigned to telecommunication services into consideration in 2015: the L-Band (1427–1518 MHz) and C-Band (3400–3600 MHz) [54,55]. There is no auction of L-Band before, therefore the value of the spectrum is achieved through forecast. The feature of propagation of 1.4 GHz lies between 800/900 MHz and 2.1 GHz. However, the extent of the 1.4 GHz is superior to that of the 2.1 GHz. The status quo reveals that the usage of 1.4 GHz is the same as the usage of 800/900 MHz. Hence, L-Band is anticipated to take features analogous to the sub-frequency below 1 GHz. This signifies that the value of the 1 GHz spectrum represents the upper-bound of 1.4 GHz.

2.6.4. C-Band

C-Band lies between 3400 and 3600 MHz. The EC Decision 2014/276/EU on the 2nd of May 2014 clarified the orchestrated technical requirements and arrangements of the frequency within 3.6 GHz spectrum. This is because the C-band has been used for the satellite but not for the IMT for 5G. Thus, only the spectrum value was estimated.

2.7. Cost Function

The spectrum consists of two parts: the cost of spectrum acquisition and equipment construction. The cost for spectrum acquisition is determined by the value of a frequency segment that is calculated with the value of spectrum, 10 MHz per segment according to frequency-devising duplexing (FDD), and 30 million users of mobile communication in Taiwan. The spectrum value is determined by the revenue from the auction for the spectrum provider, such as NCC, however is the variable cost for the users such as the mobile operators, which will be determined during bidding. In contrast, the equipment construction cost is fixed for the spectrum. The cost of the equipment construction can be obtained by multiplying the price of one base station by the number of base stations in each band.

The cost for the base station after 2020 can be predicted based on the highest price of the 4G base stations, that is, NTD 1.8 million. Consequently, the estimated cost for the 5G base station after 2020 can be estimated from the price for 4G multiplied by the average inflation rate, that is, NTD 1,887,317.

Based on the data for 4G in Taiwan, we can speculate on the minimum number of base stations required for 5G. The area covered by one base station for the 600 MHz band is 125% of that for the 700 MHz band. Hence, the number of base stations for the 600 MHz band is 80% of that of the 700 MHz band [56]. The features of the propagation at 800 MHz and 900 MHz were very similar [57]. The feature of the propagation of the L-Band lies in an intermediary value between that of the 900 MHz band and the 2.1 GHz band. The area covered by the L-Band is 150% of that of the 2.1 GHz band and 75% of that of the 900 MHz band. The number of base stations for the L-Band is thus 133% of that for the 900 MHz band [58]. The coverage of the C-Band is approximately 187% smaller than that of the 2.3 GHz band, and the area covered by the 2.3 GHz band is 180% smaller than that of the 900 MHz band, the number of base stations required for the C-Band is thus 336.6% of that for the 900 MHz band [59].

2.8. Finance of the Companies

The returns for the operators can be calculated using the NPV [60] and average revenue per user (ARPU) [5]. The cost function here is the result provided by the revenue deduct cost. To assess the value of an investment, we often apply the Weighted Average Cost of Capital (WACC) to the discount rate for the future cash flow to derive the NPV for investing a business [8,60].

2.8.1. WACC

WACC can be applied to evaluate the total cost of the company’s capital level and determine the overall financing cost. The most important factor is the discount rate while monitoring cash flow from the extension of the present processes of a company. Hence, the capital cost should be considered as the minimal required return on investments (ROI) [5,60].

2.8.2. NPV

NPV can be applied to assess an investment plan by transforming the forthcoming cash flow into the value of the initial investment. Because NPV considers all cash flows during the investment, NPV represents the basic rate for several companies while assessing investments. NPV is calculated as follows:

NPV = [C_t/(1 + r)^t] − C₀,

(2)

where C_t represents the net cash flow within the time span, C₀ represents the initial cost of the investment, r represents the discount rate, and t represents the number of time spans [5,60].

2.9. Spectrum Auction in Taiwan

According to “Regulations for Administration of Mobile Broadband Businesses” in Taiwan [4], a company who plans to run operations has to obtain a concession license from NCC and the operating region shall pertain nationwide [61]. The spectrum bidding process starts with the announcement of the spectrum to be released by the MOTC. Then, the NCC declares the spectrum and number of segments that will be issued. Accordingly, operators can choose bands to bid and submit applications with tender bonds. The NCC will examine these applications and announce qualified bidders. The bidding process is based on SMRA [3], that is, the bidders can “switch freely” among different spectrum segments. All bidding targets terminate simultaneously.

2.10. Hybrid Metaheuristics for Optimization

There are many well-known optimization algorithms originating from different scientific disciplines, such as random optimization (RO) from mathematics, SA from physics, and GA from biologics, each of which has the underlying spirit of the specific discipline. Researchers have proved the effects of RO, SA, and GA for spectrum splitting. However, in location-based bidding, there are several permutations of segments, and the aggregated result can vary and converge to the local optimum. This is interpreted as a viable characteristic of the nondeterministic result of the authorities’ revenue because it is opportunity revenue [5,6]. Chiang has classified the open ecosystems into five types of topologies, among those the type of multi-agents with unified rules which can be optimized with SA has the best efficiency and the type of multi-agents without unified rules which can be optimized with GA is in the second place [5]. Therefore, we apply SA and GA to the coalition game in the case of spectrum splitting and auction and compare the results.

Particle Swarm Optimization (PSO) is a well-known optimization method that originates from social behavior science. Because of the problem of local optimization and the emphasis of self-organization without centralized control [61,62], it is more suitable for partially centralized agents with some unified arrangements [6]. The Fruit Fly Algorithm (FFA) is a type of cooperative coevolution based on the Parisian approach. A global fitness function evaluates the quality of the population as a whole; only then the fitness of an individual is calculated as the difference between the global fitness values of the population with and without the particular fly, whose individual fitness function has to be evaluated; the fitness of each fly is considered as a level of confidence [63]. The main difference between the FFA and with PSO is that the FFA is not based on any behavioral model but only builds a geometrical representation. It belongs to the topology of decentralized autonomous agents as formulized in [6]. Neither Particle Swarm Optimization (PSO) nor Fruit Fly Algorithm (FFA) are suitable for coalition games for spectrum splitting and auction so that we will not consider it in this study.

2.10.1. Random Optimization (RO)

The name RO is attributed to Matyas [64], who made an early presentation of RO along with basic mathematical analysis, which is also known as direct-search, derivative-free. RO works by iteratively moving to better positions in the search space, which are sampled using, for instance, a normal distribution surrounding the current position.

Rastrigin and Schumer [65] proposed optimization methods that did not require gradient properties using adaptive step size random search in a single mode. Thus, RO can be used on functions that are not continuous or differentiable, so that it is not convex. Baba [66] further proved that the global minimum can be reached with a probability of one, even if the performance function is multimodal or the differentiability is not provided.

The RO algorithm showed that for a minimization problem, each round of the variable selection was independent. Starting from a random vector, RO can minimize the risk resulting from randomness. RO is easy to understand and can generally serve as a baseline for optimization. Most importantly, the cooperative game based on Shapley value and Pareto Optimality needs a property of random walk which is the advantage of RO.

Normally, there are two basic strategies for heuristic search: “divide-and-conquer” and “iterative improvement” [67]. Because RO applies an iterative one that fits the purpose of continuous improvement and does not conquer, it favors symmetry in the auction.

2.10.2. Simulated Annealing (SA)

Kirkpatrick et al. [68] proposed a simulated annealing algorithm to connect statistical physics to the optimization problem and approximate the global optimum of a given function. The term “annealing” denotes an analogy with thermodynamics involving the method that metals cool and alter their properties.

SA uses temperature to control the magnitude of the changes in the objective function and is effective in handling certain optimization problems that are out of control using combinatorial methods when the number of objects becomes large [69]. Lin et al. use nearest-neighbor heuristics in accelerated algorithms of SA optimization problems [70]. Chiang et al. adopted a Boltzmann machine to find the next best temperature to eliminate the local optimum and reach the global optimum in relation to the equilibrium for the global state [71]. Therefore, it is more suitable for the goal of this study.

SA concurrently extends both divide-and-conquer and iterative improvements. When properly used, SA can maximize revenue and minimize costs simultaneously [70,71]. Because it seeks the equilibrium and does not emphasize the conquer only, it can sustain the stability of the 5G ecosystem in the auction.

2.10.3. Genetic Algorithm (GA)

John Holland [72] developed a GA, which has been largely used in solving optimization problems. Within the GA, individuals adapting to their environment will have a better chance of surviving and reproducing [73]. It stands for the process of evolution, where the fittest ones are chosen for reproduction to generate progeny of the next generation.

Although GA is deemed robust, there are still some problems, such as the difficulty in choosing the population size, crossover, and mutation rate, because they are principally experimental parameters [74]. Furthermore, GA cannot always achieve an optimal solution, whereas premature convergence may occur when the fitness values of potential solutions stop improving, and an optimal value has not been reached [74,75]. According to our experiences, the biggest problem of GA is that, when the initial condition is wrong, the evolution process will be out of order in terms of Chaos Theory. In a coalition game, the fittest one (i.e., the elite) could be out of control of the authority and damage the equilibrium. GA applies a conquer, and it is possible that the GA does not match the goal of symmetry.

The term metaheuristic was created and is now well accepted for general techniques which are not specific to a particular problem. Many combinatorial optimization problems are very hard to solve optimally, the quality of the results obtained by somewhat unsophisticated metaheuristics is often impressive. A new kind of approximate algorithm has emerged which tries to combine basic heuristic methods in higher level frameworks aimed at efficiently and effectively exploring a search space [67]. These methods are today commonly called metaheuristics. Such a combination of one metaheuristic with components from other metaheuristics is called a hybrid metaheuristic. For the optimization of spectrum splitting and auction, we develop different kinds of hybrid metaheuristics for cooperative evolution with multiple objectives including revenue, number of segments, quality features, etc. and test them with respect to the goal of coalition game of the 5G ecosystem. For example, we can build the hybrid metaheuristics in using the RO to compensate the effect of divide-and-conquer of GA, for the optimization problems of coalition games.

3. Tasks for the Spectrum Providers

3.1. Prediction of Spectrum Value

3.1.1. Prediction of the Spectrum Value for 600 MHz Band

The final average cost of the bid of the USA in 2017 was USD 0.88/MHz-pop in 19.6 billion USD for the spectrum of 70 MHz at 600 MHz (Table 1). The consumer price index (CPI) in February 2017 was 99.96, and the CPI in February 2018 was 102.15. Consequently, the inflation rate is as follows:

Inflation Rate = (CPI₂₀₁₈ − CPI₂₀₁₇)/CPI₂₀₁₇ = 2.19%

(3)

The spectrum value of 600 MHz band after 2020 can be estimated using the inflation rate shown in Equation (3):

After-Inflation = Cost × (1 + Inflation-Rate)^year = $28.41

(4)

The cost an operator spends will be NTD 8.523 billion/segment, that is, equal to NTD 28.41 × 10 MHz × 30 million consumers.

3.1.2. Prediction of the Spectrum Value for 800 MHz Band

Since 2012, the EU has initiated auctions for a spectrum of 800 MHz [5]. The potential value of the spectrum of the 800 MHz band after 2020 can be predicted using the outcome of the EU’s auctions. We use the average inflation rate from the month of the auction to the same month of 2017 to calculate the spectrum value of the 800 MHz band. The potential value of the 800 MHz band should be the average price of NTD 27.534/MHz-pop, therefore the price for a segment will be NT$ 8.2602 billion in Taiwan.

3.1.3. Prediction of the Spectrum Value of L-Band

Plum performed three levels of prices for L-Band: high, medium, and low [76]. The high-level stands for €0.60/MHz-pop and is the mean value of the upper-bound price of those below 1 GHz. The medium-level price stands for €0.40/MHz-pop, where the lower-bound price is that of those below 1 GHz. The low-level stands for €0.25/MHz-pop, which signifies the price that ought to be significantly lower than that of others below 1 GHz. Consequently, the spectrum value of 1.4 GHz can be outlined in Table 2.

Table 2. The spectrum value of 1.4 GHz.

Low	1.4 GHz Value <<< sub 1 GHz Value	0.25 €
Medium	Value of GHz stands for the lower bound of sub 1 GHz spectrum	0.40 €
High	value of GHz lies ca. in the middle of the upper bond of sub-1 GHz	0.60 €

Considering the practical status in Taiwan, we used a high price level for the prediction. The potential spectrum value of L-Band was for NTD2 5.247/MHz-pop. From June 2011 to June 2017, the average inflation rate was 0.888%. Accordingly, the predicted spectrum value of L-Band after 2020 was NTD 27.338/MHz-pop, and the fee for one segment was NTD 8.2014 billion.

3.1.4. Prediction of the Value of Spectrum of C-Band

The value of the spectrum of C-Band can be estimated by referring to the economic indicators that result from the auction, national population, purchasing power parity (PPP), and CPI from the World Bank. Hereby, the global average price is approximately 0.0152 €/MHz-pop, and the EU average price is 0.0084. However, the global average price remains at approximately 0.0116, drawing on the average price data since 2010 [5]. In a report by the Commission for Communications Regulation (ComReg), the prices were predicted to be €0.0152/MHz-pop according to the worldwide data, €0.0084/MHz-pop according to the EU’s data, €0.0116/MHz-pop according to the worldwide data after 2010, and €0.0380/MHz-pop according to the EU’s data after 2010.

Because the C-Band had been used for the satellite but not for IMT for 5G previously, we applied the price calculated using the worldwide data after 2010, that is, NTD 0.415/MHz-pop. The inflation rate was 0.96%. Hence, the spectrum value after 2020 was NTD 0.431/MHz-pop, and the price for one segment was NTD 129.3 million.

3.2. Estimation of the Cost Function

The costs for the spectrum consist of the spectrum acquiring cost and equipment construction cost. The spectrum values discussed above indicate the revenue from auction for the spectrum provider, such as NCC, but is the variable cost for the users, such as the mobile operators, which will be determined during the bidding as the forward spectrum auction can be conducted in the USA for the 600 MHz band.

In contrast, equipment construction costs are fixed. According to the discussion and based on the data provided by Telecom Technology Center (TTC) [77], we can estimate the following numbers. The number of base stations for 600 MHz was 1120 (equal to 1400 × 0.8). The number of base stations for 800 MHz was 1900 (equal to that for 900 MHz). The number of base stations for L-Band was 2527 (equal to 1900 × 1.33). The number of base stations for the C-band was 6395 (equal to 1900 × 3.366) [78]. We can calculate the cost of the equipment per segment by multiplying the number of required base stations and the cost of NTD 1,887,317 together.

3.3. Quality Features and Costs of the Spectrum

Because the purpose of this study is to enhance the QoS, we need metrics to measure quality. According to ISO 25000, the basic attributes of quality include efficiency, throughput, and latency. The throughput depends on the bandwidth in relation to the number of segments that will be determined in the auction. Therefore, in this phase, we considered the efficiency factor that affected the quality in this phase.

The higher the frequency, the higher the latency, the shorter the transmission distance, and the lower the penetration rate. Therefore, we assigned the quality score to each frequency band with a quality score of 4 to 600 MHz because of the lower frequency and a score of 1 to 3600 MHz (L-band) because of the highest frequency. The quality score is a comparative value among the frequency bands used to measure the QoS. Finally, the price and quality score are calculated per unit of total cost, as shown in Table 3.

Table 3. Quality features and costs of the spectrum.

Per Segment	600 MHz	800 MHz	1400 MHz	3600 MHz
Price/Seg.	8.523	8.2602	8.2014	0.1293
Equip Cost	2.1138	3.5859	4.7693	12.0694
Total Cost	10.6368	11.8461	12.9707	12.1987
Price/Cost	0.8	0.7	0.62	0.01
Qos Score	4	3	2	1
Score/Cost	0.38	0.25	0.15	0.08

3.4. Financial Status of the Operators

The number of consumers/SIM-cards for mobile communication is approximately 30 million in Taiwan. In Taiwan, there are five mobile operators: CHT, TWM, FET, TSTAR, and APTG. After the information announced by the NCC regarding the market share in the 3rd quarter of 2017 (Table 4), we calculated the number of consumers that each operator ought to serve, as shown in Table 4 [79].

Table 4. Market share of the mobile operators in Taiwan.

	CHT	TWM	FET	TSTAR	APTG
Customers	10,941	7566	7494	2073	1926
Market Share	36.47%	25.22%	24.98%	6.91%	6.42%
Rank	1	2	3	4	5

Thereafter, the ARPU was calculated for each operator and transformed into that of 2020. The average inflation rate was 0.74% within Q3 of 2017. Thereafter, the ARPU was 682.4 for the CHT, 647.7 for TWM, 775 for FET, 502.3 for TSTRT, and 462.5 for APTG [79]. The forecasts of ARPU in 2020 were 697.66 for CHT, 662.19 for TWM, 792.33 for FET, 513.53 for TSTAR, and 472.84 for APTG (Table 5).

Table 5. The average revenue per user (ARPU) of the mobile operators.

ARPU	CHT	TWM	FET	TSTAR	APTG
2017 Q3	682.4	647.7	775	502.3	462.5
2020	697.66	662.19	792.33	513.53	672.84
Rank	1	2	3	4	5

Considering the discounting rate, a positive NPV indicates that the return rate of investment is higher than the discount rate. We often apply WACC to set the discount rate [5] and consider the capital cost of WACC as the minimal required ROI [5]. We documented the WACC of each mobile operator in Taiwan for five consecutive days starting from 6 March 2019 [5]. Because there were no available data on TSTAR, we used the WACC of the other four operators to estimate the discount rate, that is, 6.22%.

The monthly revenue of each operator can be calculated by the ARPU times the number of customers of each operator (Table 6). The NPV of the 15-year (180 months) license can be calculated using the discount rate of 6.22%, as shown in Equation (5). NPV considering future cash flow for operators to invest in spectrum auctions [4,58].

$$NPV={\displaystyle \sum }_{t=1}^{180}\frac{ARP{U}_t X User No.}{{\left(1+6.22\%\right)}^t}$$

(5)

Table 6. Revenue of the mobile operators in Taiwan.

	CHT	TWM	FET	TSTAR	APTG
ARPU	697.66	662.19	792.33	513.53	472.84
Customer	10,941	7566	7494	2073	1296
Revenue	7,633,098	5,010,130	5,937,721	1,064,548	612,801
Ranking	1	2	3	4	5

3.5. Standard for Spectrum Auction in Taiwan

The NCC adopted SMRA for the 4G broadband auction [3]. In the 1st phase, buyers competed against quantity-based bidding in multiple rounds. Because each mobile operator only bids the spectrum, but not the segments inside, we assumed that the base prices of all segments were deemed to be equal in the first round. In the 2nd phase, the competition was based on location-based bidding. All nominated bidders must issue a letter of intent for the frequency location. When the frequency location carried out did not overlap, the nominated frequency location was determined based on the aforementioned letter. If the segment location could not be determined, a one-round session was held, and the price for bidding was counted based on units of NTD 1 million. The contiguous spectrum eliminated fragmentation and avoided degrading the provisioning efficiency [80], that is, a plus for quality. We assumed that companies preferred to pay more to obtain segments of the same frequency.

4. Methodology and Experiment

4.1. Optimization Program

According to SMRA, the optimization program was designed in three phases. The first one that decides the best iteration of such—to decide the optimal iteration of auction—is called the time algorithm (TA), where the number of iterations indicates the rounds required by the participants to improve their decisions, it is the most significant part of our optimization approach to realize the continuous improvement of the organic agents’ behavior and the ecosystem as a whole. The second one allocates the optimal segment in each frequency—to obtain the optimal revenue for the number of segments cut in each band—and the total costs for the operators, that is, the expected outcome of the provider, such as NCC, was named with the segment algorithm (SA). The last one that works to maximize revenue by allotting segments to the operators under cost restriction is called bidding algorithm (BA) that will determine the perceived value of benefits and costs producing meaningful attributes, whereby this study emphasizes quality.

As shown in Figure 1, the segment algorithm can approximate the optimal number of segments within each band for the auction. These numbers are used to simulate the bidding scenario and calculate the cost and revenue of the operators. If the revenue is higher than the cost, the operator will consider the segment, and the segment is assigned to this operator. Otherwise, the segment is assigned to another operator. Repeat this procedure to the next auction segment until all operators are unable to bid or all segments are allocated.

Figure 1. Flow of three layers optimization.

We develop two kinds of hybrid metaheuristics, SA and RO, as well as GA and RO, and then compared the effects of both. For the time and segment algorithms, we used SA and RO, respectively. Because the decision in bidding may be influenced by random noise such as news and disturbance of sudden deficit of cash flow, and even irrational behaviors of some operators, we used RO for the bidding algorithm, which could minimize the risk resulting from randomness and lead to an improvement in bidding. For this purpose, we had to modify the algorithms for spectrum splitting and auction for SMRA.

4.1.1. Simulated Annealing (SA)

To cope with the time and segment algorithms, we formulated simulated annealing as described below (see Table 7).

Table 7. Simulated Annealing.

Step 1	Let i = 0, to choose an initial solution x randomly.
Step 2	To generate a solution y from the neighbor S(x) of the current solution x randomly. Set i = i + 1.
Step 3	To replace x by y with the probability $p\left(x\to y\right)=1-min\left[1,e^{\frac{∆f\left(x\to y\right)}{C_i}}\right]$, where $∆f\left(x\to y\right)=f\left(x\right)-f\left(y\right)$ $C_i$ is the cost function with initially higher temperature that makes ∆f near zero and easily switches current best solution from x to y After the process evolves several times and gets cooler, only ∆f large enough will alter existing solutions
Step 4	If i ≥ maximum iteration number, or Ci is lower than the specific threshold then stop, else go to Step 2.

4.1.2. Genetic Algorithm

To cope with the optimization of spectrum splitting and auction, the GA consists of the steps as shown in Table 8.

Table 8. Genetic Algorithm.

Step 1	To create the initial population S(t), where t = 0. And, to determine the size of the population (POP) and the number of generations (GEN).
Step 2	To figure out the fitness of each member
Step 3	To calculate the probability of selection for each population S(t), where the probability can be formulated as $p\left(S_i\left(t\right)\right)=f\left(S_i\left(t\right)\right)/{\displaystyle \sum }(f\left(S_i\left(t\right)\right))$
Step 4	To select a pair for the reproduction with $p\left(S_i\left(t\right)\right)$
Step 5	To apply the crossover, mutation as well as inversion operators to the selected pairs. To replace them with the resulting progeny to set up a new population, S(t + 1). If the size of the new population is equal to POP, then go to step 6, else go to step 4.
Step 6	If current generation equals to GEN, then stop, else go to step 2.

4.1.3. Hybrid metaheuristics of SA and RO

The bidding RO revolves around Segment SA until the termination condition is satisfied. The segment SA randomly changes the number of segments and throws the value to the bidding RO to obtain the next optimum result repeatedly. Then, time SA changes the number of iterations and throws it to the segment SA again to obtain the variance, optimal revenue, and the optimal number of segments. When the time SA stops, the outcome of the minimal variance becomes optimum, which includes the optimum spectrum splitting, the optimum assignment to each operator, and the optimal auction revenue.

4.1.4. Hybrid Metaheuristics of GA and RO Algorithm

As previously mentioned, the algorithm consists of three parts: the time GA, segment GA, and bidding RO. The time GA produces the optimal number of iterations for improvement. Segment GA was used to optimize the number of segments from spectrum splitting. The bidding RO assigns segments to operators.

First, the time GA randomly obtains the population size of iterations and then passes the variables to segment GA. After segment GA obtains the iteration variables, it randomly obtains the population size of the segments and throws these variables to the bidding RO. Then, the bidding RO randomly assigns segments to the operators while receiving the segment variables, and the optimization process proceeds with respect to the cost function. The bidding RO processes are repeated until all the operators are unable to bid or all segments are assigned. Bidding RO indicates a process resulting from dealing with the population of segment GA. Thus, the best value becomes optimum for this population. After sorting, segment GA removes some segments at a certain rate and completes the population using crossover and mutation, and passes the new population to the bidding RO. The entire process repeats sorting, elitism, crossover, and mutation until the maximum generations appear. Segment GA is repeated according to the number of iterations of time GA, and then the program records the variance. The complexity should be O (2m + n), where m is the number of spectrum providers and n is the number of users.

4.2. Experiment Results

According to previous studies [1,6,71], we set the cooling rate of SA as 0.95; for GA, we set the mutation probability as 20%, population size as 10, and crossover probability as 30%.

4.2.1. Result of SA and RO

In the simulation, all variances are produced through optimal revenues under different number of iterations. The goal of the time algorithm is to find the best stable number of iterations; thus, we chose the one with minimum variance. As a result of time SA, the best round was 11, and the minimal variance was 7158.

The revenue of the spectrum auction is shown in Table 7, which presents the optimal revenue of the combination of segments by the iteration from time SA. We obtain the maximum revenue of NTD 185.94 billion at the 4th combination. That means the best result can be finished at most in eleven rounds of improvement of bidding in our case.

The segment combinations obtained by bidding RO are shown in Table 7, which represents the combination of segments corresponding to the optimal iteration in the 11 rounds. From Table 7, we also learn that the maximum revenue is NTD 185.94 billion by the combination of 12 segments at 600 MHz, 6 segments at 800 MHz, 4 segments at 1400 MHz, and 10 segments at 3600 MHz spectrum. The total cost, including equipment construction costs, resulting from this combination is 372.59 billion (Table 9).

Table 9. Result of SA revenue and bidding RO.

No.	1	2	3	4	5	6	7	8	9	10	11
Rev.	158.41	175.47	183.62	185.94	184.4	177.43	161.26	177.04	167.5	177.44	161.88
600	4	5	6	12	8	7	8	8	7	10	11
800	11	8	6	6	6	6	9	5	3	3	3
1400	4	8	10	4	8	8	2	8	10	8	5
3600	5	9	7	10	8	20	18	15	8	14	18

The combination of the spectrum segments for each operator presents the optimal situation obtained by assigning the segments to the operators in the 4th round of the improvement, as shown in Table 10. Comparing the expected optimal value for NCC resulted from segment SA and the actual perceived value for the operators resulted from bidding RO, the operators obtain fewer segments and lower quality score as planned for NCC; however, they can reduce the total costs and increase the number of segments and quality scores per total costs than that planned for NCC. It indicates that the operators are in symmetric position as NCC in the auction. From the benefit of the operators, Company C wins highest number of segments and quality score and obtains the highest quality score per cost. The result matches the situation of market share as shown in Table 10. Moreover, even Companies D and E have relatively better performance in the auction in comparison to their market shares. This indicates that the optimization results in symmetric positions of the operators corresponding to their market share.

Table 10. Result of SA revenue and bidding RO.

		600	800	1400	3600	Total	Util. Rate	Min Price	Total Cost	V/P
Opt	S	12	6	4	10	32		185.94	382.59	0.09
	Q	48	18	8	10	84				0.23
A	S	6	1	2	0	9	0.28	75.8	101.61	0.09
	Q	24	3	4	0	31	0.37			0.31
B	S	1	3	0	2	15	0.15			0.09
	Q	4	9	0	2	15	0.18			0.21
C	S	2	2	2	1	7	0.22	50.1	83.11	0.08
	Q	8	6	4	1	19	0.23			0.23
D	S	1	0	0	0	1	0.03	8.52	10.63	0.09
	Q	4	0	0	0	4	0.05			0.38
E	S	1	0	0	0	1	0.03	8.52	10.63	0.09
	Q	4	0	0	0	4	0.05			0.38
Total	S	11	6	4	3	24	0.75	176.56	276.55	0.09
	Q	44	18	8	3	73	0.87			0.26
Rev.		93.75	49.56	32.81	0.39	176.56	0.08	185.94	285.99	0.26

4.2.2. Result of GA and RO

The maximal iteration number from the time GA was 5 and the variance is 290. The maximum revenue from the segment GA was NTD 208.92 billion from the 3rd round. The optimal combination in each of the 5 rounds. The maximal revenue results from the combination of 7 segments of 600 MHz, 8 segments of 800 MHz, 10 segments of 1400 MHz, and 9 segments of 3600 MHz spectrum, also shown in Table 11.

Table 11. GA revenue and bidding RO.

No.	1	2	3	4	5
Rev.	176.21	185.62	208.92	185.93	175.93
600	9	7	7	8	9
800	3	6	8	8	8
1400	9	9	10	6	4
3600	7	20	9	19	3

The combination of the spectrum segments for each operator from bidding RO is shown in Table 12 that presents also the optimal situation obtained by assigning the segments to the companies in the 3rd round. Comparing the expected optimal value for NCC resulted from segment SA and the actual perceived value for the operators resulted from bidding RO, the operators obtained fewer of segments and lower quality score than planned for NCC; however, they can reduce the total costs and increase the number of segments and quality score per total costs than that planned for NCC. This indicates that the operators also have a symmetric position to NCC in the auction.

Table 12. Result of GA and RO.

		600	800	1400	3600	Total	Util. Rate	Min Price	Total Cost	V/P
Opt	S	7	8	10	9	34		208.92	408.72	0.08
	Q	28	24	20	9	81				0.2
A	S	4	2	2	1	9	0.26	67.14	104.38	0.09
	Q	16	6	4	1	27	0.33			0.26
B	S	3	1	2	0	6	0.18	50.23	69.7	0.09
	Q	12	3	4	0	19	0.23			0.27
C	S	2	5	0	0	7	0.21	58.35	80.5	0.09
	Q	8	15	0	0	23	0.28			0.29
D	S	0	0	0	1	1	0.03	0.13	12.2	0.08
	Q	0	0	0	1	1	0.01			0.08
E	S	0	0	0	1	1	0.03	0.13	12.2	0.08
	Q	0	0	0	1	1	0.01			0.08
Total	S	9	8	4	3	24	0.71	175.99	278.99	0.09
	Q	36	24	8	3	71	0.88			0.25
Rev.		76.71	66.08	32.81	0.39	175.99	0.08	208.92	311.92	0.23

Comparing the benefit of the operators, Company A wins highest amounts of segments and quality score and obtains the highest quality score per cost but lower quality score than Companies B and C. Company C wins the best combination of segments with one contiguous spectrum in 800 MHz with the highest quality to costs ratio. The result does not match the situation of market share (Table 4); however, the consequence is that Company C has the highest APRU (Table 5), that is, higher revenue from its customers so that it is in the best position in the bidding. Companies D and E fall into the worst situation in which they can only obtain one segment in 3600 MHz that leads to the highest cost of equipment construction and the worst QoS with regard to penetration rate. This indicates that GA results in asymmetric positions of the operators in the auction.

4.3. Analysis and Discussion

The results of both hybrid metaheuristics reflect the difficulties the operators are facing. Operators on the one hand need to collect available information and learn the actions of other competitors; on the other hand, they need to adjust their strategic reaction according to the result of each round and make decisions based on their financial constraints. According to the regulation SMRA, operators can bid in location-based bidding for the frequency location selected from several permutations of segments, so the aggregated result could vary and converge to the local optimum. It is this viable characteristic that affects the non-deterministic result of the revenue of the authority.

Due to the random walk property of RO, the experiment converges quickly while assigning the segments to each operator. This conforms to the status of the operators’ decision in the last moment that all decisions could be based on limited information and the speculation of competitors’ reaction and are required to be made in the instant.

The result of SA shows that the governmental authority can earn more revenue with optimal expected combination of segments, i.e., 185.94 billion for 32 segments in total and a quality score of 84, and brings the total cost of 372.95 billion, as shown in Table 13. The attributes include 0.09 segments per costs and 0.23 quality score per cost. With bidding RO, the actual perceived benefits counts for 24 segments in total with a quality score of 73 and the total cost is 285.99. The meaningful attributes are 0.09 segments per costs and 0.26 quality score per cost, utilization rate are 0.71 for segments and 0.88 for quality. In the auction, the operators obtain fewer segments and lower quality score than the provider NCC can expect, but still can reduce the costs and increase the quality to cost ratio. Further, the utilization rate shows the operators have accepted the resources just fitting for their usage and avoid waste. It means, following the strategy of SA, the operators are in the symmetric position to improve the outcome of spectrum splitting and auction and favor themselves especially to win the more of quality with lower costs.

Table 13. Result of Simulated Annealing and Random Optimization.

		600	800	1400	3600	Total	Util. Rate	Min Price	Total Cost	V/P
Opt	S	12	6	4	10	32		185.94	382.59	0.09
	Q	48	18	8	10	84				0.23
Total	S	11	6	4	3	24	0.75	176.56	276.55	0.09
	Q	44	18	8	3	73	0.87			0.26
Rev.		93.75	49.56	32.81	0.39	176.56	0.08	185.94	285.99	0.26

The result of GA shows that the governmental authority can earn even more revenue with optimal expected combination of segments, that is, 208.92 billion for a total of 34 segments and a quality score of 81, resulting in a total cost of 408.72 billion, as shown in Table 14. The expected attributes include 0.08 segments per costs and 0.20 quality score per cost. With bidding RO, the actual perceived benefits count for a total of 24 segments with a quality score of 71 and a total cost of 311.92. The meaningful attributes are 0.09 segments per costs and 0.23 quality score per cost, utilization rate are 0.75 for segments and 0.87 for quality. In the auction, the operators obtain fewer segments and quality scores than the provider NCC expects, but can still reduce costs and increase the quality to cost ratio. This indicates that following the strategy of GA, the operators are in a position to improve the effect of spectrum splitting and auction and favor themselves.

Table 14. Result of Genetic Algorithm and Random Optimization.

		600	800	1400	3600	Total	Util. Rate	Min. Price	Total Cost	V/P
Opt	S	7	8	10	9	34		208.92	408.72	0.08
	Q	28	24	20	9	81				0.2
Total	S	9	8	4	3	24	0.71	176.56	278.99	0.09
	Q	36	24	8	3	71	0.88			0.25
Rev.		76.71	66.08	32.81	0.39	175.99	0.08	208.92	311.92	0.23

Comparing the results of SA and GA, the GA strategy does not only bring more benefits to the provider, for example, NCC, but also higher costs that the users have to suffer and its usage of segments is higher than that of SA and the quality is lower than that of SA. This indicts, GA wastes resources and gains low quality so that it is not a ideal way for sustainable development. This appears that the GA supports egoism of the provider NCC and fittest one so that the operators are in the asymmetric position in the game and thus suffer from costs, which indicates that the provider and the fittest one such as Company C receive most benefits unilaterally and that the coalition wants to avoid. Secondly, GA delivers lower value/price ratio, that is, number of segments per cost and quality score per cost than SA does. This indicates that a system following SA is more efficient than that following a GA from the perspective of inclusive sharing economy. For the situations of deadlock of policy or difficulty of building the law or a consensus, it is a possible way to enter the 5G ecosystem. Last but not least, we found in the simulation that the approach with SA and RO runs significantly faster than that with GA and RO with the same computer setting. It could be a sign that the way of SA and RO can accelerate the decision in the coalition game of auction.

5. Conclusions

For the enhancement of the QoS for 5G IoT, the goal of this study is to optimize the 5G spectrum splitting and auction from the perspective of inclusive sharing economy. Drawing on the Theory of Quality Value Transformation, we developed a 5G ecosystem as a sustainable organic ecosystem constituted of planners and providers and put forward altruism to sustain the equilibrium of 5G coalition. Thereafter, we adopted Pareto Optimality as the target for optimization for inclusive sharing economy and consider the optimization as a cooperative evolution process. For the measure of quality and value of services, we defined the metrics with respect to Pareto Frontier for value creation based on ISO 25000.

On the top of the aforementioned open cooperation evolution framework for the coalition game, we conducted a survey to find the available bands for 5G and estimated their values and the cost of the equipment construction for each available band. We also gathered information about the number of customers each mobile operator serves in Taiwan to calculate their market shares and APRU and then derived their NPVs for the investment on the 5G spectrum.

Taking SMRA as the standard rule for the optimization of spectrum auction, we developed a novel optimization program with three phases, viz. the time algorithm determines the best number of the bidding runs, the segment algorithm handles the optimal number of segments in each bind that can bring the maximal revenue, and lastly, the bidding algorithm assigns the segments to the operators according to the perceived value of benefits, including quality and costs that produce meaningful attributes. Then, we designed two hybrid metaheuristics for the cooperative evolution with multiple objectives of quality, usability, and costs. For the overall optimization, we applied SA and GA to implement the time and segment algorithms, respectively, and then used RO to implement the bidding algorithm.

The experimental results show that the coalition game for 5G spectrum auction is a group dynamic decision. With RO, the bidding converges quickly while assigning segments to each operator owing to its random walk property. SA and GA can not only bring maximal revenue with optimal combination of segments and higher quality and usability by the improvements in each round, but also minimize the costs. However, the strategy based on GA tends to support the egoism of the dominating partner such as NCC and the fittest participant so that the other operators suffer from higher costs and lower quality, which is what the coalition game likes to avoid. This indicates that splitting and auction following SA are more efficient and effective than those following GA from the perspective of inclusive sharing economy.

The contributions of this study can be summarized as follows:

• Application of HyQVS organic ecosystem framework for 5G community for inclusive sharing economy.
• Application of Pareto Optimality as the strategy for the optimization with cooperative coevolution.
• Definition of metrics for quality and value of mobile spectrum based on ISO 25000 with respect to Pareto Frontier.
• Establishment of coalition game for spectrum splitting and auction and putting forward altruism to sustain the equilibrium of 5G IoT ecosystem to solve the deficit of conventional behavior economics that is a significant breakthrough of this study.
• Development of hybrid metaheuristics for the cooperation evolution with combination of SA, GA, and BO for the optimization of spectrum splitting and auction for multiple objectives of QoS, value, and costs.
• Evaluating the effects of different hybrid metaheuristics for the optimization of spectrum splitting and auction from the aspect of inclusive sharing economy.
• Suggestion for the ICT resource sharing in regions with different cultures and policies based on the analysis of the results of this study.

The ecosystem for the coalition game for 5G spectrum splitting and auction in this study involves planners and providers, viz. governmental authorities, and users, viz. the mobile operators. It is a kind of small-scale ecosystem that belongs to the topologies of multiple organic agents with or without unified rules as described in [6]. The sharing infrastructure of base station and equipment construction to save the cost of all carriers and even special advantages to new operators to maintain the equilibrium of the market, building more sophisticated models as well as renewing the analysis are also beneficial to the practices. For the distributed autonomous agents, we should apply PSO and FFO and test their effects of the ecosystems for SIoT and MIoT services in the future.