Neutral Host Technology: The Future of Mobile Network Operators

The neutral host network (NHN) is a new self-contained network envisioned by fifth generation (5G) of cellular networks, which offers wireless connection to its subscribers from a variety of service providers, including both conventional mobile network operators and non-conventional service providers. The NHN infrastructure, which is operated and maintained by a third neutral party, is rented or leased to network operators looking to scale up their network capacities and coverage in a cost-effective way. This paper highlights NHN as an emerging communication technology for private networks and discuss its opportunities and challenges in realizing multi-tenanted space such as factory, hospitals, stadiums, and universities. The paper also investigates the current state of the art in NHN and elaborates on the underlying enabling technologies for the NHN. Lastly, an efficient radio access network (RAN) slicing scheme based on the multi-arm bandit approach has been proposed to allocate radio resources to various slices, which maximizes resource utilization while guaranteeing the availability of resources to meet the capacity needs of each multi-tenanted operator. The simulation results show that the proposed Thompson’s sampling (TS)-based approach performs best in finding the optimal RAN slice for all the operators.

infrastructures such that traditional data transmission meth-23 ods should be replaced by intelligent systems with virtual-24 ized computation and storage capabilities. The reasons for 25 The associate editor coordinating the review of this manuscript and approving it for publication was Rongbo Zhu . this paradigm shift in telecommunication networks can be 26 uncovered through trend analysis of data traffic and cus-27 tomer behavior. According to Cisco [2], 80% of all mobile 28 data traffic is generated indoors from crowded areas such 29 as concerts, sports, hospitals, universities, stations, shopping 30 malls, enterprises, and is rising at a pace of 20% every 31 year. On the other hand, user behavior is shifting away from 32 traditional services including text messaging and voice calls 33 towards high computational and data-intensive activities such 34 as there-dimensional (3D) gaming, networking sites, 4K and 35 360-degree video streaming, augmented reality (AR) and 36 virtual reality (VR) experiences. Similarly, the introduction 37 of the industry 4.0 campaign has completely revolutionized 38 the industries and factories towards cyber-physical systems 39 The term, neutral host (NH), has been popular for repre-70 senting the multi-tenancy approach in 5G communication. 71 In the 5G research, the NH is a connectivity solution that 72 allows multiple tenants (for example, conventional mobile 73 network carriers or non-conventional service vendors) to 74 access and utilize the shared resources based on an agreement 75 with a third-party infrastructure owner, operator, and main-76 tainer. NH is commonly used to improve wireless coverage 77 and capacity in big events or other congested areas. The 78 main goal of NH is to achieve deployment flexibility and 79 enhance spectrum efficiency, while curtailing the deployment 80 costs of the systems. In addition, NHN has been also con-81 sidered as an additional model to the ''pure'' private net-82 works, often managed by a corporation for internal IoT or 83 telephony applications. For instance, a 5G-enabled factory 84 may be primarily focused on machine vision and robotics, 85 but it may also permit smartphone roaming by staff mem-86 bers and guests onto the coverage of the local network. The 87 rural/remote areas, metropolitan areas, workplaces, arenas, 88 and hotels, roadway and rail track coverage, industrial plants 89 and major transport hubs, short-term events and gatherings 90 (for example, festivals, huge construction projects), and some 91 types of residential and business venues are the primary appli-92 cations for neutral host network (NHN) deployment. In Fig. 1, 93 we have categorized NHN applications into four categories, 94 namely, indoor, rural, roads and rails, and dense urban areas. 95 The most significant benefit of an NH is that its primary 96 service is the infrastructure itself, which allows the fixed 97 expenses of installations over as many operators as possible. 98 Moreover, when a single network is constructed, wastage of 99 resources is prevented, and space/area is utilized properly. 100 There are, however, difficulties that must be overcome in 101 order to build a model that can serve these capabilities while 102 ensuring that mobile network operators (MNOs) can continue 103 to provide the same or higher service quality compared to they 104 would if they installed their own networks while retaining 105 network neutrality. Shared spectrum management, seamless 106 integration between the MNOs slice in the NH and their own 107 network, interoperability, security considerations, common 108 standardization, billing, etc are some challenges that need to 109 be solved for NH implementation.

110
The remainder of this paper is organized as follows. 111 In Section II, we start with an NH technology definition 112 and discuss its state of the art. In Sections III and IV, 113 we elaborate on various opportunities and challenges in 114 realizing the multi-tenanted space such as factories, hos-115 pitals, stadiums, and universities. Then, the enabling tech-116 nologies are presented in Section V, whereas the framework 117 that are already present in the 5G is briefly described in 118 Section VI. In Section VII, we propose an efficient radio 119 access network (RAN) slicing scheme based on the multi-arm 120 bandit approach, which allocates radio resources to various 121 slices to maximize resource utilization while guaranteeing the 122 availability of resources to meet the capacity needs of each 123 RAN slice. In the simulation, we consider an extremely 124 dynamic environment, in which Thompson     shared infrastructure, wastage of resources is avoided, and, 307 at the same time, space is used more optimally. NH deals with 308 complicated business models related to multi-operator sup-309 port and offers a more affordable option for network deploy-310 ments to enable both customers and businesses to achieve 311 better wireless performances at a great value as shown 312 in Fig. 3 [33].

314
The traditional ''outside-in'' coverage from neighboring cel-315 lular macro towers effectively serves many indoor areas. 316 Thus, many venues have unstable wireless coverage and 317 VOLUME 10, 2022 limited bandwidth, which leads to a highly unsatisfactory 318 indoor user experience. Modern construction materials, such 319 as low-emissivity glass, worsen this indoor coverage dif-320 ficulty by inhibiting outdoor mobile signals from infiltrat-321 ing inside buildings. Another impending concern is the 322 increased use of high-band frequency spectrum for 5G, such The local spectrum licenses for NHN enable private networks 346 for the enterprises with a major concern on their internal

369
NHN provides significant market potentials for service 370 providers, venue owners, businesses, and traditional opera-371 tors. There are, nonetheless, issues that need to be resolved, 372 so that design model can deliver services with guarantees to 373 the host client. NHN should be able to continue offering the 374 same or better service quality that host client would if they 375 deployed their own networks, while maintaining the network 376 neutrality towards host client. In this context, we discuss 377 some of the major challenges as follows. The NHN environment may be required to serve several 380 3GPP/ non-3GPP use cases, including IoT, private industrial 381 networks, and other access methods. Security at many lev-382 els from physical access to SCs to authentication and fraud 383 management is needed [34]. Flexible authentication methods 384 supporting different device types that support a wide vari-385 ety of standardized and proprietary authentication methods 386 are required. Moreover, subscriber confidentiality as well as 387 encryption are mandatory for NH solutions. Typically, an MNO offers individual SIM cards to its cus-391 tomers from which MNO manages SIMs/eSIMs profiles of 392 the customers. However, in the case of an NHN, the respon-393 sibility for SIM provisioning and profile management ques-394 tioning whether the MNO would need their own SIMs/eSIMs 395 profiles, or this would be the NH operator's responsibility. 396 Furthermore, a surge in network devices could put a strain on 397 the current capacity for numbering and mobile network codes 398 (MNC) addresses.
The network equipment of the operator is subject to vari-401 ous tests and certifications, including security, integration, 402 and performance to meet their operation standard. In an 403 NH scenario, a single standardized method is needed that 404 incorporates all of the requirements and standards of various 405 operators' types of equipment. Moreover, it can be compli-406 cated to bring together the various actors to establish the 407 ecosystem of consultants, mobile carriers, device manufac-408 turers, regulators, cloud service providers, and end users.  The NHN user needs to be supported by seamless and 411 continuous coverage going in and out of NH connectivity 412 sites. To allow for movement between NHN and public cell 413 coverage, inter-connectivity and roaming mechanisms would 414 need to be in place, so that the NHN could assist automated 415 network selection without user intervention. Moreover, the 416 NH may be required to facilitate international roaming traffic 417 in specific instances [31].  An NH should create the process to start and close new 451 slices for exiting and new operators in real-time. More-452 over, resource management is another big challenge for NH. 453 As host clients share a common infrastructure, it is important 454 to ensure that the operation of host clients is not affected by 455 the traffic of others. In other words, there should be isolation 456 between distinct slices and the ability for different slice own-457 ers to use their resources simultaneously without interfering 458 with one another [38]. Network slicing is a revolutionary network design that allows 466 physical resources to be shared by splitting them into multiple 467 logical networks and allocating them to various users in 468 mutual isolation. Each network slice can specify its logical 469 topology, service level agreements (SLAs), reliability, and 470 security level in several ways to fulfill the needs of differ-471 ent services, industries, or subscribers. Hence, each logical 472 network can be dedicated to a certain service or industry 473 user providing network services that are very flexible and 474 VOLUME 10, 2022 ting and virtualization of radio functions to integrate with slicing between Multi-radio access technologies (Multi-RAT) 529 will be critical for the NH. Hence, industry and research 530 initiatives are focusing on integrating LTE and 5G New 531 Radio (5G NR) technologies with various slicing methodolo-532 gies (e.g., allocating shares of available airtime to different 533 customers) from various wireless technologies (e.g., Li-Fi, 534 Wi-Fi 6). The NH framework goes beyond a novel RAN 535 controller supporting LTE and Wi-Fi by adding capabilities 536 to multiple RAN controllers from multiple vendors and radio 537 technologies.

539
SDN is an architecture that gives networks more programma-540 bility and flexibility by separating the control and data plane 541 from physical hardware such as routers and switches. It is 542 becoming more common in research to refer it as network 543 automation and programmable entity. SDN simplifies the 544 management of infrastructure through network configuration 545 and monitoring. It enables direct programmable network con-546 trol for applications and network services, which may be 547 exclusive to one organization or shared so that it can be used 548 by several other entities. There are three key parts of SDN, 549 which may or may not be physically close to one another, are 550 as follows: the first is applications that communicate network 551 information or requests for the allocation or availability of 552 particular resources. To ascertain the final destination of data 553 packets, the second SDN controllers, which has visibility and 554 authority over the whole controlled network, interact with 555 the applications. Within SDN, the load balancers are also the 556 controllers. Lastly, the networking component is the third part 557 that takes instructions on packet routed from the controllers. 558 In the case of an NH deployment, it is not enough for 559 the SP to maintain control over coexisting with multiple 560 tenant deployments on the data plane. Multiple control plane 561 instances must also be created to present an abstract view of 562 the resources that make up the logical network and provide 563 control mechanisms. This control must be carefully coordi-564 nated to ensure isolation and avoid resource overlap between 565 the various networks.

567
NFV is a network architectural idea utilizing virtualization 568 technology to virtualize functions of the entire classes of net-569 work nodes into linkable building blocks or chained together 570 to create and deliver communication services. This implies 571 that virtual machines utilize software to carry out the same 572 networking tasks as conventional hardware. Software, rather 573 than hardware, handles load balancing, routing, and firewall 574 security. An SDN controller enables network engineers to 575 program every different component of the virtual network, 576 and even automate the network provisioning. NFV enables 577 NH virtual network functions to run on a normal generic 578 server under the management of an SDN, which is signif-579 icantly cheaper compared to buying specialized hardware 580 devices. A virtualized network makes NHN network con-581 figuration and management considerably easier. Best of all, 582 because the network is run on virtual machines that are easily  and the service/application layer. The computing, network, 637 and radio resource components are under the infrastructure 638 layer. Here, NH uses NFVI for distributed computing and 639 radio management. The edge computing capabilities also lie 640 in this layer, which provides real-time access to radio network 641 information. The second layer, the orchestration and control 642 layer serves as the logical core of the whole system. It is 643 further subdivided into several functional blocks for man-644 agement, control, and orchestration. To make communica-645 tion between infrastructure owners and tenants easier and 646 to enforce the necessary security and billing, a dashboard 647 with a graphic user interface (GUI) and an AAA component 648 are implemented. For security purposes, network slices from 649 various tenants are securely isolated, and maximum data and 650 information segregation is maintained. A crucial function of 651 the platform is played by the slice manager, who provides the 652 necessary logic for the creation and administration of slices 653 on-the-fly. There is also a resource placement component 654 that is used to determine the best distribution of VNFs to 655 be deployed over a particular slice. By combining the NFVO 656 with MEC components that manage mobile edge (ME) appli-657 cations, the orchestration capabilities of the platform are 658 increased to facilitate NFV/MEC integration following the 659 ETSI MEC specification. The multi-tier orchestrator compo-660 nent gives an abstracted view in front of multiple underlying 661 orchestrators. SDN-based RAN controllers manage the radio 662 components and enforce RAN slicing and RAN function 663 virtualization for 5G NR, LTE, and Wi-Fi. The infrastructure 664 abstraction entity helps to support multi-RAT controllers and 665 technologies. Lastly, a service or an application layer consti-666 tutes a set of tools aimed at facilitating service design and 667 composition to the service providers, tenants, and any other 668 associated third-party groups. Here, the software develop-669 ment kit (SDK) for network function developers and service 670 providers allows combining several functions for new service 671 deployments. This part is also in charge of adding new fea-672 tures and services to the NFVO. A 5G Service & Apps index 673 is offered to hold previously produced and published network 674 services [15]. 676 We have learned from the above discussions that network 677 slicing is one of the primary functions that will enable an 678 NHN with the required flexibility through the creation of 679 several logical networks, known as network slices, on top of 680 a single shared physical infrastructure. Using certain control 681 plane (CP) and/or user plane (UP) functionalities, each net-682 work slice can be utilized to provide services for a specified 683 operator or service category. In this section, we discuss RAN 684 slicing schemes that enable radio resources from across the 685 whole spectrum to be assigned to the host clients from various 686 slices [46], [47]. In multi-tenant NH settings, three important 687 requirements should be considered while allocating the slice 688 resources to each hosting client such as isolation, customiza-689 tion, and resource utilization.   slice category, assessing slice elasticity between tenant 721 operators, and determining the best slice option for the 722 subscriber of tenant operators. In this situation, we require a 723 real-time dynamic network slicing algorithm that automates 724 the optimal RAN slice for frequent and randomized environ-725 ments to maximize spectrum efficiency and slice isolation 726 over time. The algorithm must monitor the real-time effect of 727 the slice decisions and incorporate this additional information 728 into the future RAN slicing decisions (i.e., an ''explore-and-729 exploit'' strategy). We model our slicing problem as a bandit 730 problem, a well-known dynamic optimization technique in 731 reinforcement learning. In this approach, the NH agent is 732 given a variety of slice options, often known as the arm. 733 In most cases, each arm has an unknown reward distribution 734 to the agent, who must select arms to maximize cumula-735 tive reward over time. Hence, we model this problem as a 736 multi-armed bandit problem, as shown in Fig. 6  p(n) = betarand(Success(n) + 1, Fails(n) − Success(n) + 1); 8: end for 9: Choose action and collect reward: 10: n * = max p(n); 11: rew = reward(n * ); 12: Update success probability: 13: Success(n * ) = Success(n * ) + rew;

14:
Fails(n * ) = Fails(n * ) + 1; 15: end for probability from each n's posterior distribution. Then, using 788 n * = max p(n), we choose the variant with the highest success 789 probability. Then, in step 11, we observe and collect the 790 reward for the chosen action before updating the distribution 791 parameters Success(n * ) = Success(n * )+rew and Fails(n * ) = 792 Fails(n * ) + 1 in steps 13 and 14. 793 We develop an NH system-level simulation platform using 794 MATLAB. The RAN slicing decision observations are mod-795 eled and generated using a set of random multi-objective 796 bandit cases, in which multi-dimensional reward distri-797 butions (i.e., throughput and QoS of each slice) follow 798 finite trials with different mean distribution values, Success, 799 of each reward. It should be noted that Success is inde-800 pendent for each trial and follows the uniform distribution 801 between 0 and 1. We create M = 5, 000 samples of RAN 802 slices for various sizes (4 and 8) of the tenant operators. 803 To analyze the performance, we compare popular multi-arm 804 bandit approaches, Upper Confidence Bound (UCB), and 805 eGreedy approaches. In the simulations, we observe how per-806 formance changes with time (i.e., time steps) for N ∈ {4, 8}, 807 which corresponds to the number of arms (i.e., operators), 808 • eGreedy Approach: In bandit problems, the eGreedy 809 approach is common and straightforward, with the agent 810 selecting the most promising arm based on past experi-811 ences with probability (1 − e) and a random arm with 812 probability e. When e = 0, this algorithm becomes 813 greedy; when e = 1, this algorithm becomes strictly 814 exploratory. It is possible to maximize your cumulative 815 reward by selecting the right e.

816
• UCB Approach: The arms are measured based on con-817 fidence by the UCB algorithm. In other words, the algo-818 rithm assigns a UCB value to each arm, which is the sum 819 of the expected reward and an exploration bonus. The 820 potential value from experimentation is the exploration 821 VOLUME 10, 2022 continues to explore the random slice option half of its given 839 time. A similar conclusion can also be drawn for the average 840 achieved QoS by the tenant operators, as shown in Fig 8. 841 We can observe that all four operators have high QoS sat-842 isfaction for the TS approach (98%) and lower for UCB 843 (87%) and lowest for eGreedy (78%). It is noted that the 844 QoS unfairness across the slices is more severe in UCB and 845 eGreedy, because they are highly subject to the high spectrum 846 efficiency configuration rather than the fairness of each slice 847 category.

849
An NH access mode is one of the new network architectures 850 envisioned by 5G and beyond, where a third neutral party 851 builds and operates part of the network offering private and 852 public connectivity. It is a cost-effective alternative to provide 853 better mobile performance to both, individual customers and 854 enterprises, in remote and hard-to-reach areas or occasional 855 high-density areas as NH serves such areas with only one 856 infrastructure for all operators. We have discussed NHN 857 as a next-generation communication technology for smart 858 private network communication and presented its opportu-859 nities and challenges in realizing the multi-tenanted space 860 and its enabling technologies. In support, we have proposed 861 an efficient RAN slicing scheme based on the multi-arm 862 bandit approach, which allocates radio resources to various 863 slices to maximize resource utilization, while guaranteeing 864 the availability of resources to meet the capacity requirement 865 of each RAN slice. Through the simulation, we have shown 866 that the proposed TS multi-arm bandit algorithm, which max-867 imizes the cumulative rewards over time, can provide higher 868 spectral efficiency and average QoS, compared to the UCB 869 and eGreedy approaches.