Blockchain-Empowered Service Management for the Decentralized Metaverse of Things

The future of networking will be driven by the current emerging trends of combining the physical and virtual realities in cyberspace. Considering the ambient pandemic challenges, the role of virtual and augmented reality will definitely grow over time by transforming into the paradigm of the Metaverse of Things, where each person, thing or other entity will simultaneously exist within multiple synchronized realities. In this paper, we propose a novel framework for future metaverse applications composed of multiple synchronized data flows from multiple operators through multiple wearable devices and with different quality requirements. A new service quality model is proposed based on a customizable utility function for each individual data flow. The proposed approach is based on dynamic fine-grained data flow allocation and service selection using non-fungible tokens, which can be traded over the blockchain among users and operators in a decentralized mobile network environment.

Thus, the IoT creates a highly automated environment driven 22 by the application of various AI algorithms with collaborative 23 and smart-interconnected sensors and actuators. 24 One of the most appealing trends currently is the develop-25 ment of a metaverse. A metaverse is the evolution of AR/VR 26 technologies toward interconnected virtual worlds [1].  verses are developed based on the most advanced means of 28 visualization, sensing and wireless communications. In addi-29 tion, metaverse leverages the latest achievements of AI and 30 blockchain technologies to achieve a truly immersive user 31 experience with synchronized realities [2], [3]. Currently, 32 metaverses are limited mostly to virtual worlds, where users 33 can be engaged only through the VR headset and purchase 34 the underlying blockchain infrastructure. 84 To the best of our knowledge, there are currently no 85 research works that address the problem of decentralized 86 quality of experience (QoE) management for applications 87 composed of multiple data flows, which are transmitted by 88 different MNOs and to various UEs. In addition, there are no 89 works that consider the decentralized resource allocation in 90 mobile networks by using NFTs as the tradable representation 91 of network resources in the blockchain. 92 Thus, the main contributions of this article are as follows: 93 1) We propose a novel multi-flow synchronized service 94 provisioning based on NFTs and blockchain for meta-95 verse and IoT applications. 96 2) We propose a new service management model for 97 decentralized multi-flow applications that allows us to 98 derive the unique QoS metric for each individual data 99 flow.

100
3) We propose a new method of dynamic multi-flow ser-101 vice selection for the decentralized multioperator net-102 work environment. 103 The remainder of this paper is organized as follows. 104 Section II provides an overview of the key enablers of the 105 MoT. Section III describes the system model and the pro-106 posed approach in detail. Section IV covers the simulation 107 results and the discussion. Section V concludes the paper. 109 In this section, we review the key technological trends, which 110 are essential blocks for the MoT concept, such as IoT, digital 111 twins, blockchain, AI, 6G, AR/VR, etc. To meet the stringent criteria posed on the reliability and 115 latency of IoT applications in a fast-evolving environ-116 ment, 5G networks represent the key technological aspect 117 to be addressed jointly with various application use cases, 118 including enhanced mobile broadband (eMBB), massive 119 machine type communication (mMTC), ultra-reliable com-120 munication and low latency communication (URLLC). These 121 5G-oriented services will greatly enhance the efficiency 122 and flexibility of intelligent systems by replacing the tradi-123 tional cable setups and simplifying the digital transformation 124 process [11]. 125 The preliminary results delivered in [11] investigate the 126 impact of different radio configurations at the physical and 127 media-access control layers, which provides an important 128 conclusion that the dedicated reserved bandwidth for the 129 eMBB and URLLC use cases in industrial applications can 130 strongly impact the effectiveness of the services in terms of 131 the throughput and latency. Next, the authors in [12] proposed 132 simultaneous support of eMBB and URRLC services via an 133 explicit prioritization method, which allows perfect isola-134 tion and stable performance characteristics for URLLC-based 135 industrial applications even in a dynamically changing envi-136 ronment, while eMBB traffic with slightly weaker require-137 ments would experience dynamic transitions in terms of 138 application performance. 139 Finally, some concluding remarks regarding the eMBB 140 and URLLC performance in the isolated scenarios of 5G 141 IIoT applications were comprehensively described in [13]. 142 Applications of IIoT allowed real-time monitoring by 143  [6], [7]. The tactile internet can 187 support most MoT applications by providing an ultra-reliable 188 network with extremely low service latency [8]. Low latency 189 is vital for applications, such as self-driving vehicles, 190 AI-assisted smart medical devices, and manufacturing robots, 191 where milliseconds can literally prevent disasters. These 192 stringent requirements can be satisfied by deploying local 193 5G/6G networks with edge computing capabilities located 194 therein [9]. To unleash the full potential of the Internet of 195 Skills concept in the MoT environment, we need to extend 196 the traditional concept of the tactile internet with haptic 197 communications providing immersive user engagement by 198 enriching the sensory experience in many critical aspects of 199 industries [10]. Thus, the underlying idea of the IoS is natu-200 rally aligned with our proposed MoT concept.

201
It is clear that synchronization between physical and virtual 202 worlds will heavily rely on the use of AI-based data process-203 ing algorithms to render the corresponding views and gener-204 ate control commands in both directions. Therefore, security 205 and robustness to external attack will be one of the important 206 challenges for the MoT in the future [25].

207
A typical use case where the IoS can be applied with the 208 MoT is a remote driving scenario, where an unmanned car 209 is controlled over multiple parallel data flows, and including 210 high-quality video streaming from and on-board cameras to 211 VR headset, ultra-low latency data flows for the various sen-212 sory telemetry and ultra-low latency data flows for remote 213 control by a steering wheel and pedals. Similar applications 214 can be found in other areas, such as remote surgery, machin-215 ery maintenance, and entertainment.  All nodes (or more than 51%) in the blockchain must update 265 the new state of the distributed ledger, which ensures the 266 security and immutability of each transaction [49].

PoW is the first and most famous consensus algorithm. 281
It is implemented in the Bitcoin blockchain. The main idea 282 of PoW is the complex cryptographic puzzle (hash), which 283 takes considerable time and energy to solve. This procedure is 284 called mining. The nodes with the most computing power are 285 more likely to win the contest for mining the next block [55]. 286 In addition, each block contains a hash function of the pre-287 vious block, which makes all blocks linked in the chain [56]. 288 These features prevent any attack on the system if the number 289 of blockchain nodes is large enough because no one is able to 290 accommodate enough computing power within a short time 291 frame to replace the part of the chain with faked transac-292 tions [57]. Apart from Bitcoin, the largest PoW blockchains 293 are Ethereum 1.0, Litecoin, Monero, etc. Although PoW is 294 the most decentralized and secure consensus algorithm, it has 295 corresponding drawbacks, such as low throughput and high 296 energy consumption. Therefore, it is not suitable for the pro-297 posed multi-flow service provisioning in a decentralized MoT 298 system.

299
2. PoS has emerged as an energy-efficient consensus 300 algorithm that replaced the mining process (i.e., complex 301 cryptographic computing) by the validation process. Thus, 302 instead of competing by computational performance, valida-303 tors (nodes) compete with each other by their financial stakes 304 locked in the system. This type of approach does not require 305 complex computations because the contest will be won by the 306 ''richest'' validator. At first glance, it may be seen as a risky 307 option because extremely wealthy users can afford to accom-308 modate 51% of the total stake in the network to centralize the 309 entire blockchain and alter transactions for their own benefit. 310 However, practically, this type of attack would be a huge risk 311 of losing all the staked funds. In addition, PoS uses a random 312 selection of validators among all eligible nodes who stake a 313 minimal amount of money. This adds an additional security 314 mechanism, which requires potential attackers to not only 315 stake 51% of the total stake but also distribute it among more 316 than half of the nodes eligible for validation. Therefore, such 317 an attack is possible only for very small blockchains and is 318 very unlikely for large and widely used PoS blockchains, such 319 as Ethereum 2.0, Polkadot, and Cosmos [49]. Moreover, PoS 320 provides much higher transaction throughput, much lower 321 latency and unlimited scalability through a sharding mech-322 anism, which make it a good candidate for MoT applications. 323 3. Delegated Proof-of-Stake has emerged as a faster ver-324 sion of the PoS, where the number of validators is decreased 325 to the small group of delegates, which are elected by the 326 majority of nodes [58]. This approach increases the transac-327 tion throughput and decreases the latency of the blockchain 328 by reducing the minimal number of required validators to 329 very few nodes. This, however, causes a less decentralized 330 system, which makes the whole network less secure. The 331 most widely used DPoS blockchains are EOS, TRON and 332 Cardano. Digital Currency) [65]. 359 The numerical comparison of above-mentioned consen-360 sus algorithms in terms of throughput, latency and security 361 aspects is provided in Table 2. presently, NFTs are already adopted mostly by the gaming 380 and entertainment industry, where they are used to replicate 381 different virtual assets in games and metaverses [46], [47]. Despite of numerous related research works, currently there 391 is not any complete solution for MoT. In Table 1, we sum-392 marize the existing research background in the context of 393 MoT, and it's particular components such as IoT/IoS, 5G/6G, 394 metaverse and blockchain. It is clearly seen that combination 395 of the above-mentioned components enables the future devel-396 opment of MoT, as emphasized in the Table 1. 397 Therefore, a general framework is needed to combine all 398 existing technologies under a common umbrella of decentral-399 ized multi-flow service management to provide a new secure 400 and trustful networking ecosystem for the envisioned concept 401 of the MoT.

402
The difference between IoT (IIoT) and the MoT is that the 403 former is synchronizing any physical thing with its digital 404 twin, while the latter is actually synchronizing realities in the 405 physical and virtual world. The definition of ''thing'' within 406 the new concept is more flexible because it can now refer 407 to physically connected instances in the real world and to 408 absolutely unique virtual things that are not present in the 409 real world. From the industrial perspective, it also allows us 410 to synchronize a physical infrastructure of robots, production 411 lines, sensors and actuators with the corresponding virtual 412 copy of the same environment by using a unique NFT for each 413 entity.

414
Unlike the IoT, which can rely on any and even bad internet 415 connectivity in some cases, the MoT will rely on the precise 416 synchronization of multiple independent data flows to pro-417 vide a comprehensive replication of real world context and 418 surroundings in virtual reality. Thus, data from accelerom-419 eters, hyroscopes and haptic sensors need to be transmit-420 ted simultaneously with the data from smartphones, AR/VR 421 headsets and cameras to transfer and visualize the relative 422 movement of people, machines or vehicles in the virtual 3D 423 space. We call this process a ''multi-flow service provision,'' 424 which is described later in the paper.

428
In this section, we describe the proposed blockchain-based 429 framework for the envisioned concept of the MoT, which aims 430 to synchronize physical and virtual realities via the under-431 lying network infrastructure by using the novel analytical 432 definition of service quality evaluation and management for 433 multi-flow decentralized applications. The conventional concept of network slicing, defined in 436 3GPP TS 29.531 [18], assumes that a separate virtual net-437 work infrastructure with dedicated QoS requirements is pro-438 vided for each type of service, such as mMTC, URLLC and 439 eMBB. Here, we extend this vision to metaverse slicing, 440 which assumes a synchronized transmission of data flows 441 VOLUME 10, 2022 Thus, the total data rate at the discrete time interval t can 480 be described as a function of (3) as follows: where τ is the time frame between two discrete intervals t and 483 t + 1. By representing (2) with values of (4), we can obtain 484 a vector of target total data rates for the dedicated scheduling 485 period as follows: Thus, the problem of scheduling in the proposed system 488 model can be described by matching the vector in (5) with the 489 vector of available bandwidth resources for the equivalent set 490 of discrete time intervals t ∈ (0, T ) as follows: In general, to match data rate (5) and bandwidth (6), we need 493 to derive the values of spectral efficiency, which depend on 494 the channel characteristics between the base station and UE. 495 However, considering the frequent fluctuations of the channel 496 conditions due to the expected mobility of UEs and the decen-497 tralized service provision by multiple operators via different 498 communication networks, we can redefine (6) as a vector of 499 available data capacity per time interval as follows: If the available data capacity in a vector (7) is C i > τ R i , 502 scheduling can be performed without any need for adjust-503 ment. However, in the opposite situation when C i < τ R i , 504 we need to apply advanced scheduling based on different QoE 505 metrics.

506
Hence, it is important to define a model for the service 507 quality evaluation for each data flow, considering the indi-508 vidual features of each application.

509
The typical scheduling procedure in 3GPP-based mobile 510 networks is performed in a dynamic manner according to 511 the instantaneous channel conditions, user priority and ser-512 vice type. Nevertheless, within the proposed framework, 513 As we discussed in section II, there are several blockchains 546 that can be used for MoT, and each blockchain has its own 547 pros and cons. Most of them are actively developing so that 548 the current ''status quo'' in terms of performance can be 549 easily changed anytime. Nevertheless, the essential require-550 ments for MoTs are the support of smart contracts and NFTs 551 with reasonable throughput and latency. Therefore, among 552 the blockchains in Table 2, we can outline Ethereum 2.0, 553 Polkadot and Cosmos as more decentralized and expensive, 554 and BSC, Hyperledger and XRP Ledger as cheaper but more 555 centralized MoT solutions.

556
As mentioned above, the key feature of the proposed 557 framework is in the decentralized spatial representation of 558 the data flows and their association with UEs, services and 559 MNOs. In particular, a decentralized model allows UE to 560 choose multiple MNOs for different services simultaneously 561 and change them in quasi-real-time through blockchain smart 562 contracts. Thus, an overall data flow in this scenario can be 563 represented as follows: Note that matrix F represents the same total data flow in both

570
(2) and (8), and the following condition should be satisfied as 571 follows: follows: where q is an integral QoE value of the data flow, p is the 619 service price for the data flow normalized by time and deter-620 mined by the algorithm presented in [67], q min is the lowest 621 QoE value acceptable by the user, and p max is the highest 622 service price that the user is willing to pay. An additional 623 parameter µ ∈ (0, 1) is used to determine the preference of 624 the end user in terms of the trade-off between the service price 625 and the service quality.

626
The utility function (11) is unique for each individual UE i 627 and service type j, and thus, can be presented in a similar way 628 to (1) as follows: An integral quality parameter q can be represented as a 631 weighted sum of all other parameters as follows: where R, T , L are the throughput, latency and packet loss, 635 respectively, as defined by the 5G quality identifier in 3GPP 636 TS 23.501 [68]. By amending the corresponding values of 637 β (R) , β (T ) , β (L) , in equation (13)

671
where i is an index of UE. Thus, UEs evaluate their integral 672 utility functions as follows: Then, each UE selects the smart contract (14) with the highest 675 utility function (15) among all available bids for each type of 676 service j at each discrete time interval t: The high-level protocol diagram of the dynamic service 688 selection between MNOs and end users is displayed in Fig. 3. 689

690
This section illustrates the performance characteristics of the 691 proposed multi-flow service model for the case study of MoT 692 networks consisting of a large number of UEs with different 693 service requirements. The main setting of the radio resource 694 allocation of the physical frame using carrier aggregation is as 695 follows. We consider channel bandwidth W = 5MHz consist-696 ing of 25 resource blocks allocated in two sub-GHz frequency 697 carriers (3.6 GHz and 3.62 GHz). The total number of RBs is 698 thus 50 RBs for each dedicated time slot (1 ms). The UEs 699 follow the Manhattan mobility model with a velocity of 1m/s 700 in the grid road topology. The dedicated KPIs are collected 701 based on the Monte Carlo simulation approach, averaging 702 N = 1000 of the simulation trials.

703
In the current simulations, we omit the numerical evalua-704 tion of the blockchain performance for the following reasons. 705 First, there is no technical possibility to deploy and test a 706 blockchain, with the meaningful number of nodes. Conse-707 quently, any implementation will be limited to either few 708 nodes or to fully virtual blockchain simulation on a single 709 machine. As result, all numerical values of the throughput and 710 latency in a simulated blockchain will not be representative. 711 Nevertheless, as mentioned in section III.B, there are already 712 existing blockchains, which can satisfy the requirements of 713 MoT, which makes the whole our concept feasible. There-714 fore, in the current paper, we provide the pure simulation 715 results to assess the trade-off between possible scenarios of 716    Comparing the results for throughput (Fig. 4) and latency 724 (Fig. 5), we observe that they follow theoretical expectations 725 FIGURE 7. Operator's profit for boundary cases (i.e., throughput-and latency-sensitive UEs) and the use case with equally balanced data flow. and behave exactly as needed in terms of balancing both met-726 rics. For the latency-sensitive data flows, we do not observe 727 high throughput because each data block is transmitted imme-728 diately in available resource blocks to reduce latency, while 729 for the throughput-sensitive and latency-tolerant data flows, 730 we tend to wait until a larger number of resource blocks will 731 be available to ensure that large data blocks will have enough 732 resources for transmission. Furthermore, balanced data flows 733 are in the middle in both Fig. 4 and Fig. 5 because they 734 must follow both requirements simultaneously. Nevertheless, 735 we observe a clear relation between the number of data flows 736 and all performance indicators. The difference in throughput 737 between the 3 types of data flows is much higher when the 738 number of data flows is low and diminishes with increasing 739 data flow number (Fig. 4). This is clear because the total 740 throughput is limited and divided between all data flows. 741 Therefore, with a lower range of throughputs, we observe a 742 lower difference between service types. In Fig. 5, we observe 743 the opposite situation for latency because latency is increased 744 for a higher number of data flows, and thus, we observe 745 even more differences between the different service 746 types.

747
Apart from the pure technical parameters, we also consider 748 the economic aspects of the network because any decentral-749 ized system can be sustainable only in the case of a profitable 750 business model. In Fig. 6, the normalized throughput per price 751 unit is compared for different metrics, and we can observe the 752 same relation between the service types as in Fig. 4, which is 753 expected. However, when the number of data flows is larger, 754 all service types converge to the same value of normalized 755 throughput per price unit, which means that during a high 756 load, throughput will depend more on the service price rather 757 than on the type of service.

758
From the perspective of the MNOs in Fig. 7, we observe 759 that their profit from the latency-sensitive data flows is con-760 sistently higher than that from the throughput-sensitive data 761 flows. Balanced data flows are again averaged between bor-762 derline types. Naturally, the profit of operators rises with 763 the growing number of users regardless of the service type, 764 but for a very high number of data flows, it becomes more 765 constant. 766 99034 VOLUME 10, 2022 Since simulation results are provided for borderline cases, 767 we can expect that the results for any other combina-768 tion of β (R) , β (T ) , β (L) will be within a range between 769 throughput-sensitive and latency-sensitive data flows in terms 770 of the metrics depicted in Fig. 4 -Fig. 7.

772
In this paper, we have proposed a novel approach for man- Ofcom, and acts as Policy Advisor on issues related to digital, skills and 1123 education. He has had ample coverage by National and International press 1124 and media. He is featured on Amazon Prime. He is a frequent keynote, 1125 panel and tutorial speaker, and has received numerous awards. He has pio-1126 neered several research fields, contributed to numerous wireless broadband, 1127 IoT/M2M and cyber security standards, holds a dozen patents, organized 1128 and chaired numerous conferences, was the Editor-in-Chief of two journals, 1129 has more than 300 highly-cited publications, and authored several books. 1130 He is a Top-1% Cited Scientist across all science fields globally. He is the 1131 Co-founder of the Smart Cities pioneering company World sensing, where 1132 he was the CTO from 2008 to 2014.