Chained Orchestrator Algorithm for RAN-Slicing Resource Management: A Contribution to Ultra-Reliable 6G Communications

Received 29 September 2022, accepted 19 October 2022, date of publication 28 October 2022, date of current version 4 November 2022. Digital Object Identifier 10.1109/ACCESS.2022.3218061 Chained Orchestrator Algorithm for RAN-Slicing Resource Management: A Contribution to Ultra-Reliable 6G Communications JOSE J. RICO-PALOMO 1, JESUS GALEANO-BRAJONES 1, DAVID CORTES-POLO 2, JUAN F. VALENZUELA-VALDES 3, AND JAVIER CARMONA-MURILLO 1 1Department of Computing and Telematics System Engineering, Universidad de Extremadura, 06006 Badajoz, Spain 2Department of Signal Theory and Communications and Telematics Systems and Computing, Rey Juan Carlos University, Móstoles, 28933 Madrid, Spain 3Department of Signal Theory, Telematics and Communications, CITIC, Universidad de Granada, 18014 Granada, Spain Corresponding author: Jose J. Rico-Palomo ([email protected]) This work was supported in part by the Spanish National Program of Research, Development, Innovation, under Grant RTI2018-102002-A-I00; and in part by the Junta de Extremadura under Project IB18003 and Grant GR21097. ABSTRACT The exponentially growing trend of Internet-connected devices and the development of new applications have led to an increase in demands and data rates flowing over cellular networks. If this continues to have the same tendency, the classification of 5G services must evolve to encompass emerging communications. The advent of the 6G Communications concept takes this into account and raises a new classification of services. In addition, an increase in network specifications was established. To meet these new requirements, enabling technologies are used to augment and manage Radio Access Network (RAN) resources. One of the most important mechanisms is the logical segmentation of the RAN, i.e. RAN-Slicing. In this study, we explored the problem of resource allocation in a RAN-Slicing environment for 6G ecosystems in depth, with a focus on network reliability. We also propose a chained orchestrator algorithm for dynamic resource management that includes estimation techniques, inter-slice resource sharing and intra-slice resource assignment. These mechanisms are applied to new types of services in the future generation of cellular networks to improve the network latency, capacity and reliability. The numerical results show a reduction in blocked connections of 38.46% for eURLLC type services, 21.87% for feMBB services, 12.5% for umMTC, 11.86% for ELDP and 11.76% for LDHMC. INDEX TERMS 6G, RAN-slicing, reliability, capacity, latency, resource management, channel estimation. I. INTRODUCTION The long-awaited all-connected society is rapidly becoming part of our lifestyle. The world is flooded with mobile phones, tablets, laptops, wearables, industrial systems, smart cities, and other devices connected to the Internet, and growth prospects do not seem stagnant. In recent years, the number of smartphones has increased by 93 million, and the number of connected devices now exceeds 5.22 billion worldwide [1]. This increment in the number of devices is reflected in the data traffic, which has already recorded a volume of more The associate editor coordinating the review of this manuscript and approving it for publication was Meng-Lin Ku . than 55 Exabytes per month by 2021. To meet these demands, 5G technology was introduced. This generation of cellular networks standardises different types of services (eMBB, URLLC and mMTC) to classify applications according to demands, connection requirements and traffic, among others. However, the development of new applications, such as autonomous vehicles and tele-medicine, requires new latency specifications below 1 ms [2], [3]. High-capacitydemanding applications, such as Virtual and Augmented Reality (VR/AR), add complexity to an already problematic scenario, along with the number of connected, lowpower and synchronised devices used by Industry 4.0. These applications require a much larger volume of data, 113662 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ VOLUME 10, 2022 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management in addition to increasing the number of devices connected to the network [4]. These data reflect the huge volume of traffic generated by this type of communication and suggest that conventional technologies may struggle to meet the demands. High mobility, low power consumption of devices and high density are challenges that cellular technology must address. Therefore, the future generation of cellular networks (6G) was devised to address this scenario. 6G Communications offers higher specifications than its previous generation, and a more varied and specialised division of service types to provide dedicated services to new and emerging applications. In addition, given its versatility, it can implement mechanisms to manage its resources. Dynamic resource reconfiguration through RAN-slicing is one of its key point. Because of the greater specificity of services that 6G Communications can offer, access networks can be provisioned using slices to meet the demands of emerging applications, having the ability to particularise user demands more than in 5G. Specifically, critical communications, such as autonomous vehicles, minimally invasive tele-medicine applications, and industrial Internet, require near-zero latencies and maximum reliability. All of the above demands higher requirements than those offered by previous generations of cellular networks. The particularisation and deployment of slices focused on these types of services make a more efficient use of the resources available on the network, in addition to having the capacity to be deployed in real time. Also, intelligent management mechanisms help 6G networks deploy priority-based RAN-slicing mechanisms that increase the reliability of communication. The main contribution of this work is the development of an orchestration algorithm for 6G RAN-Slicing, with the objective of ensuring ultra-reliable cellular communication to reduce the number of blocked connections by increasing the average network capacity and latency. This solution is based on the concatenation of resource estimation techniques, dynamic resource management in inter-slice environments, and reallocation of resources between different slices. The performance of the proposed solution was tested by simulations and compared with the standardised baseline link planning and resource allocation for 5G-NR without RANSlicing. The numerical results showed an improvement in network reliability depending on the type of service: 38.46% for eURLLC, 21.87% for feMBB, 12.5% for umMTC, 11.86% for ELDP, and 11.76% for LDHMC. This paper is organised as follows: an introduction to 6G and RAN-Slicing technology is presented in Section II. Section III presents a taxonomy of related works in the research field of this study. Section IV details the modelling used in the simulations. Section Vdescribes the proposed RAN-Slicing algorithm, and Section VI details the experimentation performed to assess the performance of the proposed approach. The last part of the work, in Section VII, comprises the main conclusions of the research and suggests possible approaches to further investigate the matter. II. RAN-SLICING AS A KEY ENABLER TECHNOLOGY FOR 6G COMMUNICATIONS New applications developed recently and new requirements make necessary a new classification of service types [5]. 6G Communications establishes a new paradigm that aims to provide full wireless coverage to meet the objective of connectivity anywhere and anytime. It will be able to serve a large number of users with extremely high data rates and exceptionally low latencies, joining satellite, terrestrial, aerial and quantum communications, among others. The 6G ecosystem would also continue the trends of previous generations, which included new services with the addition of new technologies. The services proposed by 5G will evolve to address the latest applications and traffic characteristics. These new services are as follows [6]: •Further enhance Mobile Broadband (FeMBB): Applications which require a high bandwidth and a lot of capacity to meet the demands. This includes technologies such as Holographic Verticals, Full-Sensory Digital Reality (VR/AR), Tactile/Haptic Internet and UHD/EHD Videos. •Long Distance and High Mobility Communications (LDHMC): Users who move at high speed while they often are far away from the network access point. Examples include hyper-high-speed railway (HSR), space travel applications and deep-sea sightseeing, among others. •Extremely Ultra Reliable and Low Latency Communications (eURLLC): This service handles the communications that are critical and has a high priority, where a near-zero error rate must be ensured. Some of these technologies include Fully Automated Driving and Industrial Internet. •Extremely Low-Power Communications (ELPC): Applications that must have a minimum power consumption but the connection to the grid must be guaranteed. E-Health technologies and nano devices, robots and sensors are examples of such applications. •Ultra-Massive Machine-Type Communications (umMTC): Ensuring sufficient capacity for the establishment of thousands of connections is one of the most important features. Internet-of-Everything (IoE) and smartcities are two technologies that fall under this type of service. Early preliminary studies established higher requirements than 5G. To meet these requirements, several enabling technologies are being considered for inclusion in the 6G ecosystem. These include THz-communications, very-large-scale antenna arrays, laser and visible light communications, spatial satellite links, and core/RAN slicing. Artificial Intelligence also plays an important role in this type of networks, as well as cloud/edge/fog computing, blockchain, Software Defined Networking (SDN) and Network Function Virtualisation (NFV). One of the most promising enabling technologies in 6G is RAN-Slicing [7]. This key enabler technology involves VOLUME 10, 2022 113663 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management FIGURE 1. Schematic diagram of RAN-Slicing segmentation for each type of service. the segmentation of the network infrastructure into logically self-contained networks. Each slice, which is designed and deployed for a specific type of service, consists of functions and resources abstracted from underlying communication and network resources. The concept was conceived for the 5G core network; however, to meet user experience expectations, it was necessary to upgrade to the RAN. Figure 1shows a schematic segmentation of the RAN into slices depending on the service type. In a conventional RAN, the transmission employs a best-effort strategy without resource reservation, which cannot guarantee Quality of Service (QoS). However, RANSlicing implements resource management mechanisms to meet the demands of the users. These mechanisms manage the resources of the different slices to ensure an increase in the Key Performance Indicators (KPIs), to meet the requirements of the next generation of cellular networks. These are developed to supply specific needs and are deployed when necessary. According to the literature, RAN-Slicing techniques can be developed using artificial intelligence, optimisation problems, dynamic resource management, among others, and a combination of these [8], [9], [10], [11]. These resources can be distributed among those available for a slice or ceded by another slice. Therefore, the complete state of the RAN and its slices must be known. For this purpose, orchestrators are used, which are devices that are aware of the state of the RAN, the BSs that compose it and its resources. This paradigm endows the network with great flexibility and versatility for resource reconfiguration, which is necessary for a new scenario that cellular networks must face. Owing to the variety of services that will be differentiated in 6G, the mechanisms to be deployed on the RAN in operation must be able to reconfigure these resources in real time. They must also be able to differentiate the criticality of each user and reallocate resources proportionally to the priority of their traffic. III. RELATED WORKS RAN-slicing technology has been extensively studied in 5G, and some challenges are posed in the literature that must be fulfilled. Many of these studies require knowledge of communication channel characteristics to avoid overload and congestion [12]. In addition, a new line is opened focusing on resource reconfiguration algorithms in the slices, considering the QoS; channel variations can affect the QoS of the most critical services [13]. In [14], the need for resource estimation prior to establishing the connection was highlighted. Other challenges were raised in [15], which stated that there is a necessity to develop mechanisms for resource allocation and sharing in a slice-based RAN. Furthermore, in [16], a target for dynamic resource allocation was discussed, showing the need for developing algorithms to reallocate resources between different slices. Evolving to 6G Communications, a major challenge is to achieve dynamic network orchestration and slice resource management according to real-time network information and service requirements [6]. Some authors also agree on the goal of managing and sharing resources in the slices [8]. In [17], the issue of coexistence between different types of services and resource management in beyond-5G and 6G networks was discussed. To overcome these challenges, different RAN-Slicingbased solutions and mechanisms have been proposed in the literature and can be divided into four blocks: user-centric solutions, inter-slice-based and intra-slice-based techniques and orchestrator algorithms, which combine the above solutions by means of resource planning algorithms. Table 1 shows a taxonomy of the related works presented in this Section, according to the proposed RAN-Slicing solution. User-centric mechanisms respond to resource estimation techniques in which the network allocates resources according to the requirements of users at a specific time. In the taxonomy, the different solutions offered to the challenges posed by RAN-Slicing in next-generation cellular networks are presented. Some authors use resource estimation techniques to reduce the number of communication deadlocks, such as [9], which poses an optimisation problem to maximise the resources allocated to UEs by using resource reservation operations, depending on the network demands. In [18], the authors presented a novel latency-sensitive 5G RAN slicing solution based on partitions of radio resources among slices, considering the rate and latency demands of applications. A resource reservation scheme in factory-like environments was proposed in [19] using optimisation techniques. Numerous studies have been carried out in intra-slice solutions, which are solutions based on resource sharing within a slice. In [20], the authors presented a statistical model that characterises resource sharing in a RAN-Slicing scenario, considering the available resources in layer 3. A solution for resource sharing is also presented in [21], but making use of genetic algorithms to maximise long-term network utility in Slice as a Service (SlaaS). A resource allocation slicing policy for inter-slice isolation across Mobile Virtual Network Operators (MVNOs) was investigated in [22] with a multi-objective optimisation that minimises the inter-slice 113664 VOLUME 10, 2022 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management TABLE 1. Taxonomy of related works divided by type of proposed solution (Prop.) and compared with the proposal developed in this article. interference generated by the simultaneous multiplexing of resource blocks. Intra-Slice-based resource management solutions propose techniques which share resources between two or more slices. In [23], the authors presented a complete solution for dynamic RAN-Slicing resource allocation, where the optimal slice configuration was computed through a joint evaluation of the slice Service Level Agreements (SLAs) and the real-time evolution of the served traffic of the users. The research carried out in [10] also needs to be highlighted, in which an optimisation problem is proposed to maximise the dynamic allocation of resources in eMBB and URLLC service types. In other works, a customised shape-based heuristic algorithm for users to improve resource utilisation and QoS fulfilment was presented [24]. The solutions based on orchestration algorithms, that is algorithms that manage access to segmented network resources, which are presented in [8] and [11], are focused on 6G Communications and use machine learning techniques to improve the QoS performance for the users and in the whole network, respectively. In [25], an orchestration algorithmbased solution deployed in an experimental architecture was evaluated, achieving high flexibility and scalability by employing SDN and NFV technologies. Our contribution is encompassed in orchestration-based solutions and provides a chained algorithm for RAN-Slicing resource management, which is applied to 6G Communications to improve network reliability, capacity, and latency, and its performance has been tested by simulations. It combines user-centric channel-estimation techniques and a resource pooling mechanism for dynamic resource allocation in inter-slice domains and intra-slice resource reassignment. This proposed heuristic solution allows resource reconfiguration at service time when the network is in operation. In addition, in [8] and [21], proposals whose objective is real-time resource reconfiguration were presented, although the proposals use machine learning and genetic algorithms, respectively, contrary to our proposal. In addition, these proposals focus on network metrics, such as the reconfiguration of VNFs to minimise the computation time of network devices and network utility. In our proposal, usercentric resources are studied in comparison to other studies. Furthermore, in [23], the study of mechanisms that can be deployed at service time was also carried out, but by means of a small-scale experimental network. However, our proposal focuses on a dense urban environments and is tested by simulations. The aforementioned chaining of mechanisms is carried out by a slice orchestrator, who knows the state of all cells and slices. Compared to other works, the considered scenario in which the proposed techniques are deployed is a heterogeneous network, e.g., in [11], another RAN-Slicing strategy was developed, but a mechanism for a single-cell was considered. In contrast to the proposal presented in this paper, in [25] and [20], focus was placed on network slicing solutions at higher layers, without considering the cellular network or considering restrictive admission control at the network layer, respectively. The solution proposed in this work was developed to increase the reliability of the RAN, minimise latency, and maximise the capacity of the links between the BSs and UEs. IV. SYSTEM MODEL This section presents the modelling of the system used for the simulations. The simulated network is composed of several layers. The first is the RAN, which is composed of a heterogeneous Backhaul Network (BN) consisting of NBSs (macro and small) and their links, distributed over the simulation map. The BN is represented by a set of BSs as BNBS = {BS1,BS2,...,BSN}. The second layer is the slice orchestrator, which controls the resource management logic of all the BSs. The links VOLUME 10, 2022 113665 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management TABLE 2. Summary of metrics used in the simulations. FIGURE 2. Diagram of the deployed network used for the simulations. are dedicated to each BS and are considered lossless. The last tier of the system model is composed of a group of K User Equipment (UE), defined by U= {U1,U2,...,UK}, randomly distributed for the scenario and following a Fluid Flow (FF) mobility model. These UEs are modelled by a MIMO array of antennas. Figure 2shows the layout of the network scenario. A. COMMUNICATION MODEL The communication between UE uand BS iis established by means of a millimetre wave (mmWave) link Wu,i, which is generated by a link planning policy based on Signal-ToInterference-Plus-Noise-Ratio (SINR). This link has a maximum capacity set by antenna technology and bandwidth resources that can be served by the BS. Furthermore, the link has latency, which refers to the time it takes a signal to reach from a sender to a receiver. This definition is explained in Section IV-A4. 1) SINR LINK PLANNING The link planning algorithm consists of evaluating the SINR level of all BS ∈ {BNBS}and selecting the one that offers the highest value. When a UE u∈ {U}needs to connect to a BS i∈ {BNBS}, the SINR is calculated as follows: SINRu=PRX(i,u)(mW) "N P j=1j6=i PRX(j,u)(mW)#+PN0(mW) (1) where PRX(i,u)is the power received by the BS ifor the UE uin milliwatts, PN0is the noise power in milliwatts, and N P j=1j6=i PRX(j,u)is the interference, i.e. the sum of the power received, by all BS j,∀j∈ {BNBS} − {i}that works at the same frequency. The received power PRX is calculated using the well-known link budget formula: PRX (dBm)=PTX (dBm)+GTX (dBm) +GRX (dBm)−PL(dB) (2) where PTX represents the transmit power in dBm, GTX and GRX are the transmitter and receiver gain respectively, and PL is the path losses of the link. 2) PROPAGATION CHANNEL AND PATH LOSSES The free-space path losses (PL) follow the ABG model standardised by 3GPP in [26]. This model is a large-scale propagation path loss model. It can be parameterised in terms of distance, frequency, and shadow factor. The formula that describes its behaviour is: PL(dB)=PLABG(f,d)[dB]=10αlog10(d 1m)+β +10γlog10(f 1GHz)+XABG σ(3) where PLABG(f,d) denotes the path loss in dB over frequency fand distance d.αand γare coefficients showing the dependence of path loss on distance and frequency, respectively. βis an optimised offset value for path loss. XABG σis the shadow factor of the ABG model. For each propagation scenario, the α,β,γ, and σvalues vary. Table 3shows the parameters used for the simulations depending on the scenario (Scen.) and environment (Env.) type, where dis the distance range in meters, fis the frequency range in GHz and βand σare expressed in dB. UMa,Umi and Ind. correspond to Urban MacroCell, Urban microcell and Indoor scenarios, respectively. 3) LINK CAPACITY MODEL Link capacity is the product of the spectral efficiency Sc and the bandwidth BWuassigned to UE u,, and represents the maximum amount of data that can be transmitted over a communication link. Is measured in Mbps. The spectral 113666 VOLUME 10, 2022 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management TABLE 3. Parameters of the ABG model used in the simulations [27]. efficiency, measured in bps/Hz, is defined by: Sc=log2det INRX +SINR NTX H∗HT0 (4) where det[] is the determinant of [], INRX is the identity matrix whose dimensions are the number of receiver MIMO antennas, NTX is the number of transmitter antennas and H is the channel matrix, which is generated randomly using a complex normal distribution. HT0is the conjugate transpose of the matrix [28]. The rows and columns of the channel matrix are defined by the numbers of receiver and transmitter antennas, respectively. To estimate the channel conditions, it is necessary to randomly generate Hmatrices using complex normal distribution N(µ, σ 2), according to the parameters detected in the reception of the signal: H0= NH X M XH1(Ntx ×Nrx )∼N(µ, σ 2) √2 +jH2(Ntx ×Nrx )∼N(µ, σ 2) √2(5) where H0is the estimated Hmatrix composed of NHmatrices of size Ntx ×Nrx, summed Mtimes, which corresponds to the number of samples used. The final capacity was the average of all samples obtained. H1and H2correspond to the real and imaginary parts of the H-matrix, respectively. The objective of the proposed mechanism is to offer the highest link capacity between between UE and the BS. Therefore, it is necessary to determine the highest BW assigned to the UE and increases the spectral efficiency by finding the highest SINR that can be offered by the BS to which it is connected: max(Cu,i)=max(BWu,i,SINRu,i) (6) 4) LATENCY MODEL The latency model use is a composition of three values affecting uplink communication [29]. First, the propagation time Tprop, which is the time required for the wave with the information to travel from the transmitter to the receiver. Then, there is the tail time Ttail, which is the time required for information to wait in the BS queue. Finally, the handling time Thand , that is the response time of the BS computing devices: Ttotal =link latency =Tprop +Ttail +Thand (7) assuming that Thand =1 µ(1−β), and Ttail =β µ(1−β)follows aGI|M|1 queue model.1In our system modeling, the base station queues are considered infinite; therefore, handling and tailoring times are negligible. The propagation time is defined as follows: Tprop =2(tslot −E[Tv]) 1+ferr δ(f,d) √2σ(8) where tslot is the slot time between resource blocks defined by the standard, ferr is the error function and δ(f,d)=PTX + PN0−PL(f,d). E[Tv] refers to the propagation characteristics produced by mobile blockers.2This implies that there are moments in time when the link has No-Line-of-Sight (NLoS). These mobile blockers are modelled following an M|G1|∞ queue, where the arrival is interpreted as the crossing between a blocker and the LOS link. This blockage time distribution can be approximated by using the mean waiting times: E[Tv]=E[TLOS]E[TNLOS ] E[TLOS]−E[TNLOS ](9) where E[TLOS] is the mean time that the link is not blocked and E[TNLOS] is the mean time that the link is blocked by a mobile blocker. In order to estimate the link latency, the following equation is used: L0 u,i=tpilot (u,i)−T0 prop(Wu,i),∀i∈ {BNBS },∀u∈ {U} (10) where L0 u,iis the estimated latency, tpilot (u,i) is the time taken for a pilot signal to travel from UE uto BS iand T0 prop(Wu,i) is the estimated propagation latency of the link. The objective of the proposed mechanisms is to provide the lowest possible latency in the link between the UE and BS. Therefore, it is necessary to minimise the impact of mobile blockers and determine the highest power received by the BS to which the UE is connected: min(Lu,i)=min(E[Tv]),max(δ(f,d)) (11) B. SERVICES AND TRAFFIC MODELS Numerous demands generated by the users are modelled according to various traffic models defined by 3GPP [30] and other entities, compiled in [31]. Table 4presents the definition of the implemented traffic models and their priorities. Each traffic model has been assigned a priority P(a real number between 1 and 10) according to the capacity and latency requirements, depending on the type of services defined in the previous sections. In addition, the type of service assigned to traffic models is indicated. 1GI|M|1 is a queue in which inter arrival times follow a general arbitrary distribution (G), service times follow an exponential distribution (M), and 1 denotes that the model has a single server. The services times for handling and processing queues are µand βrespectively. 2Mobile blockers are objects that temporally interpose themselves in the Line-of-Sight (LoS) of the link between the transmitter and receiver (e.g. pedestrians, vehicles, etc). VOLUME 10, 2022 113667 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management TABLE 4. Implemented traffic models in simulation framework. FIGURE 3. Schematic operation diagram of DC technology. UE demands, defined by Du, will consume the available resources of the UE-BS link Wu,i. If the BS or Wu,ihave lack of available resources to be satisfied, the demand will be blocked and discarded. These blockages are used as a metric of reliability, i.e., the number of errors in the traffic flows [42]. C. MULTI-CONNECTIVITY MODEL According to [43], multi-connectivity (MC) can be used to enhance user throughput, coverage, and/or reliability. In terms of latency, to reduce it from the communication system. Complementary dual links are used for load balancing in the case of capacity or latency requirements. Dual Connectivity (DC) technology (standardised in 3GPP Release 12 [44]) has been implemented in the simulation tool. As stated in Release 12, the UE must be configured to ‘‘utilise radio resources provided by two distinct schedulers, located in two NodeBs connected via a non-ideal backhaul’’, i.e., the UE is simultaneously connected to two non-collocated nodes (master and secondary). These links do not have to operate at the same frequency or be of the same cell type. In fact, the UE is trying to be connected to a MacroBS and SmallBS simultaneously. The secondary links were activated and deactivated according to the chosen schedule and according to the needs of the UE at any given time. A schematic of this operation diagram is shown in Figure 3. The schedule of these links considers the needs of the UE connection. If the demand cannot be served because there are no resources available on the link, the scheduler commands a secondary link to be opened. When the demand is served and terminated, it is closed. FIGURE 4. Decision diagram of the slice orchestrator algorithm. V. PROPOSED ALGORITHMS In this section, a chained orchestration algorithm for 6G RAN-Slicing resource management is described. It is based on dynamic resource management in a network, focusing on the capacity, latency, and reliability. This algorithm acts in a cascading process, that is, by chaining several techniques one after the other to reduce the number of blocked connections of the users. A flow diagram of the orchestrator decisions is shown in Figure 4. This orchestrator solution is divided into three blocks, depending on the part of the network in which it operates: •Resource estimation mechanisms: This block covers user-centric techniques based on channel estimation. This estimation can be used to determine the latency or capacity. 113668 VOLUME 10, 2022 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management FIGURE 5. Process of sending and receiving pilots. The UE, without dropping the main link or stopping the information flow, continuously sends and receives pilots from other nearby BSs. •RAN-centric intra-slice algorithms: In this block, a resource pooling technique is proposed for transferring resources between BSs inside the same slice. This mechanism is based on a DC. •RAN-centric inter-slice algorithms: For the transfer of resources between different slices, different algorithms for efficient RAN bandwidth management are proposed in this block. The type of service and criticality of the communication will be the determining factors that will enable this set of mechanisms. The slice orchestrator established in the RAN must know the status of the entire network and incoming demands. Knowing the behaviour, the algorithms can be run in cascade, i.e., triggered when the previous mechanism fails to take effect. A. USER-CENTRIC MECHANISMS FOR RESOURCE ESTIMATION The proposed resource estimation mechanism is based on the needs of the service and its criticality. If a UE needs lower latency or more capacity than its link can offer, it will connect to another BS that can guarantee these resources. These resources were guaranteed by reducing the number of blocked connections. Therefore, communication errors are reduced, i.e., reliability is increased. These solutions are based on channel estimation, which provides the UE with a complete view of the state of the RAN, in agreement with the two metrics to be evaluated. The use of pilot signals is necessary to estimate link and channel properties. Pilots are signals that are used for supervisory, control, equalisation, continuity, synchronisation, or reference purposes. This method delegates the computation of the estimation to the UE and not to the RAN. For this purpose, the UE sends a pilot to the BS, which is returned by the BS. The operating scheme is illustrated in Figure 5. 1) LATENCY ESTIMATION The time taken for the pilot to reach from the UE to the BS is the estimated propagation latency tprop. The UE knows only the total time between sending its pilot and receiving it from the BS. This time is denoted as tpilot . This time is broken down into the sum of: (i) the time it takes for the pilot of the UE to reach the BS (the same as tprop), (ii) the time it takes for the BS to process that pilot and send its own Tproc and (iii) the time it takes for the pilot of the BS to reach the UE tBS. tprop and tproc are propagation times following the Equation 8, and are calculated as a function of transmit power PTX and frequency link f:tpilot =tprop(PTXUE ,f)+tproc + tBS(PTXBS ,f) Knowing the total time (pilot time), and knowing from the pilots the frequency of the link and the transmit power of the BS, t0 BS can be calculated. Because the processing time is negligible, the link latency between the UE and all BSs can be estimated. 2) CAPACITY ESTIMATION Capacity modelling, as explained before, is a function of the number of transmitter and receiver antennas (Ntx and Nrx, respectively), bandwidth BWu, channel properties Hand SINR. To estimate the link capacity C0, the UE must know the characteristics of receiving system, such as the bandwidth and the number of MIMO antennas at the link peer. Pilot signals provide this information through the same mechanism, returning some information from the BS. The number of receiving MIMO antennas is available in the link information, and the bandwidth is known to be the bandwidth allocated to the UE if it is connected. This bandwidth is allocated using a simple resource management policy. B. RESOURCE MANAGEMENT IN RAN-SLICING Each BS i∈BNBS has a certain amount of resources, grouped in a tuple Ri= {Ci,Li,u}. Each Cicorresponds to the available capacity per BS, and each Li,udefines the estimated latency that BS ican offer to UE u, depending on its position and propagation channel, based on resources management planning (with or without a slice schedule). R(u) are the resources allocated to UE u. The base case of this resource management planning is defined by Algorithm 1. This algorithm proposed by 3GPP is the baseline for numerical results [45]. The algorithm works as follows: a user uis chosen from the set of active users in the scenario (Line 1). The user is connected to the BN (Line 2). This connection establishment uses a well-known schedule based on SINR, following Eq. 1. The signal level of a base station and the interference received by all the others are evaluated. After evaluation, u is connected to the one that offers the highest SINR level. This candidate base station is referred to as i(Line 3). Once the connection is established, ubegins to generate traffic demands Du(Line 4). These demands will be characterised by the required throughput and the minimum latency it needs to be established (Line 5 and 6 respectively). Both metrics are encompassed in a tuple R(u), which represents the resources needed to satisfy the demand (Line 7). If the available resources by BS i(Ri) are greater than or equal to those needed to satisfy the demand (Line 8), the demand VOLUME 10, 2022 113669 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management FIGURE 6. Operation of proposed RAN-based resource allocation techniques. (a) intra-slices techniques for reallocation of resources between different slices in the same BS (when there are no resources available within the same slice). (b) inter-slices techniques for sharing resources between different BSs within the same slice (DC). Algorithm 1 Base Case Algorithm for Resource Management (Without RAN-Slicing) Input: U←Set of UEs BN ←Backhaul Network (set of BSs) Ri= {Ci,Li,k} ← Available resources from BS i begin [1]: foreach uin Udo [2]: Connect uto BN [3]: BS i←candidate BS ∈BN [4]: Du←ugenerates a traffic connection [5]: C(u)←required throughput by Du [6]: L(u)←min. latency required by Du [7]: R(u)← {C(u),L(u)} [8]: if R(u)≤Rithen [10]: Assign R(u) to u [9]: Ri←Ri−R(u) else [11]: U0←subset of Uconnected to BS i [12]: R0(u)←Ri len(U0)+1 [13]: Assign R0(u) to u [14]: Assign R0(u) to each u0in U0 [15]: Ri←0 end if end foreach end can be established. The resources that uneeds are allocated to it (Line 9) and the available resources by the BS iare updated, subtracting those it has allocated (Line 10). When Ri<R(u), i.e., BS ihave no available resources to satisfy the demand, two events can occur. On the one hand, if Ducannot support a reduction in QoS and cannot be served with R0(u) because of their priority and the criticality of their type of service, the connection will be blocked. On the other hand, if the traffic can be served with R0(u)<R(u) even if the QoS decreases, the connection will be established with fewer resources (lower QoS). In this case, the resources of all users connected to that base station are reallocated. First, a subset of users who comprehend the users connected to BS iis extracted. This subset is denoted by U0(Line 11). Second, a newly assigned resource is calculated using the formula presented in Line 12. The available resources Riare divided by the number of connected users and the number that needs a connection. These new resources are assigned to u(Line 13), and consequently reassigned to all users of U0, which were previously connected to BS i(Line 14). Finally, the resources available at the base station are updated (Line 15). If any of the demands of users belonging to U0do not support a decrease in their QoS, the demand is blocked. The total complexity of the algorithm is O(n2+n) for each user uin U. The connection of user uto the BN has complexity O(n2+n), depending on the number of BSs in the scenario. This algorithm is based on Eq. 1: for each evaluated BS i∈BN, it is necessary to iterate all BSs j6= i. In the worst case, after the connection, all users in U0must be iterated to reallocate their resources. The complexity of this operation is O(n). 1) RAN-SLICING IN CELLULAR NETWORKS Consider S= {S1,S2,...,SM}a set of Mslices in the RAN. Each slice Smis distributed over the BN. Depending on the type of service it has been assigned (in this case, eURLLC, LDHMC, ELPC, feMBB and umMTC), Smwill have access to a set of resources Rm, which are represented by a tuple Rm= {Cm,Lm}, entailing portions of the BSs resources. In turn, each slice hosts a set of UEs U0 Sm∈U of dimension N0≤dim(U). UEs belonging to U0 Smare 113670 VOLUME 10, 2022 J. J. Rico-Palomo et al.: Chained Orchestrator Algorithm for RAN-Slicing Resource Management JOSE J. RICO-PALOMO was born in Badajoz, Spain, in 1996. He received the B.Sc. degree in telecommunications engineering, specializing in telematics engineering, from the Centro Universitario de Mérida of University of Extremadura, Spain, in 2017, and the M.Sc. degree in telecommunications engineering from the School of Technology of University of Extremadura, in 2018. He is currently pursuing the Ph.D. degree. In 2018, he completed a Research Fellowship at the CénitS Supercomputer Center. He is with the GITACA Research Group, Department of Computing and Telematics System Engineering, University of Extremadura. His current research interests include 5G and 6G communications, and propagation channels and models for next generation cellular networks and mobility management. JESUS GALEANO-BRAJONES received the B.Sc. degree in telecommunication engineering, specialising in telematics, the B.Sc. degree in computer science engineering from the Universidad de Extremadura, Spain, in 2019, the M.Sc. degree (Hons.) in research in 2020. He is currently pursuing the Ph.D. degree with the Department of Computing and Telematics System Engineering, Universidad de Extremadura. He is a member of the GITACA Research Group. His research interests include optimisation algorithms, next-generation networks, artificial intelligence, and network security. DAVID CORTES-POLO received the degree in computer science and the Ph.D. degree in telematics from the University of Extremadura, Spain, in 2015. He was a Research and Teaching Assistant at University of Extremadura, from 2011 to 2014. Since 2011, he has been the Network Manager with COMPUTAEX Foundation and the CénitS Center. His main research interests include IP-based mobility management protocols, performance evaluation, and quality of service support in future mobile networks. JUAN F. VALENZUELA-VALDES was born in Marbella, Spain. He received the degree in telecommunications engineering from the Universidad de Málaga, Spain, in 2003, and the Ph.D. degree from the Universidad Politécnica de Cartagena, Spain, in May 2008. In 2004, he joined the Department of Information Technologies and Communications, Universidad Politécnica de Cartagena. In 2007, he joined the Head of Research with EMITE Ing. In 2011, he joined the Universidad de Extremadura, and in 2015, he joined the Universidad de Granada, where he is currently an Associate Professor. He was a Co-Founder of EMITE Ing.—a spin-off company. He also holds several national and international patents. His publication record is composed of more than 80 publications, including 40 JCR indexed articles, more than 30 contributions in international conferences and seven book chapter. His current research interests include wireless communications and efficiency in wireless sensor networks. He has also been awarded several prizes, including the National Prize to the Best Ph.D. in mobile communications by Vodafone and the I-Patents Award by the Spanish Autonomous Region of Murcia for innovation and technology transfer excellence. JAVIER CARMONA-MURILLO received the Ph.D. degree in computer science and communications from the University of Extremadura, Spain, in 2015. From 2005 to 2009, he was a Research and Teaching Assistant. He has spent research periods with the Centre for Telecommunications Research, King’s College London, U.K., and Aarhus University, Denmark.Since 2009, he has been an Associate Professor with the Department of Computing and Telematics System Engineering, Universidad de Extremadura. His current research interests include 5G networks, mobility management protocols, performance evaluation, and the quality of service support in future mobile networks. VOLUME 10, 2022 113677