Exploring Relay Technology for 5G and Beyond Systems through the use of a Network Digital Twin

Exploring Relay Technology for 5G and Beyond Systems through the use of a Network Digital Twin Master Thesis submitted to the Faculty of the Escola T`ecnica d’Enginyeria de Telecomunicaci´o de Barcelona Universitat Polit`ecnica de Catalunya by Ferran Font Pons In partial fulfillment of the requirements for the master in Advanced Telecommunication Technologies ENGINEERING Advisor: Dr. Jordi P´erez-Romero Barcelona, June 2023 Abstract This research project focuses on leveraging User Equipment (UE) as relaying devices to enhance the Radio Access Network (RAN) infrastructure, aiming to improve RAN performance in terms of capacity, outage probability, and spectral efficiency while reducing resource consumption. Extensive simulations and experiments were conducted using the Network Digital Twin (NDT) simulator, incorporating various models to replicate realistic scenarios. Initially, experiments were performed under a basic traffic model, which was later enhanced with a traffic model based on real Access Point (AP) traffic measurements. The findings provide valuable insights into resource consumption, network congestion, and link conditions, emphasizing the benefits of utilizing relays to optimize resource utilization, mitigate congestion, and enhance overall network performance. 2 Acknowledgements I would like to express my heartfelt gratitude to my supervisor, Dr. Jordi P´erez-Romero, whose invaluable support and guidance made this work possible. His patience and assistance were instrumental not only during the entire project but also throughout the various subjects I pursued under his supervision. I would also like to extend my thanks to all the members of the Mobile Communications Research Group for their continual advice and support. Furthermore, I am deeply appreciative of my parents for their unwavering love and sacrifices in preparing and educating me for my future. Additionally, I would like to acknowledge my friends, and in particular, my brother Marc, for their inspiration and unwavering support throughout my academic journey. This project is for you, avi! 3 Contents Abstract 2 Acknowledgements 3 List of Figures 6 List of Tables 8 1 Introduction 10 1.1 Statementofpurpose.............................. 11 1.2 Outlineofthework............................... 11 1.3 Technicalremarks................................ 12 1.4 GanttDiagram ................................. 12 2 Relay Technology 13 2.1 Fundamentals of relay technology . . . . . . . . . . . . . . . . . . . . . . . 13 2.1.1 TypesofRelay ............................. 14 2.1.2 RelayConnection............................ 14 2.2 Integrated Access Backhaul (IAB) . . . . . . . . . . . . . . . . . . . . . . . 15 2.2.1 IABArchitecture............................ 16 2.3 Device-to-Device (D2D) communications . . . . . . . . . . . . . . . . . . . 17 3 Network Digital Twin (NDT) 19 3.1 BenefitsofaNDT ............................... 19 3.1.1 Reducing cost of operation . . . . . . . . . . . . . . . . . . . . . . . 19 3.1.2 Improved decision-making . . . . . . . . . . . . . . . . . . . . . . . 20 3.1.3 Enhanced safety in evaluating network innovations . . . . . . . . . 20 3.2 RANNDT.................................... 20 3.2.1 NDT Architecture for the RAN . . . . . . . . . . . . . . . . . . . . 21 4 NDT simulator 23 4.1 Scenariodescription............................... 23 4.2 Functionalities.................................. 24 4.2.1 Trafficgeneration............................ 24 4.2.2 Propagationmodel ........................... 28 4.2.3 Mobilitymodel ............................. 29 4.2.4 Connectivity model . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 4.3 Spectralefficiency................................ 30 4.3.1 MCS................................... 30 4.3.2 Inefficiency factors . . . . . . . . . . . . . . . . . . . . . . . . . . . 31 4.3.3 Finalmodel............................... 32 4.3.4 Outage probability . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 4.4 PRBsocupation................................. 34 4.5 Performance indicators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 4 4.6 Simulationsteps................................. 37 4.7 Simulationparameters ............................. 38 5 Results and discussion 39 5.1 Analysis on resource consumption under the basic traffic model . . . . . . 39 5.1.1 Average network load . . . . . . . . . . . . . . . . . . . . . . . . . . 40 5.1.2 Congestionstudy............................ 42 5.1.3 Access link occupation . . . . . . . . . . . . . . . . . . . . . . . . . 45 5.1.4 Outage probability . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 5.2 Analysis on resource consumption under the measurement-based traffic model 47 5.2.1 Average network load . . . . . . . . . . . . . . . . . . . . . . . . . . 48 5.2.2 Congestionstudy............................ 51 5.2.3 Access link occupation . . . . . . . . . . . . . . . . . . . . . . . . . 53 5.2.4 Activityfactor ............................. 55 5.2.5 Outage probability . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 5.3 Smaller-scale scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 5.3.1 Coverage optimization . . . . . . . . . . . . . . . . . . . . . . . . . 60 5.3.2 Capacity optimization . . . . . . . . . . . . . . . . . . . . . . . . . 63 6 Conclusions and future work 65 References 67 5 List of Figures 1 Project’s Gantt diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 2 Scheme of access and backhaul link . . . . . . . . . . . . . . . . . . . . . . 13 3 SchemeRelaysituations ............................ 15 4 User plane and control plane protocol stack of a multi-hop IAB network according to 3GPP Rel-16 . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 5 D2DCommunications ............................. 18 6 Scenariodescription. .............................. 23 7 APs positions in the different floors. . . . . . . . . . . . . . . . . . . . . . . 26 8 APs coverage in the different floors. . . . . . . . . . . . . . . . . . . . . . . 28 9 Pedestrians walking area . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 10 TDDstructure ................................. 32 11 Net spectral efficiency map of the ground floor for the three base stations. . 33 12 Best Spectral Efficiency map . . . . . . . . . . . . . . . . . . . . . . . . . . 34 13 SchemeofPRBsusage ............................. 35 14 Total PRBs used under the basic traffic model . . . . . . . . . . . . . . . . 40 15 Total average users under the basic traffic model . . . . . . . . . . . . . . . 41 16 Average PRBs used per user under the basic traffic model . . . . . . . . . 42 17 Resource occupation for the three base stations under the basic traffic model. 43 18 Network congestion under the basic traffic model . . . . . . . . . . . . . . 44 19 Distribution of access link occupation under the basic traffic model . . . . 45 20 Distribution of users connected to a relay under the basic traffic model . . 46 21 Outage Probability under the basic traffic model . . . . . . . . . . . . . . . 47 22 Total PRBs used under the measurement-based traffic model . . . . . . . . 48 23 Total average users under the measurement-based traffic model . . . . . . . 49 24 Average PRBs used per user under the measurement-based traffic model . 50 25 Resource occupation for the three base stations under the measurementbasedtrafficmodel................................ 52 26 Network congestion under the measurement-based traffic model . . . . . . 53 27 Distribution of access link occupation under the measurement-based traffic model ...................................... 54 28 Distribution of users connected to a relay under the measurement-based trafficmodel................................... 55 29 PRBs occupation in BS1 with 50 relays varying the activity factor. . . . . 56 30 Outage probability under the measurement-based traffic model . . . . . . . 57 31 Net spectral efficiency map of the ground floor when varying number of available relays under the measurement-based traffic model. . . . . . . . . 58 32 Spectral efficiency map of the 2nd floor of the C2, C3 and C4 buildings without any active relay (a) and activating the possible relays in these buildings in one simulation (b). . . . . . . . . . . . . . . . . . . . . . . . . 59 33 CDF of the spectral efficiency of the 2nd floor of the C2, C3 and C4 buildings without any active relay and when activating randomly placed relays. .......................................... 60 6 34 Spectral efficiency map of the 2nd floor of the C2, C3 and C4 buildings without any active relay (a) and activating the possible relays in these buildings to improve the coverage (b). . . . . . . . . . . . . . . . . . . . . . 61 35 CDF of the spectral efficiency in the 2nd floor of buildings C2, C3, and C4 is evaluated under three scenarios: without any active relay, with randomly placed relays, and with optimally placed relays. . . . . . . . . . . . . . . . 62 36 Spectral efficiency map of the 2nd floor of the C2 building without any active relay (a) and activating different relays in the floor below (b). . . . . 62 37 Spectral efficiency map of the 2nd floor without any active relay (a) and activating optimally placed relays in this floor to improve network capacity (b)......................................... 63 7 List of Tables 1 Userscategory. ................................. 24 2 MCS index table for PDSCH. Source [29] . . . . . . . . . . . . . . . . . . . 31 3 PRBs available. Source [32] . . . . . . . . . . . . . . . . . . . . . . . . . . 36 4 Configured parameters for the simulation. . . . . . . . . . . . . . . . . . . 38 5 Users category under the basic traffic model. . . . . . . . . . . . . . . . . . 39 6 Comparison of net spectral efficiency levels seen by the users for different trafficmodels................................... 50 8 Revision history and approval record Revision Date Purpose 0 20/05/2023 Document creation 1 13/06/2023 Document revision 2 22/06/2023 Document revision 3 28/06/2023 Document revision DOCUMENT DISTRIBUTION LIST Name Ferran Font Pons Jordi Pe´rez-Romero Written by: Reviewed and approved by: Date 27/06/2023 Date 28/06/2023 Name Ferran Font Pons Name Jordi P´erez-Romero Position Project Author Position Project Supervisor 9 offers backhaul services but also supports the provision of existing cellular services within the same entity. For instance, reference [7] offers a comprehensive overview of the multi-hop IAB techniques implemented in the 3GPP Release 16 standard, along with a discussion on its design strategies. In [8], a scheme is developed that combines node placement and resource allocation to maximize the downlink sum rate. Furthermore, reference [9] employs simulated annealing algorithms to optimize joint scheduling and power allocation in IAB networks. 2.2.1 IAB Architecture The IAB standard, adopts the split architecture initially introduced in 3GPP Release 15. The IAB donor, represented by a gNodeB, comprises multiple digital units (DU) and a control unit (CU). The CU serves as the controller for the DU and establishes a connection with the 5G core through the NG interface. Each DU is linked to the CU of the IAB donor via the F1 interface, enabling the CU to govern the higher-layer signaling for all associated child IAB nodes. An IAB node consists of two components: the mobile termination (MT) and the DU. Within the IAB architecture, the MT portion of an IAB node is virtually perceived as a user equipment (UE) by the parent IAB node. The MT can solely receive the downlink (DL) backhaul link from a DU and transmit the uplink (UL) backhaul link towards the DU [10]. The control and user plane protocol stack of a multi-hop IAB network, is depicted in Figure 4. In this scheme, the IAB donor node serves as the connection point to the wider network using traditional means such as fiber or microwave. It provides services to the IAB nodes as well as other UEs directly linked to it. 16 Figure 4: User plane and control plane protocol stack of a multi-hop IAB network according to 3GPP Rel-16. Source[11] 2.3 Device-to-Device (D2D) communications Device-to-Device (D2D) communications Introduce an alternative approach for implementing relaying technology. In this context, devices such as cell phones, tablets, and laptops can serve as transmission relays for one another within the network. This implementation extends the benefits of relaying technology to portable wireless devices, maximizing their efficiency and connectivity. It unlocks the full potential of cellular systems [12]. The concept of D2D communications was initially introduced by 3GPP in LTE Rel.12 under the Proximity Services (ProSe) idea, which focused on addressing public safety-related scenarios. This concept was later extended to encompass vehicle-to-vehicle communications as well. With the advent of 5G NR, Release 16 brought further advancements and enhancements to D2D capabilities [13]. D2D operation involves the establishment of a direct communication link between UEs without the need for the traditional BSs as an intermediary. LTE and 5G NR defined a new PC5 interface between UEs and introduced a sidelink radio link arrangement for direct transmissions between devices. 17 Figure 5: D2D Communications. Figure 5 illustrates the various communication links involved in D2D communication, including the sidelink established between different UEs using the PC5 interface. The sidelink communication offers two options for resource allocation, depending on the entity responsible for the allocation. •Scheduled resource allocation: The allocation of resources in the sidelink is done dynamically by the base station, ensuring efficient utilization. •UE autonomous resource selection: The UE autonomously chooses resources from designated resource pools and performs transport format selection to transmit sidelink control information and data. The relay communication can be established using in-band D2D, where the D2D link and the cellular link share the same spectrum, or through out-of-band D2D, where the D2D link and the cellular links operate on different frequency bands or even different technologies [14]. Recently, in 3GPP Release 18, the connectivity models have incorporated the functionality of D2D relaying, specifically under the concept of UE to network (U2N) relaying[15]. Various scenarios for the use of relay UEs, such as in-home, smart farming, smart factories, or public safety, have been identified, along with specific requirements and key performance indicators [16]. 18 3 Network Digital Twin (NDT) The NDT is a virtual and continually updated depiction of the network, providing the capability to analyze, diagnose, and emulate the physical network within a risk-free environment. By utilizing the NDT, network operators can make informed control decisions for the physical network [1]. This allows operators to safely explore novel techniques and configurations without the need to undertake risky operations on the actual network infrastructure, providing a secure platform for experimentation and innovation. As previously mentioned in Section 1, the design of the NDT requires a high-fidelity replica of the network, which should be accomplished with efficient execution time and optimal utilization of computational resources. This entails striking a balance between the complexity and accuracy of the NDT while ensuring a high degree of flexibility to update it in line with the constant changes occurring in the physical network [2]. While the development of NDTs for 5G and B5G networks is still in its early stages, researchers have conducted some studies on establishing digital twins for various subsystems of a network. In 2019, Dong et. al. [17] developed a digital twin for a 5G mobile edge computing (MEC) network. They utilized the offline twin to train optimization algorithms for resource allocation and energy-saving based on reinforcement learning. The findings were then applied to the actual MEC network. Similarly, Y. Dai et. al. [18] created a digital twin for an edge network in a mobile edge computing system. They employed a twin edge server to evaluate the state of entity servers, enabling the twin mobile edge computing system to provide data for training offloading strategies. Furthermore, Nguyen et. al. [19] explored the deployment of a digital twin for complex 5G networks, discussing its potential implementation and benefits. Finally, a digital twin platform for automatic and intelligent management of data center networks is introduced in [20]. They proposed simplified workflows for network service management based on the digital twin platform. 3.1 Benefits of a NDT The use of a NDT offers numerous benefits for network operators. The main distinction between a NDT and a simulation platform lies in their approach to network management. While both can simulate network behavior, an NDT goes beyond simulation by providing a real-time representation of the physical network. This allows for closed-loop network automation and enables operators to make data-driven decisions and optimize network performance based on real-time interactions between the digital twin and the physical network [21]. 3.1.1 Reducing cost of operation Operating large-scale networks is a complex task. However, due to the lack of effective simulation platforms, network optimization designs often have to be tested directly on 19 the physical network. This approach carries the risk of disrupting the network’s daily operation and potentially compromising the quality of services it provides. By using a NDT platform, network operators can safely emulate and assess potential optimization solutions before implementing them in the physical network. This allows for risk-free evaluation and reduces the need for costly deployments on the real network. 3.1.2 Improved decision-making Traditional network management primarily concentrates on the deployment and ongoing management of services, lacking adequate support for predictive maintenance methods. However, a NDT integrates data acquisition, big data processing, and machine learning (ML) modeling to evaluate the network’s current status, predict future trends, and enhance predictive maintenance strategies. By accurately replicating network behaviors in different scenarios, operators can assess multiple evolution options as frequently as needed, enabling informed decision-making processes. 3.1.3 Enhanced safety in evaluating network innovations Deploying and testing new features in an operational network can be complex and risky. It is crucial to conduct a thorough service impact analysis before activating these features. However, using a digital twin network enables the evaluation of innovative network capabilities without disrupting the ongoing operations of the physical network. This empowers researchers to efficiently explore novel network protocols and machine learning applications. Moreover, network operators can quickly and safely deploy new technologies, minimizing potential risks associated with their implementation. 3.2 RAN NDT The Radio Access Network (RAN) subsystem plays a vital role in wireless communication systems by connecting user devices to the core network. However, the management and optimization of RAN deployments have become increasingly challenging in the context of complex 5G networks and beyond [22]. The emergence of the NDT concept offers a unique and promising solution to address these challenges. By providing a virtual representation and analysis of the RAN, the NDT helps the network operators gain valuable insights and effective tools for optimizing and efficiently managing the network. The applications of NDT in the context of the RAN encompass various management functions, including planning, operation, and optimization. In the planning phase, NDTs can assess different RAN deployment topologies and configurations prior to actual deployment. This allows for their evaluation under diverse conditions, such as varying traffic load levels [23]. In terms of network operation, NDTs contribute to real-time monitoring and anomaly detection by leveraging the prediction capabilities of the NDT models. Network optimization processes can also benefit from NDTs, as they enable the fine-tuning 20 of different parameter configurations and the exploration of new policies for optimization functions like load balancing, capacity sharing for network slicing, and more [24]. Another use case of NDTs in the RAN involves training ML models [25]. These ML models can address various challenges across different areas of the RAN, including physical layer processing, Medium Access Control (MAC), Radio Resource Management (RRM), Radio Network Management (RNM), and Self-Organizing Networks (SON) [2]. Specifically, Reinforcement Learning (RL) techniques have gained a special interest in the RAN due to their ability to optimize decision-making tasks. RL has been successfully employed in tasks such as cell selection, channel selection, resource allocation, power allocation, small cell activation/deactivation for energy efficiency, and adaptive modulation and coding [26]. By leveraging NDTs for RL training, the RAN benefits from enhanced decision-making and optimized performance, while avoiding disruptions to the live network. This approach enables extensive exploration and refinement of RL solutions within a controlled virtual environment, empowering network operators to deploy more efficient and effective RAN management strategies. 3.2.1 NDT Architecture for the RAN The architecture of the RAN NDT must enable accurate characterization of real RAN behavior and performance assessment, while also being adaptable for various applications and scenarios. The RAN NDT architecture comprises three primary modules [2]: the data repository, responsible for collecting and storing data from the actual RAN; the service mapping models, which represent the different components and operations of the real RAN; and the digital twin management, which oversees the functioning of the NDT. These modules are: •Data repository: It is crucial for collecting data from the real RAN environment and storing it for utilization by the NDT. This ensures that the NDT has an accurate and up-to-date representation of the actual RAN, while also providing access to historical data that can be leveraged by the NDT models. The data stored in the repository encompasses various types, including configurations, operational states, topology information, traces, key performance metrics, and more. •Service mapping models: It encompasses a variety of models that represent different elements and functionalities within the RAN. A modular design approach is adopted to enhance the programmability of network services and increase operational and deployment agility. These models are constantly updated with parameters derived from the data repository, reflecting the real behavior of the RAN. Examples of these elements and functionalities include the scenario topology model, traffic generation model, mobility model, and many others. Each model plays an important role in simulating and analyzing specific aspects of the RAN. •NDT management: It performs the essential task of managing the NDT. This 21 encompasses various aspects of the NDT’s life cycle, including deployment, operation, optimization, maintenance, and termination of the different models within the NDT. It also involves the integration of data from the data repository into the models and the validation of their performance, as well as managing the interactions between the models. A detailed description of the various modules and models that represent distinct elements and functionalities within the RAN for our specific NDT will be provided in Section 4. 22 4 NDT simulator In this section, the study is focused on examining and analyzing the deployment of a network utilizing UEs as relaying devices. The aim is to create an enhanced RAN that can provide improved performance to users, such as higher spectral efficiency, throughput, and energy efficiency. To accomplish this objective, a Network Digital Twin (NDT) simulator is presented as well as all the improvements provided. This first simulator was previously developed within the Mobile Communication Research Group (GRCM) at the UPC. The first version of a RAN NDT simulator is adopted, which includes the characterization of the propagation modelling in outdoor and indoor areas as well as the user mobility, among many other features. 4.1 Scenario description The simulation is focused on an urban environment, such as the Campus Nord of the UPC in Barcelona, as the computation zone for the analysis. This NDT simulator considers a scenario where a number of base stations are deployed providing service to different User Equipment (UE). Specifically, as depicted in Figure 13, the Campus Nord encompasses an area with different walking streets, squares, and with a total of twenty-four, three-floor buildings. Each floor of the building has a height of 3.5 m. Figure 6: Scenario description. 23 The model includes a 5G NR deployment with a total of 3 BSs placed above rooftop levels of surrounding buildings, as seen in Figure 13. The locations of the 3 base stations correspond to the actual ones deployed by one of the mobile network operators. It also establishes the number of available relays in the simulation. The number available is determined by a parameter that represents the number of stationary UEs that will act as relays. The selection of relays will be made from a database comprising 4000 distinct relays, each having predetermined positions and pre-calculated propagation values. These values contribute to making the simulation process faster. 4.2 Functionalities 4.2.1 Traffic generation The traffic demand in the actual network can be realistically emulated at various levels. This emulation takes into account the temporal and spatial distributions within the network area. The traffic generation process can be based on established models found in literature, such as session generation following a Poisson distribution and session durations modeled by an exponential distribution, as used in Section 4.2.1.1. Alternatively, it can utilize measurement reports generated by the UEs during their network connections, as explained in Section 4.2.1.2. 4.2.1.1 Basic traffic model During the simulation, a set of users is introduced, categorized as stationary UEs or pedestrians. Stationary UEs can be located either indoors or outdoors, and they are the ones considered for potential use as relays. On the other hand, pedestrians can be located walking the streets and the squares areas at a constant speed. The simulation program sets a fixed number of users and calculates the distribution percentages for each type of user as indicated in the following table: Total users Stationary 0.6*Total users Pedestrians 0.4*Total users Table 1: Users category. The traffic generation model employed in this first study assumes that sessions of this set of users are generated in accordance with a Poisson distribution. The provided formula is employed during each simulation time to calculate the time increment from the current time until the generation of the next session: 24 − 1 λlog(1 −rand),(1) where the generation rate for each category λ(sessions per second) is determined by multiplying the session generation rate of the UE by the number of users in that category and rand is a mathematical function that generates a random value between 0 and 1. The duration of each session is modeled as an exponential random variable with a predefined average µseconds. 4.2.1.2 Traffic model based on measurements In the previous section, a basic traffic model was introduced. However, it is important to note that this simplistic traffic model does not accurately reflect the realistic conditions of our specific scenario. In order to establish a more realistic traffic model, we collected measurements from various Wi-Fi access points (APs) situated across the entire campus. These measurements were obtained during a typical weekday, specifically during class periods in November 2019. By incorporating these real-world data, we aimed to create a more accurate representation of the traffic patterns and dynamics within the campus environment. Despite the APs measuring Wi-Fi traffic, there is an assimilation made that this traffic, would be similar to the potential traffic that could be present in the cellular network. In Figure 7, a detailed depiction is provided on the precise location of the access points on multiple levels of the building. The positions of these access points are represented for the ground floor, first floor, and second floor, offering a comprehensive visual representation of their distribution throughout the different buildings. 25 •(14/15) represents the inefficiency caused by the cyclic prefix of 5G transmissions. This value is obtained because each slot consists of 14 symbols and the duration of the slot in milliseconds depends on the numerology, or the subcarrier spacing (SCS). Each symbol has a total duration of (1/SCS) + Tpwhere Tpis the duration of the cyclix prefix. Therefore, the total net information sent in each slot is: εcyc =14 · 1 SCS Tslot =14 15 (8) Notice that Tslot is inverseley proportional to the subcarrier spacing. Therefore, the computation of these values is always 15. •0.86 represents the inefficiency attributed to control channels for the transmitting frequency in FR1 - DL in accordance to [30]. •(52/70) denotes the ratio of DL symbols in a Time Division Duplex (TDD) frame structure. The structure used in our configuration is: Figure 10: TDD structure. Source [31] The structure consists of 3 DL slots, 1 special slot with 10 DL symbols, and 1 slot for UL transmission. By recognizing that each slot contains 14 symbols, it becomes straightforward to calculate that 52 symbols are utilized for DL out of a total of 70 symbols. 4.3.3 Final model By utilizing the precedent values, we can calculate the actual net spectral efficiency considering the inefficiencies. The formula for computing the net spectral efficiency is as follows: Seff net = (m·r)·ε(9) where mand rare the modulation order and the code rate, respectively, obtained from Table 2 and εis the inefficiency factor. 32 As detailed in Section 4.1, the scenario encompasses three base stations within the computation area. Figure 11 illustrates a map displaying the net spectral efficiency levels on the ground floor for each of these base stations, providing a visual representation of the spectral efficiency distribution across the area. Figure 11: Net spectral efficiency map of the ground floor for the three base stations. Additionally, by overlaying the previous maps, it becomes possible to generate a map that highlights the base stations offering the highest levels of spectral efficiency throughout the map locations. Figure 12 visually represents this concept, which can also be interpreted as a map indicating the serving BS without considering the involvement of relays. It provides a clear depiction of the BSs that directly serve the respective areas due to their greater conditions, without any relay assistance. In other words, the map highlights the BSs that provide optimal coverage and performance in those respective areas. 33 Figure 12: . Best Spectral efficiency map. 4.3.4 Outage probability The outage probability of a communication channel relates to the probability of not being able to support a particular information rate due to fluctuations in channel capacity. It quantifies the likelihood that the information rate falls below a predetermined threshold, indicating that the receiver is located beyond the coverage range of the base stations. Consequently, the minimum required information rate to ensure reliable service cannot be guaranteed in such cases. Within the simulator, it is assumed that a minimum spectral efficiency of 1 b/s/Hz is needed to ensure satisfactory communications. If a user’s achieved spectral efficiency falls below this threshold, it indicates that the user is experiencing an outage. In other words, the user is unable to achieve the desired data rate, resulting in a degradation of the communication link. 4.4 PRBs ocupation The simulator strives to generate a realistic model for various functionalities, aiming to closely resemble real-world scenarios. However, it is important to note that the first version of the simulator did not consider the rate and capacity limitations of the base stations. This means that while the simulation may accurately capture other aspects of the system, it did not incorporate the specific constraints and limitations imposed by the base station’s rate and capacity capabilities. By using the spectral efficiency metrics, as explained in section 4.3, in this thesis we can effectively model the capacity usage and physical resource allocation within the given scenario. These metrics enable us to quantify the efficiency of data transmission and assess the utilization of available resources, providing valuable insights into the overall performance and resource management of the system. Once the net spectrum efficiency values are computed, the next step is to determine the number of PRBs consumed by each user. To compute the amount of PRBs consumed, the following procedure can be followed: 34 NP RBs =Rb (b/s) 12 ∆f Seff net (10) where Rb represents the data rate required by the user to guarantee the service and ∆f is the bandwidth of each one of the subcarriers. 12∆fconforms to the bandwidth of the resource block. It is crucial to emphasize the necessity of calculating the consumption of PRBs in each individual link that is being utilized. For instance, as depicted in Figure 13, when the UE is directly connected to the BS, we only need to determine the PRBs consumed in that specific link. On the other hand, when the UE is connected through a relay, we have both the access link and the backhaul link, and we need to calculate the number of PRBs used in each of them. These calculations are independent for each link and the values depend on the available bandwidth and the spectral efficiency of the respective link. Figure 13: Scheme of PRBs usage. By utilizing equation 10, it is possible to calculate the number of PRBs utilized in the link by each user, regardless of whether they are connected through a relay or a direct link. However, it is necessary to compare this value against the total number of PRBs available at each one of the links. To do this comparison, it is important to note that the amount of PRBs available at the direct link, access link, and the backhaul link depends on the bandwidth and the subcarrier spacing (SCS) used. In Table 3 it is represented the different combinations possibles for 5G NR that can be employed, according to [32] . 35 Bandwidth (MHz) SCS (kHz) 5 10 15 20 25 30 40 50 60 70 80 90 100 15 25 52 79 106 133 160 216 270 N.A N.A N.A N.A N.A 30 11 24 38 51 65 78 106 133 162 189 217 245 273 60 N.A 11 18 24 31 38 51 65 79 93 107 121 135 Table 3: PRBs available. Source [32] In our specific case, in the direct link the configuration involves a bandwidth of 100 MHz and a subcarrier spacing of 30 kHz. Looking into the previous table, this configuration provides a total of 273 PRBs available at each BS. A similar scenario occurs when the relay is transmitting. As the bandwidth and subcarrier spacing configuration is the same as when the BSs are transmitting, in both the access link and the backhaul link, the relay has access to a total of 273 available PRBs. We can proceed to conduct various experiments pertaining to resource occupancy, energy efficiency, and other relevant factors. These experiments will be thoroughly discussed and presented in Section 5, enabling us to gain valuable insights into the performance and optimization of the system. 4.5 Performance indicators As previously mentioned, the simulations gather a range of performance indicators for each user category. All the statistical data is collected individually for every user type. By employing the spectral efficiency and the outage probability metrics described in Section 4.3, we can extract several indicators that facilitate the analysis of system performance. The key values derived from this metric are: •Average spectral efficiency •Percentiles 95 and 5: These percentiles help provide insights into the spread and distribution of the data, especially in situations where extreme values or outliers may be present. •Cumulative Distribution Function: The CDF provides a cumulative view of the probability distribution, allowing us to determine the probability of an event occurring up to a specific value or within a given range. There are some additional valuable indicators that will help in the performance analysis. Some examples include the probability of connection to a relay, the duration of connection time to a relay, and the duration of connection time to a base station (BS). Each indicator can be analyzed using metrics such as averages, percentiles, and CDF analysis. 36 4.6 Simulation steps The simulation process involves a series of steps that are followed to ensure an accurate and reliable outcome. These steps are carefully designed and executed to simulate a specific scenario with precision. It is important to highlight that the simulations only focus on the downlink traffic, excluding any consideration of the uplink traffic. The steps carried out are: •Once the computation zone for the analysis is created and a number of base stations are deployed to provide service to a set of users. To accelerate simulation time, a relay database is accessible, containing precalculated parameters. During each simulation, a subset of relays is randomly chosen from this database. These chosen relays will be available during the simulation to serve the different users. To ensure the validity of the relays, it is essential that these relays are situated in locations that are not in outage, meaning they achieve a spectral efficiency greater than 1 b/s/Hz from their respective serving base stations. In Section 4.3, it is provided a more comprehensive understanding of the spectral efficiency concept, and the outage probability concept is more thoroughly explain in Section 4.3.4 . •Throughout the simulation, traffic sessions continue to be generated by the set of users. The traffic model generated can be either in accordance with the basic traffic model, as explained in Section 4.2.1.1, where every session that is generated, each kind of users follows a distinct set of actions: –Stationary non-relay UE : A location, either indoors or outdoors, is selected randomly at the beginning of each session and remains consistent throughout the entire session. The stationary UEs have the flexibility to be positioned anywhere within the scenario, without any specific restrictions. –Stationary relay UE : It is a similar case as the previous one, but now some stationary UEs are randomly associated to one of the available relays. –Pedestrian UE : A valid location is randomly selected, and the user initiates movement based on its designated model. The specific criteria for determining valid locations for each user type is more thoroughly explained in Section 4.2.3. On the other hand, the generation traffic model can be based in the measurements taken for the access points, as explained in Section 4.2.1.2. In this model, the generation rate is updated every 15 minutes based on the corresponding measurements and the position of each generated user is determined based on the probability of each AP’s coverage area. The set of users generated using this model are all stationary UEs. Furthermore, to enhance the realism of the simulation, additional pedestrian UEs are generated using the basic traffic model. •After generating the different sessions, UEs with active sessions move in accordance with their respective models, establishing connections with the most suitable base station or relay that covers their current position with the better condition, as explained in Section 4.2.4. It is also computed the amount of PRBs used by each user link depending on the link conditions, as mentioned in Section 4.4. 37 •Upon completion of the simulation, a collection of statistics is stored. To enhance the accuracy and reliability of the results, the simulation is conducted multiple times, each with different configurations of relays and UEs. These multiple realizations enable the aggregation and averaging of the collected statistics, providing a more robust and comprehensive understanding of the simulated scenario. The simulation experiments comprised 50 simulation runs, each lasting 10.000 seconds, for every configured setup and number of relays. The outcomes from these 50 simulations were subsequently averaged to derive the final statistics. 4.7 Simulation parameters All the BS transmitters work at 3.7 GHz over a total bandwidth of 100 MHz with a total transmitted power of 22 dBm. It is considered that beamforming with ideal beam steering is used with an antenna gain of 26 dB for the BS transmitter and 10 dB for the UEs. Only the downlink direction is considered and it is assumed that the interference between cells is insignificant. Despite multiple cells sharing the same channel, an ideal interference coordination is assumed. The active relays during the simulation transmit at a frequency of 3.5 GHz, employing the same bandwidth as the BS transmitters, but with a total transmitted power of -10 dBm along with no antenna gain. The table below provides the summary of the specifications for the generic parameters involved in the simulations: Base Station Transmitted Power 22 dBm Antenna Gain 26 dB Frequency 3.7 GHz Bandwidth 100 MHz Relay UE Transmitted Power - 10 dBm Antenna Gain 0 dB Frequency 3.5 GHz Bandwidth 100 MHz Direct Link UE Antenna Gain 10 dB Noise Figure 9 dB Relay Link Relay UE Antenna Gain 10 dB Table 4: Configured parameters for the simulation. 38 5 Results and discussion In this section, we present the simulations and experiments conducted using the NDT simulator discussed in Section 4. The objective of these simulations is to assess the impact of relays on resource consumption and other related factors. In Section 5.1 different simulations done using the basic traffic model are presented. Despite not accurately representing real-life scenarios, these simulations will still serve as a fundamental means of comprehending the network and assessing various aspects related to performance. After that, in Section 5.2 the same set of simulations are carried out using the generation traffic model based on the measurements of the access points. The objective of these simulations is to generate a more precise depiction of the traffic patterns that will help to create a more realistic environment. Finally, in Section 5.3, a smaller-scale scenario comprising a limited number of buildings is examined, and a series of simulations are conducted. This section aims to draw conclusions on various aspects, including identifying optimal relay locations based on traffic patterns and coverage gaps, assessing the impact on resource consumption, and more. In essence, this section focuses on a ”microscopic” analysis within a smaller area, enabling the detection of elements that might go unnoticed in a larger scenario. The simulations utilize a traffic model generated from access point measurements, providing valuable insights into the specific dynamics of this scaled-down environment. 5.1 Analysis on resource consumption under the basic traffic model Initially, a series of simulations were conducted using the basic traffic model to simulate the traffic. To achieve this, the model sets a fixed number of 2000 users and calculates the distribution percentages for each type of user as indicated in the following table: Total users Stationary 1200 Pedestrians 800 Table 5: Users category under the basic traffic model. The traffic model incorporates key parameters, with a session generation rate (λ) of 1 session per hour per UE. The duration of each session follows an exponential distribution with an average duration (µ) of 300 seconds, as per the predefined settings. Furthermore, the mobility model introduces a 10 % probability of changing directions when the pedestrians are reaching an intersection. On the other hand, when the pedestrians are located in one of the squares, the implementation considers that the pedestrians maintain a consistent direction for a duration determined by an exponential distribution, 39 with an average of 10 seconds. It also considers random direction changes of the pedestrian within a range of +45°to -45°relative to its current direction. Additionally, pedestrians can be located walking the streets and the squares areas at a constant speed of 3 km/h. 5.1.1 Average network load To start with, we conduct a test to calculate the usage of the available PRBs. Figure 14 illustrates the average PRBs usage of the total users achieved with different amounts of relays for each of the 3 BSs deployed. Figure 14: Total PRBs used under the basic traffic model. To highlight, there is a significant difference in the allocation of resources between BS2 and the rest of the base stations. The usage of PRBs in BS2 is significantly higher, approximately 4.5 times greater, compared to the other base stations. This observation is further supported by Figure 15, which illustrates that a majority of the users are connected to BS2. Another notable observation is the decrease in PRB usage as the number of relays increases. For instance, when 250 relays are introduced, the PRB usage in BS1 decreases by 30.5% compared to the scenario without relays. Similar trends can be observed for the other base stations. The results obtained are based on the design decision that a UE will be connected to either a relay or a BS based on the links that offer better spectral efficiency. This design choice contributes to a reduction in the utilization of PRBs, as the UEs are strategically connected to the most efficient links. In addition, an analysis was conducted to determine the distribution of connected users among the different BSs, as illustrated in Figure 15. The results indicate that 76% of the users are connected through BS2, which aligns with the findings in Section 5.1.1. This 40 can be attributed to the better spectral efficiency provided by BS2, as shown in Figure 12, which covers a significant portion of the scenario. The average number of connected users remains stable for the three base stations throughout the simulations, even when increasing the availability of relays. This suggests that the presence of relays does not significantly impact the total number of users. This can be attributed to the fact that there were already very few users located outside the coverage area even without the presence of relays. Figure 15: Total average users under the basic traffic model. Moreover, Figure 16 illustrates the average number of PRBs used per user. 41 Section 5.1. 5.2.1 Average network load The initial experiment to perform using the new generation traffic model involves examining the average usage of PRBs in each of the base stations. Figure 22 illustrates the distribution of PRBs usage as the number of available relays varies. Similar to the findings in Section 5.1.1, BS2 demonstrates a significant allocation of resources compared to the other base stations. Additionally, it is evident that the PRBs usage has increased for each BS when compared to the basic traffic model. Figure 22: Total PRBs used under the measurement-based traffic model. From the figure, it is also evident that with the deployment of 250 relays, there is a reduction in PRBs usage of 32.9%, 26.4%, and 38.8% for BS1, BS2, and BS3, respectively, compared to the scenario with no relays deployed. Furthermore, let’s discuss the distribution of the average number of users connected to each base station throughout the simulations. Figure 23 illustrates that the majority of users are still connected to BS2, accounting for 73% of the total users. This shows a slight decrease compared to the findings in Section 5.1.1. In this configuration, the total number of users in the network has increased for all base stations, particularly with a significant 18% increase in BS1. Although not prominently evident from the graph, the data also indicates that, on average, when 250 relays are deployed, there is an additional user connected to each base station. This represents a user who, in the absence of relays, would have experienced an out-of-coverage situation but is now able to connect to the network due to the presence of relays. 48 Figure 23: Total average users under the measurement-based traffic model. In the previous figures, we observed that both the number of users connected to the network and the utilization of network resources have increased with this configuration. However, it is important to consider the resource usage per user to gain a deeper understanding of the network behavior. Figure 24 presents the average PRBs usage per user. As depicted in the figure, the average resource usage per user has increased for each base station compared to the case with the basic traffic model. This indicates that, on average, users do not experience as favorable conditions as initially anticipated when a more realistic model is implemented. Specifically, the resource usage has increased by 27.2%, 25%, and 20.1% for BS1, BS2, and BS3, respectively. It is also noteworthy that the trend of decreasing average PRBs usage per user is maintained when introducing relays. For example, introducing just 100 relays in the scenario results in a 26.1% reduction in average PRBs usage per user for BS1. 49 Figure 24: Average PRBs used per user under the measurement-based traffic model. Upon initial examination of the network with the new traffic model, it is evident that there has been a significant increase in the utilization of PRBs as well as the average usage per user. To gain further insights, we can analyze the spectral efficiency levels experienced by different users. The table below illustrates the average net spectrum efficiency levels for both the basic traffic model and the access point traffic model, where no relays are deployed: Basic traffic model Measurementbased traffic model BS1 2.42 b/s/Hz 1.91 b/s/Hz BS2 2.63 b/s/Hz 2.14 b/s/Hz BS3 1.97 b/s/Hz 1.52 b/s/Hz Table 6: Comparison of net spectral efficiency levels seen by the users for different traffic models. Upon analyzing the values presented in the table, it is evident that the spectral efficiency levels experienced by users are slightly lower with the new traffic model. This observation can be attributed to the fact that the new distribution generates more users in areas with lower coverage. As a result, there is an increase in the average occupancy of PRBs in the network, as well as the average PRB usage per user. 50 5.2.2 Congestion study Similar to the previous configuration, we conducted a study on the PRB occupation of the network for each link of the base stations. The results are presented in Figure 25, with three separate plots corresponding to each base station. Upon observing the figure, it is evident that the distribution of PRB occupations has not changed significantly. BS2 continues to have the majority of its resources occupied, while the other two base stations still have a considerable amount of unused resources. It is important to note that in the plot for BS2, when no relays are deployed or only a few are present, the occupation exceeds 100%. This indicates critical congestion experienced by BS2 under such conditions. However, by introducing relays to the network, we are able to alleviate this problem and reduce resource usage. 51 Figure 25: Resource occupation for the three base stations under the measurement-based traffic model. 52 Figure 26 illustrates the overall network congestion, incorporating the thresholds described in Section 5.1.2. An additional threshold is introduced for when the network usage exceeds its total capacity. This new traffic generation model can cause congestion levels to surpass 100%, indicating an overloaded network unable to handle the current traffic. In a properly functioning network, congestion should ideally be kept below 100% to maintain efficient data transmission and prevent delays or packet loss. However, when relays are introduced, the probability of critical congestion is likely to quickly decrease towards 0. Despite higher overall congestion levels with this model, the introduction of relays can greatly improve overall network performance by reducing the amount of resources needed, resulting in faster data transmission, reduced latency, fewer packet losses, and an enhanced user experience. This optimization allows for efficient utilization of network resources, ensuring smooth and reliable connectivity for applications and services. Figure 26: Network congestion under the measurement-based traffic model. 5.2.3 Access link occupation In the previous section, we discussed the occupation and the congestion of the links provided by different BSs, including direct links to different UEs and backhaul links to relays. It is important to consider the occupation of the access link that connects a relay to a UE, as it operates independently and utilizes its own resources. Figure 27 depicts the occupation distribution of PRBs on the access links of relays in a simulation with 50 deployed relays. The results indicate that, on average, only 1% of the 53 available capacity on each access link is utilized. The maximum observed occupation is 5.5% of the total capacity. These findings are consistent with the observation that, on average, each relay has a user connected for approximately 15% of the simulation time. Furthermore, it is evident that only a small number of users are connected to a particular relay, with many relays having just one UE connected. Figure 27: Distribution of access link occupation under the measurement-based traffic model. Consequently, we can confidently state that, with this configuration, the access links will not experience congestion issues. However, as done in the previous configuration, it is essential to study the number of users connected to different relays to understand the utility of the relays. Figure 28 illustrates that the number of users connected to relays progressively increases as the number of available relays in the scenario increases. For example, when 200 relays are deployed, an average of 23 users are connected through relays, which accounts for 14.5% of the total users in the network. It is noteworthy that, when compared to the values obtained in Section 5.1.3, the ratio of users connected to relays out of the total users has significantly increased. This can be attributed to the fact that, as mentioned earlier, in a more realistic scenario, the UEs face worse conditions compared to the previous configuration. Therefore, the introduction of relays in the scenario provides a significant opportunity for improving conditions and connectivity to the network through the relays. 54 Figure 28: Distribution of users connected to a relay under the measurement-based traffic model. 5.2.4 Activity factor Another test that we can now conduct with the new configuration is related to the activity factor. The concept of the activity factor was initially introduced in Section 4.2.1.2 and it refers to a parameter that accounts for the ratio of users engaged in transmission compared to the total number of users connected to each AP. In Figure 29, we display the occupation of PRBs for BS1 in a network with 50 relays, while varying the activity factor. 55 Figure 29: PRBs occupation in BS1 with 50 relays varying the activity factor. It can be noted that as the activity factor is increased, more users become active in transmitting data, leading to a higher level of network resources occupancy. This increased activity among the users can result in congestion issues, with the network reaching concerning levels of congestion, particularly when the activity factor is set to 0.20. 5.2.5 Outage probability In the final study, we examine the outage probability using the generation traffic model based on measurements, considering different numbers of relays available in the network. The outage threshold remains set at 1 b/s/Hz to ensure satisfactory communications. As depicted in Figure 30, the introduction of relays significantly reduces the probability of experiencing an outage. For instance, by adding just 50 relays to the network, the probability of being in outage decreases by 19.5%. Comparing these results with the values obtained in the previous configuration, as shown in Figure 21, several observations can be made. Firstly, the outage probability without relays is 66% higher in this model. This can be attributed to the less favorable conditions in the more realistic scenario. Secondly, despite the higher probability, the introduction of relays leads to a rapid decrease in the probability compared to the previous case, resulting in a total reduction of 61.7% when 250 relays are deployed. 56 Figure 30: Outage probability under the measurement-based traffic model. To demonstrate these findings, Figure 31 illustrates the net spectral efficiency map of the ground floor under two scenarios. The first plot corresponds to the absence of deployed relays, while the second plot represents the scenario with 100 available relays. Examining the figure, it is evident that certain buildings, particularly indoors, exhibit a reduction in outage probability as indicated by the white areas in the maps. 57 network resources, maximizing the overall capacity of the system and reducing congestion. Therefore, the deployment of optimally placed relays serves as a key strategy for capacity optimization, enabling better utilization of available resources and enhancing the overall performance of the network. 64 6 Conclusions and future work This project aimed to study and analyze the deployment of a network utilizing UEs as relaying devices. The goal was to achieve an improved RAN performance, including higher capacity, lower outage probability, and overall spectral efficiency. The simulations and experiments conducted using the NDT simulator have provided valuable insights into the impact of relays on resource consumption and other related factors. The simulations using the basic traffic model served as a fundamental means of comprehending the network and assessing various aspects related to performance, while the simulations using the generation traffic model based on access point traffic measurements generated a more precise depiction of traffic patterns in a realistic environment. The smaller-scale scenario analysis in a limited number of buildings allowed for microscopic analysis, revealing elements that might go unnoticed in a larger scenario. The analysis of resource consumption under the basic traffic model showed a significant difference in the allocation of resources between different base stations, with PRB usage in BS2 being significantly higher. The introduction of relays resulted in a decrease in PRB usage for all base stations, highlighting the strategic connection of users to the most efficient links. Moreover, the simulations demonstrated that the introduction of relays resulted in a reduction in link congestion. With the availability of 100 relays, the probability of facing congestion decreased by 60%, indicating the effectiveness of relay deployments in alleviating network congestion. Under the measurement-based traffic model, the average network load increased compared to the basic traffic model, with higher PRB usage in each base station. The majority of users were still connected to BS2, but the total number of users in the network increased, and the resource usage per user also increased. However, the introduction of relays led to a decrease in the average number of PRBs used per user. When only 100 relays were introduced, there was a significant reduction of 26.1% in the average PRBs usage per user for BS1. Overall, the deployment of relays improved the network’s performance by enhancing link conditions. With the deployment of just 50 relays, the network’s resource requirements can be minimized, resulting in a significant reduction of approximately 50% in the probability of critical congestion. Furthermore, the analysis of access link occupation indicated that the utilization of access links connecting the relays to UEs remained low with approximately 14.5% of the total users in the network being connected to a relay when having 200 relays deployed. The maximum observed occupation on the access links was 5.5% of the total capacity, with an average resource utilization of only 1% per access link. This low utilization contributed to maintaining congestion levels at a minimum. Additionally, the simulations included an assessment of the outage probability, which quantifies the chances of the achieved spectral efficiency of a link dropping below a specific threshold. By deploying relays, the outage probability was significantly diminished as it improved the link conditions for users located near the relays. The introduction of 250 relays resulted in a 61.7% reduction in the probability of experiencing an outage. Furthermore, the average net spectral efficiency rose from 2.94 b/s/Hz to 3.32 b/s/Hz when relays were deployed, indicating an overall enhancement of approximately 13% throughout the 65 entire scenario. The analysis of the small-scale scenario highlights the critical importance of strategically deploying relays for capacity and coverage optimization in wireless communication systems. By carefully determining the relay placement, coverage gaps are efficiently minimized, leading to an impressive 71% reduction in outage areas. This optimization translates into improved spectral efficiency and a significantly reduced probability of experiencing outages. The quantified 46% reduction in PRB usage per user showcases the efficient utilization of network resources, leading to enhanced capacity utilization. These findings underscore the significance of relay deployment as a key strategy for optimizing network capacity. By maximizing resource utilization and improving overall network performance, the deployment of optimally placed relays proves to be a valuable approach in achieving capacity optimization objectives in wireless communication systems. These conclusions highlight the benefits of utilizing relays in improving resource consumption, reducing congestion, and enhancing network performance. The findings provide valuable insights for network planning and optimization, aiding in the identification of optimal relay locations based on traffic patterns and coverage gaps. Future research opportunities lie in the continued improvement of the NDT simulator to accurately simulate real-world scenarios across various aspects. One potential direction for enhancement is the exploration of the impact of interference between base stations and relays, as well as the enabling of multihop communication among relays. Another aspect to consider for improvement is to enhance the NDT’s capability to capture real signal level or CQI measurements in different buildings, allowing for the adjustment of coverage maps. Furthermore, there is scope for refining handover criteria and developing more effective relay selection algorithms. These enhancements would contribute to advancing the understanding and optimization of relay deployments in 5G networks. 66 References [1] Xuemin Shen, Jie Gao, Wen Wu, Mushu Li, Conghao Zhou, and Weihua Zhuang. Holistic network virtualization and pervasive network intelligence for 6g. IEEE Communications Surveys Tutorials, 24(1):1–30, 2022. [2] Irene Vil`a, Oriol Sallent, and Jordi P´erez-Romero. On the design of a network digital twin for the radio access network in 5g and beyond. Sensors, 23(3):1197, 2023. [3] Christian Hoymann, Wanshi Chen, Juan Montojo, Alexander Golitschek, Chrysostomos Koutsimanis, and Xiaodong Shen. Relaying operation in 3gpp lte: challenges and solutions. IEEE Communications Magazine, 50(2):156–162, 2012. [4] Letian Rong, Salah-Eddine Elayoubi, and Olfa Ben Haddada. Impact of relays on lteadvanced performance. In 2010 IEEE International Conference on Communications, pages 1–6. IEEE, 2010. [5] Zhe Ren, Abdallah Bou Saleh, ¨ Omer Bulakci, Simone Redana, Bernhard Raaf, and Jyri H¨am¨al¨ainen. Joint interference coordination and relay cell expansion in lteadvanced networks. In 2012 IEEE Wireless Communications and Networking Conference (WCNC), pages 2874–2878. IEEE, 2012. [6] 4g lte relay: Lte advanced relaying. https://www.electronics-notes.com/ articles/connectivity/4g-lte-long-term-evolution/lte-relaying.php. [7] Oumer Teyeb, Ajmal Muhammad, Gunnar Mildh, Erik Dahlman, Filip Barac, and Behrooz Makki. Integrated access backhauled networks. In 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall), pages 1–5. IEEE, 2019. [8] Jiun Y Lai, Wu-Hsiu Wu, and Yu T Su. Resource allocation and node placement in multi-hop heterogeneous integrated-access-and-backhaul networks. IEEE Access, 8:122937–122958, 2020. [9] Felipe G´omez-Cuba and Michele Zorzi. Twice simulated annealing resource allocation for mmwave multi-hop networks with interference. In ICC 2020-2020 IEEE International Conference on Communications (ICC), pages 1–7. IEEE, 2020. [10] 3GPP. 5G;NG-RAN;Architecture description . Technical Specification (TS) 38.401, 3rd Generation Partnership Project (3GPP). Version 15.2.0. [11] Charitha Madapatha, Behrooz Makki, Chao Fang, Oumer Teyeb, Erik Dahlman, Mohamed-Slim Alouini, and Tommy Svensson. On integrated access and backhaul networks: Current status and potentials. IEEE Open Journal of the Communications Society, 1:1374–1389, 2020. [12] Mohsen Nader Tehrani, Murat Uysal, and Halim Yanikomeroglu. Device-to-device communication in 5g cellular networks: challenges, solutions, and future directions. IEEE Communications Magazine, 52(5):86–92, 2014. 67 [13] Jordi P´erez-Romero and Oriol Sallent. Leveraging user equipment for radio access network augmentation. In 2021 IEEE Conference on Standards for Communications and Networking (CSCN), pages 83–87. IEEE, 2021. [14] Arash Asadi, Qing Wang, and Vincenzo Mancuso. A survey on device-to-device communication in cellular networks. IEEE Communications Surveys & Tutorials, 16(4):1801–1819, 2014. [15] 3GPP. 5G;Service requirements for the 5G system. Technical Specification (TS) 22.261, 3rd Generation Partnership Project (3GPP). Version 18.6.0. [16] Jordi P´erez-Romero, Oriol Sallent, and Olga Ruiz. On relay user equipment activation in beyond 5g radio access networks. In 2022 IEEE 96th Vehicular Technology Conference (VTC2022-Fall), pages 1–6. IEEE, 2022. [17] Rui Dong, Changyang She, Wibowo Hardjawana, Yonghui Li, and Branka Vucetic. Deep learning for hybrid 5g services in mobile edge computing systems: Learn from a digital twin. IEEE Transactions on Wireless Communications, 18(10):4692–4707, 2019. [18] Yueyue Dai, Ke Zhang, Sabita Maharjan, and Yan Zhang. Deep reinforcement learning for stochastic computation offloading in digital twin networks. IEEE Transactions on Industrial Informatics, 17(7):4968–4977, 2020. [19] Huan X Nguyen, Ramona Trestian, Duc To, and Mallik Tatipamula. Digital twin for 5g and beyond. IEEE Communications Magazine, 59(2):10–15, 2021. [20] Hanshu Hong, Qin Wu, Feng Dong, Wei Song, Ronghua Sun, Tao Han, Cheng Zhou, and Hongwei Yang. Netgraph: An intelligent operated digital twin platform for data center networks. In Proceedings of the ACM SIGCOMM 2021 Workshop on NetworkApplication Integration, pages 26–32, 2021. [21] Cheng Zhou, Hongwei Yang, Xiaodong Duan, Diego Lopez, Antonio Pastor, Qin Wu, Mohamed Boucadair, and Christian Jacquenet. Digital twin network: Concepts and reference architecture. Internet Engineering Task Force, Internet-Draft draft-zhounmrg-digitaltwin-network-concepts-06, 2021. [22] Mohammad Asif Habibi, Meysam Nasimi, Bin Han, and Hans D Schotten. A comprehensive survey of ran architectures toward 5g mobile communication system. IEEE Access, 7:70371–70421, 2019. [23] Linbo Hui, Mowei Wang, Liang Zhang, Lu Lu, and Yong Cui. Digital twin for networking: A data-driven performance modeling perspective. IEEE Network, 2022. [24] Hamed Ahmadi, Avishek Nag, Zaheer Khar, Kamran Sayrafian, and Susanto Rahardja. Networked twins and twins of networks: An overview on the relationship between digital twins and 6g. IEEE Communications Standards Magazine, 5(4):154– 160, 2021. [25] Juan Deng, Qingbi Zheng, Guangyi Liu, Jielin Bai, Kaicong Tian, Changhao Sun, Yujie Yan, and Yitong Liu. A digital twin approach for self-optimization of mo68 bile networks. In 2021 IEEE Wireless Communications and Networking Conference Workshops (WCNCW), pages 1–6. IEEE, 2021. [26] Jingjing Wang, Chunxiao Jiang, Haijun Zhang, Yong Ren, Kwang-Cheng Chen, and Lajos Hanzo. Thirty years of machine learning: The road to pareto-optimal wireless networks. IEEE Communications Surveys & Tutorials, 22(3):1472–1514, 2020. [27] 3GPP. Study on channel model for frequencies from 0.5 to 100 GHz. Technical Report (TR) 38.901, 3rd Generation Partnership Project (3GPP). Version 16.1.0. [28] Siemens AG. TDD UE-UE Interference Simulations. R4-030189 document of the 3GPP TSG-RAN Working Group 4 meeting 26,, 2003. [29] 3GPP. Physical layer procedures for data . Technical Specification (TS) 38.214, 3rd Generation Partnership Project (3GPP). Version 16.2.0. [30] 3GPP. 5G;NR;User Equipment (UE) radio access capabilities . Technical Specification (TS) 38.306, 3rd Generation Partnership Project (3GPP). Version 17.2.0. [31] Configuring a downlink signal under tdd frame structure. https://rfmw. em.keysight.com/wireless/helpfiles/n7631/Content/Main/Config%20DL% 20Under%20TDD.htm. [32] 3GPP. Base Station (BS) radio transmission and reception. Technical Specification (TS) 38.104, 3rd Generation Partnership Project (3GPP). Version 16.4.0. 69