Toward the integration and convergence between 5G and TSN technologies and architectures for industrial communications: A survey

IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 259 Toward the Integration and Convergence Between 5G and TSN Technologies and Architectures for Industrial Communications: A Survey Jorge Sasiain , David Franco , Asier Atutxa , Jasone Astorga , and Eduardo Jacob ,Senior Member, IEEE Abstract—Communication networks are in constant evolution to adapt to the ever-changing demands of modern services and applications. On the one hand, technologies within the framework of the fifth generation of communication networks (5G), currently in a steady course towards 6G, are being designed with unprecedented performance and flexibility capabilities to empower the digital transformation of heterogeneous vertical industries. On the other hand, Time-Sensitive Networking (TSN) is being developed to enable converged network infrastructures able to satisfy the most stringent communication requirements in regards to bounded low latency and reliability. In verticals such as manufacturing, a joint usage of 5G and TSN is very promising to enable revolutionary use cases that coexist with the constraints imposed by industrial communications and systems. In this paper, we exhaustively survey the technology landscape concerning the integration between 5G and TSN technologies, including the perspectives of both industry and academia. In the first part of this paper, we provide a comprehensive review of relevant standards and outcomes from industry initiatives. We first examine the capabilities provided by TSN and their application to industrial communications. Likewise, we contextualize the 5G architecture and its key enabling technologies, and later provide a comprehensive review of the architecture and mechanisms to enable the integration between TSN and 5G. In the second part of this paper, we extensively survey the state of the art from academic literature concerning proposals that contribute to enabling, improving, and/or demonstrating different aspects of the integration or interworking between 5G and TSN technologies. Index Terms—Time-sensitive networking (TSN), 5G, 5G-TSN, time-sensitive communications, time synchronization, virtualization, software-defined networking (SDN), management and orchestration, Industry 4.0. Manuscript received 12 July 2023; revised 16 January 2024 and 15 May 2024; accepted 21 June 2024. Date of publication 3 July 2024; date of current version 17 February 2025. This work was supported in part by the National Project funded by MICIU/AEI/ 10.13039/501100011033 titled “Enabling Native-AI Secure Deterministic 6G Networks for Hyper-Connected Environments” (6G-INSPIRE) under Grant PID2022-137329OB-C44, and in part by the Basque Government through the Project “Social Network for Machines” (SONETO) of the ELKARTEK Program under Grant KK-2023/00038. (Corresponding author: Jorge Sasiain.) The authors are with the Department of Communications Engineering, Faculty of Engineering, University of the Basque Country UPV/EHU, 48013 Bilbao, Spain (e-mail: [email protected]; da[email protected]; asier[email protected]; [email protected]; [email protected]). Digital Object Identifier 10.1109/COMST.2024.3422613 I. INTRODUCTION A. Background and Motivation THE FOURTH industrial revolution, known as Industry 4.0, [1] encompasses the digital transformation of the manufacturing sector. It drives the evolution towards fully interconnected scenarios in the manufacturing industry between the physical and virtual worlds to achieve greater automation and smartization of the industrial processes. This trend received the term of Smart Factories; a vision of smart production systems and production lines interconnected by wired or wireless means, and characterized by flexibility, reconfigurabily, cost-efficiency, and agility [2].Itisalso commonly referred to as Smart Manufacturing [3]. In this context, Industry 4.0 embraces a series of technologies and paradigms primarily focused around ubiquitous communication scenarios driven by Internet of Things (IoT) and Industrial Internet of Things (IIoT), and around the merging of physical and virtual assets through concepts like Cyber-Physical Systems (CPS) and digital twins [4]. Further, Industry 4.0 systems can be empowered by advanced technologies such as Artificial Intelligence (AI), Machine Learning (ML), Big Data analytics, and Augmented Reality and Virtual Reality (AR/VR) [5]. Altogether, the Industry 4.0 revolution is driving the evolution of industrial automation in the manufacturing sector [6]. However, this technological ecosystem has to coexist with the real-time and deterministic latency constraints imposed by the use cases within the industrial automation field including closed-loop control systems for process and factory automation, connected robotics, motion control of machines and robots, remote operation, etc. This landscape implies complex communication scenarios with stringent yet heterogeneous requirements that demand a highly flexible, adaptable, and efficient network substrate. Traditionally, the industry has relied on proprietary fieldbus protocols, such as PROFINET, Ethernet/IP, Modbus, and EtherCAT, to achieve the required levels of determinism for time-sensitive communications. However, the lack of interoperability and the closed nature of these solutions make them very limited in their ability to support modern Industry 4.0 applications [7]. To bridge this gap, the past decade has seen the emergence of Time-Sensitive Networking (TSN), which encompasses a set of open standards, continuously being rolled out as part of IEEE 802.1, to enable deterministic c 2024 The Authors. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License. For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ 260 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 communications over Ethernet. Overall, TSN has achieved very significant progress towards a technology that is capable of converging multiple types of traffic over a shared infrastructure whilst providing performance guarantees to the most critical time-sensitive applications. Furthermore, the open and interoperable nature of TSN, along with its versatile configuration capabilities, greatly facilitates its coexistence with external networks and its integration with other technologies. This enablement of a converged network infrastructure is key for meeting the ubiquitous connectivity requirements demanded by modern Industry 4.0 applications [8]. In parallel, the fifth generation of communication networks (5G) has emerged with the aim of satisfying a highly demanding yet highly heterogeneous set of requirements, which include ultra-high bandwidth (enhanced Mobile Broadband; eMBB), ultra-low latency and reliability (Ultra-Reliable and Low-Latency Communication; URLLC), and ultra-high connection density (massive Machine-Type Communications; mMTC). A key difference with previous generations is that the target is not almost exclusively human users, but also vertical industries. In order to be able to tailor 5G networks to this wide range of vertical use cases, 5G networks are being designed to support an unprecedented level of flexibility. As such, many technologies that have emerged in recent years have become key 5G building blocks [9], including Network Functions Virtualization (NFV), Software Defined Networking (SDN), network slicing, and Multi-access Edge Computing (MEC). These technologies represent the trend towards the virtualization, softwarization, and programmability of networks and services, paving the path towards a networking and computing infrastructure that can be specifically tailored to meet the varying requirements and demands of each individual vertical. In order to truly maximize the potential of Industry 4.0 and beyond systems, both 5G technology—including its future evolutions—and TSN technology are expected to play a key role. In this way, 5G is not a replacement for TSN; instead, they are complementary technologies. On the one hand, 5G enables wireless scenarios that are not possible over a wired Ethernet infrastructure [10], whether it is to enable more costefficient and flexible deployments and reduce maintenance costs, or to integrate wireless equipment into the production lines, such as mobile robots, Automated Guided Vehicles (AGVs), and wireless field devices. The URLLC capabilities developed in 5G have opened up the possibility of realizing these wireless communication scenarios maintaining an unprecedented level of latency and reliability guarantees [11]. On the other hand, TSN’s pursuit of network convergence is fundamentally aligned with 5G architectures; virtualization and edge computing paradigms enable the coexistence of mixed-criticality applications in a shared network and compute infrastructure, whereas novel network orchestration frameworks [12] present a solution to the ever-increasing complexity of managing and operating such heterogeneous networks. The interest in 5G and TSN by both industry and academia is evident, having propelled extensive standardization and research work alike. Besides, given the foreseen potential of enabling deterministic low-latency wireless communications for verticals such as industrial automation, the integration and confluence of these two technologies is also receiving considerable attention. Notably, the 3rd Generation Partnership Project (3GPP), the partnership of Standards Development Organizations (SDOs) currently responsible for the standardization of 5G, has been standardizing the architecture and mechanisms to enable the integration with TSN in its latest releases. Together with these standardization efforts, endeavors carried out by several organizations and industrial alliances, and extensive research work by academia, have demonstrated significant breakthroughs in recent years, a trend that is expected to continue in response to the requirements of the continuously evolving vertical industries. B. Paper Organization The remainder of this paper is structured as follows: Section II categorizes and examines the state of the art regarding surveys and tutorials in the scope of integration between 5G and TSN technologies, and outlines the contributions of this paper. Section III reviews TSN technology according to the main IEEE standards, and contextualizes TSN in the industrial automation and manufacturing vertical. Section IV reviews the architecture, capabilities, and key technology pillars of 5G networks, emphasizing on the enabling technologies and standardization efforts to make possible deterministic communications compliant with TSN over 5G networks. Section Vprovides a comprehensive state- of-the-art survey of 5G and TSN integration, analyzing and discussing literature that corresponds to the integration of 5G and TSN networks and to the interworking between 5G technology pillars and TSN. Section VI presents the most significant research challenges and directions regarding the topics covered in this survey. Finally, a brief conclusion to the paper is provided in Section VII. The structure of core sections of this paper is represented in Figure 1. A list with the main acronyms used in this paper is provided in Table I. II. RELATED WORK AND CONTRIBUTIONS Both TSN and especially 5G have been thoroughly investigated by the scientific community. Because of this, several research works exist that comprehensively survey either of the two technologies, such as [13],[14], and [15]. Among the related survey and/or tutorial works found in the literature involving these topics, Table II reflects the coverage of those that are deemed to have the most overlap with our work. As per the scope of this paper, the list of research papers included in Table II has been narrowed down to the ones that include interworking aspects between TSN and 5G, and/or between TSN and 5G-related technologies. Table II hints the depth at which various high-level scopes are covered by the research papers, and whether they contribute with a review or tutorial of technologies and standards, and/or with a survey of research literature. It should be noted that the primary scope of some of the included research papers may be outside of the extent of Table II and thus not indicated, and that some papers jointly carry a theoretical review and some state-of-the- art proposal, meaning that their contributions may be discussed in later sections of this paper instead of, or, in addition to, SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 261 Fig. 1. Organization of the core sections of this paper. Orange: review of technologies and standards; green: survey of academic literature; blue: perspectives. appearing here. The remainder of this section expands on the contributions of these surveys, and outlines the contributions of our work in comparison to said state of the art and the gaps encountered. A. Survey and Tutorial Works Related to the Integration between 5G and TSN Networks A number of research papers address the integration between TSN and 5G networks to different degrees. Some works such as [19],[21],[23],[24], and [28] specifically assess the evolution of 5G standards to support time-sensitive communications. Among the first papers surveying this topic, [19] and [21], describe initial standardization activities by 3GPP to support TSN services in 5G networks. References [23] and [24] also analyze the 3GPP standards in this context, in this case in a moderately more mature state. On the other hand, [28] comprehensively reviews 5G URLLC features introduced in the standards, and analyzes how they can contribute to support low latency and high reliability in industrial automation. Continuing this trend, additional existing papers describe different aspects of the architecture-level integration between 5G and TSN, possibly focusing on specific aspects such as latency, synchronization accuracy, or converged configuration. For example, [22] provides a preliminary analysis on the usage and performance of the procedures and QoS mechanisms available in 5G for the support of time-sensitive communications. In this line, [34] focuses on the configuration aspects of TSN streams, likewise providing an overview of the 5G QoS mechanisms, and analyzing their mapping to TSN traffic classes. References [36] and [42] provide a concise overview of the system integration between 5G and TSN. In regards to [42], this paper also presents a brief survey of implementations of 5G-TSN integration in the literature. References [26] and [27] provide literature review addressing future research directions of 5G and TSN, but do not investigate the integration of the two technologies. Overall, comprehensive coverage of literature contributions on this topic appears to be largely missing from the available survey papers. Notably, however, [47] illustrates an exhaustive categorization of research works in the scope of 5G and TSN integration with the aim of quantifying research trends, but does not delve into their contributions. Reference [45] analyzes deterministic communications with an eye on the upcoming 6G networks. This paper envisions 262 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE I LIST OF ACRONYMS the role of several technologies in the evolution towards 6G, including TSN, DetNet, URLLC, edge computing, and also covering the application layer and end-to-end security. Insightful takeaways can be found in this article, but due to its more generalist nature, detailed TSN integration aspects are sparse. Other works such as [25],[31],[33],[38], and [48] SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 263 TABLE II RELATED SURVEY AND TUTORIAL RESEARCH WORKS R: REVIEW OR TUTORIAL OF TECHNOLOGIES AND STANDARDS;S:SURVEY OF ACADEMIC LITERATURE;BOLD NOTATION:SIGNIFICANTLY COVERED also provide generalist views on the matter, contextualizing the topic of deterministic communications in the scope of private 5G deployments for the industry vertical. Although discussed, the integration between TSN and 5G is not the core contribution of these papers, so it is not covered in extensive detail. The enablement of end-to-end time synchronization to support TSN over 5G networks is an inherent requirement of their integration, so many works, including some of the aforementioned ones, address time synchronization support together with other architecture and configuration aspects. There are, however, several articles that specifically concentrate on reviewing this topic: [16],[23],[24],[29],[35],[40], and [46].[16] and [29] evaluate the challenges and requirements of clock synchronization accuracy for wireless TSN, and discuss how novel advances and capabilities in two wireless technologies—Wi-Fi 802.11AX, and 5G—contribute to the accuracy of the synchronization achieved. The other papers target exclusively 5G systems. References [23] and [24] provide a detailed perspective from 3GPP standards in regards to timesynchronizationtosupporttime-sensitivecommunications. References [35] and [40] focus on surveying mechanisms to enhance Propagation Delay Compensation (PDC), which can be decisive to the accuracy of the time synchronization distributed through the wireless 5G network. Finally, [46] discusses recent advances in 3GPP standards in regards to time synchronization resiliency in a more generalist manner, and looks ahead for the challenges to be solved in 6G. Another scope of the integration between 5G and TSN technologies corresponds to the use of TSN within the 5G Transport Network (TN). The following articles are considered to review this topic to different degrees, often without being the main contribution of the paper: [11],[19],[20],[33],[36], and [50]. Reference [20] does focus exclusively on this matter, briefly studying the applicability of different TSN standards and features to transport fronthaul traffic with deterministic 264 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 requirements in 5G. Additionally, a broader view on the matter from an standards perspective is one of the contributions of [11]. Finally, [50] also concentrates on this topic, as it presents an interesting perspective on the opportunities and challenges of integrating TSN into the O-RAN architecture. Overall, however, fewer works with a survey or tutorial nature can be found in the literature covering the use of TSN in the 5G TN, in comparison with those that analyze the use of 5G to extend a TSN network, corresponding to the previous paragraphs of this subsection. This is especially true concerning research papers published from 2020 onwards. B. Survey and Tutorial Works Related to the Integration of TSN With Key 5G Enabling Technologies On the contrary, despite the increasing recognition of the benefits of enabling time-sensitive communications within the framework of technologies such as SDN and virtualization, purely survey or tutorial works concerning the landscape and state of the art of the use of these technologies together with TSN appear to be highly scarce. TSN together with SDN is addressed by [17],[18],[27], and [32]. References [17] and [27] study the applicability of TSN and SDN for networks in Industry 4.0 or in vertical industries, but discuss the contributions of each of the two technologies independently from each other, as opposed to their integration. The primary contributions in this regard are probably [18] and [32]. The former reviews challenges and research directions of software-defined time-sensitive networks to be applied to the operation and management of CPSs. Reference [32] also sketches a preliminary architecture, so it is covered in a later section of this paper. In addition, [11] presents a brief literature review surveying the adoption of the SDN paradigm for TSN. However, overall, the coverage of state of the art concerning the usage of TSN and SDN together is limited and not completely up-to-date. Finally, a related paper worth pointing out is [51], which studies time synchronization mechanisms that can be borrowed for the coordinated management of SDN networks. In regards to integrating TSN and virtualization, [43] is centered on reviewing the available capabilities to enable a cloudification of specific TSN functionality to enhance reliability. Reference [52] provides a technology and literature review towards real-time container-based virtualization. It does not directly address TSN technology, so it has been left out of Table II. Even though [49]’s primary contribution is to present experimentation results, it briefly reviews challenges and existing mechanisms for TSN support in container-based applications. Overall, no research work with a survey nature has been found to widely investigate approaches to combine TSN and virtualization, even though the topic is gaining remarkable traction in the literature in order to support the requirements of new technological trends such as virtualized controller functionality in CPSs and in industrial automation networks, and digital twins. Reference [39] addresses the progress towards zero-touch cognitive management of industrial networks. The use of TSN is considered only very briefly and not directly addressed. References [19] and [33] describe analytics exposure to applications in 5G networks, and zero-touch practices for the management of private industrial 5G networks, respectively. Both of these papers have a generalist nature and also introduce the support of TSN in 5G networks. However, these topics are only presented as independent capabilities of 5G networks rather than delving into a more holistic perspective. Overall, it has been found that the applicability of novel intelligent, zero-touch, and analytics-powered orchestration approaches and technologies from the framework of 5G to the management of complex 5G-TSN networks is barely present in survey contributions of the literature, even though it could be a key matter in the evolution towards 6G. C. Additional Remarks Among the research papers in Table II,[11],[19],[20], [26],[29],[32],[35],[36],[40],[43],[45], and [47] do not focus on any specific vertical use case. Meanwhile, the rest target the industrial automation and manufacturing vertical like our work does. Still, while the targeted environment may have implications, the technology and architecture aspects are largely application scenario-agnostic. It should also be noted that due to extension reasons, given the broad and overarching scope of the topics at hand, the following related aspects are narrowly covered, or not covered, by this paper. Example articles that primarily cover such topics are included in the list below for the reader’s reference. •Taxonomy of applications and architectures in Industry 4.0 and in industrial networks [4],[53],[54]. •Modeling and evaluation of TSN mechanisms and algorithms where 5G integration does not intervene [55],[56], [57],[58],[59]. •Security aspects of TSN and of 5G-TSN integration [60], [61]. •Integration of TSN with wireless access network technologies other than 5G [62],[63],[64]. •Protocol and architecture aspects that specifically target use cases not related to industrial automation in manufacturing [65],[66]. •Private 5G network architectures for industrial environments [67] [33]. •Deterministic Networking (DetNet) [68],[69]. D. Contributions of This Survey The outline of the contributions of this paper is described as follows: •An in-depth review, from the perspective of SDOs and industrial organizations, of the technologies and network architectures that are shaping a new generation of innovative industrial applications, focusing on TSN standards, on the 5G technology ecosystem, and on the enabling technologies for their integration. •A comprehensive state-of-the-art survey of academic papers on the interworking between TSN and key pillars of 5G networks over the period between 2016 and September 2023. This includes contributions in the matters of SDN and network programmability, virtualization SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 265 technologies and frameworks, edge computing architectures, end-to-end orchestration, zero-touch automation, and network capability exposure. •A comprehensive state-of-the-art survey of academic papers on the integration between 5G networks and TSN to support end-to-end time-sensitive communications over wired and wireless infrastructures, spanning the period between 2018 and September 2023. This includes contributions in the areas of: performance evaluation and optimization of 5G-TSN integrated architectures, enablement of time synchronization over 5G, converged configuration of 5G and TSN domains, and TSN for the 5G TN. •An outline of the overall vision and the challenges ahead of end-to-end communications that integrate 5G and TSN technologies, based on synthesized insights and lessons learned derived from the literature survey. To the best of our knowledge, according to the related work explored in this section, this is the first work providing a holistic comprehensive review of the integration and convergence of TSN with the 5G architecture as well as with a wide spectrum of enabling technologies in the 5G ecosystem. Moreover, we consider that no existing work surveys the state of the art of such topics in an exhaustive manner and from a TSN technology-centric perspective, despite the spike in interest and in the number of research works undergone by the scientific community since the latter part of the last decade. III. TIME-SENSITIVE NETWORKING AND ITS ROLE IN INDUSTRIAL COMMUNICATIONS This section provides a comprehensive review of the TSN features from a standards perspective from Sections III-A–III-E. The remaining Sections III-F–III-H introduce the basic architecture of industrial networks and the communication patterns involved in such networks. Key contributions in the matter from SDOs and other entities are examined in order to analyze how TSN technology can be implemented and configured to satisfy the heterogeneous communication requirements present in these scenarios. A. TSNOverview TSN standards are developed by the Time-Sensitive Networking Task Group, which is one of the four subgroups of the IEEE 802.1 Working Group [70]. TSN has its origins in 2004, when the IEEE 802.3 Residential Ethernet Study Group was established. This Study Group was a year later moved to the IEEE 802.1 Audio Video Bridging (AVB) Task Group, before being renamed in 2012 to the current Time-Sensitive Networking Task Group to better reflect an scope that goes beyond the originally conceived audio and video transmission by including control applications in industry and automotive areas [71]. TSN communication occurs between end stations (Talkers and Listeners) across a layer 2 network of TSN-capable devices, generally referred to as TSN bridges. Each communication channel between a Talker and one or more Listeners is known as a TSN stream or TSN flow, which is associated to a given TSN traffic class characterized by a set of features (periodicity, payload size, etc.) and Quality of Service (QoS) properties (latency, bandwidth, tolerance to loss, etc.). Traffic classes are assigned to queues according to the Priority Code Point (PCP) in the VLAN tag within the Ethernet header. As the PCP field has a length of 3 bits in the VLAN header, this enables the distinction of up to 8 different traffic classes or priorities. The mere use of PCP to prioritize the queuing of frames is referred to as strict priority. However, TSN standards define additional mechanisms for the treatment of Ethernet traffic. In many cases, a TSN stream is implemented through the assignment of a periodic time window in which the traffic belonging to said stream is allowed to be transmitted across the TSN networking devices. The end stations report their stream requirements to the TSN control plane, which takes care of computing the global schedule that accommodates all streams, and of configuring the bridges involved in the communication path accordingly. TSN standards can be primarily grouped in four categories according to the features they provide: timing and synchronization, bounded low latency, high reliability, and resource management. Timing and synchronization standards provide the means to establish a common source of time between all entities involved in a TSN communication, in order to pave a way to configure consistent traffic scheduling policies. Bounded low latency standards provide traffic shaping mechanisms, often leveraging time synchronization, that allow to apply a different treatment to different traffic streams according to their characteristics and requirements. This ensures that the delivery of critical frames is carried within a bounded timespan. Besides, preemption mechanisms included in this category of standards can further reduce interference. High reliability standards include path redundancy mechanisms to mitigate the effects of network failures or suboptimal network conditions, and policy enforcement mechanisms to protect the network from non-compliant traffic. This group of standards target high availability, ultra-reliability, and frame loss avoidance for critical traffic. Finally, resource management standards deal with the mechanisms and interfaces that make it possible to allocate TSN resources to traffic streams and push the necessary configuration onto the involved entities. TSN has standardized both distributed and centralized models. As TSN provides a toolbox of standards, the usage of each of these individual features in a specific application scenario can be selected based on its characteristics and its requirements [72]. Among the base set of TSN standards, IEEE Std 802.1Q-2022 [73] standardizes multiple aspects regarding bridge forwarding, scheduling, and configuration. Previous revisions of 802.1Q, such as IEEE Std 802.1Q-2018 [74] and IEEE Std 802.1Q-2014 [75], have been progressively rolled into it. IEEE Std 802.1Q-2022 has had some amendments corresponding to enhancements to specific features since its release. Other TSN base standards include, but are not limited to, 802.1AS in regards to time synchronization, and 802.1CB providing frame replication mechanisms. At large, TSN defines three main traffic classes in the 802.1Q standard. Best Effort (BE) is used on low-priority traffic that does not require any form of traffic shaping and has 266 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE III CLASSIFICATION OF MAIN TSN STANDARDS no time-based constraints. Next, Stream Reservation (SR) is used on traffic with intermediate priority. SR was introduced by AVB standards, and is enabled by credit-based shaping mechanisms. SR subclasses are also defined as SR Class A, SR Class B, etc., from higher to lower priority. However, even though traffic shaping based on credit can be adequate for audio and video traffic, TSN usage in industrial control and automotive scenarios require additional mechanisms to satisfy the requirements of periodic applications that demand traffic to be transmitted according to a time schedule [76]. As such, as time-based shaping and time synchronization mechanisms were standardized in TSN, a new traffic class with the highest priority, Scheduled Traffic (ST), was also introduced. ST provides the least interference and the greatest temporal guarantees in the delivery of frames, as separate streams in the ST class can be assigned isolated time windows for transmission. Subsequent subsections cover the most significant IEEE 802 TSN standards and their features classified into the four aforementioned categories. Table III classifies the main standards in order of appearance in this section. The reader interested in learning more about a specific standard is referred to the referenced TSN standard documents published by IEEE. B. Timing and Synchronization Standards Time synchronization is crucial for enabling deterministic communications and hence is a prerequisite to the TSN mechanisms that enable time-sensitive traffic shaping. TSN time synchronization is based on the IEEE 1588 Precision Time Protocol (PTP) standard, which is a protocol for providing precise clock synchronization in the sub-microsecond range, primarily targeting Ethernet-based LANs. The original PTP standard dates from 2002, with the most recent revision, IEEE 1588-2019, often referred to as PTPv2.1, being released in 2019 [77]. A PTP domain is composed of several interconnected PTP- enabled nodes, forming a PTP network. Each of these nodes is known as a PTP clock. PTP clocks are characterized by having one or more PTP-enabled ports through which the time synchronization information is propagated. Each individual port of a PTP clock can be in three possible states at any given time, which are Master Port,Slave Port, and Passive Port. A Passive Port is simply a port that is not PTP-enabled. On the other hand, time synchronization information consists on timestamped PTP messages that are transmitted between the Master Port of a clock and the Slave Port of another clock, in a procedure that allows the latter clock to become synchronized to the former. The IEEE 1588 standard defines several types of clocks according to the functionality they implement and to the ports they contain. At the top of the hierarchy of a PTP domain, aGrand Master (GM) clock is elected, either manually, or through negotiation by executing the Best Master Clock (BMC) algorithm. The BMC algorithm compares the accuracy of two given clock sources to determine the most accurate one according to several attributes. The BMC algorithm determines the GM this way, and then constructs a time-synchronization spanning tree. The GM clock acts as the root of the tree and provides the primary time source in the PTP domain, periodically initiating the synchronization procedure towards the downstream clocks through one or more Master Ports. Devices with one PTP-enabled port in a domain are known as Ordinary Clocks, and those with multiple PTP-enabled SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 267 ports in a domain are known as Boundary Clocks. The single Slave Port of an Ordinary Clock allows it to synchronize its clock from an upstream clock. Meanwhile, a Boundary Clock has one Slave Port connected to an upstream Master Clock from which it receives the clock reference, and one or more Master Ports through which it provides its clock reference to downstream clocks. Finally, a Transparent Clock corresponds to a device that neither provides a clock source nor updates their own, but is still capable of relaying PTP messages while applying a correction to their time reference (i.e., updating the correctionField field) to compensate for the transit time in the device. The PTP standard supports customization through the definition of profiles. The TSN standard for time synchronization is IEEE 802.1AS, known as generalized Precision Time Protocol (gPTP), which is precisely a profile of the PTP standard with adaptations and simplifications for its use in time-sensitive applications over IEEE 802 layer 2 networks. The latest revision of the 802.1AS standard is 802.1AS- 2020 [78], published in 2020. 802.1AS standardises a PTP Relay Instance, corresponding to a TSN bridge acting as a Boundary Clock, and a PTP End Instance, corresponding to a TSN end station acting as an Ordinary Clock. According to 802.1AS, a group of interconnected gPTP-capable devices forms a time-aware network. These devices are referred to as time-aware systems. A time-aware system can support multiple PTP instances and thus can be part of multiple gPTP synchronization domains. The 802.1AS standard provides additional robustness, demanding each node to support gPTP in the MAC level. 802.1AS also introduces reliability improvements, such as the ability to have multiple gPTP time domains to allow nodes to be synchronized to multiple GM clocks, as well as the possibility to make multiple redundant connections to these GM clocks. For further insight into the differences between PTP and gPTP, the reader is referred to Clause 7.5 of 802.1AS. Time synchronization in a gPTP domain is achieved through the execution of two separate messaging procedures between each pair of PTP instances separated by one hop. The first procedure consists on measuring the link propagation delay between two instances, and is performed by exchanging timestamped messages as depicted in the example of Figure 2.The initiator of the exchange eventually collects four timestamps that enable it to calculate the link delay as shown in the figure. If the frequencies of the clocks of each instance differ, an additional RateRatio factor is taken into account, calculated as the ratio between the clock frequency of the initiator relative to the clock frequency of the responder. It can be quantified by executing the procedure multiple times. Once the propagation delay of each link has been calculated, clock synchronization can take place. Sync—and Follow_Up if operating in two-step mode—messages periodically sent by the GM clock are received on Slave Ports and forwarded via Master Ports as depicted in Figure 3. Before a device forwards synchronization messages to the next device this way, it updates its CorrectionField field by adding to it the calculated propagation delay of the link as well as the residence time in the device. The residence time is calculated as the difference Fig. 2. Example of calculation of RateRatio and link delay between two clocks. The values are calculated by the initiator and the delay value is determined relative to its clock frequency. Symmetric link and constant link delay is assumed. between a timestamp at the egress of the device and a timestamp at the ingress. The device also recalculates the message’s CumulativeRateRatio field, which is the product of RateRatio factors during the transit across all devices. C. High Reliability Standards The two primary standards concerning the provision of reliability guarantees in TSN are 802.1CB and 802.1Qci. The 802.1CB-2017 standard [79] defines the Frame Replication for Elimination and Reliability (FRER) protocol. FRER is intended to provide redundancy to critical traffic by providing mechanisms to replicate the frames of a stream into multiple member streams that can be sent through paths with the least possible overlap. Precisely, the execution of an optimal path selection strategy to transmit the streams is key to maximizing the effectiveness of FRER when fully disjoint paths are not available [57]. FRER also ensures that the redundant duplicate frames are eliminated in the destination or in intersection points of the 274 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE IV CHARACTERISTICS OF TRAFFIC TYPES FOR INDUSTRIAL COMMUNICATIONS TABLE V CHARACTERISTICS OF TRAFFIC TYPE CATEGORIES IN INDUSTRIAL AUTOMATION SYSTEMS IIC [84] considers that traffic scheduling based on time synchronization may still be used for cyclic traffic to guarantee bandwidth and bounded latency, given that the presence of isochronous traffic already demands it. These traffic types that represent industrial automation communications can be mapped to TSN requirements and configuration. The mappings between traffic types and traffic classes (i.e., the traffic priorities marked in the PCP field of the VLAN tag) proposed by these sources, also compared to the list of traffic types standardized in 802.1Q (Annex I), is presented in Table VII. On the other hand, Table VIII shows the mapping between each traffic type and the usage requirements of the primary TSN standards. As reflected in Table VIII, 802.1Qbv along with 802.1AS are used to offer dedicated cycle times to the traffic types with the most critical requirements in terms of bounded low latency. Moreover, time-based 802.1Qci guarantees that no interfering frames occupy their reserved cycle times. 802.1Qbu and 802.3Qbr could also be used in complementary fashion to 802.1Qbv as discussed in Section III-D. On the other hand, asynchronous and sporadic traffic classes tend to have more loose requirements, relying primarily on frame preemption from 802.1Qbu and 802.3Qbr and on rate-based TSN mechanisms that do not require time synchronization. The most critical traffic out of these traffic classes, such as Cyclic- Asynchronous and Events: Control, shall be configured as express traffic. Bandwidth reservation based on 802.1Qav can be applied mainly to video and voice traffic classes that have the lowest latency constraints, to provide them with bandwidth guarantees over best effort traffic. As far as redundancy is concerned, the usage of frame replication via 802.1CB is primarily considered according to the network topology as well as to the criticality and loss tolerance of the applications. Finally, 802.1Qcc applies to most industrial communication traffic classes, as it provides the mechanisms to configure features from 802.1Qav, 802.1Qbv, 802.1Qci, and 802.1CB, across the TSN domain. IV. THE 5G TECHNOLOGICAL LANDSCAPE IN THE CONTEXT OF TIME-SENSITIVE NETWORKING The cutting-edge capabilities introduced by 5G networks have caused a paradigm shift in how verticals can leverage the network to enable innovative and groundbreaking applications. To address the exponential increase of demands raised by these heterogeneous vertical applications, an unprecedented level of flexibility, scalability, and agility is required from the network and from the service provisioning mechanisms. This section first reviews, in Sections IV-A and IV-B, 5G mechanisms to support deterministic and reliable communications over the wireless medium, as well as procedures to facilitate the interoperation between 5G networks and TSN networks. Then, Sections IV-C–IV-E review key technologies and frameworks that have driven the evolution towards network architectures that can accommodate the aforementioned heterogeneity and complexity, together with the existing technological support to adopt these architectures and paradigms in TSN networks. A. An Open and Programmable Service-Based 5G Architecture That Meets Stringent Requirements The necessity to satisfy an ever-growing set of diverse communication requirements highlighted several limitations of SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 275 TABLE VI CLASSIFICATION OF TRAFFIC TYPES FOR INDUSTRIAL COMMUNICATIONS TABLE VII TSN TRAFFIC CLASSES FOR INDUSTRIAL COMMUNICATION TRAFFIC TYPES TABLE VIII APPLICABILITY OF TSN STANDARDS TO INDUSTRIAL COMMUNICATION TRAFFIC TYPES the monolithic, static, and proprietary black box approach of the 4G network architecture. The inflexibility of 4G networks crucially limits their scalability and their ability to provide critical capabilities to novel use cases [122], and leads to higher CAPEX/OPEX costs and higher complexity [123]. Release 15 of the 3rd Generation Partnership Project (3GPP) 276 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE IX OVERVIEW OF MAIN 5GC NFS marks the transition between 4G and 5G. As such, this 3GPP release standardizes the first version of the 5G System (5GS), defined as the “3GPP system consisting of 5G Access Network (5G-AN), 5G Core (5GC) Network and UE” [124]. 5G evolution has continued with the standardization of Release 16 and Release 17, and its currently in its evolution towards 5G Advanced with the standardization of Release 18. In order to overcome the limitations of 4G, 5G standardization efforts have been executed starting from the native adoption of softwarization, cloudification, and disaggregation for the functionalities of the 5GS architecture. The first step towards disaggregation started in 3GPP Release 14 with the introduction of Control and User Plane Separation (CUPS), which enables more flexible deployments by physically decoupling of control and user plane functionality in the Core Network. 5G networks, however, have fully revamped the architecture through the definition of the Service-Based Architecture (SBA). The 5GC SBA is built upon the design principles of modularity and open interfaces, targeting service agility and adaptability. Emerging technological trends such as NFV, SDN, cloud-native, and intelligent end-to-end orchestration are a perfect fit for the SBA architecture. With SBA, the 5GC is implemented in the form of flexibly deployable and configurable Network Functions (NFs) that interact via service-based interfaces. An overview of the primary NFs that make up the 5GC since Release 15 is provided in Table IX. The 5G-AN also follows similar trends of pursuing openness and leveraging virtualization technologies. Building upon the concept of Cloud-RAN (C-RAN) introduced in 4G networks to decouple base band processing functions carried by a centralized pool of Base Band Units (BBUs) from radio functions carried by the Remote Radio Heads (RRHs)—known as Radio Units (RUs) in 5G—, the 5G-AN has introduced a functional split in the BBUs. This functional split enables the relocation of lower layer BBU functions to the antenna site, relieving the fronthaul link. As such, a BBU within the 5G base station (gNodeB, gNB) can be further split into a Distributed Unit (DU) at the antenna site, and a Centralized Unit (CU), which handle lower and higher layers of the 5G New Radio (5G NR) protocol stack, respectively. A CU can be connected to a pool of multiple DUs and thus support multiple gNBs. Overall, this improves the flexibility and cost-efficiency of Radio Access Network (RAN) deployments, and, in this line, facilitates the virtualization of the RAN in an architecture known as Virtual RAN (vRAN). Based on vRAN and seeking an open, interoperable, and intelligent mobile RAN, the O- RAN Alliance [125] is working on the development of the O-RAN architecture, which is also steadily gaining industry adoption. The allocation of RAN resources for transmitting data (i.e., scheduling) in 5G entails challenges inherent to the wireless medium given its time-variant transmission characteristics, which has implications on the propagation latency and jitter and on the reliability of the transmissions. URLLC features introduced in the 5G NR specifications provide latency and reliability enhancements to 5G communications in order to support new vertical applications with stringent requirements, such as in industrial automation [126]. In other words, URLLC capabilities of 5G networks contribute to the support of wireless data transmission with ultra-low bounded delivery times and with ultra-high reliability guarantees. Hence, URLLC capabilities are synergistic with the capabilities that TSN has been conceived to provide to wired, Ethernet-based communications, and are complementary to the system architecture aspects later reviewed in Section IV-B for integrating 5G and TSN networks. 5G NR uses the Orthogonal Frequency-Division Multiple Access (OFDMA) modulation scheme. The transmission SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 277 resources are quantized into Resource Blocks (RBs). In the frequency domain, each RB consists of a total of 12 orthogonal sub-carrier frequencies, allowing simultaneous transmissions from different users on different sub-carriers. In the time domain, RBs are divided into multiple time slots called Transmit Time Intervals (TTIs). In the default transmission scheme, each TTI uses 14 Orthogonal Frequency-Division Multiplexing (OFDM) symbols. The length of a TTI depends on the number of symbols per TTI and on the duration of such OFDM symbols. Flexible Sub-Carrier Spacing (SCS) and mini-slot scheduling are two significant Release 15 URLLC capabilities that target latency improvements. While fixed at 15 kHz in 4G, the SCS can exist at 15, 30, 60, 120, and 240 kHz in 5G. This feature can reduce transmission latency thanks to decreasing OFDM symbol duration. Additionally, while in 4G the default slot-based scheduling of Physical Downlink Shared Channel (PDSCH) and Physical Uplink Shared Channel (PUSCH) uses a separation of 14 OFDM symbols, 5G supports scheduling packets for transmission in mini-slots that use 2, 4, or 7 symbols. This is also known as sub-slot-based transmission. Combining flexible SCS with sub-slot scheduling, transmission intervals can be scaled between 8.93 µs and 1 ms, which allows for more stringent and granular periodicities in the scheduling of TSN traffic streams [127]. Semi-static scheduling based on Configured Grant (CG) for uplink, or on Semi-Persistent Scheduling (SPS) for downlink or for uplink, can reduce transmission latency. By default, each Dynamic Grant (DG) uplink transmission requires a Scheduling Request from the UE and a grant from the gNB first, which accumulates latency. Support for CG uplink transmissions for periodic traffic are introduced in 5G, eliminating the need to send Scheduling Requests by preallocating resources for the UE instead. Release 16 enhances CG in uplink and SPS in downlink with support for configuring multiple CG and SPS resources with different scheduling parameters and periodicities, which is key for supporting cyclic industrial control traffic flows with heterogeneous characteristics [28]. Other Release 16 URLLC features concerned with improving latency are: PUSCH enhancements that allow a CG or DG can schedule multiple PUSCH repetitions in consecutive slots, which eliminates the time gap between repetitions; sub-slot- based Hybrid Automatic Repeat Request acknowledgement (HARQ-ACK) feedback supporting more than one Physical Uplink Control Channel (PUCCH) carrying HARQ-ACK within a slot; enhancements to Logical Channel Prioritization (LCP) restriction, allowing to restrict the mapping of logical channels to specific CG configurations; and intra-UE prioritization that enables a UE to drop an already-scheduled transmission whose resources overlap in time with a more time-critical flow. In regards to URLLC capabilities that target reliability improvements via Protocol Data Unit (PDU) redundancy, Release 16 includes Packet Data Convergence Protocol (PDCP) Packet Duplication and higher-layer multi connectivity. With PDCP packet duplication, PDUs are duplicated at the PDCP layer and can be submitted via independent transmission paths. To enable this, additional Radio Link Control (RLC) entities are added to the radio bearer, leading to one primary logical channel and multiple secondary ones. After the transmission of a PDCP PDU is acknowledged, others are discarded. Higher-layer multi connectivity consists on duplicating transmissions over disjoint user plane paths. These redundant paths can be set up by having the UE establish two PDU sessions using Dual Connectivity, or via a single PDU session that uses redundant tunnels between the RAN and the UPF or between the RAN and two Intermediate UPFs. The ability to setup redundant transmission paths can be leveraged to implement TSN features, particularly FRER [128]. Other URLLC capabilities for enhanced reliability include Release 16 improvements to Multiple Transmission and Reception Point (Multi-TRP) technology, which can enhance Signal to Interference plus Noise Ratio (SINR) in presence of interference or poor channel conditions through joint transmissions and receptions. In Release 15, a new Channel Quality Indicator (CQI) table with a target Block Error Rate (BLER) of 10−5for channel state reporting, and a new 64-QAM Modulation and Coding Scheme (MCS) table for very low spectral efficiency are introduced. Additionally, to improve robustness of Physical Downlink Control Channel (PDCCH), Release 16 introduces Downlink Control Information (DCI) formats with configurable sizes, and a higher PDCCH Aggregation Level (AL), which increases decoding capability of PDCCH by the UE. As far as architectural aspects of 5G are concerned, a cornerstone technology introduced by 5G networks to simultaneously comply to diverging use case requirements is network slicing [129]. Fundamentally, network slicing means that a shared physical network infrastructure can be partitioned— sliced—into multiple isolated logical networks, each tailored to guarantee a QoS that meets the requirements of a specific application or vertical use case [130] [131] [132]. Relying on the foundations provided by NFV and SDN technologies, these network slices are comprised by a set of NFs and associated programmable resources, and can be dynamically created, scaled, and terminated according to demand. Largely, network slices can be used to provide dedicated logical networks with granular and customized configuration according to specific functional (e.g., URLLC), administrative, or business requirements [133], through procedures carried within both the Core Network and the RAN [134]. Finally, another key capability that distinguishes 5G technology from its predecessors is network programmability, which is critical for facilitating vertical customers to translate their QoS requirements into tailored network services and configurations. Tightly linked with the SBA, 3GPP has defined 5GC exposure capabilities through standard APIs over the Network Exposure Function (NEF), enabling external systems to communicate with the 5GC in a secure manner and influence network services and resources. In this line, frameworks such as Common API Framework (CAPIF) [135] and Service Enabler Architecture Layer (SEAL) [136], introduced in Release 15 and Release 16 respectively, augment these exposure capabilities through the provision of enriched interfaces, procedures and auxiliary services, which, for example, opens 278 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE X LOWEST BOUNDS OF MAXIMUM END-TO-END LATENCY FOR SIGNIFICANT INDUSTRIAL VERTICAL APPLICATIONS ACCORDING TO TS 22.104 up new opportunities for industrial applications in 5G nonpublic networks (NPN) [137]. These openness capabilities of 5G networks have propelled a new ecosystem of Network Applications, or NetApps for short. By consuming the APIs exposed by the 5G control plane, NetApps constitute a middleware layer for vertical applications towards 5G functionalities, simplifying the deployment of vertical applications with enhanced intelligence by virtue of taking full advantage of 5G capabilities [138]. B. Standards for Enabling Time-Sensitive Communications Over 5G Networks The primary reference for integrating 5G and TSN networks is 3GPP standardization activities. The need for supporting TSN integration and Time Sensitive Communications (TSC) in 5G systems is reflected in 3GPP Technical Specification (TS) 22.104 [139],[140]. This document discusses the communication requirements driven by industrial vertical domains, including: factory automation, process automation, and human-machine interaction for the Factories of the Future; electric-power distribution and central power generation; and, in the Release 17 revision, connected hospitals. This TS arrives to an extensive list of end-to-end performance requirements that the 5GS shall comply to in order to support the needs of cyber-physical control applications in these vertical domains, which feature communication services aligned with those targeted by TSN systems as described in Section III-I. Key Performance Indicators (KPIs) provided with TS 22.104 for different kinds of 5G industrial applications include end-to-end latency, availability and reliability, UE density, service area, clock synchronization, and positioning accuracy. In regards to clock synchronization, a maximum error of 900 ns is targeted for both Working Clock and Global Time, as well as support for up to four simultaneous Working Clock synchronization domains on a UE. Latency and reliability KPIs are provided according to vertical application. Service availabilities of up to 8 nines (99.999999%) and mean time between failures of up to 10 years are present. Some of the most stringent latency KPIs defined in TS 22.104 are shown in Table X. The specifications for achieving these time-sensitive capabilities in 5G networks are mainly covered in TS 23.501 and TS 23.502. Extensions to provide support for TSC are introduced in 3GPP 5G Release 16 [124],[141],bythe means of the explicit integration between the 5GS and TSN. Further enhancements to TSC are added in Release 17 [142], [143]. In fact, the capabilities and features of TSC including 5G-TSN integration are expected to continue evolving with Release 18 and future releases, as 5G technology matures into 5G Advanced and, eventually, 6G. The remainder of this subsection delves into the architecture aspects, mechanisms, and procedures standardized as of the writing of this article. 1) 3GPP Standards Support for an Integrated 5G-TSN Architecture: Release 16 introduces an architecture to support TSC by defining mechanisms and procedures that permit to expose the 5GS as a logical TSN bridge. This bridge can be used to interconnect field devices, AGVs, mobile robots, cameras, etc. to external TSN networks wirelessly through a 5G UE, hence seeking to remove the dependency from inflexible wired connections to interconnect the assets of production lines in industrial automation networks. The dedicated entities that enable this modelling of the 5GS are the Device-Side TSN Translator (DS-TT), the Network-Side TSN Translator (NW-TT), and the TSN Application Function (TSN AF). The TSN AF is responsible for the control plane interactions with the CNC in the TSN domain, performing the exchange of port and bridge management information between both domains. DS-TT and NW-TT are located at the edges of the 5GS, in the UE side and in the UPF side, respectively. Being synchronized to both the 5G time domain and to the TSN time domain, they are the entities responsible for the interoperation and synchronization between both domains. A 5GS TSN bridge (5GS bridge for short) is composed of a single UPF and associated NW-TT, and of one or more UEs and associated DS-TTs. Each UE establishes a PDU session of type Ethernet terminating in the UPF of the bridge. Multiple PDU sessions from the same UE terminating in different UPFs are modelled as separate bridges, with a different DS-TT being part of each bridge for each separate PDU session established by the UE (Figure 8). Hence, there is a one-to-one relationship between Bridge ID and UPF ID. Meanwhile, the TSN AF is responsible for storing the association between UE/DSTT ports, their MAC addresses—which are known by the SMF after PDU session establishment—and Bridge ID. This information is used for binding TSN traffic with a respective UE’s PDU session. The information model encompassing a 5GS bridge that can be configured in the 5GS by the TSN network and/or reported from the 5GS to the TSN network includes the attributes listed below. Information received in the 5GS from the TSN network can be mapped to 5GS information to configure 5GS QoS for TSN traffic. •Bridge description, including Bridge ID and list of ports of DS-TT and NW-TT. •Capabilities of the 5GS bridge that are reported to the TSN network, as defined in 802.1Qcc. This comprises bridge delay per port pair per traffic class, propagation delay per port, and VLAN configuration information. SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 279 Fig. 8. The 5G System acting as one or more logical TSN bridges allowing to connect wireless field devices in a production line to a wired TSN network. An UE that establishes redundant PDU sessions towards separate UPFs sets up connectivity through multiple logical TSN bridges. •LLDP configuration information, and neighbor discovery information and configuration. •Scheduled traffic information according to the 802.1Qbv standard, including traffic classes and their priorities per port and gate control information. •Information related to the 802.1Qci standard, including the Stream Parameter Table that describes PSFP support, and PSFP stream filters and stream gates. To allow for the transmission of TSN scheduled traffic over a 5GS bridge, the 5GS shall also support a hold and forward buffering mechanism implemented in DS-TT and NW-TT ports. This mechanism targets achieving the behavior provided by TAS as defined in the 802.1Qbv standard, transmitting or holding back frames according to the state of the corresponding transmission gate. However, both the Release 16 and 17 standards do not further specify on the details, leaving them up to implementation. Apart from the port and bridge management information exchange between CNC and TSN AF, the 5GS also shall support transparent transfer of this information between TSN AF and DS-TT or NW-TT. This is done through standardized Port Management Information Containers (PMIC) and Bridge Management Information Containers (BMIC), which contain much of the information described above. Release 17 renames BMIC to User plane node Management Information Container (UMIC). The support for TSC and for integration with TSN networks was enhanced in Release 17 with the addition of new capabilities and features. Many of these target the support for non-TSN TSC, which is not considered the focus of this research paper. For example, Release 17 adds support for TSC PDU sessions of type IP and support for IEEE 1588-based time synchronization. In this context, Release 17 introduces a new Network Function called Time Sensitive Communication and Time Synchronization Function (TSCTSF). The TSCTSF enables the new time synchronization mechanisms, as well as non-TSN TSC scenarios. In Release 17, Management Information Containers may be exchanged between NW-TT and TSCTSF and also include time synchronization parameters for the purpose of enabling the TSCTSF to manage DSTT and NW-TT in the absence of TSN AF. Lastly, Release 17 also introduces support for UE-UE TSC communications, treating the UE-UE TSC stream as a combination of one uplink stream and one or more downlink streams and carrying their procedures separately. 2) 3GPP Standards Support for TSN Time Synchronization: Supporting TSN time synchronization is a critical requirement for enabling the integration between 5G and TSN. Within the 3GPP standardization activities, support for TSN time synchronization is introduced in Release 16 and further expanded in Release 17. Release 16 introduces the exposure of the 5GS as a logical TSN bridge modelled as a time-aware system compliant to the 802.1AS standard, as represented in Figure 9. Within the 5GS, 5G entities including the UE, gNB, and UPF are synchronized to the 5G GM clock according to 5G-AN synchronization mechanisms [144] that utilize the ITU-T G.8275.1 profile for telecommunication applications based on PTP, and a reference time signaling mechanism from the gNB to synchronize an UE. Meanwhile, TSN entities outside the 5GS are instead synchronized to the TSN GM in the TSN domain. The TSN translators (i.e., NW-TT and DS-TT) at the edges of the 5GS are synchronized to both GM clocks, and, implementing support for 802.1AS operations, are responsible for linking the 5G time domain and the TSN time domain. Release 16 supports the election of the GM instance and the determination of the time synchronization tree via manual configuration. DS-TT ports are configured as Master Ports 280 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 Fig. 9. Modeling of the 5G System to support downlink TSN time synchronization from a TSN domain behind the UPF towards an end station behind an UE, including the transportation of timing messages from Master Ports to Slave Ports. NW-TT and DS-TT are the entities responsible for linking the 5GS time domain and the TSN time domain. and forward timing information to the downstream TSN nodes behind the UEs, whereas each NW-TT port is configured as a Slave Port to receive timing information from the upstream gPTP time-aware network to which the UPF of the bridge is connected to. The procedures to enable TSN time synchronization across the 5GS are implemented according to the 802.1AS standard. In Release 16, gPTP synchronization is always initiated downlink from the GM in the TSN domain. When a NWTT receives a gPTP message, it needs to calculate the message’s updated CorrectionField taking into account the link delay from the upstream TSN node, and the updated CumulativeRateRatio accounting for the frequency offset between 5GS time and TSN time. Such fields are updated in the gPTP messages that the NW-TT forwards through the 5GS towards the DS-TT. The residence time in the 5GS is calculated via the creation of ingress and egress timestamps (TSi and TSe) by the NW-TT and DS-TT respectively. The TSi is added to the gPTP messages by the NW-TT and later removed by the DS-TT. Finally, this residence time is used to calculate the final value of the CorrectionField when the DS-TT forwards the gPTP message to the downstream TSN node. In Release 16, the connection to multiple TSN domains is also supported, with the NW-TT and DS-TT entities being able to handle the timestamping and forwarding of gPTP messages according to the above procedure for one or more domains. Release 17 expands the support of TSN time synchronization capabilities in the 5GS. Apart from operating as a 802.1AS time-aware system, the 5GS can also operate as other types of PTP instances, including IEEE 1588 compliant Boundary Clock, End-to-End Transparent Clock, and Peer- to-Peer Transparent Clock (Figure 9). Hence, from Release 17, the 5GS system supports both gPTP and PTP-based time synchronization with external time domains. Additionally, in the gPTP and PTP Boundary Clock schemes, the 5GS can be configured to operate as the GM clock for the gPTP or PTP domain, as opposed to the clock source being located in the wired TSN network. Support for uplink time synchronization, including the UE-UE scenario, is also introduced, corresponding to a GM located behind an UE rather than in the network behind the UPF. Having multiple TSN domains behind the UE each with their own GM for uplink time synchronization is also supported. Additionally, Release 17 introduces support of the BMC algorithm for determining the GM clock and the time synchronization tree, as opposed to resorting to only preconfigured GM and port states. Finally, time synchronization exposure capability towards an arbitrary AF is introduced in Release 17, allowing the AF to request time synchronization services (indirectly via NEF if the AF is non-trusted) that may include a specific time synchronization error budget. To meet the strict time synchronization requirements of time-sensitive communications, TR 38.330 [145] in Release 17 establishes the use of two techniques that help compensate the propagation delay within the 5GS in the residence time that is added to the CorrectionField of the gPTP or PTP messages. On the one hand, Round-Trip Time (RTT)-based PDC employs 5G NR reference signals transmitted between gNB and UE in both directions. Both gNB and UE measure the time difference in the reference signal received from the other entity, and either of the two provides the other entity with its own measurement, so that it can perform PDC based on the measurements from itself and from the other entity. On the other hand, Timing Advance (TA) sent from gNB to UE to adjust the timing of uplink transmissions based on the estimated propagation delay can also be used to compensate this delay. 3) 3GPP Standards Support for the QoS Configuration and Translation for TSN Traffic: 5G standards support the integration with TSN networks that apply the fully centralized model as defined in the 802.1Qcc standard. QoS configuration to support TSN traffic in the 5GS involves the use of TSC Assistance Information (TSCAI) and of 5G QoS flows. The aim is to achieve a translation of parameters from the wired, time-domain-based Ethernet, to the wireless, frequencydomain-based 5G through the most appropriate mapping of SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 281 TSN stream requirements to 5G radio resources. For this purpose, 3GPP has standardized support for QoS configuration that may be implemented in two ways according to availability of PSFP information provided by the CNC, or the lack thereof. •If PSFP information is always provided to the TSN AF by the CNC, then this information is used to setup the QoS flows in the 5GS. PSFP information can also be used to derive TSCAI for each QoS flow. •If PSFP information is not provided by the CNC, preconfigured QoS flows that do not require PSFP information are used, and may be setup during PDU session establishment. The means to derive TSCAI are not specified for this case. It may however be still possible to derive QoS information (priority and delay) of TSN streams from scheduled traffic information (i.e., the TAS GCL) provided by the CNC. TSCAI includes a description of TSC flow traffic patterns that can be provided to the gNB to allow it to efficiently configure 5G-AN resources for the scheduling of deterministic traffic flows (for example via CGs or SPS). TSCAI parameters are derived from the PSFP information that the TSN AF receives from the CNC. TSN AF forwards per-flow TSC Assistance Container (TSCAC) with traffic pattern parameters calculated from this information to the SMF. These parameters are Flow Direction (uplink or downlink), Periodicity (time period between transmissions, or bursts), and Burst Arrival Time. Release 17 introduces a new parameter, Survival Time, which refers to the tolerance against the lack of arrival of anticipated messages. The SMF builds the TSCAI by adjusting these parameters according to the difference between 5G and TSN clocks, and forwards the result to the 5G-AN. Whereas TSCAI contains QoS configuration used specifically for the traffic characteristics of TSC flows, all 5G QoS flows are identified by a QoS Flow Identifier (QFI) and characterized by a QoS Profile. A QoS Profile firstly classifies flows according to Resource Type, that is, between Guaranteed Bit Rate (GBR) ones and Non-Guaranteed Bit Rate (Non-GBR) ones. These QoS Profiles describe several QoS characteristics, which, for GBR flows, include at least the following: Allocation and Retention Priority (ARP), which determines priority level and preemption capability, uplink and downlink Guaranteed Flow Bit Rate (GFBR) and Maximum Flow Bit Rate (MFBR), and a 5G QoS Identifier (5QI). To determine the right QoS Profile for each TSN stream, the TSN AF is preconfigured with a QoS mapping table that associates TSN traffic classes to bridge delays—which can be calculated after PDU session establishment—and to priority levels. A 5QI references a combination of additional QoS characteristics that impact the forwarding treatment of a given flow in the RAN. Precisely, QoS Profiles and 5QIs can be considered mechanisms to implement 3GPP RAN network slicing in 5G networks [146]. 3GPP has standardized several 5QI values [124], including some with a Resource Type of Delaycritical Guaranteed Bit Rate (Delay-critical GBR) aimed at use cases such as time-sensitive industrial communications. The 5GS supports the use of both standardized or customized 5QIs, and, as such, TSC QoS flows may also use either option. Apart from the aforementioned Resource Type characteristic that determines the allocation of resources, additional QoS characteristics are included in the 5QI, such as Priority Level for the scheduling of resources, upper bounds for the Packet Delay Budget (PDB) and for the Packet Error Rate (PER), and Maximum Data Burst Volume (MDBV). These can be mapped accordingly, via a PCF QoS mapping table, from TSC QoS parameters such as the maximum burst size of the aggregated TSC streams allocated to a 5G QoS flow that is provided by the CNC, and the maximum flow bitrate derived from the calculation of TSCAI. The PDB of a 5QI has to satisfy— i.e., be equal to or higher than—the bridge delay reported by the 5GS bridge to the CNC for its associated traffic class. PDB-based QoS can be applied using the hold and forward mechanism. Figure 10 summarizes the steps involved in the configuration of a TSN stream, focusing on the configuration of 5GS bridges that are included in the path. This follows procedures defined by either the 802.1Qcc standard for the fully centralized model, or by 3GPP standards. For additional details on the 5GS procedures involving TSN bridges, including the intervention of 5GC NFs like the PCF, SMF, and AMF, the reader is referred to Annex F of TS 23.502 [143]. Based on the 3GPP standards, 5G-ACIA white paper [147] provides a detailed description of the 5G QoS model and of 5G procedures for enabling the configuration and management of QoS features for time-sensitive flows. Some key takeaways for the industrial automation use case in particular are: •The ability for a UE, that could for example be connected to a field device, to request the modification of a PDU session in order to assign different QoS parameters or additional QoS flows. •The ability for an external application, such as Manufacturing Execution System (MES) and Enterprise Resource Planning (ERP), to flexibly manage QoS Profiles and monitor the QoS parameters via the consumption of the NEF APIs of the 5GC. •The support for adaptive QoS applications in which two or more alternative QoS Profiles with different priorities may be specified, so the application can survive events of temporary QoS degradation in the network providing a lower quality service (e.g., adjusting the resolution of a video stream). •The interworking with different deployment configurations of 5G network slicing for the provisioning of tailored QoS flows. 4) Time-Sensitive Networking for Fronthaul: While one direction of standardization and research work involving the usage of 5G and TSN together is the 5GS bridge for interconnecting TSN devices over 5G, another direction addresses the usage of TSN to achieve deterministic low-latency transmissions in the 5G TN. As such, rather than studying how a 5G network can be incorporated as a wireless subcomponent of an industrial TSN network, this research direction leverages the adoption of TSN within the wired infrastructure—i.e., between RAN and UPF—of a 5GS that could itself act as a 5GS bridge as part of an end-to-end TSN network. The 282 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 Fig. 10. Fully centralized TSN network configuration for a TSN stream spanning a TSN bridge and a 5GS bridge. cornerstone standard for using TSN as a building block for 5G TNs is the 802.1CM-2018 standard [148], a collaborative effort between Common Public Radio Interface (CPRI) and IEEE 802.1. 802.1CM is the TSN Profile for Fronthaul, and, much like the case of the TSN-IA profile, it selects TSN features and configurations for, in this case, transporting timesensitive fronthaul streams. Even though this approach may not get as much focus as the 5GS bridge in the context of enabling time-sensitive industrial communications over 5G, using TSN in the TN can be a key enabler for achieving end-to-end deterministic communications [149], specifically for the fronthaul connection, which has very stringent packet loss and delay requirements. Furthermore, both approaches for combining 5G and TSN can be employed in complementary fashion. 802.1CM describes how to meet the strict requirements of Ethernet-based fronthaul networks using TSN mechanisms. Following the 5G functional split, a fronthaul link connects a RU with a DU, and is implemented through the CPRI protocol, or its enhanced version, evolved CPRI (eCPRI). Both CPRI and eCPRI are supported in the 802.1CM standard, in which they are called Class 1 traffic and Class 2 traffic respectively. CPRI and eCPRI define different functional split options, which indicate the line in which the division between DU and CU functionalities takes place. Different kinds of information flows are transmitted through the fronthaul link, as per the CPRI and eCPRI specifications. This includes User data, Control and Management (C&M) data, and Synchronization data. User data includes In- Phase/Quadrature (I/Q) data, and, in the case of Class 2 traffic, also other data related to the User plane. Synchronization data is provided separately from the other two. For time synchronization, the standard specifies support for the ITU-T G.8275.1 telecom profile and one or more of the related clocks as a requirement. TABLE XI CHARACTERISTICS OF CLASS 1 (CPRI) INFORMATION FLOWS TABLE XII CHARACTERISTICS OF CLASS 2(ECPRI) INFORMATION FLOWS The respective CPRI/eCPRI specifications define the requirements for the remaining two types of data, in the form of maximum latency and frame loss probability. This leads to three types of fronthaul traffic classes according to priority, which are High Priority Fronthaul (HPF) with the lowest oneway latency requirement of 100 µs, Medium Priority Fronthaul (MPF) with a one-way latency requirement of 1 ms, and Low Priority Fronthaul (LPF), with the most relaxed requirement of 100 ms one-way latency. Table XI and Table XII summarize the requirements of the information flows and their mapping to fronthaul flow types, for Class 1 traffic and Class 2 traffic respectively. To satisfy such requirements, 802.1CM defines two possible profiles, called Profile A and Profile B. The two profiles are applicable to both Class 1 and Class 2 traffic, and essentially differ in the TSN features adopted. Specifically: SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 283 •Profile A: the strict priority algorithm, as defined in the 802.1Q standard, is employed in order to treat User Plane data (higher priority) and Control and Management Plane data (lower priority) differently. The highest priority is assigned to HPF traffic, followed by MPF traffic. •Profile B: apart from making use of the strict priority algorithm like in Profile A, frame preemption mechanisms of 802.1Qbu and 802.3br are also used. This allows non-fronthaul traffic to be preemptable by fronthaul traffic. It is worth noting the existence of ammendment 802.1CMde-2020 [150] for the 802.1CM standard. It introduces enhanced synchronization features and other updates that specifically address 5G fronthaul interfaces standardized after the release of 802.1CM-2018. C. Virtualization and Cloudification of Time-Sensitive Applications Virtualization has proven to be a technology that can significantly improve flexibility and cost-efficiency in comparison to monolithic approaches, being specifically a focal technology in the industrial automation vertical that concerns this work to develop novel Industry 4.0 and beyond applications [151],[152]. As industrial control functions move to private edge cloud infrastructures enabled by 5G [153], paradigms such as NFV, MEC, and cloud-native can be adopted in the design of the architecture of industrial networks and applications. However, virtualization platforms also entail additional complexity and often come with a disregard for optimized performance that may hamper their viability to host time-critical applications with low latency and bounded delivery time requirements in the communications. Furthermore, general-purpose systems do not focus on providing real-time capabilities, requiring specific tuning for example in embedded systems for industrial control applications that are commonly based on Linux [154],[155]. This subsection first reviews available technologies and mechanisms that can address these challenges, and later introduces the virtualization and edge computing paradigms mentioned above. 1) Technologies and Mechanisms for Deterministic End Stations with Virtualization: The Linux ecosystem provides several TSN features to accommodate TSN-compliant applications. Beginning with VLAN tagging support, multiple virtual interfaces with the desired VLAN PCP can be associated to a physical interface. Internal Linux packet priorities, set with socket option SO_PRIORITY, can be flexibly mapped to VLAN PCPs. For enabling time synchronization, Linux PTP [156] is the most popular implementation of PTP. It includes support for the gPTP profile and supports hardware and software timestamping; the former also requires NIC and driver support. Linux PTP includes two utility daemons, ptp4l and php2sys, to synchronize the hardware clock of the NIC and the system clock, respectively. Traffic shaping functionality is implemented through Traffic Control (TC) Queuing Disciplines (Qdiscs), which are associated to a specific network interface to control the scheduling, shaping, and policing of transmitted traffic. CBS functionality is implemented in the MultiQueue Priority (MQPRIO) Qdisc, whereas TAS functionality is implemented in the Time Aware Priority (TAPRIO) Qdisc. The configuration of these Qdiscs includes feature-specific parameters, and the mappings between Linux priorities, traffic classes, and the transmission queues of the network interface. Additionally, the Earliest TxTime First (ETF) Qdisc can be used alongside the TAPRIO Qdisc to provide time-based scheduling using the SO_TXTIME socket option. The TSN Documentation Project for Linux [157] gathers comprehensive documentation and configuration instructions regarding the features introduced in this paragraph. To provide real-time capabilities to the standard Linux kernel, the PREEMPT_RT patch [158] was developed. This real-time preemption patch targets a maximization of the amount of points in the kernel code where its execution can be interrupted, thus reducing the delay upper bound that real-time tasks may experience. PREEMPT_RT is being progressively mainlined in the Linux kernel. An alternative approach to achieve real-time capabilities is the use of a real-time cokernel that runs in parallel to the Linux kernel and specifically executes the time-critical workloads. Examples of the cokernel approaches are Xenomai [159] and RTAI [160]. Lastly, a potentially less intrusive alternative to satisfy softer realtime requirements is the low latency Ubuntu kernel [161], which is configured with the maximum preemption and timer granularity available in the mainline kernel. Task scheduling in Linux is carried out according to policies. The SCHED_FIFO, SCHED_RR (Round Robin), and SCHED_DEADLINE policies are called real-time policies as tasks scheduled with these policies always have priority over those scheduled with normal ones. The SCHED_DEADLINE policy in particular incorporates awareness of the deadline of the task. Virtualization technology potentially increases the amount of mixed-criticality applications executing processes in the same host system, highlighting the need to use realtime scheduling policies that can guarantee temporal isolation between the virtualized components [162],[163]. Another potential source of latency in virtualized applications is related to the network stack. Virtualization introduces a virtual switching layer responsible for directing the packets to and from the respective virtual components that use an emulated virtual NIC, all of which can be causes of additional overhead [164]. Enhanced Platform Awareness [165] configuration such as CPU and interrupt affinity can for example optimize the servicing of packet processing interrupts. In this context, PCI Passthrough and SR-IOV Passthrough techniques can be used to provide a virtualized component with direct access to a physical network interface. SR-IOV allows the partitioning of the NIC at hardware level into Virtual Functions that can be shared between guests. Still, flexibility is somewhat reduced as the guest remains tied to a particular NIC. Furthermore, typical NICs available in the market usually do not support both TSN and Passthrough [166]. Last but not least, acceleration technologies such as Data Plane Development Kit (DPDK) [167] and eXpress Data Path (XDP) [168] can also be used to circumvent overhead introduced by the network stack, especially when a virtualization 290 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE XIII TAXONOMY RELATED TO THE DESIGN AND EVALUATION OF TSN-OVER-5G ARCHITECTURES the enablement of time synchronization on integrated 5G-TSN networks, neither those that predominately tackle procedural aspects of the configuration of TSN traffic in such networks, as they will be surveyed in subsequent subsections. Table XIII shows a taxonomy for the research papers surveyed in this section. Regarding QoS mapping between TSN and 5G, there are several complementary scopes researched in the literature, such as bridge delay estimation and the configuration of 5QIs. When it comes to optimizing latency and determinism over 5G, most proposals can be identified under semi-static scheduling schemes, such as those based on SPS and CG, though other approaches can also be found. Meanwhile, mechanisms proposed for enhancing reliability include NR-DC and other redundant user plane path configurations, CoMP, and Multi-TRP. 1) Network Architectures for Integrating TSN into 5G Networks: Authors of [234] propose different hybrid scenarios that combine 5G and Industrial Ethernet based on TSN in order to address reliability and latency requirements of industrial communication systems. Predating much of the standardization work by SDOs such as 3GPP and IEEE/IEC, this article evaluates how different network topologies can be applied to different wireless communication use cases in the Smart Factories of the future, ranging from mobile robotics and closed loop control to massive sensor networks. Such 5G-TSN integration scenarios introduce aspects such as the virtualization of a centralized controller in an edge cloud or wireless interconnection between remote production sites. Conducted in parallel to 3GPP Release 16 standardization work, [235] explores preliminary approaches for an architecture, and protocol design aspects, for the integration of 5G with TSN. Key features that allow the 5GS to operate as a TSN switch are described, as well as architecture and technology-level considerations for enabling end-to-end time synchronization, transmission scheduling, and 802.1Qbv and 802.1Qcc-based QoS management for TSN traffic with stringent latency and reliability requirements. Reference [236] proposes an harmonized usage of several technologies for the realization of a 5G network architecture for a manufacturing factory, that comprises out-factory and in-factory domains. The use of edge computing, URLLC, and TSN technologies is proposed for the latter, where QoS monitoring and dualconnectivity redundant transmission mechanisms are used to enhance end-to-end delay and reliability. Authors evaluate their architecture with a practical remote control use case. Likewise, different options for deploying 5G NPN networks using edge computing architectures are presented in [237], where the interworking with TSN is also explored. In [238], authors propose a 5G architecture tailored for industrial TSN services. Their contributions are mainly twofold: on the one hand, radio access technologies including New Radio Dual Connectivity (NR-DC), and Multi-TRP, a Coordinated Multipoint (CoMP) capability introduced in 5G, are studied for improving network quality at different levels, primarily with regards to reliability and transmission rate; on the other hand, in order to improve data security and agility, an on-premises MEC deployment is used to implement a disaggregated RAN and the user plane of the 5G network. 2) Analysis and Estimation of 5G-TSN System Latency: In [22], authors look into how the 5GS bridge delay can be estimated, in order to select 5QIs with QoS parameters that are adequate to support the specific end-to-end latency requirements of deterministic industrial applications over 5G. PDB, GFBR, and MDBV are estimated considering reference requirements of a closed loop control use case. Authors conclude by stating the inconvenience of not having standardized 5QI values with lower PDB in the sub-millisecond range, the problematic of periodicity mismatch between the requested GCL and the supported 5G numerology, and the desirability of being able to estimate the bridge delay before the establishment of the QoS flow. In [239], authors perform and end-to-end latency performance of an integrated 5G- TSN system using a simulated 5GS bridge model. Authors quantitatively evualuate the impact of PCP-based prioritization SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 291 and scheduled traffic based on 802.1Qbv on TSN-over-5G performance. Their results show that PCP prioritization for the transmission of frames is particularly beneficial for reducing jitter, whereas the use of 802.1Qbv on top of it significantly reduces latency of high-priority traffic, especially for uplink due to CGs. Their results also indicate that jitter can be significantly reduced at the cost of slightly higher average latency as the length of the GCL time slots becomes larger. Authors conclude the article by highlighting the challenge presented by the need to translate the time-domain scheduling of 802.1Qbv to the OFDM modulation used in 5G. In [240], authors focus on the calculation of the PDB for different industrial scenarios that have adopted 5G-TSN communications. Authors propose three communication scenarios involving a PLC and a field device: (1) single 5G link from wireless device to wired PLC; (2) wireless device to local wireless PLC not crossing UPF; and (3) wireless device to remote wireless PLC crossing UPF. For elaborating the PDB formulas, authors use TSN traffic classes with predefined packet size and packet rate values, take into account end-to-end delay including delay introduced outside the 5GS, and consider the possibility of multiple 5GS bridge hops in the end-to-end communication. Authors propose a 5QI table with PDB values derived from their calculations, although they state that the delay introduced by TSN translators needs further investigation. 3) Evaluation of Traffic Scheduling and Forwarding over 5G: As opposed to 3GPP standardization, the research community has generally focused more on the optimal integration TAS-based traffic shaping of 802.1Qbv rather than PSFP defined in the 802.1Qci standard, deriving QoS and TSCAI parameters from the former. In fact, many research works specifically focus on analyzing the performance of 802.1Qbv-based traffic scheduling through the allocation of 5G radio resources. In [127], authors quantitatively evaluate the performance of transporting TSN traffic over a 5G network. They implement a system-level simulation that models a preliminary 5GS bridge, applying several URLLC functionalities that are mainly from Release 15. They present end-to-end latency and throughput performance results for scheduling multiple TAS streams with different 5G QoS configurations. They highlight open research challenges regarding trade-offs in the optimal configuration and scheduling of resources between the 5G and TSN systems. In a similar vein, a simulation model is proposed in [241] to investigate the challenges of 5G and TSN integration in regards to uplink and downlink transmission scheduling using the 802.1Qbv standard. Their simulation results using preconfigured transmission slots (such as with CG) indicate that a mismatch between application periodicity and the 5G time slots causes degradation in delay and jitter due to variable transmission start times. Meanwhile, in [242], authors provide a comprehensive performance evaluation and comparison of semi-persistent and dynamic packet scheduling (DPS) strategies to support downlink time-sensitive flows in 5G. Authors highlight the benefits in terms of adaptivity to channel conditions of DPS methods; however, the absence of control channel overhead using SPS results in better performance. Finally, in [243], authors consider a 5G-TSN network that accepts both URLLC/TSN and eMBB traffic, and take into consideration both latency and throughput using TAS scheduling. They evaluate an scheme based on the allocation of RBs to TSN streams followed by eMBB streams in a manner that maximizes the overall throughput while satisfying the end- to-end delay of TSN traffic. 4) Proposal of Mechanisms for TSN-compliant Resource Allocation and Traffic Scheduling over 5G: Aforementioned work [242] also proposes optimizations to both SPS and DPS cases to satisfy the latency constraints of TSN traffic over 5G. Authors leverage the information provided in TSCAI and employ Release 15 URLLC features. For performing DPS resource allocation, authors present link adaptation strategies based on CQI reports from the UE. For SPS, authors propose a mechanism to collect SINR samples at the UE, in order to estimate and select an adequate MCS. Authors also analyze the impact on performance of the update frequency of the MCS in use. The mismatch problem between 5G CG and TSN periodicity is addressed in [244], where authors propose a scheduling scheme where two or more CGs per TSN flows operate alternately. Their algorithm involves the activation and deactivation of CGs for each flow over time. This CG shift is autonomously triggered by the UE based on gNB information when the threshold of the transmission latency requirement would be surpassed due to cumulative latencies caused by periodicity misalignment. Also analyzing TSN stream scheduling performance in an integrated 5G-TSN system, [245] proposes packet prioritization mechanisms for the appropriate scheduling and treatment of those streams. Authors analyze the behavior of traffic forwarding and arrive to conditions that meet TSN-compliant scheduling. As previously mentioned works, they emphasize on the challenges that arise from mapping TSN traffic scheduling and forwarding rules to 5G resources and mechanisms. Based on their analysis, authors introduce and evaluate a scheduling framework using dynamic LCP (dLCP). Their purpose with dLCP is to overcome the limitations of time invariant LCP against the requirement of logical channel priority updates after every GCL time slot transition. In [246], authors focus on the resource allocation conflict problem in a 5G-TSN network where the same radio resources are allocated to multiple TSN flows due to different packet periodicities. Authors introduce a novel CG scheduling scheme that relies on the configuration of several grants per flow according to periodicity and other characteristics derived from the information provided by the TSN network. Unlike previous works that focus on 802.1Qbv, the specific TSN mechanism used to acquire this information is transparent to the proposed scheme so long as TSCAI and 5QI can be derived. Their results reflect an increase in the number of TSN flows that can be served whilst meeting their latency requirements satisfactorily. Finally, authors in [247], tackle this resource allocation challenge in an uplink 5G-TSN network by instead introducing an adaptive SPS algorithm. TSN flows are first divided into clusters according to QoS characteristics and turned into aggregate flows according to this classification. RB preallocation is performed for time-sensitive traffic, whereas non-critical sporadic traffic uses dynamically allocated resources. The designed algorithm focuses on optimizing the preallocation of resources according 292 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 to flow requirements in order to maximize system resource utilization. 5) Implementation of Reliability Mechanisms over 5G Networks: End-to-end reliability and robustness is predominately improved through the provision of redundant data paths that leverage 5G NR and URLLC features, and the FRER mechanism of the 802.1CB TSN standard. In [236], in order to duplicate the transmitted data streams over 5G, the use of redundant links between base stations and UPFs leveraging URLLC is briefly introduced; however, authors do not delve into the implementation details. Spatial diversity via CoMP is thoroughly evaluated in [235] as well for providing reliability without relying on packet retransmission. Authors evaluate CoMP technology for the provision of wireless URLLC communications required to support TSN-over-5G in industrial factory automation scenarios. Additionally, [238] targets a reliability improvement in the wireless data transmissions using CoMP based on Multi-TRP operation. Authors of [237] propose the use of FRER to provide high reliability in TSN- over-5G communications spanning a factory TSN domain and an edge cloud domain. They analyze different architectural approaches to deploy FRER functionality in order to provide end-to-end redundancy between an industrial end device connected to a TSN network, and an application instance in the edge cloud. For this scenario, they propose the deployment of FRER functionality of the edge cloud domain side in three possible ways; in a virtualized manner packaged together with the TSN Talker/Listerner functions for the applications; virtualized, but in a separate entity to Talker/Listener functions; or implemented in a SmartNIC. This work is further surveyed in Section V-G in relation to virtualization of TSN functionalities. In [248], authors propose experiments over actual smart manufacturing testbeds to empirically evaluate the benefits of using FRER over wireless 5G. In this case, FRER is used to transmit replicated traffic over two independent 5G user plane paths using two different UEs. They highlight several benefits that FRER can provide to wireless networks such as 5G: protection against network node failures; protection against UE failures or UE performance variations; and robustness against wireless links with latency variations. Authors focus on evaluating end-to-end latency robustness improvements introduced by FRER when 5G URLLC functionality is not available, where they find that guaranteed latency can be reduced by up to several milliseconds. Likewise, in [249], authors also investigate FRER as a solution for the lack of URLLC availability in commercial products in order to improve worstcase latency in addition to significantly increasing reliability. Their performance evaluation in a production environment shows that their redundant path implementation using FRER provides higher reliability against outages. Additionally, it improves the overall latency, as it mitigates the effect of the additional delay introduced by HARQ retransmissions in each path. 6) Traffic Steering Optimization in User Equipment: With aforementioned works focusing on the architecture integration and evaluation within the RAN and Core Network levels of the 5GS bridge, in [250], authors focus instead on the delay minimization of traffic steering in edge compute devices that include the UE component of the 5GS bridge. Authors propose the use of XDP programming to reduce the latency introduced by the network stack in the device when steering Ethernet PDU and gPTP traffic to the TSN NIC or to userspace applications, and to the DS-TT, respectively. Their proposed algorithm filters the traffic according to destination MAC address. Packets with a gPTP multicast destination address are redirected to the DS-TT in userspace bypassing the kernel network stack of the device. Authors conduct an experimental evaluation of their XDP algorithm, arriving to an steering latency of PDU traffic 100 times lower than with a conventional layer 2 software bridge. Summary and Lessons Learned: •The integration of 5G and TSN is empowered by 5G architectural enabling technologies and URLLC:As demonstrated by several preliminary TSN and 5G integration studies in industrial environments, the possibility of introducing a 5G network into an industrial TSN deployment begins with the adoption of architectural enablers like private 5G networks and edge computing. Architecture-level technologies highlighted in the literature include the availability of several 5G NPN configurations, redundant transmissions based on Dual Connectivity, Multi-TRP, and/or URLLC, the support for traffic steering to local data networks, and 5G LAN technology to support the grouping of devices that can communicate with each other as though in a Local Area Network. As such, 5G-TSN is expected to thrive in this kind of technology ecosystem to satisfy the requirements of flexible and innovative industrial applications. •The characterization of TSN traffic requirements in 5G QoS language is a fundamental requirement: As the 5GS does not understand configuration parameters of specific TSN functionalities, ensuring that it is aware of the characteristics of TSN flows traversing it is fundamental. 3GPP establishes guidelines for estimating the bridge delay and for deriving 5G QoS parameters like PDB, GFBR, and MDBV, but implementation details are not thoroughly elaborated, and, additionally, depend on the end-to-end communication use case. Thus, even when not accounting for factors such as bridge delay estimation accuracy or the representation of traffic periodicity with 5G mechanisms, the mapping of QoS parameters between the wired TSN network and the 5GS requires careful tinkering and, possibly, the use of 5G QoS configuration, like 5QI, that is tailored to the specific use case rather than depending on catch-all standardized 5G QoS profiles. •The 5G-TSN integration must also meet stringent performance requirements: While 3GPP standards detail the architecture-level integration of 5G and TSN, achieving a performant integration requires not only the support, but also the optimal execution, of 5G NR and URLLC features. With some industrial applications demanding a delivery time of even less than 1 ms, bounding the delay contribution of all elements across the end-to- end path is critical. In this regard, one of the main challenges is the latency and jitter minimization over the SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 293 wireless 5G medium (that is, across the 5G RAN) in order to guarantee the bounded latency requirements of the TSN flows. Many studies highlight that fitting TSN periodicity, such as dictated by a 802.1Qbv schedule, into the 5G scheduling mechanisms based on OFDM, entails a matching problem that must be carefully addressed. Among these issues, the unpredictable conditions of the wireless medium, the potential mismatch between traffic periodicity values and 5G transmission intervals, and potential conflicts during the scheduling in the 5GS of multiple TSN flows with different periodicities, are highlighted in the literature. •Configurations based on 5G redundant user plane transmission can be effectively integrated with TSN FRER to improve resilience and worst-case latency: In regards to specifically achieving the required reliability and robustness of the architecture, one research line investigates the usage of CoMP options and Multi-TRP for spatial diversity, while another research line analyzes the setup of redundant user plane paths through NR-DC or through redundant tunnels between RAN and UPF. The former is more closely linked to increasing capacity and transmission rate, whereas the latter is more pertinent to improve reliability as it also provides the architectural support to implement FRER. In this regard, FRER is found to be effective at providing additional robustness in communication over 5GS bridges, with the most critical path generally considered to be the link between UE and gNB, due to its wireless nature. This kind of redundancy is found to improve worst-case latencies as a consequence of duplicating transmissions through redundant wireless paths and discarding redundant copies at the egress point. The works addressing these configurations mainly propose the duplication of UEs, rather than the duplication of only the number of PDU sessions that a single UE establishes through redundant RANs and UPFs. This provides resilience against UE failure as well, but implies duplication of the UE resources connected to the TSN end stations. •The optimization of TSN scheduling and performance over 5G is an open research question that progresses in harmony with the standardization of new 5G and URLLC capabilities: The literature showcases a variety of proposals to address the resource allocation and scheduling challenge in 5G. The most basic approach involves the bare use of preconfigured transmission slots with CG and SPS, which provides some guarantees through the preallocation of resources for TSN flows in comparison with the default dynamic scheduling. However, issues such as the ones pointed out in the previous item need to be more effectively addressed. For the periodicity mismatch problem, one work proposes the use of a scheduling scheme with alternating CGs as to adapt to flow dynamicity. Another work looks higher in the 5G NR stack, evaluating the use of dynamic LCP for the time-variant prioritization of logical channels in the RLC layer to support TAS-based TSN. They are promising approaches for dynamic uplink TSN flows and for TAS GCLs that accommodate heterogeneous traffic, respectively. For the mitigation of the resource allocation conflict problem, one work proposes a scheduling scheme based on the provisioning of several CGs for each TSN flow to ensure that RBs are not allocated to more than one flow. Another work proposes an adaptive SPS algorithm over aggregated TSN flows that treats time-sensitive and best effort traffic differently in order to optimize the usage of resources. These examples also highlight that there can be different optimization goals in the design of the 5G-TSN network, ranging from TSN flow admission to 5G resource usage, which expands even more the research space. Finally, another paper looks at improving traffic steering delay in the UE through userspace packet processing in the form of XDP. This contribution is complementary to the other ones in the sense that the 5G- TSN integration shall be approached from an end-to-end perspective, including the 5G RAN, the 5G UEs, and the 5G TN, the latter of which is surveyed in Section V-E. C. QoS Configuration and Mapping Strategies for TSN Flows Over 5G This subsection gathers research works primarily concerned with architecture and procedural aspects of the 5GS configuration and mapping for TSN traffic, as opposed to those schemes focused on performance characterization, which were surveyed in Section V-B. Still, it should be noted that the distinction in some cases may be narrow. A taxonomy for this subsection is proposed in Table XV. Several approaches for mapping TSN and 5G QoS can be identified, whereas other proposals can be classified under network optimization at a broader end- to-end network management level and often employ learning schemes, such as Reinforced Learning (RL)-based. 1) QoS Mapping and Translation between TSN and 5G: Authors of [34] delve into how traffic class, frame preemption, and 5G QoS configuration can be jointly carried out in a 5G-TSN integrated network, according to IEEE TSN and 3GPP standards. Based on the mechanisms and capabilities provided by the 5GS, authors suggest parameter mappings and constraints for configuring 5G QoS Profiles. In [251], authors design and implement, via simulation, a translator tool between TSN and 5G traffic that is able to map parameters between the two technologies according to the 3GPP specifications. In their work, they describe their implementation of the translation workflows in the simulator, and finally evaluate the result using an industrial automotive use case. They find that their tool can be useful for evaluating the performance of 5G-TSN network configurations, particularly in regards to channel propagation delays. Authors in [252] propose and evaluate a QoS-aware algorithm that systematically maps traffic flows between the TSN and 5G network domains. Their algorithm uses as input the requirements of applications with specific constraints that include deadline, jitter, and bandwidth. Each application is mapped to a 5G Resource Type according to the binary requirements of these three types of constraints. The numerical values of bandwidth and deadline, in addition to the application’s PER requirement, 294 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE XIV TAXONOMY RELATED TO THE QOSCONFIGURATION AND MAPPING STRATEGIES FOR TSN FLOWS OVER 5G TABLE XV TAXONOMY OF TIME SYNCHRONIZATION SCHEMES IN 5G-TSN NETWORKS are used to calculate the mapping to a standardized 5G QFI for the given Resource Type. In [247], authors propose a dynamic and load-aware QoS mapping method for 5G-TSN networks. A clustering-based algorithm that aggregates flows with similar QoS characteristics is designed to reduce QoS mapping complexity whilst ensuring the consistency of QoS classes. Additionally, authors introduce an algorithm that maps flows to several 5QIs considering network load to improve resource allocation efficiency. 2) Injection of Scheduling Configuration in 5G-TSN Networks: Both TAS and PSFP mechanisms are explored in [253], as authors propose an heterogeneous architecture integrating 5G and TSN. The analyzed industrial scenario comprises a set of wireless field devices that connect to a TSN GW towards a wired network via 5G. On the one hand, authors design a scheduling mechanism that can reserve 5G resources in advance using SPS to omit signaling delays for the highestpriority sporadic emergency events. On the other hand, to offset the jitter introduced by the 5G network, they propose the use of PSFP at the ingress of the wired TSN network. Authors do so taking into account the delay experienced by the data in the 5G network together with its end-to-end delay constraints to assign a suitable priority. As such, unlike other previously covered works, they concentrate on configuration injection in the direction from 5G to TSN. In [254], authors study a similar industrial scenario extending the configuration challenge to the NFV domain as well, as they formulate a mapping and scheduling problem of 5G-TSN resources to incoming VNFs. Admitted VNFs are deployed in TSN GWs located in an edge layer that also contains 5G base stations. The communication flows occur between VNFs and devices in a field layer that connect via 5G to the TSN GWs. The arrival time of VNFs is mapped to 5G slots. Apart from GCL computation for the VNF streams, authors also apply preemption-based scheduling for the arriving VNFs accounting for their priority. Authors of [255] address the challenge of granularly mapping traffic in a converged 5G-TSN network, primarily focusing on an architectural model to inject the configuration. Authors propose to move the mapping functionality from the TSN AF in the 5GS—as defined in 3GPP standards—to a centralized CNC, taking advantage of its global network knowledge. In the proposed architecture, which contains a data plane comprised of one or more 5GS and/or non-TSN bridges, the CNC calculates the 5QIs in each 5GS bridge for the application flows between end stations. 5QIs are derived from to deadline, jitter, bandwidth, and PER requirements received from the CUC. Finally, the CNC communicates the 5QIs to the corresponding PCFs and jointly configures 5GS and non-5GS bridges in the network. 3) Joint configuration and scheduling over 5G and TSN resources: In [256], authors pursue the optimization of end-to- end scheduling in a 5G-TSN network carrying heterogeneous traffic through a joint configuration approach. Firstly, authors derive global and technology-specific constraints that the scheduling strategy must adhere to. This includes, for example, 5G transmission constraints according to 5G NR TTIs and RBs. TSN constraints consider scheduling based on 802.1Qbv TAS. Based on these constraints, authors tackle end-to-end schedule optimization across the different domains, focusing on resource utilization and support of BE traffic, given that the gating mechanisms of TAS guarantee time-bounded latency requirements. Some recent research works adopt learning methods for the configuration of 5G-TSN networks according to constraints and optimization goals. In [257] and [258], SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 295 authors aim for a dynamic and zero-touch configuration of a 5G NPN featuring 5G-TSN integration based on the ZSM reference model. The survey of these works is carried out in Section V-H. Notably, their proposed smart controller is designed to act as a TSN AF that dynamically configures each 5G QoS flow according to compliance with expected KPIs. In [259], authors present an algorithm based on RL— the Q-learning technique specifically—and on particle swarm optimization that aims to iteratively arrive to the configuration that maximizes the time-sensitive flow admission ratio in 5G- TSN networks. Authors use CQF as the TSN traffic shaping mechanism and establish 5GS latency constraints, end-to-end latency constraints, and CQF time slot constraints. Finally, authors conduct an evaluation of their proposed algorithm through the simulation of 5G-TSN network topologies with specific 5G numerology parameters and assuming semi-static scheduling. Likewise, in [260], authors investigate the usage of CQF in 5G-TSN networks. Authors consider constraints related to CQF and to its mapping to 5G resources, including flow periodicity and latency in the 5GS in regards to satisfying the scheduling rules of CQF time intervals. Based on these constraints, authors formulate flow schedulability and resource utilization as joint optimization objectives. To address the formulated problem, authors propose a DRL-based algorithm designed to learn the relationship between different time intervals and the utilization of 5G sub-carriers, and finally conduct a simulation to evaluate its performance. In regards to the use of learning methods, it is worth pointing out previously surveyed work [253] for its contribution of a risk-sensitive learning algorithm aimed at reducing the worst-case delay in PSFP-based 5G-TSN networks by adjusting reserved RBs and PSFP priorities. Summary and Lessons Learned: •There are different possible TSN mechanisms to extend TSN to the 5G domain: When a TSN network integrates 5GS bridges, the literature proposes 802.1Qbv TAS, 802.1Qci PSFP, and 802.1Qch CQF as viable TSN mechanism choices; nevertheless, further observations can be made. The only proposal considering PSFP employs it in order to dynamically manage the injection of TSN flows coming from a 5GS bridge into the entry point of a wired TSN network. In this way, the policing functions of PSFP are found useful to balance the assigned queue priorities against the jitter introduced by the 5G network for each flow. Overall, TAS is the most popular choice. When a 5GS acts as a virtual TSN bridge, the concept of GCL can be extended naturally to the 5G domain for the joint configuration of both domains. The main challenges of this approach regarding 5G scheduling were discussed in Section V-B. In spite of 3GPP standardization suggesting the use of PSFP’s stream identification and gate-based policing configuration to derive 5G QoS for TSN flows, several works have preferred to use characteristics from 802.1Qbv to map TSN QoS to 5G QoS. However, the QoS characteristics that can be derived from a GCL are potentially less granular, so, for example, one proposal of a QoS mapping algorithm has recoursed to explicitly providing additional application requirements. Hence, it could be worthwhile to consider the use of PSFP for accurate QoS and TSCAI derivation, even if a TAS GCL is also used. Meanwhile, the proposals for using CQF support it for its simplification of the scheduling in comparison to TAS, avoiding complex GCL computation, and allowing the accommodation of several flows in the same time slot. With CQF, the imposed TSN scheduling constraint is simplified to finding a time slot length that satisfies the residence time in the 5GS bridge. However, as the articles that propose CQF focus on maximizing flow admission, more research is pertinent regarding if the end-to-end flow latency attained with CQF is adequate enough to satisfy stringent industrial communication requirements. Indeed, several sources, including the TSNIA as seen in Section IV, suggest that isochronous and cyclic-synchronous traffic types require the use of TAS. Still, given that CQF guarantees a fixed worst-case delay so long as constraints are met, its use in 5G-TSN networks is promising in the cases where the time slot length satisfies the end-to-end latency requirements of the involved traffic. •Cognitive approaches can be adopted to optimize the configuration of 5G-TSN networks: Several works formulate the configuration of the 5G-TSN as an optimization problem with end-to-end constraints related to latency, periodicity, and 5G numerology, in which the number of scheduled flows, either alone or jointly with the resource usage efficiency, aims to be maximized. Other proposals convey the problem inversely, taking a 5G-TSN network with configured QoS flows, and dynamically applying closed-loop reconfiguration in order to optimize the network according to the KPIs that characterize the QoS flows. In all cases, authors find RL techniques useful for this purpose, as they constitute AI/ML techniques suited to learn by observing the environment and to find the most adequate configuration of several parameters that maximizes the given goals. RL variants found in the various algorithms proposed in the literature include DRL, automata learning, risk-sensitive learning, and Q- learning. Furthermore, works that propose a closed-loop management of the 5G-TSN network propose approaches based on the ZSM architecture, as it provides a framework that integrates monitoring, learning, prediction, and reconfiguration capabilities. More research can be expected going forward in the scope of applying learning mechanisms to the design and optimization of 5G-TSN networks, especially as most of the current proposals are still in theoretical or simulation stage. D. Evaluation of Time Synchronization in 5G-TSN Networks Highly accurate time synchronization between the 5G and TSN clock domains is an essential requirement to allow for time-sensitive industrial control communications across a 5GS bridge. Research works addressing time synchronization in 5G-TSN networks are surveyed in this subsection. As seen in Table XVI, contributions in the scope of time synchronization 296 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE XVI TAXONOMY RELATED TO THE EVALUATION OF TIME SYNCHRONIZATION IN 5G-TSN NETWORKS focus on the evaluation and optimization of synchronization accuracy employing PDC and/or clock drift mitigation schemes. A variety of mechanisms, strategies, and involved factors can be identified. 1) Analysis of Over-the-air Time Synchronization: Prior to the consolidation of 3GPP Release 16, a mechanism for 802.1AS-based time synchronization with mobile networks is presented in [261], that is also compatible with 4G. They leverage the fact that inner synchronization mechanisms between base station and UEs are already required in 4G and 5G networks. To synchronize each UE to TSN time, authors propose the calculation of a time offset between the timing information received in the base station and the time of a Reference UE that is wiredly synchronized to the TSN network using 802.1AS. This offset can be distributed to the UEs via a multicast OPC UA service that the UEs subscribe to. They achieve a 0.55 ms synchronization accuracy over a 4G testbed. A joint comparative analysis of 802.1AS over Wi-Fi and over 5G is conducted in [262], seeking performance results and the identification of factors that impact such performance. They highlight the asymmetric and medium sharing nature of wireless links and the lack of hardware timestamping support in wireless cards as particular challenges for delay estimation. Most specifically for 5G, they remark that the 3GPP architecture based on TSN translators can bring critical performance improvements. 2) Evaluation of Time Synchronization Based on 3GPP Architecture: Several works can be found in the literature that make use of the TSN time synchronization procedures standardized from Release 16 onwards. Precisely, in [263], authors perform an exhaustive analysis of 802.1AS time synchronization for converged wired and wireless networks, targeting 5G-TSN integrated networks. Authors model the behavior of 802.1AS mechanisms and investigate synchronization accuracy and worst-case parameters taking into account requirements specified by the IEEE 802.1AS standard itself, the IEEE/IEC TSN-IA profile, and 3GPP. Their investigation identifies limited timestamp resolution, physical layer jitter, and concatenation of time-aware systems as the primary sources of time synchronization inaccuracy. In [264], authors carry out an evaluation of the TSN time synchronization mechanism introduced in Release 16, using the 802.1AS time-aware system model supported in this release of 3GPP. They evaluate the time synchronization accuracy in the presence of potential sources of inaccuracy, such as clock drifts and different air-interface errors associated with reference time indication. They specifically evaluate the conditions that need to be met to achieve below 1 µs accuracy. In regards to scalability, authors also calculate that a maximum of 8 TSN domains can be supported. Authors of [265] consider a synchronization scenario for 5G-TSN networks that employs neither downlink synchronization from the TSN network, nor uplink synchronization from the UE. Instead, authors present and evaluate the use of a shared GM clock for both 5G and TSN networks that is located within the 5GS bridge at the gNB and has access to a GPS source of time. Authors identify challenges related to the mapping between gPTP and 5G RAN synchronization mechanisms and to the execution of the BMC algorithm, and propose two strategies to address them: a shallow integration that maintains the boundaries in NW-TT and DS-TT aligned with 3GPP standards, and a more transparent deep integration using a modified 5GS bridge that shifts the synchronization boundaries to gNB and UE. 3) Evaluation of Frequency Offset and Clock Drift: It has been found that the evaluation of frequency offsets and resulting clock drifts is one of the main research areas within the scope of time synchronization in 5G-TSN networks according to 3GPP standards. Indeed, aforementioned work [264] analyzes the effects of clock drift over the 5G air interface between UE and gNB. Some takeaways are: higher SCS is more robust to cumulative synchronization error; reference time indication granularity must be in the order of 10 ns; and frequent time synchronization is key against clock drift. in [266], authors conduct a performance evaluation of the Release 16 TSN time synchronization scheme, focusing specifically on estimating the impact of the NW-TT, DS-TT, and 5G GM rate ratios, i.e., frequency offsets. Their measurements conclude that the main contribution to time synchronization error is dependant on the frequency offset between 5G GM and DS-TT, which involves the air interface delay. To mitigate these errors, authors suggest a method that takes advantage of the fact that the 5G GM is distributed to the UE in order to compute the rate ratio between DS-TT and 5G GM. Meanwhile, the authors in [267] adopt Release 17 standards SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 297 as they evaluate time synchronization performance, modeling the 5GS as a Transparent Clock, and carrying out a device-to- device synchronization analysis. Authors present a simulated scenario of two cooperative mobile robots and execute it with different parametrizations. Their results highlight that even small frequency offsets between NW-TT and DS-TT can have a significant impact on the calculation of the residence time inside the 5GS bridge. They conclude that, ultimately, this can result in significant synchronization errors. 4) Propagation Delay Estimation and Compensation using TA-based or RTT-based methods: Compared to wired networks, the link propagation delay in a wireless network is much more challenging to estimate, due to variable transmission characteristics that also depend on UE position and movement. As such, some research works specifically concentrate on this requirement of 5G-TSN time synchronization. Reference [268] presents an analytical evaluation of time synchronization error sources across the end-to-end path of a 5GS in the context of integrated 5G-TSN networks. Authors analyze how a 5GS modeled as a 802.1AS time-aware system can be designed to support a 900 ns time synchronization error budget. Authors split their analysis between transport and radio network paths. For the former, they conclude that ITU-T time synchronization standards [269] are adequate for industrial network scenarios. For the latter, they analyze RTT- based and TA-based propagation delay estimation methods and find the former to achieve a lower time error. Likewise, [35] examines these two methods for PDC according to the time synchronization error budget established in Release 17, taking into account that UE-UE communications require a twice as tight unidirectional synchronization budget. Meanwhile, [270] proposes and evaluates a procedure based on UE localization estimation using TA in order to achieve a more accurate PDC. More specifically, their proposed mechanism is based on obtaining TA measurements from multiple gNBs rather than just from the one serving the UE. This results in enhanced TA granularity, which leads to a more precise estimation of the UE localization and its distance to the gNB. 5) Propagation Delay Estimation and Compensation using other methods: [271] regards the 5GS as a Transparent Clock and adopts a delay compensation method based on data packet relay where the 5G network is only responsible for the forwarding of timestamp messages, offloading the complexity of translating between time domains. The residence time of TSN timing messages forwarded through the 5GS in a PDU session is estimated using a clock domain compensation algorithm that uses timestamps in the NW-TT and DS-TT. A delay compensation method based on Time Difference of Arrival (TDoA) is briefly presented in poster [272]. Authors present an algorithm that focuses on overcoming the non-line-of-sight error. Simulating a network with 7 gNBs, authors obtain a 8.9% time synchronization accuracy improvement compared to the RTT- based method. Finally, in [273], authors expand the problem formulation to a scenario with multiple 5GS bridges that accumulate synchronization errors, eyeing the use of 5G-TSN in multi-field manufacturing systems scenarios. Timestamp compensation between each 5GS bridge is performed with a method based on cumulative rate ratios. Authors evaluate the impact of SCS in the synchronization error, and, via simulation, arrive to 5G-TSN clock synchronization accuracy results for different SCS values and number of 5GS bridges in the network. Based on these results, authors propose an SCS optimization scheme that integrates synchronization accuracy, and bandwidth utilization determined by SCS. Summary and Lessons Learned: •Very accurate end-to-end time synchronization is required in the 5GS to support wireless TSN communications: Many applications using TSN require highly accurate time synchronization in the sub-microsecond range. Furthermore, with the introduction of time-sensitive UE-UE communications in Release 17, bidirectional synchronization involving two UE-UPF paths in total is required. The literature identifies the minimization of the contribution of the 5GS to the end-to-end synchronization error as a key challenge. It has been found that the main contributions in this regard can be classified in two complementary categories: the estimation of the propagation delay in the 5GS and methods for its compensation, and the mitigation of the difference in clock frequency offsets—and resulting clock drifts—in the 5GS that introduces errors in the residence time computation. •The interval between time synchronization events is a decisive factor to the error caused by clock drifts inside the 5GS: The literature generally agrees that the best solution to prevent drifting between the clocks at the ingress and egress of the 5GS, or between the clocks of the UE and the gNB when focusing on the wireless path, is to execute frequent re-synchronization procedures, as otherwise even small frequency drifts can have a significant impact on the residence time calculation error. Some research works analyze time synchronization intervals in the ranges between tens of milliseconds to hundreds of milliseconds, and the results indicate that the clock frequency offset increases linearly with the synchronization interval, regardless of the SCS in use. On the other hand, but to a lesser extent, a higher SCS has been found to lead to lower frequency offset between clocks for the same interval between time synchronization events. Complementarily, the frequency offset between TSN GM and 5G GM can be estimated in order to mitigate it. The DS-TT could directly make use of the 5G GM time acquired by the UE—which can be made most accurate through frequent re-synchronization events as aforementioned—and calculate its rate ratio with the UE, and the same for NW-TT and UPF. The gPTP procedure carried out through the 5GS bridge is maintained in order to calculate the residence time in the 5GS, except it instead leverages these rate ratio calculations for the conversion from 5G to TSN time. Hence, both minimization of the drifting between the clocks in the 5GS, and the derivation of the rate ratio between 5G and TSN time domains, shall be taken into consideration to arrive to the most satisfactory result. •The optimization of the propagation delay estimation and compensation mechanisms is critical for the time 298 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 TABLE XVII TAXONOMY RELATED TO THE USAGE OF TSN IN 5G TRANSPORT NETWORKS synchronization accuracy: The two main mechanisms to estimate and compensate the propagation delay between UE and gNB studied by the research community are the TA-based method and the RTT-based method, aligned with 3GPP standards. Overall, the results are more favourable for the latter, which was specifically introduced in 3GPP Release 17 for TSC. Furthermore, some authors agree that the RTT-based method is more promising due to having more room for optimizations. A theoretical study reports an improvement from 514 ns to 188 ns from TA-based to RTT-based estimation. Another study presents several samples of between 276 ns and 536 ms depending on the SCS, with higher SCS values achieving better accuracy, but, in this case, the estimation method used is not specified. On the other hand, another work proposes a TA-based approach that, by receiving measurements from more than one gNB, accuracy can be improved by up to around 55 ns. In another brief study that involves multiple gNBs, the use of TDoA, normally used for positioning, yields promising results, but more research is necessary, especially with a lower amount of gNBs. Meanwhile, a method that treats the 5GS as a Transparent Clock and is only responsible for the forwarding of timestamped messages in PDU sessions is found effective at reducing the complexity of TSN translators, but results indicate synchronization accuracy in the microsecond level. It is noteworthy that while there are several factors that contribute to the estimation accuracy (e.g., the PDC method, the SCS, the UE mobility including its speed, the network size, the granularity of the reported time reference, whether it is the UE or the gNB who performs the PDC, etc.), the surveyed works do not always analyze or report all of these factors, so more research is arguably necessary. An important consideration is the synchronization error budget requirements established in 3GPP Releases 16 and 17—900 ns, but 450 ns for the UE-UE scenario— , which may dictate the validity of different methods and configurations. Another open question consists on determining the UE-gNB distance threshold, perhaps accounting for other air-interface factors, that warrants the use of PDC, as, for low distances, the propagation delay could be considered negligible. E. TSN for 5G Transport Networks This subsection presents research works that analyze the usage of TSN functionalities in 5G TNs, and concentrates specifically on the contributions provided by the use of TSN. Like the former subsections, this subsection surveys related works published from 2018 onwards, aligning in this case with the publication date of 802.1CM. As reflected by Table XVII, the literature showcases research works that tackle, among other aspects, latency and jitter minimization in the fronthaul, and TN convergence by bringing TSN to the backhaul and crosshaul as well. 1) Delay and Jitter Minimization of Fronthaul Traffic: In [274], authors first analyze the delay requirements in eCPRI-based 5G fronthaul networks and what are the components that contribute to end-to-end delay between BBU and RRH. Then, authors analyze TSN functionality and the appropriate design of fronthaul networks using such functionality in order to comply to such requirements. Authors focus on worst-case delay and packet jitter minimization using 802.1Qbu-based preemption and 802.1Qbv. In [275], authors evaluate the performance of a Ethernet TSN network enabled by 802.1Qbu and 802.1Qbv for the transport of fronthaul traffic. Authors benchmark the latency performance of their proposal against strict priority and round-robin scheduling. Their results over several experiments demonstrate that the either of TSN standards is adequate for protecting the highpriority fronthaul flows when the network is overloaded by low-priority traffic. An individual comparison between 802.1Qbu and 802.1Qbv also reveals that the former achieves the lowest average latency and jitter values, whereas the latter achieves the lowest maximum jitter value while still providing below 100 µs latency to comply with eCPRI. 2) TSN for Sporadic Backhaul Traffic: Authors of [276] propose the adoption of the ATS standardized in 802.1Qcr for 5G backhaul networks. In asynchronous TSN, authors encounter an optimal trade-off between latency guarantees and network scalability for the sporadic backhaul traffic with real-time constraints, avoiding the network-wide synchronization and coordination needs of synchronous traffic shaping. Authors address the flow allocation problem in ATS-based backhaul networks, which involves optimal path selection and configuration for each traffic class according to delay, jitter, frame loss, and reliability constraints. 3) Enabling a Converged Transport Network: Some research works propose the use of TSN as a technology to enable convergence in the crosshaul, which refers to the continuum of fronthaul, midhaul, and backhaul that comprises the TN. Reference [277] describes an Ethernet-based 5G TN that incorporates fronthaul express traffic and lower priority midhaul/backhaul traffic over the same Ethernet infrastructure. Authors integrate a deterministic scheduler called SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 299 FUSION [278], 802.1CM-based preemption, and IEEE 1588 time synchronization to enable the aggregation and convergence of multiple fronthaul, midhaul and backhaul streams in the same network. Compared to strict priority and roundrobin schedulers, authors achieve fronthaul delay and PDV that is not impacted in presence of an increasing number of midhaul/backhaul streams. In [279] and in subsequent work [280], authors demonstrate a 5G platform that can simultaneously accommodate periodically generated fronthaul traffic and massive IoT backhaul traffic. This is enabled by the use of time aware shaping and adaptive mobile fronthaul compression. For the former, authors propose the use of a variation of TAS in the aggregation points of fronthaul and IoT traffic that does not require time synchronization. Instead, their mechanism detects the transmission and maximum length of fronthaul bursts by observing real-time network statistics, then autonomously sets the cycle time and a time slot for the latency-sensitive fronthaul traffic. This time slot can additionally be shrunk within each cycle by detecting when each fronthaul burst ends, to optimize the transmission of IoT traffic. In their platform evaluation in [280], authors achieve a76µs fronthaul latency over a 10 km TN. Lastly, [149] also deals with the incorporation of TSN in the whole 5G transport network, but as it focuses on the enablement of end-to-end TSN network slice orchestration, it is mainly surveyed in Section V-H. The architecture proposed in this work contemplates the use of TSN for the crosshaul including within the TN of a 5GS modeled as a TSN bridge. 4) Integration of TSN into the O-RAN Architecture: Authors of [50] review the requirements and options towards a TSN-enabled O-RAN architecture. Authors highlight the open nature and crosshaul convergence provided by O-RAN, while focusing the study on the fronthaul—OpenFronthaul in O-RAN, located between the O-RU and the O-DU—due to being the interface that contains the traffic with the most restrictive requirements in terms of latency. Authors propose placement options of the TSN control plane entities within the O-RAN components, and analyze possible mechanisms to achieve the tight time synchronization required between O-RUs and O-DUs. Authors also analyze the TSN standards most appropriate for its usage in O-RAN, describing how the 802.1CM profile could be leveraged, and highlighting the joint usage of 802.1Qbv and 802.1Qbu in order to provide deterministic delay whilst optimizing the available capacity for the different kinds of traffic in the TN. Summary and Lessons Learned: •TSN mechanisms are effective for minimizing delay and jitter in fronthaul networks and achieving the latency budgets specified in eCPRI: The literature showcases that the usage of 802.1Qbu-based frame preemption and of 802.1Qbv for the aggregation of multiple fronthaul streams over a GCL, leads to better performance in regards to delay and jitter compared to the mere use of strict priorities or to the use of other mechanisms. This quality is particularly relevant for its direct impact on the size of the fronthaul link in terms of nodes, and on the length of the fiber links that can be used, without compromising the 100 µs latency budget established in the eCPRI standard and the 802.1CM profile for the most stringent fronthaul traffic. To achieve the time synchronization required by 802.1Qbv, either IEEE 1588 PTP or 802.1AS gPTP is proposed by different works. In regards to 802.1Qbv and 802.1Qbu, the literature agrees that the best performance is achieved when both mechanisms are jointly used. However, the use of only one of the two mechanisms may suffice to comply with the 100 µs latency budget depending on the network characteristics. On the other hand, the effectiveness of other mechanisms such as 802.1Qch or 802.1Qci is unclear, as research works available so far only explore those two standards along with simply using strict priority. •Adopting TSN for the whole TN enables network convergence: While the most stringent time-sensitive requirements across the TN are located in the fronthaul link, one of the main strengths of TSN lies in its enablement of a converged network infrastructure that is capable of simultaneously treating traffic with mixed characteristics in the most appropriate manner. Several research works demonstrate that TSN can be used to converge and aggregate the fronthaul, midhaul, and backhaul traffic. 802.1Qbv TAS-based mechanisms are the most popular choices for this purpose, managing the priority of each kind of traffic and the length of the GCL schedule and time slots accordingly, including one proposal that adopts traffic monitoring and learning methods to optimize TAS configuration. Furthermore, a converged crosshaul based on TSN can be applied to the O-RAN architecture as introduced by one of the research papers surveyed in this section, thus enabling a deterministic TN with open interfaces. The O-RAN architecture provides softwaredefined controllers to accommodate the deployment of the TSN control plane entities as applications, while standardizing well-defined interfaces along the crosshaul where TSN mechanisms can be applied. However, this is considered a novel research line and more studies are deemed appropriate in this direction. F. Software-Defined Time-Sensitive Networks Research regarding the convergence of SDN and TSN architectures is extensive in the literature. In general, these initiatives leverage the centralized configuration model standardized in TSN and their alignment with the control plane centralization pillar of SDN. As explained in Section III-E, in the centralized configuration models, the CNC entity is responsible for the configuration of the whole TSN network based on stream requirements that come from the network devices, or, in the fully centralized model specifically, from the CUC entity. Due to its flexibility, SDN has been utilized in TSN networks to serve multiple, often complementary, purposes. The taxonomy proposed for this subsection is shown in Table XVIII. The primary scope is the software-defined network configuration, where a variety of design approaches and techniques can be identified. Additionally, SDN-aligned schemes that tackle the network self-configuration and 306 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 •The resource allocation and VNF placement challenge of NFV is augmented in TSN networks: The VNF placement problem in NFV has been studied in the literature, aiming to arrive at optimal strategies to admit incoming VNFs by weighing resource utilization costs of computing and networking resources whilst adhering to specific VNF constraints. When NFV meets a TSN network where VNFs have varying requirements of timeliness and determinism, a new dimension is introduced to this problem, as the availability and optimization of TSN resources has to be taken into consideration as well. For example, an implementation based on 802.1Qbv requires that all communication streams between VNFs are assigned a traffic class and time slot that satisfies their requirements. In this regard, another matter to be addressed is that, by default, NFV considers flows between VNFs as bidirectional and virtual, whereas TSN, by definition, defines unidirectional streams that are mapped to specific physical paths across the network. Therefore, the mapping is not straightforward. •Userspace networking techniques are promising to minimize the latency of the networking stack: Several works converge to the use of kernel-bypassing techniques for accelerated packet processing in order to deliver the latency and determinism requirements of TSN applications. The most studied technology for this purpose is DPDK, often used together with OVS. OVS-DPDK provides virtual networking where applications in the userspace can receive frames from the NIC without context switching between spaces and without CPU interrupts. As such, some works that address this topic report significant latency improvements when adopting DPDK over kernel space networking for the communication path of VMs and containers alike. OVS-DPDK has also been proposed for the virtualization of a high-performance TAS bridge, an approach that can be considered synergistic with approaches to encapsulate PSFP surveyed in Section V-F using hardware acceleration through specialized P4 switches. •Virtualization provides a converged substrate to flexibly support 5G functions along with industrial control functions in 5G-TSN environments: As seen in Section IV, the design of 5G networks involves the use of servicebased and microservices-based architectures that thrive with the use of virtualization. On the other hand, several works surveyed in this section fruitfully push the use of virtualization in industrial control applications with strict requirements in terms of deterministic communications and reliability. This manifests the viability of deployments that can not only converge the communication infrastructure through TSN, but also the computing infrastructure through virtualization. A MEC deployment close to the production lines could, for example, simultaneously allocate a private 5G control plane including use case-specific NFs, and industrial control functions through vPLCs that support TSN-based C2C, C2D, and D2Cmp communications, all in a cost-efficient and flexible manner. While the literature reflects that the performance results achieved so far might not be suitable for the processes with the tightest latency-sensitive control loops, many industrial control infrastructures shall expect a significant flexibility and scalability improvement with the adoption of the virtualization, cloud-native, and microservices IT paradigms. H. Orchestration and Closed-Loop Automation of TSN Networks This subsection surveys research works that introduce end- to-end network management aspects that characterize 5G networks and 5G-TSN networks, such as: end-to-end orchestration, network self-configuration and zero-touch closed-loop automation, network slicing, and network capability exposure. These network management schemes are reflected in the taxonomy of Table XX, where contributions can be primarily classified between network optimization,network orchestration, and network self-configuration and automation. 1) End-to-end Orchestration Involving TSN Resources: In [334], authors describe an edge cloud collaborative intelligent platform for industrial applications called IndustEdge that integrates a TSN-based data plane substrate. IndustEdge’s architecture consists of: an infrastructure layer that comprises field devices, a TSN data plane between the devices and the edge, and an edge cloud that provides K3s and Kubernetes nodes; a platform layer, which includes an orchestrator that coordinates TSN configuration according to the placement and characteristics of the applications, and additional tools to boost efficiency and programmability such as edge cloud and edge-edge collaboration modes, an SDK, and analytics; and an application layer for the microservices. In addition to the prototype platform, authors propose an emulation platform of IndustEdge. A performance evaluation is also carried out through two industrial use cases. Reference [335] proposes a task deterministic network (TDN) architecture to enable bounded latency and zero jitter communications in scenarios of the Industry 4.0 ecosystem that involve 5G and MEC technologies. With the authors’ focal point being the task offloading of industrial applications to MEC systems, the primary deterministic communication scenarios contemplated are multiple-to-one device-to-MEC access, and cross-domain interconnection of MEC systems. The proposed hierarchical TDN architecture consists of: an heterogeneous access network and data plane, which includes a 5G RAN in the factory network with a local MEC deployment, and a transport network with additional MEC deployments over TSN and DetNet domains; a control plane formed by sub-controllers that collaborate seamlessly for the management of the data plane including cross-domain configuration of traffic shaping mechanisms; and an application plane featuring a fast path between the network and the applications. Authors of [336] present AIDA, an AI-powered architecture for data-driven IIoT applications. The proposed orchestration architecture focuses on: (1) a TSN network and control plane with reconfiguration capabilities aimed at addressing the dynamicity of IIoT and microservices-based industrial applications in Kubernetes clusters on the edge; (2) a monitoring architecture encompassing SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 307 TABLE XX TAXONOMY RELATED TO THE ORCHESTRATION AND CLOSED-LOOP AUTOMATION OF TSN NETWORKS the infrastructure, platform, and containerized microservices; and (3) real-time AI/ML pipelines that target the provision of decision-making capabilities to IIoT processes based on the processing of collected sensor data. Authors introduce preliminary performance results of their architecture based on two proposed industrial control use cases. Additionally, previously surveyed works [237] and [330] can also be highlighted here for their orchestration contribution based on Kubernetes. Other research papers briefly propose the use of TSN for specific domains such as the TN within an end-to-end 5G or Beyond 5G (B5G) network to enrich the flexibility of their proposed architecture and support additional vertical use cases. Such is the case of [337],[338] and [339]. Implementation details regarding TSN are however not elaborated. 2) Network Slicing in TSN Networks: [149] explores the incorporation of TSN technologies into end-to-end network slicing, focusing on the orchestration of a TSN-based TN fabric that can serve 5G and non-5G flows concurrently. The authors’ proposed architecture is concerned with the integration of 3GPP network slicing management functions and an SDN-based TSN control plane. A TSN slice orchestrator entity is introduced as an entry point atop the TSN control plane (CNC/CUC) to provide it with slice visibility so that additional slice isolation mechanisms can be enforced prior to TSN scheduling configuration such as 802.1Qbv. This orchestrator also interfaces with an additional controller that takes care of slice management procedures in the 5G network. For slice definition in the 5G management domain, an augmented network slice template is proposed to accommodate TSN features. Authors finalize with a preliminary implementation and evaluation of their approach. In [340], authors focus on the use of network slicing in private 5G networks in order to provide isolation among production lines in Industry 4.0. In this case, rather than seeking the design of a slice-aware TSN network, TSN is used as a tool for its ability to fully isolate the performance of the different slices in the TN whilst providing deterministic low latency. Authors employ asynchronous TSN based on the ATS of the 802.1Qcr standard, which is evaluated against bare Ethernet. Authors’ experimental evaluation focusing on the midhaul between CUs and gNB-DUs highlights the effectiveness of infrastructure slicing together with TSN traffic regulation to realize complete slice isolation and minimize performance degradation in URLLC slices due to traffic excess in other slices. 3) Self-configuring TSN Networks: Some works in the literature explore the challenges towards self-configuring TSN networks, either to address network and schedule reconfiguration, or aiming to entirely remove the requisite from end stations to declare their stream requirements. The synergies with SDN architectures are clear in these proposals. The framework introduced in [96] is based on an agent that continuously monitors network changes and adapts the network configuration accordingly in real-time. Authors propose the introduction of learning capabilities in the TSN switches to feed the agent intelligence about network traffic. Another example is [341], where authors focus on the real-time reconfiguration of GCL schedules using an architecture also based on a configuration agent. Note that these two articles were published at the time that 802.1 Qcc was being standardized. More recently, [342] explores required functionality for an end-to-end self-configuring TSN network. Centralized knowledge of network topology leveraging the fully centralized model of 802.1 Qcc, and specialized TSN agents in the end stations for the configuration of parameters such as TC Qdiscs and time synchronization, are proposed. In this line, [343] presents a SDN-based dynamic self-configuration approach for TSN-based IoT networks. Edge switches in the network are enhanced with the capability to monitor traffic and learn traffic characteristics. Authors’ proposed approach consists on treating incoming traffic as BE by default, and, once its traffic characteristics are extracted, the CNC computes its optimal path and TAS configuration according to the stream’s priority 308 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 and time-sensitive requirements. Finally, [344] specifically focuses on the holistic monitoring of TSN networks, which could be used to feed AI-based decision-making. Authors identify three main sources of information that can be contrasted to provide insights ranging from device congestion or malfunction to network misconfiguration: real-time network traffic, configuration status of network devices, and centralized CNC/CUC information. Authors design an architecture to extract, preprocess, and store this information. 4) Zero-touch Closed-loop Automation of TSN and 5G- TSN Networks: Some recent works adopt ETSI ZSM-based approaches for the zero-touch configuration of 5G networks with TSN capabilities. In [276], authors propose the use of Deep Reinforcement Learning (DRL) based on ZSM and on the Framework for Abstraction and Control of Traffic Engineered Networks (ACTN) [345] for the optimization of the management and orchestration of 5G TNs. They employ asynchronous TSN based on ATS, targeting the backhaul link specifically with the goal of flow acceptance maximization. In the workflow devised by the authors, incoming flow requests are parsed to determine their characteristics, and then traffic predictive data analytics are leveraged to drive the allocation of incoming flow requests according to network status. In their work in [257] and [258], authors address the challenge of the configuration of a 5G NPN featuring 5G-TSN integration following the automation principles of the ZSM framework. They propose an automata learning approach to construct a behavioral model of the 5G NPN as a digital twin in order to find suitable 5GS configurations for given TSN traffic flows and desired KPIs. Authors introduce a smart controller that monitors the 5G network and learns sequences of observed states for given 5GS configurations and TSN traffic patterns. Using this model, the controller can then arrive to candidate network configurations that can be applied to the real 5G NPN when there are KPI deviations between an observed and desired state. The new network state then feeds the monitoring module of the controller in a closed loop manner. In [307], authors apply the NDT paradigm to 5G and 6G networks that integrate TSN. To construct the model in the NDT, simulation, emulation, and AI/ML over collected data are proposed in a complementary manner. The proposed NDT sits between the CNC and the CUC. The CNC shares information about the physical network to the NDT and can dynamically deploy probes in the physical network to collect specific data requested by the NDT. Some challenges of the NDT are briefly presented, including the real-time communication between the NDT and the physical network, and the data collection from heterogeneous and dynamic data sources. 5) Exposure and Abstraction of TSN Control to Verticals: [346] showcases the design and implementation of a TSN Controller (TSNC) NetApp for the exposure to vertical applications of the quality-aware management of a wireless TSN network. The TSNC contains the TSN control plane including CNC and CUC, but also real-time monitoring and telemetry functions to ensure compliance with requested KPIs. The devised data plane consists of WiFi gateways coupled with a TSN Agent (TSNA) that interacts with the TSNC NetApp and applies the network configuration changes in the wireless TSN data plane. The TSNC features a northbound interface that receives requests from vertical applications, which are collected by the CUC in the form of stream characteristics and expected KPIs. Monitoring rules in the TSNC are derived from the registry of vertical applications maintained in the CUC. While not its core contribution, aforementioned work [258] also introduces the management of vertical application-facing concepts including configuration parameters, traffic patters, and expected KPIs of TSN sessions. Summary and Lessons Learned: •TSN is a valuable technology for the transport layer in edge clouds and MECs with deterministic requirements: Industrial applications requiring reduced response times have long adopted edge computing architectures. However, some authors find that the availability of edge resources near the devices is not enough to guarantee the determinism in the communication to the edge/MEC node, or between collaborating nodes in an edge/MEC system. TSN has been found to be a promising technology to fill this gap, including for the realization of performance-isolated network slices over such infrastructures (e.g., for the isolation of industrial production lines). One of the main challenges, better tackled with the usage of SDN architectures, lies on the design of coordinated multi-domain orchestration capabilities that can control end devices, MEC or edge nodes, and TSN resources in an unified manner. Furthermore, in 5G networks, RAN resources also become part of this puzzle. On the other hand, while the literature focuses on architectural aspects in these proposals, the underlying TSN mechanisms to employ are not comprehensively examined. In one case, the use of CQF is briefly proposed. Other works hint at the use of TAS-based scheduling. Still, how to distribute the time synchronization between the different domains is not examined. Conversely, a proposal for network slicing in the TN using TSN finds ATS suitable given the nonperiodic traffic patterns. It is worth noting that, in 5G, MEC connectivity can be considered a subcase of the 5G TN, and, as later seen in Section V-E, both synchronous and asynchronous traffic shaping mechanisms may be appropriate in these scenarios. •Addressing configuration complexity of TSN and 5G- TSN networks through self-learning and zero-touch approaches is challenging but promising: Concepts including self-configuration, zero-touch automation, realtime monitoring, and digital twin, represent synergistic trends being adopted by novel TSN network management proposals to address the configuration complexity of such networks. Some works adopt the ZSM paradigm for this purpose, with two research lines being identified. In one, historical data combined with predictive analytics on upcoming workloads is used for traffic forecasting to manage online TSN flow allocation. In the other, real-time monitoring of the TSN traffic is used to automatically reconfigure the network to meet requested KPIs when there are deviations. Initial proposals opt for RL-based solutions, including DRL or automata learning, though further research is needed to determine SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 309 the ML method that yields the best performance. In this regard, the utilization of the digital twin paradigm and in particular of the NDT is also promising. If a faithful NDT can be constructed, it provides a side-effect free scenario to evaluate the behavior of TSN configuration, for example through RL methods, before applying the optimal configuration candidate to the physical network. Notably, NDTs and closed-loop automation impose the requirement of real-time monitoring, which has to be carefully implemented to ensure the validity and timeliness of the collected data without introducing overhead in the actual traffic. •Closed-loop automation based on KPI monitoring facilitates the abstraction of underlying TSN complexity to vertical applications: When it comes to abstracting end stations from the underlying mechanics of the TSN network, the 802.1Qcc standard provides a relevant baseline support through the definition of the CUC entity and the UNI. End stations report their traffic requirements in terms of KPIs to a centralized entity, rather than themselves executing the network-wide configuration to support them. As illustrated by some preliminary research works in the literature, its combination with real-time monitoring of said KPIs effectively materializes an intentbased approach to orchestrate TSN networks, facilitating its usage by heterogeneous vertical applications that can focus on their business logic rather than on specific TSN mechanisms. A promising complementary research line has also been found to aim for fully autonomous TSN networks, where end hosts do not have to report traffic requirements and thus all knowledge about traffic characteristics and TSN procedures can be foregoed. Still, more research is needed to examine how this approach scales with high dynamicity in the allocation and deallocation of TSN streams, and with TSN networks integrated with 5G, where the classification and configuration of TSN flows is more complex. I. Industrial Application Deployments and Prototypes Using TSN over 5G This subsection describes real-world deployments carried out by the research community in which different kinds of industrial use cases are enabled by the utilization of TSN technology over wireless 5G connectivity. This subsection includes works that carry out performance evaluation over a real testbed or prototype, as opposed to, for example, Section V-B, which covered works in which the focal point is the execution of mathematical and/or simulation-based analyses. Note that some of the works included in this subsection have also been referenced previously in this section covering other specific contributions. Reference [248] describes five use cases for evaluating and validating 5G for smart manufacturing, conducted across three sites and over six different 5G network deployments with different characteristics (frequency bands, bandwidth, SA/NSA architecture). Such use cases include collaborative robotics, process monitoring, and AGVs, and highlight benefits of adopting 5G, such as the offloading of machine control functionality to an edge cloud, and overall more flexible and performant deployments. URLLC features, FRER, and TSN- over-5G are evaluated. Reference [347] presents and evaluates a prototype setup integrating 5G in a TSN network. In the prototype, authors design a robotic application in which a Robot PLC transmits real-time sensor data to a factory edge cloud that controls the robot’s behavior and movement. A pre-commercial 5GS with URLLC support is used to implement the 5GS bridge that wirelessly interconnects two TSN bridges connected to the end stations that represent the robot and the edge cloud. Authors evaluate 802.1Qbv-based traffic shaping considering ST and BE traffic classes. They describe an approach to synthesize end-to-end GCL configuration in the devices according to latency thresholds, and leverage 802.1Qbv in the hardware TSN devices to reduce the jitter introduced separately by the sending end station and the 5G network. A method based on Pulse-per-second signals is used to measure clock offsets. Reference [249] demonstrates an Autonomous Mobile Robot (AMR) use case in a lineless mobile assembly system. AMRs collaboratively perform various screw installation tasks on a truck chassis, and use 5G connectivity to communicate with an edge system that executes computationally demanding tasks such as motion planning and multi-robot coordination. Apart from evaluating FRER as discussed in the previous section, authors apply the 3GPP-supported hold and forward mechanism to compensate jitter in the wireless transmissions. Reference [348] presents a 5G and TSN-enabled industrial control system developed with 3GPP Release 15 hardware and evaluates achievable latency bounds over such testbed consisting on a Ball-Balancing Table (BBT) that comprises electrical and mechanical dynamics, sensors, and actuators. Raspberry Pi-based BBT components send sensor data through a TSN network connected to an edge device, which sends back control signals to the BBT to control the path of the ball. Communication from and to the edge device is done via 5G. 802.1Qbv-based traffic scheduling supported by 802.1AS time synchronization is evaluated in this industrial system for the prioritization of time-sensitive traffic over artificial BE traffic. VI. OPEN CHALLENGES AND RESEARCH LINES From the extensive state of the art survey carried out in this paper, the leading research directions can be identified, and the potentially most critical or challenging matters to be addressed in the near future can be outlined. This section discusses how the integration between current 5G and TSN technologies shall continue progressing and evolving in order to pave the way for innovative applications and paradigms in the industrial automation vertical. A. Standardization and Adoption of URLLC Capabilities With TSN standards being developed for time-sensitive communications over wired Ethernet networks, their application to wireless 5G networks entails significant challenges. On the one hand, a wireless communication channel features dynamic propagation characteristics and can experiment 310 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 signal degradation due to interferences and other kinds of channel impairments. Thankfully, a significant effort of 5G standardization and research focuses on the enablement of ultra-reliable and ultra-low latency communications, which clearly synergizes with supporting the requirements of the most stringent TSN applications in the industrial automation vertical. In fact, we have seen how several URLLC capabilities are highly adopted in recent research studies as a foundation for their proposals, such as semi-static scheduling schemes like CG and SPS, flexible numerology in the SCS and the transmission slots, and schemes and configurations that enable redundant transmissions such as redundant user plane paths based on NR-DC, and Multi-TRP. Hence, to ensure an smooth technological development, SDOs and industry initiatives shall continue to dedicate efforts to expand current 5G NR and URLLC capabilities in the new releases of 5G and future generations through the technological surveillance of vertical industry requirements. Likewise, future research studies shall follow standardization progress closely and stay alert to novel technological advances in this line which could unveil new opportunities to optimize time-sensitive communications over 5G and the upcoming generations. B. Scheduling and Transmission of TSN Traffic over 5G The encompassing issue of augmenting a wired TSN network with wireless 5G-TSN virtual bridges revolves around how to achieve an optimal scheduling of time-sensitive streams using 5G resources. Based on a given 5G NR numerology, TSN streams have to be mapped to the available 5G RBs in the frequency domain. From the surveyed literature, it is evident that there is no one-size-fits-all solution to this challenge, with variables ranging from the periodicities of the streams, to their maximum delay budget, to the traffic heterogeneity, and to the end-to-end network topology. While some affirmations can be made from the literature such as the usefulness of semi-static scheduling to pre-allocate resources to time-sensitive streams, and the benefits of sub-slot scheduling and high SCS to enable more granular stream periodicities, different kinds of novel approaches are being brought to light. Indeed, proposals including adaptive scheduling algorithms [243],[247], CG-based allocation schemes [244],[246], traffic prioritization schemes [245], and modulation table adjustment algorithms [242], at different levels of the 5G NR stack, showcase that there is still ongoing innovation in the matter. Furthermore, the complexity of this challenge has propelled that many researchers are taking advantage of AI/ML techniques [258], often based on RL [259],[260], to tackle this challenge and to arrive at the optimal 5G QoS configuration by learning from the network behavior. With several contrasting proposals emerging from the research community, not only there is room to delve further into their optimization and performance characterization, but, as they often feature complementary approaches or mechanisms (e.g., leveraging different 5G NR and URLLC capabilities), assessing the feasibility and effectiveness of harmoniously combining them can also be substantially explored. On the other hand, as far the synthesis of a global 802.1Qbv GCL is concerned, which is the most adopted approach to coordinate the end-to-end scheduling in 5G-TSN networks, it could be valuable to more actively investigate the suitability of thoroughly researched methods for wired TSN networks. For instance, the usage of SDN-based stream classification and clustering algorithms (e.g., [290],[291]) in 5G QoS mapping strategies for TSN streams could facilitate the synthesis of global GCLs that are also suitable for 5GS bridges. However, this can be a double-edged sword. While embracing the extensive GCL computation research in wired TSN networks can be highly beneficial, it is critical to account for the inherent differences in the wireless 5GS bridges. Consequently, an expressive interplay between the TSN control plane and the 5GS that can accurately capture an unified vision of the wired and wireless domains, despite their diverging characteristics in behavior and resource and QoS management, can be a pivotal factor in the design of effective end-to-end 5G-TSN path and schedule computation schemes. C. Highly Accurate TSN Time Synchronization Over 5G Jointly with the scheduling of time-sensitive streams, the main challenge to integrate TSN in 5G networks is concerned with achieving a highly accurate time synchronization over the wireless medium, as indicated by the number of research papers that address this topic. Stringent communication requirements imposed by industrial control applications and implemented upon TSN capabilities such as TAS, demand for TSN time synchronization across the 5GS with an accuracy in the sub-microsecond range. The literature identifies PDC as the most significant challenge towards this goal for propagation distances above a certain threshold. With the propagation delay estimation accuracy depending on a varying range of factors as discussed in Section V-D, more research is deemed appropriate to more precisely characterize the impact of all these factors and the effectiveness of available and potential new methods in different network conditions. Furthermore, given the nature of the problem, optimization schemes based on AI/ML and closed-loop automation could be explored, aimed at finding a suitable configuration to maximize 5G resource efficiency whilst the synchronization accuracy threshold for the given synchronicity budget requirements and network characteristics can be attained. Another relevant aspect to consider is the evolution of time synchronization schemes supported in 3GPP standards. While Release 16 introduced the support of downlink time synchronization, Release 17 expanded this support by considering new configurations to distribute timing information. At large, this leads to four different time synchronization schemes that can be appropriate for different use case scenarios: (1) downlink from GM behind network side towards UE side; (2) uplink from GM behind UE side towards network side; (3) UE-to- UE with GM behind one of the UEs; and (4) GM inside 5GS towards network side and UE side. With a significant portion of the academic contributions targeting the downlink time synchronization scheme specifically or being aligned in timeline Release 16 standards (e.g., [23],[266],[270]), SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 311 it is important that upcoming research efforts analyze the implications of the remaining configurations more deeply. Lastly, the robustness and resiliency against failures of the time synchronization is also a critical aspect to be mindful of in industrial applications with stringent synchronization requirements. In the scope of wired TSN networks, the 802.1ASdm standard, currently in draft phase, is amending 802.1AS to introduce hot standby mechanisms for obtaining a single clock source from to two gPTP domains over redundant paths, a primary one and an standby one, without requiring a BMC algorithm execution. On the other hand, proposals for adopting such time synchronization redundancy mechanisms in 5G-TSN networks appear to be missing in the literature. However, it can be worthwhile to study the feasibility of supporting them over redundant 5G network configurations. D. Mobility and Handover Management for Mobile UEs in 5G-TSN Networks Innovative Industry 4,0 applications employ all kinds of mobile vehicles and robots in their production processes, as seen for example in the prototype deployments described in Section V-I. As the integration between TSN and 5G is a novel research topic, the community has understandably focused on evaluating and optimizing the scenarios where the UE and DSTT have low movement speeds and their 5G connectivity fixed to a single gNB. However, while one of the main incentives of introducing 5G network into Industry 4.0 is the reduction of cabling equipment, many use cases such as motion control, remote control, and cooperative workpiece carrying, often involve mobile units that span large areas requiring coverage from multiple cells, and that can additionally reach speeds of up to 75 km/h [140]. On the one hand, how the PDU session QoS and the time synchronization quality is affected by UE speed can be further examined, as it directly impacts the planning of the network and the choice of 5G QoS profiles to be configured to support TSN streams. On the other hand, the support of large coverage distances may require from handover events, whether its a cell change while remaining anchored to the same CU, a complete gNB change, or even a handover with UPF rellocation, with each type potentially having different feasibility limits in the context of 5G-TSN networks. As such, it is considered timely to investigate mechanisms that can help mitigate the impact. Similarly, the procedures that would need to be triggered in the 5G and TSN control planes to respond to these events can be studied, in order to ensure the continuity of the TSN streams and the synchronization as seamlessly as possible. E. Open Source Software for 5G-TSN Integration Even though open source 5G solutions have existed for years and are constantly being updated and polished, as of the writing of this paper, there is no open source 5G software that supports the integration with TSN. Such support would require the implementation of the architecture and procedural aspects of the 5GS bridge model, including the TSN translator entities, NW-TT and DS-TT, the TSN AF, and additional procedures across the 5GC and 5G-AN. Among examples of existing 3GPP-compliant open source implementations, free5GC [349] and Open5GS [350] provide modular cores, srsRAN [351] offers a RAN implementation, and the OpenAirInterface (OAI) [352] project includes both core and RAN. Hence, support for TSN integration in any of these projects would require either implementation over a solution like OAI, or the coordination between an open source 5GC project and an open source RAN project. A first approach could begin with the support of Release 16 features, such as downlink time synchronization and the essential interactions for QoS configuration and management between CNC, TSN AF, NW-TT, and DS-TT. It could eventually be expanded to support the extensions standardized in newer 3GPP releases, potentially adopting consolidated advancements that arise from the research community as well. The availability of such open source solution could pave the way to take the step forward from theoretical evaluation and simulation analysis to faithful testbed prototypes. F. Integration of TSN Support Into Virtualization and Orchestration Architectures As shown in this paper, industry initiatives and academic research works alike reflect a clear spike in the interest to adopt virtualization technologies in industrial networks. This is especially true for the agile container-based virtualization, following the paradigm shift towards microservices. Moreover, several academic works have surfaced in recent years addressing the usage of TSN for time-sensitive applications in end stations with virtualization support. In this regard, the literature showcases promising first steps towards the integration of TSN into Kubernetes [329],[330]; container-based virtualization can be deemed more favourable in this context than hypervisor-based virtualization due to the complexity of addressing the overhead introduced by the latter in order to comply with TSN requirements. However, in addition to Kubernetes, there are consolidated architectural frameworks to deploy applications with either type of virtualization under rich and flexible orchestration capabilities, namely NFV and MEC, that are backed up by extensive SDO work and industry adoption. Initiatives led by SDOs to integrate TSN support into these architectures, potentially leading to the implementation of such support into open source NFV and MEC platforms, could facilitate the growth of innovative industrial applications. An approach to integrate of TSN into either Kubernetes, NFV, or MEC shall address the interoperability and cooperation between the architectures’ control planes. Nevertheless, it shall also guarantee that orchestration mechanisms are made aware of TSN assets and real-time infrastructure capabilities in order to adequately configure the applications and the network as a continuum by applying procedures defined in TSN standards. Upcoming TSN standards regarding configuration enhancements, such as 802.1Qdj and 802.1Qcw, could play an important role. Furthermore, specific research questions may arise. For example, in the highly dynamic environments that virtualization often features, exploring the feasibility of alternative traffic shaping algorithms more friendly to network reconfiguration, such as 802.1Qch CQF, could be interesting. 312 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 Another potential topic to investigate could be the applicability of SFC to represent TSN streams, given that, service function chains, which can be implemented for example using the Source Routing paradigm, can be made unidirectional [353]. Finally, a complementary research direction that is only beginning to surface is the integration of TSN into the virtualized and interoperable O-RAN architecture [50]. This is an equally promising challenge that can be approached in an equivalent manner to the integration of TSN into NFV or MEC, in this case for bringing TSN to the TN of the 5G RAN. G. Accelerated SDN to Enhance Time-Sensitive Communications With the performance of time-sensitive communications being potentially hampered by virtualization, the use of kernel bypass and userspace packet processing approaches have been studied in recent years to accelerate the networking stack of virtualized applications in real-time platforms. This trend is slowly moving to the TSN ecosystem. Some promising proposals surveyed in this paper include the use of XDP for the traffic steering in 5G UEs [250], the design of a OVSDPDK-based Kubernetes CNI supporting TSN [329], and the design of a TAS VNF that is built using DPDK [333]. With these research works being mostly preliminary studies and approaches, further research shall be carried out. As most surveyed works opt for DPDK over XDP, the latter could be further investigated; even though existing studies indicate that its packet processing performance may be lower than DPDK, it boasts flexibility advantages due to providing a trade-off between kernel bypass and kernel networking stack [354]. Overall, an important matter is performance characterization in different scenarios in order to support the requirements of TSN-aware and virtualized industrial applications. The consideration of hardware offload via FPGA-based SmartNICs to accelerate DPDK or XDP could be beneficial, as the literature showcases software implementations. In addition, the integration of these capabilities into existing platforms such as Kubernetes can be highly important towards industry adoption. Another technology showing promising potential in this regard is the P4 programming language. P4 provides a powerful and expressive packet processing language with a rich ecosystem of available hardware targets, that aims to decouple the hardware from the software implementation [355]. Research works so far dealing with the use of P4 in TSN networks have focused on the implementation of PSFP functionality in the data plane [299],[300]. However, it could be promising to employ P4 for the encapsulation of other TSN functionalities. This likely warrants feasibility studies in matters ranging from the clock synchronization in P4 targets and the propagation of time synchronization messages in P4 networks, to the management of network reconfigurations that may impact the P4 data plane. H. Autonomous Zero-Touch 5G-TSN Networks The literature reflects complementary proposals that can play a crucial role in the autonomous configuration and optimization of TSN and 5G-TSN networks. This includes SDN-based network reconfiguration schemes (e.g., [343]), holistic real-time monitoring frameworks (e.g., [344]), and a versatile suite of ML algorithms to arrive at optimal network configurations (e.g., [253],[259]). Even though further research on these topics can be delved into independently, the biggest potential likely comes from their contribution as part of a whole in the architecting of closed-loop, zerotouch network management frameworks—for example based on ZSM—, or in the adoption of the NDT paradigm for TSN networks. Additionally, there is a significantly challenging step to take in the engineering of these solutions so that they contemplate the inclusion wireless 5G domains in addition to traditional wired domains. Some recent research studies and proposals, including [257],[258],[276], and [307], take the first steps to address this challenge. Given the resonance of zero-touch network management in today’s 5G and the upcoming 6G [356], as well as in current and future Industry 4.0 use cases in manufacturing and automation [39], substantial profits can be gained by continuing research in this direction pertaining TSN networks. While its unveiling in the future cannot be determined with absolute certainty, some hints can be taken. On the one hand, a more faithful reflection of 5G-TSN integration mechanisms, such as the accounting of time synchronization requirements, can enhance currently proposed solutions. Likewise, the consideration in the system’s decision-making of previously discussed URLLC capabilities, which have been deemed highly beneficial by other research studies, could be another natural step forward in this matter. Additionally, while current ZSM-based proposals fundamentally approach the problem from the perspective of static network characteristics with changing application traffic and KPI requests, a future research line could look at how the closed-loop management framework can learn to predict varying network channel conditions in the 5G domain. This would allow to anticipate their effects and take appropriate actions (e.g., reconfiguration of resources), in order to maintain compliance with traffic requirements in spite of events like channel degradation, obstruction of the propagation path, etc. It is unclear if achieving this would require network sensing capabilities that will only arrive with 6G technology [357], but a preliminary approach could look to leverage the network insight provided by current 3GPP-standardized capabilities, such as the NWDAF and MDAF NFs. I. Exposure of 5G-TSN Network Capabilities Towards Industrial Applications In order to achieve a smooth adoption of deterministic 5G- TSN hybrid deployments in industrial environments, mainly in the context of 5G NPNs, developing adequate exposure capabilities from the network towards the industrial applications and management systems can play a crucial role. For this purpose, the network shall be able to provide rich and flexible service-based interfaces that can be used by an industrial consumer entity to provide application requirements, to request the activation of services, and to monitor network metrics; these capabilities shall be implemented in a way SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 313 that the consumer is abstracted from the internal complexity of the network and the underlying mechanisms it utilizes. With this in mind, it could be highly profitable to research more exhaustively how a satisfactory exposure of TSN traffic management capabilities in 5G networks can be fulfilled based on standardized 5G frameworks and services such as NEF, CAPIF, and SEAL. In this regard, the perspective of upcoming standardization endeavors is promising, as the 3GPP Technical Report (TR) 23.700-25 [358], which outlines how Release 18 is expected to evolve with regards to TSC and URLLC as 5G matures into 5G-Advanced, identifies several key issues to be studied, including: (1) how the 5GS can report time synchronization status and quality to UEs, to devices attached to UEs, and to third-party AFs; (2) how to enable an AF to request a time synchronization service in a specific coverage area; (3) how to enable an AF to explicitly provide the required PER to the NEF or PCF to assist in the selection of a 5QI; and (4) how to allow applications with low latency requirements to adapt transmission scheduling according to RAN feedback, and how to enable the RAN to provide such feedback to the applications (AFs). Additionally, SDOs and the research community alike should play close attention to the reports of industry alliances such as 5G-ACIA [137], given their profound vision on the demands of emerging Industry 4.0 applications. J. End-to-End Multi-Technology Orchestration Solutions As seen throughout this paper, complex communication scenarios demanded by innovative industrial applications imply the necessity of managing separate technology domains in a coordinated manner, including: •Applications running inside of heterogeneous systems and platforms, located either in end stations in a wired TSN domain, in end stations behind a 5G UE, or in an edge cloud. •Wired TSN systems supporting production lines or the communication with centralized control and/or an edge cloud. •5G Systems acting as virtual TSN bridges within the above TSN systems to provide wireless connectivity. •The TN between the RAN and UPF of a 5G System. •The inclusion of architectures with 5G network slicing capabilities, where a single or multiple slices may be available to allocate different service flows. Several articles addressing different technology domains within the scope of this paper share this view on configuration and orchestration complexity, e.g., [34],[38],[52],[237], [257], and [340]. Ultimately, end-to-end orchestration in this context encompasses not only the innumerable procedures and mechanisms related to 5G-TSN scheduling and network management, but also the lifecycle management of industrial CPSs and applications under paradigms like edge computing, MEC, or NFV. Crucially, all of these points should be carried out in an unified manner as a seamlessly integrated whole, to satisfy each individual requirement across the end-to- end infrastructure. Furthermore, the dynamicity introduced by 5G scenarios, such as through the rapid deployment and reconfiguration of VNFs, is unprecedented in traditional deterministic industrial networks. This demands for robust and complex orchestration capabilities, including a proper reconfiguration strategy for TSN mechanisms, and a network topology-aware management of TSN resources. For all these reasons, the state of the art shall continue advancing towards pervasive multi-domain orchestration solutions such as the proposals in [149],[334],[335], and [336], building upon the ever-evolving ecosystem of 5G frameworks and capabilities developed by SDOs. VII. CONCLUSION In this paper, we have comprehensively reviewed the main technologies and standards pertaining the integration of TSN and 5G technologies in the context of industrial automation in manufacturing. Additionally, we have thoroughly surveyed the existing literature in the matter, showcasing the emergence of extensive research studies and proposals to improve the current state of the art. Finally, we derived open challenges and several promising research directions that are complementary towards the shared goal of supporting the ever-evolving requirements of innovative Industry 4.0 applications. With encouraging research space in diverging but complementary scopes ranging from the TSN-compliant scheduling of 5G resources, to the support of TSN in platforms with virtualization, and to the end-to-end orchestration and zero-touch management of 5G- TSN networks, we believe that the integration and convergence between TSN and 5G will continue to be a highly relevant topic in the coming years. Furthermore, we believe that a joint effort from SDOs, industry alliances, and the research community, will lead to the best outcomes. We hope that our work in this paper can contribute to providing further guidance for the continuation of research in this area in the road towards upcoming paradigms and technologies like Industry 5.0 and 6G. REFERENCES [1] H. Lasi, P. Fettke, H.-G. Kemper, T. Feld, and M. Hoffmann, “Industry 4.0,” Bus. Inf. Syst. Eng., vol. 6, no. 4, pp. 239–242, 2014. [2] E. Hozdi´ c, “Smart factory for industry 4.0: A review,” Int. J. Mod. Manuf. Technol., vol. 7, no. 1, pp. 28–35, 2015. [3] A. Kusiak, “Smart manufacturing,” Int. J. Prod. Res., vol. 56, nos. 1–2, pp. 508–517, 2018. [4] D. G. Pivoto, L. F. de Almeida, R. da Rosa Righi, J. J. Rodrigues, A. B. Lugli, and A. M. Alberti, “Cyber-physical systems architectures for Industrial Internet of Things applications in industry 4.0: A literature review,” J. Manuf. Syst., vol. 58, pp. 176–192, Jan. 2021. [5] A. Angelopoulos et al., “Tackling faults in the industry 4.0 era—A survey of machine-learning solutions and key aspects,” Sensors, vol. 20, no. 1, p. 109, 2019. [6] M. Wollschlaeger, T. Sauter, and J. Jasperneite, “The future of industrial communication: Automation networks in the era of the Internet of Things and industry 4.0,” IEEE Ind. Electron. Mag., vol. 11, no. 1, pp. 17–27, Mar. 2017. [7] S. Vitturi, C. Zunino, and T. Sauter, “Industrial communication systems and their future challenges: Next-generation Ethernet, IIoT, and 5G,” Proc. IEEE, vol. 107, no. 6, pp. 944–961, Jun. 2019. [8] J. Farkas, B. Varga, G. Miklós, and J. Sachs, “5G-TSN integration meets networking requirements for industrial automation,” Ericsson Technol. Rev., vol. 96, no. 7, pp. 45–51, 2019. [9] B. Blanco et al., “Technology pillars in the architecture of future 5G mobile networks: NFV, MEC and SDN,” Comput. Stand. Interfaces, vol. 54, pp. 216–228, Nov. 2017. 314 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, VOL. 27, NO. 1, FEBRUARY 2025 [10] Z. M. Temesvári, D. Maros, and P. Kádár, “Review of mobile communication and the 5G in manufacturing,” Procedia Manuf., vol. 32, pp. 600–612, 2019, doi: 10.1016/j.promfg.2019.02.259. [11] A. Nasrallah et al., “Ultra-low latency (ULL) networks: The IEEE TSN and IETF DetNet standards and related 5G ULL research,” IEEE Commun. Surveys Tuts., vol. 21, no. 1, pp. 88–145, 1st Quart., 2018. [12] C. Benzaid and T. Taleb, “AI-driven zero touch network and service management in 5G and beyond: Challenges and research directions,” IEEE Netw., vol. 34, no. 2, pp. 186–194, Mar./Apr. 2020. [13] Y. Seol, D. Hyeon, J. Min, M. Kim, and J. Paek, “Timely survey of time-sensitive networking: Past and future directions,” IEEE Access, vol. 9, pp. 142506–142527, 2021. [14] L. Deng, G. Xie, H. Liu, Y. Han, R. Li, and K. Li, “A survey of real-time Ethernet modeling and design methodologies: From AVB to TSN,” ACM Comput. Surv., vol. 55, no. 2, pp. 1–36, 2022. [15] A. A. Barakabitze, A. Ahmad, R. Mijumbi, and A. Hines, “5G network slicing using SDN and NFV: A survey of taxonomy, architectures and future challenges,” Comput. Netw., vol. 167, Feb. 2020, Art. no. 106984. [16] D. Cavalcanti, J. Perez-Ramirez, M. M. Rashid, J. Fang, M. Galeev, and K. B. Stanton, “Extending accurate time distribution and timeliness capabilities over the air to enable future wireless industrial automation systems,” Proc. IEEE, vol. 107, no. 6, pp. 1132–1152, Jun. 2019. [17] L. Silva, P. Pedreiras, P. Fonseca, and L. Almeida, “On the adequacy of SDN and TSN for industry 4.0,” in Proc. IEEE 22nd Int. Symp. Real-Time Distrib. Comput. (ISORC), 2019, pp. 43–51. [18] N. K. Haur and T. S. Chin, “Challenges and future direction of time-sensitive software-defined networking (TSSDN) in automation industry,” in Proc. 12th Int. Conf. Secur., Privacy, Anonym. Comput., Commun., Stor., 2019, pp. 309–324. [19] A. Ghosh, A. Maeder, M. Baker, and D. Chandramouli, “5G evolution: A view on 5G cellular technology beyond 3GPP release 15,” IEEE Access, vol. 7, pp. 127639–127651, 2019. [20] J. Zou, S. A. Sasu, J. Messenger, and J.-P. Elbers, “Options for timesensitive networking for 5G fronthaul,” in Proc. 45th Eur. Conf. Opt. Commun. (ECOC), 2019, pp. 1–3. [21] T. Striffler, N. Michailow, and M. Bahr, “Time-sensitive networking in 5th generation cellular networks-current state and open topics,” in Proc. IEEE 2nd 5G World Forum (5GWF), 2019, pp. 547–552. [22] A. Larrañaga, M. C. Lucas-Estañ, I. Martinez, I. Val, and J. Gozalvez, “Analysis of 5G-TSN integration to support industry 4.0,” in Proc. 25th IEEE Int. Conf. Emerg. Technol. Fact. Autom. (ETFA), vol. 1, 2020, pp. 1111–1114. [23] I. Godor et al., “A look inside 5G standards to support time synchronization for smart manufacturing,” IEEE Commun. Stand. Mag.,vol.4, no. 3, pp. 14–21, Sep. 2020. [24] S. Jun, Y. Kang, J. Kim, and C. Kim, “Ultra-low-latency services in 5G systems: A perspective from 3GPP standards,” ETRI J., vol. 42, no. 5, pp. 721–733, 2020. [25] A. Aijaz, “Private 5G: The future of industrial wireless,” IEEE Ind. Electron. Mag., vol. 14, no. 4, pp. 136–145, Dec. 2020. [26] L. C. Mutalemwa and S. Shin, “Achieving bounded ultra-low latency in 5G and beyond: Challenges and future research directions,” in Proc. Korean Inst. Commun. Sci. Conf., 2021, pp. 38–41. [27] S. Scanzio, L. Wisniewski, and P. Gaj, “Heterogeneous and dependable networks in industry–A survey,” Comput. Ind., vol. 125, Feb. 2021, Art. no. 103388. [28] F. Hamidi-Sepehr et al., “5G URLLC: Evolution of high-performance wireless networking for industrial automation,” IEEE Commun. Stand. Mag., vol. 5, no. 2, pp. 132–140, Jun. 2021. [29] Ó. Seijo, I. Val, M. Luvisotto, and Z. Pang, “Clock synchronization for wireless time-sensitive networking: A march from microsecond to nanosecond,” IEEE Ind. Electron. Mag., vol. 16, no. 2, pp. 35–43, Jun. 2022. [30] Y. Kang, S. Lee, S. Gwak, T. Kim, and D. An, “Time-sensitive networking technologies for industrial automation in wireless communication systems,” Energies, vol. 14, no. 15, p. 4497, 2021. [31] J. Sachs and K. Landernäs, “Review of 5G capabilities for smart manufacturing,” in Proc. 17th Int. Symp. Wireless Commun. Syst. (ISWCS), 2021, pp. 1–6. [32] H. Chahed and A. J. Kassler, “Software-defined time sensitive networks configuration and management,” in Proc. IEEE Conf. Netw. Funct. Virtualiz. Softw. Defin. Netw. (NFV-SDN), 2021, pp. 124–128. [33] J. Prados-Garzon, P. Ameigeiras, J. Ordonez-Lucena, P. Muñoz, O. Adamuz-Hinojosa, and D. Camps-Mur, “5G non-public networks: Standardization, architectures and challenges,” IEEE Access,vol.9, pp. 153893–153908, 2021. [34] L. Martenvormfelde, A. Neumann, L. Wisniewski, and L. Schreckenberg. “Co-configuration of 5G and TSN enabling end-to- end quality of service in industrial communications.” 2021. [Online]. Available: https://opendata. uni-halle. de//handle/1981185920/41505 [35] H. Xu, J. Xin, S. Xu, and H. Zhang, “RAN enhancement to support propagation delay compensation of TSN,” in Proc. IEEE 9th Int. Conf. Inf., Commun. Netw. (ICICN), 2021, pp. 179–182. [36] X. Li, Y. Chi, X. Jia, and Y. Xie, “Research on integration of 5G and TSN,” in Mechatronics and Automation Technology. Amsterdam, The Netherlands: IOS Press, 2022, pp. 128–135. [37] T. Fedullo, A. Morato, F. Tramarin, L. Rovati, and S. Vitturi, “A comprehensive review on time sensitive networks with a special focus on its applicability to industrial smart and distributed measurement systems,” Sensors, vol. 22, no. 4, p. 1638, 2022. [38] A. Mahmood, S. F. Abedin, T. Sauter, M. Gidlund, and K. Landernäs, “Factory 5G: A review of industry-centric features and deployment options,” IEEE Ind. Electron. Mag., vol. 16, no. 2, pp. 24–34, Jun. 2022. [39] M. Friesen, L. Wisniewski, and J. Jasperneite, “Machine learning for zero-touch management in heterogeneous industrial networks-a review,” in Proc. IEEE 18th Int. Conf. Fact. Commun. Syst. (WFCS), 2022, pp. 1–8. [40] J. Xin, S. Xu, H. Zhang, and S. Xiong, “A review of time sensitive communication technology,” in Proc. Int. Conf. Comput. Eng. Artif. Intell. (ICCEAI), 2022, pp. 844–848. [41] M.-A. Kourtis et al., “5G NPN performance evaluation for I4.0 environments,” Appl. Sci., vol. 12, no. 15, p. 7891, 2022. [42] F. Bette, “Overview of 5G new radio and time-sensitive networking bridge architecture,” Sep. 2022, doi: 10.13140/RG.2.2.23244.49288. [43] M. Abuibaid et al., “Cloudification of time-sensitive networking reliability functions: Challenges and potential solution directions,” IEEE Commun. Stand. Mag., vol. 6, no. 4, pp. 30–37, Dec. 2022. [44] S. Senk, H. K. Nazari, H.-H. Liu, G. T. Nguyen, and F. H. Fitzek, “Open-source testbeds for integrating time-sensitive networking with 5G and beyond,” in Proc. IEEE 20th Consum. Commun. Netw. Conf. (CCNC), 2023, pp. 1–7. [45] G. Prateek Sharma et al., “Toward deterministic communications in 6G networks: State of the art, open challenges and the way forward,” IEEE Access, vol. 11, pp. 106898–106923, 2023. [46] D. Chandramouli, P. Andres-Maldonado, and T. Kolding, “Evolution of timing services from 5G-A towards 6G,” IEEE Access, vol. 11, pp. 35150–35157, 2023. [47] Z. Satka, M. Ashjaei, H. Fotouhi, M. Daneshtalab, M. Sjödin, and S. Mubeen, “A comprehensive systematic review of integration of time sensitive networking and 5G communication,” J. Syst. Architect., vol. 138, May 2023, Art. no. 102852. [48] R. Muzaffar, M. Ahmed, E. Sisinni, T. Sauter, and H.-P. Bernhard, “5G deployment models and configuration choices for industrial cyber-physical systems—A state of art overview,” IEEE Trans. Ind. Cyber-Phys. Syst., vol. 1, pp. 236–256, Sep. 2023, doi: 10.1109/TICPS.2023.3311394. [49] M. Maliosz, I. Moldován, M. Máté, C. Simon, and J. Harmatos, “Deterministic local cloud for industrial applications,” in Proc. IEEE 19th Int. Conf. Fact. Commun. Syst. (WFCS), 2023, pp. 1–8. [50] E. Municio, G. Garcia-Aviles, A. Garcia-Saavedra, and X. Costa-Pérez, “O-RAN: Analysis of latency-critical interfaces and overview of time sensitive networking solutions,” IEEE Commun. Stand. Mag.,vol.7, no. 3, pp. 82–89, Sep. 2023. [51] T. Mizrahi and Y. Moses, “Timing in software-defined and centrallymanaged networks,” 2019, arXiv:1904.06676. [52] V. Struhár, M. Behnam, M. Ashjaei, and A. V. Papadopoulos, “Realtime containers: A survey,” in Proc. 2nd Workshop Fog Comput. IoT (Fog-IoT), 2020, pp. 1–9. [53] M. Alabadi, A. Habbal, and X. Wei, “Industrial Internet of Things: Requirements, architecture, challenges, and future research directions,” IEEE Access, vol. 10, pp. 66374–66400, 2022. [54] J. Kaiser, D. Mcfarlane, G. Hawkridge, P. André, and P. Leitão, “A review of reference architectures for digital manufacturing: Classification, applicability and open issues,” Comput. Ind., vol. 149, Aug. 2023, Art. no. 103923. [55] A. Nasrallah et al., “Performance comparison of IEEE 802.1 TSN time aware shaper (TAS) and asynchronous traffic shaper (ATS),” IEEE Access, vol. 7, pp. 44165–44181, 2019. [56] M. A. Ojewale, P. M. Yomsi, and B. Nikoli´ c, “Multi-level preemption in TSN: Feasibility and requirements analysis,” in Proc. IEEE 23rd Int. Symp. Real-Time Distrib. Comput. (ISORC), 2020, pp. 47–55. [57] D. Ergenç and M. Fischer, “On the reliability of IEEE 802.1 CB FRER,” in Proc. IEEE Conf. Comput. Commun., 2021, pp. 1–10. SASIAIN et al.: TOWARD THE INTEGRATION AND CONVERGENCE 315 [58] Y. Zhang, Q. Xu, L. Xu, C. Chen, and X. Guan, “Efficient flow scheduling for industrial time-sensitive networking: A divisibility theory-based method,” IEEE Trans. Ind. Informat., vol. 18, no. 12, pp. 9312–9323, Dec. 2022. [59] T. Stüber, L. Osswald, S. Lindner, and M. Menth, “A survey of scheduling algorithms for the time-aware shaper in time-sensitive networking (TSN),” IEEE Access, vol. 11, pp. 61192–61233, 2023. [60] R. A. Peña, M. Pascual, A. Astarloa, D. Uribe, and J. Inchausti, “Impact of MACsec security on TSN traffic,” in Proc. 37th Conf. Design Circuits Integr. Circuits (DCIS), 2022, pp. 1–6. [61] R. Sethi, A. Kadam, K. Prabhu, and N. Kota, “Security considerations to enable time-sensitive networking over 5G,” IEEE Open J. Veh. Technol., vol. 3, pp. 399–407, Sep. 2022. [62] T. Adame, M. Carrascosa-Zamacois, and B. Bellalta, “Timesensitive networking in IEEE 802.11 BE: On the way to low-latency WiFi 7,” Sensors, vol. 21, no. 15, p. 4954, 2021, doi: 10.1109/TICPS.2023.3311394. [63] I. Val, Ó. Seijo, R. Torrego, and A. Astarloa, “IEEE 802.1 AS clock synchronization performance evaluation of an integrated wired– wireless TSN architecture,” IEEE Trans. Ind. Informat., vol. 18, no. 5, pp. 2986–2999, May 2022. [64] D. Cavalcanti, C. Cordeiro, M. Smith, and A. Regev, “WiFi TSN: Enabling deterministic wireless connectivity over 802.11,” IEEE Commun. Stand. Mag., vol. 6, no. 4, pp. 22–29, Dec. 2022. [65] M. Ashjaei, L. L. Bello, M. Daneshtalab, G. Patti, S. Saponara, and S. Mubeen, “Time-sensitive networking in automotive embedded systems: State of the art and research opportunities,” J. Syst. Archit., vol. 117, Aug. 2021, Art. no. 102137. [66] Y. Peng, B. Shi, T. Jiang, X. Tu, D. Xu, and K. Hua, “A survey on in-vehicle time-sensitive networking,” IEEE Internet Things J., vol. 10, no. 16, pp. 14375–14396, Aug. 2023. [67] P. Trakadas et al., “A cost-efficient 5G non-public network architectural approach: Key concepts and enablers, building blocks and potential use cases,” Sensors, vol. 21, no. 16, p. 5578, 2021. [68] X. Yang, D. Scholz, and M. Helm, “Deterministic networking (DetNet) vs time sensitive networking (TSN),” in Proc. Semin. IITM, 2019, pp. 79–84. [69] N. Ambrosy, T. Kampa, U. Jumar, and D. Großmann, “5G and DetNet: Towards holistic determinism in industrial networks,” in Proc. IEEE Int. Conf. Ind. Technol. (ICIT), 2022, pp. 1–6. [70] Time-Sensitive Networking (TSN) Task Group, IEEE Standard 802.1, 2019. [Online]. Available: https://1.ieee802.org/tsn/ [71] D. Bruckner et al., “An introduction to OPC UA TSN for industrial communication systems,” Proc. IEEE, vol. 107, no. 6, pp. 1121–1131, Jun. 2019. [72] J. L. Messenger, “Time-sensitive networking: An introduction,” IEEE Commun. Stand. Mag., vol. 2, no. 2, pp. 29–33, Jun. 2018. [73] IEEE Standard for Local and Metropolitan Area Networks–Bridges and Bridged Networks, IEEE Standard 802.1Q-2022 (Revision of IEEE Standard 802.1Q-2018), 2022. [74] IEEE Standard for Local and Metropolitan Area Network–Bridges and Bridged Networks, IEEE Standard 802.1Q-2018 (Revision of IEEE Standard 802.1Q-2014), 2018. [75] IEEE Standard for Local and Metropolitan Area Networks–Bridges and Bridged Networks, IEEE Standard 802.1Q-2014 (Revision of IEEE Standard 802.1Q-2011), 2014. [76] G. Alderisi, G. Patti, and L. L. Bello, “Introducing support for scheduled traffic over IEEE audio video bridging networks,” in Proc. IEEE 18th Conf. Emerg. Technol. Fact. Autom. (ETFA), 2013, pp. 1–9. [77] IEEE Standard for a Precision Clock Synchronization Protocol for Networked Measurement and Control Systems, IEEE Standard 1588- 2019 (Revision of IEEE Standard 1588-2008), 2020. [78] IEEE Standard for Local and Metropolitan Area Networks–Timing and Synchronization for Time-Sensitive Applications, IEEE Standard 802.1 AS-2020 (Revision of IEEE Standard 802.1 AS-2011), 2020. [79] IEEE Standard for Local and Metropolitan Area Networks–Frame Replication and Elimination for Reliability, IEEE Standard 802.1CB- 2017, 2017. [80] IEEE Standard for Local and Metropolitan Area Networks–Bridges and Bridged Networks–Amendment 28: Per-Stream Filtering and Policing, IEEE Standard 802.1Qci-2017 (Amendment to IEEE Standard 802.1Q-2014 as amended by IEEE Standard 802.1Qca-2015, IEEE Standard 802.1Qcd-2015, IEEE Standard 802.1Q-2014/Cor 1-2015, IEEE Standard 802.1Qbv-2015, IEEE Standard 802.1Qbu-2016, IEEE Standard 802.1Qbz-2016), 2017. [81] D. Ergenç, C. Brülhart, J. Neumann, L. Krüger, and M. Fischer, “On the security of IEEE 802.1 time-sensitive networking,” in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), 2021, pp. 1–6. [82] P. Meyer, T. Häckel, S. Reider, F. Korf, and T. C. Schmidt, “Network anomaly detection in cars: A case for time-sensitive stream filtering and policing,” 2021, arXiv:2112.11109. [83] IEEE Standard for Local and Metropolitan Area Networks—Virtual Bridged Local Area Networks Amendment 12: Forwarding and Queuing Enhancements for Time-Sensitive Streams,” IEEE Standard 802.1Qav- 2009 (Amendment to IEEE Standard 802.1Q-2005), 2010. [84] Time Sensitive Networks for Flexible Manufacturing Testbed Characterization and Mapping of Converged Traffic Types, Ind. Internet Consort. (IIC), Boston, MA, USA, 2019. [85] R. Hummen, S. Kehrer, and O. Kleineberg, “TSN–time sensitive networking,” Hirschmann Autom., Rochester Hills, MI, USA, White Paper 00027, 2016. [86] IEEE Standard for Local and Metropolitan Area Networks—Bridges and Bridged Networks—Amendment 25: Enhancements for Scheduled Traffic, IEEE standard 802.1Qbv-2015 (Amendment to IEEE standard 802.1Q-2014 as amended by IEEE standard 802.1Qca-2015, IEEE standard 802.1Qcd-2015, IEEE standard 802.1Q-2014/Cor 1-2015), 2016. [87] IEEE Standard for Local and Metropolitan Area Networks–Bridges and Bridged Networks—Amendment 34:Asynchronous Traffic Shaping, IEEE Standard 802.1Qcr-2020 (Amendment to IEEE Standard 802.1Q- 2018 as amended by IEEE Standard 802.1Qcp-2018, IEEE Standard 802.1Qcc-2018, IEEE Standard 802.1Qcy-2019, IEEE Standard 802.1Qcx-2020), 2020. [88] J. Specht and S. Samii, “Urgency-based scheduler for time-sensitive switched ethernet networks,” in Proc. 28th Euromicro Conf. Real-Time Syst. (ECRTS), 2016, pp. 75–85. [89] IEEE Standard for Local and Metropolitan Area Networks—Bridges and Bridged Networks—Amendment 26: Frame Preemption, IEEE Standard 802.1Qbu-2016 (Amendment to IEEE Standard 802.1Q- 2014), 2016. [90] IEEE Standard for Ethernet Amendment 5: Specification and Management Parameters for Interspersing Express Traffic, IEEE Standard 802.3br-2016 (Amendment to IEEE Standard 802.3-2015 as amended by IEEE Standard 802.3bw-2015, IEEE Standard 802.3by- 2016, IEEE Standard 802.3bq-2016, IEEE Standard 802.3bp-2016), 2016. [91] A. Arestova, K.-S. J. Hielscher, and R. German, “Simulative evaluation of the TSN mechanisms time-aware shaper and frame preemption and their suitability for industrial use cases,” in Proc. IFIP Netw. Conf. (IFIP Netw.), 2021, pp. 1–6. [92] IEEE Standard for Local and Metropolitan Area Networks–Bridges and Bridged Networks–Amendment 29: Cyclic Queuing and Forwarding, IEEE 802.1Qch-2017 (Amendment to IEEE Standard 802.1Q-2014 as amended by IEEE Standard 802.1Qca-2015, IEEE Standard 802.1Qcd(TM)-2015, IEEE Standard 802.1Q-2014/Cor 1-2015, IEEE Standard 802.1Qbv-2015, IEEE Standard 802.1Qbu-2016, IEEE Standard 802.1Qbz-2016, IEEE Standard 802.1Qci-2017), 2017. [93] W. Quan, J. Yan, X. Jiang, and Z. Sun, “On-line traffic scheduling optimization in IEEE 802.1Qch based time-sensitive networks,” in Proc. IEEE 22nd Int. Conf. High Perform. Comput. Commun., IEEE 18th Int. Conf. Smart City, IEEE 6th Int. Conf. Data Sci. Syst. (HPCC/SmartCity/DSS), 2020, pp. 369–376. [94] B. Wang, F. Luo, and Z. Fang, “Performance analysis of IEEE 802.1 Qch for automotive networks: Compared with IEEE 802.1 Qbv,” in Proc. IEEE 4th Int. Conf. Comput. Commun. Eng. Technol. (CCET), 2021, pp. 355–359. [95] IEEE Standard for Local and Metropolitan Area Networks–Virtual Bridged Local Area Networks Amendment 14: Stream Reservation Protocol (SRP), IEEE Standard 802.1Qat-2010 (Revision of IEEE Standard 802.1Q-2005), 2010. [96] M. Gutiérrez, A. Ademaj, W. Steiner, R. Dobrin, and S. Punnekkat, “Self-configuration of IEEE 802.1 TSN networks,” in Proc. 22nd IEEE Int. Conf. Emerg. Technol. Fact. Autom. (ETFA), 2017, pp. 1–8. [97] IEEE Standard for Local and Metropolitan Area Networks–Bridges and Bridged Networks—Amendment 31: Stream Reservation Protocol (SRP) Enhancements and Performance Improvements, IEEE Standard 802.1Qcc-2018 (Amendment to IEEE Standard 802.1Q-2018 as amended by IEEE Standard 802.1Qcp-2018), 2018. [98] M. Bjorklund, “YANG-a data modeling language for the network configuration protocol (NETCONF),” Internet Eng. Task Force, RFC 6020, 2010. [99] R. Enns, M. Bjorklund, J. Schoenwaelder, and A. Bierman, “Network configuration protocol (NETCONF),” Internet Eng. Task Force, RFC 6241, Jun. 2011. [100] A. Bierman, M. Bjorklund, and K. Watsen, “RESTCONF protocol,” Internet Eng. Task Force, RFC 8040, Jan. 2020.