Industry 4.0 imposes strict requirements on Fifth Generation (5G) system, such as high reliability, availability, and low latency. Guaranteeing such requirements means that there are supposed to be a low number of system failures. Such rareness can cause problems when access to a broader range of these failures is necessary (e.g., finding optimal scheduler or learning in Deep Reinforcement Learning (DRL)). This work will investigate the possibility of using Generative Adversarial Network (GAN) to generate rare events in wireless communication data that might cause failure events. Conventional training methods fall short when trained on such a dataset, as they will overfit the common values while ignoring the rare values. We propose an alternative training method for GANs, called incremental learning, that aims at increasing learning in the rare sections without sacrificing the learning of the rest of the dataset. / Industry 4.0 ställer strikta krav på 5Gsystemet, såsom hög tillförlitlighet, tillgänglighet och låg latens. För att säkerställa uppfyllandet av kraven ovan på systemet, måste antalet systemfel vara sällsynta. I vissa fall som t.ex. skapandet av en optimal ”scheduler” eller inlärning av DRL kan det vara problematiskt att ha ett system med sällsynta systemfel. Detta är sant, eftersom det kommer att vara nödvändigt och nästintill ett krav att ha tillgång till ett brett utbud av systemfel. Denna studie kommer undersöka möjligheten att använda GAN för att generera sällsynta händelser i trådlös kommunikationsdata. Konventionell träning misslyckas när den tränas på en sådan datamängd, eftersom den kommer att vara överanpassad för de vanliga värdena samtidigt som de sällsynta värdena ignoreras. Vårt förslag är att använda en så kallad ”incremental learning” som en alternativ metod för GANs. Inom ”incremental learning” strävar man efter att öka inlärningen i de sällsynta fallen utan att offra inlärningen av de resterande datamängd.

Dans cette thèse, nous étudions plusieurs scénarios de communication pour les futurs réseaux sans fil. Plus particulièrement, cette thèse porte son attention sur comment la communication directe entre équipements mobiles (D2D) peut améliorer les performances des technologies existantes dans les systèmes sans fil. Le premier scénario étudié durant cette thèse est celui de la communication par multidiffusion d’un message commun entre un émetteur et plusieurs récepteurs. Il peut être illustré par le streaming vidéo, les messages d’alerte à destination de la police ou des pompiers ou des ambulanciers. Le second scénario étudié est celui d’une transmission à contraintes critiques en latence et en fiabilité. Ce dernier est illustré par son implication primordiale dans les futures technologies telles que les voitures connectées, avec pour but d’éviter des accidents, ou bien les machines connectées pour améliorer les services hospitaliers tels que la télé-chirurgie entre autres. Le dernier scénario étudié est celui de la localisation d’un groupe d’équipement dans un réseau densément peuplé tel qu’on peut trouver dans le contexte des objets connectés en masse. En général les objets communiquent entre eux à un niveau local et sont intéressés par des services communs et locaux. Plus concrètement, dans cette thèse, nous montrons les bienfaits de la communication D2D dans les trois scénarios précédents. Dans le cas du premier scénario de multidiffusion, contrairement à la tendance habituelle d’avoir un taux de transmission qui diminue en fonction du nombre d’équipements mobiles (en particulier, car l’équipement émetteur doit adapter sa transmission à l’équipement récepteur en plus mauvaise condition), en ajoutant la communication D2D, on observe que ce même taux de transmission augmente en fonction du nombre d’équipements mobiles présents. Dans le deuxième scénario où la communication est soumise à des contraintes de fiabilité et de latence exigeantes, nous déduisons une politique de retransmission optimale et proposons une autre politique semi-optimale qui est beaucoup moins gourmande en temps et qui a prouvé son optimalité dans plusieurs cas pratiques. Enfin dans le dernier scénario, nous proposons une méthode de localisation d’équipements mobile et l’étudions dans plusieurs environnements (avec et sans visibilité directe dans les cas intra-muros et extérieurs). L’identification de ces zones est ensuite utilisée pour créer de petites cellules virtuelles adaptatives aux situations changeantes et non prédictibles, dans le but de réduire les coûts liés aux infrastructures actuelles. / This thesis studies D2D communication in realistic and challenging scenarios for future wireless systems. In particular, the thesis focuses on how may D2D communication help other technologies to enhance their performance. The first wireless scenario is the one of multicasting, used for example in video streaming or common alert message transmission for police, firefighters or ambulances. The second wireless scenario is the critical one of URLLC expected to be used to avoid cars crashes in the upcoming V2X context, and also when connecting machines together in environments like connected hospitals, airports, factories (industry 4.0), and last but not least in e-health context in order to enhance medical tele-surgery. The last wireless scenario is the one of UE group localization in the context of massive IoT, where devices are interacting with each other and are mostly confined in local groups, needing local services. In the multicast channel scenario, where a transmitter wishes to convey a common message to many receivers, it is known that the multicast rate decrease as the number of UEs increases. This vanishing behavior changes drastically when enabling the receivers to cooperate with each other via D2D. Indeed, the multicast rate increases with high probability when the number of receivers increases. This chapter also analyzes the outage rate of the proposed scheme in the same setting. Extensions regarding firstly resource utilization and secondly considering the use of HARQ are also analyzed. Next chapter addresses one of the major challenges for future networks, named URLLC. Specifically, the chapter studies the problem of HARQ with delayed feedback, where the transmitter is informed after some delay on whether or not his transmission was successful. The goal is to minimize the expected number of retransmissions subject to a reliability constraint within a delay budget. This problem is studied at two levels: (i) a single transmitter faced with a stochastic i.i.d. noisy environment and (ii) a group of transmitters whom shares a collision channel. Then the chapter that follows provides a cooperative UE mapping method that is highly accurate. Four different channel models are studied in this chapter: LOS and NLOS for indoor and outdoor environments. The results show significant improvement compared to already existing methods. Identifying the dense local areas in real time and informing the network allows the Base Station (BS) to increase the capacity through highly directive beams, and therefore, avoids the deployment cost of new infrastructure.

In 5G, Ultra-reliable and low-Latency communications (URLLC) service is envisioned to enable use cases with strict reliability and latency requirements on wireless communication. For the upcoming 6G network, machine learning (ML) also stands an important role that introduces intelligence and further enhances the system performance. This thesis explores the deployment of reinforcement learning (RL), a popular sub-field of ML, to optimize the application-layer availability and reliability of URLLC service in factory automation scenarios. In conventional RL methods, the decision variables are typically optimized in the same control loop. However, wireless systems’ parameters can be optimized either on a cell level or globally, depending on the inter-cell dynamics’ impact on their optimal value. Although global optimizations can provide a better performance, such optimizations introduce major practical limitations on the control loop’s delay. Besides, global optimization of all decision variables leads to excessive signalings, and thus, it is costly in terms of communication overhead. In this thesis, we propose a more flexible hierarchical reinforcement learning (HRL) framework that enables the implementation of multiple agents and multi-level policies with different time scales for each optimization. Therefore, we selected a use case from the prior art, optimizing the maximum number of retransmissions and transmission power to industrial devices, and solved it with our HRL framework. Our simulation results on factory automation scenario shows that HRL framework achieves similar performance as the ideal RL method, which highly improves the availability and reliability compared to the baseline solutions. Besides, the new HRL framework allows a more flexible allocation of agents. By allocating the low-level agents close to the base stations, our framework also significantly decreases the overhead of signal transmissions compared to the one-agent RL method. / Inom 5G är tjänster kallade “Ultra-reliable and low-latency communication” URLLC tänkta att möjliggöra trådlös kommunikation i användningsfall med strikta krav på tillförlitlighet och latens. För framtidens 6G nätverk har även maskininlärning ML en viktig roll som introducerar intelligens och ytterligare förbättrar systemens prestanda. Den här avhandlingen utforskar implementeringen av förstärkande inlärning (reinforcement learning eller RL), ett populärt underområde av ML, för att optimera tillgängligheten och tillförlitligheten av URLLC-tjänster i automatiserade fabriker. I traditionella RL-metoder optimeras beslutsvariablerna vanligtvis i samma kontrollslinga. Parametrarna för trådlösa system kan dock optimeras antingen på cellnivå eller globalt, beroende på inverkan av dynamiken mellan cellerna på deras optimala värde. Även om globala optimeringar kan ge bättre prestanda introducerar sådana optimeringar stora praktiska begränsningar på kontrollslingans latens. Dessutom leder global optimering av beslutsvariablerna till ökad signalering och är därför kostsamt. I denna avhandling föreslår vi ett mer flexibelt ramverk med hierarkisk förstärkande inlärning HRL som möjliggör implementering av flera agenter och flernivå-policys med olika tidsskalor för varje optimering. Därför valde vi ett tidigare känt användningsfall, optimeringen av det maximala antalet återsändningar samt överföringseffekten till industriella enheter, och löste det med vårt HRL ramverk. Resultaten från våra simuleringar på fabriksscenariot visar att HRL-ramverket uppnår liknande prestanda som den ideala RL-metoden, vilket i hög grad förbättrar tillgängligheten och tillförlitligheten jämfört med standardlösningarna. Dessutom tillåter det nya HRL ramverket en mer flexibel fördelning av agenter. Genom att allokera lågnivåagenterna nära basstationerna minskar vårt ramverk också avsevärt kostnaden för signalöverföringar jämfört med RL-metoden med endast en agent.

To enable wireless control of factories, such that sensor measurements can be sent wirelessly to an actuator, the probability to receive data correctly must be very high and the time it takes to the deliver the data from the sensor to the actuator must be very low. Earlier, these requirements have only been met by cables, but in the fifth generation mobile network this is one of the imagined use cases and work is undergoing to create a system capable of wireless control of factories. One of the problems in this scenario is when all data in a packet cannot be sent in one transmission while ensuring the very high probability of reception of the transmission. This thesis studies this problem in detail by proposing methods to cope with the problem and evaluating these methods in a simulator. The thesis shows that splitting the data into multiple segments and transmitting each at an even higher probability of reception is a good candidate, especially when there is time for a retransmission. When there is only one transmission available, a better candidate is to send the same packet twice. Even if the first packet cannot achieve the very high probability of reception, the combination of the first and second packet might be able to. / För att möjliggöra trådlös kontroll av fabriker, till exempel trådlös sändning av data uppmätt av en sensor till ett ställdon som agerar på den emottagna signalen, så måste sannolikheten att ta emot datan korrekt vara väldigt hög och tiden det tar att leverera data från sensorn till ställdonet vara mycket kort. Tidigare har endast kablar klarat av dessa krav men i den femte generationens mobila nätverk är trådlös kontroll av fabriker ett av användningsområdena och arbete pågår för att skapa ett system som klarar av det. Ett av problemen i detta användningsområde är när all data i ett paket inte kan skickas i en sändning och klara av den väldigt höga sannolikheten för mottagning. Denna uppsats studerar detta problem i detalj och föreslår metoder för att hantera problemet samt utvärderar dessa metoder i en simulator. Uppsatsen visar att delning av ett paket i flera segment och sändning av varje segment med en ännu högre sannolikhet för mottagning är en bra kandidat, speciellt när det finns tid för en omsändning. När det endast finns tid för en sändning verkar det bättre att skicka samma paket två gånger. Även om det första paketet inte kan uppnå den höga sannolikheten för mottagning så kan kanske kombinationen av det första och andra paketet göra det.

Unter dem Überbegriff Industrie 4.0 wird in der industriellen Fertigung die zunehmende Digitalisierung und Vernetzung von industriellen Maschinen und Prozessen zusammengefasst. Die drahtlose, hoch-zuverlässige, niedrig-latente Kommunikation (engl. ultra-reliable low-latency communication, URLLC) – als Bestandteil von 5G gewährleistet höchste Dienstgüten, die mit industriellen drahtgebundenen Technologien vergleichbar sind und wird deshalb als Wegbereiter von Industrie 4.0 gesehen. Entgegen diesem Trend haben eine Reihe von Arbeiten im Forschungsbereich der vernetzten Regelungssysteme (engl. networked control systems, NCS) gezeigt, dass die hohen Dienstgüten von URLLC nicht notwendigerweise erforderlich sind, um eine hohe Regelgüte zu erzielen. Das Co-Design von Kommunikation und Regelung ermöglicht eine gemeinsame Optimierung von Regelgüte und Netzwerkparametern durch die Aufweichung der Grenze zwischen Netzwerk- und Applikationsschicht. Durch diese Verschränkung wird jedoch eine fundamentale (gemeinsame) Neuentwicklung von Regelungssystemen und Kommunikationsnetzen nötig, was ein Hindernis für die Verbreitung dieses Ansatzes darstellt. Stattdessen bedient sich diese Dissertation einem Co-Design-Ansatz, der beide Domänen weiterhin eindeutig voneinander abgrenzt, aber das Informationsalter (engl. age of information, AoI) als bedeutenden Schnittstellenparameter ausnutzt. Diese Dissertation trägt dazu bei, die Echtzeitanwendungszuverlässigkeit als Folge der Überschreitung eines vorgegebenen Informationsalterschwellenwerts zu quantifizieren und fokussiert sich dabei auf den Paketverlust als Ursache. Anhand der Beispielanwendung eines fahrerlosen Transportsystems wird gezeigt, dass die zeitlich negative Korrelation von Paketfehlern, die in heutigen Systemen keine Rolle spielt, für Echtzeitanwendungen äußerst vorteilhaft ist. Mit der Annahme von schnellem Schwund als dominanter Fehlerursache auf der Luftschnittstelle werden durch zeitdiskrete Markovmodelle, die für die zwei Netzwerkarchitekturen Single-Hop und Dual-Hop präsentiert werden, Kommunikationsfehlerfolgen auf einen Applikationsfehler abgebildet. Diese Modellierung ermöglicht die analytische Ableitung von anwendungsbezogenen Zuverlässigkeitsmetriken wie die durschnittliche Dauer bis zu einem Fehler (engl. mean time to failure). Für Single-Hop-Netze wird das neuartige Ressourcenallokationsschema State-Aware Resource Allocation (SARA) entwickelt, das auf dem Informationsalter beruht und die Anwendungszuverlässigkeit im Vergleich zu statischer Multi-Konnektivität um Größenordnungen erhöht, während der Ressourcenverbrauch im Bereich von konventioneller Einzelkonnektivität bleibt. Diese Zuverlässigkeit kann auch innerhalb eines Systems von Regelanwendungen, in welchem mehrere Agenten um eine begrenzte Anzahl Ressourcen konkurrieren, statistisch garantiert werden, wenn die Anzahl der verfügbaren Ressourcen pro Agent um ca. 10 % erhöht werden. Für das Dual-Hop Szenario wird darüberhinaus ein Optimierungsverfahren vorgestellt, das eine benutzerdefinierte Kostenfunktion minimiert, die niedrige Anwendungszuverlässigkeit, hohes Informationsalter und hohen durchschnittlichen Ressourcenverbrauch bestraft und so das benutzerdefinierte optimale SARA-Schema ableitet. Diese Optimierung kann offline durchgeführt und als Look-Up-Table in der unteren Medienzugriffsschicht zukünftiger industrieller Drahtlosnetze implementiert werden.:1. Introduction 1 1.1. The Need for an Industrial Solution . . . . . . . . . . . . . . . . . . . 3 1.2. Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2. Related Work 7 2.1. Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2. Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3. Codesign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3.1. The Need for Abstraction – Age of Information . . . . . . . . 11 2.4. Dependability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3. Deriving Proper Communications Requirements 17 3.1. Fundamentals of Control Theory . . . . . . . . . . . . . . . . . . . . 18 3.1.1. Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 3.1.2. Performance Requirements . . . . . . . . . . . . . . . . . . . 21 3.1.3. Packet Losses and Delay . . . . . . . . . . . . . . . . . . . . . 22 3.2. Joint Design of Control Loop with Packet Losses . . . . . . . . . . . . 23 3.2.1. Method 1: Reduced Sampling . . . . . . . . . . . . . . . . . . 23 3.2.2. Method 2: Markov Jump Linear System . . . . . . . . . . . . . 25 3.2.3. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.3. Focus Application: The AGV Use Case . . . . . . . . . . . . . . . . . . 31 3.3.1. Control Loop Model . . . . . . . . . . . . . . . . . . . . . . . 31 3.3.2. Control Performance Requirements . . . . . . . . . . . . . . . 33 3.3.3. Joint Modeling: Applying Reduced Sampling . . . . . . . . . . 34 3.3.4. Joint Modeling: Applying MJLS . . . . . . . . . . . . . . . . . 34 3.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 4. Modeling Control-Communication Failures 43 4.1. Communication Assumptions . . . . . . . . . . . . . . . . . . . . . . 43 4.1.1. Small-Scale Fading as a Cause of Failure . . . . . . . . . . . . 44 4.1.2. Connectivity Models . . . . . . . . . . . . . . . . . . . . . . . 46 4.2. Failure Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 4.2.1. Single-hop network . . . . . . . . . . . . . . . . . . . . . . . . 49 4.2.2. Dual-hop network . . . . . . . . . . . . . . . . . . . . . . . . 51 4.3. Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.3.1. Mean Time to Failure . . . . . . . . . . . . . . . . . . . . . . . 54 4.3.2. Packet Loss Ratio . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.3.3. Average Number of Assigned Channels . . . . . . . . . . . . . 57 4.3.4. Age of Information . . . . . . . . . . . . . . . . . . . . . . . . 57 4.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 5. Single Hop – Single Agent 61 5.1. State-Aware Resource Allocation . . . . . . . . . . . . . . . . . . . . 61 5.2. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 5.3. Erroneous Acknowledgments . . . . . . . . . . . . . . . . . . . . . . 67 5.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 6. Single Hop – Multiple Agents 71 6.1. Failure Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 6.1.1. Admission Control . . . . . . . . . . . . . . . . . . . . . . . . 72 6.1.2. Transition Probabilities . . . . . . . . . . . . . . . . . . . . . . 73 6.1.3. Computational Complexity . . . . . . . . . . . . . . . . . . . 74 6.1.4. Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . 75 6.2. Illustration Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 6.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 6.3.1. Verification through System-Level Simulation . . . . . . . . . 78 6.3.2. Applicability on the System Level . . . . . . . . . . . . . . . . 79 6.3.3. Comparison of Admission Control Schemes . . . . . . . . . . 80 6.3.4. Impact of the Packet Loss Tolerance . . . . . . . . . . . . . . . 82 6.3.5. Impact of the Number of Agents . . . . . . . . . . . . . . . . . 84 6.3.6. Age of Information . . . . . . . . . . . . . . . . . . . . . . . . 84 6.3.7. Channel Saturation Ratio . . . . . . . . . . . . . . . . . . . . 86 6.3.8. Enforcing Full Channel Saturation . . . . . . . . . . . . . . . 86 6.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 7. Dual Hop – Single Agent 91 7.1. State-Aware Resource Allocation . . . . . . . . . . . . . . . . . . . . 91 7.2. Optimization Targets . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 7.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 7.3.1. Extensive Simulation . . . . . . . . . . . . . . . . . . . . . . . 96 7.3.2. Non-Integer-Constrained Optimization . . . . . . . . . . . . . 98 7.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 8. Conclusions and Outlook 105 8.1. Key Results and Conclusions . . . . . . . . . . . . . . . . . . . . . . . 105 8.2. Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 A. DC Motor Model 111 Bibliography 113 Publications of the Author 127 List of Figures 129 List of Tables 131 List of Operators and Constants 133 List of Symbols 135 List of Acronyms 137 Curriculum Vitae 139 / In industrial manufacturing, Industry 4.0 refers to the ongoing convergence of the real and virtual worlds, enabled through intelligently interconnecting industrial machines and processes through information and communications technology. Ultrareliable low-latency communication (URLLC) is widely regarded as the enabling technology for Industry 4.0 due to its ability to fulfill highest quality-of-service (QoS) comparable to those of industrial wireline connections. In contrast to this trend, a range of works in the research domain of networked control systems have shown that URLLC’s supreme QoS is not necessarily required to achieve high quality-ofcontrol; the co-design of control and communication enables to jointly optimize and balance both quality-of-control parameters and network parameters through blurring the boundary between application and network layer. However, through the tight interlacing, this approach requires a fundamental (joint) redesign of both control systems and communication networks and may therefore not lead to short-term widespread adoption. Therefore, this thesis instead embraces a novel co-design approach which keeps both domains distinct but leverages the combination of control and communications by yet exploiting the age of information (AoI) as a valuable interface metric. This thesis contributes to quantifying application dependability as a consequence of exceeding a given peak AoI with the particular focus on packet losses. The beneficial influence of negative temporal packet loss correlation on control performance is demonstrated by means of the automated guided vehicle use case. Assuming small-scale fading as the dominant cause of communication failure, a series of communication failures are mapped to an application failure through discrete-time Markov models for single-hop (e.g, only uplink or downlink) and dual-hop (e.g., subsequent uplink and downlink) architectures. This enables the derivation of application-related dependability metrics such as the mean time to failure in closed form. For single-hop networks, an AoI-aware resource allocation strategy termed state-aware resource allocation (SARA) is proposed that increases the application reliability by orders of magnitude compared to static multi-connectivity while keeping the resource consumption in the range of best-effort single-connectivity. This dependability can also be statistically guaranteed on a system level – where multiple agents compete for a limited number of resources – if the provided amount of resources per agent is increased by approximately 10 %. For the dual-hop scenario, an AoI-aware resource allocation optimization is developed that minimizes a user-defined penalty function that punishes low application reliability, high AoI, and high average resource consumption. This optimization may be carried out offline and each resulting optimal SARA scheme may be implemented as a look-up table in the lower medium access control layer of future wireless industrial networks.:1. Introduction 1 1.1. The Need for an Industrial Solution . . . . . . . . . . . . . . . . . . . 3 1.2. Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2. Related Work 7 2.1. Communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 2.2. Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3. Codesign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 2.3.1. The Need for Abstraction – Age of Information . . . . . . . . 11 2.4. Dependability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 2.5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 3. Deriving Proper Communications Requirements 17 3.1. Fundamentals of Control Theory . . . . . . . . . . . . . . . . . . . . 18 3.1.1. Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 3.1.2. Performance Requirements . . . . . . . . . . . . . . . . . . . 21 3.1.3. Packet Losses and Delay . . . . . . . . . . . . . . . . . . . . . 22 3.2. Joint Design of Control Loop with Packet Losses . . . . . . . . . . . . 23 3.2.1. Method 1: Reduced Sampling . . . . . . . . . . . . . . . . . . 23 3.2.2. Method 2: Markov Jump Linear System . . . . . . . . . . . . . 25 3.2.3. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.3. Focus Application: The AGV Use Case . . . . . . . . . . . . . . . . . . 31 3.3.1. Control Loop Model . . . . . . . . . . . . . . . . . . . . . . . 31 3.3.2. Control Performance Requirements . . . . . . . . . . . . . . . 33 3.3.3. Joint Modeling: Applying Reduced Sampling . . . . . . . . . . 34 3.3.4. Joint Modeling: Applying MJLS . . . . . . . . . . . . . . . . . 34 3.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41 4. Modeling Control-Communication Failures 43 4.1. Communication Assumptions . . . . . . . . . . . . . . . . . . . . . . 43 4.1.1. Small-Scale Fading as a Cause of Failure . . . . . . . . . . . . 44 4.1.2. Connectivity Models . . . . . . . . . . . . . . . . . . . . . . . 46 4.2. Failure Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 4.2.1. Single-hop network . . . . . . . . . . . . . . . . . . . . . . . . 49 4.2.2. Dual-hop network . . . . . . . . . . . . . . . . . . . . . . . . 51 4.3. Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 4.3.1. Mean Time to Failure . . . . . . . . . . . . . . . . . . . . . . . 54 4.3.2. Packet Loss Ratio . . . . . . . . . . . . . . . . . . . . . . . . . 55 4.3.3. Average Number of Assigned Channels . . . . . . . . . . . . . 57 4.3.4. Age of Information . . . . . . . . . . . . . . . . . . . . . . . . 57 4.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 5. Single Hop – Single Agent 61 5.1. State-Aware Resource Allocation . . . . . . . . . . . . . . . . . . . . 61 5.2. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 5.3. Erroneous Acknowledgments . . . . . . . . . . . . . . . . . . . . . . 67 5.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 6. Single Hop – Multiple Agents 71 6.1. Failure Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 6.1.1. Admission Control . . . . . . . . . . . . . . . . . . . . . . . . 72 6.1.2. Transition Probabilities . . . . . . . . . . . . . . . . . . . . . . 73 6.1.3. Computational Complexity . . . . . . . . . . . . . . . . . . . 74 6.1.4. Performance Metrics . . . . . . . . . . . . . . . . . . . . . . . 75 6.2. Illustration Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 6.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78 6.3.1. Verification through System-Level Simulation . . . . . . . . . 78 6.3.2. Applicability on the System Level . . . . . . . . . . . . . . . . 79 6.3.3. Comparison of Admission Control Schemes . . . . . . . . . . 80 6.3.4. Impact of the Packet Loss Tolerance . . . . . . . . . . . . . . . 82 6.3.5. Impact of the Number of Agents . . . . . . . . . . . . . . . . . 84 6.3.6. Age of Information . . . . . . . . . . . . . . . . . . . . . . . . 84 6.3.7. Channel Saturation Ratio . . . . . . . . . . . . . . . . . . . . 86 6.3.8. Enforcing Full Channel Saturation . . . . . . . . . . . . . . . 86 6.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 7. Dual Hop – Single Agent 91 7.1. State-Aware Resource Allocation . . . . . . . . . . . . . . . . . . . . 91 7.2. Optimization Targets . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 7.3. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 7.3.1. Extensive Simulation . . . . . . . . . . . . . . . . . . . . . . . 96 7.3.2. Non-Integer-Constrained Optimization . . . . . . . . . . . . . 98 7.4. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 8. Conclusions and Outlook 105 8.1. Key Results and Conclusions . . . . . . . . . . . . . . . . . . . . . . . 105 8.2. Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 A. DC Motor Model 111 Bibliography 113 Publications of the Author 127 List of Figures 129 List of Tables 131 List of Operators and Constants 133 List of Symbols 135 List of Acronyms 137 Curriculum Vitae 139

A key distinction between today's and tomorrow's wireless networks is the appetite for reliability to enable emerging mission-critical services such as ultra-reliable low-latency communication (URLLC) and hyper-reliable low-latency communication (HRLLC), the staple mission-critical services in IMT-2020 (5G) and IMT-2023 (6G), for which reliable and resilient communication is a must. However, achieving ultra-reliable communication is challenging because of these services' stringent reliability and latency requirements and the stochastic nature of wireless networks. A natural way of increasing reliability and reducing latency is to provision additional network resources to compensate for uncertainty in wireless networks caused by fading, interference, mobility, and time-varying network load, among others. Thus, an important step to enable mission-critical services is to identify and quantify what it takes to support ultra-reliable communication in mobile networks -- a process often referred to as dimensioning. This dissertation focuses on resource dimensioning, notably spectrum, for ultra-reliable wireless communication. This dissertation proposes a set of methods for spectrum dimensioning based on concepts from risk analysis, extreme value theory, and meta distributions. These methods reveal that each ``nine'' in reliability (e.g., five-nines in 99.999%) roughly translates into an order of magnitude increase in the required bandwidth. In ultra-reliability regimes, the required bandwidth can be in the order of tens of gigahertz, far beyond what is typically available in today's networks, making it challenging to provision resources for ultra-reliable communication. Accordingly, this dissertation also investigates alternative approaches to provide resources to enable ultra-reliable communication services in mobile networks. Particularly, this dissertation considers multi-operator network sharing and multi-connectivity as alternatives to make additional network resources available to enhance network reliability and proposes multi-operator connectivity sharing, which combines multi-operator network sharing with multi-connectivity. Our studies, based on simulations, real-world data analysis, and mathematical models, suggest that multi-operator connectivity sharing -- in which mobiles multi-connect to base stations of operators in a sharing arrangement -- can reduce the required bandwidth significantly because underlying operators tend to exhibit characteristics attractive to reliability, such as complementary coverage during periods of impaired connectivity, facilitating the support for ultra-reliable communication in future mobile networks. / Doctor of Philosophy / A key distinction between today's and tomorrow's wireless networks is the appetite for reliability to enable emerging mission-critical services in 5G and 6G, for which ultra-reliable communication is a must. However, achieving ultra-reliable communication is challenging because of these services' stringent reliability and latency requirements and the stochastic nature of wireless networks. Reliability often comes at the cost of additional network resources to compensate for uncertainty in wireless networks. Thus, an important step to enable ultra-reliable communication is to identify and quantify what it takes to support mission-critical services in mobile networks -- a process often denoted as dimensioning. This dissertation focuses on spectrum dimensioning and proposes a set of methods to identify suitable spectrum bands and required bandwidth for ultra-reliable communication. These methods reveal that the spectrum needs for ultra-reliable communication can be beyond what is typically available in today's networks, making it challenging to provide adequate resources to support ultra-reliable communication services in mobile networks. Alternatively, we propose multi-operator connectivity sharing: mobiles simultaneously connect to multiple base stations of different operators. Our studies suggest that multi-operator connectivity sharing can reduce the spectrum needs in ultra-reliability regimes significantly, being an attractive alternative to enable ultra-reliable communication in future mobile networks.

The advancement of technology has led to an increase in the demand for ultra-low end-to-end network latency in real-time applications with a target of below 10ms. The IEEE 802.1 Time-Sensitive Networking (TSN) is a set of standards that supports the required low-latency wired communication with ultra-low jitter for real-time applications. TSN is designed for fixed networks thus it misses the flexibility of wireless networks.To overcome this limitation and to increase its applicability in different applications, an integration of TSN with other wireless technologies is needed. The fifth generation of cellular networks (5G) supports real-time applications with its Ultra-Reliable Low Latency Communication (URLLC) service. 5G URLLC is designed to meet the stringent timing requirements of these applications, such as providing reliable communication with latencies as low as 1ms. Seamless integration of TSN and 5G is needed to fully utilize the potential of these technologies in contemporary and future industrial applications. However, to achieve the end-to-end Quality of Service (QoS) requirements of a TSN-5G network, a significant effort is required due to the large dissimilarity between these technologies. This thesis presents a comprehensive and well-structured snapshot of the existing research on TSN-5G integration that identifies gaps in the current research and highlights the opportunities for further research in the area of TSN-5G integration. In particular, the thesis identifies that the state of the art lacks an end-to-end mapping of QoS requirements and traffic forwarding mechanisms for a converged TSN-5G network. This lack of knowledge and tool support hampers the utilisation of ground-breaking technologies like TSN and 5G. Hence, the thesis develops novel techniques to support the end-to-end QoS mapping and traffic forwarding of a converged TSN-5G network for predictable communication.Furthermore, the thesis presents a translation technique between TSN and 5G with a proof-of-concept implementation in a well-known TSN network simulator. Moreover, a novel QoS mapping algorithm is proposed to support the systematic mapping of QoS characteristics and integration of traffic flows in a converged TSN-5G network. / PROVIDENT

Future mobile networks are anticipated to not only enhance communication performance but also facilitate new sensing-based applications. This highlights the essential role of integrated sensing and communication (ISAC) in sixth-generation (6G) and beyond mobile networks. The seamless integration of sensing and communication poses challenges in deployment and resource allocation. Cell-free massive multiple-input multiple-output (MIMO) networks, characterized by multiple distributed access points, offer a promising infrastructure for ISAC implementation. However, the effective realization of ISAC necessitates joint design and resource allocation optimization. In this thesis, we study ISAC within cell-free massive MIMO systems, with a particular emphasis on developing power allocation algorithms under various scenarios. In this thesis, we explore two scenarios: utilizing existing communication signals and incorporating additional sensing signals. We propose power allocation algorithms aiming to maximize the sensing performance while meeting communication and power constraints. In addition, we develop two maximum a posteriori ratio test (MAPRT) target detectors under clutter-free and cluttered scenarios. Results indicate that employing additional sensing signals enhances sensing performance, particularly in scenarios where the target has low reflectivity. Moreover, although the clutter-aware detector requires more advanced processing, it leads to better sensing performance. Furthermore, we introduced sensing spectral efficiency (SE) to measure the effect of resource block utilization, highlighting the integration advantages of ISAC over orthogonal resource sharing approaches.  In the next part of the thesis, we study the energy efficiency aspects of ISAC in cell-free massive MIMO systems with ultra-reliable low-latency communications (URLLC) users. We propose a power allocation algorithm aiming to maximize energy efficiency of the system while meeting communication and sensing requirements. We conduct a comparative analysis between the proposed power allocation algorithms and a URLLC-only approach which takes into account only URLLC and power requirements. The results reveal that while the URLLC-only algorithm excels in energy efficiency, it is not able to support sensing requirements.   Moreover, we study the impact of ISAC on end-to-end (including radio and processing) energy consumption. Particularly, we present giga-operations per second (GOPS) analysis for both communication and sensing tasks. Two optimization problems are formulated and solved to minimize transmission and end-to-end energy through blocklength and power optimization. Results indicate that while end-to-end energy minimization offers substantial energy savings, its efficacy diminishes with sensing integration due to processing energy requirements. / Framtida mobila nätverk förväntas inte bara förbättra kommunikations-prestanda utan även mögliggöra nya applikationer baserade på sensorer. Dettaunderstryker den avgörande rollen för Integrerad avkänning och kommunika-tion (ISAC) i sjätte generationens (6G) och efterföljande mobila nätverk. Densömlösa integrationen av sensorer och kommunikation medför utmaningar iutrullning och resursallokering. Cellfria massiva flerantennsystem (MIMO-nätverk), kännetecknade av flera distribuerade åtkomstpunkter, erbjuder enlovande infrastruktur för implementering av ISAC. Dock kräver den effektivarealiseringen av ISAC samverkande design och optimering av resursallokering.I denna avhandling studerar vi ISAC inom cellfria massiva MIMO-system,med särskild tonvikt på att utveckla effektallokeringsalgoritmer under olikascenarier.Vi utforskar två scenarier: att utnyttja befintliga kommunikationssignaleroch att inkludera ytterligare sensorssignaler. Vi föreslår effektallokeringsalgo-ritmer med målet att maximera sensorsprestandan samtidigt som kommunika-tions och effektbegränsningar uppfylls. Dessutom utvecklar vi två detektorerbaserade på maximum a posteriori ratio test (MAPRT) under störningsfriaoch störda scenarier. Resultaten visar att användning av ytterligare sensors-signaler förbättrar sensorsprestandan, särskilt i scenarier där målet har lågreflektivitet. Dessutom, även om den störkänsliga detektorn kräver mer avan-cerad bearbetning, leder den till bättre sensorsprestanda. Vidare introducerarvi sensorerspektral effektivitet (SE) för att mäta effekten av resursblocksan-vändning och framhäva integrationsfördelarna med ISAC över ortogonala re-sursdelningsmetoder.I den andra delen av avhandlingen studerar vi energieffektivitetsaspek-terna av ISAC i cellfria massiva MIMO-system med användare med ultra-tillförlitlig låg-latens (URLLC) kommunikation. Vi föreslår en effektalloke-ringsalgoritm med syfte att maximera systemets energieffektivitet samtidigtsom kommunikations- och sensorskraven uppfylls. Vi utför en jämförande ana-lys mellan de föreslagna effektallokeringsalgoritmerna och ett URLLC-ensamttillvägagångssätt som tar hänsyn enbart till URLLC- och effektkrav. Resul-taten avslöjar att medan URLLC-ensamma algoritmen utmärker sig i energi-effektivitet, kan den inte stödja sensorskraven. Dessutom studerar vi effektenav ISAC på slut till slut (inklusive radios och bearbetning) energiförbruk-ning. Särskilt presenterar vi giga-operationer per sekund (GOPS) analys förbåde kommunikations- och sensorsuppgifter. Två optimeringsproblem formu-leras och löses för att minimera överförings- och slut till slut energi genomblocklängd- och effektoptimering. Resultaten indikerar att medan slut till slutenergiminimering erbjuder betydande energibesparingar, minskar dess effek-tivitet med sensorintegrationen på grund av bearbetningsenergikrav. / <p>QC 20240513</p>

Abstract Future cellular networks need to support the ever-increasing demand of bandwidth-intensive applications and interconnection of people, devices, and vehicles. Small cell network (SCN)-based communication together with proximity- and social-aware connectivity is conceived as a vital component of these networks to enhancing spectral efficiency, system capacity, and quality-of-experience (QoE). To cope with diverse application needs for the heterogeneous ecosystem, radio resource management (RRM) is one of the key research areas for the fifth-generation (5G) network. The key goals of this thesis are to develop novel, self-organizing, and low-complexity resource management algorithms for emerging device-to-device (D2D) and vehicle-to-vehicle (V2V) wireless systems while explicitly modeling and factoring network contextual information to satisfy the increasingly stringent requirements. Towards achieving this goal, this dissertation makes a number of key contributions. First, the thesis focuses on interference management techniques for D2D-enabled macro network and D2D-enabled SCNs in the downlink, while leveraging users’ social-ties, dynamic clustering, and user association mechanisms for network capacity maximization. A flexible social-aware user association technique is proposed to maximize network capacity. The second contribution focuses on ultra-reliable low-latency communication (URLLC) in vehicular networks in which interference management and resource allocation techniques are investigated, taking into account traffic and network dynamics. A joint power control and resource allocation mechanism is proposed to minimize the total transmission power while satisfying URLLC constraints. To overcome these challenges, novel algorithms are developed by combining several methodologies from graph theory, matching theory and Lyapunov optimization. Extensive simulations validate the performance of the proposed approaches, outperforming state-of-the-art solutions. Notably, the results yield significant performance gains in terms of capacity, delay reductions, and improved reliability as compared with conventional approaches. / Tiivistelmä Tulevaisuuden solukkoverkkojen pitää pystyä tukemaan yhä suurempaa kaistanleveyttä vaativia sovelluksia sekä yhteyksiä ihmisten, laitteiden ja ajoneuvojen välillä. Piensoluverkkoihin (SCN) pohjautuvaa tietoliikennettä yhdistettynä paikka- ja sosiaalisen tietoisuuden huomioiviin verkkoratkaisuihin pidetään yhtenä elintärkeänä osana tulevaisuuden solukkoverkkoja, joilla pyritään tehostamaan spektrinkäytön tehokkuutta, järjestelmän kapasiteettia sekä kokemuksen laatua (QoE). Radioresurssien hallinta (RRM) on eräs keskeisistä viidennen sukupolven (5G) verkkoihin liittyvistä tutkimusalueista, joilla pyritään hallitsemaan heterogeenisen ekosysteemin vaihtelevia sovellustarpeita. Tämän väitöstyön keskeisinä tavoitteina on kehittää uudenlaisia itseorganisoituvia ja vähäisen kompleksisuuden resurssienhallinta-algoritmeja laitteesta-laitteeseen (D2D) ja ajoneuvosta-ajoneuvoon (V2V) toimiville uusille langattomille järjestelmille, sekä samalla mallintaa ja tuottaa verkon kontekstikohtaista tietoa vastaamaan koko ajan tiukentuviin vaatimuksiin. Tämä väitöskirja edistää näiden tavoitteiden saavuttamista usealla keskeisellä tuloksella. Aluksi väitöstyössä keskitytään häiriönhallinnan tekniikoihin D2D:tä tukevissa makroverkoissa ja laskevan siirtotien piensoluverkoissa. Käyttäjän sosiaalisia yhteyksiä, dynaamisia ryhmiä sekä osallistamismekanismeja hyödynnetään verkon kapasiteetin maksimointiin. Verkon kapasiteettia voidaan kasvattaa käyttämällä joustavaa sosiaaliseen tietoisuuteen perustuvaa osallistamista. Toinen merkittävä tulos keskittyy huippuluotettavaan lyhyen viiveen kommunikaatioon (URLLC) ajoneuvojen verkoissa, joissa tehtävää resurssien allokointia ja häiriönhallintaa tutkitaan liikenteen ja verkon dynamiikka huomioiden. Yhteistä tehonsäädön ja resurssien allokoinnin mekanismia ehdotetaan kokonaislähetystehon minimoimiseksi samalla, kun URLLC rajoitteita noudatetaan. Jotta esitettyihin haasteisiin voidaan vastata, väitöstyössä on kehitetty uudenlaisia algoritmeja yhdistämällä graafi- ja sovitusteorioiden sekä Lyapunovin optimoinnin menetelmiä. Laajat tietokonesimuloinnit vahvistavat ehdotettujen lähestymistapojen suorituskyvyn, joka on parempi kuin uusimmilla nykyisillä ratkaisuilla. Tulokset tuovat merkittäviä suorituskyvyn parannuksia erityisesti kapasiteetin lisäämisen, viiveiden vähentämisen ja parantuneen luotettavuuden suhteen verrattuna perinteisiin lähestymistapoihin.

Edge Computing is considered one of the key capabilities in next generation (5G) networks, which will enable inundation of latency, throughput, and data sensitive edge-native applications. Edge application developers require infrastructure at the edge to host the application workload and network connectivity procedures to connect the application users to the nearest edge where the application workload is hosted. Distributed nature of edge infrastructure and the requirement on network connectivity makes it attractive for communication service providers (CSPs) to become Edge Service providers (ESP); similarly, hyper-scale cloud providers (HCPs) are also planning to expand as ESP building on their cloud presence targeting edge application developers. CSPs across the globe follow a standard approach for building interoperable networks and infrastructure, while HCPs do not participate in telecom standardization bodies. Standards development organizations (SDOs) such as the European Telecommunication Standardization Institute (ETSI) and 3rd Generation Partnership Project (3GPP) are working to provide a standard architecture for edge computing solution for service providers. However, the current focus of SDOs is more on architecture and not much focus on application developer experience and the Application Programming Interfaces (APIs). On the architecture itself, there are different standards and approaches available which overlap with each other. APIs proposed by different SDOs are not easily consumable by edge application developers and require simplification. On the other hand, there are not many widely known standards in the hyper-scale cloud and public cloud industry to integrate with each other except the public application programming interfaces (APIs) offered by cloud providers. To scale and succeed, edge service providers need to focus on interoperability, not only from a cloud infrastructure perspective but from a network connectivity perspective as well. This work analyzes standards defined by different standardization bodies in the 5G edge computing area and the overlaps between the standards. The work then highlights the requirements from an edge application developer perspective, investigates the deficiencies of the standards, and proposes an end-to-end edge solution architecture and a method to simplify the APIs which fulfil the need for edge-native applications. The proposed solution considers CSPs providing multi-cloud infrastructure for edge computing by integrating with HCPs infrastructure. In addition, the work investigates existing standards to integrate cloud capabilities in network platforms and elaborates the way network and cloud computing capabilities can be integrated to provide complete edge service to edge application developers or enterprises. It proposes an alternative way to integrate edge application developers with cloud service providers dynamically by offering a catalog of services. / Edge Computing anses vara en av nyckelfunktionerna i nästa generations (5G) nätverk, vilket möjliggör minskad fördröjning, ökad genomströmning och datakänsliga och kantnära applikationer. Applikationsutvecklare för Edge Computing är beroende av kantinfrastruktur som är värd för applikationen, och nätverksanslutning för att ansluta applikationsanvändarna till närmaste kant där applikationens är placerad. Även om kantapplikationer kan vara värd för vilken infrastruktur som helst, planerar leverantörer av kommunikationstjänster (CSP:er) att erbjuda distribuerad kantinfrastruktur och anslutningar. På liknande sätt planerar även molnleverantörer med hög skalbarhet (HCP) att erbjudakantinfrastruktur. CSP:er följer standardmetoden för att bygga nätverk och infrastruktur medan HCP:er inte deltar i standardiseringsorgan. Standardutvecklingsorganisationer (SDO) som europeisk telekommunikations standardiseringsinstitut (ETSI) och 3rd Generation Partnership Project (3GPP) arbetar för att tillhandahålla en standardarkitektur för Edge Computing för tjänsteleverantörer. Men nuvarande fokus är mer på arkitektur och inte mycket fokus är riktat mot applikationsutvecklares erfarenhet och API:er. I själva arkitekturen finns det olika standarder och tillvägagångssätt som överlappar varandra. API:er föreslagna av olika SDO:er är inte lättillgängliga för utvecklar av kantapplikationer och måste förenklas. Å andra sidan finns det inte många allmänt kända standarder i hyperskalära moln och offentlig molnindustri som går att integrera med varandra förutom de offentliga gränssnitten för applikationsprogrammering (API:er) som erbjuds av molnleverantörer. För att kunna betjäna omfattande applikationsutvecklare måste CSP:er erbjuda multimolnfunktioner och därmed komplettera sin egen infrastruktur med kapaciteten för HCP:er. På liknande sätt kommer HCP:er att behöva integrera anslutningstjänster utöver infrastruktur för att erbjuda kantfunktioner. Den här arbetet beskriver olika standarder definierade av olika standardiseringsorgan i Edge Computing-området för 5G, och analyzerar överlappningar mellan standarderna. Arbetet belyser sedan kraven från ett utvecklingsperspektiv av kantapplikationer, undersöker bristerna i standarderna och föreslår en lösningsarkitektur som uppfyller behovet för kantbyggda applikationer. Den föreslagna lösningen beaktar CSP:er som tillhandahåller flermolnsinfrastruktur för Edge Computing genom att integreras med HCP:s infrastruktur. Arbetet undersöker vidare befintliga standarder för att integrera molnfunktioner i nätverksplattformar och utvecklar på vilket sätt nätverks- och molntjänster kan integreras för att erbjuda kompletta tjänster till utvecklare av kantapplikationer. Arbetet föreslår ett alternativt sätt att dynamiskt integrera utvecklare av kantapplikationer med leverantörer av molntjänster genom att erbjuda en katalog av tjänster.