Back to search

National Technical University of Athens (NTUA)

Compute and radio resource orchestration and data fusion in next-generation multiple access (NGMA) 6G-enabled computing paradigms

Abstract

dc:description

In the evolving landscape of digital transformation, next-generation wireless networks are anticipated to accommodate an unprecedented surge in data generation and computational demands, driven by the proliferation of intelligent services and devices. Within this context, the forthcoming sixth-generation (6G) wireless networks envision a deeply integrated communication-computation fabric, wherein a continuum of heterogeneous resources, distributed across multiple layers, collaboratively sustains the requirements of data-driven and latency-sensitive applications. While this multi-tiered computing infrastructure offers enhanced adaptability and scalability, it also introduces complex interdependencies and inherent competition among network entities over limited radio and compute resources. These challenges, compounded by the uncertainties and heterogeneity that characterize such systems, render conventional centralized solutions insufficient. Consequently, there is a critical need for autonomous and context-aware frameworks that can support distributed decision-making and efficient resource orchestration across the disaggregated computing continuum.In this thesis, we address the problem of compute and radio resource orchestration in heterogeneous, decentralized, and multi-layer computing environments envisioned in 6G networks. Leveraging the Next-Generation Multiple Access (NGMA) paradigm, which enables flexible coordination of radio resources across users, devices, and service layers, the proposed frameworks are designed to model the strategic interactions of network entities operating under resource constraints and competitive dynamics. Adopting a distributed and self-organizing perspective, we develop novel models that explicitly account for agents’ interdependent decisions, diverse objectives, and the inherent limitations in information exchange. To this end, learning, game, and prospect-theoretic tools are employed, enabling the design of low-complexity, scalable solutions that reflect rational and strategic behaviors, providing a realistic approach to resource management in complex 6G computing paradigms. First, to address the challenges inherent in decentralized resource management, the application of distributed Artificial Intelligence (AI) is investigated within the context of Hierarchical Federated Learning (HFL) networks. As the demand for collaborative model training across end-user devices continues to grow, the problem of user association and uplink transmission power allocation in wireless HFL networks is addressed by introducing a satisfaction-based game-theoretic model. This approach effectively balances the tradeoff between global model accuracy, energy consumption, and transmission time. The decentralized nature of this solution ensures scalability, as users are empowered to independently select their optimal association and power transmission strategies, all while guaranteeing that their minimum Quality of Service (QoS) requirements are satisfied. Furthermore, the proposed framework ensures cost-efficiency by minimizing the overall energy expenses associated with transmission, without compromising the system's performance. The integration of Non-Orthogonal Multiple Access (NOMA) enhances communication efficiency, optimizing both energy utilization and network throughput within the system.Subsequently, to address the need for efficient aggregation and coordination in such distributed learning systems, Over-the-Air Computation (AirComp) is introduced as an alternative communication and data fusion technique, particularly suited for large-scale Internet-of-Things (IoT) and learning environments. By leveraging the superposition property of the wireless channel, AirComp enables simultaneous aggregation and reduces communication latency compared to conventional multiple access schemes. The integration of Reconfigurable Intelligent Surfaces (RIS) enhances aggregation accuracy and energy efficiency by optimizing the wireless propagation environment. Within this framework, RIS phase shifts, receive beamforming at the central server, and local computing frequency allocation at client devices are jointly optimized, addressing the tightly coupled communication and computation challenges inherent in distributed learning. The resulting non-convex optimization problem is decomposed into tractable subproblems and solved using Semidefinite Programming, Difference-of-Convex programming, and alternating optimization strategies, ensuring computational efficiency and scalability. %The proposed approach significantly reduces overall latency and energy consumption, demonstrating the practical potential of AirComp-enabled federated learning.Building on the hierarchical edge-cloud architecture for wireless FL, the focus shifts to the broader challenge of task offloading and scheduling in such systems. With increasing demand for real-time, adaptive computing across diverse applications, relying on local processing is no longer feasible. The integration of approximate and delayed computing paradigms within the edge-cloud system presents an effective solution, where tasks are processed at the edge faster with reduced accuracy or forwarded to the cloud for precise but delayed execution. This hybrid approach mitigates latency and improves network efficiency, but it also requires efficient resource allocation to meet QoS demands. In this context, the SatisfOptimizing Game is introduced, a satisfaction-based game-theoretic model that enables users to autonomously choose their offloading strategies while balancing latency and accuracy requirements. The edge system, in turn, employs a Genetic Algorithm for task scheduling to optimize resource utilization.Beyond the optimization of communicational and computational resources, the thesis extends to the economic dimension of the computing continuum, where multiple cloud, edge, and fog providers coexist and compete to attract users. Recognizing that resource allocation decisions are increasingly driven by pricing incentives, a holistic market-based framework for symbiotic resource pricing is introduced. Unlike existing approaches that focus on a single layer of the continuum, this framework captures the interdependence among heterogeneous service providers, enabling both competitive and cooperative market behaviors. Through non-cooperative and coalition-based game formulations, the proposed model analyzes pricing equilibria and equitable profit distribution across providers, revealing how cooperation can enhance market stability and user satisfaction in multi-provider computing ecosystems. This analysis offers a unifying economic perspective, complementing the technical frameworks and highlighting the interplay between market incentives and resource efficiency.Finally, moving from terrestrial multi-layer architectures to the broader computing continuum, the work is adapted to include airborne and satellite-assisted layers, introducing new opportunities and risks in resource-constrained environments. In Space-Air-Ground Integrated Networks (SAGINs), offloading decisions must account not only for latency and energy tradeoffs but also for uncertainty from task rejection and failure, particularly at Unmanned Aerial Vehicle (UAV)-mounted servers with limited computational capacity. To navigate these challenges, a fully distributed and risk-conscious offloading framework is introduced, wherein IoT devices autonomously decide whether to offload tasks to nearby UAV-mounted MEC servers or to distant cloud servers via satellite relays. Unlike conventional models assuming rational, risk-neutral behavior, the proposed approach incorporates Prospect Theory to reflect users' loss-averse perceptions, under resource contention. Treating the UAV as a shared, failure-prone resource, the framework captures the dynamics of a Common-Pool of Resources, enabling devices to balance total overhead with risk.All methodologies are rigorously tested through simulations under diverse network conditions, validating their practical relevance and performance. Metrics such as energy consumption, latency, accuracy, and convergence are comprehensively assessed, revealing consistent advantages over baseline and state-of-the-art methods. The evaluations demonstrate the frameworks' capacity for scalable and adaptive resource management, with notable improvements in resource utilization and QoS in various 6G-enabled computing environments.

Degree

thesis:*
Grantor dc:publisher
National Technical University of Athens (NTUA)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Charatsaris, Panagiotis
  • Χαρατσάρης, Παναγιώτης

Subjects

dc:subject × 32

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/62196

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Charatsaris, Panagiotis; Χαρατσάρης, Παναγιώτης. Compute and radio resource orchestration and data fusion in next-generation multiple access (NGMA) 6G-enabled computing paradigms. National Technical University of Athens (NTUA), 2026. http://hdl.handle.net/10442/hedi/62196