Back to search

University of Thessaly (UTH)

Predictive and energy-aware management of heterogeneous disaggregated 6G RAN

Abstract

dc:description

Fifth-generation (5G) mobile networks introduce cloud-native and disaggregated architectures to support the three distinct service categories of enhanced mobile broadband (eMBB), ultra-reliable low-latency communication (URLLC), and massive machine-type communication (mMTC). Looking beyond 5G, 6G evolves these service categories into Immersive Communications (IMC), Hyper-Reliable Low-Latency Communications (HRLLC), and Massive Connectivity (MC), respectively, while introducing three new service categories: Ubiquitous Connectivity (UC) for seamless coverage across heterogeneous and non-terrestrial environments, Integrated AI and Communication (IAC) embedding AI as a native component of the network control plane, and Integrated Sensing and Communication (ISAC) unifying sensing and communication within shared spectrum and hardware. While, 6G defines four design directions: sustainability, connecting the unconnected, security and resilience, and ubiquitous intelligence.This thesis addresses core challenges in the dynamic control and management of disaggregated 5G and beyond-5G networks, targeting the UC, IMC, HRLLC, and IAC service categories while advancing the sustainability design direction of 6G. Specifically, this thesis addresses the following five research questions: (RQ1) How can non-3GPP access technologies be integrated directly into the disaggregated RAN stack to enable unified capacity extension and service differentiation at the access layer, advancing the UC service category of 6G? (RQ2) Given a heterogeneous multi-RAT RAN resulting from the integration of non-3GPP access technologies, how can AI-driven control mechanisms within the O-RAN architecture enable proactive and intelligent traffic steering across the available RATs, advancing both UC and IAC? (RQ3) As intelligent traffic steering increases the aggregated traffic processed by the RAN, how can cloud-native User Plane functions be autonomously scaled within an O-RAN deployment to sustain service performance under dynamic traffic conditions, targeting IMC and HRLLC? (RQ4) Once dynamic RAN User Plane scaling is established, how can different scaling strategies be designed to jointly optimize performance, latency, and energy efficiency in disaggregated O-RAN deployments, advancing HRLLC and the sustainability design direction of 6G? (RQ5) Given that energy-aware scaling relies on knowing the power consumption of RAN components, to what extent can network KPIs predict this consumption for the DU and RU, and how can such predictions support autonomous anomaly detection in online O-RAN deployments, advancing the sustainability design direction and IAC?Addressing RQ1, a novel architecture is proposed that integrates a non-3GPP WiFi Distributed Unit into a disaggregated 5G C-RAN stack, enabling RAN-level convergence of heterogeneous RATs and advancing the UC service category. A user-based slicing scheme is introduced using SDN and application-aware prioritization within the WiFi MAC layer, enabling per-user QoS differentiation. Experimental validation on a real-world testbed demonstrates improved system capacity and service differentiation without requiring additional licensed spectrum.Addressing RQ2, the WiFi DU integration is elevated into a fully O-RAN-compliant framework, advancing both UC and IAC. An xApp-based traffic steering mechanism is developed that utilizes LSTM-based demand forecasting within the Near-RT RIC to proactively distribute traffic across heterogeneous, multi-RAT DUs. Evaluation on a real-world testbed demonstrates that the mechanism dynamically adapts to traffic fluctuations, ensuring fair resource distribution and maintaining throughput without degradation under high contention.Addressing RQ3, a proactive horizontal scaling mechanism for the CU-UP component is proposed, targeting the IMC and HRLLC service categories. An xApp on the Near-RT RIC integrates LSTM-based traffic forecasting to anticipate demand and initiate scaling before congestion occurs. Implementation and evaluation on O-RAN-compliant testbeds using real traffic traces demonstrate enhanced service continuity, reduced packet loss, and more efficient resource utilization compared to static provisioning.Addressing RQ4, two complementary AI-driven xApps are designed for proactive CU-UP scaling, jointly advancing the sustainability design direction and HRLLC. The first implements vertical scaling by dynamically adjusting CPU resources to minimize power consumption, achieving up to 23% energy savings. The second performs latency-aware horizontal scaling by optimally placing CU-UP instances across the infrastructure, achieving up to 16% energy savings with significantly reduced end-to-end latency.Addressing RQ5, an AI-driven framework is proposed to advance the sustainability design direction and the IAC by predicting the power consumption of O-DU and O-RU components using only 3GPP-compliant network KPIs. Hybrid XGBoost and CatBoost models achieve a MAPE of 1.02% for the O-RU and 1.22% for the O-DU. Residual-based anomaly detection achieves a 100% detection rate for injected measurement anomalies, followed by automatic calibration.All proposed mechanisms are designed and evaluated on real-world 5G testbeds under real-time operating conditions, confirming their practical applicability. These five contributions collectively form a path from the physical extension of the RAN through heterogeneous access integration to

Degree

thesis:*
Grantor dc:publisher
University of Thessaly (UTH)
Year dc:date
2026

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kafalas, Dimitrios
  • Κεφαλάς, Δημήτριος

Subjects

dc:subject × 20

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/63116

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Kafalas, Dimitrios; Κεφαλάς, Δημήτριος. Predictive and energy-aware management of heterogeneous disaggregated 6G RAN. University of Thessaly (UTH), 2026. http://hdl.handle.net/10442/hedi/63116