AI-driven digital twin for traffic prediction and calibration in 5G NR networks

Thumbnail Image

Files

master_Pham_Duc_2026.pdf (1.51 MB)

URL

Journal Title

Journal ISSN

Volume Title

School of Electrical Engineering | Master's thesis
Unless otherwise stated, all rights belong to the author. You may download, display and print this publication for Your own personal use. Commercial use is prohibited.

Authors

Date

2025-12-30

Department

Major/Subject

Cloud and Network Infrastructures

Mcode

Degree programme

Master's Programme in ICT Innovation
Master's Programme in ICT Innovation
Master's Programme in ICT Innovation

Language

en

Pages

53

Series

Abstract

Digital Twin application has been gaining more and more popular across various domains. In mobile networking, digital twins offer the ability to replicate the behaviour of a live system and evaluate different possible outcomes corresponding to different configurations. However, maintaining an accurate twin is challenging due to the dynamic nature of radio conditions, hardware variability, and the tight performance requirements of modern applications. This motivates the need for mechanisms that continuously align the digital twin with the physical network and support proactive performance optimisation. This thesis examines how periodically collected probe measurements and machine-learning predictions can be combined with a LENA based digital twin to support configuration evaluation in an indoor 5G NR network. The collected data reveals significant short-term variability in throughput and latency, and the measurement setup introduces constraints that affect the modelling fidelity of the twin. In particular, the use of TCP for throughput measurements prevents realistic congestion modelling in LENA. These factors limit the extent to which the digital twin can replicate all observed network behaviours. To operate within these constraints, the thesis develops a workflow in which real measurements are used to calibrate the twin, and a machine-learning model predicts short-term throughput degradations. When a degradation is predicted, the corresponding throughput level is reproduced in the twin through base station transmitting power adjustments, enabling the evaluation of candidate Time Division Duplexing (TDD) configurations under controlled conditions. The evaluation is performed using the calibrated system-level simulator for several TDD reallocations. The results show that the synchronized digital twin can reproduce degraded throughput conditions within a 10% deviation threshold and correctly reflect the uplink-downlink trade offs associated with 3:7 (UL:DL) and 1:4 (UL:DL) configurations. Overall, the findings demonstrate that the proposed workflow can detect throughput degradations in advance, reproduce the predicted operating state in the digital twin, and evaluate alternative TDD configurations before they are applied to the physical network. The study also confirms that LENA provides a sufficiently accurate basis for constructing a digital twin under normal operating conditions. This is supported by the close correspondence between measured and simulated signal-to-interference-plus-noise ratio (SINR), total received power, and throughput after calibration. However, this alignment does not extend to degraded states, where SINR cannot be reliably synchronized. Finally, the thesis identifies several directions for further improvement, such as UDP-based data collection for accurate backhaul congestion modelling, more expressive prediction models, and reinforcement-learning integration for closed-loop autonomous optimisation.

Description

Supervisor

Mähönen, Petri

Thesis advisor

Costa Requena, Jose

Keywords

5G, digital twins, AI, traffic optimisation, LENA NR, NS3

Other note

Citation

Permanent link to this item

https://urn.fi/URN:NBN:fi:aalto-202601201638

Collections

[dipl] Sähkötekniikan korkeakoulu / ELEC