Signal-Aware Predistortion Configuration Using Unsupervised Learning for 5G–Advanced/6G Radios
(2026)Department of Automatic Control
- Abstract
- Digital Predistortion (DPD) is an essential component of modern mobile communication systems ensuring that distortion, created by the nonlinearities of the Power Amplifier (PA) is reduced while allowing them to be operated at high efficiency levels. This thesis has investigated the use of an unsupervised machine learning method known as Gaussian Mixture Model (GMM) as a way to define and populate a suitable number of DPD configurations to satisfy the required use cases for specific radio products. The goal was to minimize the required resources by the DPD as well as the actual DPD configuration time spent in the hardware integration process. The proposed method was compared against a baseline implementation and evaulated on the mentioned... (More)
- Digital Predistortion (DPD) is an essential component of modern mobile communication systems ensuring that distortion, created by the nonlinearities of the Power Amplifier (PA) is reduced while allowing them to be operated at high efficiency levels. This thesis has investigated the use of an unsupervised machine learning method known as Gaussian Mixture Model (GMM) as a way to define and populate a suitable number of DPD configurations to satisfy the required use cases for specific radio products. The goal was to minimize the required resources by the DPD as well as the actual DPD configuration time spent in the hardware integration process. The proposed method was compared against a baseline implementation and evaulated on the mentioned metrics. The result of the thesis was a reduction in resources compared to the baseline while maintaining system requirements quantified by Adjacent Channel Leakage Ratio (ACLR). However, the proposed method had a significant increase in hardware integration time compared to the baseline. Overall, the most promising method was a combination between machine learning and traditional optimization which yielded the lowest DPD resources while still maintaining lower integration time than the strictly machine learning based approach. We conclude that a clear trade-off exists between DPD resource utilization and hardware integration time, where the optimal configuration is dependent on system-level priorities. These findings highlight the potential of hybrid approaches, combining machine learning with conventional optimization, as an effective strategy for balancing performance and implementation complexity in practical systems. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9246995
- author
- Magnusson, Oskar and Bergenfeldt, Arvid
- supervisor
- organization
- year
- 2026
- type
- H3 - Professional qualifications (4 Years - )
- subject
- keywords
- Digital Predistortion, Unsupervised Learning, 6G, Digital Radio Tuning, 5G–Advanced, Machine Learning, State-of-the-art, Optimization, Clustering, GMM, Artificial Intelligence
- report number
- TFRT-6316
- other publication id
- 0280-5316
- language
- English
- id
- 9246995
- date added to LUP
- 2026-08-25 10:31:17
- date last changed
- 2026-08-25 10:31:17
@misc{9246995,
abstract = {{Digital Predistortion (DPD) is an essential component of modern mobile communication systems ensuring that distortion, created by the nonlinearities of the Power Amplifier (PA) is reduced while allowing them to be operated at high efficiency levels. This thesis has investigated the use of an unsupervised machine learning method known as Gaussian Mixture Model (GMM) as a way to define and populate a suitable number of DPD configurations to satisfy the required use cases for specific radio products. The goal was to minimize the required resources by the DPD as well as the actual DPD configuration time spent in the hardware integration process. The proposed method was compared against a baseline implementation and evaulated on the mentioned metrics. The result of the thesis was a reduction in resources compared to the baseline while maintaining system requirements quantified by Adjacent Channel Leakage Ratio (ACLR). However, the proposed method had a significant increase in hardware integration time compared to the baseline. Overall, the most promising method was a combination between machine learning and traditional optimization which yielded the lowest DPD resources while still maintaining lower integration time than the strictly machine learning based approach. We conclude that a clear trade-off exists between DPD resource utilization and hardware integration time, where the optimal configuration is dependent on system-level priorities. These findings highlight the potential of hybrid approaches, combining machine learning with conventional optimization, as an effective strategy for balancing performance and implementation complexity in practical systems.}},
author = {{Magnusson, Oskar and Bergenfeldt, Arvid}},
language = {{eng}},
note = {{Student Paper}},
title = {{Signal-Aware Predistortion Configuration Using Unsupervised Learning for 5G–Advanced/6G Radios}},
year = {{2026}},
}