Every time we stream a video, send a message, or make a call, our phones exchange signals with a vast network of radio towers and base stations. This network relies on radio frequency power amplifiers, power-hungry devices that boost the signals before they are transmitted. But there is a catch: When these amplifiers operate efficiently, they also distort the signal. Fixing the distortion presents a tough engineering challenge, requiring highly accurate modelling of the distortions with minimal processing complexity.
Arne Fischer-Bühner’s research focuses on improving this balancing act. He studied how neural networks, machine learning models that learn patterns from data, can be most effectively used to correct the signal distortions. The underlying technique, known as digital predistortion (DPD), is already used in today’s mobile networks, but Fischer-Bühner’s work shows how it can be advanced to be more powerful and more efficient.
“An emerging approach with the potential to overcome the limitations of today’s DPD systems is artificial neural networks due to their excellent ability to model complex nonlinear problems,” he says.
Neural networks meet domain-specific requirements
Yet, to unlock the potential of neural networks, careful adoption to the domain-specific requirements of DPD is necessary. In his dissertation, Fischer-Bühner describes several specialised neural network techniques that address key challenges such as improving the modelling of power amplifier memory effects, compensating for long-term transient behaviour, and reducing computational complexity of the model.
“Our experimental results demonstrate that with careful design, these models can correct the amplifier’s distortions with a very high accuracy, more effectively than traditional methods, while also being lightweight to process,” says Fischer-Bühner.
His results could help mobile operators and equipment manufacturers build more energy-efficient, greener, and more scalable mobile communication systems. Especially as we move to beyond 5G technologies with dense deployments of multi-antenna radios that include a large number of power amplifiers.
Fischer-Bühner’s work was carried out in the context of a research collaboration between Nokia Bell Labs, Belgium, and Tampere University, as part of a European Union funded Marie Sklodowska-Curie Innovative Training Network funded (Horizon 2020 Grant No. 860921). Fischer-Bühner currently continues to work as a Researcher at Nokia Bell Labs.
Public defence on Friday 7 November
The doctoral dissertation of Dipl.-Ing. Arne Fischer-Bühner in the field of communications engineering entitled Advanced Neural Network Modeling Techniques for RF Power Amplifier Linearization will be publicly examined at the Faculty of Information Technology and Communication Sciences at Tampere University at 12:00 p.m. on Friday 7 November 2025 at Hervanta campus, Tietotalo auditorium TB109.
The Opponent will be Prof. Pere Gilabert from Universitat Politècnica de Catalunya, Barcelona, Spain. The Custos will be Prof. Mikko Valkama from Tampere University, Finland. The work has been co-supervised by D.Sc. (Tech) Lauri Anttila from Tampere University and PhD Manil Dev Gomony from Nokia Bell Labs, Antwerp, Belgium.
The doctoral dissertation is available online.
The public defence can be followed via remote connection.