Signals and Systems Research Group

The Department of Numerical Analysis at the Faculty of Informatics of the Eötvös Loránd University has a decades-long professional history in the fields of approximation theory, mathematical modeling, and numerical analysis. The Harmonic Analysis School, established at the department over 30 years ago, has provided scientific motivation for numerous university and academic doctoral dissertations, contributing to the continuous expansion of the department's research portfolio. Following this tradition, the Signals and Systems Research Group was established, and in recent years, several young colleagues have joined. Within the research group, alongside classical mathematical and computer science studies, we focus on the development of model-driven machine learning algorithms in response to the growing influence of artificial intelligence (AI) and emerging trends in computer science. Both MSc and doctoral students actively participate in these research projects.

Research group homepage: institute link

▶ Biomedical signal processing

• ECG segmentation, heartbeat classification, data compression.
• EEG signal processing, epileptic seizure detection, sleep stage classication.
• Multimodal biosignal analysis: ECG and PPG based blood pressure estimation, statistical analysis of physiological datasets.

▶ Digital signal processing

• Autonomous vehicle control: tire sensor data processing, road surface classification, estimation of tire-road forces, skidding detection.
• Signal processing for telecommunications: OFDM communication, transceiver and receiver data simulation, channel estimation, modell-based decoding, physical layer encryption.
• System identification: investigating dynamical systems, transfer function identification, rational function approximation, embedded hardware implementations using microcontrollers, FPGAs.

▶ Tomographic imaging

• Thermal image reconstruction for nondestructive testing.
• CT image reconstruction, segmentation, artifact reduction.

▶ Model-driven machine learning

• Fusing mathematical models with data-driven artificial intelligence: Variable Projection Networks (VPNet), ODE Network, WaveletKernelNet, OptNet, Wiener-Hammerstein networks.
• Deep unfolding: unrolling iterative numerical methods into neural network layers.

• Stationary and non-stationary signal analysis, classical signal and image processing techniques, dynamical modeling and state space representation of physical systems.
• Transparent AI model development, mathematical formulation and integration of domain-specific knowledge via parametrized (trainable) orthogonal transformations.
• Numerical methods including optimization and regularization techniques, computationally efficient CPU and GPU implementations, reproducible research in Python, code porting in embedded systems.

• Dr. habil. Péter Kovács (research group leader, bibliography: MTMT)
• Prof. Dr. Sándor Fridli, full professor
• Dr. Gergő Bognár, assistant professor
• PhD students: Mátyás Szabari, Gergő Ungvári, Hoque Sajedul

• TKP2021-NVA-29
• OTKA Project Nr.: 146721 
• EFOP-3.6.3-VEKOP-16-2017-00001 
• TÁMOP-4.2.2.A-11/1/KONV-2012-005 
• OTKA Project Nr.: 115804 
• 2 industrial projects with Mediso Medical Imaging Systems Ltd, Silicon Austria Labs GmbH

International collaborators:

• Institute of Signal Processing, Johannes Kepler University (Linz, Austria)
• Silicon Austria Labs (Linz, Austria)
• Department of Medical Informatics, University Medical Center, Georg-August University (Göttingen, Germany)
• Center for Artificial Intelligence and Cybersecurity, Faculty of Engineering, University of Rijeka
• Institute of Neurophysiology, RWTH Aachen University Hospital (Aachen, Germany)
• Institute for Biomedical Informatics, University Hospital Cologne (Köln, Germany)
• Department of Anesthesiology, University Hospital Würzburg (Würzburg, Germany)
• Department of Computing Sciences, Tampere University, (Tampere, Finland)
• Department of Health Technology, Technical University of Denmark (Copenhagen, Denmark)

National collaborators:

• HUN-REN SZTAKI, Systems and Control Lab
• HUN-REN Institute of Technical Physics and Materials Science, Nanosensors Laboratory
• HUN-REN Wigner Research Centre for Physics, Data and Compute Intensive Sciences Research Group
• ELTE IK, Savaria Institute of Technology

Industrial collaborators 

• RECENDT Research Center for Non-Destructive Testing GMBH (Austria)
• Siemens Digital Industries Software, Test Division (Belgium)
• BKK Modeling and Data Analysis Department (Hungary)

• P. Kovács, S. Fridli, F. Schipp: Generalized Rational Variable Projection With Application in ECG Compression, IEEE Transactions on Signal Processing (2020) [DOI]
• P. Kovács, G. Bognár, C. Huber, M. Huemer: VPNET: Variable Projection Networks, International Journal of Neural Systems (2022) [DOI]
• T. Dózsa, F. Deuschle, B. Cornelis, P. Kovács: Variable projection support vector machines and some applications using adaptive Hermite expansions, International Journal of Neural Systems (2023) [DOI]
• T. Dózsa, J. Radó, J. Volk, A. Kisari, A. Soumelidis, P. Kovács: Road abnormality detection using piezoresistive force sensors and adaptive signal models, IEEE Transactions on Instrumentation and Measurement (2022) [DOI]
• S. Baumgartner, G. Bognár, O. Lang, M. Huemer: Neural network npproaches for data estimation in unique word OFDM systems, IEEE Transactions on Vehicular Technology (2023) [DOI]
• D. Selimovic, J. Lerga, P. Kovács, J. Prpic-Oršic: Improved parametrized multiple window spectrogram with application in ship navigation systems, Elsevier Digital Signal Processing (2022) [DOI]
• T. Dózsa, M. Szabari, A. Soumelidis, P. Kovács: Pole identification using discrete Laguerre expansion and variable projection, 22nd International Federation of Automatic Control (IFAC) World Congress (2022) [DOI]

kovika@inf.elte.hu