---

---

01.09.2026 18:28

Tensor-Based Signal Processing: TU Ilmenau Researcher Honored for World’s Best Ph.D. Thesis

Barbara Aichroth Marketing und Kommunikation
Technische Universität Ilmenau

Dr. Liana Khamidullina, a postdoctoral researcher at Technische Universität Ilmenau, has been awarded the Best Ph.D. Award by the European Association for Signal Processing (EURASIP). In her award-winning doctoral thesis, she developed new methods for processing and analyzing multidimensional signal data for interdisciplinary applications, including wireless communications, radar technology, and biomedicine. She received the award today (September 1, 2026), during the opening and awards ceremony of the international EUSIPCO 2026 conference in Bruges, Belgium.

Completed in 2023, her thesis, entitled “Tensor Decompositions and Algorithms for Efficient Multidimensional Signal Processing,” was conducted at TU Ilmenau’s Communications Research Laboratory under the supervision of Prof. Martin Haardt. The thesis focuses on tensor-based signal processing methods. These methods consider the specific structure of complex data that simultaneously contains information across multiple dimensions – for example, space, time, and frequency.

Understanding complex data better

Many modern measurement and communication systems generate data not just in one or two dimensions. In a medical measurement, for example, information from many sensors may be collected simultaneously at different points in time and across different frequencies. Such data can be described as multidimensional structures.

This is where Liana Khamidullina’s dissertation comes in. In her doctoral thesis at TU Ilmenau, she developed and analyzed methods for processing such multidi-mensional data using so-called tensors.

This representation offers a crucial advantage: the characteristic structure of multidimensional data is preserved during processing. As a result, information can be estimated more accurately, different signal sources can be separated more effectively, and complex data models can be identified more unambiguously.

These advantages can be leveraged in a wide range of interdisciplinary applications, including facial recognition, image compression and noise reduction for hyperspectral images, data mining, social network analysis, machine learning, pattern recognition, array signal processing, wireless communications, and biomedical signal processing.

Liana Khamidullina demonstrates the versatility of these methods in her dissertation through three applications.

Novel Tensor Decompositions Extend Established Mathematical Methods

“Her derivation of the Multilinear Generalized Singular Value Decomposition (ML-
GSVD) as a new tensor decomposition is particularly impressive,” says Prof. Martin Haardt. “It extends the well-known Generalized Singular Value Decomposition (GSVD) – an important mathematical tool in signal processing – from two matrices to more than two matrices. Put simply, these methods help break down complex data into characteristic components and reveal the structures contained within it. Unlike other extensions of the GSVD already described in the literature, the ML-GSVD preserves the essential properties of the GSVD, such as the orthogonality of the factor matrices in the first mode.”

The researcher demonstrates its practical significance, among other things, through three different applications of the ML-GSVD in wireless MIMO communication systems. MIMO stands for “Multiple Input, Multiple Output” and refers to communication systems in which multiple transmitting and receiving antennas are used simultaneously. Rather than considering just a single signal path, this makes it possible to process multiple signals at the same time. Tensor-based analysis enables the structure contained in the different signal paths to be specifically exploited.

Precise Localization with Multiple Radar Antennas

For radar data processing, Liana Khamidullina also developed new robust methods for what is known as Block-Term Decomposition (BTD) and extended these to a coupled BTD, which allows multiple interconnected datasets to be analyzed jointly.

“Particularly noteworthy is the COBRAS algorithm developed by Dr. Khamidullina, short for coupled block-term decomposition for multi-static radar systems,” says Prof. Haardt. It enables high-resolution three-dimensional localization of targets in multistatic MIMO radar systems. Several receiving arrays detect the same transmitted signals at different locations. COBRAS combines this jointly acquired information and takes into account an accurate wavefront model for the near field. This makes it possible to determine not only direction but also distance with high precision. The approach can be applied to arbitrary array geometries.

Joint Analysis of EEG and MEG Data

Finally, in collaboration with Jena University Hospital, Liana Khamidullina investigated EEG and MEG data. While EEG measures the brain’s electrical activity using electrodes placed on the scalp, MEG detects the magnetic fields generated by this activity. The two methods provide different but complementary information.

For the study, electrical stimuli were applied to the median nerve of eight healthy volunteers. The median nerve runs from the upper arm to the hand and is involved, among other things, in flexion and grasping movements. The resulting brain responses were recorded using EEG and MEG.

Using the methods she developed, Khamidullina was able to jointly analyze data from four measurement systems and extract temporal, spectral, and spatial signatures of brain activity. Her work thus provides a foundation for gaining new insights into how the brain functions.

Prof. Martin Haardt concludes: “The results of the dissertation make significant contributions to multi-antenna signal processing and to the development of efficient multidimensional algorithms for processing multichannel data. Their broad range of interdisciplinary applications also opens up new avenues for research.”

International Selection Process

The Best Ph.D. Award is presented by the European Association for Signal Processing (EURASIP). Each year, experts commissioned by EURASIP select up to three outstanding doctoral theses from different areas of signal processing. This year’s selection considered dissertations from the 2021 to 2023 graduating cohorts.

The evaluation takes into account, among other factors, the scientific significance of the dissertation, publications resulting from the work in international journals and at conferences, the number of citations, and assessments by independent international reviewers. The number of downloads of the dissertations from the EURASIP database is also considered.

Based on these criteria, Dr. Liana Khamidullina was selected as one of this year’s award recipients, alongside Dr. Ricardo Augusto Borsoi of Université Côte d’Azur, France, and the Federal University of Santa Catarina, Brazil, and Dr. Samuel Rey Escudero of Universidad Rey Juan Carlos, Madrid, Spain.

Dr. Liana Khamidullina currently works as a postdoctoral researcher in the Communications Research Laboratory at TU Ilmenau. There, she continues her pioneering research in the field of tensor-based signal processing

---

Wissenschaftliche Ansprechpartner:

Prof. Martin Haardt
Head of Communications Research Laboratory
+49 3677 69-2613
martin.haardt@tu-ilmenau.de

---

Originalpublikation:

Khamidullina, L., 2024. Tensor decompositions and algorithms for efficient multidimensional signal processing. Ilmenau. https://doi.org/10.22032/dbt.59389

---

<
Dr. Liana Khamidullina, a postdoctoral researcher at TU Ilmenau, accepting the EURASIP Best Ph.D. Aw ...
Quelle: jdesign_www.jdes.be
Copyright: EUSIPCO

---

Merkmale dieser Pressemitteilung:
Journalisten, Wissenschaftler
Elektrotechnik, Informationstechnik
überregional
Forschungsergebnisse, Wettbewerbe / Auszeichnungen
Englisch

---