From wireless communication networks to medical imaging and radar systems, many modern technologies depend on the ability to detect just a few meaningful signals buried in massive amounts of data. This task — known as sparse signal recovery — is essential for making sense of complex measurements, but existing algorithms can be painfully slow, especially when the signals come in naturally grouped patterns, known as block sparsity.

In our recent work [1], we developed a new algorithm called Fast Variational Block-Sparse Bayesian Learning (F-BSBL) that makes this process dramatically faster and more efficient. By rethinking how uncertainty and structure are handled in sparse models, we achieved up to a hundredfold speedup over existing approaches — without sacrificing accuracy. Our method also unifies several previously separate learning strategies into a single, elegant mathematical framework. Tested on challenging problems such as direction-of-arrival estimation, F-BSBL delivers both speed and precision, paving the way for real-time signal analysis in next-generation communication, sensing, and data-driven technologies.

1. J. Möderl, E. Leitinger, B. H. Fleury, F. Pernkopf and K. Witrisal, "Fast Variational Block-Sparse Bayesian Learning," in IEEE Transactions on Signal Processing, doi: 10.1109/TSP.2025.3611234

Contact:
Jakob Möderl

In many real-world scenarios - like when a person carries a radio device - direct line-of-sight (LoS) signals from surrounding anchors (e.g., base stations or sensors) can get blocked by the person's own body. This makes accurate positioning difficult, especially in challenging indoor or cluttered environments.

Our recent work tackles this problem by treating the person not just as an obstacle, but as a key part of the solution. We model the person as an extended object (EO) that can scatter and attenuate signals. Instead of relying only on LoS measurements, we combine:

• Active measurements between the device and fixed anchors  
• Passive measurements between pairs of anchors that reflect off the person  

We developed a Bayesian estimation framework and an extended object data association algorithm that can make sense of these complex, indirect signals. 

The result shows that the proposed joint estimation algorithm (AP-EOPDA) is much more reliable and accurate in positioning - even when the LoS is completely blocked [1].

This approach outperforms traditional methods that assume the point-target or rely solely on LoS signals. It opens the door to more robust and energy-efficient localization, especially in environments where unavailable LoS is common.

Feel free to reach out if you're working on related problems or want to know more about the technical details!

[1] Hong Zhu and Alexander Venus and Erik Leitinger and Klaus Witrisal, "Multi-Sensor Fusion of Active and Passive Measurements for Extended Object Tracking", in 33rd European Signal Processing Conference (EUSIPCO 2025), DOI:10.48550/arXiv.2504.18301.

Contact:
hong.zhunoSpam@tugraz.at
https://www.linkedin.com/in/hong-zhu-74173427a/