Congratulations Hanie!

• Friday, 21 June 2024

On 18 June 2024, Hanie Moghaddasi successfully defended her thesis "Model-based feature engineering of atrial fibrillation". Congratulations!

Swarming lab opened!

• Friday, 21 June 2024

The TU Delft Swarming Lab is an interfaculty facility between Delft Robotics Institute (DRI), Faculty of EEMCS, Faculty of AE and MAVLab, established to foster research collaboration and community building on swarms. 

Although the lab has been operational since 2023,  it was formally opened on 20 June 2024 by Guido de Croon and Raj Rajan (co-directors). It is housed in the TU Delft Science Centre, which also gives it large visibility to the general public.

The opening presented the facilities as well as some of the amazing students and researchers who have contributed to the lab thus far.

Justin Dauwels appointed (co)director of TU Delft Safety & Security Institute

• Monday, 27 May 2024

Justin Dauwels is appointed as codirector of the TU Delft Safety & Security Institute, together with Eleonora Papadimitriou.

One of the first activities was the signing of a framework agreement with the Dutch police to work on safety & security issues, which led to the establishment of a joint Model-Driven Decisions Lab (MoDDL). Several PhD projects will be funded in the context of this Lab.

Greetings from ICASSP'24 Seoul

• Monday, 22 April 2024

A small but high-level delegation of the SPS group went to ICASSP'24 Seoul this year, along with over 4400 participants! It's been hard to find each other in this crowd, but still we managed to make this "group" picture.

STOI incorporated in Matlab

• Friday, 22 March 2024

STOI is a widely used measure for "short-time objective intelligibility" in (compressed) speech. It was developed by Cees Taal, Richard Hendriks (SPS), Richard Heusdens, and Jesper Jensen in 2010.

It allows to assess the performance of speech compression algorithms, without using listening panels. It is currently popular as a target feature in machine learning algorithms.

STOI already won best paper awards, and has now been incorporated in the 2024 edition of Matlab.

• C.H.Taal, R.C.Hendriks, R.Heusdens, J.Jensen, "An Algorithm for Intelligibility Prediction of Time-Frequency Weighted Noisy Speech," IEEE Transactions on Audio, Speech, and Language Processing, 2011.

Startup Innatera in the news

• Thursday, 7 March 2024

Innatera is a startup that originated in 2019 in the SPS group and has now grown to 65 employees. Recently they completed a neural network-based microcontroller, based on spiking neural networks. This enables highly power efficient processing (e.g. real-time image recognition) for Internet-of-Things applications.

The chip was demonstrated at the Consumer Electronics Show (CES) in Las Vegas.

Alle-Jan van der Veen wins EURASIP Technical Achievement Award

• Friday, 1 March 2024

Alle-Jan has been awarded the prestigious Technical Achievement Award of EURASIP for his “contributions to subspace-based array signal processing”.

The EURASIP Technical Achievement Award honors a person who, over a period of years, has made outstanding technical contributions to theory or practice in technical areas within the scope of the Society, as demonstrated by publications, patents, or recognized impact in this field.

EURASIP is the European Association for Signal Processing. The award will be presented at EUSIPCO'2024 in August in Lyon.

Portrait of Medical Delta Professor Johan Frijns (appointed at EEMCS, Bioelectronics group)

• Friday, 16 February 2024

Portrait of Medical Delta Professor Johan Frijns 

More than 800,000 people in the Netherlands are hard of hearing. They suffer so much from hearing loss that it limits their daily lives. Prof. Dr. ir. Johan Frijns treats people with hearing loss, conducts research on hearing implants, and gladly shares his knowledge about electrical stimulation of the nervous system. "We shouldn't want to reinvent everything in every little corner. What we learn in one place, we can also use in another."

Johan Frijns is a professor of Otology and Physics of Hearing in the Department of Otorhinolaryngology at LUMC. He heads the Center for Audiology and Hearing Implants Leiden (CAHIL) and the Cochlear Implant Rehabilitation Centre Leiden (CIRCLE). He was recently appointed as a Medical Delta professor with a position at the Faculty of Electrical Engineering, Mathematics, and Computer Science at TU Delft.

Read more: Portrait and video Johan Frijns: “When a deaf child suddenly hears and learns to talk, this also has a huge impact on the people around him.” | Medical Delta

New book with contribution of Geert Leus

• Wednesday, 24 January 2024

Sparse Arrays for Radar, Sonar, and Communications

Science Advances paper co-authored by Geert Leus

• Friday, 19 January 2024

"Four-dimensional computational ultrasound imaging of brain hemodynamics" by Michael D. Brown et al. appeared in Science Advances on 17 January, 2024.

Read the paper at the following link.

Dr. Bahareh Abdi (EEE) awarded the TU Delft Open Education Stimulation Fund

• Wednesday, 17 January 2024

Dr. Bahareh Abdi (EEE) was awarded with the TU Delft Open Education Stimulation Fund for her project “Enhancing Electrical Engineering Education: A Digital Twin and Interactive Manual Approach for Dynamic Hands-On Learning.” Dr. Abdi will receive €20k over one year period to develop interactive textbook equivalent of BSc EE lab manuals with added digital content to improve students’ learning experience.

In this project, Dr. Abdi collaborates with Dr. Seyedmahdi Izadkhast (EEE), Dr. İlke Ercan (EEE), and Prof. Dr. Ir. Alle-Jan van der Veen (SPS) from the Department of Microelectronics, and Dr. Serdar Asut from the Faculty of Architecture who has expertise on the subject. The departure point of the project is Integrated Project-3 in the BSc EE curriculum. The outcome of the project will inform the improvement of other integrated projects and lab courses. 

Mario Coutino wins best PhD thesis award

• Tuesday, 12 December 2023

Mario Coutino has won the IEEE Signal Processing Society Best PhD Dissertation Award. Mario received his PhD in Apr 2021 with a dissertation titled "Advances in graph signal processing: Graph filtering and network identification" (promotor: Geert Leus (SPS)).

Congratulations!

Justin Dauwels named IEEE SPS Distinguished Lecturer for 2024

• Tuesday, 28 November 2023

Justin Dauwels has been named IEEE SPS Distinguished Lecturer for 2024. The Distinguished Lecturer Program provides means for IEEE chapters to have access to individuals who are well known educators and authors in the fields of signal processing. The selection is quite rigorous and only 5 persons are appointed as DL each year.

Lecture topics for Justin are:

• Deep Generative AI
• Machine Learning for Applications in Neurology
• Machine Learning for Applications in Psychiatry
• Perception Error Modelling for Autonomous Driving

New NWO-TTW project for Geethu and Justin (SPS)

• Monday, 27 November 2023

Laser satellite communications is a promising technology to support worldwide access to telecommunications services. A major technological challenge is atmospheric turbulence impacting the propagation of the laser beams. Its effect can be mitigated by adaptive optics and geographic diversity. The DAILSCOM project aims to provide a map of the effective optical channel performance over Europe. This map is needed to design ground network technology and estimate communications service availability. Since our current understanding and ability to estimate the channel performance are limited, we will develop novel physics-informed machine learning algorithms to formulate the optical link performance map.

Project PIs are: Rudolf Saathof (AE), Justin Dauwels, Geethu Joseph, Sukanta Basu (CiTG)

Justin Dauwels IEEE Signal Processing Magazine Area Editor

• Monday, 27 November 2023

Justin Dauwels has been appointed Columns & Forums Area Editor for IEEE Signal Processing Magazine. Congratuations!

Richard Hendriks Senior Associate Editor

• Friday, 10 November 2023

Richard Hendriks has been appointed as Senior Associate Editor of IEEE Transactions on Audio Speech and Language Processing. Congratulations!

Model-Driven Decisions Lab initiated

• Friday, 29 September 2023

On 28 September, a Memorandum of Understanding was signed between the TU Delft and the national police, to collaborate on research around safety and security, and forensic research. Part of this initiative will be a new research lab, the "Model-Driven Decisions Lab". This lab will start up with 5 PhD positions. Scientific director is Justin Dauwels (SPS).

SPS Summer outing 2023

• Monday, 17 July 2023

Best paper award for Costas Kokke

• Wednesday, 14 June 2023

Costas Kokke received the best contribution award at ISCS, the International Symposium on Computational Sensing. Congratulations!

IEEE SPS Best Thesis Award for Geethu Joseph

• Wednesday, 14 June 2023

Geethu Joseph received the IEEE Signal Processing Society Best PhD Thesis Award at ICASSP'23. Annually, there are 2 such awards. The 2nd award went to Elvin Isufi, former PhD student of Geert Leus. Congratulations to both!

These awards were installed 3 years ago; until now, there have been 3 recipients.

Where is everyone?

• Monday, 12 June 2023

Greetings from ICASSP'23, Rhodes, Greece!

Best paper award for Sofia Kotti!

• Monday, 12 June 2023

We are proud of our PhD student Sofia-Eirini Kotti who won the best paper award for her work “Modeling nonlinear evoked hemodynamic responses in functional ultrasound” at the Data Science and Learning Workshop during #icassp2023. Special thanks to Aybüke Erol for her contributions to this research!

Best Paper Award at the SITB'23 symposium

• Monday, 15 May 2023

At the 43rd Symposium on Information Theory and Signal Processing in the Benelux, our MSc student Alan Hamo has won the award for Best Student Presentation, for the paper "Machine learning algorithm to predict cardiac output based on arterial pressure measurement". Congratulations!

Paper abstract

Cardiac output (CO) plays a crucial role in determining the delivery of oxygen to tissues and is a key metric in hemodynamic optimization. The gold standard method for measuring cardiac output is through thermodilution using pulmonary artery catheter, but it is an invasive procedure associated with complications during placement and the need for a skilled expert to perform the measurements. An alternative approach is to estimate cardiac output by utilizing arterial blood pressure (ABP) measurements, which is a minimally invasive technique. However, the relationship between ABP and CO is not yet fully understood. In this study, we aim to utilize regression-based machine learning techniques and feature engineering to estimate CO from ABP. Hemodynamics and waveform features, along with demographic information of the patient, were integrated to enhance the  accuracy of the model.

MSc thesis supervisor: Justin Dauwels. Joint work with Drs. Niki Ottenhof and Jan-Wiebe Korstanje of the Erasmus MC. 

Where is everyone?

• Friday, 12 May 2023

43rd Symposium on Information Theory and Signal Processing in the Benelux (SITB 2023)

Assistant professor Raj Rajan appointed secretary of "De Nederlandse Vereniging voor Ruimtevaart" (Netherlands Space Society)

• Wednesday, 19 April 2023

NVR has approximately 1200 members. They include aerospace professionals, students, and people with a personal interest. As of today, the association has been around for over 70 years. Yet, it remains in touch with the newest Dutch and international developments surrounding space and astronautics. The NVR aims to be an association for all ages, and the network for anyone excited by modern aerospace and space travel.

Raj Rajan has joined the Board of NVR as Secretary, starting Feb 2023.

IEEE SPS Student Scholarship Program

• Thursday, 2 March 2023

The IEEE Signal Processing Society (SPS) awards scholarships of up to a total of US$7,000 for up to three years of consecutive support to students who have expressed interest and commitment to pursuing signal processing education and real-world career experiences.

Students and graduate students from all 10 IEEE Regions are encouraged to apply!

This is not a full scholarship but rather a supplementary fund in order to support the signal processing field.  Apply before 30 June 2023.

Congratulations Class of 2022!

• Saturday, 31 December 2022

In 2022, SPS (CAS) graduated 48 MSc students, an all-time record for us! In particular, Justin Dauwels (12) and Richard Hendriks (8) had very popular topics, while also biomedical is seeing increasing interest. Four graduates stayed with us to to a PhD.

Congratulations to all the young Engineers!

CAS becomes SPS

• Wednesday, 28 December 2022

The Circuits and Systems (CAS) group was formed over 20 years ago out of a merger of the Electronic Techiques and the Network Theory group. The resulting group had a focus on system theory and digital VLSI design, in particular design and verification tools.

Over these years, the signal processing part of the group has grown, to a point where now it is time to change our name. Starting from 1 January 2023, the group is called Signal Processing Systems (SPS). Happy New Year!

Nature paper for Gerard Janssen

• Thursday, 17 November 2022

GPS signals are not available everywhere. Especially in urban environments, the satellites may be blocked or the signals experience multipath, leading to timing errors and inaccurate positions in areas where accuracy matters most.

In the SuperGPS project of Gerard Janssen e.a., it is shown how relatively inaccurate wireless transceivers can be networked; the use of wideband signals and carrier phase recovery techniques results in a localization accuracy at decimeter-scale or better; two orders of magnitude better than GPS in certain situations. The perspective is that such transceivers can be embedded in 4G/5G mobile communication networks (the employed protocols are compatible), such that it is relatively inexpensive to roll out this solution.

In the Nov issue of Nature, the project partners describe experimental results acquired on a testbed at The Green Village at TU Delft. An associated article provides an introduction and a perspective. The topic was also covered by IEEE Spectrum.

2022 4TU.NIRICT Community Day, November 14

• Thursday, 20 October 2022

Identified are 3 themes around which we would like to bring NIRICT researchers together to discuss and to explore per theme the opportunities for interdisciplinary collaboration. The themes are related to the societal challenges we are currently facing and require interdisciplinary effort in order to be addressed effectively. We believe that productive discussions during the event could be instrumental in forming new research lines, but also project consortia around various national R&D investment programs addressing the knowledge and innovation agenda’s (KIAs) of the Netherlands (https://www.topsectoren.nl/missiesvoordetoekomst).

Themes of the day:
• Health and ICT People live longer, more healthcare is needed. Work pressure in healthcare is rapidly increasing. More data is collected by smart devices. How can the ICT community contribute solutions to keep our society healthy?
• Energy and ICT The current explosion of the energy prices shows that we are facing an unstable situation with respect to energy production and consumption. We produce more solar energy on one hand and consume more energy due to air conditioners and electric cars. We need smart solutions both in production and consumption. Moreover, due to its significant share in energy consumption, ICT systems from the end user devices to power-hungry data centers need to become more energy efficient.
• Agriculture and ICT We just had another summer with extreme heat waves and droughts. The climate changes we are facing have tremendous consequences for agriculture. The population is growing and the pressure on nature is increasing. What smart solutions do we see to deal with the challenges to keep our planet healthy and at the same time to keep the food production in line with the population growth?
The Community Day starts at 12:30hrs with a walk-in lunch. At 13:30hrs the official part will start with a number of short presentations, followed by several parallel sessions regarding the three themes of the day. You can indicate your preference on the registration form. Afterwards there will be a networking drink from 17:00-18:00hrs.

It would be great to welcome many of you on November 14th, because together we strengthen the ICT community of The Netherlands, so join us for the 4TU.NIRICT Community Day 2022, November 14, in Van der Valk Hotel Utrecht and register here by November 6, 2022.
Your data will only be used for registration and will be discarded within one week after the 4TU.NIRICT Community Day. If you have any questions, don’t hesitate to contact us: m.mommers@tue.nl

Congratulations Ids

• Friday, 9 September 2022

CAS PhD student Ids van der Werf won the Best Presentation Award at the 2022 IEEE-SPS / EURASIP Summer School "Data and Graph Driven Learning for Communications and Signal Processing". Congratulations!

Ids started just 3 weeks ago as a PhD student; the presentation was actually based on his MSc thesis work on Sound Field Reconstruction.

Where is everyone?

• Thursday, 1 September 2022

CAS outing 2022

• Monday, 11 July 2022

Congratulations for Miao Sun

• Friday, 17 June 2022

PhD defense - 15 June 2022

New NWO project: GRASPA

• Friday, 17 June 2022

Six research projects receive funding within the Open Technology Programme (OTP) this month. The projects receive a total of about 4 million euros from NWO; organisations involved in the research projects invest a total of 1.3 million euros. Among these 6 projects is a project from prof. Geert Leus from the CAS group; Graph Signal Processing in Action (GraSPA).

Graph signal processing (GSP) is the exciting research field that extends concepts from traditional signal processing to signals living in an irregular domain that can be characterized through a graph. GSP is extremely promising for applications in transportation networks, smart grid, wireless communications, social networks, brain science and recommender systems, to name a few. This project focuses on the non-trivial extension of GSP to time-varying or dynamic networks, where either the connections or the nodes can change. We will develop innovative tools to estimate such time-varying graphs from data and devise new graph filtering schemes for denoising, interpolation, and prediction. The developed techniques will be applied to brain activity monitoring, which is crucial to understand the working of the brain, as well as recommender systems, which are omnipresent in our daily lives.

More information: Four million euro for six technological research projects | NWO

New NWO-TTW project: GraSPA

• Friday, 3 June 2022

Graph signal processing (GSP) is the exciting research field that extends concepts from traditional signal processing to signals living in an irregular domain that can be characterized through a graph. GSP is extremely promising for applications in transportation networks, smart grid, wireless communications, social networks, brain science and recommender systems, to name a few. This project focuses on the non-trivial extension of GSP to time-varying or dynamic networks, where either the connections or the nodes can change. We will develop innovative tools to estimate such time-varying graphs from data and devise new graph filtering schemes for denoising, interpolation, and prediction. The developed techniques will be applied to brain activity monitoring, which is crucial to understand the working of the brain, as well as recommender systems, which are omnipresent in our daily lives.

This project will fund 2 PhD students. PIs are Geert Leus and Elvin Isufi.

Happy Secretaries Day!!!

• Thursday, 21 April 2022

Our congratulations and best wishes!

Raj Thilak Rajan elected to the IAF-SCAN committee

• Wednesday, 16 March 2022

Raj Thilak Rajan was elected to the Space Communications and Navigation Committee (SCAN) of the International Astronautical Federation (IAF).

The SCAN committee is responsible, within the IAF, for all aspects of satellite-based communication and navigation systems and technology. The committee will organise the SCAN symposium at each IAC and, when appropriate, will organise specialised plenary sessions and Highlight Lectures. It will provide annual inputs to the IAF report to the United Nations.

For more information: see link.

Justin Dauwels wins the Vincent Bendix Automotive Electronics Engineering Award

• Wednesday, 2 February 2022

A paper by Justin Dauwels (from the time he was in Singapore) has won the Vincent Bendix Automotive Electronics Engineering Award.

The paper is “An Optimal Controller Synthesis for Longitudinal Control of Platoons with Communication Scenarios in Urban Environments and Highways”.

SAE is a foundation for automotive engineers, founded in 1904 by Henri Ford, and with 100,000+ members.

Alle-Jan van der Veen appointed IEEE Signal Processing Society Vice President

• Wednesday, 26 January 2022

Alle-Jan van der Veen has been appointed IEEE Signal Processing Society Vice President for Technical Directions, for the term 2022-2024.  In this capacity, he is member of ExCom and chairs the Technical Directions Board (comprising the 12 Technical Committees of the SP Society).

Minister of Education "meets" Jorge

• Friday, 21 January 2022

On 20 January, our new Minister of Education Robbert Dijkgraaf interviewed a few young scientists. Jorge Martinez was one of them.

On twitter: "Jonge wetenschappers inspireren mij enorm. Ik sprak ze over hun ideeën voor de toekomst van de wetenschap en hun werkdruk en de werkcultuur. Daarnaast vertelden zij over interne en externe stressfactoren en het pad dat zij voor zich zien, ook buiten de wetenschap. Wordt vervolgd!"

ISOCC 2021 ISE President Best Paper Award

• Wednesday, 8 December 2021

At this year's edition of ISOCC 2021, our former excellent bachelor and master student, Jun Feng, obtained the "2021 ISE President Best Paper Award" for his paper: "A Versatile and Efficient 01.-to-11 Gb/s CML Transmitter in 40-nm CMOS."

Ph.D. candidates Mohammadreza Beikmirza and Milad Mehrpoo contributed to Jun's outstanding work. Jun is now a Ph.D. candidate at KU-Leuven. We congratulate Jun, Mohammadreza, and Milad for this award and send our best wishes for their future research.

New DAI Labs for Raj Rajan

• Thursday, 11 November 2021

Raj was awarded a DAI Labs, with 3ME (Manon Kok). This will fund 2 PhD positions for CAS, and generate some TU-wide publicity.

The awarded proposal is "Sensor AI" and investigates sensor networks, sensor fusion, distributed learning, sensor swarms. The 3ME part also looks at human motion estimation and including physical knowledge into learning systems.

New Medical Delta professors

• Friday, 5 November 2021

Congratulations Wouter Serdijn, Natasja de Groot and Andrew Webb on your inauguration as Medical Delta Professor

Innatera: Spiking neural network architecture

• Wednesday, 6 October 2021

A short video about the Spiking Neural Hardware Architecture Innatera is working on. (Innatera is a startup originating in the CAS group.)

Welcome Geethu Joseph

• Monday, 16 August 2021

The Department of Microelectronics and CAS are welcoming dr. Geethu Joseph, who starts in August 2021 as a new Assistant Professor. Her expertise is in signal processing for communications, in particular machine learning, compressed sensing and sparse recovery on graphs and networks, as well as dynamical systems theory.

Rising star: Justin Dauwels

• Thursday, 1 July 2021

As associate professor within the Circuits and Systems group at the faculty of Electrical Engineering, Mathematics and Computer Science, rising star Justin Dauwels is fascinated by how data-driven intelligent systems can tackle many of society’s problems. Neural networks and graphical models are his hammer, and he applies it to topics ranging from autonomous cars to digital health and beyond. But neuroscience has always been a main theme ever since his Erasmus exchange to the Institute of Neuroinformatics in Zürich. ‘I apply the models that I studied as an engineer to help unravel and understand human behaviour,’ he says.

Healthcare for the underprivileged

We want to be able to answer what-if questions prior to the operation; if I remove some tissue here, what will that accomplish?

Planning epilepsy operations

Next generation deep learning

But it is not all about applications. He will also delve into the fundamentals of deep learning. ‘People start to realise that neural networks, on their own, won’t be able to address all problems. I have been working on graphical models for more than twenty years and these are a promising path towards next generation deep learning systems. It’s about building a better hammer, one that combines the fast, instinctive thinking of neural networks with the slow thinking of graphical and logical models. Just like in humans, it makes sense to verify an intuitive answer and to correct it if necessary.’

New TTW project for Richard Hendriks and Jorge Martinez

• Thursday, 24 June 2021

This week Richard Hendriks received the news that his project proposal 'Personalized Auditory Scene Modification to Assist Hearing Impaired People' was granted funding from NWO TTW. Co-applicant is Jorge Martinez. The project will fund 2 PhD students. Congratulations!

Hearing impairment has become a serious problem for a large and increasing portion of the population as is shown by its prevalence: 11 % of the Dutch population suffers from severe hearing loss. These numbers show that hearing impairments form an important societal problem. It comes with many daily life problems in both private situations and working environments. Moreover, hearing-impaired people are often less confident in practical situations, need more assistance and have a worse quality of life. The underlying reasons are that hearing-impaired people suffer from a) the inability to understand speech in acoustic challenging situations, and, b) the inability to correctly localize sound sources. The goal of this project is to make the hearing impaired user fully benefit from an improved intelligibility and sound localization. To do so, we will develop an algorithm to personalize the presented auditory scene for improved speech intelligibility and sound localization for hearing impaired users.

The rising stars of the TU Delft, featuring Dante Muratore

• Tuesday, 8 June 2021

After his PhD in what he calls “hardcore analogue microelectronics”, rising star Dante Muratore knew he wanted to continue his career working on systems that are closer to an actual application. A postdoc position at Stanford University, in which he worked on the electronics for an artificial retina to treat medical conditions leading to the loss of vision, brought him just that. Then, wanting to come back to Europe and to continue doing bioelectronics at the highest level possible, an opening at TU Delft crossed his path. ‘It was the easiest choice I ever made,’ he says.

Read more about Dante in the link below!

Mario: cum laude PhD defense

• Wednesday, 21 April 2021

Today Mario Coutino defended his PhD thesis, and received the title "cum laude". Congratulations!

"Tech for Health" featuring Natasja de Groot

• Sunday, 28 March 2021

Please donate to our research on better understanding of cardiac arrhythmia

Geert Leus wins EURASIP Technical Achievement Award

• Friday, 26 February 2021

Geert Leus will be this year's recipient of the European Association for Signal Processing (EURASIP)'s Technical Achievement Award.
The award will be presented at EUSIPCO 2021 Dublin, Ireland.

As noted in the announcement by Maria Sabrina Greco, EURASIP Director for Awards,
The Technical Achievement Award honors a person who, over a period of years, has made outstanding technical contributions to theory or practice in technical areas within the scope of the Society, as demonstrated by publications, patents, or recognized impact in this field.

This year’s recipient is Geert Leus, “For contributions to signal processing for communications and sensing".

Cardiac mapping of very young children reveals conduction disorders related to atrial fibrillation

• Friday, 8 January 2021

Research of Medical Delta professor prof. dr. Natasja de Groot (Erasmus MC and TU Delft)

New Associate Professor: Justin Dauwels

• Saturday, 12 December 2020

Per 1 January 2021, Prof. Dauwels starts at CAS. His expertise is in statistical signal processing and machine learning. Welcome!

EUSIPCO 2020 - virtual event on 18-22 January 2021

• Sunday, 6 December 2020

The organizing committee of EUSIPCO 2020 has decided that EUSIPCO 2020 will be a full virtual conferencedue to the on-going COVID-19 pandemic and resulting travel restrictions.

We are excited to be able to provide a virtual venue for EUSIPCO 2020 and hope you will join us and learn about the latest developments in research and technology for signal processing.

Please register today for a reduced fee of E 50, or sign up for one of the tutorials for E 25.

CAS startup Innatera Nanosystems raises €5 million

• Wednesday, 25 November 2020

Innatera, started by CAS members Sumeet, Amir and Rene, is making the next step

Congratulations for Aydin

• Thursday, 22 October 2020

On 20 October, Aydin defended his PhD thesis. Congratulations!

New project: PCaVision

• Thursday, 27 August 2020

Prostate cancer detection using ultrasound

SSCS WYE Webinar

• Tuesday, 4 August 2020

Webinar: To Academia, or to Industry, That is the Question. Presented by: Kofi Makinwa and Shin-Lien Lu

Abstract:

REGISTER TODAY!

Medical Delta Professors appointed

• Monday, 22 June 2020

Medical Delta appointed 9 new MD Professors, with joint appointments at LUMC and TU Delft, or Erasmus MC and TU Delft. Three of these are connected to the MicroElectronics Department: Wouter Serdijn, Andrew Webb, and Natasja de Groot.

Prof. Dr. Natasja de Groot (Erasmus MC, TU Delft) researches the use of sensors and catheters to more accurately diagnose and treat cardiac arrhythmias. At TU Delft, she will have an affiliation with CAS and BE.

Prof. Dr. ir. Wouter Serdijn (TU Delft, Erasmus MC) researches the use of bioelectronics in medical research. At EMC, he will have an affiliation with Neuroscience.

Prof. Dr. Andrew Webb (LUMC, TU Delft) researches how imaging can be more widely available for medical purposes. He is a professor in MRI at LUMC, and already had a part-time appointment at CAS.

The new Medical Delta professors introduce themselves and their research in a short video. This can be viewed here:

TU Delft launches first eight TU Delft AI Labs

• Wednesday, 17 June 2020

TU Delft is setting up eight new AI Labs to investigate how artificial intelligence (AI) can accelerate scientific progress. To this end, scientists researching AI will be working together with scientist who use AI in their research. The first of these eight interdisciplinary AI labs will be followed by another sixteen in the course of 2020 and 2021.

Rapid developments in AI, data science and digitalisation can accelerate scientific progress in all fields, ranging from medical science to infrastructural research, and across all levels, from fundamental to applied research. TU Delft is boosting collaboration between AI scientists and scientists in other domains, with the launch of a series of TU Delft AI Labs.

Within the MACHINA Lab, for example, researchers in machine learning work together with materials scientists on the analysis of existing materials and the development of new materials. Within the AidroLab, researchers in geometric deep learning are working with researchers in water management on subjects such as how to improve flood forecasting in the urban environment. Researchers in the CiTyAI-Lab will use a wide variety of data sources to map the impact of the city's 'fabric' on its inhabitants in order to improve the living environment.

Education

Investments

DeTAIL: Delft Tensor AI Lab

Real-life biomedical data is often high-dimensional. Current signal processing solutions artificially segment such high-dimensional data into shorter one- or two-dimensional arrays, causing information loss by destroying correlations between these data. At the same time, advances in biomedical sensor and imaging technology – such as substantially larger recording durations of wearable sensor technology and the unprecedented increase in spatial and temporal resolution of the latest neuroimaging techniques – have led to ever increasing data sets. Tensors (multi-dimensional arrays) are the data structure of choice in artificial intelligence research to exploit the full potential of these data in a timely manner.

Within the DeTAIL Lab, we focus on both the development and application of novel low-rank tensor methods for biomedical signal processing, thereby enabling a much faster training of AI models from large datasets without any loss of accuracy.

We will exploit an as of yet unused property of real-life data; the fact that different modes of data may be correlated. Using tensor decompositions, we can find these correlations as well as compress the data, speeding up computations significantly.

Our findings will, for example, be applied to detect events, such as epileptic seizures, through the classification of multichannel time series data based on labelled training data. We also aim to reveal hidden structure, such as functional networks, in neuroimaging data. As biomedical innovation is a defining characteristic of the TU Delft, we will develop an interfaculty elective course on AI tensor methods to satisfy the expected continual increase in demand for such knowledge

Opening Airport Technology Lab

• Tuesday, 26 May 2020

This Friday, May 29 at 15:30, the opening of the Airport Technology Lab will be done by Ron Louwerse (Director of the Rotterdam The Hague Airport), Prof. dr. ir. Tim van der Hagen (President of the Board and Rector Magnificus of TU Delft) and Henk Jan Gerzee (Chief Digital Officer of the Royal Schiphol Group).

The Microwave Sensing, Signals, and Systems section (Department of Microelectronics, EEMCS faculty), chaired by Professor DSc. Alexander Yarovoy, participates in the Laboratory and is going to use their professional expertise and advanced radar facilities to develop modern remote sensing techniques to extend the sensing capability of airport radars for constant weather monitoring. This real-time monitoring with high spatial and temporal resolution is aiming to improve air traffic planning (departing/arriving on time, less fuel, less noise, less pollution, etc.) and safety.

The Airport Technology Lab (ATL) is an organization for the development, test, and demonstration of the innovative products and services for airports. The digitization of everything that happens at airports and the application of Artificial Intelligence is increasing exponentially. With all that data and the right IT platform, companies can create and test new services and products. The most successful innovations will help to provide passengers with more comfort and convenience, make the airliners maintain smoother and more efficient flights, and the handlers will optimize their processes. Ultimately, the airport also contributes (extra) to its social responsibility to reduce air pollution, CO2 emissions and noise pollution.

New project: ADACORSA

• Tuesday, 25 February 2020

Autonomous drone flight, beyond visual line of sight, safe and reliable

Wouter Serdijn appointed theme leader of Delft Health Initiative 2.0's NeuroTech theme

• Wednesday, 4 September 2019

The Delft Health Initiative has laid the foundation for connected health-oriented research at TU Delft and will continue to focus expertise, develop talent and to connect researchers to national and international initiatives. Wouter Serdijn will lead this for Neurotechnology.

NWA Idea Generator grant for Richard Hendriks

• Tuesday, 9 July 2019

Restored sound localization for hearing impaired people

The inability of hearing impaired people to localize sound has a big impact on their well-being and self- reliance. Compared to normal-hearing people, hearing-impaired people cannot efficiently use the same localization information. In this project will be investigated whether inaudible localization information can be transformed into a different audible form.

With a sum of 50,000 euros each, NWO granted 37 out-of-the-box research ideas with the potential to make an impact in society.

The applicants receive funding from the Idea Generator programme of the Dutch National Research Agenda (NWA). A total of 1.85 million euros was available.

Congratulations for Miao

• Friday, 31 May 2019

Best presenter at the WIC/IEEE Benelux Symposium, Ghent

First Microelectronics Synergy Grants

• Friday, 22 February 2019

According to Professor Geert Leus who heads the ME Research Committee, the Synergy Grants are also intended to kick-start the research of young faculty, as it can be quite challenging for them to obtain funding at the beginning of their research careers. The grants cover half the costs of a PhD candidate, with the rest coming from existing research funding. ‘The submitted proposals were carefully evaluated by the ME Research Committee on the basis of their scientific quality, their clarity and feasibility, the synergy between the participating sections, and the relationship to the departmental themes. The ME Management Team (MT) then decided to award Synergy Grants to the top three proposals.’

Changes

The aim of the grants is to encourage newly emerging combinations of technologies and to facilitate cross-overs between them, thus strengthening and broadening the department's research portfolio. This goal fits seamlessly within the research strategy of ME, which has defined itself around the four themes of Health & Wellbeing, XG, Safety & Security and Autonomous Systems to better address societal challenges.

Winners

Last week, the winners were received by the ME MT. They received flowers from the head of the department (Kofi Makinwa) and had the opportunity to briefly present their proposals to the assembled MT. Below are short descriptions of the successful proposals.

Akira Endo & Sten Vollebregt: ‘The aim of our project TANDEM: Terahertz Astronomy with Novel DiElectric Materials is to develop advanced dielectric materials to realize superconducting microstrip lines with very low losses in the frequency ranges of 2-10 GHz and 100-1000 GHz. The PhD candidate will combine the dielectric deposition, characterization, material expertise and facilities of the ECTM group and the Else Kooi Laboratory, and the submillimetre wave device measurement capability of the THz Sensing Group and SRON. The aim is not only to realize low loss dielectrics, but also to understand the underlying physics that governs these losses. If successful, these microstrips will be immediately applied to enhance the sensitivity of the DESHIMA spectrometer on the ASTE telescope in Chile.’

Bori Hunyadi: ‘On one hand, the vast complexity of the human brain (10^11 neurons and 10^14 connections) enables us to process large amounts of information in the fraction of a second. At the same time, imperfections of the wiring in this vast network cause devastating neurological and psychiatric conditions such as epilepsy or schizophrenia. Therefore, understanding brain function is one of the greatest and most important scientific challenges of our times. Brain function manifests as various physical phenomena (electrical or e.g. metabolic) at different spatial and temporal scales. Therefore, the PhD candidate working on this grant will develop a novel multimodal and multiresolution brain imaging paradigm combining EEG and a novel imaging technique, fUS. The specific engineering challenge is to understand and describe the fUS signal characteristics, deal with the large amount of data it records using efficient computational tools; and finally, formulate the specification of a dedicated non-invasive, multimodal, wearable EEG-fUS device.’

Virgilio Valente & Massimo Mastrangeli: ‘The seed money of the Synergy Grant will partially support a joint PhD candidate to investigate the tight integration of an heart-on-chip device with dedicated electronic instrumentation in the same platform. Our aim is to bring sensing and readout electronics as close as possible to a cardiac tissue cultivated within a dedicated micro physiological device. The grant helps promoting the logical convergence between current departmental research activities at ECTM and BE and within the Netherlands Organ-on-Chip Initiative (NOCI) on the development of instrumented organ-on-chip devices.’

New STW project: TOUCAN

• Friday, 15 February 2019

Develop cheap 3D ultrasound using compressive sensing

New postdoc: David Aledo

• Friday, 4 January 2019

CAS welcomes David Aledo as a new postdoc, working with Rene van Leuken on the Prystine project. David received his PhD from the Polytechnic University of Madrid (UPM), Spain.

New project: Medical Delta Cardiac Arrhythmia Lab

• Thursday, 3 January 2019

with Erasmus MC; Medical Delta will fund 2 PhD students as part of a larger program aimed to unravel and target electropathology related to atrial arrhythmia

Image formation for future radio telescopes

• Monday, 24 December 2018

Radio astronomy is an interesting application area for array signal processing. We developed a new image formation tool called PRIFIRA, inspired by Sparse Bayesian Learning. Featured in ETV Maxwell 22.1.

New Assistant Professor

• Wednesday, 3 October 2018

CAS welcomes Borbala Hunyadi, a new Assistant Professor, working on Bio Signal Processing

Guillermo Ortiz selected as "Best 2017/2018 Graduate of EEMCS"

• Friday, 21 September 2018

Guillermo Ortiz is selected as "Best 2017/2018 Graduate of EEMCS" by the Dean, and is nominated to compete for the Best Graduate of TU Delft (election on 6 November).

Guillermo did his thesis work on the topic of Graph Signal Processing, which was graded with a 10. Part of his work is already accepted for publication in the GlobalSIP 2018 conference, and has been submitted to an IEEE journal.

Congratulations!

Shahrzad Naghibzadeh EURASIP 3MT contest finalist

• Sunday, 9 September 2018

The 3MT (three minute thesis) contest is an international contest for students to explain their PhD thesis within 3 minutes. It is held across many universities and countries. In the EURASIP version, students in Signal Processing are invited to submit their 3-minute video, and the best ones are invited to present their work on stage during the EUSIPCO conference.

In 2018, Shahrzad Naghibzadeh was one of 10 selected students to present her work, in the conference auditorium. By ballot of the over 200 people in the audience, she ended up in the top-three (see picture). The final selection was done by an award committee; the #1 place went to Virginie Ollier.

Congratulations!

Geert Leus in IEEE ICASSP Top Downloads

• Wednesday, 15 August 2018

An 2015 ICASSP paper co-authored by Geert Leus made it to the Top-20 Downloads list of ICASSP papers over 2015-2017.

The paper is "Compressed Sensing Based Multiuser Millimeter-Wave Systems: How Many Measurements Are Needed?", by Ahmed Alkhateeb, Geert Leus, and Robert Heath. Last year, the related journal paper also won a best (young author) paper award.

The overview of top-downloaded papers was published in IEEE Signal Processing Magazine, July 2018.

Jie Zhang wins best student paper award

• Thursday, 12 July 2018

On 10 July, Jie Zhang won one of the three best student paper awards of the IEEE Sensor Array Multichannel Signal Processing workshop (175 att.), for the paper "RATE-DISTRIBUTED BINAURAL LCMV BEAMFORMING FOR ASSISTIVE HEARING IN WIRELESS ACOUSTIC SENSOR NETWORKS", coauthored with Richard Hendriks and Richard Heusdens. Congratulations!

Best paper award for Aydin

• Friday, 8 June 2018

Aydin Rajabzadeh was selected as winner of the best student paper presentations at SPIE's Photonics Europe International Symposia held 22-26 April 2018 in Strasbourg, France, with the paper "A computationally efficient approximation of the transfer matrix model for analysis of FBG reflected spectra". Congratulations!

Leiden Marathon

• Monday, 28 May 2018

This weekend, Mario and Tarik ran the half-marathon of Leiden, and Tuomas even the full one!

(This is the event where several people had to be hospitalized because of the heath...)

CAS on the run

• Monday, 12 March 2018

Success for Tarik, Mario, Thom and Elvin after the (half) marathon of The Hague on 11 Mar 2018.

URSI Benelux Forum-2018 poster awards

• Friday, 26 January 2018

The Annual Benelux URSI meeting took place on January 25 in Delft. Around 20 posters were presented. Sharzad and Patrick won the first and second Student Poster Award.

Shahrzad Naghibzadeh proposed "PRIFIRA - General regularization using priorconditioning for fast radio interferometric imaging". while Patrick Fuchs presented "Local Maxwell two-dimensional first order electrical properties tomography".

New project: PRYSTINE

• Monday, 22 January 2018

Rene van Leuken, together with his team (Sumeet Kumar, Amir Zjajo), and Said Hamdioui at CE acquired part of a new EU project "PRYSTINE". The aim is to design programmable compute hardware for automatic driving functions, across two application targets: data fusion for robust perception; and acceleration of AI frameworks for decision making. The emphasis is on low-power compute platforms.

While the overall budget of the project is 50 ME and spans over 60 partners, Delft will sign up for 1.8 ME.

Geert Leus Editor-in-Chief of Elsevier Signal Processing

• Friday, 22 December 2017

Starting 1 January 2018, Geert is the new Editor-in-Chief of Elsevier Signal Processing (IF: 3.1)

Best student paper award for Mario Coutino and Elvin Isufi

• Monday, 11 December 2017

Mario and Elvin win the Student Paper Award contest (3rd place) at CAMSAP’17 for their paper on "Distributed Edge-Variant Graph Filters".

3D Ultrasound using a single transducer element

• Saturday, 9 December 2017

Pim van der Meulen, Geert Leus and our collaborators at Erasmus MC show how this is possible using a phase mask and compressed sensing,

Tadashi and Geert won best paper award

• Monday, 23 October 2017

The IEEE/OES Japan Chapter Young Researcher Award was won by Tadashi Ebihara for the paper "Doppler-Resilient Orthogonal Signal-Division Multiplexing for Underwater Acoustic Communication", T. Ebihara; G. Leus; IEEE Journal of Oceanic Engineering, Volume 41, Issue 2, pp. 408-427, 2016.

Tahashi was a 3-month visitor at CAS in 2013 when he worked on this paper with Geert.

TU Delft "Female Fellowship" Tenure Track Academic Positions

• Monday, 16 October 2017

All academic levels; apply before Jan 8, 2018.

Geert Leus appointed IEEE SPS Distinguished Lecturer

• Monday, 4 September 2017

Geert Leus has been selected to serve as an IEEE Signal Processing Society Distinguished Lecturer for the term 1 January 2018 through 31 December 2019. This is considered a prestigious sign of recognition.

Best Student Presentation Award for Jamal Amini

• Tuesday, 30 May 2017

At the 2017 Symposium on Information Theory and Signal Processing (Delft, 11-12 May), organized by the IEEE Benelux Chapter, Jamal Amini received a best student presentation award. Congratulations!

Dielectric Shimming

• Thursday, 23 February 2017

Introductory article in ETV "Maxwell"

Rob Remis appointed Associate Professor

• Thursday, 26 January 2017

Per 1 January, Rob Remis has been promoted by the Dean to the rank of UHD (Associate Professor). Congratulations!

Two IEEE SPS Best Paper Awards

• Monday, 12 December 2016

Selected for the 2016 IEEE Signal Processing Society Best Paper Award:
Cees H. Taal, Richard C. Hendriks, Richard Heusdens, and Jesper Jensen
“An Algorithm for Intelligibility Prediction of Time–Frequency Weighted Noisy Speech”
IEEE Transactions on Audio, Speech, and Language Processing, Volume 19, No. 7, September 2011

Selected for the 2016 IEEE Signal Processing Society Young Author Best Paper Award:
Ahmed Alkhateeb, Omar El Ayach, Geert Leus and Robert W. Heath, Jr.
“Channel Estimation and Hybrid Precoding for Millimeter Wave Cellular Systems”
IEEE Journal of Selected Topics in Signal Processing, Volume 8, No. 5, October 2014.

The awards will be presented at the Awards Ceremony at ICASSP 2017 in New Orleans, LA

Rob Remis wins STW Open Mind 2016 award

• Monday, 28 November 2016

At their annual congres, STW awarded 5 grants (each 50 kE) to research teams to enable them to explore 'risky research' ideas. Martin van Gijzen, Andrew Webb and Rob Remis presented one of the winning proposals: an affordable MRI instrument based on permanent magnets (as opposed to superconducting magnets) for detecting hydrocephalus.

Short movie presenting the idea.

Vacancy: Team manager for Electrical Engineering Education (EEE)

• Thursday, 6 October 2016

The Faculty of EEMCS is creating a special team to fully focus on teaching using our unique and innovative ‘Delft method’. This method integrates practical and theoretical electrical engineering education and trains students to be hands-on, theoretically versed electrical engineers ready for a future career in science or industry.

We are looking for a team manager specialising in Electrical Engineering Education (EEE) who will be both a group leader and a teacher in his/her capacity as the role model of EE Education.

More information

Geert Leus named 2016 EURASIP Fellow

• Tuesday, 6 September 2016

To recognize outstanding achievements in the broad field of Signal Processing, each year the European Association for Signal Processing (EURASIP) elevates a select group of up to maximum four signal processing researchers to "EURASIP Fellow", the Association's most prestigious honor.

The EURASIP Board of Directors (BoD) has awarded prof. Geert Leus as one of the 2016 Fellows, "for contributions to signal processing for communications".

The award consists of a certificate presented during the Opening and Awards Ceremony at EUSIPCO 2016, held in Budapest (Hungary) on August 30, 2016.

Best paper award for Geert Leus

• Monday, 18 July 2016

IEEE Sensor Array and Multichannel Signal Processing Workshop, Brazil ("Stationary Graph Processes: Nonparametric Spectral Estimation" with Antonio Marques en Alejandro Ribeiro)

7 July 2016: Opening of CryoLab for Extremely Sensitive Electronic Measurements

• Friday, 8 July 2016

The CryoLab of TU Delft's Faculty of EEMCS has been opened on Thursday 7 July by the dean Rob Fastenau. TU Delft scientists from the Tera-Hertz Sensing Group, Jochem Baselmans and Akira Endo, will be leading a team of young scientists and engineers working in the lab on astronomical instrumentation. The first instrument, DESHIMA (Delft SRON High-redshift Mapper), is being developed to be operated on the ASTE telescope in the Atacama Desert in Chile. The goal of the research is to create 3D charts of so-called submillimetre galaxies that, in contrast to 2D charts, also show distance and time.

The large number of superconducting detectors, and the advanced electronics developed at SRON, allows DESHIMA to map a very large volume of space at once. While Endo leads the development of DESHIMA, Baselmans will soon install the next cryostat for testing novel THz array antennas, that will enable his upcoming instrument MOSAIC to target multiple galaxies at once. In the future, the CryoLab is envisioned to also host new coolers from QuTech. Superconducting electronics used for astronomical instrumentation and quantum electronics have much in common, because they both push the limits of what can be observed.

Rob Remis elected best teacher at EWI

• Friday, 10 June 2016

By student election (1700 votes), Rob Remis was elected as best teacher for Fac. EWI in 2016. A decade ago, Rob won already once the title 'Best teacher in EE'. This has now been extended to comprise the full faculty (EE, Mathematics, Computer Science). Later this year, Rob will compete for the title of 'Best teacher of TU Delft'.

The annual election is organised by the student associations of the Faculty (ETV, Christiaan Huygens), based on voting and written motivations.

New project "tASk-cognizant sParse sensing for InfeREnce" approved

• Tuesday, 24 May 2016

STW project by Geert Leus

New project "Earlier recognition of cardiovascular diseases" approved

• Wednesday, 20 April 2016

Atrial fibrillation (AF) is a progressive disease and associated with severe complications such as stroke. Early treatment of AF is of paramount importance as it inhibits disease progression from the treatable (recurrent intermittent) to the untreatable (permanent) stage of AF. However, early treatment is seriously hampered by lack of accurate diagnostic instruments to recognize patients who will develop new onset AF or progress to a severer form of the disease.

The goal of this project is to develop age and gender based, bio-electrical diagnostic tests, the invasive and non-invasive AF Fingerprint, which consists of electrical atrial signal profiles and levels of atrial specific tissue/blood biomarkers. In daily clinical practice, this novel diagnostic instrument can be used for early recognition or progression of AF by determination of stage of the electropathology. As such, AF Fingerprinting enables optimal AF treatment, thereby improving patient?s outcome.

The project is a collaboration between Erasmus University (Dept. Cardiology), VU Medical Center (Dept. Physiology), and TU Delft (Sections CAS and Bioelectronics), and will fund 4 PhD students.

New book by Amir Zjajo: Brain-Machine Interface

• Friday, 11 March 2016

low-power analog front-end circuits for brain signal conditioning and quantization and digital back-end circuits for signal detection

Michel Antolovic granted PicoQuant Young Investor Award

• Thursday, 3 March 2016

On February 14, 2016, Michel Antolovic was granted the prestigious PicoQuant Young Investigator Award at Photonics West in San Francisco for his paper titled 'Analyzing blinking effects in super resolution localization microscopy with single-photon SPAD imagers�. The paper shows the first localization super resolution images obtained with a SPAD camera. The analysis includes specific timing properties of fluorescing molecules in vitro with unprecedented accuracy thanks to one of the world�s single-photon fastest cameras that was created in the AQUA laboratory. The timing properties are aimed to be used for optimizing fluorophore blinking or separation of fluorophores, enabling multichannel super resolved imaging.

Sundeep's PhD Defense

• Thursday, 28 January 2016

On 25 January, Sundeep Chepuri defended his PhD thesis on "sparse sensing". Here are some pictures

Happy 2016!

• Thursday, 14 January 2016

Here are some pictures of the New Year Reception of the Microelectronics department

TU Delft Female Fellowship Tenure Track Openings

• Friday, 9 October 2015

Academic openings at all professor levels

QuTech enters in collaboration with Intel

• Thursday, 3 September 2015

Intel and QuTech, the quantum institute of TU Delft and TNO, have finalised plans for a ten-year intensive collaboration, along with financial support for QuTech totalling approximately $50 million.

New STW project: "Good vibrations"

• Thursday, 9 July 2015

Today STW announced that Rob's proposal "Good Vibrations" in the Open Technology Program will receive funding. The project will utilize the power of so-called Krylov subspace reduction techniques and develop solution methodologies for wave field problems in complex media.

The project will fund 1 PhD student: Jorn Zimmerling.

CAS Outing a big success.

• Saturday, 4 July 2015

The CAS outing this year focused on socializing, team building and having a great time and it was a big success. The group went to Outdoor Valley in Bergschenhoek for a bit of serious fun. Thanks all for participating and a big hug for the organizers.

Sundeep Chepuri wins ICASSP Best Student Paper Award

• Tuesday, 21 April 2015

The ICASSP paper "SPARSE SENSING FOR DISTRIBUTED GAUSSIAN DETECTION" by Sundeep Chepuri and Geert Leus won the best student paper award. This is quite a prestegious achievement. Congratulations!

Else Kooi Award ceremony at ICT Open

• Wednesday, 25 March 2015

The 2015 Else Kooi Award has been granted to Dr Daniele Raiteri for his scientific research on Technology-Aware Circuit Design for Smart Sensors on Plastic Foils. The Else Kooi Award is an annual award for young researchers in the field of applied semiconductor research conducted in the Netherlands. The award comes with a prize of 5,000 euros.

Raiteri has received the award during a special ceremony at the ICT.OPEN symposium on March 25th. The award was presented by the board of the Else Kooi Award foundation professor Nauta, chair of the foundation (TU Twente) and professor Edoardo Charbon. Edoardo Charbon from the microelectronics department of the EEMCS faculty holds the position of secretary of the Else Kooi Award Foundation.

Dr Raiteri?s research is focused on organic semiconductors. This emerging technology has specific features which severely complicate the design of circuits and systems, such as low transconductance, gain and speed, as well as high component variability. Dr Raiteri has devised several new solutions that have shown to be extremely robust to variability, achieving significantly better gain-bandwidth products in amplifiers and exceptional signal-to-noise ratios in voltage-controlled oscillators.

Photos by: Thijs ter Hart

Dony's PhD defense

• Friday, 20 March 2015

Some pictures appear here!

New STW project: "SuperGPS"

• Tuesday, 17 February 2015

Gerard Janssen acquired, with his colleagues Jeroen Koelemeij (VU Amsterdam, PI) and Christian Tiberius (CiTG), a new STW project called SuperGPS.

The project addresses the problem that currently, GPS is not sufficiently accurate and reliable to enable autonomous driving. The central question is: "?How do we realize highly accurate and reliable positioning using extremely accurate time-frequency reference signals, distributed through hybrid optical-wireless networks??.

The project aims at a hybrid optical-wireless system for accurate positioning, navigation, and network synchronization, to complement or even replace satellite navigation technology. This system is accomplished through a terrestrial grid of radio antenna ?pseudolites?, synchronized with extreme accuracy through the fiber-optic telecommunications network. The key deliverable of the project is a pilot demonstration of SuperGPS technology under real-life circumstances.

The technology will be developed with support and feedback from potential users in telecommunications (Royal KPN N.V.), mobility (TNO and Volvo), and Dutch high-tech manufacturers, as well as stakeholders from the scientific and R&D community, including the Dutch metrology institute VSL, the Dutch ?keepers of atomic time? UTC, and physicists and astronomers in need of better time and frequency signals.

Best paper award for Amir, Carlo and Rene

• Monday, 9 February 2015

Amir Zjajo, Carlo Galuzzi and Rene van Leuken won the Best Paper Award for the paper "Noise Analysis of Programmable Gain Analog to Digital Converter for Integrated Neural Implant Front End" at the International Conference on Biomedical Electronics and Devices (Biodevices 2015; Rome, Italy).

New STW project "From coil to antenna"

• Sunday, 21 December 2014

The STW project "From coil to antenna: development of innovative transmit array elements for MRI of the body at 7 Tesla" has been granted funding. Rob Remis is one of the applicants. The PI is Prof.dr. A.G. Webb (Leids Universitair Medisch Centrum), also Univ. Utrecht is involved.

L3SPAD honored

• Thursday, 18 December 2014

The STW HTSM "L3SPAD: A Single-Photon, Time-Resolved Image Sensor for Low-Light-Level Vision" program has received funding. The program is led by Edoardo Charbon.

Description

Best MSc student of TU Delft

• Thursday, 13 November 2014

This afternoon, Jorn Zimmerling won the competition for best MSc student of TU Delft of this year. Jorn was an MSc student of Rob Remis and Paul Urbach, and is now a PhD student with Rob at CAS.

TU Delft news article (in dutch)

QuTech appointed as 'national icon'

• Friday, 7 November 2014

The Ministery of Economic Affairs has named 4 innovations as 'national icon'; QuTech is one of them. "National icons are innovations which generate future welfare and help to solve mondial problems." The icons will receive a national support podium, including a minister or secretary of state as ambassador.

In the Department of Microelectronics, prof. Edoardo Charbon and dr. Ryochi Isihara are 2 of the 5 EWI faculty members directly involved in QuTech.

MP Jan Vos visits PARSAX

• Friday, 7 November 2014

On Friday 7 November, Jan Vos, MP for the PvdA, visited the TU Delft Climate Institute. The theme of the visit was climate change, TU Delft's research and the usefulness of and need for climate monitoring. The programme included a demonstration of cloud simulations in the Virtual Lab and a visit to the PARSAX radar. Thanks to the rain, it was possible to obtain good live measurements.

Board of Directors of EURASIP

• Thursday, 4 September 2014

On Sep 1, Alle-Jan was elected as incoming member of the Board of Directors of EURASIP, with a 4-year term starting 1 January 2015. EURASIP is the European Signal Processing association.

Stefan Wijnholds finalist for Christiaan Huygensprijs 2014

• Friday, 27 June 2014

Yesterday, 25 June 2014, Stefan Wijnholds received an "honorable mention" as finalist for the Christiaan Huygensprijs 2014, rewarding the best PhD thesis work in ICT over the past 4 years. The awards were handed by the Minister of Education (dr. Jet Bussemaker).

After a tough preselection by each university, a list of 32 candidates at a national level were judged by the jury. Out of these, 4 finalists were nominated who received a certificate in a ceremony in Voorburg.

Stefan received the honor for his PhD thesis on calibration and imaging for the LOFAR radio telescope. He was a PhD student with Alle-Jan van der Veen in 2006-2010, while being employed by ASTRON in Dwingeloo.

New STW project for Rob Remis

• Tuesday, 20 May 2014

Rob Remis was granted an STW project "Dielectric enhanced MRI". The main applicant of this project is Andrew Webb (Leiden Univ.), coapplicants are Rob Remis (CAS) and Bert-Jan Kooij (MS3). This will fund 2 PhD students in Delft.

The project aims to improve MRI imaging by inserting "bags" with dielectric materials between the magnets and the body. This should provide better illumination, in particular when using high-tesla fields. This has already been applied in practice but the effect is theoretically poorly understood. The project should provide the EM theory related to this case.

Millad's SAM'14 paper in final selection for award

• Monday, 5 May 2014

Millad's paper "Application of Krylov based methods in Calibration for Radio Astronomy" has made it to the final round of the IEEE Sensors and Multichannel (SAM) 2014 student paper competition. This selection has been made based on ranking by the TPC members. The final poster competition is Sunday, June 22, 2014 in A Coru?a.

New project: DRIFT

• Friday, 2 May 2014

The future SKA radio telescope will produce large amounts of correlation data that cannot be stored and needs to be processed quasi real-time. Image formation is the main bottleneck and requires order 350 peta-flops using current algorithms. Another bottleneck is the transportation of station data (samples) to the central location where they are correlated.

The project aims to reduce the transportation bottleneck by time-domain compressive sampling techniques, allowing the recovery of full correlation data from significantly subsampled antenna signals, and to introduce advanced algebraic techniques to speed up the image formation. Ideally, we would even skip the intermediate covariance reconstruction.

The project is funded by NWO in the "Big Bang, Big Data" program and is carried out in context of the ASTRON-IBM DOME project.

Prof. Bastiaan Kleijn in Delft for 2 months

• Thursday, 1 May 2014

Professor Bastiaan Kleijn is a part-time professor in the CAS group. He will be physically present in the period 1 May-1 July 2014.

His expertise is speech and audio signal processing. He will be collaborating with Richard Heusdens and Richard Hendriks.

New project on improved hearing aids

• Thursday, 1 May 2014

Richard Hendriks has acquired a new STW project aimed at improving the intelligibility of speech for users of hearing aids.

With a prevalence of about 11 %, severe hearing loss has become a serious problem in our society. While the current generation of hearing aids can be of a great help in certain situations, they generally are not able to provide the hearing-aid user a natural impression of the acoustical scene. An often-reported problem for hearing impaired people is the inability to understand speech in complex acoustical environments as well as the inability to localize sound.

Due to the development of wireless technology, it is possible to equip hearing aids with more powerful noise reduction algorithms to further increase the intelligibility. However, these more powerful multichannel noise reduction algorithms sacrifice naturalness of the sound environment, also when state-of-the-art binaural noise reduction algorithms are used.

This project aims at developing signal processing algorithms to help hearing aid users in these situations, by providing them a natural impression of the acoustical scene.

European Conf Antennas Propagation

• Thursday, 10 April 2014

TU Delft is platinum sponser and exhibitor at the EuCAP 2014 - The 8th European Conference on Antennas and Propagation, to be held at the World Forum in The Hague, The Netherlands, on 7 to 11 April 2014.

The Microelectronics (ME) department from the faculty of Electrical Engineering, Mathematics and Computer Science, includes research groups actively engaged on teaching and research in the field of antennas and propagation.

Located within the microelectronics department, the mission of the THz Sensing Group is to introduce breakthrough antenna technology that will revolutionize THz Sensing for Space based and Earth based applications. In the long term the research will enable multi Tera-bit wireless communications.

Alle-Jan van der Veen appointed EURASIP Fellow

• Saturday, 5 April 2014

The award is for contributions to array signal processing applied to communications and radio astronomy. In 2014, four researchers have been recognized as Fellow.

New CAS website

• Saturday, 5 April 2014

Today, CAS switched to a new website. Please enjoy, and let us know if something is not working right (contact: Alle-Jan van der Veen). CAS members can apply for a user account to maintain their own bio-page.

Georg Kail new postdoc at CAS

• Tuesday, 4 March 2014

Georg Kail is a new postdoc at CAS, working with Geert Leus on distributed localization

Przemek Pawelczak new Assistant Professor

• Friday, 17 January 2014

In July 2012, Przemek Pawelczak was awarded a VENI research grant from NWO. This grant (EUR 250k) allows the researcher to fund his own research for up to 3 years. The topic of the research is "Intelligent spectrum use in emergency networks", and it will explore statistical methods to guarantee quality of communication in Cognitive Radio Emergency Networks.

Following this, Przemek was appointed as Assistant Professor in the Embedded Software group and started in January 2013.

A new professor

• Tuesday, 14 January 2014

On 26 Sept 2012, the Board of Directors of TU Delft has decided to appoint Geert Leus as Antoni van Leeuwenhoek Professor in the CAS group. This is a `personal' full professorship aimed to promote young, excellent academics to Professor at an early age so that they can develop their academic careers to the fullest possible extent

MSc SPS Thesis presentation

• Monday, 26 August 2024
• 10:00-11:30
• HB17.140 (Seminar room)

Automated Epilepsy Diagnosis beyond IEDs by Multimodal Features and Deep Learning

Automated diagnosis of epilepsy for differentiating epileptic EEGs without Interictal Epileptic Discharges (IEDs) from normal EEGs remains a critical challenge in clinical settings. Current state-of-the-art methods use algorithms that can effectively detect epilepsy seizures which improves the current treatment methods for people suffer from epilepsy. Electroencephalograms (EEGs) analyzed by neurologists which are not able to meet the criteria are further looked into to obtain an efficient classification. However, this manual process can be time-consuming and prone to errors.

The main objectives of this research include the development of a robust multi-processing feature extraction pipeline, the application of VGG16 model / XGBoost classifier, and the validation of the proposed methods on comprehensive EEG datasets. Specifically, the focus is on detecting epilepsy in EEG data without Interictal Epileptic Discharges (IEDs) which poses a significant challenge due to the complex nature of the EEG signals in such cases.

This thesis presents an automated epilepsy diagnosis approach using a multi-algorithmic feature extraction pipeline. The final models include the development of a robust multi-processing feature extraction pipeline, the application of advanced machine learning / deep learning algorithms, and the validation of the proposed methods on comprehensive EEG datasets. The results, achieved using an XGBoost classifier with leave-one-subject-out (LOSO) cross-validation, demonstrate comparable performance to state-of-the-art epilepsy detectors. The study emphasizes the detection of epilepsy without IEDs, optimizing models through nested cross-validation, and evaluates their performance on the Temple University Hospital (TUH) and Erasmus MC (EMC) Rotterdam datasets.
 

MSc SPS Thesis presentation

• Friday, 23 August 2024
• 14:30-16:00
• HB17.140 (Seminar room)

Performance assessment of IVA-based subgroup identification methods and their application in experimental fUS data

Using independent vector analysis (IVA) to analyze and find subgroups in functional magnetic resonance imaging (fMRI) and functional ultrasound (fUS) data requires a lot of manual labour. Recently methods like subgroup identification using IVA (SI-IVA) and IVA for common subspace identification (IVA-CS) have tried to reduce this labour through automation. However, both methods did not test for accuracy. This thesis shows through simulations that these proposed methods are not accurate or robust enough to be trusted and that spectral clustering is a better alternative for automatic subgroup identification. Spectral clustering is then incorporated into the analysis of experimental fUS data of two groups of mice to try and identify these automatically. In this analysis, adaptive constrained IVA (acIVA) was used to incorporate references, further improving the interpretability of the results as components are directly linked to prior constraints. However, applying subgroup analysis showed that the mice could not be clustered based on their response to the stimuli. Still, spectral clustering is more accurate in the simulations making it a promising alternative for automatic subgroup identification. Furthermore, combining spectral clustering with acIVA makes the results more interpretable due to constrained components not being subject to permutation ambiguity.

MSc SPS Thesis presentation

• Thursday, 22 August 2024
• 13:00-14:30
• HB17.140 (Seminar room)

Towards Optimizing PMCW Radar: Low Complexity Correlation and Enhanced Frame Design

The rapid development of Advanced Driver Assistance Systems (ADAS) necessitates enhanced performance in automotive radar systems, with Phase Modulated Continuous Wave (PMCW) radar emerging as a key technology due to its high resolution, interference resistance, and robust performance. Despite these advantages, PMCW radar faces challenges such as high computational complexity and Doppler-induced range sidelobes. This thesis addresses these challenges by proposing a novel correlation method to reduce computational complexity and enhance processing efficiency, ensuring reliable target detection. Additionally, we tackle Doppler-induced range sidelobes by introducing code diversity and novel frame designs for MIMO systems, leveraging cyclic shifts and Hadamard matrices to balance sidelobe attenuation and sequence set size requirements. Through extensive analysis and simulations, the proposed methods demonstrate significant improvements in radar performance, especially in detecting weak targets behind strong reflectors. The findings contribute to developing more efficient, reliable, and scalable PMCW radar systems for advanced automotive applications.

MSc SPS Thesis presentation

• Thursday, 25 July 2024
• 14:00-15:30
• HB17.140 (Seminar room)

Advanced Digital Signal Processing for Probabilistic Constellation Shaping and Partial Response Signaling

Probabilistic constellation shaping (PCS) and partial response signaling (PRS) promise higher-speed communications but challenge the receiver-side digital signal processing (Rx DSP) developed for conventional QAM. This thesis focuses on the carrier recovery in Rx DSP, which is crucial for coherent optical communication, and proposes a carrier recovery scheme using generalized maximum likelihood estimation with negligible pilot overhead (approximately 0.2%). Through simulations and 100 GBaud experiments with PCS-64QAM, our scheme shows greater stability and accuracy, and doubles computational efficiency compared to others. Additionally, in a popular PRS scheme, Tomlinson-Harashima precoding combined with polybinary shaping (THP + Polybinary), we discuss the disputed issues regarding 2M modulo operations and investigate their impact on carrier recovery. Based on simulations and 96 GBaud experiments with THP + Polybinary-16QAM, our carrier recovery scheme enhances both accuracy and stability, effectively mitigating issues caused by zero symbols and residual inter-symbol interference. With the proposed scheme, gains from PCS and PRS can be further utilized in practical optical communication to achieve ultra-fast transmission. For the future, modifying the timing recovery and the blind equalization in Rx DSP to suit PCS/PRS systems is prioritized to further maximize PCS/PRS gains.

Conferences

• Monday, 24 June 2024
• Aula, TU Delft

7th Graph Signal Processing Workshop (GSP 2024)

Following a series of successful workshops since 2016, we are pleased to announce that the 7th Edition of the Graph Signal Processing Workshop will be held June 24-26, 2024 in Delft, The Netherlands (campus TU Delft). The workshop will provide a warm welcome to experts and practitioners from academia and industry in the field of graph signal processing (GSP). The goal of GSP is to generalize classical signal processing and statistical learning tools to signals on graphs (functions defined on a graph).

PhD Thesis Defence

• Tuesday, 18 June 2024
• 15:00-16:30
• Aula Senaatszaal

Model-based feature engineering of atrial fibrillation

MSc SPS Thesis presentation

• Monday, 17 June 2024
• 16:00-18:00
• HB17.140 (Seminar room)

Precipitation Nowcasting using Deep Generative model

Abstract: Intense precipitation can have extensive economic outcomes, from disrupting outdoor activities to causing severe infrastructural damage, such as landslides, and endangering public safety. The urgency to mitigate these impacts underscores the need for improved early warning systems. Enhanced short-term weather prediction, or nowcasting, is critical for addressing these severe weather events effectively. Traditional meteorological forecasting methods, while foundational, are often constrained by simplistic physical assumptions and fail to capture the complex, nonlinear patterns of intense weather events. These methods also struggle with high computational demands and lack the resolution needed to detect crucial microscale atmospheric phenomena for accurate short-term forecasts.

To address these challenges, this research introduces a novel deep learning approach utilizing a streamlined architecture that combines a Vector Quantized Variational Autoencoder (VQVAE) and an Autoregressive (AR) Transformer. This model aims to predict weather conditions up to 180 minutes ahead, using data analyzed at 30-minute intervals. The proposed model displays comparable performance with the state-of-theart conventional methods and other deep learning nowcasting models in predicting precipitation and sometimes extreme events. This study seeks to enhance forecasting accuracy and efficiency, providing valuable contributions to the field of meteorological nowcasting.

MSc SPS Thesis presentation

• Monday, 10 June 2024
• 16:00-18:00
• HB.20.150

Precipitation Nowcasting using Deep Generative model

Abstract: Intense precipitation can have extensive economic outcomes, from disrupting outdoor activities to causing severe infrastructural damage, such as landslides, and endangering public safety. The urgency to mitigate these impacts underscores the need for improved early warning systems. Enhanced short-term weather prediction, or nowcasting, is critical for addressing these severe weather events effectively. Traditional meteorological forecasting methods, while foundational, are often constrained by simplistic physical assumptions and fail to capture the complex, nonlinear patterns of intense weather events. These methods also struggle with high computational demands and lack the resolution needed to detect crucial microscale atmospheric phenomena for accurate short-term forecasts. To address these challenges, this research introduces a novel deep learning approach utilizing a streamlined architecture that combines a Vector Quantized Variational Autoencoder (VQVAE) and an Autoregressive (AR) Transformer. This model aims to predict weather conditions up to 180 minutes ahead, using data analyzed at 30-minute intervals. The proposed model displays comparable performance with the state-of-theart conventional methods and other deep learning nowcasting models in predicting precipitation and sometimes extreme events. This study seeks to enhance forecasting accuracy and efficiency, providing valuable contributions to the field of meteorological nowcasting.

Conferences

• Monday, 27 -- Tuesday, 28 May 2024
• Aula, TU Delft

44th Benelux Symposium on Information Theory and Signal Processing (SITB'24, Delft)

PhD Thesis Defence

• Tuesday, 21 May 2024
• 10:00-11:30
• Aula Senaatszaal

PhD Thesis Defence

• Monday, 6 May 2024
• 12:30-14:00
• Aula Senaatszaal

Multi-agent exploration under sparsity constraints

MSc SPS Thesis presentation

• Tuesday, 30 April 2024
• 10:00
• HB18.090

A SystemC SNN model for power trace generation

Power analysis can be used to retrieve key information as secure systems leak data-dependent information over side channels. A proposed solution to break the correlation between side channel information and secret information was to replace a vulnerable part of the cryptography implementation with a neural network. This uses the inherent properties of a neural network to disrupt the correlation by breaking the linear power characteristics assumed by leakage models.

To test this neural network without physically creating a hardware implementation a simulation must be performed that provides both the data and the power information. Currently neural network simulators do not generate a power trace and analog circuit simulators generate more information traces than required increasing the simulation time.

This thesis describes the creation of a complete SystemC spiking neural network model that generates both data and power information. The information generated by this model was compared and verified with results acquired by the Cadence Spectre analog circuit simulation platform. The results indicate that the created SystemC SNN model works and generates comparable data and power traces as the Spectre simulator.

Signal Processing Seminar

• Monday, 29 April 2024
• 11:00-12:00
• HB 17.140

Harmonics to the Rescue: Why Voiced Speech Is Not a WSS Process

Signal Processing Seminar

• Thursday, 25 April 2024
• 11:00-12:00
• HB 17.140

Modelling Error Correction in Sparse Bayesian Learning via Grid Optimization

Signal Processing Seminar

• Thursday, 11 April 2024
• 11:00-12:00
• HB 17.140

Multiple Sensors Facilitated Automotive Perception System

MSc SPS Thesis presentation

• Friday, 5 April 2024
• 14:30-16:31
• Snijderszaal (LB01.010)

Multiplicative Contrast Source Inversion method in Electrical Properties Tomography (CSI-EPT) based on Jacobi matrix inversion

The biological tissue electrical properties of conductivity and permittivity affect the interactions of electromagnetic fields in the body. These properties vary throughout the different tissues as the tissue structure and composition varies. In this thesis, medical imaging and diagnosis is used as primary example to motivate exploration of a novel regularization approach to an MRI-based electrical properties tomography (EPT) method. Total variation (TV) regularization has been shown to perform noise reduction in the iterative Contrast Source Inversion EPT (CSI-EPT) method. The Jacobi matrix inversion regularization, an alternative to the known conjugate gradient formulation, is elaborated and applied to an E-polarized MRI fields scenario such that this thesis presents the Jacobi step regularized CSI-EPT. The alternative regularization method outperforms the known regularization method in the reconstruction qualities of noise-suppression and edge-preservation in the simulated MRI experiments using a virtual body model. Further advancements are also described, such as multiple inner-iterations Jacobi regularization and an anatomical prior initialization of the contrast function. Important future research topics are the incorporation and evaluation of the Jacobi step regularization into more advanced CSI-EPT versions, which are the three-dimensional and transceive phase based algorithms to correct realistic MRI data.

Signal Processing Seminar

• Thursday, 4 April 2024
• 11:00-12:00
• HB 17.140

Extracting Hemodynamic Activity With Low-Rank Spatial Signatures in Functional Ultrasound Using the Tensor Block Term Decomposition

Signal Processing Seminar

• Thursday, 28 March 2024
• 11:00-12:00
• HB 17.140

Compositional Generative Modelling with Energy-based Models/Diffusion Models

1. Srikar Chaganti, “Target Localization using Distributed Radar under Communication Constraints” (MSc. Introduction)
2. Yanbo Wang, “Compositional Generative Modelling with Energy-based Models/Diffusion Models”

Signal Processing Seminar

• Thursday, 21 March 2024
• 11:00-12:00
• HB 17.140

Underwater Acoustic Communications: TNO-TUDelft Collaboration Opportunities

This talk/initiative aims to offer students the chance to collaborate with TNO, contributing to innovative research that shapes the field of underwater acoustic communications.

About Costas Pelekanakis

Costas received his Diploma from the Department of Electronic and Computer Engineering, Technical University of Crete, Greece, in 2001 and his M.Sc. and Ph.D. degrees in Ocean Engineering from MIT, USA, in 2004 and 2009, respectively. From 2009 to 2015, he worked with the Acoustic Research Laboratory at the National University of Singapore as a Research Fellow. In Singapore, he also worked as Lecturer for the Master of Defence Technology and Systems Program at the Temasek Defence Systems Institute from 2011 to 2014.  From 2015 to 2023, he was a Senior Scientist at the NATO Centre for Maritime Research and Experimentation (CMRE) in La Spezia, Italy. He is currently a Senior Scientist at the Netherlands Organization for Applied Scientific Research (TNO) in the Hague, the Netherlands. In 2001, Costas was awarded the MIT Presidential Fellowship and in 2018, he was the co-recipient of the NATO Scientific Achievement Award. In 2019, he was the co-recipient of the IET Premium Award for Best Paper in Radar, Sonar & Navigation. His broad research area is underwater acoustic communications. Costas also serves as an Associate Editor for the IEEE Journal of Oceanic Engineering and is currently a member of the IEEE Oceanic Engineering Society (OES) Administrative Committee. He is a Senior member of the IEEE.

MSc SPS Thesis presentation

• Thursday, 21 March 2024
• 16:00-18:00
• HB17.140 (Seminar room)

Sparse Non-uniform Optical Phased Array Design

This thesis addresses the design and optimization of sparse non-uniform optical phased arrays (OPAs) for advanced automotive LiDAR systems. As autonomous driving technologies advance, the demand for high-resolution, reliable, and compact LiDAR systems has become increasingly critical. Traditional uniform OPAs, while effective, face limitations regarding power consumption. This work introduces an innovative approach to designing sparse non-uniform OPAs that achieve desired performance metrics essential for automotive applications, including beamwidth, field of view, and sidelobe levels, while minimizing element count and, consequently, energy consumption.

Through mathematical modelling and simulation, we formulate the problem of sparse OPA design as an optimization problem, leveraging techniques from compressive sensing to identify the most efficient element arrangements. We propose using the sparse array synthesis method to formulate the sparse OPA design problem, utilizing algorithms such as LASSO, thresholding, and iterative reweighted l1-norm minimization to achieve optimal sparse configurations. Our results demonstrate substantial improvements in effectiveness, offering a practical solution to the constraints posed by current LiDAR systems. This thesis contributes to the field by providing a comprehensive framework for the design of sparse non-uniform OPAs, highlighting the trade-offs and benefits of various design strategies. The findings advance our understanding of OPA design principles.

MSc SPS Thesis presentation

• Wednesday, 20 March 2024
• 13:30-15:30
• HB17.140 (Seminar room)

Sparse Non-uniform Optical Phased Array Design

Extreme precipitation, like floods and landslides, poses major risks to safety and the economy, underscoring the need for sophisticated weather forecasting to predict these events accurately, enhancing readiness and resilience. Nowcasting, which uses real-time atmospheric data to predict short-term weather, is key in addressing this challenge. Traditional nowcasting systems, reliant on extrapolation from rainfall radar observations and constrained by simplistic physical assumptions, often struggle to detect complex, nonlinear weather patterns. This gap has opened the door for deep learning models, which have shown significant promise in improving the accuracy and reliability of short-term weather predictions, making them a focal point of recent research and the basis of this thesis's approach.

This thesis introduces a deep generative model designed for the nowcasting of extreme precipitation events up to 3 hours ahead, utilizing a Vector-Quantized Variational Autoencoder (VQ-VAE) to compress radar data into a low-dimensional latent representation, and an Autoregressive Transformer for predicting future radar images. Additionally, a binary classifier works in conjunction with the Autoregressive Transformer to identify extreme versus non-extreme weather events, using these classifications to inform an Extreme Value Loss (EVL) function. This loss function aims to improve the accuracy of predicting extreme weather events by addressing the data imbalance between normal and extreme precipitation occurrences. The proposed model displays comparable performance with the state-of-the-art conventional methods and other deep learning nowcasting models in predicting extreme events.

Signal Processing Seminar

• Thursday, 7 March 2024
• 11:00-12:00
• HB 17.140

Robust Tensor Decomposition Methods for Denoising Speckle Noise

Signal Processing Seminar

• Thursday, 29 February 2024
• 11:00-12:00
• HB 17.140

Cauchy-Schwarz Divergence Information Bottleneck for Regression

The information bottleneck (IB) approach is popular to improve the generalization and robustness of deep neural networks. Essentially, it aims to find a minimum sufficient representation t from input variable x that is relevant for predicting desirable response variable y, by striking a trade-off between I(x;t) and I(y;t), where I refers to the mutual information (MI). However, optimizing IB remains a difficult problem. In this talk, we study the IB principle for the regression problem and develop a new way to parameterize IB with deep neural networks, by leveraging the favorable properties of the Cauchy-Schwarz (CS) divergence. By doing so, we move away from the mean squared error (MSE) loss-based regression and ease estimation of MI terms by avoiding variational approximations or distributional assumptions. We investigate the improved generalization ability of our proposed CS-IB and demonstrate strong adversarial robustness guarantee. We observe that the solutions discovered by CS-IB always achieve the best trade-off between prediction accuracy and compression ratio. We additionally extend CS-IB to structured data such as graphs, and demonstrate its effectiveness to predict the age of patients based on their brain functional MRI (fMRI) data with a graph neural network.

Signal Processing Seminar

• Thursday, 22 February 2024
• 11:00-12:00
• HB 17.140

Cooperative Collision Avoidance for Multi-agent Systems

MSc SPS Thesis presentation

• Monday, 29 January 2024
• 13:31-15:30
• HB17.140 (Seminar room)

Towards Robust Object Detection in Unseen Catheterization Laboratories

Deep-learning-based object detectors, while offering exceptional performance, are data-dependent and can suffer from generalization issues. In this thesis, we investigated deep neural networks for detecting people and medical instruments in the vision-based workflow analysis system inside Catheterization Laboratories (Cath Labs). The central problem explored in this thesis is the fact that the performance of the detector can degrade drastically if it is trained and tested on data from different Cath Labs. 

Our research aimed to investigate the underlying causes of this specific performance degradation and find solutions to mitigate this issue. We employed the YOLOv8 object detector and created datasets from clinical procedures recorded at Reinier de Graaf Hospital (RdGG) and Philips Best Campus, supplemented with publicly accessible images. An aggregated version of object detection metrics was created for multi-camera system evaluation. Through a series of experiments complemented by data visualization, we discovered that the performance degradation primarily stems from data distribution shifts in the feature space. Notably, the object detector trained on non-sensitive online images can generalize to unseen Cath Labs, outperforming the model trained on a procedure recording from a different Cath Lab. The detector trained on the online images achieved an mAP@0.5 of 0.517 on the RdGG dataset. Furthermore, by switching to the most suitable camera for each object, the multi-camera system can further improve detection performance significantly. An aggregated 1-camera mAP@0.5 of 0.679 is achieved for single-object classes on the RdGG dataset.

THz Symposium

• Thursday, 25 -- Thursday, 25 January 2024
• 09:30-17:00
• Lecture Hall D@ta, 36.HB.01.630

Dutch Symposium on Terahertz Science and Technology - 3rd Edition

You are cordially invited to join us for the third edition of the Dutch Terahertz Symposium on January 25, 2024. It is a co-organized event between the Delft and Eindhoven University of Technology. With these series of symposiums, we aim to create an environment to exchange information and allow the Dutch Terahertz community to find collaborations and funding opportunities.

The 2024 Dutch Terahertz Symposium aims to bring together the Dutch research community and industry interested in the field of terahertz science and technology. There will be six featured talks from both academic and industrial institutions about the latest developments in the field. You can find the list of speakers and topics attached to this email. There will be plenty of opportunities for interaction with each other and inspired further community building.

Save the date to join a full day of interesting presentations, Q&As, and panel discussions!

Signal Processing Seminar

• Thursday, 25 January 2024
• 11:00-12:00
• HB 17.140

The Virtual Array: A Possible Use Case in Array Design

Physics-assisted multi-robot exploration of spatio-temporal dispersal phenomena

• Tuesday, 23 January 2024
• 17:00-18:00

Physics-assisted multi-robot exploration of spatio-temporal dispersal phenomena

The problem of exploring a dispersal of a potentially hazardous or toxic material in air using robots has a number of applications for e.g., environmental monitoring, infrastructure inspection, or civil protection, to name only a few. Especially in situations when explored substances pose a health risk to human operators, autonomous solutions are of a great interest. However, the key challenge that arises on a path towards autonomy in this context is a rather complicated  dynamics of the dispersed material, coupled with specifics of spatial aperture and low temporal resolution of olfactory (chemical) sensors used for perception. While the former precludes tele-operation (or makes it rather challenging), the latter requires perception and autonomy schemes that are able to cope with very low information rate acquired through olfactory sensing. To address these challenges the proposed solution incorporates two elements that will be discussed in this talk. First, a mobile swarm of robotic sensor carriers is used to increase spatial sampling, and thus capture spatial dynamics more efficiently. Second, a prior information about the dispersal process in terms of domain-specific knowledge is used to support data processing and autonomy. Specifically, the dispersal process is modeled with an advection-diffusion partial differential equation (PDE). The advection, or plainly speaking, the wind – a  dominant transport mechanism in a majority of practically relevant applications – is modeled using Navier-Stockes equations and explored with a robotic swarm. Furthermore, using a probabilistic (Bayesian) formulation of the PDE models, the resulting representation can be relaxed to additionally allow for more control over model mismatches.  Using data samples collected by multiple robots, the multi-robot exploration then includes two steps: (i) a cooperative solution to an inverse problem of identifying parameters of the PDEs given measurements, and (ii) exploration – the design of an optimal sampling scheme for multiple robotic platforms. This work will describe the used models, discuss the developed probabilistic inference schemes, their advantages and limitations, as well as demonstrate their performance in simulations and  in experiments.

Multi-feature-based Automatic Targetless Camera-LiDAR Extrinsic Calibration

• Thursday, 18 January 2024
• 10:00
• HB20.150 (ME Meeting room)

In autonomous driving, environmental perception, crucial for navigation and decision-making, depends on integrating data from multiple sensors like cameras and LiDAR. Camera-LiDAR fusion combines detailed imagery with precise depth, improving environmental awareness. Effective data fusion requires accurate extrinsic calibration to align camera and LiDAR data under one coordinate system. We aim to calibrate the camera and LiDAR extrinsic automatically and without specific targets. Targetless, non-automated calibration methods are time-consuming and labor-intensive. Existing advanced methods have proven that automatic calibration methods based on edge features are effective, and most focus on the extraction and matching of single features. The proposed method matches 2D edges from LiDAR's multi-attribute density map with image-derived intensity gradient and semantic edges, facilitating 2D-2D edge registration. We innovate by incorporating semantic feature and addressing random initial setting through the PnP problem of centroid pairs, enhancing the convergence of the objective function. We introduce a weighted multi-frame averaging technique, considering frame correlation and semantic importance, for smoother calibration. Tested on the KITTI dataset, it surpasses four current methods in single-frame tests and shows more robustness in multi-frame tests than MulFEAT.

Our algorithm leverages semantic information for extrinsic calibration, striking a balance between network complexity and robustness. Future enhancements may include using machine learning to convert sparse matrices to dense formats for improved optimization efficiency.

Signal Processing Seminar

• Thursday, 14 December 2023
• 11:00-12:00
• HB 17.140

Error-correcting codes and cryptography: from theory to practice

Error-correcting codes (ECCs) and cryptography schemes are indispensable to the reliability and security of numerous classic and emerging systems. Advanced ECCs play essential roles in the performance of 5G/6G wireless communications, hyper-scale distributed storage, in/near-memory computing, and quantum computing. They are also key enablers of high-density next-generation memories and wafer-scale integration, which are important pillars of the CHIPS Act. With the development of new technologies, such as quantum computers and cloud computing, traditional cryptography schemes are no longer secure and/or pose privacy concerns. There are imminent needs for post-quantum cryptography and homomorphic encryption for privacy-preserving cloud computing. Protecting the circuit chips implementing advanced functions from counterfeiting is also necessary to preserve the semiconductor supremacy achieved by the CHIPS Act.

This talk presents our recent contributions on related topics. Efficient and high-speed generalized integrated interleaved ECC decoders are developed to meet the tight latency and excellent error-correcting capability requirements of next-generation hyper-speed memories. Code construction and decoder designs are jointly optimized to enable low-latency failure recovery and continued scaling of distributed storage. Our generalized logic locking includes many previous designs as special cases and achieves better resilience towards various attacks. Logic locking that can effectively protect chips implementing fault-tolerant functions, such as machine learning, are also developed by exploiting algorithmic specifics. Last by not least, high-speed and low-complexity hardware accelerators are designed for homomorphic encryption and post-quantum cryptography.

Biography

Xinmiao Zhang received her Ph.D. degree in Electrical Engineering from the University of Minnesota. She is currently a Professor at the Ohio State University. She was a Senior Technologist at Western Digital/SanDisk 2013-2017. Prior to that, she was a Timothy E. and Allison L. Schroeder Associate Professor at Case Western Reserve University. Prof. Zhang’s research spans the areas of VLSI architecture design, digital storage and communications, cryptography, security, and signal processing.

Prof. Zhang is a recipient of the NSF CAREER Award 2009, the College of Engineering Lumley Research Award at The Ohio State University 2022, the Best Paper Award at ACM Great Lakes Symposium on VLSI 2004, and Best Paper Award at International SanDisk Technology Conference 2016. She authored “VLSI Architectures for Modern Error-Correcting Codes” (CRC Press, 2015), and co-edited “Wireless Security and Cryptography: Specifications and Implementations” (CRC Press, 2007). Prof. Zhang was elected the Vice President-Technical Activities of the IEEE Circuits and Systems Society (CASS) 2022-2023 and served on the Board of Governors of CASS 2019-2021. She was also the Chair (2021-2022) and a Vice-Chair (2017-2020) of the Data Storage Technical Committee (DSTC) of the IEEE Communications Society.  She served on the technical program and organization committees of many conferences, including ISCAS, ICC, GLOBECOM, SiPS, GlobalSIP, MWSCAS, and GLSVLSI. She has been an Associate Editor for the IEEE Transactions on Circuits and Systems-I (TCAS-I) 2010-2019 and IEEE Open Journal of Circuits and Systems since 2019. She will be the Associate Editor-in-Chief of TCAS-I 2024-2025.

Signal Processing Seminar

• Friday, 8 December 2023
• 14:00-15:00
• HB 17.140

Future of Automotive Radar and Signal Processing Challanges

The last decade has seen a rapid adoption of automotive radars to improve the safety of vehicles and to enable a higher level of driving autonomy. Even though major progress has been achieved in this effort, high-level automation has remained partially evasive due to the insufficient resolution of automotive radars.

In this presentation, we will be discussing several of the proposals to tackle those challenges, specifically the ones related to digitally modulated radars. The presentation will start with a brief discussion of the evolution of automotive radar and the alternatives to increase angular resolution, we will be focusing on the opportunities and challenges we envision for the adoption of imaging radars and specifically digitally modulated from the perspective of signal processing. Lastly, the presentation will finish with a live demonstration of one of the NXP corner radar sensors.

Presenters: Tarik Kazaz, Juan Osorio, Karan Jayachandra

MSc SPS Thesis presentation

• Friday, 1 December 2023
• 14:00-16:00
• Lecture hall Boole

Compressed Sensing in Low-Field MRI: Using Multiplicative Regularization

Magnetic resonance imaging (MRI) is a non-invasive tool to image the body’s anatomy and physiology, but suffers from long scan times. Compressed Sensing (CS) is used to accelerate MRI scans by incoherently taking fewer measurements and using a nonlinear optimization algorithm to image the undersampled data. Convex optimization techniques are generally used for image reconstruction. Minimizing a data fidelity term along with two regularization terms, which are a total variation (TV) based and a wavelet transform based function, is a standard procedure in CS-MRI. Regularization parameters are needed to balance the different terms, but it is impossible to know upfront what the optimal regularization parameters are to get the desired output. A consequence is that the algorithm of choice needs to be executed many times for many different values of the regularization parameters, which is a time-consuming process requiring knowledge of the algorithm.

In this work we rewrite and implement the regularization functions in a multiplicative manner by multiplying the data fidelity term with the regularization terms, thereby eliminating the need to tune the regularization parameters. Moreover, we include a region of support (ROS) mask to further accelerate reconstruction. The performance of different combinations of regularization functions and reconstruction algorithms are validated on a simulation study and various experiments on a low-field MRI scanner. This also shows the capability of CS applied to low-field MRI, which has lower signal-to-noise ratio compared to conventional MRI. Of all proposed methods, a nonlinear conjugate gradient method applied to the fully multiplicatively regularized objective function shows the most robust performance.

MSc SPS Thesis presentation

• Monday, 27 November 2023
• 14:00
• HB17.140 (Seminar room)

Sparse Millimeter Wave Channel Estimation From Partially Coherent Measurements

“This project develops a channel estimation technique for millimeter wave (mmWave) communication systems. Our method exploits the sparse structure in mmWave channels for low training overhead and accounts for the phase errors in the channel measurements due to phase noise at the oscillator. Specifically, in IEEE 802.11ad/ay-based mmWave systems, the phase errors within a beam refinement protocol packet are almost the same, while the errors across different packets are substantially different.

We show that standard compressed sensing algorithms that treat phase noise as a constant fail when channel measurements are acquired over multiple beam refinement protocol packets. Most of the methods that have addressed this problem treat phase noise as purely random, missing the inherent structure within the measurement packets. We present a novel algorithm called partially coherent matching pursuit for sparse channel estimation under practical phase noise perturbations. The proposed approach leverages this partially coherent structure in the phase errors across multiple packets. Our algorithm iteratively detects the support of sparse signal and employs alternating minimization to jointly estimate the signal and the phase errors.

We numerically show that our algorithm can reconstruct the channel accurately at a lower complexity than the benchmarks, and derive a preliminary support detection bound as a performance guarantee.”

Microelectronics Colloquium

• Thursday, 23 November 2023
• 15:30-16:30
• EWI Snijderszaal (LB 1.010)

Neuromorphic On-Device Intelligence for Energy-Efficient AIoT

In the swiftly advancing realm of Artificial Intelligence of Things (AIoT), the integration of edge smart devices and communication networks is becoming increasingly central to our digital infrastructure. In this context, the need for energy-efficient computing—characterized by low latency and low power consumption—is paramount, not only to enhance user experience across various AIoT applications but also to contribute to carbon neutrality. Neuromorphic computing emerges as a promising, sustainable solution, offering both efficiency and effectiveness. This presentation will delve into our recent research in neuromorphic computing, focusing on its application in speech recognition, eye tracking, and robotic control. Our work underscores the potential of neuromorphic technology to substantially reduce latency and energy consumption while managing complex tasks with negligible accuracy loss. In addition, we will delve into the application of AI for the correction of non-linearity in wideband RF power amplifiers, a critical aspect of advanced RF signal processing for emerging 6G and WiFi 7 technologies vital for connecting data-intensive AIoT devices in the future. By integrating neuromorphic computing, we aim to make AIoT devices more accessible, thereby enhancing the quality of life and fostering a sustainable, environmentally friendly future.

MSc SPS Thesis presentation

• Friday, 17 November 2023
• 14:00-16:30
• HB20.150 (ME Meeting room)

Subgraph Matching via Fused Gromov-Wasserstein Distance

Subgraph matching is a fundamental problem in various fields such as machine learning, computer vision, image processing, and bioinformatics, where detecting specific substructures within an object is often crucial. In these domains, not only structure plays an essential role, but also the feature information on nodes should be incorporated, thus highlighting the necessity for comprehensive analytical approaches.

In this thesis, we propose two novel subgraph matching frameworks using the Fused Gromov-Wasserstein (FGW) distance, namely the Subgraph Optimal Transport (SOT) and the Sliding Subgraph Optimal Transport (SSOT). Both frameworks integrate a dummy node strategy to handle the discrepancy between two graphs of different sizes. The SSOT extends upon the SOT by incorporating a sliding window framework and Wasserstein pruning to enhance the performance, especially for sparse large graphs. Our frameworks can be easily implemented and are adaptable for problems of exact matching, top-k approximate matching, and inexact matching.

We further propose a normalized FGW distance to cater to the practical interests and enhance the performance evaluation. We adopt the Frank-Wolfe algorithm for optimization and develop computation-reducing techniques by isolating the dummy node.

By conducting experiments on both synthetic and real-world datasets, we demonstrate that the SOT method achieves excellent performance on small graphs, and the SSOT method improves the accuracy over the SOT on large graphs. Both these two methods show the ability to outperform the state-of-the-art methods in noisy environments in terms of accuracy and efficiency.

Microelectronics Research Day 2023

• Thursday, 9 November 2023
• 09:30-17:30
• EEMCS

Microelectronics Research Day 2023

TU Delft Microelectronics Research Day
Fully booked!
Registering 2023 not for possible anymore

EE-NL Day

• Wednesday, 8 November 2023
• 10:30-17:30
• EEMCS

EE-NL Day

EE-NL Day

MSc SPS Thesis presentation

• Tuesday, 7 November 2023
• 14:00-16:00
• HB17.140 (Seminar room)

Malleable Kernel Interpolation for Scalable Structured Gaussian Process

Gaussian process regression (GPR), a potent non-parametric data modeling tool, has gained attention but is hindered by its high com- putational load. State-of-the-art low-rank approximations like struc- tured kernel interpolation (SKI)-based methods offer efficiency, yet lack a strategy for determining the number of grid points, a pivotal factor impacting accuracy and efficiency. In this thesis, we tackle this challenge.

We explore existing low-rank approximations that facilitates the computation, dissecting their strengths and limitations, particularly SKI-based methods. Subsequently, we introduce a novel approxima- tion framework, MKISSGP, which dynamically adjusts grid points us- ing a new hyperparameter of the model: density, according to changes in the kernel hyperparameters in each training iteration.

MKISSGP exhibited consistent error levels in the reconstruction of the kernel matrix, irrespective of changes in hyperparameters. This robust performance forms the bedrock for achieving accurate approx- imations of kernel matrix-related terms. When employing our rec- ommended density value (i.e., 2.7), MKISSGP achieved a comparable level of precision to that of precise GPR, while requiring only 52% of the time compared to the current state-of-the-art method.

MSc SPS Thesis presentation

• Friday, 27 October 2023
• 10:00
• HB17.140 (Seminar room)

ADS-B Based Trajectory Prediction for Aerial Vehicles

The evolution of aerial vehicle technology necessitates robust trajectory prediction models. These models are crucial for maintaining safe airspace and enabling autonomous operations. Automatic dependent surveillance–broadcast (ADS-B) is a surveillance system that enables aircraft to receive data from navigation satellites and periodically broadcasts it, enabling it to be tracked. Moreover, using ADS-B data for more general aerial vehicles has become a popular trend because it can provide real-time high-resolution aircraft state information and share this information with other vehicles in real-time for the aviation safety ecosystem.

In this project, we delve into ADS-B-based trajectory prediction for both aircraft and drone motion trajectories with the overarching goal of improving prediction accuracy. We initially implement several model-based

Kalman filters—including interactive multiple models (IMM)—to assess the accuracy of aircraft trajectory predictions across different model structures. The results reveal that the IMM filter outperforms the single model predictions in terms of root mean square error (RMSE).

Furthermore, we implement the Gaussian process (GP) with a sliding window scheme to predict online drone trajectories. Recognizing the high computational complexity of the GP, we also introduce a low-rank

approximation method, structured kernel interpolation (SKI) GP, aiming to conserve computational resources. Finally, we compare the prediction performances of the IMM filter, classical GP, and SKI GP on real drone

trajectories. The results highlight that the classical GP method enhanced prediction accuracy, achieving an RMSE of less than 1.7m, which is 50% lower compared to the model-based IMM filter. Additionally, the SKI GP

realizes a 25% reduction in computation time compared to the classical GP, despite a slight compromise in prediction accuracy.

ME colloquium

• Thursday, 26 October 2023
• 15:30-16:30
• EEMCS, Snijderszaal

Generative AI

Generative AI

Generative AI refers to a category of artificial intelligence models that are designed to generate new content, such as text, images, audio, or other types of data. Probably the best known example of generative AI is ChatGPT, the fastest consumer application to hit 100 million monthly active users. Generative AI models use machine learning algorithms to learn patterns and structures from existing data and then produce new data that is similar in style or content to what they have been trained on.

In this presentation, I will talk about projects in our group on deep generative models. I will briefly present novel kinds of deep generative models that we are developing in our team. Next I will explain how we are designing such models for rainfall nowcasting, where we integrate physical laws into the deep generative models.  At last, I will talk recent AI related initiatives that my team is involved in.

MSc SS Thesis Presentation

• Thursday, 26 October 2023
• 10:00-10:45
• HB 20.150

Enhancing Fiber Direction Estimation from Electrograms

For the heart to pump blood throughout the body, electrical impulses that trigger the cellular contraction must be generated and spread through the myocardial tissue. These signals propagate faster along the longitudinal cardiac fiber direction than the transverse direction, conferring the heart with anisotropic conduction properties. Therefore, the arrangement of the fibers within the tissue governs the impulse propagation. Given the variability of the fiber direction across the heart and between patients, incorporating it into electrophysiological models would enhance our understanding of the mechanisms and progression of different heart conditions, such as atrial fibrillation (AF). The study of this common cardiac arrhythmia relies on analyzing electrical recordings of the heart, known as electrograms (EGMs), which, if integrated with the patient’s fiber architecture into cardiac models, can enable effective personalized treatment. Over the years, researchers have proposed different approaches to estimate the fiber direction from EGMs. However, these methods have been evaluated in different, usually simplistic, cardiac tissue models, making their comparison, and therefore selection of the most accurate approach for clinical and research applications, challenging.

The current study aims to identify the best fiber direction estimation method under consistent and realistic conditions. To achieve this goal, synthetic EGMs and local activation time (LAT) maps were generated from 2D and 3D monodomain models that mimicked the muscle bundle, atrial bilayer, and ventricular transmural fiber rotation structures. A comparison analysis of existing fiber direction estimation methods, first as described by their authors and then standardized to have the same spatial resolution, showed the superior performance of the techniques based on fitting an ellipse to local conduction velocity or conduction slowness vectors from a whole LAT map. The estimation accuracy of these methods can be further improved by increasing the number of vectors to which the ellipse is fitted. Nonetheless, given the influence of underlying layers in the epicardial recordings, the estimation error increases in the tissue models where fibers in the epicardial and endocardial layers run perpendicularly. The effect on the estimate of such architecture, characteristic of the inferior side of the right atria and the ventricles, can be accounted for by combining epicardial electrical recordings obtained after pacing either in the endocardium or the epicardium. Although a preliminary assessment of the estimation methods was carried out with human EGMs, future studies should focus on validating the methods in a controlled experimental framework and refining them for more localized fiber direction estimation. All in all, the automation of the techniques and their integration into electrophysiological models brings us a step closer to creating valuable clinical tools for diagnosing and treating electropathologies.

PhD Thesis Defence

• Wednesday, 27 September 2023
• 12:00-14:00
• Aula Senaatszaal

Pitch-Matched Integrated Transceiver Circuits for High-Resolution 3-D Neonatal Brain Monitoring

Peng Guo will defend his PhD thesis entitled

Pitch-Matched Integrated Transceiver Circuits for High-Resolution 3-D Neonatal Brain Monitoring

on Wednesday, Sept. 27th. The layman's talk will be at 12:00, the defence starts at 12:30.

The thesis can be downloaded from the TU Delft repository. Peng's work was part of the MIFFY project.

Livestream of the defence: link

Promotors: Michiel Pertijs and Nico de Jong

Abstract: This thesis presents the design and implementation of integrated ultrasound transceivers for use in transfontanelle ultrasonography (TFUS). Two generations of ultrasound transceiver ASICs integrated with PZT transducer arrays intended for TFUS are presented. In the first generation, a novel AFE design that combines an LNA with the continuous TGC function is realized in a bid to mitigate the gain-switching and T/R switching artifacts. Besides, a new current-mode micro-beamforming design based on boxcar integration (BI) is also implemented to reduce the channel count within a compact layout. In the second generation, the AFE is derived from the first version, while the design focuses on RX backend circuitry and channel-count reduction, including a passive BI-based µBF merged with a charge-sharing SAR ADC, which digitizes the delayed-and-summed signals, and a subsequent multi-level data link, which concatenates outputs of four ADCs. In total, a 128-fold reduction in channel count is finally achieved. The techniques we developed have established the groundwork and removed the initial barriers for an electronics architecture suitable for a wearable 3D TFUS device.

MSc SPS Thesis presentation

• Wednesday, 27 September 2023
• 10:00
• HB.20.150

Friction Identification on Gantry Stage

In an era marked by the demand for unprecedented levels of precision in engineering applications, the profound impact of friction forces on motion control systems cannot be underestimated. This thesis extensively investigates the frictional behavior of the Proton Motion Stage, an advanced high-precision motion control system developed by Prodrive Technologies. This research conducts both experimental investigations and computational simulations, offering valuable insights into its friction behavior across diverse conditions and scenarios. 

The research begins with an analysis of existing models used to describe friction behavior in precision engineering systems. A critical evaluation of empirical models highlighting strengths and limitations is presented, and the LuGre friction model is selected for further research. Subsequently, the chosen model is used to simulate the behavior of the Proton Motion Stage. The simulation setup is described, including the incorporation of the LuGre friction model and the identification of system parameters. The accuracy of the identification is above 99%. The sensitivity analysis of the parameters is also conducted to enable a comprehensive exploration of friction dynamics. Finally, the research delves into static and dynamic parameter experiments, where cable slab forces' position-dependent impacts and velocity-friction maps that capture the intricate Stribeck effect are presented, and closed-loop and open-loop setups to dissect friction behavior during rapid motion changes are employed. Residual analysis of histogram and 90% confidence autocorrelation and cross-correlation is also presented to study the quality of identification. Overall, this thesis combines theory and practice to enhance our understanding of friction in precision engineering systems.

MSc SPS Thesis presentation

• Wednesday, 27 September 2023
• 10:00
• HB17.140 (Seminar room)

Small end-to-end OCR model

Optical Character Recognition (OCR) is a pivotal technology used to extract text information from images, finding wide-ranging applications in document digitization and medical records management. The integration of machine learning has ushered in an era of swift and precise OCR models. Broadly, OCR comprises two key components: detecting the bounding boxes around text instances and recognizing the characters within them. Presently, prevailing OCR models are primarily intricate two-stage systems necessitating real-time operation on remote servers. Nevertheless, end-to-end models exhibit superior performance from a data utilization perspective. There exist scenarios where offline models prove indispensable, such as in environments with restricted internet access or locales with stringent data privacy and security requirements.

This project delves into various end-to-end models, leveraging the PaddleOCR end-to-end model as a foundational reference to devise a compact OCR model tailored for edge devices. Through meticulous optimization of the backbone architecture and the introduction of diverse Feature Pyramid Network (FPN) structures within the stem network, we achieved a remarkable reduction in model size, down to 19MB. This represents a substantial advancement, constituting merely one-tenth of the original PaddleOCR end-to-end model's footprint.

By leveraging an extensive database and conducting a series of fine-tuning experiments specifically tailored for end-to-end OCR tasks involving curved text images, the model exhibits an impressive precision rate of 47.3% and an f-score of 45.3%. This achievement highlights the effectiveness of the customized loss function relative to the original model, despite its reduced size. Notably, this performance is comparable to certain end-to-end models with larger backbones. Furthermore, an Android demo has been carefully developed to demonstrate the model's capabilities on mobile devices, achieving an average processing time of 433 milliseconds per image.

MSc SPS Thesis presentation

• Wednesday, 20 September 2023
• 09:00-11:00
• HB.20.150

Efficient Content-Based Image Retrieval from Videos Using Compact Deep Learning Networks with Re-ranking

The rise of streaming and video technologies has underscored the significance of efficient access and navigation of digital content, particularly for scholars in fields like history and art. Scholars actively seek streamlined approaches to index, retrieve, and explore digital content, with a focus on locating specific instances. The process of searching for specific instances in video search is complex that requires the analysis of video sequences and the identification of relevant video segments. Advanced techniques and algorithms are necessary to ensure effective content-based retrieval of the required information.

In response to the escalating demand for accurate and swift access to relevant visual data within the vast spectrum of video resources, our research has been dedicated to the development of novel, efficient content-based image retrieval methods tailored for videos by integrating deep learning methodologies. Our comprehensive system contains two crucial components: keyframe extraction and content-based image retrieval. Keyframe extraction involves identifying significant frames within videos, while content-based image retrieval enables the retrieval of similar frames to a query image through feature extraction and ranking.

A unique aspect of our research lies in the exploration and analysis of a diverse range of feature extraction techniques derived from compact deep learning networks. We have compared our proposed method with state-of-the-art retrieval systems, evaluating performance metrics in terms of both accuracy and speed. Our method harnesses the power of compact deep learning network features in the initial ranking stage, effectively sublisting frames, and subsequently introduces re-ranking using a larger network. This innovative approach promises to deliver the best of both worlds: exceptional efficiency without compromising retrieval accuracy.

Repository link:  http://resolver.tudelft.nl/uuid:751092b8-1b3d-4335-98bd-cc26e69d374c

Signal Processing Seminar

• Wednesday, 20 September 2023
• 10:30-11:30
• HB 13.140

Automotive Radar for Autonomous Driving: Signal Processing Meets Deep Learning

Millimeter-wave automotive radar emerges as one of key sensing modalities for autonomous driving, providing high resolution in four dimensions (4D), i.e., range, Doppler, and azimuth and elevation angles, yet remain a low cost for feasible mass production. In this talk, we will address the challenges in automotive radar for autonomous driving, examine how signal processing and deep learning can be combined to optimize the performance of automotive radar systems, and outline future research directions. Our focus will be on the generation of high-resolution radar imaging using multi-input multi-output (MIMO) radar and frequency-modulated continuous-wave (FMCW) technology. We will examine the challenges of waveform orthogonality, mutual interference, and sparse antenna array design and present our recent innovations in the field, including sparse array interpolation via forward-backward Hankel matrix completion, fast direction-of-arrival estimation via unrolling iterative adaptive approach, and adaptive beamforming via deep reinforcement learning, leading to the generation of high-resolution low-level automotive radar imaging, represented in bird's-eye view (BEV) format, providing rich shape information for object detection and recognition with deep neural networks. However, the radar BEVs are in general hardly shift-invariant over both angle and range since not every pixel is generated equally. The talk will highlight the importance of physics-aware machine learning in perception task on high-resolution radar imaging. We will show how incorporating radar domain knowledge and signal structure into deep neural network design can lead to more accurate and reliable object detection and recognition. Finally, we will discuss future research directions, including integrated sensing and communication, and collaborative radar imaging via an automotive radar network.

Bio:

Shunqiao Sun received the Ph.D. degree in Electrical and Computer Engineering from Rutgers, The State University of New Jersey under supervision of Prof. Athina Petropulu in Jan. 2016. He is currently an assistant professor at The University of Alabama, Tuscaloosa, AL, USA. From 2016-2019, he was with the radar core team of Aptiv, Technical Center Malibu, California, where he has worked on advanced radar signal processing and machine learning algorithms for self-driving vehicles and lead the development of DOA estimation techniques for next-generation short-range radar sensor which has been used in over 120-million automotive radar units. His research interests lie at the interface of statistical and sparse signal processing with mathematical optimizations, automotive radar, MIMO radar, machine learning, and smart sensing for autonomous vehicles. Dr. Sun has been awarded 2016 IEEE Aerospace and Electronic Systems Society Robert T. Hill Best Dissertation Award for his thesis “MIMO radar with Sparse Sensing”. He authored a paper that won the Best Student Paper Award at 2020 IEEE Sensor Array and Multichannel Signal Processing Workshop (SAM). He is Vice Chair of IEEE Signal Processing Society Autonomous Systems Initiative (ASI) (2023-2024). He is an associate editor of IEEE Signal Processing Letters and IEEE Open Journal of Signal Processing. He is a Senior Member of IEEE.

MSc SPS Thesis presentation

• Wednesday, 13 September 2023
• 10:00-12:00
• Pulse-Hall 9, 33.A2.300

Indoor in-network asset localization using Crownstone network

Indoor localization is the process of determining the location of an asset within an indoor environment. Crownstone, a subsidiary company of Almende, wants to fulfill specific tasks within the Crownstone network like developing indoor localization algorithms to get the room-level location of an asset for smart building, home automation and healthcare applications. Crownstones act as sensors that receive the Bluetooth messages transmitted by the asset and measure the strength of the received signal(RSSI).
There are two most widely used and researched asset localization methods, namely, model-based(MB) and data-driven(DD) methods have their drawbacks, especially because of the presence of obstacles, signal interference, reflections, noise that influence the RSSI signals and the dependency of the methods on the knowledge of sensor positions. In this thesis, a centralized multilateration(MB-C) algorithm as well as a simple averaging consensus based distributed(MB-D) algorithm is implemented. Algorithms are tested on real data collected at the Almende office(test environment) divided into a finite number of rooms.
To deal with the challenges posed by the MB algorithms like selecting a model, learning parameters and an additional step of mapping the position output to a room location, an Ensemble based centralized machine learning(DD-C) data driven algorithm is proposed that classifies the asset in one of the rooms with an accuracy of 65%. Algorithm is further improved by distributed data handling leading to a classification accuracy of 77%. To perform in-network localization, a consensus based distributed ML algorithm (DD-D) is proposed that performs local predictions within the Crownstone network using the same globally trained model giving a classification accuracy of 73%. The results show that the proposed DD algorithms perform better than the MB algorithms in terms of accuracy and are comparable in terms of prediction time. They also indicate that the proposed DD algorithms are more scalable, robust against noise but are computationally expensive.
Thesis link: http://resolver.tudelft.nl/uuid:7b0eb6bb-323f-44e8-b994-28dd8c9c9467

MSc SPS Thesis presentation

• Tuesday, 12 September 2023
• 13:00-15:00
• HB17.140 (Seminar room)

Direction Finding and Localization with Bluetooth 5.2

The potential of indoor localization using the Bluetooth Low Energy approach increases with the introduction of the Constant Tone Extension (CTE) feature in BLE 5.1. These small and energy-efficient beacons transmit signals that Bluetooth-enabled devices can detect, allowing for proximity and positioning calculations. This technology supports novel applications such as indoor wayfinding. By applying the new feature with an appropriate antenna array, it is possible to estimate both the Angle of Arrival (AoA).

Nevertheless, estimating AoA poses significant challenges, particularly in indoor scenarios. Throughout the span of this project, an in-depth analysis is being conducted on several elements, including multipath propagation and frequency offset. We first look into the I/Q data processing and then dive into the effect of frequency offset. Following that, several AoA estimation algorithms and multipath-mitigation strategies are discussed. Finally, we model the overall AOA estimation problem, followed by a positioning algorithm based on the estimated angles.

In this project, we propose an end-to-end indoor BLE positioning solution. Matlab simulation is performed to assess its performance. The simulation reveals that, with appropriate settings, the Toeplitz Reconstruction (TR) method is the best. In the setting of a 4-by-4 uniform rectangular array (URA), over 90% of the results show a position inaccuracy that is smaller than 0.14 meters.

After the simulation, we conduct a real-world experiment to assess the practicality and effectiveness of the solutions.  The TR approach demonstrates a position error of less than 0.4 m, which is lower than previous BLE positioning research.

Finally, we suggest a few future research directions. This involves optimizing parameters, considering other antenna-affecting elements, etc.

Repository link:  http://resolver.tudelft.nl/uuid:ff13dbcd-da99-4599-ad7f-f3b7282681bb

MSc SPS Thesis presentation

• Thursday, 31 August 2023
• HB17.140 (Seminar room)

Using Tensor Decompositions To Obtain Biomarkers From Auditory Event-Related Potentials

Brain disorders in children pose significant challenges to their development, impacting cognition, speech, movement, and behavior. The uncertainty surrounding prognostic information at the time of diagnosis leaves families with numerous questions about the future. The Child Brain Lab at Erasmus MC Sophia Children's Hospital conducts IQ, electroencephalogram (EEG), speech, and movement tests in playful environments, enhancing scientific research and healthcare practices for a better understanding of disease progression.

The Otolaryngology department at the Child Brain Lab focuses on auditory-related potentials (ERPs) obtained from EEG measurements to predict the future development of children with brain disorders. Analyzing ERP data from experiments like Mismatch Negativity (MMN) and Acoustic Change Complex (ACC) yields insights into developmental trajectories and connections between hearing, language, and brain development.

This thesis aims to explore alternative methodologies for extracting comprehensive information from ERPs, overcoming limitations of the commonly used peak amplitude and latency analysis. Tensor decompositions are employed to exploit structural information present in the data, using data fusion methods to combine multiple datasets for improved classification and deeper insights into group differences.

Simulations on artificial ERP data demonstrate that data fusion methods perform better on two ERP tensors compared to single tensor decomposition when group differences are shared between datasets. On a real dataset, tensor decompositions show promise for classifying subjects based on auditory event-related potentials while giving more insights into the neurological sources.

This report proposes an alternative method for analyzing ERP data, highlighting the potential of tensor decompositions and data fusion techniques. 

MSc SPS Thesis presentation

• Monday, 28 August 2023
• 15:00
• Online

Path planning for Lunar rovers An Artificial Potential Field-based algorithm for the path planning of a walking Lunar rove

Abstract: 

Abstract E↵ective path planning is a key challenge for Lunar rovers. This allows for safe autonomous navigation over complex and unknown areas. Lunar Zebro (LZ), a project of the Delft Univer[1]sity of Technology, is developing a robot to be the first European rover to walk on the Moon. This tiny rover, no bigger than an A4 sheet of paper, aims to explore a wide area to monitor solar radiations.

The thesis derives a path plannning algorithm for the local navigation of LZ. State-of-the[1]art path planning methods for terrestrial and non-terrestrial robots are studied. Metrics based on the needs of LZ are defined to compare the algorithms. Artificial Potential Field (APF)- based methods are identified as the most promising for LZ. APF methods are a category of path planning approaches in which robot motion is influenced by virtual forces generated by the destination point and obstacles. A Monte-Carlo simulation in a lunar environment is run to compare APF-based algorithms. The most relevant algorithm is picked up and improved (Bacteria-Aritificial Obstacle (B-AO) algorithm) with respect to success rate, path length and computing time. Finally it is implemented on a LZ prototype and tested in a challenging envi[1]ronment. Testing is conducted on a real lunar testbed made of sand, rocks and craters. A rock and crater abundance model is established considering rock and crater coverage of 2% and 15% respectively to represent the lunar surface as close as possible.

The B-AO algorithm shows 200% higher success rate and 50% lower computing time than the conventional APF algorithm, for only 5% longer path length than the optimal algorithm A*. It also outperforms state-of-the-art APF-based algorithms by more than 15% in reachabil[1]ity and 10% in path length for a similar or shorter planning time. Field testing results exhibit the robustness of the B-AO algorithm to real-world uncertainties in di↵erent scenarios. They also show that near-optimal paths are computed in real-time with limited available processing power. The bacterial approach of the B-AO algorithm makes it faster to execute and smaller to store than path planning algorithms used on existing or past non-terrestrial rovers.

Keywords: Path planning, Moon rover, Artificial Potential Field methods, Lunar Zebro
 

MSc SPS Thesis presentation

• Friday, 25 August 2023
• 10:30-13:00
• HB17.140 (Seminar room)

Coded Excitation for Doppler Ultrasound Imaging of The Brain

Doppler ultrasound imaging of cerebral blood flow faces challenges arising from a low signal-to-noise ratio (SNR) and a wide dynamic range. Echo signals received from blood cells are significantly weaker compared to surrounding tissues, such as the skull or brain soft tissue, resulting in inhibited visualization of small blood vessels and deep brain areas. To address this issue, this thesis explored the feasibility of employing and improving coded excitation techniques to enhance the SNR of Doppler ultrasound images. Furthermore, an optimized code for Doppler ultrasound imaging is designed, represented by a generalized encoding matrix. 

The research begins with the definition of a linear signal model that incorporates the encoding matrix. Subsequently, a trace-constraint optimization problem is formulated based on maximizing the Fisher information matrix to find the optimized encoding matrix. The feasibility and performance of the optimized encoding matrix are assessed through simulations on both small and large array settings, which operate above Nyquist sampling frequency and under Nyquist sampling frequency respectively. The imaging results indicate that the optimized code exhibits higher SNR in deep image regions compared to existing coded excitation methods like Barker code while using the same number of transmissions, bit length, and same average transmit energy, albeit with a trade-off of decreased axial resolution. Nonetheless, this resolution degradation can be mitigated through the application of the iterative imaging technique LSQR. Finally, the optimized code is tested in a clinical transducer setting, and a blood flow simulation is conducted. The outcomes showcase the capacity of the proposed optimized code to enable higher SNR in Doppler ultrasound imaging and more accurate and informative clinical assessments. 

MSc SPS Thesis presentation

• Thursday, 24 August 2023
• 10:00
• HB13.140, Boardroom EEMCS

Ultrasound Imaging through Aberrating Layers using a Virtual Array

Ultrasound images are typically generated using the Delay-And-Sum (DAS) method, which assumes a homogeneous propagation medium. When an aberrating layer is situated between the sensor array and the imaging target, this assumption does not hold, and DAS is replaced with model-based methods. These methods are computationally expensive and require to accurately model the aberrations caused by the layer. This thesis investigates novel methods for image formation and aberration estimation. The effect of the layer is described using a set of transfer functions from the sensor array to a virtual array placed after the layer. In the first part, we assume the transfer functions are known, and we propose a new method for image formation. The transfer functions allow to map the signal from the sensor array to the virtual array, and the DAS method is used on the virtual array signal. This technique is equivalent to model-based matched filtering in terms of image quality, without requiring expensive matrix computations. In the second part, the transfer functions are unknown, and a novel technique is introduced for their estimation. Using pulse-echo data, a focus-quality metric is computed to quantify the accuracy of the transfer function estimate. The transfer functions are modeled using a dictionary and the dictionary coefficients are iteratively updated to increase the defined metric. The optimization leads to improved focus quality and sharper images. In the case the layer model requires a limited dictionary, the proposed algorithm generates an accurate estimate of the transfer functions.

MSc SPS Thesis presentation

• Wednesday, 23 August 2023
• 10:00
• Boardroom, HB13.140

Prediction of Post-induction Hypotension Using Machine Learning

Abstract: 

Anesthesia-related hypotension is a significant concern during surgery, occurring shortly after induction and potentially leading to severe complications. Since the anesthetic drug is believed to have an important role in the occurrence of post-induction hypotension (PIH), anesthesiologists now advocate for the appropriate selection of anesthetics dosage to avoid PIH.To facilitate such decision-making, an accurate prediction of PIH associated with a certain dosage of anesthetics is necessary.

This thesis presents a high-accuracy prediction model for PIH that supports anesthesia decision-making. The model is trained on data from the VitalDB database of 320 patients undergoing general anesthesia. The target output of this classification model is the occurrence of PIH, as defined through comprehensive analysis that incorporates clinical operations. Besides demographic data and vital signs, our model incorporates the dosage of propofol administered during the induction period as an input variable, mimicking real-world anesthetic plans. By employing the model in the target control infusion system of anesthesia, the anesthetics dosage can be varied as input, providing outcome predictions as security suggestions. An ensemble algorithm is employed to balance the prediction performance and the ability to elucidate the positive relationship between propofol and PIH risk, forming an anesthetics advice model. Compared to previous PIH prediction studies, our prediction model is validated in a more reliable nested cross-validation approach and achieves a higher performance (precision of 0.83 and recall of 0.84). We believe utilizing demographic and dynamic vital signs to predict HIP can be useful in determining the appropriate anesthetic dosage plan, offering potential improvements in patient care and safety.

MSc SPS Thesis presentation

• Tuesday, 22 August 2023
• 09:30
• HB17.140 (Seminar room)

Deep learning-empowered Content-based Video Image Retrieval (CBVIR)

The advent of streaming and video has sparked a revolutionary shift in the presentation of materials across various fields, such as history, art, and media copyright protection. In this context, scholars and rights holders are seeking efficient solutions to index, retrieve, and browse through digital content searching for a specific instance. Unlike searching a specific instance in an image, searching in a video requires more than analyzing the visual features of an image and then comparing these features to a database, for it includes processing video sequences and retrieving video segments.

Motivated by the urgent need and promising applications across diverse disciplines, we present a novel deep-learning-empowered content-based video image retrieval (CBVIR) system with a strong emphasis on real-world applications. This system offers high efficiency and considerable accuracy, addressing the challenges associated with accessing and utilizing video materials effectively. Our initial approach revolves around the extraction of informative keyframes that effectively capture essential objects within the video. This process, known as Key Frame Extraction (KFE), enables us to distill the most crucial visual representations for further analysis. After the extraction of keyframes, the relatively smaller dataset allows for content-based image retrieval (CBIR) to be conducted, retrieving similar images from a database solely based on the content of the query image. In this project, a wide range of methods is investigated and analyzed, including traditional representation, handcrafted image feature extraction, and up-to-date machine learning-based image representations. Our contribution is striking a balance between high-level and low-level image representation for this task; targeting efficiency improvement, enhanced color-based KFE module is proposed and implemented, achieving high efficiency ratio and satisfactory accuracy and targeting accuracy, a traditional and deep learning-based hybrid feature is proposed, achieving valid efficiency ratio and highest accuracy. Overall, an automatic retrieving system requiring much less user engagement is provided together with a system GUI prototype.

MSc SPS Thesis presentation

• Tuesday, 22 August 2023
• 13:00
• HB17.140 (Seminar room)

LiDAR and Radar-Based Occupancy Grid Mapping for Autonomous Driving Exploiting Clustered Sparsity

Occupancy grid maps are fundamental to autonomous driving algorithms, offering insights into obstacle distribution and free space within an environment. These maps are used for safe navigation and decision-making in self-driving applications, forming a crucial component of the automotive perception framework. An occupancy map is a discretized representation of a chosen environment that is constructed using point cloud information obtained from sensor modalities like LiDAR and radar. In this project, we formulate the problem of estimating the occupancy grid map using sensor point cloud data as a sparse binary occupancy value reconstruction problem. We utilize the inherent sparsity of occupancy grid maps commonly encountered in automotive scenarios. Besides, the spatial dependencies between the grid cells are exploited to provide a better reconstruction of the boundaries of the objects inside the range of the map and to suppress the false alarms emerging from the reflections coming from the road. To address sparsity and spatial correlation jointly, we propose an occupancy grid estimation method that is based on pattern-coupled sparse Bayesian learning. The proposed method shows enhanced detection capabilities compared to two benchmark methods, based on qualitative and quantitative performance evaluation with scenes from the automotive datasets nuScenes and RADIal. 

MSc SPS Thesis presentation

• Monday, 31 July 2023
• 10:00-12:00
• HB17.140 (Seminar room)

Machine learning algorithm to estimate cardiac output based on arterial blood pressure measurements.

Cardiac output (CO), a vital hemodynamic parameter that reflects the blood volume pumped by the heart per minute, is crucial for determining tissue oxygen delivery and the heart's ability to meet the body's demands. Researchers developed various methods to measure cardiac output, including thermodilution using pulmonary artery catheters (PAC), also called Swan-Ganz catheters, the gold standard for cardiac output measurements. Such an approach involves an invasive procedure associated with complications, and it requires specialized equipment and expertise, limiting its use to critically ill patients undergoing operations in intensive care units (ICUs). An alternative, less invasive way to estimate CO is by analyzing arterial blood pressure (ABP) waveform. However, the relationship between cardiac output and blood pressure is unknown. This study uses machine learning and feature engineering techniques to discover the relationship between CO and ABP. We used the sparse identification non-linear dynamics (SINDy) algorithm to discover features that significantly contribute to the relationship between CO and ABP. Additionally, we investigated the optimum number of cardiac cycles needed to achieve the best performance providing insights into the temporal dynamics of CO estimation. The proposed approach achieved clinically acceptable performance regarding radial limits of agreement and bias. Further, the proposed approach was validated on an external dataset and achieved comparable performance. Finally, the learned model was interpreted as a differential equation describing the blood flow where CO acts as an external force to the system. All materials used in this study, including code, model, raw data, processed data, and extracted features, are available on GitHub to facilitate further development.

MSc SPS Thesis presentation

• Friday, 21 July 2023
• 10:00-12:30
• Timmanzaal (LB01.170)

Finding Representative Sampling Subsets on Graphs: leveraging submodularity

In this work, we deal with the problem of reconstructing a complete bandlimited graph signal from partially sampled noisy measurements. For a known graph structure, some efficient centralized algorithms are proposed to partition the graph nodes into disjoint subsets such that sampling the graph signal from any subsets leads to a sufficiently accurate reconstruction on average. Furthermore, we consider the situation when the graph is massive, where processing the data centrally is no longer impractical. To overcome this issue, a distributed framework is proposed that allows us to implement centralized algorithms in a parallelized fashion. Finally, we provide numerical simulation results on synthetic and real-world data to show that our proposals outperform state-of-the-art node partitioning techniques.

MSc SPS Thesis presentation

• Monday, 17 July 2023
• 13:00-14:30
• Lipkenszaal (LB01.150)

Privacy Analysis of Decentralized Federated Learning

Privacy concerns in federated learning have attracted considerable attention recently. In centralized networks, it has been observed that even without directly exchanging raw training data, the exchange of other so-called intermediate parameters such as weights/gradients can still potentially reveal private information. However, there has been relatively less research conducted on privacy concerns in decentralized networks.

In this report, we analyze privacy leakage in optimization-based decentralized federated learning, which adopts generally distributed optimization schemes such as ADMM or PDMM in federated learning. By combining local updates with global aggregations, it was proved that optimization-based approaches are more advantageous compared to the traditional average consensus-based approaches, especially in scenarios where the data at the nodes are not independent and identically distributed (non-IID).

We further extend the privacy bound in distributed optimization to the  decentralized learning framework. Different from the fact in the centralized learning framework the leaked information is the local gradients of each individual participant at all rounds, we find that in decentralized cases the leaked information is the difference of the local gradients within a certain time interval.  Motivated by the gradient inversion in centralized networks, we then design a homogeneous attack to iteratively optimize dummy data whose gradient differences are close to the true revealed gradient differences. Though the gradient difference information still brings privacy concerns,  we show that it is  more challenging for adversaries to reconstruct private data using the difference of gradients than using the gradients themselves in the centralized case.

To deal with the privacy attack, we propose several potential defense strategies such as early stopping, inexact update and quantization etc. The main advantage of these approaches is that they introduce error/noise/distortion into decentralized federated learning for protecting private information from being revealed to others without affecting the training accuracy. In addition, we also show that the larger the batchsize is, the more difficult for the adversary to reconstruct the private information.

MSc SPS Thesis presentation

• Monday, 17 July 2023
• 09:00-11:30
• HB17.140 (Seminar room)

Simplicial Unrolling ElasticNet for Edge Flow Signal Reconstruction

The edge flow reconstruction task improves the integrity and accuracy of edge flow data by recovering corrupted or incomplete signals. This can be solved by a regularized optimization problem, and the corresponding regularizers are chosen based on prior knowledge. However, obtaining prior information is challenging in some fields. Thus, we consider exploiting the learning ability of neural networks to acquire prior knowledge. In this thesis, we propose a new optimization problem for the simplicial edge flow reconstruction task, the simplicial ElasticNet, which is a regularized optimization problem that combines the advantages of the L1 and L2 norm. It is solved iteratively by the multi-block ADMM algorithm, and the convergence conditions are illustrated. By unrolling the simplicial ElasticNet's iterative steps, we propose a neural network with high interpretability and low requirement for the number of training data for the reconstruction task of simplicial edge flows. The unrolling network replaces the fixed parameters in the iterative algorithm with the learnable weights in the neural networks, thus exploiting the neural network's learning capability while preserving the iterative algorithm's interpretability. The core component of this unrolling network is simplicial convolutional filters with learnable weights to aggregate information from the edge flow neighbors, thus enhancing the learning and expressive ability of the network. We conduct numerical experiments on real-world and artificial datasets to validate the proposed approach. It is demonstrated that the simplicial unrolling network is significantly more advantageous than the traditional iterative algorithms and standard non-model-based neural networks in the case of limited training data.

MSc SPS Thesis presentation

• Tuesday, 11 July 2023
• 13:00
• HB13.140, Boardroom EEMCS

Estimating Transmembrane Currents and Local Activation Times from Atrial Epicardial Electrograms

Estimating the transmembrane currents travelling through the epicardium and local activation times based on atrial epicardial electrograms can greatly help in the study of cardiac arrhythmias such as atrial fibrillation. This work focuses on the accurate estimation of the aforementioned signals and features. To do this, two least squares-based regression methods were used to estimate transmembrane currents from electrograms and then find their local activation times by searching for the maximum negative slope. The first least squares optimization method consists of using standard least squares, while the second consists of regularized least squares, by combining both lasso and ridge regression, to deal with signal sparsity and multicollinearity, respectively. Furthermore, to improve estimation results, multiresolution analyses based on wavelet decompositions and principal components analysis were used to filter out parasitic components that were present in the estimated transmembrane currents by separating them from the main activation complex of the decomposed signals.

Using these algorithms on simulated data, it was shown that promising results can be achieved for both transmembrane current estimations and LAT estimations. Several wavelet support sizes were tested on the simulated data to observe performance changes. These were compared to an already existing LAT estimation algorithm. The results mainly confirm the efficiency of the proposed methods on severely diseased tissue corrupted by conduction blocks and noise.

MSc SPS Thesis presentation

• Tuesday, 4 July 2023
• 10:30-12:00
• SPS Seminar room, HB17.140

Comparative analysis of clutter filtering techniques on freehand µDoppler ultrasound imaging

Micro-Doppler (µDoppler) ultrasound imaging is a high frame rate ultrasound imaging modality that provides high spatiotemporal resolution ultrasound images of blood flow. It is sensitive to slow blood flow and particularly suitable for capturing fast-changing phenomena like rapid blood flow. Clutter filtering is an essential step in µDoppler data processing to reject tissue clutter signals and keep blood flow information as much as possible. 3D freehand µDoppler imaging is an emerging ultrasound technique that can construct full spatial vasculature images with a panoramic view that conventional 2D ultrasound is not able to provide. As freehand implies the continuous and nonuniform movement of the probe, it becomes more challenging for clutter filtering to acquire high-quality images.
This thesis explores and compares different state-of-art clutter filtering techniques on freehand in-vivo µDoppler imaging of the human brain. Specifically, Singular Value Decomposition (SVD), Robust Principle Component Analysis (Robust PCA), and Independent Component Analysis (ICA) clutter filtering techniques have been investigated. The aim is to test and compare their performance on in-vivo µDoppler ultrasound data with freehand probe movement and understand how freehand motion affects the threshold selection criteria. Besides that, a newly proposed method that combines ICA clutter filtering and clustering is included in this thesis to bring another perspective for sorting independent components corresponding to blood flow and rejecting unwanted ones consisting mostly of tissue clutter signals. 

MSc SS Thesis Presentation

• Thursday, 29 June 2023
• 09:00-09:45
• HB 17.140

Rank Detection Based on Generalized Eigenvalue Threshold in Arbitrary Noise

Rank detection is crucial in array processing applications, as many algorithms rely on accurately estimating the rank of the data matrix to ensure optimal performance. Under Gaussian white noise, rank can be detected through eigenvalue analysis. However, in arbitrary noise, prewhitening the data matrix with the noise covariance matrix is necessary, and rank detection is achieved by examining the generalized eigenvalues. Existing methods often assume the noise covariance structure or require a large number of noise samples. This thesis focuses on addressing the rank detection problem in scenarios with limited noise samples and arbitrary noise environments.

Firstly, we investigate the largest generalized eigenvalue threshold for the prewhitened data sample covariance matrix according to the random matrix theory. We develop a rank detection algorithm based on the threshold via a sequential test, and provide the performance analysis. A series of simulations demonstrate its superiority over conventional methods such as Minimum Description Length (MDL) and Akaike’s Information Criterion (AIC). Secondly, since the Short-time Fourier Transform (STFT) is commonly used for non-stationary signal analysis, we extend our rank detection method to the STFT domain. The correlations introduced by the STFT have a significant impact on the distribution of the noise. Therefore, we develop a technique to remove correlations among time-frequency bins based on exact expressions of these correlations.

After successfully eliminating these correlations, our proposed rank detection method achieves enhanced reliability and performance in the STFT domain. Lastly, we evaluate the effectiveness of our rank detection method in speech enhancement applications. Simulations confirm that utilizing the estimated rank improves speech quality compared to using the known number of sources.

MSc SS Thesis Presentation

• Thursday, 29 June 2023
• 15:30-16:30
• Executive room YesDelft

Self-calibrated plant counting in early crop stand scenarios using deep clustering

In recent years, the agricultural sector has seen significant techno- logical improvements under the flag of precision agriculture, assisting farmers in the manageability that coincides with large-scale farming. Moreover, precision agriculture aims to enable plant-specific farming on the macro scale that is demanded by the current global population growth. By more closely matching the individual needs of the plants, farmers are able to increase crop yield while reducing the environmen- tal footprint as well as the economic cost of farming due to savings in fertilizers and pesticides.

Visual inspection of arable land is a key factor in maintaining trace- ability of plant growth and health in precision farming. More specif- ically, plant counting, size measurement and plant localisation are of great use for farmers in yield prediction, growth tracking, and obtain- ing insight in the emergence ratio of the crop.

Most of the state-of-the-art plant counters, or object counters in gen- eral, rely on human annotations (labour intensive and error prone) as exemplars for the counting model. Self-supervised object counting, however, is a machine learning paradigm independent of human la- belled data, enabling an object counter to learn solely from raw photo- graphic data. Furthermore, the generative character of self-supervised learning models implies the potential to generalize well on unseen data to the model, such as new plant species or variations in plant size in the case of plant counting.

In this master’s thesis in cooperation with Tective Robotics, a study is performed towards the design of self-calibration based self-supervised object counting and localisation model for the scenario of early crop stand scenarios. More specifically, a novel self-calibrator has been developed to estimate the planting distance in between crops and a threshold for small object noise filtering (weeds, loose leaves ect). Im- plications of the self-calibrator are robustness to variations in plant size and allignement, accurate segmentation of occluded plant clusters and small-sized weed suppression.

Model testing on UAV orthomosaic arable land imagery collected by Tective Robotics B.V. has shown outstanding performance (R2 = 0.94) of the newly developed plant counting model, without the need of any labelled training data. The plant counter is comparable in performance to some of the commercially available plant counters. The plant counter, embedded in the entire data processing pipeline for Geotiff orthomosaics, has been made available on GitHub.

MSc SS Thesis Presentation

• Wednesday, 28 June 2023
• 10:00-10:45
• Snijderszaal (Room number LB01.010)

Coherent integration for imaging and detection using active sonar

Existing sonar systems typically rely on a minimum signal strength of a single echo, which limits their performance in low signal-to-noise conditions. This thesis explores the concept of coherent integration for active sonar, with the aim of improving imaging and detection capabilities under low signal-to-noise conditions. The goal is to provide signal processing methods that achieve long-time coherent integration of the received echoes, thereby maximising the processing gain.  Additionally, this research explores waveform design by comparing the performance of pseudo-random noise with chirps.

Two applications are seen in this thesis: moving target detection, which involves static sonar sensors, and synthetic aperture imaging, where the sensors move while the imaging scene remains static. For moving target detection, a processing methods is proposed which achieves coherent integration for constant velocity targets in a computationally efficient manner, and improves the detection performance by implementing a clutter filtering stage. For the second application, a processing method for imaging from a moving sensor pair is proposed. The resulting point-spread function for a circular sensor trajectory is investigated, from which a set of design rules are established. Additionally, a least squares algorithm is applied, which shows that the resulting image can be improved in terms of resolution and sidelobe interference.

Finally, the imaging and detection methods are tested and verified using an in-air demonstrator.

IEEE SPS Webinar

• Wednesday, 26 April 2023
• 17:00-18:00
• webex (see link)

Adaptive and Fast Combined Waveform-Beamforming Design for mmWave Automotive Joint Communication-Radar

Millimeter-wave (mmWave) joint communication-radar (JCR) will enable high data rate communication and high-resolution radar sensing for applications such as autonomous driving. Existing mmWave JCR systems, however, suffer from a limited angular field-of-view and low estimation accuracy for radars due to the use of directional communication beams. In this presentation, we propose an adaptive beamforming design for mmWave JCR with a phased-array architecture that permits a trade-off between communication and radar performances.

To enable fast estimation of the mmWave radar channel in the Doppler-angle domain, we use a convolutional compressed sensing framework and optimize the radar waveforms within this framework. Our optimization accounts for the space-time sampling constraints that are specific to phased-array radars.

We evaluate the JCR performance trade-offs using a normalized mean square error (MSE) metric for radar estimation and a distortion MSE metric for data communication, which is analogous to the distortion metric in the rate-distortion theory. Numerical results demonstrate that our proposed JCR design enables the estimation of short- and medium-range radar channels in the Doppler-angle domain with a low normalized MSE, at the expense of a small degradation in the communication distortion MSE.  

To attend, use the registration link provided below.

Symposium

• Friday, 21 April 2023
• 09:00-18:00
• EEMCS - Lecture Hall Chip

Heterogeneous system integration - Driving the EU Chip Act ambitions

The Netherlands have a strong national ecosystem for quantum, photohics and semiconductor technologies, well connected to international key players. This symposium aims to build on this strength by intensifying collaboration among these domains.

Heterogeneous integration plays a crucial role in enabling future quantum, photonics and semiconductor technologies by creating new functionalities and business opportunities through the integration of different chips, technologies and materials into a single system.

This symposium will discuss the importance of heterogeneous integration and its potential for creating more industry and business value. It also aims to cultivate human resources for heterogeneous integration, further strengthening the Dutch ecosystem.

Join us to explore the exciting opportunities that heterogenous system integration can offer for the Dutch ecosystem and beyond, and to be part of the conversation on driving the EU Chip Act ambitions.

• Thursday, 30 March 2023
• 14:00
• Van der Poelzaal (LB01.220)

Array Processing in Atrial Fibrillation: Application of different signal models and LAT estimation techniques

Thesis link : http://resolver.tudelft.nl/uuid:82dd26ec-d49b-4780-8b14-7ce2960c1b1b

Signal Processing Seminar

• Thursday, 16 March 2023
• 11:00-12:00
• HB 17.140

Robust Pareto-Optimal Radar Receive Filter Design for Noise and Sidelobe Suppression

Signal Processing Systems Seminar

• Tuesday, 28 February 2023
• 11:00
• HB17.140

Community Detection in Multilayer Networks: Algorithms and Applications

Abstract:
Modern data analysis and processing tasks typically involve large sets of structured data, where the structure carries critical information about the nature of the data.  Typically, graphs are used as mathematical tools to describe the structure of such data. Traditional network models employ simple graphs where the nodes are connected to each other by a single, static edge.  However, in many contemporary applications, this relatively simple structure cannot capture the diverse nature of the networks, e.g., multiple types of entities and interactions between them. Multilayer networks (MLNs) allow one to represent the interactions between a pair of nodes through multiple types of links. MLNs can further be categorized based on the homegeneity of the nodes and complexity of topological structure as: i) multiplex networks where each layer has the same set of entities of the same type and inter-layer edges are not shown as they are implicit; ii) heterogenous multilayer networks where the set and types of entities may be different for each layer and the relationships of entities across layers are shown using explicit inter-layer edges. A core task in the complexity reduction of these high-dimensional networks is community detection. In this talk, a joint nonnegative matrix factorization approach is proposed to detect the community structure in both multiplex and multilayer networks. The proposed approach considers the heterogeneity of layers and formulates community detection as a regularized optimization problem. The proposed approach is evaluated for both social networks and a fully connected multi-frequency brain network model.
 

Bio:
Selin Aviyente received her B.S. degree with high honors in Electrical and Electronics engineering from Bogazici University, Istanbul. She received her M.S. and Ph.D. degrees, both in Electrical Engineering: Systems, from the University of Michigan, Ann Arbor. She joined the Department of Electrical and Computer Engineering at Michigan State University in 2002, where she is currently a Professor and Associate Chair for Undergraduate Studies. Her research focuses on statistical and nonstationary signal processing, higher-order data representations and network science with applications to neuronal signals. She has authored more than 150 peer-reviewed journal and conference papers. She is the recipient of a 2005 Withrow Teaching Excellence Award, a 2008 NSF CAREER Award and 2021 Withrow Excellence in Diversity Award. She is currently serving as the chair of IEEE Signal Processing Society Bioimaging and Signal Processing Technical Committee, on the Steering Committees of IEEE SPS Data Science Initiative and IEEE BRAIN. She has served as an Associate Editor and Senior Area Editor for IEEE Transactions on Signal Processing, IEEE Transactions on Signal and Information Processing over Networks, IEEE Open Journal of Signal Processing and Digital Signal Processing.

SPS Seminar

• Friday, 24 February 2023
• 12:30-13:20
• Teams

Understanding the importance of AI quality management

Organizations that want to benefit from AI while managing its risks need to overcome three challenges: adopting AI at scale, complying with regulations, and demonstrating the responsible use of AI. All these challenges can be addressed with appropriate AI quality management. Hence, organizations are looking for talent who understand both: AI technology and quality management. This lecture will introduce you to the core concepts of AI quality management. Awareness is the first step to update your skills and boost your AI careers. As a notified body, TÜV SÜD is a leading several initiatives around AI standardization and regulation. Join and gain an insight view in some of the latest developments and ongoing discussions.

Bio: In his role as CTO Digital Service at TÜV SÜD, Dr. Saerbeck oversees the technology roadmap and key implementation projects of digital testing services, including novel continuous testing services, targeting the demands of a connected smart industry. He has a long track record in academia and industry in the domains of smart sensor networks, robotics, and AI. After completing his PhD with Philips Research, Dr. Saerbeck started an interdisciplinary research team on human-machine interaction within the Institute of High Performance Computing, which developed novel technologies for several industries, including aerospace, manufacturing and retail. He is an awardee of the prestigious A*STAR Independent Investigatorship given by Agency for Science, Technology and Research, Singapore. Dr. Saerbeck has a passion for applied research, promoting translation of academic results in formal verification and artificial intelligence to make today’s connected smart systems safe, secure, and reliable.

PhD Thesis Defence

• Tuesday, 31 January 2023
• 15:00-16:30
• Aula Senaatszaal

Ultrasound Imaging through Aberrating Layers

Whereas aberrating layers are typically viewed as forming an impediment to medical ultrasound imaging, they can surprisingly also be used to our benefit. As long as we can model the effect of an aberrating layer,we can utilize ‘model-based imaging’, the imaging technique explored throughout this thesis, to reconstruct ultrasound images where traditional beamforming methodswould fail, employing the ever increasing computational power available to us nowadays. Not only does this allow us to image through layers, but it also leads to interesting applications, such as 3D ultrasound imaging with spatially undersampled data, using an aberrating ‘coding mask’. The formulation of a measurement model, a fundamental part of model-based imaging, also gives insight into the imaging problem mathematically, and allows us to investigate methods for estimating the effect of an aberrating layer ‘blindly’, i.e., without explicitly measuring it.

In this thesis, we thus investigate (a), imaging through a layer when the layer’s aberration effect is known, and how it can be applied to imaging with spatially undersampled data, and (b), methods and algorithms for estimating the effect of the aberrating layer without knowing it a priori.

In the first part of this thesis, we illustrate how using model-based imaging can be utilized for 3D ultrasound imaging using a single ultrasound transducer, and equipping it with a plastic coding mask. The plastic mask acts as an analog coder, that scrambles the transmitted and received waves in a manner that is location dependent. As a result, the temporal shape of an ultrasound echo can be used instead of the traditional method of using phase differences between sensors in a sensor array. Imaging is instead accomplished using model-based imaging. By measuring the pulse-echo response of each pixel, we can form an image by solving a regularized linear least squares problem, which takes into account the measured pixel-specific pulse-echo signals. The proposed device and imaging method is then verified experimentally.

In the following chapter, a coding mask design method is proposed for the aforementioned imaging device. A measurement model is formulatedwhere themask geometry is an explicit parameter to be optimized. After forming this model, a numerical optimization method is proposed and numerically tested. Our numerical experiments show that optimized mask geometries exhibit an energy focusing effect on the region-of-interest, whilst simultaneously decorrelating echo signals between pixels.

In the second part of this thesis, in contrast, we consider methods for calibrating propagation models when the pulse-echo response per pixel is not known. The most important calibration challenge we consider is that of imaging through an aberrating layer in front of an ultrasound array. This could be subcutaneous fat or the human skull, for example. In this thesiswe formulate ameasurement model consisting of a partwhere wave propagation is known (i.e., the assumed homogeneous region behind the aberrating layer, where the contrast image of interest is located), and an unknown propagation part, consisting of the Green’s functions from an array sensor to any point on the the interface of the aberrating layer and the imaging medium. We then investigate methods for finding this set of Green’s functions without explicitly measuring them (so called ‘blind’ calibration).

The first proposed method exploits the singular value decomposition of the measurement data in combination with the assumed Toeplitz structure of the matrices representing the aberrating layer’s Green’s functions. However, the method is lacking in practicality since an additional set ofmeasurements is required with a phase screen mounted on the interface of the aberration layer and the imaging medium. The second method resolves these practical issues by utilizing a covariance matching technique. A sufficiently large set of measurements is obtained where each measurement is different due to e.g. moving particles such as blood flow or micro-bubbles. Using the covariance of the data, algorithms are then defined that can estimate the transfer functions of the aberrating layer from the measurement covariance data.

Finally, we propose a method for estimating the electro-mechanical impulse response of an ultrasound sensor, by simply measuring its pulse-echo response from a flat plate reflector in front of the sensor. Estimating the one-way (electro-mechanical) impulse response then becomes a de-autoconvolution problem, for which we propose a method by solving a semi-definite relaxation of the de-autoconvolution problem.

Signal Processing Seminar

• Tuesday, 31 January 2023
• 11:00-12:00
• HB 17.140

A framework for augmented reality visualization of simulated and experimental ultrasound images

In a number of clinical applications, such as breast cancer surgery, ultrasound medical images are acquired and analyzed prior to or during the intervention. However, these images are no longer available during the intervention, or are only visible on a computer screen, thus imposing important constraints to practitioners. Moreover, 2D ultrasound images are still the clinical standard, while the tissues of interest are naturally 3D.

The main idea of this work is to construct a framework whose objective is to fill the gap between 2D multimodal acquisitions and real-time visualization in augmented reality of 3D reconstructed volumes, at the time of intervention. In particular, the talk will focus on: i) an experimental platform able to visualize ultrasound images in augmented reality in real time, and ii) an efficient ultrasound image simulation method mimicking freehand scanning of a virtual 3D volume, allowing real-time visualization of the simulated images in augmented reality.

MSc SPS Thesis presentation

• Tuesday, 31 January 2023
• 15:45
• HB17.140

Digital self-timed neuron design for Spiking Neuron Networks

Spiking Neural Networks(SNN) have been widely leveraged by neuromorphic systems due to their ability to closely mimic biological neural behavior, where information is exchanged and received between neurons in the form of sparse events(spikes). Such neuromorphic systems are highly energy-efficient because the use of a global clock can be avoided by asynchronous event-driven operations. Neurons, as the basic processing units of neuromorphic systems, are required to be lowpower and high-speed for the implementation of complex networks. In this work, two fully event-driven digital Integrate-and-Fire(IF) neuron design is presented. Both design exploits the hierarchical structure, which allows the synaptic weights can be accumulated by local compute units in parallel. Instead of using handshake protocols, the proposed design generates on-demand event pulses to drive the weight accumulation, so we call it self-timed. Both neurons are designed by SystemVerilog and synthesized in TSMC 28nm technology. According to the synthesis results, both designs can finish the accumulation of 1024 6-bit weights within 100ns, with a power consumption of 0.055pJ per spike and 0.23pJ per spike respectively.

MSc SPS Thesis presentation

• Monday, 30 January 2023
• 15:45
• HB17.140

On-chip Self Timed SNN Custom Digital Interconnect System

A Spiking neural network (SNN) is a type of artificial neural network which encodes information using spike timing, network structure, and synaptic weights to emulate the information processing function of the human brain. Within an SNN, it is always required to support the spike transmission that travels between neurons(array). This thesis aims to design a customized high-speed interconnect system which supports multi-point communication in a neuromorphic computing system. The burst-mode two-wire protocol in point-to-point communication is applied in this interconnect system, which is designed in high-level modelling with SystemC. In order to improve the utilization of hardware resources, a virtual channel system is involved.

Furthermore, this system could be extended to a variable number of neuron arrays to support different types of spiking neural networks. Also, optimization methods are adopted to increase the transmission rate of the system and save unnecessary energy consumption. The interconnect system could achieve a throughput of 3.802 Gbits/s with the given MNIST use case, based on the evaluation of simulation results.

MSc ME Thesis presentation

• Tuesday, 24 January 2023
• 11:00
• HB13.140, EWI

Physics-Informed Data Augmentation for Human Radar Signatures

In recent years, neural networks (NNs) have seen a surge in popularity due to their ability to model complex patterns and relationships in data. One of the challenges of using NNs is the requirement for large amounts of labelled data to train the model effectively. In many real-world applications such as radar, labelled data may be scarce due to the high cost of acquiring measurements together with privacy and security concerns.
To overcome the lack of data, researchers have resorted to data augmentation (DA), a technique that aims to solve the problem at the root by generating new training samples by leveraging the available ones. In computer vision, image transformations and generative networks are used to perform DA. These techniques, however, may lead to the production of physically unfeasible samples that may hinder the generalization capabilities of classifiers in domains where the data has an underlying physical meaning.
Physics-informed machine learning aims to incorporate physical prior knowledge and governing equations of the target domain into the machine learning pipeline to improve the performance of NNs in fields with limited available data but with well-defined physical models. In this thesis, physics-informed DA in the radar domain is addressed to improve the task of classifying armed and unarmed walking individuals through micro-Doppler spectrograms. 
To begin with, the usage of model-driven micro-Doppler radar simulations to improve the existing generative augmentations is investigated. After introducing several generative NN architectures, the quality and diversity of the produced synthetic spectrograms are evaluated together with their effect on the downstream classification task.
Next, the impact of visual transformations on micro-Doppler spectrograms is studied. Their effect on the underlying physics of the micro-Doppler spectrograms and their impact on feature extraction is assessed. Based on the results, a new DA technique is devised to improve the feature extraction process by informed segmentation of the input spectrogram, halving the size of the NN model and reducing the risk of overfitting.

IEEE Autonomous Systems Initiative Webinar

• Wednesday, 21 December 2022
• 17:00-18:00
• zoom

Active inference in cognitive neuroscience

In the cognitive neurosciences and machine learning, we have formal ways of understanding and characterising perception and decision-making; however, the approaches appear very different: current formulations of perceptual synthesis call on theories like predictive coding and Bayesian brain hypothesis. Conversely, formulations of decision-making and choice behaviour often appeal to reinforcement learning and the Bellman optimality principle. On the one hand, the brain seems to be in the game of optimising beliefs about how its sensations are caused; while, on the other hand, our choices and decisions appear to be governed by value functions and reward. Are these formulations irreconcilable, or is there some underlying imperative that renders perceptual inference and decision-making two sides of the same coin.

Speaker biography

Karl Friston is a theoretical neuroscientist and authority on brain imaging. He invented statistical parametric mapping (SPM), voxel-based morphometry (VBM) and dynamic causal modelling (DCM). These contributions were motivated by schizophrenia research and theoretical studies of value-learning, formulated as the dysconnection hypothesis of schizophrenia. Mathematical contributions include variational Laplacian procedures and generalized filtering for hierarchical Bayesian model inversion. Friston currently works on models of functional integration in the human brain and the principles that underlie neuronal interactions. His main contribution to theoretical neurobiology is a free-energy principle for action and perception (active inference).

Friston received the first Young Investigators Award in Human Brain Mapping (1996) and was elected a Fellow of the Academy of Medical Sciences (1999). In 2000 he was President of the international Organization of Human Brain Mapping. In 2003 he was awarded the Minerva Golden Brain Award and was elected a Fellow of the Royal Society in 2006. In 2008 he received a Medal, College de France and an Honorary Doctorate from the University of York in 2011. He became of Fellow of the Royal Society of Biology in 2012, received the Weldon Memorial prize and Medal in 2013 for contributions to mathematical biology and was elected as a member of EMBO (excellence in the life sciences) in 2014 and the Academia Europaea in (2015). He was the 2016 recipient of the Charles Branch Award for unparalleled breakthroughs in Brain Research and the Glass Brain Award, a lifetime achievement award in the field of human brain mapping. He holds Honorary Doctorates from the University of Zurich and Radboud University.

Zoom link

Please register for the webinar, here.

IEEE SPS BISP-TC Webinar

• Tuesday, 6 December 2022
• 15:00-16:00
• webex

Tensor Decompositions in Functional Neuroimaging

Brain data are inherently large scale, multidimensional, and noisy. Indeed, advances in imaging and sensor technology allow recordings of ever-increasing spatio-temporal resolution. Multidimensional, as time series data are recorded at multiple locations (electrodes, voxels), from multiple subjects, under various conditions. Finally, the data are noisy: the recorded observations are a mixture of ongoing brain activity, physiological, and non-physiological noise sources. Tensors (higher order arrays) are the natural representations of such multidimensional data. Tensor decompositions, in general, aim to write a large and high-order tensor in terms of the product and summation of several smaller and low-rank tensors (including vectors and matrices). A tensor decomposition with a well-chosen number of terms and ranks can approximate the original data tensor using much fewer entries; even to capture the underlying sources separately in its individual components. This talk will first give an introduction to multilinear algebra and tensor decompositions, discuss current challenges in large-scale brain data analysis, and finally highlight some successful applications of tensor decompositions in EEG and functional ultrasound (fUS) data processing.

Registration

Follow this link.

IEEE SPS Autonomous Systems Initiative

• Wednesday, 30 November 2022
• 17:00-18:00
• zoom

Computational Self-awareness and Self-organization: A Paradigm for Building Adaptive, Resilient Computing Platforms

Self-awareness and self-organization have a long history in biology, psychology, medicine, engineering and (more recently) computing. In the past decade this has inspired new self-aware/self-organizing strategies for building resilient computing platforms that can adapt to the (often conflicting) challenges of resiliency, energy, heat, cost, performance, security, etc. in the face of highly dynamic operational behaviors and environmental conditions. I will begin by outlining a computational self-awareness paradigm that enables adaptivity and which supports system resilience. Computational self-awareness is achieved through introspection (i.e., modeling and observing its own internal and external behaviors) combined with both reflexive and reflective adaptations via cross-layer physical and virtual sensing and actuations applied across multiple layers of the hardware/software system stack.

Next I will outline strategies for combining computational self-awareness with self-organization for life-cycle management of dependable distributed computing platforms.   Our ongoing NSF/DFG Information Processing Factory (IPF) project applies principles inspired by factory management that combine self-awareness and self-organization for continuous operation and optimization of highly-integrated-but-distributed embedded computing platforms.  While each IPF computational component exhibits autonomy through self-awareness, collections of IPF entities can self-organize; the resulting emergent behavior must be controlled to ensure guaranteed service even under strict safety and availability requirements. I will outline two use cases: i) End-to-end computational pipelines for a single autonomous IPF component, and ii) Truck platooning as an exemplar for distributed-but-coupled IPF autonomous systems. The talk will conclude with the opportunities and challenges arising from adopting computational self-awareness and self-organization for making complex computational systems more resilient and self-adaptive.

Zoom link: see below.

MSc ME Thesis presentation

• Wednesday, 30 November 2022
• 14:00-16:00
• HB17.140

Self-timed interconnect for SNN - From Point to Point Communication to Multi-array Segmented-bus solution

Spiking Neural Networks use Address Event Representation to communicate among different Neuron Arrays. To mimic the behavior of the human neural system and meets the requirement for large Neuron Array communication, the AER interconnect should be area-saving, have low power, and operates at high speed.

This thesis aims to build self-timed interconnects for point-to point and multi-array communication. The whole system is designed at the RTL level using SystemVerilog. For point-to-point communication, two transmitters are implemented and compared according to their synthesis results. In the multi-array communication structure, we develop a generalized segmented-bus topology and the element fence, to control its segments. Different timing problems in the design are analyzed and corresponding solutions are proposed. The whole system can operate at around 1Gbps in a self-timed manner without any timing problems.

MSc ME Thesis Presentation

• Tuesday, 29 November 2022
• 15:00-15:40
• HB17.140

A New Logarithmic Quantization Technique and Corresponding Processing Element Design for CNN Accelerators

Convolutional Neural Networks (CNN) have become a popular solution for computer vision problems. However, due to the high data volumes and intensive computation involved in CNNs, deploying CNNs on low-power hardware systems is still challenging. The power consumption of CNNs can be prohibitive in the most common implementation platforms: CPUs and GPUs. Therefore, hardware accelerators that can exploit CNN parallelism and methods to reduce the computation burden or memory requirements are still hot research topics. Quantization is one of these methods.

One suitable quantization strategy for low-power deployments is logarithmic quantization.

Logarithmic quantization for Convolutional Neural Networks (CNN): a) fits well typical weights and activation distributions, and b) allows the replacement of the multiplication operation by a shift operation that can be implemented with fewer hardware resources. In this thesis, a new quantization method named Jumping Log Quantization (JLQ) is proposed. The key idea of JLQ is to extend the quantization range, by adding a coefficient parameter ”s” in the power of two exponents (2sx+i ).

This quantization strategy skips some values from the standard logarithmic quantization. In addition, a small hardware-friendly optimization called weight de-zeroing is proposed in this work. Zero-valued weights that cannot be performed by a single shift operation are all replaced with logarithmic weights to reduce hardware resources with little accuracy loss.

To implement the Multiply-And-Accumulate (MAC) operation (needed to compute convolutions) when the weights are JLQ-ed and dezeroed, a new Processing Element (PE) have been developed. This new PE uses a modified barrel shifter that can efficiently avoid the skipped values.

Resource utilization, area, and power consumption of the new PE standing alone and in a systolic array prototype are reported. The results show that JLQ performs better than other state-of-the-art logarithmic quantization methods when the bit width of the operands becomes very small.

MSc ME Thesis Presentation

• Monday, 28 November 2022
• 15:00-15:40
• HB17.140

Off-chip Self Timed SNN Custom Digital Interconnect System

To support the spike propagates between neurons, neuromorphic computing systems always require a high-speed communication link.

Meanwhile, spiking neural networks are event-driven so that the communication links normally exclude the clock signal and related blocks.

This thesis aims to develop a self-timed off-chip interconnect system with ring topology that supports multi-point communication in neuromorphic computing systems. This interconnect system is implemented in high-level modeling with SystemC and involves the burstmode two-wire protocol in point-to-point communication. In order to ensure the flexibility of the system, the distributed control system is involved. Further, the system can be configured with different numbers of chiplet to fulfill various spiking neural network structures.

We also explore optimization methods, which is a bi-directional ring topology achieving the growth of throughput. Based on evaluation and simulation results, the interconnect system can achieve 4.57Gbps with the specific application scenario.

MSc CE Thesis Presentation

• Monday, 28 November 2022
• HB17.140

Mapping of Spiking Neural Network Architecture using VPR

As the new generation of neural networks, Spiking Neural Network architectures executes on specialized Neuromorphic devices. The mapping of Spiking Neural Network architectures affects the power consumption and performance of the system. The target platform of the thesis is a hardware platform with Neuromorphic Arrays with columns for neural signal processing.

The explorations for the mapping methods are based on VPR, an open-source academic CAD tool for FPGA architecture exploration.

The packing of VPR is used for mapping neurons to Neuromorphic Arrays. VPR includes two levels of mapping: pins and neurons.

An evaluation of the mapping methods is established. Based on the evaluation, the optimized mapping solution is generated. Modifications are made in VPR to adapt to SNN architectures. An ActivityCriticality input file is added to the VPR flow for the optimized mapping solution.

MSc ME Thesis Presentation

• Monday, 31 October 2022
• 10:00-10:45
• EWI - Meeting room 15.1

A Low-Noise Transimpedance Amplifier for Ultrasound Imaging with 40dB Continuous-Time Gain Compensation

This work presents a low-noise amplifier (LNA) for miniature 3D ultrasound probes. Time gain compensation (TGC) is required to provide continuously variable gain and compensate for the attenuated echo signal, resulting in decreased output dynamic range (DR). As TGC is embedded in the LNA, a power-hungry LNA is no longer needed to handle the full dynamic range of attenuated echo signal. Compared to prior art where TGC is applied after the LNA, this structure reduce die area and power consumption greatly.

The LNA with built-in TGC functionality is comprised of a transimpedance amplifier (TIA) with exponentially increasing feedback resistive network. Since a transducer with a relatively high impedance is targeted, a TIA is utilized to interface with the tranducer and sense the signal current. TGC is implemented in a continuous fashion by tunable resistors so as to alleviate imaging artifacts associated with gain switching moments. The resistive feedback network is achieved by triode transistors with exponentially decreasing gate voltages. Three parallel branches of triode transistors are varied simultaneously to obtain 40dB gain range. Each branch consists of two back-to-back triodes to mitigate non-linearity related to the body effect.

The variable-gain loop amplifier employing a current-reuse topology enables constant closed-loop bandwidth in an energy-efficient way. The first stage is a fixed-gain stage with dynamic biasing to save power at the lowest gain setting. The next two stages are variable-gain stages with variable resistive loads. The load resistor is implemented in the same fashion as the TIA’s feedback resistor to achieve intrinsic gain matching. The last stage is a buffer to provide low output impedance for stability.

The LNA has been designed in 0.18 μm CMOS technology and occupies an estimated die area of 0.0339 mm2. The effective gain range is 40 dB with ±1 dB gain error. The LNA’s noise floor at the highest gain is below 1.15 pA/rt-Hz and its harmonic distortion is better than -40 dB. During 100 μs receive period, the total power consumption is 6mW from a ±0.9 V supply. The LNA featuring small area and high power efficiency is a promising circuit for miniature 3D ultrasound probes.

MSc ME Thesis presentation

• Friday, 28 October 2022
• 09:30
• LB01.010, Snijderszaal

Autonomous Landing of an Unmanned Aerial Vehicle

Abstract: 

With the advancement in technology, Unmanned Aerial Vehicles (UAVs) have been able to safely maneuver in risky environments. During landing, the UAV should slow down while not affecting its physical design. Currently, multiple sensors are being used to increase the accuracy while landing which might weigh down some of the smaller UAVs. The usage of a single sensor to guarantee safe landing is yet on its early stages of development. It was observed that insects use Optical Flow to maneuver, and this has been used as a method to design UAV's journey and land safely. By dividing each frame, an Optical Flow Difference is calculated which is used as a metric for the UAV to move towards the landing platform.

Once the UAV has positioned itself on top of the elevated landing platform, image dilation method is used to safely land the UAV. One of the methods tested used Image Dilation Method using IMU. Using the calculated image dilation, the control input to the UAV is generated and the UAV lands. This method failed in providing safe landing and it also did not take the vision of the UAV into consideration. Then, Image Dilation Method using Features from Accelerated Segment Test (IDMF-AST) was tested. This method tracks features observed by the UAV and calculated an estimate of the image dilation. The estimate of image landing is used to control the UAV. This showed dependency on the landing design platform and the hyperparameters that are used for implementing IDMF for the type of landing.

Different landing designs were tested for different elevated landing platforms. It was observed that concentric circles on a textured landing marker provided the highest probability of safe landing. To tackle the dependency of the hyperparameters, a Classification Model was proposed to find an optimal set of hyperparameters for each possible height of the platform. This trained model was incorporated with the IDMF, thereby designing the proposed Adaptive IDMF algorithm. The Adaptive IDMF was tested against the original IDMF on the metrics of safe landing probability, time taken to safely land and simulation time. Adaptive IDMF performed better compared to IDMF providing an 190% increase in probability of safe landing, faster safe landing and lesser simulation and compilation time.

Microelectronics Colloquium

• Thursday, 27 October 2022
• 15:30-16:30
• LB 1.010 Snijderszaal

Advances in Low-Field MRI Hardware Design and Data Processing

In this talk we discuss several recent advances in low-field Magnetic Resonance Imaging (MRI). We focus on magnet and gradient coil design for a low-field MR scanner in which the strong background field is generated by permanent magnets (Halbach systems). These design problems are treated as inverse source problems, which are severely ill-posed in general. How to obtain approximate (regularized) solutions to these problems is discussed and the practical implementation of these solutions is addressed as well. Several processing algorithms that can handle compressed noisy MR input data are also presented and we illustrate the performance of these algorithms on simulated and measured low-field MR data.

MSc ME Thesis presentation

• Friday, 21 October 2022
• 13:30-16:30
• HB17.150/140 - CAS Seminar room

Energy-efficient seizure detection for wearable EEG

MSc ME Thesis presentation

• Monday, 17 October 2022
• 09:30-11:30
• HB17.140/150

A trainable activation scheme for applying Neural Radiance Fields in novel view synthesis tasks

3D scene reconstruction is a common computer vision task with many applications. The synthesized virtual environments are beneficial for many downstream applications such as 3D modeling, building inspection, virtual reality, etc. As conventional scene reconstruction methods often require expensive data collections and prior information on the geometry, a learning-based method named Neural Radiance Fields (NeRF) has gained a surge of interest within the computer vision community recently for its capacity to achieve state-of-the-art view synthesis performance and achieve photorealistic rendering from only a sequence of RGB images as the input. However, NeRF has a strict requirement for input images with accurate camera poses, which is often not available in real-life applications. To this end, we provide an end-to-end guideline for 3D scene reconstruction using NeRF under a real-world scenario the objective is to reconstruct the HDB facade in Singapore. This guideline requires only RGB images as the input and can achieve photorealistic rendering results which are competitive with the results from the conventional point cloud-based method. Besides, we build our own models upon NeRF and also improve its performance in representing fine details. We first research that the ability to represent high-frequency contents in the signal is limited by its ReLU activations in the Multi-layer Perceptron (MLP) network, and demonstrate that mapping the input to the MLPs from low dimensional space to high dimensional space significantly improves the reconstruction and view synthesis quality. Afterward, We make several attempts to perform input mapping. We first use Gaussian-distributed Fourier features to replace the original positional encoding used in NeRF. Then, we research the activations in coordinate MLP and propose an embedding-less NeRF model equipped with parameterized sine activations called SIRENeRF. Next, we extend the use of parameterized activations from sine activations to a class of non-periodic activations and propose a trainable activation scheme that not only achieves higher scene reconstruction results but also enjoys better flexibility to different datasets. Experiments show that all of the above attempts outperform the positional encoding in terms of scene reconstruction and view synthesis results.

MSc ME Thesis presentation

• Friday, 14 October 2022
• 14:30
• Online

Nowcasting of extreme precipitation using deep generative model

Extreme precipitation can often cause serious hazards such as flooding and landslide. Both pose a threat to human lives and lead to substantial economic loss. It is crucial to develop a reliable weather forecasting system that can predict such extreme events to mitigate the effect of heavy precipitation and increase resilience to these hazards.

Numerical Weather Prediction (NWP) models play the dominant role in the field of weather forecasting. However, due to their long computational time, these models had limited utility in predicting weather conditions in the following several hours. This gap is filled by nowcasting, an observation-based method that uses the current state of the atmosphere to forecast future weather conditions for several hours. Operational nowcasting systems typically apply extrapolation algorithms to rainfall radar observations based on simple physics assumptions. However, the physics constraints also limit the performance, and the methods can hardly capture non-linear patterns in the radar observations. Besides the conventional methods, deep learning models have started to play an essential role in this field. Recent works have shown the promising potential of using deep learning models to tackle the nowcasting task, which is also this thesis's focus.

The thesis work mainly studied in two directions: the development of novel deep generative models for precipitation nowcasting and the application of statistical approaches for better modeling and prediction of extreme events. For the first direction, our proposed model is inspired by recently developed deep learning models from the field of visual synthesis. The model makes use of a two-stage structure: the first stage is a Vector Quantization Variational Autoencoder (VQ-VAE) which compresses the original high-resolution radar observations into a low-dimensional latent space. The second stage works in this latent space. It contains an autoregressive Transformer that models the probabilistic distribution of latent space data. The trained Transformer can predict the latent space representation of future frames. ForT better modeling and prediction of extreme events, Extreme Value Loss (EVL) is proposed and incorporated with the autoregressive Transformer. The loss function aims at penalizing predicting extreme cases as non-extreme and predicting non-extreme cases as extreme in order to solve the high imbalance between extreme and normal precipitation data. Our results show that the proposed model shows comparable performance with the state-of-the-art conventional method and other deep learning nowcasting models. The proposed EVL has also been shown to improve the overall performance and accuracy in predicting extreme events.

MSc ME Thesis presentation

• Friday, 14 October 2022
• 10:45-12:46
• Hall Chip, HB01.600

IMU-Based Adaptive Filtering for Movement Artifact Removal from ECG Recorded With a Single Lead Wearable Device

Background and Objectives: Wearable devices (WDs) capable of recording electrocardiograms (ECGs) for prolonged periods in ambulatory settings offer the possibility of detecting non-predictable events such as epileptic seizures and atrial fibrillation. Nevertheless, these systems suffer from additional noise sources, such as movement artifacts (MA). 

Several adaptive (AF) algorithms have been proposed in the literature to suppress movement artifacts from ECG without consistent results. Adaptive algorithms for signal enhancement require a reference signal correlated with the noise and not correlated with the sign of interest. This correlation can change significantly depending on the absolute and relative location of the electrodes and the location of the reference sensor (i.e., accelerometer, gyroscope); objectively measuring the correlation is not a trivial problem.

For this reason, first, we used an algorithm to obtain a rough estimate of the movement artifacts from the recorded ECG to calculate the correlation between them and the available reference signals (three-axis accelerometer and three-axis gyroscope) and selected the one with the highest correlation. Then, we compare three adaptive filter algorithms using the signal-to-noise ratio (SNR) coefficient as the evaluation parameter.     

Methods: For testing the implemented adaptive filters, first, we used a simulated signal, then data from an open-online database, and last, a single lead ECG wearable device called AFi^® (Praxa Sense™, The Netherlands) with an embedded IMU. To induce the movement artifacts in a controlled setting, participants performed a set of predefined movements within three intensities; high (running, jumping), moderate (torso rotations, pushups), and low (walking). Then we analyzed the recorded data offline as follows:

1. Test the correlation between noise and the IMU components, and select the component with the highest correlation to be used as a reference input for the adaptive filters.
2. Compare three adaptive filters in terms of SNR improvement; the Least means squares (LMS), the Normalized least means squares (NLMS), and the Recursive least squares RLS.
3. Filter the selected reference input with wavelet decomposition, and test if there is a filter performance improvement in SNR.

Results: The implemented adaptive filters performed as expected with the simulated signals, but they showed very poor results once we used them on real data.

The RLS filter showed superior performance than the least mean squares-based filters in terms of convergence speed and the root mean squared error minimization. Nevertheless, it requires a high correlation (ρ) above ρ>0.8 between the reference input and the undesired signal or noise to provide a proper signal enhancement and morphology recovery.

The low correlation between the movement artifacts and the components of the IMU used as a reference input for the adaptive filters affected the filter performance heavily. Filtering the reference input with the wavelet decomposition did not improve the correlation or the filter performance.

Conclusions The correlation between ECG motion artifacts and movement recorded with inertial sensors appeared to be low and inconsistent. Given this, adaptive filters using inertial sensors as reference input are unsuitable for removing ECG movement artifacts. 

MSc ME Thesis presentation

• Monday, 10 October 2022
• 09:30-10:31
• LB 1.150 Lipkenszaal

Predicting noise attenuation level in the earplugs using Gaussian Process Regression

The earplug development followed by ALPINE, a hearing protection company is a trial and error method. This method leads to high material wastage. It is also a time-consuming and expensive development process. So, the objective of this thesis is to ease the error earplug development process by implementing a prediction model. The research was steered towards understanding the factors that influence the sound dampening properties in an earplug, building the dataset, and analysing the data and regression models. The regression model must be able to predict the noise attenuation provided by an earplug based on the material and design specifications for which Gaussian Process Regression (GPR) model was used. With RBF kernel function and through hyperparameter optimization, the GPR model was trained and tested on datasets.

The main finding of this thesis is that noise attenuation provided by an earplug is highly subjective. Sound perception is crucial in earplug design. Even though the most ideal sound attenuation prediction model for earplugs can be developed, in the end, everything relies on individual sound perception. So, the prediction of noise attenuation in earplugs has a higher level of uncertainty. However, with accurate age, gender, and ethnicity information, the earplugs can be modelled and designed for each group and the sound attenuation prediction model can be developed to make predictions with higher certainty.

MSc ME Thesis presentation

• Friday, 30 September 2022
• 10:00-11:00
• Timmanzaal (LB01.170)

Detecting Medical Equipment in the Catheterization Laboratory using Computer Vision

Workflow analysis aims to improve the efficiency and safety in operating rooms by analyzing surgical processes and providing feedback or support, where observations can be made and evaluated by algorithms rather than human experts. For our study, we mount five calibrated cameras from different angles in a Catheterization Laboratory (Cath Lab) to observe and analyze Cardiac Angiogram procedures. To automate the classification of workflow and personnel activities, we propose an object detection algorithm based on Scaled-YOLOv4 with a filter to improve bounding box prediction. Scaled-YOLOv4, as a state-of-the-art technique, is featured with extremely fast processing speed and decent precision. However, we find that Scaled-YOLOv4 still suffers when detecting objects with flexible appearance due to fixed anchors. This can result in the prediction of bounding boxes missing parts of the object or containing a blank environment. In this work, we design a filter following Scaled-YOLOv4 to improve the prediction of the bounding box by matching the features detected from different cameras. With the keypoints detected by SuperPoint and matched by SuperGLue, the filter adjusts the boundaries of the bounding box to include all the matched keypoints. The proposed algorithm achieves 95.1% mAP in detecting medical equipment in the Catheterization Laboratory and a real-time speed of 58 FPS on RTX 3090.

MSc ME Thesis presentation

• Tuesday, 27 September 2022
• 10:00-11:00
• LB01.190

Reconstruction and Rendering of Buildings as Radiance Fields for View Synthesis

In inspection and display scenarios, reconstructing and rendering the entire surface of a building is a critical step in presenting the overall condition of the building. In building reconstruction, most works are based on point clouds because of their enhanced availability. In recent years, neural radiance fields (NeRF) have become a common function for implementing novel view synthesis. Compared to other traditional 3D graphic methods, NeRF-based models have a solid ability to produce photorealistic images with rich details that point clouds based methods cannot offer. As a result, we decided to investigate the performance of this technique in architectural scenes and look for ways to improve it for more significant scenes.

This thesis explores the ability to reconstruct large-field scenes with NeRF-based models. NeRF introduced a fully-connected network to predict the volume density and view-dependent emitted radiance at the special location, which will be projected into an image through classic volume rendering techniques. Due to the limitation of near-field ambiguity and parameterization of unbounded scenes, the original NeRF does not perform well on 360° input view, especially when the inputs are sparse. An inverted sphere parameterization that facilitates free view synthesis is introduced to address this limitation so that the foreground and background views can be trained separately. Besides that, we also compare the performance of tracing different light geometries, ray and cone, respectively. Meanwhile, to generate the reconstructed scene precisely, raw RGB images should be pre-processed to estimate the corresponding camera parameters. Finally, customized camera paths should be prepared to generate the final rendered video.

According to our experiments, training foreground and background separately is a promising method to solve practical large-scale scene reconstruction problems. A complete wrap-around view of the target building can be obtained using adjusted camera path parameters. Furthermore, introducing conical frustum casting into the original model also provides an alternative method to implement reconstruction. We named this method mip-NeRF++, which can contribute to the final results to some extent.

MSc ME Thesis presentation

• Monday, 26 September 2022
• 09:00-11:00
• Online via Teams

Automatic Camera Extrinsics Estimation in the Catherization Laboratory

Surgical workflow analysis has gained more importance in operating rooms, which could take responsibility for the working condition, the safety of both patients and surgical personnel, as well as the working efficiency. Focusing on the optimization of the workflow, a set of cameras is installed in the Catherization Laboratory in Reinier de Graaf Gasthuis for multiple computer vision related researches. However, our cameras are calibrated only once after installation. The orientation and position of the cameras could be changed after days or months, which could lead to a wrong localization. 

Compared with the traditional calibration method (calibration patterns or markers), we propose a new image-based camera pose estimation pipeline tested in the Catheterization Laboratory. Our proposed pipeline exploits object detection model (Scaled-YOLOv4) to detect fixed objects. The mean average precision with 50% IoU threshold (mAP@.5) achieves more than 0.99 for all detected objects. Then use a self-supervised interest point detector and descriptor (SuperPoint). With the detected feature points, a feature matching technique based on graph neural networks (SuperGlue) is adopted to match the interest points detected in the target image with reference points annotated in the image databases (image DBs). The point-correspondences between the image coordinates and the 3D coordinates are applied to solve Perspective-n-Point (PnP) problem to compute the orientation and position of each camera (Camera pose). The final camera pose estimation achieves a 5.79-pixel reprojection error with a 4.97-cm Euclidean distance error. Compared with other image-based camera pose estimation techniques, our pipeline requires no 3D reconstruction or 3D point cloud in the scene model. Using the video from real procedures, we show that the pipeline is able to estimate the camera pose with high accuracy.

Microelectronics Colloquium

• Thursday, 22 September 2022
• 15:30-16:30
• TBD

Sparsity-constrained Linear Dynamical Systems

Abstract: At the intersection of control engineering and signal processing sits the upcoming field of sparse control and state estimation of linear dynamical systems. It deals with linear dynamical systems with control inputs having a few nonzero entries compared to their dimensions. Constraining the inputs to be sparse is often necessary to select a small subset of the available sensors or actuators at each time instant due to energy, bandwidth, or physical network constraints. Bringing together research from classical control theory and compressed sensing, the talk presents a comprehensive overview and critical insights into the conceptual foundations of sparsity-constrained systems, including the formulation, theory, and algorithms. We look at the concrete example of a budget-constrained external agent controlling the opinion of a social network.

MSc ME Thesis presentation

• Thursday, 22 September 2022
• 13:30-16:00
• van Katwijkzaal (HB08.150)

Frequency Domain Joint Estimation of HRF and Stimulus from fUS Data

To better understand how brain signals are processed and even how the human mind works, analyzing the hemodynamic signal model is one of the most essential steps. In the CUBE group of Erasmus MC, functional ultrasound (fUS) data of a mouse’s brain is recorded. By using this fUS dataset, this thesis will solve the problem regarding the joint estimation of hemodynamic response function (HRF) and the underlying stimulus. Usually, hemodynamic responses are investigated in the time domain, while this thesis provides another perspective

from frequency domain signal processing.

We consider the hemodynamic response as a convolutive signal mixture, then try to transform it into an instantaneous mixing model by converting the context into the frequency domain. By applying independent vector analysis (IVA), this estimation problem can be solved without facing permutation ambiguity which is a well-unknown problem regarding independent component analysis (ICA). Additional steps before and after IVA are also discussed so that a whole estimation road map is formed.

Both simulation and experimental analysis are provided to validate this estimation algorithm. Results show that by using this method, both stimulus and HRF estimation can be achieved satisfyingly in a suitable experimental setting. This thesis provides insights and future potentials for IVA to be further investigated in neural signal processing problems.

PhD Thesis Defence

• Thursday, 15 September 2022
• 12:00-13:30
• Aula Senaatszaal

Advanced Measurement Techniques and Circuits for Array-Based Transit-Time Ultrasonic Flow Meters

This thesis describes the design, prototyping and evaluation of matrix-based clamp-on ultrasonic flow meters. Several new measurement techniques are presented as well as an Application-Specific Integrated Circuit (ASIC) designed for accurate measurement of flow velocity with matrix transducers.

The influence of circuit topologies on the zero-flow performance of ultrasonic flow meters has been analyzed and an algorithm is presented to reduce the offset. With a linear transducer array, flow measurements have been performed via two different acoustic paths, demonstrating the ability to accurately measure flow with array transducers through a stainless-steel pipe wall. In order to improve signal quality, an ASIC has been designed that is able to drive and read-out 96 piezo transducer elements. The ASIC has been characterized electrically and flow measurements have been performed in combination with the linear transducer arrays.

Several new techniques, enabled using transducer arrays, have also been explored. By tapering the amplitude of the transmit signals, spurious waves can be suppressed. An auto-calibration technique has been developed that uses additional acoustic measurements to estimate the diameter of the pipe and the speed of sound in the pipe wall and liquid. Finally, a simulation study has been performed to explore the possibility of exploiting the beam-steering capabilities of transducer arrays to measure flow velocity profiles by using measurements obtained via multiple acoustic paths.

Thesis:
https://doi.org/10.4233/uuid:e2f4b411-3d8e-4b93-b037-096009c59f61

Collegerama (live stream of the defence):
https://collegerama.tudelft.nl/mediasite/play/571eb84f3e724d47b46df7e8d0eb3a7a1d

PhD Thesis Defence

• Tuesday, 13 September 2022
• 12:00-13:30
• Aula Senaatszaal

Integrated Transceiver Circuits for Catheter-based Ultrasound Probes and Wearable Ultrasound Patches

Promotors: Michiel Pertijs and Ronald Dekker

Thesis: link

Collegerama link (live stream of the defence): link

Abstract: This thesis describes the design, prototyping, and experimental evaluation of transceiver ASICs (application-specific integrated circuits) for catheter-based ultrasound probes and wearable ultrasound patches. Various circuit techniques are proposed to address requirements and implementation bottlenecks in these applications. Prototype chips are presented to demonstrate the effectiveness of these techniques. To reduce the loading effect of micro-coaxial cables in an ICE probe based on capacitive micro-machined ultrasound transducers (CMUTs), an ASIC prototype including element-level high-voltage pulses and low-noise trans-impedance amplifiers has been implemented. Besides reducing the loading effect from micro-coaxial cables, ASICs play an important role in achieving cable-count reduction, which is crucial for 3-D imaging catheters, such as forward-looking IVUS probes. Circuit techniques are proposed to implement a prototype ASIC which only requires 4 cables to interface with a 2D piezoelectric transducer array. Additionally, to address the challenges in interface electronics for wearable ultrasound patches, a prototype ASIC is presented that contains 64 reconfigurable transceiver channels that can interface with different transducer elements by employing channel-parallelizing techniques.

Special EI Colloquium

• Tuesday, 13 September 2022
• 10:00-11:00
• EWI - Lecture Hall F

Real-time High-Frame Rate imaging: Novel Methods and Applications

Prof. Piero Tortoli
Microelectronics Systems Design Laboratory
University of Florence, Italy

Medical imaging is increasingly based on High-Frame-Rate (HFR) methods, which are in principle capable of producing one frame (or even one data volume) per transmission event. However, achieving such a goal in real-time implicitly involves the transfer and processing of huge amount of data at high rates, and this can be done only through an appropriate experimental setup.

In this talk, the main characteristics of the hardware-based open scanner ULA-OP 256 are briefly reviewed, and its recent advancements, such as the data transfer acceleration obtained through an architectural change, and the possible expansion toward the control of an unlimited number of probe elements, are reported in detail. The “virtual real-time” modality will also be described as ideal to obtain the best performance from specific HFR imaging modalities. Finally, the combination of ULA-OP 256 with properly designed sparse 2-D arrays will be shown suitable for the investigation of full volumes. The talk will be concluded with the presentation of experimental results in a few sample applications, including multi-plane imaging, HFR CFM and HFR vector Doppler.

Micro- and Nanosystems at ESAT, KU Leuven

Prof. Michael Kraft
ESAT, Micro- and Nano-Systems
KU Leuven, Belgium

This seminar will give an brief overview of the activities in micro- and nanosystems at the Electrical Engineering Department (ESAT) of KU Leuven. It will describe the available infrastructure and give a short overview of current research activities in the division Micro- and Nanosystems (MNS), which currently comprises 24 PhD students, 4 postdoctoral researchers and 2 technicians.

A selection of current active projects and recent highlights will be presented, including work on:

• Coupled resonators for mass sensing applications
• Piezoelectric ultrasound technology arrays for medical imaging and underwater communication
• Micromachined probes for neuro recording and stimulation
• Multi-parameter sensing chip for bioreactor condition monitoring
• Genetic Algorithm for the design of MEMS devices (accelerometers and microgrippers)

Finally, the newly founded Leuven Institute for Micro- and Nano Integration (LIMNI) will be briefly introduced.

Note: This Colloquium precedes the PhD defence of Mingliang Tan, which will take place in the Aula on the same day at 12:00 (layman’s talk), 12:30-13:30 (defence). More information can be found here.

MSc SS Thesis Presentation

• Monday, 29 August 2022
• 09:30-10:15
• Lecture Hall F

Improving the Estimation of Epicardial Activation Times Using Spatial Information

Atrial fibrillation is a common cardiovascular disease, affecting the regular beating of the heart through chaotic contraction of the heart's upper chambers. On its own, the condition—increasingly prevalent among the elderly—is not life threatening, but it leads to an increased risk of stroke and heart failure. As of yet, there is no consensus on the physiological mechanisms responsible for initiating and sustaining atrial fibrillation. A more detailed view of cardiac activity would improve understanding of the disease, making earlier diagnosis possible and improving options for treatment. 

The contraction of the cardiac muscles is governed by electrical signals propagating through the tissue. Cardiac activity can be monitored with a high spatial resolution by measuring the electrical potential directly on the epicardium of the heart during open-heart surgery, using an array of closely spaced electrodes. From these electrograms, estimating the time of local activation of the cardiac tissue underneath each electrode provides a quantitative way of mapping the mechanisms of atrial fibrillation. Various methods exist to estimate the activation times, but the complex signals that are typical of atrial fibrillation make it difficult to obtain accurate results. This thesis proposes combining two existing methods for estimating the local activation times. Based on a model of the electrogram as a spatial convolution of local transmembrane currents, an inverse problem is formulated and solved, resulting in a less opaque view of the cardiac activity at the electrode locations by attenuating distant disturbances and emphasizing local activity. The deconvolution output is fed to the second step, where cross-correlating certain pairs of signals gives an estimate for the mutual time delay in local activation. A graph representation of the electrode array is used to define neighbor order and decide which signal pairs are correlated. The set of pairwise time delays this produces is then converted to an estimate for the local activation times, using a least-squares estimator. 

The proposed method is evaluated using different simulated cardiac settings. In a setting with one stimulation source, earlier results of the deconvolution and cross-correlation methods are confirmed, and the proposed method is seen to produce a slightly lower mean error than reference methods. In the higher-complexity triple-source setting, the latter effect is again visible. Reinforced by the performance of the different methods in increasingly noisy settings, the main merits of the proposed method for the estimation of local activation times can be said to be found in the form of increased consistency, not significantly improving on the accuracy of existing methods.

MSc SS Thesis Presentation

• Friday, 26 August 2022
• 10:00-10:45
• Van Katwijkzaal (HB 08.150)

Cooperative Localization of Unmanned Aerial Vehicles using ADS-B

As unmanned aerial systems (UAS) turn into a full-fledged industry, the sky will be much more crowded in the future. Large-scale UAV applications make reliable UAV navigation a pressing need. Traditionally, global navigation satellite system (GNSS) is extensively used as the primary positioning, navigation, and timing (PNT) service. However, GNSS is vulnerable to intentional radio interference such as spoofing, jamming, and repeating. Hence, alternative PNT (APNT) attracted many researchers' attention. 

In this thesis, instead of GNSS signals, ADS-B signals from piloted aircraft are leveraged for UAV navigation. We propose a cooperative navigation strategy for multiple UAVs in GNSS-denied environments. It consists of: 1) a system-level, leader-follower cooperative strategy; 2) a sensor fusion algorithm for individual UAV navigation based on the extended Kalman filter. Furthermore, the effects of asynchronous clocks are studied and a joint relative positioning and synchronization algorithm is applied to tackle this problem. 

Finally, Monte Carlo experiments in a multi-UAV scene are performed to verify the proposed algorithms. The results show that the proposed algorithms achieve a performance comparable to civilian GNSS on the selected data set and under the system assumptions we made. Moreover, the proposed cooperative navigation framework only needs one ground station of limited service capacity as external aid. Compared with large-scale, specialized terrestrial APNT service networks, our proposed framework is more flexible and the system can be deployed in areas without infrastructure.

MSc SS Thesis Presentation

• Thursday, 25 August 2022
• 13:00-13:45
• Room G

Robust Formation Control against Observation Losses

Distributed formation control has received increasing attention in multiagent systems. Maintaining certain geometry in space is advantageous in many applications such as space interferometry and underwater sensing. At present, there is a variety of distributed solutions for agents to converge to desired formations and track a series of prescribed maneuvers. They typically rely on the relative kinematics e.g., relative positions of the neighboring agents as state observations for the local controller. In harsh working environments, the acquisition of the relative kinematics is challenged and observation losses might occur, which can be detrimental to the optimality of formation.   

In this work, observation losses in noisy environments are addressed under a distributed formation control framework. Three types of solutions are proposed to enhance the robustness which is evaluated through the improvements of tracking error, convergence speed, and smoothness of trajectories in both random and permanent loss settings. Firstly, a relative localization technique is proposed using formation itself as a spatial constraint. Secondly, a dynamic model is established for the agents entailed by a Kalman filter-based solution. Finally, a fusion of the previous two types is inspired and it exhibits superior performance than both aforementioned types individually. 

  This work not only provides means of relative localization without additional sensor data but also shares insights into coping with random or permanent graph changes for stress-based formation control systems. This could potentially lead to the exploration of formation control with subgraphs or energy-efficient sensing as future directions.

MSc SS Thesis Presentation

• Thursday, 25 August 2022
• 10:00-10:45
• LB01.040

Distributed Gaussian Process for Multi-agent Systems

This work is focused on environmental monitoring and learning of unknown field by Gaussian Process (GP) in Multi-agent Systems (MAS). The two main problems are how to develop fully-distributed and robust algorithm to (1). optimize GP hyperparameters, and (2). aggregate GP predictions from agents.  

The state-of-the-art distributed GP hyperparameter optimization algorithm is proximal alternated direction method of multipliers (pxADMM), which requires a center station in MAS. Based on pxADMM, two fully-distributed algorithms are proposed so that the center station is no longer needed. Asynchronous behavior is also introduced into the proposed algorithms to deal with heterogeneous processing time of agents.

  Current aggregation methods are classified based on whether datasets are independent. Under independence assumption, PoE and BCM families of methods are distributed by applying discrete time consensus filter (DTCF), which is proposed to be replaced by primal-dual method of multiplier (PDMM) for faster convergence. Without independence assumption, the Nested Pointwise Aggregation of Experts (NPAE) can be distributed by NPAE-JOR in complete graph with high flooding overhead. We propose fully-distributed CON-NPAE in connected graph to eliminate flooding overhead. 

Simulation results show that the proposed hyperparameter optimization algorithms are fully-distributed at a cost of 2 to 4.5 times more iterations compared to pxADMM. The fully-dsitributed PoE and BCM based methods are accelerated, and the fully-distributed CON-NPAE makes comparable aggregations as NPAE without flooding overhead. Future work will be focused on the theoretical convergence analysis of fully-distributed pxADMM, the effect of network structure on CON-NPAE and new type of distributed NPAE based on inducing points. 

MSc SS Thesis Presentation

• Wednesday, 24 August 2022
• 12:00-12:45
• Lecture Hall Ampere

Detect and Avoid for Autonomous Agents in Cluttered Environments

Autonomous agents are the future of many services and industries such as delivery systems, surveillance and monitoring, and search and rescue missions. An important aspect in an autonomous agent is the navigation system it uses to traverse the environment. Not much emphasis has been paid in the past on autonomous agent navigation in cluttered environments. Cluttered and unknown environments such as forests and subaquatic environments need to have autonomous navigation systems developed just for them due to their uncertain and changing nature.

  Path planning algorithms are used for the navigation of an autonomous agent in an environment. The agent needs to reach a target location while avoiding the obstacles it detects along the path. Such a system is called a Detect and Avoid (DAA) system and there are different implementations for it of which some are explored in this thesis.  

The Artificial Potential Fields method or APF for short is a method for mobile agent navigation which is based on generating an attractive force on the agent from the target and a repulsive force from the obstacles. This leads to the agent reaching the target while avoiding the obstacles along the way. The Classical APF (CAPF) method works for structured environments well but not for cluttered environments. The CAPF method can be replaced with a modified version where the agent is surrounded by a set of points (called bacteria points) around its current location and the agent moves by selecting a bacteria point as a future location. This method is named the Bacteria APF (BAPF) method. This selection happens through combinatorial optimization based on the potential value of each bacteria point.  

In this thesis, we propose two distinct contributions to the BAPF method. The first one being the use of an adaptive parameter in the repulsive cost function which is determined through a brute-force search. The second addition is a branching cost function that changes the value of the repulsive potential based on predefined perimeters around each obstacle. We show through simulations on densely and lightly cluttered environments that this Improved BAPF (IBAPF) method significantly improves the performance of the system in terms of the convergence to the target by almost 200% and reduced the time it takes to converge by around 25% as well as maintain the safety of the navigation route by keeping the average distance from obstacles around the same value.

MSc SS Thesis Presentation

• Monday, 22 August 2022
• 14:00-15:00
• EEMCS Lecture room F

Search by Image: Deep Learning Based Image Visual Feature Extraction

In recent years, the expansion of the Internet has brought an explosion of visual information, including social media, medical photographs, and digital history. This massive amount of visual content generation and sharing presents new challenges, especially when searching for similar information in databases —— Content-Based Image Retrieval (CBIR). Feature extraction is the foundation of image retrieval, making research into obtaining concrete features and representations of image content a vital concern. In the feature extraction module, We first pre-process the target image and input it into a CNN to obtain feature maps for different channels. These feature maps can be aggregated into compact and global uniform descriptors by pooling. Then these global descriptors are further dimensionalised and normalized by whitening methods to obtain image feature vectors that are easy to compute and compare. In this process, the accuracy of the retrieval depends on how accurately the final feature vectors represent the meaning expressed by the target image. Therefore, various CNN network structures, pooling and whitening methods are proposed to get more concrete feature vectors.

In this thesis, our study (1) fine tunes the pre-trained CNNs, (2) optimizes the application of second-order attention information in feature map, (3) applies and compares popular feature enhancement methods in both aggregating and whitening, (4) explores how to combine all strengths, and (5) propose a new model \textit{ResNet-SOI}, which achieves 53.4(M) and 59.2(M) mAP on the challenging benchmark \textit{ROxford5k+1M} and\textit{ RParis6k+1M}, and outperforms the state-of-art methods.

MSc SS Thesis Presentation

• Friday, 22 July 2022
• 10:00-10:45
• LB 01.010 Snijderszaal; Teams

Image-Based Query Search Engine via Deep Learning

Typically, people search images by text: users enter keywords and a search engine returns relevant results. However, this pattern has limitations. An obvious drawback is that when searching in one language, users may miss results labelled in other languages. Moreover, sometimes people know little about the object in the image and thus would not know what keywords could be used to search for more information. Driven by this use case with many applications, content-based image retrieval (CBIR) has recently been put under the spotlight, which aims to retrieve similar images in the database solely by the content of the query image without relying on textual information.

To achieve this objective, an essential part is that the search engine should be able to interpret images at a higher level instead of treating them simply as arrays of pixel values. In practice, this is done by extracting distinguishable features. Many effective algorithms have been proposed, from traditional handcrafted features to more recent deep learning methods. Good features may lead to good retrieval performance, but the problem is still not fully solved. To make the engine useful in real-world applications, retrieval efficiency is also an important factor to consider while has not received as much attention as feature extraction.

In this work, we focus on retrieval efficiency and provide a solution for real-time CBIR in million-scale databases. The feature vectors of database images are extracted and stored offline. During the online procedure, such feature vectors of query images are also extracted and then compared with database vectors, finding the nearest neighbours and returning the corresponding images as results. Since feature extraction only performs once for each query, the main limiting factor of retrieval efficiency in large-scale database is the time of finding nearest neighbours. Exact search has been shown to be far from adequate, and thus approximate nearest neighbour (ANN) search methods have been proposed, which mainly fall into two categories: compression-based and tree/graph-based. However, these two types of approaches are usually not discussed and compared together. Also, the possibility of combining them has not been fully studied. Our study (1) applies and compares methods in both categories, (2) reveals the gap betweentoy examples and real applications, and (3) explores how to get the best of both worlds. Moreover, a prototype of our image search engine with GUI is available on https://github.com/YYao-42/ImgSearch

MSc SS Thesis Presentation

• Wednesday, 20 July 2022
• 13:00-13:45
• zoom

Gaussian Process enhanced Distributed Particle filtering

In many applications of multi-agent networks, the physical systems consist of massive nonlinear and non-Gaussian elements. Hence, in the first decade of this century, intensive research on distributed particle filters (DPFs) has been conducted to address the distributed estimation problems. For distributed algorithms, communication overhead is an important metric in terms of engineering feasibility. In previous work, the approach to distributed particle filtering relies on a parameterization of the posterior proba-bility or likelihood function, to reduce communication requirements. However, as more and more effective resampling algorithms are proposed, the dependence of particle filter performance on particle set size is greatly reduced, so this thesis attempts to explore the possi-bility of DPFs based on direct particle exchange. In this thesis, the Gaussian process enhanced resampling algorithm is used. Meanwhile, several metaheuristic optimization algorithms (i.e., genetic algorithm and firefly algorithm) are further adapted to seek the global optimal particle set to improve the estimation performance. Furthermore, all algorithms are simulated in target tracking scenarios and are evaluated from three aspects: time, space, and communication complexity.

MSc SS Thesis Presentation

• Thursday, 30 June 2022
• 09:30-10:15
• EWI-Lecture Hall G

Sensor-to-Cell Height Estimation for Conductivity Estimation in Cardiac Cells

The heart is one of our vital organs. It functions by periodically contracting in a rhythmic way. Sometimes, this rhythmic behaviour is affected by abnormalities in the tissue. These conditions are referred to as cardiac arrhythmias. One kind is of specific interest, called atrial fibrillation (AF). To further study AF, methods have been developed to estimate the conductivity of cardiac cells based on the measured electrical signals. In addition, other parameters of interest besides the conductivity are involved. The goal of this project was to consider one of the parameters in the electrogram (EGM) model: the sensor-to-cell height.

First, we studied the effect of the height as a parameter in the model when used for conductivity estimation. To that end, a detector was built with which we can explain the effect of all involved parameters on the ability to accurately estimate any parameters of interest.

In addition, we considered the case where the height is unknown and is estimated, thus possibly including estimation errors. The focus was on the consequences of making errors in the estimation of the height with respect to conduction block detection and conductivity estimation.

Lastly, the effort was made to estimate the height. Here, the optimisation problem of height estimation was formalised and derived as its implementable form. At first, a simplified EGM model was assumed in order to mimic and estimate cell-specific effects, i.e. the cell conductivities. Then, the height was estimated in various situations to study its behaviour and performance under different conditions. Then, also the standard EGM model was used in the same way, after which we also tested the performance of the designed algorithm in combination with existing conductivity estimation methods.

MSc SS Thesis Presentation

• Thursday, 30 June 2022
• 10:00-10:45
• Lecture hall Data

Analyzing dynamic functional connectivity using state-space models on mice fUS data

In recent years, the increase in brain research led to the development of large-scale brain imaging techniques. With large-scale brain imaging techniques, such as functional magnetic resonance imaging (fMRI), functional connectivity analyses have shown altered connectivity patterns in humans and mice with neurobiological disorders, such as autism spectrum disorder (ASD). To further investigate different mutations that contribute to ASD, a behavioral neuroscientific experiment has been performed at the neuroscientific department of Erasmus MC. During the trial, brain activity in groups of wild-type and homozygous mice is measured using functional ultrasound (fUS) while allowing for spontaneous behaviors of mice. The homozygous mice lack the SHANK2 protein, resulting in hyperactivity and autistic-like behavioral alterations associated with ASD in humans. Understanding the origin of ASD is key to providing effective treatment. However, with the introduction of novel large-scale brain imaging techniques such as fUS, new methods have to be developed that enable functional connectivity analyses. Furthermore, new insights might also be acquired from dynamic functional connectivity analyses, in which changes in functional connectivity over time are evaluated. Subsequently, the question remains if it is possible to unravel differences in brain dynamics between wild-type and homozygous mice using a dynamic functional connectivity analysis.

            First, an fUS data model is developed to model how fUS signals arise from a generative perspective. This model comprises a combination of a convolutive and a state-space model. Subsequently, inference of functional networks and their temporal dynamics can be performed. Also, a pre-processing pipeline for experimental fUS data is designed to reduce problem complexity and data cleaning. The performance of the developed methods is evaluated on the experimental data set, where a difference in brain dynamics between wild-type and homozygous mice is investigated.

            It is found that a deconvolution procedure using the non-negative least absolute shrinkage and selection operator (NNLASSO) is necessary to reconstruct the underlying activity of neural populations. After that, using the hidden Markov model (HMM) as a state-space model, it is found that functional networks and their temporal dynamics can be learned from fUS data using expectation maximization (EM). It has been discovered that the developed methods consistently decompose reconstructed neural activity into biologically plausible functional networks from experimental fUS data. Also, with 96% certainty, a difference in brain dynamics between wild-type and homozygous mice is found using this method.

            In summary, in this thesis, novel methods are developed to perform a dynamic functional connectivity analysis on experimental fUS data. Also, by performing such dynamic functional connectivity analysis for the first time on fUS data, a consistent decomposition of reconstructed neural activity into biologically plausible functional networks and a possible difference in brain dynamics between wild-type and homozygous mice are found. This research highlights the potential of fUS as a large-scale brain imaging technique in the quest to understand the origin of ASD and other neurobiological disorders.

IEEE SPS IFS-TC Webinar Series

• Wednesday, 29 June 2022
• 16:00-17:00
• zoom (need to register)

Communication Efficient Privacy-Preserving Distributed Optimization

Privacy issues and communication costs are both major concerns in distributed optimization in networks. There is often a tradeoff between them because encryption methods used for privacy-preservation often introduce significant communication overhead. In this talk, we discuss a quantization-based approach to achieve both communication efficiency and privacy-preserving in the context of distributed optimization. By deploying an adaptive differential quantization scheme, we allow each node in the network to achieve the optimum solution with low communication costs while keeping its private data unrevealed. The proposed approach is general and can be applied in various distributed optimization methods, such as dual ascent and methods based on operator splitting (PDMM and ADMM). We consider two widely used adversary models, passive and eavesdropping, and investigate the properties of the proposed approach using different applications and demonstrate its superior performance compared to existing privacy-preserving approaches in terms of privacy, accuracy, and communication cost.

This webinar is organized by the IEEE Signal Processing Society Information Forensics and Security Technical Committee. See under "additional information" for a registration link.

MSc SS Thesis Presentation

• Monday, 27 June 2022
• 09:30-10:15
• EWI-Lecture Hall Chip

Estimation of Atrial Fibre Directions Based on Epicardial Electrograms

Being able to estimate atrial tissue conductivity parameters from epicardial electrograms is an important tool in diagnosing and treating heart rhythm disorders such as atrial fibrillation. One of these parameters is the atrial fibre direction, which is often assumed to be known in conductivity estimation methods.

In this thesis, a novel method to estimate the fibre direction from epicardial electrograms is presented. This method is based on local conduction slowness vectors of a propagating activation wave, which can be calculated from a corresponding activation map of the atrial tissue. These conduction slowness vectors follow an elliptical pattern that strongly depends on the underlying conductivity parameters. The fibre direction and conductivity anisotropy ratio can therefore be estimated by fitting an ellipse to the conduction slowness vectors.

Applying the presented method on simulated data shows that it can accurately estimate the fibre direction, and that the performance of the method depends mostly on the range of wavefront directions present in the measurement area. The main advantage of the presented method over existing methods is that it still functions in the presence of conduction blocks, as long as the surrounding tissue is approximately homogeneous.

TU Delft Women+ In Engineering

• Friday, 24 June 2022
• 16:00-18:00
• Pulse Breakout

End of year event

W+IE (Women+ in Engineering) invites you to this end of year meeting where we can enjoy drinks and snacks together and learn from our guest speakers from Qorvo WiT and TU Delft Integrity Office. We'll also learn more about gender issues by playing the Dilemma game.

PhD Thesis Defence

• Thursday, 23 June 2022
• 15:00-16:30
• Aula Senaatszaal

Multiband channel estimation for precise localization in wireless networks

Can we reach decimeter accuracy on wireless localization?

MSc SS Thesis Presentation

• Friday, 17 June 2022
• 14:00-15:00
• Van der Poelzaal (LB01.220)

Link adaptation and equalization for underwater acoustic communication using machine learning

The underwater acoustic environment is amongst the most challenging mediums for wireless communications. The three distinct challenges of underwater acoustic communication are the low and nonuniform propagation speed, frequency-dependent attenuation and time-varying multipath propagation. To cope with these challenges, physical layer communication systems allow the selection of communication parameters based on environmental conditions and constraints. This is also known as link adaptation.

In this thesis, the frequency-repetition spread-spectrum (FRSS) physical layer is studied. Various channel parameters are used to classify the optimal FRSS format. Furthermore, different machine learning classifiers are implemented to solve the classification problem. It is determined that the output signal-to-noise ratio provides enough information to switch effectively between transmission formats. Among the implemented machine learning classifiers, the decision tree strikes a good balance between performance and computational complexity. It is shown that a small performance gain can be achieved when custom channel parameters are extracted from the estimated impulse response and the equalizer error sequence using a deep neural network.

Optimizing the equalization process is another method to better cope with difficult environmental conditions. Various adaptive filter algorithms are implemented for the decision feedback equalizer used in the FRSS receiver. The optimal algorithm parameters are found by means of algorithm unrolling. It is shown that the standard least mean squares algorithm cannot be outperformed by various other optimization algorithms that use linear or non-linear filters.

The Watermark channel simulator is used to study the performance of the link adaptation and equalization optimization solutions for a wide range of underwater channels.

CAS MSc Midterm Presentations

• Thursday, 16 June 2022
• 13:30-14:30
• Online [zoom]

IMU-based Adaptive Motion Artifact Reduction in Wearable ECG

CAS MSc Midterm Presentations

• Thursday, 16 June 2022
• 13:00-14:30
• Online [zoom]

Novel View Synthesis of HDB Façade in Singapore using Neural Radiance Fields

CAS MSc Midterm Presentations

• Thursday, 16 June 2022
• 13:30-14:30
• Online [zoom]

Extreme-value Neural Networks for Weather Forcasting

PhD Thesis Defence

• Wednesday, 15 June 2022
• 15:00-16:30
• Aula Senaatszaal

Modelling and analysis of atrial epicardial electrograms

Atrial fibrillation (AF) is a frequently encountered cardiac arrhythmia characterized by rapid and irregular atrial activity, which increases the risk of strokes, heart failure and other heart-related complications. The mechanisms of AF are complicated. Although various mechanisms were proposed in previous research, the precise mechanisms of AF are not clear yet and the optimal therapy for AF patients are still under debated. A higher success rate of AF treatments requires a deeper understanding of the problemof AF and potentially a better screening of the patients.

In order to study AF, instead of using human body surface ECGs, we use the epicardial electrograms (EGMs) obtained directly from the epicardial sites of the human atria during open heart surgery. This data is measured using a high-resolution mapping array and exhibits irregular properties during AF. Although different studies have analyzed electrograms in time and frequency domain, there remain many open questions that require alternative and novel tools to investigate AF.

Experience in signal processing suggests that incorporating the spatial dimension into the time-frequency analysis on the multi-electrode electrograms may provide improved insights on the atrial activity. However, the electrophysiologcial models for describing spatial propagation are relatively complex and non-linear such that conventional signal processing methods are less suitable for a joint space, time, and frequency domain analysis. It is also difficult to use very detailed electrophysiologcial models to extract tissue parameters related to AF fromthe high-dimensional data.

In this dissertation, we propose a radically different approach to study and analyze the EGMs from a higher abstraction level and from different perspectives to get more understanding of the characteristics of AF. We also develop a simplified electrophysiological model that can capture the spatial structure of the data and propose an efficient method to estimate the tissue parameters, which are helpful to analyze the electropathology of the tissue, e.g., cell activation time or conductivity. 

CAS MSc Midterm Presentations

• Thursday, 9 June 2022
• 13:30-14:30
• Online [zoom]

Reconstruction and Rendering of Buildings as Radiance Fields for View Synthesis

Signal Processing Seminar

• Thursday, 9 June 2022
• 13:30-14:30
• Online [zoom]

Data Processing on Expanding Graphs with Graph Filters

PhD Thesis Defence

• Wednesday, 8 June 2022
• 09:30-11:30
• Aula Senaatszaal

Direction of Arrival estimation and Self-Calibration techniques using an array of Acoustic Vector Sensors

The localization and characterization of sound have played a vital role in various ap­plications, ranging from noise control of machinery to battlefield awareness. Micro­phone arrays are commonly used to find sound sources, which implicitly inherits a se­ries of limitations. Alternatively, acoustic vector sensors (AVSs) have shown promising results in overcoming most of those limitations, specifically having a larger operation frequency while requiring a smaller number of sensor nodes. However, literature about this topic is still evolving and mainly focused on the theoretical aspects, disregarding most real-world limitations. This thesis extends the AVS arrays' theoretical framework for direction-of-arrival (DOA) estimation of far-field sources while considering practical constraints. Specifically, the study considers the DOA estimation problem using AVS ar­rays in three main scenarios: spatially under-sampled configurations, the presence of calibration errors, and sensors with a reduced number of channels.

The idea of spatial sampling by AVS arrays has a different interpretation compared to the equivalent acoustic pressure sensor (APS) arrays. Notably, it is possible to carry out unambiguous DOA estimation using a spatially under-sampled AVS array, which is the main topic of interest in the first part of this work. Here we study the effects of the grating lobes or spatial aliasing on the performance of DOA estimation. We will observe that this idea can also be extended to beamforming applications.

Subsequently, in the second part of this work, we consider the DOA estimation prob­lem using AVS arrays in the presence of calibration errors. First, identifiability conditions are derived for the solution to exist. Then two main classes of self-calibration approaches are proposed. The first calibration approach is array geometry independent and is based on sparse recovery techniques that lead to a one-step solver to estimate both the source DOAs and the calibration parameters jointly. Further, the extension of the proposed self­calibration approach in the presence of wide-band sources is also presented. The sec­ond calibration approach applies only to a uniform linear array (ULA) of AVSs, where the Toeplitz block structure of its covariance matrix is exploited to estimate the calibration errors followed by the estimation of the source DOAs.

In the last part of the thesis, an alternate configuration of an AVS is considered for DOA estimation with a reduced channel count. We refer to such an AVS as a uniaxial AVS (U-AVS). The DOA estimation performance using a U-AVS array is analyzed, and specifically, the impact of the extra degree-of-freedom originating from the fact that each U-AVS in the array can have arbitrary orientation is studied comprehensively. Further­more, all the analyses and proposed algorithms in this thesis are supported by real ex­perimental results performed with AVS arrays in an anechoic chamber.

To conclude, this research on AVS arrays paves the way to achieve an increased sit­uational awareness across our society; this could be either by detecting and localizing problems or threats occurring in an urban environment or assisting soldiers on the bat­tlefield to make a timely decision to achieve peace.

CAS MSc Midterm Presentations

• Thursday, 19 May 2022
• 13:30-14:30
• Online [zoom]

Link Adaptation and Equalization for Underwater Acoustic Communication Using Machine Learning

CAS MSc Midterm Presentations

• Thursday, 19 May 2022
• 13:30-14:30
• Online [zoom]

IMU-based Adaptive Motion Artefact Reduction in Wearable ECG

CAS MSc Midterm Presentations

• Thursday, 19 May 2022
• 13:30-14:30
• Online [zoom]

Finding Camera Position and Orientation using Marker-Less Computer Vision in the Catheterization Laboratory

MSc SS Thesis Presentation

• Monday, 16 May 2022
• 09:30-10:15
• Timmanzaal (LB01.170)

Multiple Subbands Ranging Signals Design and Investigation on Frequency Dependence of the Subband Channel Impulse Responses within an Ultra-wideband Channel

This thesis covers two topics. The first one is signal design for accurate Time-of-Arrival estimation using a number of frequency separated signals. Rather than use a full UWB band, we will using sparse subband signals spanning the full band to construct a new virtual UWB signal. To evaluate the performance of the constructed signal, Cramerrao lower bound and auto-correlation are used. And given a given fixed bandwidth the number of subbands within a 1 GHz UWB channel, optimal subbands’ allocation will be found based on the evaluation results. Our results show that when three 50 MHz subbands are used to construct a virtual 1GHz UWB signal, a lower CRLB and better auto-correlation performance can be reached when subband are close to the edges of the virtual band. However, the autocorrelation still has multiple peaks, which poses a serious challenge for accurate time estimation.

The second topic is to investigate the frequency dependence of the channel impulse response of subbands with different frequency separations. We propose a covariance calculation method to determine the frequency dependence which changes with frequency separation. To validate the method, different artificial UWB channels with distinct paths are given. The results show that covariances between the subband CIRs stay at high level when measured at the direct path and the majority interference caused by other paths can be eliminated by wider bandwidth subband. Given UWB channels measured from 5 to 10 GHz with a link-budget of 120dB, the frequency dependence of the direct path and reflections are determined, different bandwidths and frequency separations are used, the results show that the channel impulse response of the subbands will become different when measured at different center frequencies, where the difference increases with increased frequency separation of the subbands. The correlation of the direct path is maintained over larger frequency distances than that of reflected paths.

CAS MSc Midterm Presentations

• Thursday, 12 May 2022
• 13:30-14:30
• Online [zoom]

Re-ranking for Improved Image Query-Based Search

CAS MSc Midterm Presentations

• Thursday, 12 May 2022
• 13:30-14:30
• Online [zoom]

Detecting Medical Equipment in the Catheterization Laboratory using Computer Vision

CAS MSc Midterm Presentations

• Thursday, 12 May 2022
• 13:30-14:30
• Online [zoom]

Distributed Gaussian Process on Multi-Agents Network for Environmental Monitoring

PhD Thesis Defence

• Monday, 2 May 2022
• 10:00-11:30
• Aula Senaatszaal

Image Reconstruction for Low-Field MRI

This thesis presents imaging algorithms for a prototype low-field MRI system, in particular regularization, handling missing data, and deep learning approaches.

CAS MSc Midterm Presentations

• Thursday, 28 April 2022
• 13:30-14:30
• Online [zoom]

Frequency Domain Joint Estimation of HRF and Stimulus from Functional Ultrasound Data

CAS MSc Midterm Presentations

• Thursday, 28 April 2022
• 13:30-14:30
• Online [zoom]

Cooperative Localization using ADS-B for Unmanned Aerial Vehicles

CAS MSc Midterm Presentations

• Thursday, 28 April 2022
• 13:30-14:30
• Online [zoom]

Development of a Detect and Avoid Control System for Autonomous Agents in Cluttered Environments using the Artificial Potential Field Method

MSc SS Thesis Presentation

• Tuesday, 26 April 2022
• 10:15-11:00
• EWI Snijderszaal

Estimating atrial activity in epicardial electrograms: a beamforming perspective

The most common serious heart rhythm disease is atrial fibrillation. It is not fatal on its own but does increase the risk of heart failures and strokes. There is little understanding about the mechanisms behind the disease, so more insight is desired. Using an array of electrodes, measurements are being performed of the electrical atrial activity directly on the heart tissue. These signals are, however, not clean and suffer from far-field interference coming from the ventricles.

During normal sinus rhythm the atrial and ventricular activities are separated in time and easy to distinguish. In case of atrial fibrillation this is not always the case. Luckily, there is a major difference between both signals: the ventricular signal comes from far away and arrives therefore approximately simultaneously at all electrodes. A simple, but effective way to remove this ventricular activity is to use a bipolar electrode. It produces the difference between two normal unipolar electrodes, thus removing the common ventricular signal component.  

The bipolar electrode, however, distorts the atrial signal component, which in some orientations can even lead to removing it altogether. This bipolar electrode is known as a differential beamformer from the field of array signal processing. There are more complex beamformers that can keep the atrial component undistorted and therefore produce better results than the bipolar electrode.  

This thesis proposes a Fourier-domain signal model for all available electrodes relying on an atrial and ventricular transfer function. It is possible to estimate these transfer functions from the data blindly. Three beamformers are derived utilizing the signal model and the transfer functions. The bipolar electrode is extended to multiple electrodes like the other beamformers as well.

This thesis proposes a Fourier-domain signal model for all available electrodes relying on an atrial and ventricular transfer function. It is possible to estimate these transfer functions from the data blindly. Three beamformers are derived utilizing the signal model and the transfer functions. The bipolar electrode is extended to multiple electrodes like the other beamformers as well.  

Experiments with simulated data show that the complex beamformers indeed keep the atrial activity undistorted and are still able to remove the ventricular activity effectively when using multiple electrodes, except for very complex data where the signal model is not valid. For low numbers of electrodes the beamformers are not useful, they hardly remove the ventricular activity while keeping the atrial component undistorted, where the bipolar electrode does the opposite.  

The electrograms are also used to estimate local activation times of the cells underneath the electrodes which says something about the health of the cardiac tissue. Besides the mentioned filtering, this thesis proposes a method to estimate those moments in time by looking at the time-domain version of the atrial transfer function, called the atrial impulse response. For simple data, it performs well compared to state-of-the-art methods, but for more complicated data, it does not.

MSc SS Thesis Presentation

• Monday, 25 April 2022
• 14:00-14:45
• (online)

Towards Gridless Sound Field Reconstruction

Sound pressure varies over space and time. Knowledge about this exact behavior has many applications, e.g., room compensation, dereverberation and sound field reconstruction. Inside enclosures, the sound field is influenced by the surroundings, such as the geometry of the enclosure and the materials used. Reconstructing a satisfying sound field in the whole enclosure by extrapolating from few measurements is thus not an obvious task.

The sound field in a room can be represented by a weighted sum of room modes. Thus, we can estimate the room modes and compute the sound field from it. To estimate the room modes, compressive sensing literature uses on-the-grid, sparse reconstruction methods. However, these on-the-grid methods are known to suffer from basis mismatch.

In this work, we investigate the use of a gridless framework for estimating room modes using atomic norm minimization, a gridless method. The advantage of this approach is that it does not suffer from this basis mismatch problem. We derive a bound for the sound field reconstruction problem in a one-dimensional room with rigid walls and relate this to the frequency separation that is required by the atomic norm. We conclude that for perfect reconstruction of the room modes based on the investigated gridless approach, additional prior knowledge about the signal model is required. For example, knowledge about the shape of the room modes can be used. We show how recovery is possible in a one-dimensional setting by exploiting both the structure of the sound field and the acquisition method.

CAS MSc Midterm Presentations

• Thursday, 21 April 2022
• 13:30-14:30
• Online [zoom]

Personnel Activities Observation in the Catheterization Laboratory

CAS MSc Midterm Presentations

• Thursday, 21 April 2022
• 13:30-14:30
• Online [zoom]

Autonomous Landing of an UAV

CAS MSc Kick-off Presentations

• Thursday, 21 April 2022
• 13:30-14:30
• Online [zoom]

Machine Learning of Ultrasound Data: Cardiovascular Parameters Detection Using Carotid Artery Ultrasound Measurements

CAS MSc Midterm Presentations

• Thursday, 14 April 2022
• 13:30-14:30
• Online [zoom]

The Impact of Jamming and Spoofing on GNSS Signals

CAS MSc Midterm Presentations

• Thursday, 14 April 2022
• 13:30-14:30
• Online [zoom]

Energy-Efficient Seizure Detection for Wearable EEG

CAS MSc Midterm Presentations

• Thursday, 14 April 2022
• 13:30-14:30
• Online [zoom]

Distributed Particle Filtering

CAS MSc Midterm Presentations

• Thursday, 7 April 2022
• 13:30-14:30
• Online [zoom]

Modular Inductive Biases: To What Extent Can They Improve Neural Reasoning and Generalization?

CAS MSc Midterm Presentations

• Thursday, 7 April 2022
• 13:30-14:30
• Online [zoom]

Distributed System Design for Space-based Correlators

Signal Processing Seminar

• Thursday, 7 April 2022
• 13:30-14:30
• Online [zoom]

Single-Pulse Estimation of Target Velocity on Planar Arrays

CAS MSc Midterm Presentations

• Thursday, 31 March 2022
• 13:30-14:30
• Online [zoom]

Relative Affine Localization for Robust Formation Control

CAS MSc Midterm Presentations

• Thursday, 31 March 2022
• 13:30-14:30
• Online [zoom]

Search by Image: A Digital History Search Engine in Online Image Repositories

MSc SS Thesis Presentation

• Tuesday, 29 March 2022
• 13:00-13:45
• zoom (and HB17.090)

Few-shot emotion recognition using intelligent voice assistants and wearables

Emotion Recognition is one of the vastly studied areas of affective computing. Attempts have been made to design emotion recognition systems for everyday settings. The ubiquitous nature of Intelligent voice assistants (IVAs) in households, make them a great anchor for the introduction of emotion recognition technology to consumers. The existing systems lack such pipelines and rely on dictionary-based architectures in their design. Further, these systems lack conversational properties and are merely an extension of information retrieval engines.

In this setting, we propose to introduce and develop emotion recognition pipelines that are suited to the interactions, common with these IVAs. To augment the existing emotion recognition pipelines which rely on audio information, we look at physiological information derived from wearables. Our proposed model uses multimodal embeddings with a Siamese Network to achieve the task of emotion recognition from a few samples. Physiological signals of blood volume pulse (BVP) and electrodermal activity (EDA) are used as additional input embeddings to two audio embeddings arising from the speech samples. We employ the state-of-the-art training schedules for Siamese Networks, which use a very limited amount of training on support datasets via sample pair comparisons. The performance of the model is evaluated using weighted binary accuracy and f1 scores.

The proposed model is applied on two datasets that denote two unique experimental settings - the K-EmoCon dataset and RECOLA dataset. We demonstrate an improvement in the state-of-the-art accuracy with the K-EmoCon dataset with accuracies of 63.97% and 66.91% on arousal and valence dimensions respectively. Further, on the RECOLA dataset, the model performs moderately well with 53.81% and 53.87% respectively for arousal and valence dimensions. In addition to this, we present a study of the effects of variation of available support set for training from the dataset. We make some salient observations for these experiments across individual participants and also identify how the label distributions affect the performance of the model. Further, we investigate the impact of real-world noise samples from the DEMAND dataset on the two datasets. We observe that the proposed model is robust and performs sustainingly well even in the presence of imputed noise.

CAS MSc Midterm Presentations

• Thursday, 24 March 2022
• 14:00-15:00
• Online [zoom]

Distributed Optimization using Stochastic PDMM: Convergence, Transmission Losses and Privacy

CAS MSc Kick-off Presentations

• Thursday, 24 March 2022
• 14:00-15:00
• Online [zoom]

IMU-based Adaptive Motion Artifact Reduction in Wearable ECG

Signal Processing Seminar

• Thursday, 24 March 2022
• 14:00-15:00
• Online [zoom]

High Performance Control System Architecture with an Output Regulation Theory-based Controller and Two-Stage Optimal Observer for the Fine Pointing of Large Scientific Satellites

This study describes the functional architecture and the control law of an attitude control system designed for large scientific satellites with high accuracy and stability pointing requirements, such as those necessary during the fine pointing operations. The satellite dynamical motion has been reproduced by an accurate software simulator implementing the error models of the sensors and actuators currently being part of the on-board system configuration of state-of-the-art space missions. Moreover the main disturbances affecting the satellite motion during its mission are considered. The control system architecture includes an optimal estimator, the well-known Multiplicative Extended Kalman Filter, modified for the time delay correction of the star tracker measurements by means of a forward state propagation for a correct implementation of measurement update. This estimator is cascaded with a second one specifically used for disturbance acceleration estimation. The estimation task is then coupled with a linear optimal control algorithm combined into an error feedback output regulator, based on a space system design involving the exosystems of the disturbances and reference profiles. The reported control system has analytically been proved to be globally stable and the reported simulations highlight the high performances of the algorithm, capable of fully satisfying a stringent pointing error requirement. Lastly, the proposed attitude control system is shown to be capable of tracking not only step reference trajectories but also sinusoidal time varying signals within a disturbed environment and promises robust performances also in case of possible actuators faults.

CAS MSc Information Market

• Thursday, 24 March 2022
• 12:30-13:30
• EWI HB 17.110

If you are a first year MSc EE student (Signals & Systems, WiCos), then on Thu 24 March, come visit the CAS group at the 17th floor on the EWI tower, meet the professors, and learn about graduation topics for next year.

Signal Processing Seminar

• Thursday, 17 March 2022
• 13:30-14:30
• Online [zoom]

RL-based Path Planning and Coverage for Autonomous UAVs in Unknown Environments

Unmanned Aerial Systems (UASs) have become a relevant sector in the aerospace industry. In the last decade, the increase in the capabilities of Unmanned Aerial Vehicles (UAVs), paired with a drop in their price, has led them to be used in many different applications where they are employed for their versatility and efficiency. A challenge that is being addressed in this field is that of autonomous UAVs fleets, i.e., the coordinated use of multiple UAVs to perform a common task. A particularly interesting application of UAV fleets is their use in the exploration and mapping of unknown or critical environments. This topic brings with it a significant number of challenges, from the design of the policy used to coordinate the fleet to the path planning algorithm that each UAV uses to move in the environment while exploring it. In my research, I studied and implemented a Reinforcement Learning (RL)-based approach for the cooperative exploration of unknown environments by a fleet of UAVs is presented. Different approaches have been considered and compared. A Reinforcement Learning-based approach has been developed taking inspiration from some of the other methods studied. Two RL agents are trained to address the exploration problem: the first has the task of coordinating the coverage task, optimizing the way the UAVs spread in the unknown area by assigning some waypoints to them. The waypoints are placed in order to optimize the distribution of the fleet and to maximize the exploration process efficiency. The second RL agent is a path planning algorithm and is used by each UAV to move in the environment to reach the region pointed by the first agent. The combined use of the two agents allows the fleet to coordinate in the execution of the exploration task.

CAS MSc Midterm Presentations

• Thursday, 17 March 2022
• 13:30-14:30
• Online [zoom]

Analyzing Dynamic Functional Connectivity using State-Space Models on Mice fUS Data

CAS MSc Midterm Presentations

• Thursday, 10 March 2022
• 13:30-14:30
• Online [zoom]

Link Adaptation for Underwater Acoustic Communication using Machine Learning

CAS MSc Midterm Presentations

• Thursday, 10 March 2022
• 13:30-14:30
• Online [zoom]

Image Search Engine Based on Deep Learning for Digital History

CAS MSc Kick-off Presentations

• Thursday, 10 March 2022
• 13:30-14:30
• Online [zoom]

ICA Through Diagonalization of the Implicitly Formed Higher Order Cumulant Tensor

CAS MSc Kick-off Presentations

• Thursday, 24 February 2022
• 13:30-14:30
• Online [zoom]

Detect and Avoid Control System for Autonomous Drones in Cluttered Environments

CAS MSc Midterm Presentations

• Thursday, 24 February 2022
• 13:30-14:30
• Online [zoom]

Improving the Estimation of Atrial Activation Times Using Spatial Information

Signal Processing Seminar

• Thursday, 24 February 2022
• 13:30-14:30
• Online [zoom]

Sampling and Reconstruction in IoT Networks on Graphs

Microelectronics Colloquium

• Thursday, 24 February 2022
• 16:30-05:00
• Online via TEAMS

An inclusive EEMCS faculty: An emphatic approach.

Our faculty consists of a vibrant and diverse community. Diversity is a catalyst that allows us to achieve broad knowledge, and a base upon we can drive scientific innovation and improve education.

Moreover, diversity is one of the core values of TUDelft and our faculty and comes with great responsibility. Without equality and inclusion diversity becomes an empty gesture. But realising a safe, equal and inclusive environment requires the participation of everyone in our community. It starts by having a dialog, stablish communication channels at different levels, and debunking taboos with respect to the visible and invisible differences among each other and our students. An empathic approach for this process can play a key role in realising this ambition.

In this colloquium Jorge talks about his experience within EDIT: EEMCS Diversity & Inclusion Team. Join us to know more about EDIT, and for an informal discussion on the current advancements on addressing issues like harassment, discrimination, and gender (in)equality. Or if you want to know what are the channels and means within our faculty and our University to reach for advice or help in case you encounter any issues related to these important topics.

Signal Processing Seminar

• Thursday, 17 February 2022
• 13:30-14:30
• Online [zoom]

Sparsity-Based Vascular Ultrasound Imaging Through Compressive Spatial Coding

CAS MSc Kick-off Presentations

• Thursday, 17 February 2022
• 13:30-14:30
• Online [zoom]

Extreme Weather Forecasting Using Deep Generative Model

CAS MSc Kick-off Presentations

• Thursday, 17 February 2022
• 13:30-14:30
• Online [zoom]

Truncated Array Sound Field Synthesis

CAS MSc Midterm Presentations

• Thursday, 10 February 2022
• 13:30-14:30
• Online [zoom]

Estimation of Atrial Activity using Beamforming in Epicardial Electrograms

CAS MSc Kick-off Presentations

• Thursday, 10 February 2022
• 13:30-14:30
• Online [zoom]

Energy-Efficient Seizure Detection for Wearable EEG

CAS MSc Kick-off Presentations

• Thursday, 10 February 2022
• 13:30-14:30
• Online [zoom]

Cooperative Localization of Drones using ADS-B Data

CAS MSc Kick-off Presentations

• Thursday, 10 February 2022
• 13:30-14:30
• Online [zoom]

Distributed Gaussian Process for Multi-Agents System

CAS MSc Kick-off Presentations

• Thursday, 27 January 2022
• 13:30-14:30
• Online [zoom]

Detecting Medical Instruments in Surgical Nets with Deep Learning

CAS MSc Kick-off Presentations

• Thursday, 27 January 2022
• 13:30-14:30
• Online [zoom]

Extreme-Value Deep Generative Models for Weather Forecasting

CAS MSc Kick-off Presentations

• Thursday, 27 January 2022
• 13:30-14:30
• Online [zoom]

Link Adaptation for Acoustic Underwater Communication Using Machine Learning

Signal Processing Seminar

• Thursday, 27 January 2022
• 13:30-14:30
• Online [zoom]

Introduction

CAS MSc Kick-off Presentations

• Thursday, 20 January 2022
• 13:30-14:30
• Online [zoom]

Extreme-Value Neural Networks for Weather Forecasting

CAS MSc Kick-off Presentations

• Thursday, 20 January 2022
• 13:30-14:30
• Online [zoom]

Distributed Particle Filtering

CAS MSc Midterm Presentations

• Thursday, 20 January 2022
• 13:30-14:30
• Online [zoom]

Towards Sustainable Satellite Swarms

CAS MSc Midterm Presentations

• Thursday, 13 January 2022
• 13:30-14:30
• Online [zoom]

Towards Gridless Sound Field Reconstruction

CAS MSc Kick-off Presentations

• Thursday, 13 January 2022
• 13:30-14:30
• Online [zoom]

3D Person Tracking in the Catheterization Laboratory

CAS MSc Kick-off Presentations

• Thursday, 13 January 2022
• 13:30-14:30
• Online [zoom]

Autonomous Landing of an UAV

CAS MSc Kick-off Presentations

• Thursday, 13 January 2022
• 13:30-14:30
• Online [zoom]

Frequency Domain Joint Estimation of HRF and Stimulus from Functional Ultrasound Data

MSc SS Thesis Presentation

• Friday, 17 December 2021
• 09:30-10:15

A Decentralized Key Management System for the European Railway Signalling System

MSc Biomedical Engineering Thesis Presentation

• Thursday, 16 December 2021
• 09:30-10:05
• zoom

An Expanded IPFM Model for Heart Rhythm Analysis

Atrial Fibrillation affects millions of people worldwide. It is associated with an impaired quality of life and an increased risk of stroke, cardiac failure and mortality. Treatments exist, but early detection and treatment is crucial, due to the progressive nature of the disease. Algorithms can help with early detection.

Machine learning algorithms are commonly trained to diagnose based on ECG data, but the interpretability is low. A physiological model that simulates the heart gives more insight into the situation of the patient. Current approaches, like the IPFM model, simulate only the SA node and generate RR intervals as output, while completely neglecting the interaction between the AV and SA node. By using an IPFM model and including the AV node as well, an extended and more accurate physiological model was built to more accurately detect Atrial Fibrillation. The AV node model is able to estimate PR intervals when the P waves are annotated. This result shows that the model extension is able to capture information about the signal conduction.

When the SA node model and the AV node model are cascaded and only the R peaks are considered, the classification accuracy does not improve compared to the SA node model alone. The R peaks alone do not contain sufficient information for accurate parameter estimation. The parameters governing the behavior of the AV node seem different for NSR compared to AF, but more data is needed to confirm this. The ability of the model to predict PR intervals gives hope that the inclusion of P wave data should improve the performance of the classification with the extended physiological model.

MSc CE Thesis Presentation

• Tuesday, 14 December 2021
• 14:00-14:35
• zoom

Neuromorphic computing application of emerging memory technologies for spiking neural networks

Renewed interest in memory technologies such as memristors and ferroelectric devices can provide opportunities for traditional and non-traditional computing systems alike. To make versatile, reprogrammable AI hardware possible, neuromorphic systems are in need of a low-power, non-volatile and analog memory solution to store the weights of the spiking neural network (SNN). In addition to being used for memory, memristive memory can be read out passively and thus also replaces digital-to-analog circuitry.

In this thesis, two solutions are proposed: one is based on a generalized memristor, the other is based on ferroelectric memory. Both solutions are implemented and simulated in SystemC AMS and tested with a spiking neural network (SNN). As a final test, both memory solutions are integrated into a full-sized SNN and simulated against the MNIST dataset. The simulation results validate the capabilities of memristive and ferroelectric memory when it comes to providing a sensible weight storage solution for neuromorphic systems.

CAS MSc Kick-off Presentations

• Thursday, 9 December 2021
• 13:30-14:30
• Online [zoom]

Finding Camera Position and Orientation using Marker-Less Computer Vision in the Catheterization Laboratory

Signal Processing Seminar

• Thursday, 9 December 2021
• 13:30-14:30
• Online [zoom]

Nonlinear and Time-Varying Aspects of the Functional Ultrasound Response

CAS MSc Kick-off Presentations

• Thursday, 9 December 2021
• 13:30-14:30
• Online [zoom]

The Impact of Jamming and Spoofing on GNSS Signals

CAS MSc Kick-off Presentations

• Thursday, 9 December 2021
• 13:30-14:30
• Online [zoom]

An image processing solution to building inspection

MSc SS Thesis Presentation

• Friday, 3 December 2021
• 15:00-15:35

Analyzing Functional Ultrasound Images of the Brain Using Tensor Decompositions

Functional ultrasound (fUS) is a neuroimaging modality that offers high spatial and temporal resolution while also providing portability. In this thesis, neuroimaging data acquired with fUS at Center for Ultrasound and Brain imaging at Erasmus MC (CUBE) is processed. Due to the fact that fUS data is inherently multidimensional, we propose using tensor decompositions, tensors here referring to generalizations of matrices, for processing of fUS data.

We define two main research questions regarding fUS data analysis. First, for compressing the large-scale raw beamformed fUS data, we apply sequentially truncated multilinear singular value decomposition. This compression method is compared against ensemble averaging used in the conventional pipeline, and shown to provide a higher compression rate while preserving more temporal resolution for specific ranks. Furthermore, it is observed to denoise the data, resulting in a more precise extraction of the active region of Superior Colliculus using correlation maps.

Secondly, in order to investigate the advantage of multi-slice processing that incorporates 3-D information, blind-source separation methods are applied to single slice and two-slice fUS recordings. After applying independent component analysis (ICA) to the matricized data as a benchmark method, block termdecomposition (BTD) is used as a way of processing the data as it is, in its natural 3-D structure without vectorization. Through a simulation study, it is shown that the method is able to separate two images even when using a rank that is lower than the true rank, as well as in noisy conditions. Subsequently, BTD is applied to real 4-D fUS data formed by concatenation of slices in a new dimension. However, this method is seen to perform worse than single slice ICA in terms of extracting the active regions. In order to amplify common information between slices, a new 3-D data structure is then formed by summing the fUS data of two slices. For extraction of this common information, a BTD is then applied to the aggregate 3-D data. The findings of this decomposition reveal that both taking a longer portion of single slice data and incorporating the second slice helps to achieve better results.

CAS MSc Kick-off Presentations

• Thursday, 2 December 2021
• 13:30-14:30
• Online [zoom]

Developing a Data-Driven Model to Predict to the Noise Attenuation Level in Earplugs

Signal Processing Seminar

• Thursday, 2 December 2021
• 13:30-14:30
• Online [zoom]

Faster-Than-Nyquist Signaling Short-Packet Communication

CAS MSc Kick-off Presentations

• Thursday, 2 December 2021
• 13:30-14:30
• Online [zoom]

Sparse Recurrent Independent Mechanisms for Vision Tasks

MSc CE Thesis Presentation

• Tuesday, 30 November 2021
• 14:00-14:35
• Building 33 Pulse A0.400 Technology

Spiking CA-CFAR Implementation for Radar Target Detection

Radar systems have been used for decades to detect targets on the ground and in the air. The radar signal is transformed into a rangedoppler image that distinguishes each detected object by range and velocity to process radar data. A target detection algorithm is used to filter noise and unwanted reflections. Each target can be in a region with different noise levels; a simple threshold will yield false positives or miss detections depending on its value. To solve this problem, a Constant False Alarm Rate or CFAR is desirable.

A CFAR detector estimates the noise surrounding each target and has a dynamic threshold based on this. Spiking Neural Networks are the third generation of Artificial Neural Networks where, instead of continuous signals, the input is encoded into trains of spikes over time. These networks have a potentially efficient hardware implementation instead of the older generation Artificial Neural Networks and could run directly at the sensor edge, lowering latency and power consumption.

This thesis will explore a Spiking Cell Averaging CFAR implementation and attempt to use its desirable properties like a temporal average over multiple radar frames, mimicking the non-coherent integration sometimes done in radar processing. It is shown that some configurations will behave similarly in a simulated environment with additive white Gaussian noise.

MSc SS Thesis Presentation

• Tuesday, 30 November 2021
• 09:00-09:35
• zoom

Investigation of focal epilepsy using graph signal processing

Epilepsy is one of the most common neurological disorders worldwide. Its manifestations, the seizures, are due to a group of neurons’ abnormal and synchronous activity. The unpredictable nature of these events hinders the quality of life of those affected. In particular, focal seizures show a localized onset of abnormal activity and are the most common ones. Correct detection of the episodes can help clinicians to give the best medical treatments. This research project arises from the need to have automatic algorithms for seizure detection with a high number of correctly detected seizures for low false alarm rates.

Recent studies have shown disorganization in how brain areas interact with each other before and during a seizure. We decided to model this change in connectivity patterns by inferring graphs from EEG recordings of epileptic patients. We work with seventeen subjects suffering from focal epilepsy, and we build, for each of them, a graph of the activity preceding (preictal) and during (ictal) a seizure. After that, we exploit techniques from graph signal processing to build a detector for seizures. Last, we analyze the density of connections of the inferred graphs to indicate the seizure onsets. 

The obtained results are unsuitable for real-life applications, but they are a starting point for further research. Furthermore, we find that most the proposed ictal or preictal graphs show fewer connections in the nodes involved with the seizure onset.

MSc CE Thesis Presentation

• Monday, 29 November 2021
• 10:00-10:35
• zoom

Hardware Spiking Neural Network based Sbox AES

Hardware cryptographic algorithm implementation is easy to attack by side-channel attacks. The power-based side-channel attacks are powerful among several side-channel attacks. This attack methods use the relationship between the leakage model and power traces to reveal the secret key. Some existing countermeasures like mask and hide can protect the algorithms from attacking. However, they can not break the relationship between power traces and the leakage model. Based on the property of the neural network, the linear relationship can be easily broken. Furthermore, the spiking neural network is more hardware-friendly than a conventional neural network. The design replaces the sbox in AES with a pipeline spiking neural network-based sbox and implements it in hardware. The help of the FPGA attack platform demonstrates that the proposed design can resist DPA, CPA, Template Attacks, and Deep Learning-based attacks.

CAS MSc Kick-off Presentations

• Thursday, 25 November 2021
• 13:30-14:30
• Online [zoom]

Robust Formation Maneuvering of Multi-agent Systems

CAS MSc Kick-off Presentations

• Thursday, 25 November 2021
• 13:30-14:30
• Online [zoom]

Search by Image: A Digital History Search Engine in Online Image Repositories

CAS MSc Kick-off Presentations

• Thursday, 25 November 2021
• 13:30-14:30
• Online [zoom]

Crack Detection for Building Inspection Using UAV with Image Alignment

CAS MSc Kick-off Presentations

• Thursday, 11 November 2021
• 13:30-14:30
• Online [zoom]

Analyzing Dynamic Functional Connectivity Using State-Space Models

Signal Processing Seminar

• Thursday, 11 November 2021
• 13:30-14:30
• Online [zoom]

Low Complex Accurate Multi-Source RTF Estimation

CAS MSc Midterm Presentations

• Thursday, 28 October 2021
• 11:00-12:00
• Online [zoom]

Frequency Dependence of the Impulse response in an Ultra-wideband Channel

CAS MSc Kick-off Presentations

• Thursday, 28 October 2021
• 11:00-12:00
• Online [zoom]

Sound Field Recovery by Estimating Room Modes Off-the-Grid

CAS MSc Kick-off Presentations

• Thursday, 28 October 2021
• 11:00-12:00
• Online [zoom]

Image Search Engine Based on Deep Learning for Digital History

Microelectronics Colloquium

• Thursday, 28 October 2021
• 15:30-16:30
• Microsoft Teams Meeting

On my personal journey into artificial intelligence

In this presentation, I will start with a brief introduction to artificial intelligence (AI). I will then elaborate on two types of AI approaches that our research team is investigating: graphical models and neural networks. Next I will summarize some of the main research results of our group. I will review some of the applications of AI that we have been working on over the years, and will present some of our future research plans. I will also say a few words about the spin-off companies that have emerged from our research group. At last, I will conclude with a few thoughts on the potential impact of AI on society and will formulate a few important open research questions in the field of AI.

PhD Thesis Defence

• Wednesday, 27 October 2021
• 12:01-13:00
• Senaatszaal of Aula Building

Atrial Fibrillation Fingerprinting

Atrial fibrillation (AF) is a common age-related cardiac arrhythmia. AF is characterized by rapid and irregular electrical activity of the heart leading to a higher risk of stroke and heart failure. During AF, the upper chambers of the heart, called atria, experience chaotic electrical wave propagation. However, despite the various mechanisms introduced in the literature, there is still an ongoing debate on a precise and consistent mechanism underlying the initiation and perpetuation of AF. Some studies show that AF is rooted in impaired electrical conduction and structural damage of atrial tissue, known as electropathology. Atrial electrograms (EGMs) recorded directly from heart’s surface, provide an important diagnostic tool to localize and quantify the degree of electropathology in the tissue. However, the analysis of the electrograms is currently constrained by the lack of suitable methods that can reveal the hidden electrophysiological parameters of the tissue. These parameters can be used as local indication of electropathology in the tissue. We believe that understanding AF and improving AF therapy starts with developing a proper forward model that is accurate enough (from a physiological point of view) and simultaneously simple enough to allow for subsequent parameter estimation. Therefore, the main focus of this thesis is on developing a simplified forward model that can efficiently explain the observed EGM based on AF relevant tissue parameters. An initial step before performing any analysis on the data is to remove noise and artefacts. All atrial electrogram recordings suffer from strong far-field ventricular activities (VA). Therefore, as the first step, we propose a new framework for removal of VA from atrial electrograms, which is based on interpolation and subtraction followed by low-rank and sparse matrix decomposition. The proposed framework is of low complexity, does not require high resolution multi-channel recordings, or a calibration step for each individual patient. In the next step, we develop a simplified electrogram model. We represent the model in a compact matrix form and show its linear dependence on the conductivity vector, enabling the estimation of this parameter from the recorded electrograms. The results show that despite the low resolution and all simplifying assumptions, the model can efficiently estimate the conductivity map and regenerate realistic electrograms, especially during sinus rhythm. In the next contribution of this dissertation, we propose a new approach for a better estimation of local activation times for atrial mapping by reducing the spatial blurring effect that is inherent to electrogram recordings using deconvolution. Employing sparsity based regularization and first-order time derivatives in formulating the deconvolution problem, improved performance of transmembrane current estimation is obtained. In the final part, we focus on translating our findings from research to clinical application. Therefore, we studied the effect of electrode size on electrogram properties including the length of the block line observed on the resulting activation map, percentage of observed low voltage areas, percentage of electrograms with low maximum steepness, and the number of deflections in the recorded electrograms.

MSc SS Thesis Presentation

• Tuesday, 26 October 2021
• 14:00-16:00
• ZOOM

Temporal Synchronization of Sensors

Advanced automotive vehicles are based on the real-time fusion of an increasing number of automotive sensors. For precise fusion of different sensors, measurements need to be synchronized both temporally and spatially. This thesis aims to design a hardware temporal synchronization block as part of the PRISTINE systolic array accelerator project for multi-sensor data fusion. In this process, we study and address several temporal sensor synchronization issues that are characteristic of the considered system as well as any other typical sensor fusion system. First and foremost, we handle the problem of estimating the actual time of sensor measurement by exploring well-known filtering techniques such as Kalman, mean and median filters. A suitable filter is selected for implementation based on the statistical characteristics of the observed sensor cycle times, the complexity of the filters and the quality of obtained estimates. Next, we address the issue of reconstructing incoming sensor data streams according to the estimated sensor measurement times while maintaining minimal latency and synchronization error by employing an adaptive stream buffering technique utilized in distributed multimedia systems. An analysis of the effects of the stream synchronization algorithm’s parameters on buffering latency and synchronization error was presented. Finally, the above synchronization solution was efficiently implemented on hardware by making certain modifications and design decisions to the algorithm. A method to evaluate the whole temporal synchronization process is proposed and the obtained results on real sensor data are presented.

https://tudelft.zoom.us/j/99844909465?pwd=OEtwbCtuS2xzdTg2Sk5lQzNOdVp2Zz09 Meeting ID: 998 4490 9465 Passcode: 375074

MSc SS Thesis Presentation

• Friday, 22 October 2021
• 10:00-12:00
• EWI building 36, lecture hall H

Adaptive Graph Partition Methods for Structured Graphs

Graphs can be models for many real-world systems, where nodes indicate the entities and edges indicate the pairwise connections in between. In various cases, it is important to detect informative subsets of nodes such that the nodes within the subsets are 'closer' to each other. For example, in a cellular network, determining appropriate node subsets can reduce the operation costs. A subset is usually called a cluster. This leads to the graph clustering problem. Furthermore, plenty of systems in the real world are changing over time, and consequently, graphs as models vary with time as well. It is thus also important to update the clusters when the graph changes.

In this thesis work, we studied two problems from the cellular network background. We needed to partition graphs that have certain structures and cluster their nodes to minimize certain cost functions. In the first problem, we partitioned a bipartite graph by minimizing the so-called MinMaxCut cost function, while in the second problem, we partitioned a structured graph by minimizing the so-called Modified-MinMaxCut cost function. The structural property of the graph is incorporated in defining this new cost function. The solutions we proposed are under the framework of spectral clustering, where one relies on the eigenvectors of the graph matrices, e.g., the Laplacian matrix or the adjacency matrix, and any clustering algorithm, e.g., K-means, to partition nodes into disjoint clusters.

Furthermore, for the time-variant graph, we decomposed the problem into two steps. First, we transformed the variations in the graph topology into perturbations to the graph matrices. Then we transformed the update of the clusters into an update of the (generalized) eigenvectors of these graph matrices. We utilized matrix perturbation theory to update the generalized eigenvectors and then update the clusters. Our simulations showed that on synthetic data, the proposed method can efficiently track the eigenvectors and the clusters generated by the updated eigenvectors have almost the same cost function value as that of exact computation.

https://tudelft.zoom.us/j/95884057579?pwd=YXkzY0lCRVZuRmsvTzl3NW9xRzZ3QT09

CAS MSc Midterm Presentations

• Thursday, 21 October 2021
• 11:00-12:00
• Online [zoom]

Few-Shot Learning Using Speech and Physiological Signals for Emotion Recognition for Intelligent Voice Assistants

CAS MSc Kick-off Presentations

• Thursday, 21 October 2021
• 11:00-12:00
• Online [zoom]

Improving the Accuracy of the Image Search Engine for Digital History

Signal Processing Seminar

• Thursday, 14 October 2021
• 11:00-12:00
• Online [zoom]

Finite Impulse Response Simplicial Filters

CAS MSc Kick-off Presentations

• Thursday, 14 October 2021
• 11:00-12:00
• Online [zoom]

Improving the Estimation of Local Activation Time in Atrial Electrograms

CAS MSc Kick-off Presentations

• Thursday, 14 October 2021
• 11:00-12:00
• Online [zoom]

Removing Far-Field Artifacts from Atrial Electrograms Using Temporal and Spatial Information

MSc SS Thesis Presentation

• Tuesday, 12 October 2021
• 14:00-17:00
• EWI Lecture Hall M

Targetless Camera-LiDAR Calibration for Autonomous Systems

In recent decades, the field of autonomous driving has witnessed rapid development, benefiting from the development of artificial intelligence-related technologies such as machine learning. Autonomous perception in driving is a key challenge, in which multi-sensor fusion is a common feature. Due to the high resolution and rich information, the camera is one of the core perceptual sensor in autonomous systems. However, the camera provides no knowledge on distance (or depth), which is insufficient for the requirements of autonomous driving. On the other hand, LiDAR provides accurate distance measurements, however the information is sparse. The complementary characteristics of cameras and LiDAR have been exploited over the past decade for autonomous navigation. In order to be able to fuse the camera and LiDAR sensor system jointly, an efficient and accurate calibration process between sensors is essential. Conventional methods for calibrating the camera and LIDAR rely on deploying artificial objects, e.g., checkerboard, on the field. Given the impracticality of such solutions, targetless calibration solutions have been proposed over the past years, which require no human intervention and are readily applicable for various autonomous systems, e.g., automotive, drones, rovers, and robots.
In this thesis, we review and analyze several classic targetless calibration schemes. Based on some of their shortcomings, a new multi-feature workflow called MulFEA (Multi-Feature Edge Alignment) is proposed. MulFEA uses the cylindrical projection method to transform the 3D-2D calibration problem into a 2D-2D calibration problem and exploits a variety of LiDAR feature information to supplement the scarce LiDAR point cloud boundaries to achieve higher features similarity compared to camera images. In addition, a feature matching function with a precision factor is designed to improve the smoothness of the objective function solution space and reduce local optima. Our results are validated using the open-source KITTI dataset, and we compare our results with several existing targetless calibration methods. In many different types of roadway environments, our algorithm provides more reliable results regarding the shape of the objective function in the 6-DOF space, which is more conducive for the optimization algorithms to solve. In the end, we also analyze the shortcomings of our proposed solutions and put forward a prospect for future research in the field of joint camera-Lidar calibration algorithms.

This work is part of the EU ADACORSA project co-hosted by the TU Delft CAS group (link: https://cas.tudelft.nl/Research/project.php?id=160&pid=59)

The zoom link for the defense is https://tudelft.zoom.us/j/96202579105

CAS MSc Midterm Presentations

• Thursday, 7 October 2021
• 11:00-12:00
• Online [zoom]

Loudspeakers as Recording Devices in Public Address Systems

Signal Processing Seminar

• Thursday, 30 September 2021
• 11:00-12:00
• Online [zoom]

Information Theoretic Rank Estimation in Dynamic Contrast Enhanced Ultrasound Sequences

CAS MSc Thesis Presentation

• Thursday, 30 September 2021
• 14:00-15:00
• EWI Building 36, lecture hall J

Hybrid Posit and Fixed Point Hardware for Quantized DNN Inference

The recently introduced posit number system was designed as a replacement for IEEE 754 floating point, to alleviate some of its shortcomings. As the number distribution of posits is similar to the data distributions in deep neural networks (DNNs), posits offer a good alternative to fixed point numbers in DNNs: using posits can result in high inference accuracy while using low precision numbers. The number accuracy is most important for the first and last network layers to achieve good performance. For this reason, these are often computed using larger precision fixed point numbers compared to the hidden network layers. Instead, these can be computed using low precision posit, to reduce the memory access energy consumption and the required memory bandwidth. The hidden layer computation can still be performed using cheaper fixed point numbers.
An inference accuracy analysis is performed to quantify what the effect of this approach is on the VGG16 network for the ImageNet image classification task. Using 8 bit posit for the first and last network layer instead of 16 bit fixed point is shown to result in a top-5 accuracy degradation of only 0.24%. The hidden layers are computed using 8 bit fixed point in both cases.
The design of a parameterized systolic array accelerator performing exact accumulation is proposed that can be used in a scale-out system along with fixed point systolic array tiles. To increase hardware utilization, a hybrid posit decoder is designed to enable fixed point computation on the posit hardware. Using this hardware, the entire network can be computed using 8 bit data, instead of using 16 bits for some layers. This reduces energy consumption and the complexity of the memory hierarchy

CAS MSc Midterm Presentations

• Thursday, 23 September 2021
• 11:00-12:00
• Online [zoom]

Forecasting Multidimensional Graph Signals Using Graph Filters

CAS MSc Midterm Presentations

• Thursday, 23 September 2021
• 11:00-12:00
• Online [zoom]

Analyzing Functional Ultrasound Images of the Brain Using Tensor Decompositions

CAS MSc Midterm Presentations

• Thursday, 16 September 2021
• 11:00-12:00
• Online [zoom]

A Physiological Model for Heart Rate Analysis

CAS MSc Midterm Presentations

• Thursday, 16 September 2021
• 11:00-12:00
• Online [zoom]

The Area of a Unipolar Electrogram to Identify the Arrhythmogenic Substrate

CAS MSc Thesis Presentation

• Thursday, 16 September 2021
• 10:00-11:30
• ZOOM

Sound Zones with a Cost Function based on Human Hearing

With the aid of an array of loudspeakers, sound zone algorithms seek to reproduce multiple distinct zones of audio inside an enclosure. Typical approaches determine the loudspeaker inputs by optimizing over a cost function that models the sound pressure inside the enclosure. However, recent methods propose cost functions that include a perceptual model of the human auditory system, which further models the perception of sound. This thesis investigates such an approach by proposing a framework within which sound zones are constructed through optimization over a perceptual model. The framework is used to propose two perceptual sound zone algorithms: unconstrained and constrained perceptual pressure matching. Simulations of the proposed algorithms and a reference algorithm are presented to determine the benefits of including auditory-perceptual information in sound zone algorithms. From this, it is found that the unconstrained perceptual approach outperforms the reference in terms of various perceptual measures. In addition, it is found that adding perceptual constraints to the optimization problem allows for control of sound zones which correlates well with other perceptual quality measures.

PhD Thesis Defence

• Monday, 13 September 2021
• 14:30
• Aula Senaatszaal

In-pixel temperature sensors for dark current compensation of a CMOS image sensor

This thesis describes the integration of temperature sensors into a CMOS image sensor (CIS). The temperature sensors provide the in-situ temperature of the pixels as well as the thermal distribution of the pixel array. The temperature and the thermal distribution are intended to be used to compensate for dark current affecting the CIS. Two different types of in-pixel temperature sensors have been explored. The first type of temperature sensor is based on a substrate parasitic bipolar junction transistor (BJT). The second type of temperature sensor that has been explored is based on the nMOS source follower (SF) transistor of the same pixel. The readout system that is used for the temperature sensors and for the image pixels is based on low noise column amplifiers. Both types of in-pixel temperature sensors (IPTS) have been designed implementing different techniques to improve their accuracy. The use of the IPTSs has been proved by measuring three prototypes chips. Also, a novel technique to compensate for the dark current of a CIS by using the IPTS has been proposed.

For those who cannot attend, you can follow it by using this link:
https://collegerama.tudelft.nl/mediasite/play/84dc9775142d44db81aa38e5532c67ca1d

MSc SS Thesis Presentation

• Monday, 6 September 2021
• 14:30-15:15

Time Synchronization for Anchorless Satellite Networks

In this thesis, we propose a new class of pairwise frequency and multi-domain time synchronization and ranging algorithms and in a case study address network and mission level aspects of time synchronization on the example of the Orbiting Low Frequency Array for Radio astronomy (OLFAR), a proposed distributed radio interferometer.

The new class of frequency and multi-domain time synchronization and ranging algorithms proposed is applicable to anchorless mobile networks of asynchronous nodes. In a first step, the Frequency Pairwise Least Squares (FPLS) that estimates clock skew and relative velocity in a pairwise setup using only frequency measurements is formulated. In a second step, we extend this method to a motion model with constant acceleration. As frequency domain methods do not estimate clock offset and pairwise range, relying purely on frequency domain estimates is not feasible for most applications.

To harness the potential of frequency domain synchronization and ranging, the Combined Pairwise Least Squares (CPLS) has been developed. The combined method reduces the number of minimum required messages from 4 to 3 compared to current methods and decreases the computational complexity. Using a generic simulation with nodes in pairwise non-linear motion, we show that frequency domain methods can outperform time domain methods in clock skew and relative velocity estimation and that the proposed multi-domain method delivers better clock offset and pairwise range estimation in low to medium SNR conditions.

In the second part of our work, we apply the novel methods to OLFAR –— a spaceborne large aperture radio interferometric array platform. We address network level and mission level aspects, proposing network path planning for pairwise synchronization algorithms and determining the required resynchronization period. i

MSc SS Thesis Presentation

• Monday, 6 September 2021
• 10:00-11:30

On the Integration of Acoustics and LiDAR

Loudspeakers are placed in an environment unknown to the loudspeaker designers. The room influences the acoustic experience for the user. Having information about the room makes it possible to better reproduce the sound field as intended. Using microphone measurements, the location of acoustic reflectors can be inferred. Current state-of-the-art methods for room boundary detection focus on a two-dimensional setting. Detection of arbitrary reflectors in three dimensions increase complexity due to practical limitations, i.e. the need for a spherical array and the increase of computational complexity. The presence of horizontal reflectors cause inaccuracy for wall detection due to model mismatch. Loudspeakers may not present an omnidirectional directivity pattern, as usually assumed in the literature, thus making the detection of acoustic reflectors in some directions more challenging.

In this thesis, a LiDAR sensor is added to a smart loudspeaker to improve wall detection accuracy and robustness. This is done in two ways. First, the horizontal reflectors that are not present in the acoustic model are sought detected with the LiDAR sensor to enable elimination of their detrimental influence. Second, a method is proposed to compensate for the challenging regions for wall detection in highly directive loudspeakers, using the LiDAR sensor. Experimental results, evaluated in different simulated scenarios are shown for comparison of the proposed method and the state-of-the-art method, that exclusively uses acoustic information.

MSc SS Thesis Presentation

• Wednesday, 25 August 2021
• 10:00-10:45
• zoom

Energy-efficient Particle Filter SLAM for Autonomous Exploration

Autonomous robots are increasingly used in more and more applications, such as warehouse robots, search-and-rescue robots and autonomous vacuum cleaners. These applications are often in environments where the GPS signals are denied or inaccurate, which makes it difficult to localize the robot in an unknown environment. To overcome this problem the framework of Simultaneous Localization and Mapping (SLAM) is typically used. This solution constructs a map of the environment with the use of cameras or range sensors, while keeping track of the location of the robot in it. To extend the exploration time of these battery powered robots, the energy consumption of the SLAM algorithm could be reduced. It is assumed that if the computational load of an algorithm reduces, the energy consumption of the algorithm reduces as well. An existing paradigm to solve SLAM is the use of a particle filter, which tracks the trajectory of the robot and simultaneously maps the environment. The question answered in this thesis is how to make this algorithm more energy-efficient to be able to deploy this framework in more applications and make the existing robots more sustainable.

In this thesis two methods are investigated. In the first method, the information about the landmarks are incorporated in the trajectory estimation as spatial constraints, to try to achieve a higher accuracy with less particles and thus subsequently a smaller computational load. The proposed method is validated by simulations on synthetic datasets. This method shows improvements in terms of the estimation accuracy. However, it is more computational complex than the existing algorithms, so it is considered less energy-efficient. The second method in this thesis, is the implementation of a parallelized particle filter. This method processes the observation measurements in parallel for the different particles and communicates the information between the particles efficiently. It should reduce the computational time, to enable partial computation of the algorithm to reduce the computational load. This method shows improvements on the run time and thus on the computational load, especially for a larger number of particles and is therefore more energy-efficient. The two separate methods have been analyzed and compared with state of the art methods.

MSc SS Thesis Presentation

• Tuesday, 24 August 2021
• 10:00-10:45
• zoom

Quantifying the dynamic interactions between physiological signals to predict the exposure from chemicals

Causal inference is a familiar topic in biomedical research and a key concept in the study of connectivity in various physiological systems. This work aimed to analyse the coupling between the beat to beat parameters derived from ECG and respiration. It was the first time such an analysis was carried out in the context of finding the differences caused by chemical's exposure.

We used conditional Granger causality, a popular method to evaluate direct causal relationships. We have incorporated the cardinality constraint in the optimization function of Granger causality (GC) to deal with the high dimensionality challenge. Further, we extended the original formulation GC to evaluate the coupling between two unequally sampled signals. Finally, end to end implementation of the machine learning prediction model using causal features is well illustrated.

We found a consistent decrease in the average coupling strength of breathing parameters after the exposure. But in the case of ECG interactions, no noticeable change was observed. Surprisingly we found no significant links between the ECG and breathing parameters. The support vector machine (SVM) and random forest trained on coupling values differentiate between healthy and exposure samples. The accuracy of trained SVM and random forest on the independent test set were 78 % and 75 %, respectively.

MSc SS Thesis Presentation

• Friday, 9 July 2021
• 10:00-10:35
• zoom

Edge State Kalman Filtering for Distributed Formation Control Systems

Formation control problems consider a set of mobile agents with the underlying goal of attaining and maintaining a state where the relative positions of agents are stable in accordance with the desired configuration. Navigation for formation control is typically achieved through localization in a global reference frame, e.g., via GNSS. However, when a global reference frame is not shared among agents, a relative navigation approach is required.

Distributed filtering for relative localization in formation control systems is a relatively unexplored field. The absence of absolute positioning means motivates the need for a distributed filter that operates on the edges of the sensing graph of the multi-agent system. In this thesis, a data model for relative formation control problems and two edge-based Kalman filters are proposed. The first filter is designed for an individual edge. The second is a filter designed via decoupling of the optimal global filter which allows for the joint estimation of adjacent edges. It is shown that the joint filter is optimal under the decoupling constraints.

Monte Carlo results show that when random environmental disturbances are correlated among agents, the joint filter outperforms the local edge filter in a mean square error sense.

Lastly, systems are considered where inter-agent communications are unavailable, leading to biased prediction steps of the Kalman filters. We aim to minimize this effect through the proposal of a local Wiener filter which predicts the control actions of neighboring agents.

PhD Thesis Defence

• Friday, 25 June 2021
• 15:00-16:30
• Aula Senaatszaal

Graph filter designs and implementations

MSc SS Thesis Presentation

• Thursday, 17 June 2021
• 09:00-09:45
• zoom

SPLITTER

A data model and algorithm for detecting spectral lines and continuum emission of highredshift galaxies using DESHIMA 2.0

MSc SS Thesis Presentation

• Wednesday, 26 May 2021
• 10:00-10:45
• zoom

GNSS Chirp Interference Estimation and Mitigation