Summary
The sophisticated electronic devices ubiquitous today are largely enabled by conversion of analog signals to digital through digital signal processing (DSP). DSP is also the enabler behind the deep learning revolution, which led to systems with unprecedented performance. However, DSP comes at a cost. Internet traffic has exceeded Zettabytes. The conventional approach to DSP and machine learning is to acquire all data possible and build systems that are capable of massive training and parallel processing in order to extract the relevant information out of huge data sets. Acquiring, storing, processing and communicating pervasive massive datasets generated at an ever-increasing speed requires huge amounts of power, hardware, and physical space.
In this proposal, we put forth an ambitious goal: introducing a paradigm shift in acquisition and learning to enable systems capable of low-rate, low-power, wideband sensing, efficient training, super-resolution, and flexible operations. Our approach is based on acquiring and processing only the information needed for the required task. We achieve this goal by introducing the new concept of jointly designing all system components, where the underlying structure in the data and the specific required task are taken into account throughout the design. Realizing this new scientific frontier will entail (1) developing new theoretical frameworks of sampling theory, information theory, learning theory and quantization theory of sub-sampled channels where sampling and information are treated jointly and depend on the system task; this will form the basis for (2) the development of novel mixed analog-digital hardware prototypes that enable new technology paradigms, such as wireless ultrasound systems and joint radar-communication platforms, and (3) a broad-range of scientific and clinical applications. A successful outcome can revolutionize the way signals are acquired and processed and have a profound impact on multitude applications.
In this proposal, we put forth an ambitious goal: introducing a paradigm shift in acquisition and learning to enable systems capable of low-rate, low-power, wideband sensing, efficient training, super-resolution, and flexible operations. Our approach is based on acquiring and processing only the information needed for the required task. We achieve this goal by introducing the new concept of jointly designing all system components, where the underlying structure in the data and the specific required task are taken into account throughout the design. Realizing this new scientific frontier will entail (1) developing new theoretical frameworks of sampling theory, information theory, learning theory and quantization theory of sub-sampled channels where sampling and information are treated jointly and depend on the system task; this will form the basis for (2) the development of novel mixed analog-digital hardware prototypes that enable new technology paradigms, such as wireless ultrasound systems and joint radar-communication platforms, and (3) a broad-range of scientific and clinical applications. A successful outcome can revolutionize the way signals are acquired and processed and have a profound impact on multitude applications.
Unfold all
/
Fold all
More information & hyperlinks
| Web resources: | https://cordis.europa.eu/project/id/101000967 |
| Start date: | 01-07-2021 |
| End date: | 30-06-2026 |
| Total budget - Public funding: | 3 000 000,00 Euro - 3 000 000,00 Euro |
Cordis data
Original description
The sophisticated electronic devices ubiquitous today are largely enabled by conversion of analog signals to digital through digital signal processing (DSP). DSP is also the enabler behind the deep learning revolution, which led to systems with unprecedented performance. However, DSP comes at a cost. Internet traffic has exceeded Zettabytes. The conventional approach to DSP and machine learning is to acquire all data possible and build systems that are capable of massive training and parallel processing in order to extract the relevant information out of huge data sets. Acquiring, storing, processing and communicating pervasive massive datasets generated at an ever-increasing speed requires huge amounts of power, hardware, and physical space.In this proposal, we put forth an ambitious goal: introducing a paradigm shift in acquisition and learning to enable systems capable of low-rate, low-power, wideband sensing, efficient training, super-resolution, and flexible operations. Our approach is based on acquiring and processing only the information needed for the required task. We achieve this goal by introducing the new concept of jointly designing all system components, where the underlying structure in the data and the specific required task are taken into account throughout the design. Realizing this new scientific frontier will entail (1) developing new theoretical frameworks of sampling theory, information theory, learning theory and quantization theory of sub-sampled channels where sampling and information are treated jointly and depend on the system task; this will form the basis for (2) the development of novel mixed analog-digital hardware prototypes that enable new technology paradigms, such as wireless ultrasound systems and joint radar-communication platforms, and (3) a broad-range of scientific and clinical applications. A successful outcome can revolutionize the way signals are acquired and processed and have a profound impact on multitude applications.
Status
SIGNEDCall topic
ERC-2020-COGUpdate Date
27-04-2024
Images
No images available.
Geographical location(s)
