A)Deep Learning for Embedded Automotive Platforms:
The goal is to:
1) Research semi-supervised training methods in order to make neural networks more resilient to adverse imaging conditions, causing the performance drop (wrt. nominal imaging conditions) to decrease with one order of magnitude.
2) Research inference methods that exploit spatio-temporal coherency in video streams, to increase computational efficiency by one order of magnitude.
All research results will be validated in real-world conditions using TU/e’s Highly Automated Driving research vehicle. NXP will make available its Bluebox automotive computation platform.
B)Spread-spectrum Modulated And interfeRence resilienT RADAR:
To develop novel nm CMOS 77-81GHz radars frontends that generate ultra-fast arbitrary radar waveforms that alleviate radar interference through spread spectrum techniques. To exploit smartness concepts in the front-ends such as instantaneous receive signal analysis, self-interference cancellation using known transmit signal properties, etc. To design breakthrough performance building blocks such as beamforming receivers and GHz data converters that support reception of high rate long encrypt bitstreams for coding orthogonality with increased range and velocity resolution. Concepts will be demonstrate on NXP’s modulated radar demonstration platforms that will be made compatible and available for the test chips of this project.
Contact Gijs Dubbelman I Project leader Deep Learning for Embedded Automotive Platforms Marion Matters I Project leader Spread-spectrum Modulated And interfeRence resilienT RADAR