Complex-valued neural networks for fully-temporal micro-Doppler classification

Abstract

Micro-Doppler analysis commonly makes use of the log-scaled, real-valued spectrogram, and recent work involving deep learning architectures for classification are no exception. Some works in neighboring fields of research directly exploit the raw temporal signal, but do not handle complex numbers, which are inherent to radar IQ signals. In this paper, we propose a complex-valued, fully temporal neural network which simultaneously exploits the raw signal and the spectrogram by introducing a Fourier-like layer suitable to deep architectures. We show improved results under certain conditions on synthetic radar data compared to a real-valued counterpart.

A frequency domain method for scattering problems with moving boundaries

“…In the automotive industry, micro-Doppler sensing has recently been proposed [23,31,36,38] for the contactless detection of vital signs like breathing of infants left alone on the back seat of overheating cars. The engineering of such very high frequency radar sensing devices entails dealing with multiple challenges, as for instance the analysis of random body movements and vehicle vibrations [23,31,38,39,40,43], leading to radar signatures that can be classified for example by deep learning techniques [7,8,18]. In order to design these new sensors, an adequate full realistic simulation of the underlying high frequency scenarios is crucial.…”

Radar sensing for human healthcare: challenges and results

“…However, radar data can span a variety of formats beyond simple 2D spectrograms, across the multidimensional space of range/distance, Doppler/velocity, time, and angular information in elevation or azimuth. Additionally, unlike optical images, radar data are complex, hence representable in real-imaginary or absolute value & phase planes [16][17][18]. Identifying the most suitable radar data representation for the specific classification problem and to be paired with specific classification algorithm or neural network remains an interesting open question.…”

A MultiHarmonic Finite Element Method for Scattering Problems with Small-Amplitude Boundary Deformations

“…of glass or synthetic materials. This is modelled by the zeroth-order absorbing boundary condition (3). Therefore, within our notations, Ω 0 is the domain with boundary Γ…”

“…In the automotive industry, micro-Doppler sensing has recently been applied with success [17,24,27,29] to the contactless detection of vital signs, in particular for the breathing of children left alone on the back seat of overheating cars. Difficult challenges are then related to this kind of applications as for example the analysis of random body movements or vehicle vibrations [17,24,29,31,32,33] that can be classified thanks to their radar signature by deep learning techniques [3,4,14]. In the development life cycle of these new sensors, a full realistic simulation of the high frequency scenarios is therefore needed.…”