Holograms with neural-network backend for submillimeter-wave beamforming applications

Abstract

We present a new method to carry out localization based on distributed beamforming and neural networks. A highly dispersive hologram, is used together with a terahertz spectrometer to localize a corner-cube reflector placed in the region of interest. The transmission-type dielectric hologram transforms input pulse from the spectrometer into a complex pattern. The hologram causes complicated propagation paths which introduce delay so that different parts of the region of interest are interrogated in a unique way. We have simulated the emitted pulses propagating through the hologram. The hybrid simulation combines the finite-difference and physical optics methods in time domain and allows for evaluating the dispersion and directive properties of the hologram. The dispersive structure is manufactured of Rexolite and it has details resulting in varying delay from 1 to 19 wavelengths across the considered bandwidth. The spectrometer is configured in reflection mode with wavelets passing in to the region of interest through the hologram. A data-collecting campaign with a corner-cube reflector is carried out. The effective bandwidth for the localization is from 0.1 THz to 2.1 THz, and the measured loss is 57 dB at minimum. The collected data is used to train a fully-connected deep neural network with the known corner-cube positions as labels. Our first experimental results show that it is possible to predict the position of a reflective target in the region of interest. The accuracy of the prediction is 0.5-0.8 mm at a distance of 0.17 m.