Computational Imaging for Dynamic Metasurface based Synthetic Aperture Radars

Abstract

In this thesis, computational imaging is used to improve millimeter-wave dynamic metasurface based synthetic aperture radar (SAR) imaging. SAR image reconstruction is a computationally complex inverse problem, which can be solved via a generalized inverse [e.g. full pseudoinverse (FPI)], or by means of approximation [e.g. matched filter (MF)]. However, approximation limits the resolution. In the first part of the thesis, a computationally efficient partitioned inverse (PI) algorithm for SAR image reconstruction is proposed. The independent partitions can be reconstructed in parallel using the matched filter or the pseudo inverse. A parallelized approach leveraging a graphics processing unit (GPU) is used to achieve a dramatic speedup compared to non-GPU accelerated algorithms. In the results presented in this thesis, PI algorithms are three orders magnitude faster than the FPI and two orders faster than the MF. The improved resolution of FPI is maintained by using the pseudo inverse to find the solutions for the PI. The partitioned pseudo inverse (PPI) algorithm is based on the Moore-Penrose pseudo inverse using truncated singular value decomposition for regularization. Optimal regularization ensures that the algorithm is robust to noise. It is shown that the PPI has an improved resolution of 24% over MF even at a signal to noise ratio (SNR) of 0 dB. Experimental results using a laboratory K-band (15-26.5 GHz) ultra-wideband SAR system are presented to validate the PI algorithms. The second part of the thesis focuses on SAR imaging with dynamic metasurface antennas (DMAs). An enhanced resolution stripmap mode (ERSM) SAR approach is extended for 3D imaging using DMA, where PI algorithms are used to accelarate the reconstruction. Experimental results using a commercial prototype DMA are presented. It is shown that the cross-range resolution and ground-range resolution has improved by 26% and 42% respectively when using 3D ERSM SAR. Furthermore, the PI reconstruction resulted in a 23.5X speedup over the MF. A limitation of using DMAs for SAR imaging are strong side lobes of the antenna, which causes the images to be smeared due to a distorted point spread function (PSF). It is shown that Lucy-Richardson deconvolution using a measured PSF can be applied to correct the smearing, achieve high resolution, and restore the image. Experimental results using a liquid-crystal based DMA for near field SAR imaging are presented, where the reconstructed images are restored using Lucy-Richardson deconvolution.