We present novel realizations of computational terahertz (THz) imaging techniques based on compressive sensing and phase-retrieval algorithms and a single-pixel THz sensor. Imaging in the THz band covering 300 GHz-10 THz is being considered for key applications in biomedical imaging, security screening and non-destructive evaluation. State-of-the-art in THz imaging is based on mechanical raster scanning using a single, high-performance sensor. Such raster-scanning imagers are rather bulky and suffer from very low frame-rates, as well as mechanical noise due to the moving parts in the hardware. Alternatively, multi-detector imagers such as THz focal plane arrays (FPAs) can speed-up image acquisition time, potentially reaching real-time video rates. However, such devices require complex and expensive fabrication and they typically exhibit limited sensitivity due to additional noise introduced by the read-out circuit.
In this dissertation, we demonstrate novel THz imaging techniques based on a single THz sensor that concurrently circumvent the slow acquisition time and mechanical noise of raster scan imaging. This is achieved by using an optically reconfigurable spatial wave modulation scheme to "serialize'' the scene measurements. Subsequently, compressive sensing (CS) and reconstruction algorithms are employed to computationally generate 2D images of the scene from a set of serial measurements, each corresponding to a different spatial modulation. Similar to well-developed optical CS methods, compressive THz imaging allows far fewer measurements than the conventional Nyquist rate to accurately reconstruct sparse scenes. In addition, compressive THz reconstruction exhibits better signal-to-noise (SNR) performance compared to the FPA cameras.
To enable the study and experimental demonstration of various computational imaging algorithms, we realize a generalized compressive THz imaging setup using conventional quasi-optical components and a semiconductor-based photoconductive spatial wave modulator. In particular, the spatial modulator consists of a high-resistivity N-type 420 μm-thick Silicon wafer illuminated by computer-generated pixelated patterns or commonly called "masks". The masks are projected on the Si wafer via a commercially-available LCD projector. The selective illumination of localized regions on the Si wafer creates conductive islands that obstruct the transmission of the THz signal from the scene. As such, we individually control the transmission of object THz beam through each pixel via Si photoconductivity. For each spatial modulator configuration, we record the Fourier-domain signal using a single THz detector. Subsequently, we reconstruct high-dimensional and high SNR intensity images of the scene applying various CS algorithms. In particular, we demonstrate reconstruction of 64×64-pixel THz images with high fidelity using measurements as few as 25% of the Nyquist rate (i.e., 1,024 measurements) at 690 GHz. Furthermore, we show that the acquisition speed can be increased further by employing multiple sensors for data collection. Specifically, using two-sensor measurements, we demonstrate that the number of masks and hence, the acquisition time is halved while achieving similar reconstruction performance compared to single-pixel approach. In both methods, the finest resolution we achieved was 380 μm.
Using the general THz CS hardware, we demonstrate, for the first time, phase-sensitive image reconstruction using intensity-only measurement data and phase-retrieval algorithms. In particular, the "Phaselift'' and "Reshaped Wirtinger Flow'' algorithms are demonstrated to recover the phase of the scene by recording an oversampled set of spatially modulated, intensity-only measurements at 690 GHz. We demonstrate that the phase reconstruction produce superior resolution and contrast, and provides additional information about the object, compared to traditional intensity-only reconstruction. Motivated by this technique, we also develop a novel, free-of-motion mmW and THz antenna radiation pattern characterization system.
We also demonstrate, for the first time, reconstruction of THz images by recording only the sign of the measurements and applying single-bit CS. Analyzing some unique noise robustness characteristics through experiments, we show that single-bit CS allows faster data acquisition and performs much better that the conventional CS under low SNR illumination. These features along with single-bit recording enable use of low-cost source-detector and acquisition hardware for THz imaging.
Finally, we demonstrate a single-pixel THz imaging system using rail-based synthetic aperture radar (SAR) technique. Scanning the object in azimuth, we synthesize an aperture larger than the beamwidth, using conventional SAR reconstruction. We show image reconstruction of objects concealed within dielectric materials, achieving resolution as fine as 1 mm in down-range for a measurement bandwidth of 500-750 GHz. Additionally, the acquisition speed is much faster than a traditional raster scanning system.