Real-time signal processing algorithms for computational millimetre-wave radars

Abstract

Radar imaging using microwave and millimetre (mmW) frequencies has gained significant popularity for a wide range of applications. The corresponding wavelengths have been proven to successfully penetrate most optically opaque materials and work well in all weather conditions. At the same time, the non-ionizing nature of microwaves and mmW is safe for the human body and has played a significant role in establishing this band of frequencies, particularly for applications such as the security screening or medical imaging. However, fundamentally, microwave and mmW systems exhibit a poor resolution due to the relatively large wavelengths comparing to optics. One way to mitigate this limitation is by increasing the aperture size, which for applications involving large target scenes is intuitively essential to accommodate the scanning of the whole body. Therefore, the challenge is to design electrically large apertures that are also cost-effective and fast in terms of data acquisition time. Metasurface antennas equipped with computational imaging (CI) techniques have recently emerged as a strong candidate to simplify the hardware and reduce the cost of the system. The elements of a metasurface antenna can be crafted, either by a simple frequency sweep or by applying an external voltage stimulus, to produce a set of spatially distinct radiation patterns. This can be achieved without the need for any mechanical parts or additional electronic circuits. In addition, the hardware layer is further simplified using the aperture transfer function to encode the information from the scene into one channel.

Considering all the hardware relaxations offered by CI metasurface antennas and the fast acquisition time that is reassured, the question that arises is what happens to the signal processing layer. A signal processing algorithm that can provide an estimation of the scene in real-time (in a few seconds) is essential to design a real-time system in terms of both software and hardware. The most popular algorithms in the literature so far to deal with CI problems rely on the computation and inversion of the sensing matrix, which is basically the transfer function of the aperture. As the target scene becomes larger and more complicated, these methods do not scale well. On the other hand, algorithms in the spatial frequency domain that are realized through Fourier transformations and can be executed fast without the need of heavy computational resources. However, due to the compressed nature of the back-scattered signal in the case of a CI system, those algorithms are not directly applicable. However, recently, some progress has been made in utilizing spatial frequency domain algorithms for CI systems employing a form of signal decompression. The goal of this thesis is to, first, study the most popular algorithms for reconstructing the image for CI and non-CI systems, particularly focusing on the spatial frequency domain solution due to low complexity and fast execution. Secondly, this thesis aims to develop further signal processing algorithms in the spatial-frequency domain to address the image reconstruction for multiple-input multiple-output (MIMO) CI systems consisting of multiple metasurface panels.

The first part of the thesis is devoted to developing numerical models for conventional synthetic aperture radar (SAR) systems and using the popular spatial frequency domain range migration algorithm (RMA) to obtain an estimation of the target scene. Experimental measurements taken in the electromagnetic (EM) Imaging and Sensing Lab at the Centre for Wireless Innovation have been used to verify the developed techniques. Further speed-up of the developed algorithms is investigated by using a single general purpose graphical processing unit (GPGPU).

Additionally, in the first part, large conventional MIMO apertures are also explored for imaging in the near-field. The RMA algorithm is successfully adapted for this case, employing an additional interpolation step.

The second part of the thesis is devoted to CI realized by dynamic metasurface antennas (DMAs). A stationary MIMO DMA-based aperture is proposed to perform three-dimensional (3D) imaging in the near-field. Having reassured the advantages of such a configuration from the hardware perspective, the thesis focuses on the development of a signal processing algorithm that can address the image reconstruction step in real-time. The challenge is that due to the compressed nature of the signal captured by DMA panels, the RMA cannot be applied directly. To address this challenge, a signal decompression step suitable for MIMO CI systems is derived to transform the signal in a conventional multistatic form. Then the RMA can be used to obtain a final estimation of the target. First, the developed algorithm is validated using numerical results from the mathematical models. A resolution target is mostly used and the comparison with traditional sensing matrix reconstruction techniques is presented in terms of image quality and reconstruction time, demonstrating the superiority of the proposed technique is terms of execution time. The possibility of scaling the proposed system to synthesize an electrically large MIMO DMA-based aperture is discussed and verified through numerical results.

The last part of the thesis includes the experimental system built in the EM Imaging and Sensing Lab at the Centre for Wireless Innovation.

Thesis is embargoed until 31 July 2025.

Date of Award	Jul 2024
Original language	English
Awarding Institution	Queen's University Belfast
Sponsors	The Leverhulme Trust
Supervisor	Okan Yurduseven (Supervisor) & Hien-Quoc Ngo (Supervisor)