This thesis explores the design, fabrication, characterization, and performance of frequency-diverse antennas and frequency-scanning notch-beam antennas for high-resolution radar detection and imaging at sub-THz frequencies ranging from 220 to 330 GHz. 

Utilizing cutting-edge silicon micromachining techniques, the body of work presents innovative solutions that address the challenges of high-resolution radar imaging, focusing on improving imaging performance, reducing hardware complexity, and enhancing signal processing techniques.

At the heart of this research is analyzing, characterizing, and evaluating a minimalistic beam-shape switching frequency-scanning notch and broad-beam radar systems. A two-antenna system was designed to switch between broad and notch beam patterns, improving imaging resolution and hardware efficiency. The system's ability to operate with fewer components while maintaining high performance was further augmented through advanced signal processing algorithms like TwIST and MUSIC, which offered superior image reconstruction in noisy scenarios.

A key aspect of this research was the experimental validation of the scanning notch antenna, which demonstrated its capability for sub-THz imaging. This work evaluated various algorithms, including FISTA, MUSIC, and matched filter methods, highlighting MUSIC's limitations when multiple targets were present. To overcome the challenge of multi-target scenario failure, a novel adaptive IFFT range-gating method was introduced, which markedly enhanced the radar’s ability to distinguish closely spaced targets by separating the return signals more effectively.

Further advancing in sub-THz imaging systems, we explored designing and fabricating a Mills-cross frequency-diverse antenna using slot radiators and a direct waveguide feed network. This configuration allows for efficient radiation pattern diversity over the operating bandwidth, contributing to enhanced imaging resolution without complex mechanical scanning or phase shifters. 

Following this, we developed a more advanced wideband frequency-diverse antenna for 220 to 330 GHz, featuring an array of silicon-micromachined cross-slot radiators. The frequency-diverse antenna using a cross-slot is incorporated with direct and distributed feed networks. This configuration, which utilizes cross-slot radiators and a distributed feed network, significantly improved radiation pattern diversity and imaging resolution. The performance of these frequency-diverse antennas was evaluated using advanced imaging algorithms, with FISTA and CoSaMP emerging as the preferred algorithms for efficient, high-resolution image reconstruction under various noise conditions. These antennas are designed for short-range, high-resolution imaging and eliminate the need for phase shifters or mechanical scanning, achieving diverse radiation patterns across a broad frequency range.

Finally, this work culminated in a detailed investigation of imaging performance using these frequency-diverse antennas, where the comparison of direct and distributed feed networks provided key insights into optimizing feed designs for enhanced imaging quality and spatial resolution. Furthermore, we investigated the influence of sparse data collection on imaging performance, considering sparsity in three major areas: 1) when the imaging antenna array is sparsely populated, 2) when sparse frequency sampling is applied across the total available bandwidth, and 3) when the bandwidth is divided among multiple transmitters, each operating over a partial bandwidth, while the receiver utilizes the full bandwidth. 

These scenarios provided a comprehensive understanding of how sparsity affects overall imaging performance and resolution, enabling more efficient data acquisition without compromising image quality.

This thesis substantially contributes to advancing sub-THz radar detection and computational imaging. By integrating innovative antenna designs, adaptive signal processing techniques, and advanced fabrication methods, this research presents a comprehensive solution for achieving high-resolution radar imaging with minimal hardware complexity, paving the way for practical applications in security, medical diagnostics, and structural monitoring.

Denna avhandling utforskar design, tillverkning, karakterisering och prestanda hos frekvensdiverse antenner och frekvensskannande notchstråleantenner för högupplöst radardetektion och avbildning vid sub-THz-frekvenser i intervallet 220 till 330 GHz.

Genom att använda banbrytande kiselmikromaskineringstekniker presenterar arbetet innovativa lösningar som adresserar utmaningar inom högupplöst radaravbildning, med fokus på att förbättra avbildningsprestanda, minska hårdvarukomplexiteten och stärka signalbehandlingstekniker.

Kärnan i denna forskning är att analysera, karakterisera och utvärdera minimalistiska system med strålformsväxling i frekvensskannande notch- och bredstråleradarsystem. Ett system med två antenner designades för att växla mellan breda och smala strålmönster, vilket förbättrade avbildningsupplösning och hårdvarueffektivitet. Systemets förmåga att arbeta med färre komponenter samtidigt som hög prestanda bibehålls, förstärktes ytterligare genom avancerade signalbehandlingsalgoritmer som TwIST och MUSIC, vilka erbjöd överlägsen bildrekonstruktion i bullriga miljöer.

En viktig aspekt av denna forskning var experimentell validering av den skannande notch-antennen, som visade sin kapacitet för sub-THz-avbildning. Arbetet utvärderade olika algoritmer, inklusive FISTA, MUSIC och matched filter-metoder, och framhöll MUSICbegränsningar när flera mål var närvarande. För att övervinna utmaningen i scenarion med flera mål introducerades en ny adaptiv IFFT-range-gating-metod, som avsevärt förbättrade radarens förmåga att särskilja tätt placerade mål genom att effektivare separera de mottagna signalerna.

Vidareutveckling inom sub-THz-avbildningssystem utforskades genom design och tillverkning av en Mills-cross frekvensdivers antenn med slot-radiatorer och en direktvågledarförsörjningsnät. Denna konfiguration möjliggör effektiv diversitet i strålningsmönster över arbetsbandbredden, vilket bidrar till förbättrad avbildningsupplösning utan komplex mekanisk skanning eller fasskiftare.

Därefter utvecklade vi en mer avancerad bredbands frekvensdivers antenn för 220 till 330 GHz, med en array av kiselmikromaskinerade korsslitsradiatorer. Den frekvensdiversa antennen med korsslits kombineras med direkt och distribuerat matningsnätverk. Denna konfiguration, som använder korsslitsradiatorer och ett distribuerat matningsnätverk, förbättrade avsevärt strålningsmönsterdiversiteten och avbildningsupplösningen. Prestandan hos dessa frekvensdiverse antenner utvärderades med avancerade avbildningsalgoritmer, där FISTA och CoSaMP visade sig vara de föredragna algoritmerna för effektiv, högupplöst bildrekonstruktion under olika brusförhållanden. Dessa antenner är utformade för kortdistans och högupplöst avbildning och eliminerar behovet av fasskiftare eller mekanisk skanning, vilket uppnår olika strålningsmönster över ett brett frekvensområde.

Slutligen resulterade detta arbete i en detaljerad undersökning av avbildningsprestanda med dessa frekvensdiverse antenner, där jämförelsen av direkt och distribuerade matningsnätverk gav viktiga insikter för att optimera matningsdesigner för förbättrad bildkvalitet och rumslig upplösning. Dessutom undersökte vi inverkan av gles datainsamling på avbildningsprestanda, med hänsyn till gleshet i tre huvudsakliga områden: 1) när antennarrayen för avbildning är glesbefolkad, 2) när gles frekvenssampling appliceras över den tillgängliga bandbredden, och 3) när bandbredden delas mellan flera sändare, var och en som arbetar över en del av bandbredden, medan mottagaren använder hela bandbredden.

Dessa scenarier gav en omfattande förståelse för hur gleshet påverkar den övergripande avbildningsprestandan och upplösningen, vilket möjliggör effektivare datainsamling utan att kompromissa med bildkvaliteten.

Denna avhandling bidrar avsevärt till att driva fram sub-THz radardetektion och datorbaserad avbildning. Genom att integrera innovativa antenndesigner, adaptiva signalbehandlingstekniker och avancerade tillverkningsmetoder presenterar denna forskning en heltäckande lösning för att uppnå högupplöst radaravbildning med minimal hårdvarukomplexitet, vilket banar väg för praktiska tillämpningar inom säkerhet, medicinsk diagnostik och strukturell övervakning.

 This paper investigates the performance of a frequency-sweeping steering notch beam implemented with a simple two-element antenna array operating within the frequency range of 238 to 248 GHz. We assess the performance of this minimalistic scanning notch beam antenna array for three single-target measurement scenarios, comparing the effectiveness of different algorithms, including IFFT, FISTA, and MUSIC. The results demonstrate that the MUSIC algorithm achieves a range resolution of 13.5 mm, surpassing that of both FISTA and IFFT algorithms. Additionally, the steering notch exhibits superior angular resolution, with a resolution better than 9◦ compared to MUSIC and FISTA. This study explores the potential capabilities of this minimalistic radar and optimized signal processing techniques to reduce hardware complexity in radar systems and address evolving demands for cost-effective radar applications. 

This research investigates a unique radar concept based on a frequency-sweeping stepped-frequency continuous-wave (SFCW) radar with a two-element antenna array. The suggested array can switch between a notched and a broad beam shape. Regarding target discrimination and locating accuracy, we compare this radar system to three-element and four-element frequency-sweeping SFCW radars. Compressed sensing methods are used to reconstruct images and evaluate the performance of various antenna arrays. The range resolution for the proposed two-element beam shape switching antenna system with a bandwidth of 10 GHz is noteworthy-10 mm for point targets and 20 mm for extended targets. This range resolution is two times that of the larger aperture three-element array and four times that of the four-element array. Notably, the range resolution beats the theoretical range resolution due to the use of compressed sensing to combine information from the two beam shapes. Furthermore, the suggested system outperforms previous antenna arrays regarding angular resolution, especially when targets are close in range and angle. With a 10 mm range distance and a 2(degrees )angle difference, the system effectively discriminates three targets. In a triple-target scenario, the suggested two-element beam shape switching system outperforms a standard four-element phased array radar with the same bandwidth even with restricted computational resources.

This article experimentally demonstrates a frequency-sweeping notch-beam sub-THz radar frontend based on a two-line array antenna featuring computational imaging. Operating within 237.5 GHz and 250 GHz with 12.5 GHz bandwidth, the radar utilizes a 12 λc delay line to achieve frequency-sweeping capabilities. This configuration allows dynamic notch-beam scanning across angular ranges from − 26.5 ∘ to 28 ∘ . The radar frontend is highly compact with a total size of 20 mm× 14.3 mm× 1.2 mm, including the beam-steering network, a magic-tee for creating the 180 ∘ phase shift required for creating the notch-beam, and the antenna array, and is implemented by silicon micromachining. The radar was evaluated with single and dual-target scenarios utilizing and benchmarking different computational imaging algorithms, i.e., matched filter (MF), fast iterative shrinkage-thresholding algorithm (FISTA), and multiple signal classification (MUSIC). It was found that the MUSIC algorithm outperforms MF and FISTA in range and angular resolution in single-target scenes, achieving a range resolution of 7.8 mm and an angular resolution of 15.7 ∘ , with detection errors of less than 6.6 mm and 3.5 ∘ , respectively. Although the MUSIC algorithm maintains reliable range resolution in dual-target scenarios, it performs poorly in providing angular information.

This article experimentally demonstrates a frequency-sweeping notch-beam sub-THz radar frontend based on a two-line array antenna featuring computational imaging. Operating within 237.5 GHz and 250 GHz with 12.5 GHz bandwidth, the radar utilizes a 12 λc delay line to achieve frequency-sweeping capabilities. This configuration allows dynamic notch-beam scanning across angular ranges from − 26.5 ∘ to 28 ∘ . The radar frontend is highly compact with a total size of 20 mm× 14.3 mm× 1.2 mm, including the beam-steering network, a magic-tee for creating the 180 ∘ phase shift required for creating the notch-beam, and the antenna array, and is implemented by silicon micromachining. The radar was evaluated with single and dual-target scenarios utilizing and benchmarking different computational imaging algorithms, i.e., matched filter (MF), fast iterative shrinkage-thresholding algorithm (FISTA), and multiple signal classification (MUSIC). It was found that the MUSIC algorithm outperforms MF and FISTA in range and angular resolution in single-target scenes, achieving a range resolution of 7.8 mm and an angular resolution of 15.7 ∘ , with detection errors of less than 6.6 mm and 3.5 ∘ , respectively. Although the MUSIC algorithm maintains reliable range resolution in dual-target scenarios, it performs poorly in providing angular information.

This work considers a collocated radar scenario where a probing signal is emitted toward the targets of interest and records the received echoes. Estimating the relative delay-Doppler shifts of the targets allows determining their relative locations and velocities. However, the received radar measurements are often affected by impulsive non-Gaussian noise which makes a few measurements partially corrupted. While demixing radar signal and impulsive noise is challenging in general by traditional subspace-based methods, atomic norm minimization (ANM) has been recently developed to perform this task in a much more efficient manner. Nonetheless, the ANM cannot identify close delay-Doppler pairs and also requires many measurements. Here, we propose a smoothed l(0) atomic optimization problem encouraging both the sparse features of the targets and the impulsive noise. We design a majorization-minimization algorithm that converges to the solution of the proposed non-convex problem using alternating direction method of multipliers (ADMM). Simulations results verify the superior accuracy of our proposed algorithm even for very close delay-Doppler pairs in comparison to ANM with around 40 dB improvement.

this paper investigates a novel radar concept that is based on a minimalistic, small-aperture antenna array but features intelligent beam-shape switching and artificial-intelligence signal processing. In contrast to conventional phased-arrays, size, cost, and hardware complexity are drastically reduced by the proposed dual-antenna array which can create a broad and a frequency-scanning notched beam shape. The angular-resolution and target discrimination performance of the proposed radar concept have been validated by radar simulations for single and multiple target scenarios. For the signal processing, two convolutional neural networks (CNN) and a multilayer perceptron model are benchmarked against each other. A further CNN is implemented for estimating the number of targets, which can be used to pre-select the type of network determining range and cross-range of multiple targets. This paper shows that a small antenna aperture frontend in combination with beam-shape switching and artificial-intelligence signal processing methods is a suitable hardware-efficient radar concept for accurate multi-target location.

This paper presents the design, fabrication, and characterization of a compact silicon-micromachined beam steering frontend circuit featuring beam shape switching and frequency beam steering. The designed beam steering frontend includes a switching circuit consisting of a crossover switch and a magic-tee for beam shape switching, as well as a delay line network connected to a corporate-fed dual-line antenna array for frequency beam steering. All components are co-designed and integrated to minimize the insertion loss. The signal chain is implemented on four layers of silicon-on-insulator chips with a total footprint of 20mm×1.4mm×1.2mm. The designed frontend operates properly in the 220-260 GHz frequency band with a measured return loss of better than 15 dB and a beam steering range of 238-248 GHz with a measured return loss of better than 20 dB. The frontend has two standard WR-3.4 waveguide input ports, and each port can create two different radiation patterns, a broad and a notched beam, in which both can be steered from 238 to 248 GHz in the E-plane of the antenna array. The sidelobe level of the broad beam is reduced by amplitude tapering and remains better than 18 dB in the H-plane of the antenna array across the scanning range. Moreover, the depth of the notched beam remains better than 17 dB in the entire scanning range, making the presented beam steering frontend well-suited for tracking, surveillance, imaging, and radar applications.

This paper explores novel aspects of sparsity in sub-THz computational imaging using dual-polarization, wideband,frequency-diverse antennas fabricated through silicon micro-machining techniques. These antennas—one with a distributed feed and the other with a direct feed—operate at 220-330GHz frequency range, generating pseudo-random radiation patterns. This makes them well-suited for high-resolution imaging with reduced hardware complexity compared to traditional radar imaging setups. A forward model based on these antennas' radiation patterns is integrated with compressed sensing techniques, specifically the Compressive Sampling MatchingPursuit (CoSaMP) and Fast Iterative Shrinkage-ThresholdingAlgorithm (FISTA). The computational imaging investigation, involving three configurations of the two antennas, shows that deploying the distributed feed antenna significantly enhances imaging performance in scenarios with sparse measurement data. Furthermore, the paper comprehensively investigates two types of sparsity: the number of Tx-Rx measurements and sub-bandwidth division. Sparsity analysis on several emulation results reveals that under high SNR conditions, using only the distributed feed antenna as both Tx and Rx requires just 25% of the measured data to successfully reconstruct an image of the scene. Additionally, the sub-bandwidth allocation technique drastically reduces the complexity of wideband hardware, requiring only 2% of the data when using the distributed feed antenna. The system demonstrates impressive performance through both simulations and experimental validation, achieving range resolutions as satisfactory as 1.4 mm and angular resolutions down to 0.35°.

This paper explores novel aspects of sparsity in sub-THz computational imaging using dual-polarization, wideband,frequency-diverse antennas fabricated through silicon micro-machining techniques. These antennas—one with a distributed feed and the other with a direct feed—operate at 220-330GHz frequency range, generating pseudo-random radiation patterns. This makes them well-suited for high-resolution imaging with reduced hardware complexity compared to traditional radar imaging setups. A forward model based on these antennas' radiation patterns is integrated with compressed sensing techniques, specifically the Compressive Sampling MatchingPursuit (CoSaMP) and Fast Iterative Shrinkage-ThresholdingAlgorithm (FISTA). The computational imaging investigation, involving three configurations of the two antennas, shows that deploying the distributed feed antenna significantly enhances imaging performance in scenarios with sparse measurement data. Furthermore, the paper comprehensively investigates two types of sparsity: the number of Tx-Rx measurements and sub-bandwidth division. Sparsity analysis on several emulation results reveals that under high SNR conditions, using only the distributed feed antenna as both Tx and Rx requires just 25% of the measured data to successfully reconstruct an image of the scene. Additionally, the sub-bandwidth allocation technique drastically reduces the complexity of wideband hardware, requiring only 2% of the data when using the distributed feed antenna. The system demonstrates impressive performance through both simulations and experimental validation, achieving range resolutions as satisfactory as 1.4 mm and angular resolutions down to 0.35°.