- ItemEmbargoCoexistence of weather radars and communication systems: model to identify interfering sources and mitigation solutions(Colorado State University. Libraries, 2023) Vaccarono, Mattia, author; Chandrasekaran, Chandra V., advisor; Cheney, Margaret, committee member; Jayasumana, Anura, committee member; Popat, Ketul, committee memberElectromagnetic spectrum is a finite resource. Weather radars are one of the many sources which use electromagnetic waves. The availability of spectrum bands that can be assigned to a specific user is limited. Consequently, the electromagnetic spectrum is shared by different application in the same frequency band. This is the specific case of C-band weather radars, which operate in the 5.6GHz band, sharing the same frequencies with Radio Local Area Networks, Wireless Local Area Networks and HiperLAN systems. These telecommunication systems are continuously increasing in rural areas as broadband Internet access points. The coexistence of C-band weather radar and such systems is nowadays a primary topic in the weather radar community. The amount of interference received by weather radars are affecting the data quality, especially for polarimetric observations. Electromagnetic interference may also appear at higher frequencies, such as the X-band located around 9.3GHz. These frequencies are used by weather radars for hydrological purposes. The dense radar network deployed in Dallas Fort Worth area and the mobile radar managed by Arpa Piemonte operate at X-band and they receive interfering signals. These signals have been detected during a field measurement campaign using both the mobile weather radar and a vector signal analyzer able to perform real time analysis. A technique to identify the likely interfering sources is discussed, which can be used by the National Regulatory Authorities or Regional Agencies, such as the Physics and Industrial Risk Department of Arpa Piemonte, Italy, in charge of the telecommunication authorization processes. The model may be applied to a telecommunication tower transmitting at the same frequency of a given radar and in case of likely interference, mitigation strategies could be set during the tower installation, i.e. changing the antenna direction or tilt. Over the years, many RFI removal and mitigation tools have been discussed in the literature, but only few are currently implemented on operational weather radars. This work, instead, aims to implement mitigation solutions that can be implemented by National Weather Services. The electromagnetic interference may be removed at different levels: from the received signals to the processed radar products, such as reflectivity maps that are shown to general public. In order to make possible the interference removal also to those National Weather Services, or radar management services, which are not able to act on the radar signal processor to implement deeper mitigation tools, a RFI mitigation solution based on image processing is shown. This method does not require to access the radar signal processor, but it does not mitigate the effect of interference overlapped with weather echoes. Then, based on the interfering signals features, a mitigation solution has been developed. The interfering signals are removed before received signals are processed to obtain radar moments. The proposed method has been tested with good performances in clear air echoes at both C and X-bands. A study case has been selected to evaluate its performances during precipitation events. The proposed mitigation solution is applied to the received signals to remove interfering signals and to reconstruct the residual information. The radar reflectivity is computed and it is compared to the operational radar Z product. A Swiss C-band radar is selected as reference to validate the mitigation solution. The interfering signals are properly removed and the missing data in the received radar pulses are computed by smoothing from adjacent range gates and pulses. Actually, removing only the interfering signals the proposed solution is able to preserve the meteorological echoes which lead to a better estimate of the reflectivity values, especially in case of weak echoes (i.e. light rain or drizzle). The Interference to Signal Ratio (ISR) is considered the metric to quantitatively evaluate the mitigation performance as ISR difference between processed and received signals. The proposed mitigation solution can achieve up to 20dB suppression.
- ItemEmbargoMultiphoton spatial frequency modulated imaging(Colorado State University. Libraries, 2023) Wernsing, Keith, author; Bartels, Randy, advisor; Squier, Jeff, committee member; Wilson, Jesse, committee member; Borch, Thomas, committee memberFar-field optical microscopy has seen significant development in the last 20 years in its ability to resolve specimen information beyond the diffraction limit. However, nearly all of these super-resolution techniques are predicated on the use of fluorescence as the contrast mechanism in the sample. While the variety of fluorophores available for labeling a sample are a widely-utilized tool, in many instances non-fluorescent contrast mechanisms also provide valuable information. Multiphoton microscopy is one route to probing non-fluorescent contrast mechanisms. It has the benefit of sampling multiple contrast mechanisms at once, including second- and third-harmonic generation and Raman vibrational characteristics, as well as autofluorescence and labeled fluorescence. However, development of super-resolving techniques for coherent scattering processes like harmonic generation or coherent Raman excitation has lagged behind that of incoherent scattering processes like fluorescence. In this work I present the first technique to simultaneously enhance resolution in both real-state (e.g., fluorescence) and virtual-state (e.g. harmonic generation) molecular excitation mechanisms, known as multiphoton spatial-frequency modulated imaging (MP-SPIFI). Standard SPIFI works by projecting spatial cosine patterns onto the sample and gathering object spatial frequency information. Multiphoton SPIFI generates harmonics of these cosine patterns and therein gathers information beyond the frequency passband of the microscope. We demonstrate our initial results with two-photon fluorescence and SHG. An extensive model is built describing the super-resolved image formation process. We then present a method for extending the native, 1D resolution enhancement into two dimensions for an isotropic enhancement. Finally, we present development of two femtosecond, amplified pulsed laser sources tailored to boost SNR in multiphoton processes, through parabolic pulse amplification, and chirped pulse fiber broadening, in order to deliver the high average power & high peak power required by MP-SPIFI for driving nonlinear processes across a line-focus geometry.
- ItemOpen AccessSilicon photonic hardware accelerators for transformers and graph neural networks(Colorado State University. Libraries, 2023) Afifi, Salma, author; Pasricha, Sudeep, advisor; Nikdast, Mahdi, committee member; Malaiya, Yashwant, committee memberThe rapid growth of artificial intelligence (AI) applications has revolutionized the way we process data, make decisions, and interact with machines. Specifically, artificial neural networks (ANNs) have significantly evolved and now encompass various advanced neural networks such as transformers and graph neural networks (GNNs). This has enabled the development of innovative AI applications that can transform several industries, including healthcare, recommendation systems, and robotics. Transformer and transformer-based neural networks have outperformed multiple ANNs, such as convolution neural networks (CNNs) and recurrent neural networks (RNNs), across many natural language processing (NLP) tasks. Moreover, transformers are currently being integrated into vision tasks through using the vision transformer model (ViT). Similarly, GNNs have witnessed a surge of advancements over the past few years and have established their proficiency in dealing with graph-structured data. Nevertheless, each of these neural networks imposes unique challenges, hindering their inference and usage in resource-constrained systems. For instance, the transformer model's size, number of parameters, and complexity of operations lead to long inference times, large memory footprint, and low computation-to-memory ratio. On the other hand, GNNs inference challenges are due to their dense and very sparse computations. Additionally, the wide variety of possible input graphs structure and algorithms dictate the need for a system capable of efficiently adapting their execution and operations to the specific graph structure and effectively scaling to extremely large graphs. Accordingly, conventional computing processors and ANN accelerators are not tailored to cater for such challenges, and using them to accelerate transformers and GNN execution can be highly inefficient. ii Furthermore, the utilization of traditional electronic accelerators entails a number of limitations, including escalating fabrication costs due to low yields and diminishing performance improvements, associated with semiconductor-technology scaling. This has led researchers to start investigating other technologies for ANN acceleration such as silicon photonics which enables performing complex operations in the optical domain with low energy consumption and at very high throughput. While several hardware accelerators leveraging silicon photonics have been presented for networks such as CNNs, none have been customized for emerging complex neural networks such as transformers and GNNs. Due to the various challenges associated with each of these networks, designing reliable and efficient inference hardware accelerators for transformers and GNNs is a non-trivial problem. This thesis introduces two novel silicon-photonic-based hardware architectures for energy efficient and high throughput inference acceleration. As our first contribution, we propose a non-coherent silicon photonic hardware accelerator for transformer neural networks, called TRON. We demonstrate how TRON is able to accommodate a wide range of transformer and transformer-based neural networks while surpassing GPU, CPU, TPU, and several state-of-the-art transformer hardware accelerators. For GNN inference acceleration, we propose GHOST, a hardware accelerator that integrates various device-, circuit- and architecture-level optimizations which enable it to efficiently process a broad family of GNNs and real-world graph structures and sizes. When compared to multiple state-of-the-art GNN hardware accelerators, GPUs, CPUs, and TPUs, our experiments showcase how GHOST exhibits significantly better performance and energy efficiency.
- ItemOpen AccessSimulation and hardware validation of methods for synchronization of central-converter multi-motor electric actuation systems(Colorado State University. Libraries, 2023) Miller, Zane P., author; Cale, James, advisor; Chong, Edwin, committee member; Fairbank, William, committee memberReplacement of previously hydraulic and pneumatic drives with power-electronic drive systems to reduce weight and maintenance requirements is a current target of research in the aerospace industry. This includes electrification of thrust reverser actuation systems (TRAS), which redirect thrust produced by the aircraft's engines to aid with deceleration upon landing, reducing wear on the brakes. However, one challenge of developing an electromagnetic TRAS (EM-TRAS) is the requirement of speed and position synchronization of all motors in the system, despite unequal torque loading from differing wind forces. Use of a single ("central") power electronic converter to power a set of induction machines in parallel could potentially lower cost and weight requirements compared to the use of separate converters, but such a central-converter, multi-motor (CCMM) architecture requires some form of compensation for load torque differences. Previous research presented a synchronization methodology using closed-loop feedback control of variable stator resistances in parallel with each induction machine. This thesis builds on this research by presenting an alternative methodology that instead applies closed-loop feedback control to smaller-scale auxiliary converters for each motor line, coupled to the induction machines using transformers to apply adjustments to the stator voltage. This new methodology achieves similar synchronization performance with better energy efficiency, lowering power requirements for its use compared to the external resistance methodology. The author's contributions to construction of a testbed for aerospace actuation system research are also presented in this thesis, with applications including hardware validation of the external resistance CCMM EM-TRAS implementation.
- ItemOpen AccessSingle pixel computational imaging(Colorado State University. Libraries, 2023) Stockton, Patrick Allen, author; Bartels, Randy A., advisor; Pezeshki, Ali, committee member; Muller, Jennifer, committee member; Wilson, Jesse, committee memberMicroscopy has a long rich history of peering into life's smallest mysteries. Ever since the first microscope, the ability to see objects that would otherwise be impossible to see with the naked eye have allowed new discoveries and modern technology has benefited tremendously. There have been many improvements on microscopes over the centuries with each improvement unlocking more knowledge as we go. Some of these advancements are the modern objective lens correcting for numerous optical aberrations, phase contrast imaging allowing nearly transparent samples to have high contrast, the confocal pinhole allowing an easy method to get optical sectioning, and super resolution microscopy surpassing the diffraction limit by several orders of magnitude. One of the most amazing things about all these discoveries is that they all rely on the same fundamental concepts. This work focuses on expanding the capabilities of single pixel imaging. Single pixel imaging is a class of imaging that encodes spatial information on a temporal signal using a single element detector; having knowledge of the encoding allows the time signal to be reconstructed to generate a spatial image. A canonical example of single pixel imaging is laser scanning microscopy (LSM). More complicated encoding systems have been developed but the basic idea for reconstruction remains the same. There are several advantages conferred to single pixel imaging such as image formation is resistant to scattering, very fast temporal response, flexibility in detector selection at a given wavelength, and exotic imaging information. My research primarily utilizes two techniques, SPatIal Frequency modulated Imaging (SPIFI) and Coherent Holographic Image Reconstruction by Phase Transfer (CHIRPT), both are explained in detail. My research aims to expand the capability's of SPIFI by providing a method for homogenizing the anisotropic resolution observed in the higher orders, additionally, I present a method of solving the inverse problem that allows the measurement matrix to more accurately represent to true image formation process there by increasing the performance of the reconstruction. I present research for CHIRPT which takes advantage of the encoded coherent phase information of two interfering beams to measure the quantitative phase of an object. I also present a new technique utilizing CHIRPT's holographic phase information to extend optical diffraction tomography to incoherent emitters which has long been an illusive task.