東華大學圖書館 |

Machine Learning Enhanced Antenna Systems.

紀錄類型:	書目-電子資源 : Monograph/item
正題名/作者:	Machine Learning Enhanced Antenna Systems./
作者:	Friedrichs, Gaeron R.
出版者:	Ann Arbor : ProQuest Dissertations & Theses, : 2023,
面頁冊數:	176 p.
附註:	Source: Dissertations Abstracts International, Volume: 84-11, Section: B.
Contained By:	Dissertations Abstracts International84-11B.
標題:	Electrical engineering. -
電子資源:	https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30247838
ISBN:	9798379529789

Machine Learning Enhanced Antenna Systems.
Friedrichs, Gaeron R.

Machine Learning Enhanced Antenna Systems. - Ann Arbor : ProQuest Dissertations & Theses, 2023 - 176 p.

Source: Dissertations Abstracts International, Volume: 84-11, Section: B.

Thesis (Ph.D.)--University of Colorado at Boulder, 2023.

This item must not be sold to any third party vendors.

Radio direction finding (DF) has long since been of interest to the antenna and greater RF community. DF helps address problems seen in both commercial and defense applications, and the needs are ever-growing. A substantial breadth of research addresses maximizing the performance of DF systems. High quality antenna design, high performance beamforming, and high resolution processing have been demonstrated with great success, though often sacrificing size, weight, power, or cost (SWAP-C). The widespread proliferation of the Internet of Things and generally increased connectivity opens up research areas that leverage unprecedented levels of networking and connectivity. These systems require low-cost, low-fidelity RF components to enable mass proliferation, which makes engineering well-performing solutions difficult by comparison to historically specialized sensors. This thesis seeks to provide low-cost sensing solutions, leveraging machine learning and modern processing techniques, that yield high performance and are compatible with modern day edge-computing requirements.An investigation into the feasibility of using machine learning for retrofitting - the extraction of information for which a system is not conventionally designed for - is explored in the context of polarimetry. Fundamental studies are used to demonstrate the practicality of machine learning to facilitate low-cost sensing capabilities. An ultra-wideband log periodic antenna is first designed for additive manufacturing to integrate the antenna and feeding structure (balun and impedance transformer) into the body of the sensor itself. The dual-polarized system enables wideband spectrum sensing at a relatively low cost. This system, in isolation, does not sense angle-of-arrival information, therefore antenna-induced distortions imparted on the signal cannot be corrected with DF-based calibration methods, making polarimetry difficult. A neural network solution is demonstrated which facilitates good accuracy in classifying the polarization of the incident signal, whereas polarimetry is otherwise unachievable.An extension of the retrofitting motif is further explored through the lens of DF. An ultrawideband circular antenna array is investigated for its ability to perform amplitude-only DF, despite its radiation patterns being suboptimal for the task. Machine learning is once again utilized, but with careful consideration to the practical deployment aspects of the algorithm. A novel, general, architecture is developed to leverage the rotational symmetry of the uniform circular array to perform single-snapshot azimuth angle-of-arrival estimation with as minimal a footprint as possible. The architecture scales favorably, and its relative performance only increases as a function of the number of elements, when compared to competing methods. An elevation sensing capability is also shown with the same sensor, which is unprecedented for other systems with similar field-of-view (FOV) and bandwidth. These concepts are then extended to the design of a structure. The design of as low-cost, low-complexity system is achieved, while maintaining azimuth and elevation DF capabilities with a single snapshot. Fundamental studies on the number of elements and element geometries are carried out, and a particular geometry is built and tested with a custom, four-channel receiver. A systematic neural network design methodology is introduced to facilitate the construction of neural networks for DF. Experimental validation is performed showing the efficacy of integrating machine learning for azimuth and elevation estimation, with a low-element count antenna array and receiver.Finally, the synthesis of all the prior investigations is achieved. Novel use of a four-arm spiral antenna sensor is proposed, using each arm of the spiral as its own amplitude-only channel. Traditionally, without mode-forming, DF is quite difficult. Frequency rotation modeling is deployed alongside a compact neural network architecture to facilitate ultrawideband amplitude-only DF in both azimuth and elevation with a cavity backed spiral in the absence of mode/beam-forming circuitry that is conventionally required. Integration with a four-channel receiver is demonstrated and performance over multiple octaves is obtained.

ISBN: 9798379529789Subjects--Topical Terms:

649834
Electrical engineering.
Subjects--Index Terms:

Antenna

Machine Learning Enhanced Antenna Systems.
LDR:05543nmm a2200409 4500 001 2394820
005 20240513061024.5
006 m o d
007 cr#unu||||||||
008 251215s2023 ||||||||||||||||| ||eng d
020 $a 9798379529789
035 $a (MiAaPQ)AAI30247838
035 $a AAI30247838
040 $a MiAaPQ $c MiAaPQ
100 1 $a Friedrichs, Gaeron R. $0 (orcid)0000-0002-6703-1712 $3 3764318
245 1 0 $a Machine Learning Enhanced Antenna Systems.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2023
300 $a 176 p.
500 $a Source: Dissertations Abstracts International, Volume: 84-11, Section: B.
500 $a Includes supplementary digital materials.
500 $a Advisor: Filipovic, Dejan S.
502 $a Thesis (Ph.D.)--University of Colorado at Boulder, 2023.
506 $a This item must not be sold to any third party vendors.
520 $a Radio direction finding (DF) has long since been of interest to the antenna and greater RF community. DF helps address problems seen in both commercial and defense applications, and the needs are ever-growing. A substantial breadth of research addresses maximizing the performance of DF systems. High quality antenna design, high performance beamforming, and high resolution processing have been demonstrated with great success, though often sacrificing size, weight, power, or cost (SWAP-C). The widespread proliferation of the Internet of Things and generally increased connectivity opens up research areas that leverage unprecedented levels of networking and connectivity. These systems require low-cost, low-fidelity RF components to enable mass proliferation, which makes engineering well-performing solutions difficult by comparison to historically specialized sensors. This thesis seeks to provide low-cost sensing solutions, leveraging machine learning and modern processing techniques, that yield high performance and are compatible with modern day edge-computing requirements.An investigation into the feasibility of using machine learning for retrofitting - the extraction of information for which a system is not conventionally designed for - is explored in the context of polarimetry. Fundamental studies are used to demonstrate the practicality of machine learning to facilitate low-cost sensing capabilities. An ultra-wideband log periodic antenna is first designed for additive manufacturing to integrate the antenna and feeding structure (balun and impedance transformer) into the body of the sensor itself. The dual-polarized system enables wideband spectrum sensing at a relatively low cost. This system, in isolation, does not sense angle-of-arrival information, therefore antenna-induced distortions imparted on the signal cannot be corrected with DF-based calibration methods, making polarimetry difficult. A neural network solution is demonstrated which facilitates good accuracy in classifying the polarization of the incident signal, whereas polarimetry is otherwise unachievable.An extension of the retrofitting motif is further explored through the lens of DF. An ultrawideband circular antenna array is investigated for its ability to perform amplitude-only DF, despite its radiation patterns being suboptimal for the task. Machine learning is once again utilized, but with careful consideration to the practical deployment aspects of the algorithm. A novel, general, architecture is developed to leverage the rotational symmetry of the uniform circular array to perform single-snapshot azimuth angle-of-arrival estimation with as minimal a footprint as possible. The architecture scales favorably, and its relative performance only increases as a function of the number of elements, when compared to competing methods. An elevation sensing capability is also shown with the same sensor, which is unprecedented for other systems with similar field-of-view (FOV) and bandwidth. These concepts are then extended to the design of a structure. The design of as low-cost, low-complexity system is achieved, while maintaining azimuth and elevation DF capabilities with a single snapshot. Fundamental studies on the number of elements and element geometries are carried out, and a particular geometry is built and tested with a custom, four-channel receiver. A systematic neural network design methodology is introduced to facilitate the construction of neural networks for DF. Experimental validation is performed showing the efficacy of integrating machine learning for azimuth and elevation estimation, with a low-element count antenna array and receiver.Finally, the synthesis of all the prior investigations is achieved. Novel use of a four-arm spiral antenna sensor is proposed, using each arm of the spiral as its own amplitude-only channel. Traditionally, without mode-forming, DF is quite difficult. Frequency rotation modeling is deployed alongside a compact neural network architecture to facilitate ultrawideband amplitude-only DF in both azimuth and elevation with a cavity backed spiral in the absence of mode/beam-forming circuitry that is conventionally required. Integration with a four-channel receiver is demonstrated and performance over multiple octaves is obtained.
590 $a School code: 0051.
650 4 $a Electrical engineering. $3 649834
650 4 $a Electromagnetics. $3 3173223
650 4 $a Computer engineering. $3 621879
653 $a Antenna
653 $a Direction finding
653 $a Machine learning
653 $a Sensor
690 $a 0544
690 $a 0607
690 $a 0800
690 $a 0464
710 2 $a University of Colorado at Boulder. $b Electrical Engineering. $3 1025672
773 0 $t Dissertations Abstracts International $g 84-11B.
790 $a 0051
791 $a Ph.D.
792 $a 2023
793 $a English
856 4 0 $u https://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30247838