ADVANCED MACHINE LEARNING EMPOWERING MICROWAVE SENSORS WITH GENERATIVE/PREDICTIVE CAPABILITIES

• Author / Creator

  Kazemi, Nazli

• This dissertation, addressing the critical shortage of microwave planar sensors, embarks on a journey to advance noncontact sensing applications, with a particular emphasis on the application of machine learning techniques to address existing problems. It pioneers the introduction of a unique planar reflective sensor, employing complementary split ring resonators (CSRRs) as the primary sensing element, complemented by an additional loss-compensating negative resistance. Experimental results validate that this innovative design substantially amplifies the sensor's resolution without any detriment to sensitivity.

  The thesis further unveils a novel design that utilizes coupled CSRRs, resulting in a sensor with heightened sensitivity. The sensor response in passive mode undergoes processing using CycleGAN, a machine learning algorithm typically harnessed for image-to-image translation. This groundbreaking application of machine learning effectively transmutes low-quality factor sensor profiles in the passive domain into their high-quality factor counterparts in the active domain. The resolution achieved by this CycleGAN-enhanced response rivals that of an active sensor, marking a significant leap in the resolution of passive sensors. This is a clear demonstration of how machine learning can be used to address the limitations of traditional sensor designs.

  In the biomedical sensing arena, the thesis emphasizes the paramount importance of continuous glucose monitoring systems that obviate the need for invasive finger pricking, thereby ameliorating the comfort and lifestyle of diabetic patients. It advocates for a compact planar resonator-based sensor for noncontact glucose sensing, delivering real-time response and a strong linear correlation between sensor readings and blood glucose levels. The integration of the Long Short Term Memory (LSTM) algorithm, another machine learning technique, enables the prediction of glucose level fluctuations in both non-diabetic and diabetic patients. This facilitates early interventions in case of abnormal glucose trends and aids in identifying sensor anomalies. This is another example of how machine learning can be used to address real-world problems, in this case, improving the management of diabetes.

  This research makes a monumental contribution to the field of microwave planar sensors by bolstering their robustness and resolution using machine learning algorithms of CycleGAN and LSTM. This paves the way for an expansion of potential applications in material characterization and biomedical analysis. The insights extracted from this thesis are instrumental in propelling the development of advanced sensor technologies, with a focus on crafting miniaturized devices. The central role of machine learning in these advancements cannot be overstated.

• Subjects / Keywords

  • machine learning
  • microwave sensor
  • lstm
  • generative adversarial network
  • glucose
  • anomaly detection
  • cyclegan
  • permittivity

• Graduation date

  Fall 2023

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-gd21-wh77

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.