Signal parameter estimation of damped sinusoidal waveforms using deep learning
| dc.contributor.author | Idoko, Dawn | |
| dc.contributor.examiningcommittee | Swatek, David (Electrical and Computer Engineering) | en_US |
| dc.contributor.examiningcommittee | Ashraf, Ahmed (Electrical and Computer Engineering) | en_US |
| dc.contributor.supervisor | Kordi, Behzad | |
| dc.contributor.supervisor | Bridges, Greg E. | |
| dc.date.accessioned | 2022-12-21T21:42:35Z | |
| dc.date.available | 2022-12-21T21:42:35Z | |
| dc.date.copyright | 2022-12-21 | |
| dc.date.issued | 2022-12-21 | |
| dc.date.submitted | 2022-12-21T15:28:00Z | en_US |
| dc.degree.discipline | Electrical and Computer Engineering | en_US |
| dc.degree.level | Master of Science (M.Sc.) | en_US |
| dc.description.abstract | Sinusoids and damped signals are a fundamental part of different engineering fields. Analysis of these signals to give an accurate estimation of certain parameters such as frequency, damping factor, and phase angle is important in many engineering fields as an accurate estimation of these parameters is needed to ensure the smooth running of various processes. The need for higher levels of precision and accuracy in the signal-processing domain has resulted in the development of several algorithms based on different methods of operation. These algorithms can be divided into two classes, namely, parametric and non-parametric algorithms. The former assumes that the signal follows a particular model and estimates the signal parameters based on that assumption, while the latter makes no assumptions regarding the signal. Intuitively, the non-parametric class of algorithms seem to be a better choice for real-life applications as the model of the signal is usually unknown. However, algorithms under this class suffer from the issue of spectral leakage. Both classes of algorithms for signal analysis have their strengths as well as shortcomings. In this thesis, the concept of using machine learning methods in signal analysis is explored. To achieve this, the DeepFreq model is extended by modifying its architecture and applying it to damped sinusoidal signals to provide an estimate of signal parameters such as frequency and damping factor. The developed algorithm can estimate the number of frequencies as well as the value of the frequencies contained in a signal waveform with an R2 score of 0.88, even in noise levels of up to 0dB. The algorithm's performance was evaluated using data samples of sinusoidal signals within the ISM band range of 2.4GHz to 2.65GHz. The algorithm was tested on synthetic data and data from lab experiments, and the results show that the deep learning model can perform frequency and damping factor estimation for damped multi-frequency sinusoidal signals. | en_US |
| dc.description.note | February 2023 | en_US |
| dc.description.sponsorship | Manitoba Hydro, Mitacs, NSERC | en_US |
| dc.identifier.uri | http://hdl.handle.net/1993/37027 | |
| dc.language.iso | eng | en_US |
| dc.rights | open access | en_US |
| dc.subject | Deep learning | en_US |
| dc.subject | Frequency estimation | en_US |
| dc.subject | Damping factor estimation | en_US |
| dc.subject | Signal Analysis | en_US |
| dc.subject | Parametric methods | en_US |
| dc.subject | Non-parametric methods | en_US |
| dc.title | Signal parameter estimation of damped sinusoidal waveforms using deep learning | en_US |
| dc.type | master thesis | en_US |
| local.subject.manitoba | no | en_US |