서브메뉴
검색
Time Domain Source Separation Methods for Impulsive Aeroacoustic Sources
Time Domain Source Separation Methods for Impulsive Aeroacoustic Sources
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202104732
- ISBN
- 9798290650708
- DDC
- 330
- 서명/저자
- Time Domain Source Separation Methods for Impulsive Aeroacoustic Sources
- 발행사항
- [Sl] : The Pennsylvania State University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 279 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-01, Section: B.
- 주기사항
- Advisor: Harris, Jeff R.;Krane, Michael H.
- 학위논문주기
- Thesis (Ph.D.)--The Pennsylvania State University, 2025.
- 초록/해제
- 요약Source separation is a necessary signal processing task when multiple sources are present in an experiment. Methods which seek to perform source separation are often framed as an inverse problem, in which the underlying sources are estimated from a set of observations of their mixture. Many methods seek to perform source separation with little to no knowledge of both the underlying source waveforms as well as the mixing parameters, referred to as blind source separation (BSS). Solving the BSS problem is non-trivial, particularly when time delays of the source must be accounted for across the observations. Most real-world problems will experience these time delays. Mixtures which account for time delays across observations are called convolutive mixtures.Impulsive sources are common across many different acoustic problems. For source separation, impulsive sources violate many of the assumptions made by canonical source separation methods. In recent years there has been an increased interest in data-driven and machine learning techniques in acoustics. Machine learning techniques are often statistical in nature and can estimate model parameters in both a deterministic and probabilistic manner. These techniques may be leveraged for the separation of impulsive acoustic sources.Aeroacoustic sources can often be impulsive, like the emission of a vortex/edge (V/E) interaction. In the present study, microphone array signals are used which are observing the V/E interaction in the presence of an additional impulsive source. The efficacy of these machine learning techniques for source separation are evaluated by comparing the estimated V/E interaction source parameters with theory and prior experiments. Principal component analysis was found to be ineffective in separating the V/E interaction source. Robust principal component analysis, however, provided sufficient source separation to estimate the V/E source parameters. An iterative approach using robust principal component analysis for the separation of multiple sources is also introduced.Probabilistic methods are also evaluated for impulsive source separation. A simple interpretable model is developed to approximate the V/E experiment. Information about the V/E experiment and source are used to simplify the approximating model. The model is parameterized by source characteristics including time series waveforms and directivities. Statistical inversion methods maximum likelihood estimation and maximum a posterioriare used to estimate the model parameters and quantify their uncertainties. The prior information included in the Bayesian estimation framework (maximum a posteriori) is shown to increase the effectiveness of the source separation. The distributions describing the underlying source waveforms can then be used to estimate the uncertainty of other V/E interaction source characteristics.
- 일반주제명
- Sparsity
- 일반주제명
- Vortices
- 일반주제명
- Normal distribution
- 일반주제명
- Signal processing
- 일반주제명
- Microphones
- 일반주제명
- Decomposition
- 일반주제명
- Acoustics
- 일반주제명
- Parameter estimation
- 기본자료저록
- Dissertations Abstracts International. 87-01B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017358656
■00520260202104732
■006m o d
■007cr#unu||||||||
■020 ▼a9798290650708
■035 ▼a(MiAaPQ)AAI32148959
■035 ▼a(MiAaPQ)PennState28139mjs8688
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a330
■1001 ▼aSwann, Mitchell J.
■24510▼aTime Domain Source Separation Methods for Impulsive Aeroacoustic Sources
■260 ▼a[Sl]▼bThe Pennsylvania State University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a279 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-01, Section: B.
■500 ▼aAdvisor: Harris, Jeff R.;Krane, Michael H.
■5021 ▼aThesis (Ph.D.)--The Pennsylvania State University, 2025.
■520 ▼aSource separation is a necessary signal processing task when multiple sources are present in an experiment. Methods which seek to perform source separation are often framed as an inverse problem, in which the underlying sources are estimated from a set of observations of their mixture. Many methods seek to perform source separation with little to no knowledge of both the underlying source waveforms as well as the mixing parameters, referred to as blind source separation (BSS). Solving the BSS problem is non-trivial, particularly when time delays of the source must be accounted for across the observations. Most real-world problems will experience these time delays. Mixtures which account for time delays across observations are called convolutive mixtures.Impulsive sources are common across many different acoustic problems. For source separation, impulsive sources violate many of the assumptions made by canonical source separation methods. In recent years there has been an increased interest in data-driven and machine learning techniques in acoustics. Machine learning techniques are often statistical in nature and can estimate model parameters in both a deterministic and probabilistic manner. These techniques may be leveraged for the separation of impulsive acoustic sources.Aeroacoustic sources can often be impulsive, like the emission of a vortex/edge (V/E) interaction. In the present study, microphone array signals are used which are observing the V/E interaction in the presence of an additional impulsive source. The efficacy of these machine learning techniques for source separation are evaluated by comparing the estimated V/E interaction source parameters with theory and prior experiments. Principal component analysis was found to be ineffective in separating the V/E interaction source. Robust principal component analysis, however, provided sufficient source separation to estimate the V/E source parameters. An iterative approach using robust principal component analysis for the separation of multiple sources is also introduced.Probabilistic methods are also evaluated for impulsive source separation. A simple interpretable model is developed to approximate the V/E experiment. Information about the V/E experiment and source are used to simplify the approximating model. The model is parameterized by source characteristics including time series waveforms and directivities. Statistical inversion methods maximum likelihood estimation and maximum a posterioriare used to estimate the model parameters and quantify their uncertainties. The prior information included in the Bayesian estimation framework (maximum a posteriori) is shown to increase the effectiveness of the source separation. The distributions describing the underlying source waveforms can then be used to estimate the uncertainty of other V/E interaction source characteristics.
■590 ▼aSchool code: 0176.
■650 4▼aSparsity
■650 4▼aVortices
■650 4▼aNormal distribution
■650 4▼aSignal processing
■650 4▼aMicrophones
■650 4▼aDecomposition
■650 4▼aAcoustics
■650 4▼aParameter estimation
■690 ▼a0986
■71020▼aThe Pennsylvania State University.
■7730 ▼tDissertations Abstracts International▼g87-01B.
■790 ▼a0176
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17358656▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


