Optimal Frequency Resolution Research Articles

Parametrized pre-trained network (PPNet): A novel shape classification method using SPHARMs for MI detection

PurposeShape information contains descriptive knowledge for classification problems. A novel methodology for combining the valuable 3D and 3D + t shape information with the advantages of deep transfer learning is introduced. This method produces informative images from shape information, which are interpretable for deep CNNs previously trained using a large-scale image dataset. MethodsOur proposed structure can (1) generate 3D surfaces with optimal spatial resolution, (2) parametrize them using spherical harmonics shape descriptors (SPHARMs) with optimal frequency resolution, (3) employ registration to establish surface correspondences, and (4) convert the 3D corresponding point clouds to 2D images by spreading out the latitude and the longitude values of spherical components in the form of a flat coordinate system. The generated 2D images have the potential to feed into the pre-trained deep CNNs. The efficiency of the proposed method was assessed in myocardial infarction (MI) classification using two different datasets. ResultsFive well-known CNN architectures as fixed feature extractors were evaluated. Deep features derived from the output of convolutional and pooling layers were fed into the SVM classifier. We also explored the feasibility of adding additional information by assigning a color to each RGB channel. Subsequently, EfficinetNet-b0 reached the best performance, with an accuracy of 97.92% ± 1.47 using colored images containing additional information, respectively. ConclusionOur suggested approach by incorporating the transformed shape information and CNNs can be adopted as a robust computer-aided diagnosis system to assist clinicians in detecting various diseases such as MI.

Influence of frequency resolution in case of frequency response function measurement in structural dynamics

Frequency resolution is an essential parameter in acoustical testing, even if we are using numerical or experimental method, for example when determining frequency response function (FRF) of a dynamic mechanical system, or executing modal analysis based on the FRFs. Finer resolution leads to more accurate results, at the expense of longer calculation/measurement process and larger data size. This parameter is generally set based on rules of thumb, prior practice or with big margin for safety. This results in waste time and data storage if the required frequency resolution is overestimated, or even significant errors in the results, if it is underestimated. Present paper offers a direct, method for the conscious determination of optimal frequency resolution. It is based fully on theoretical considerations, and investigates amplitude and phase distortion at resonances as target parameters. Beside defining the steps of the process, it is tested on a real structure, and the results are presented as well, proving the applicability and the appropriateness of the method. With this method, development engineers get a practical tool for adjusting the parameters of dynamic measurements and simulations.

Adaptive S transform for feature extraction in voltage sags

This paper proposes an adaptive S transform (AST) to extract the feature vectors of voltage sags. With the effective window width matches the Fourier spectrum of sag signals, the standard deviation σ of Gaussian window may be determined as well. The narrowest and the widest window width of AST are obtained without additional parameters and iterative computing. Then, the optimal frequency resolution and time resolution are got respectively. Compared with ST, AST provides better time–frequency resolution to extract more precise feature vectors of eight types of voltage sags. Based on the time–frequencyrepresentation of AST, five disturbance features are extracted to construct the feature vector in this paper. In addition, four machine learning classifiers and two fuzzy clustering classifiers are used to analyze the validity and redundancy of these features. Through analyzing the classification accuracies and time costs of these classifiers with different training sets and different level of noise, it can be concluded that the machine learning classifiers perform better in classification accuracy and stability than fuzzy clustering classifiers.

A two-step parameter optimization algorithm for improving estimation of optical properties using spatial frequency domain imaging

This research was aimed at optimizing the inverse algorithm for estimating the optical absorption (μa) and reduced scattering (μs′) coefficients from spatial frequency domain diffuse reflectance. Studies were first conducted to determine the optimal frequency resolution and start and end frequencies in terms of the reciprocal of mean free path (1/mfp′). The results showed that the optimal frequency resolution increased with μs′ and remained stable when μs′ was larger than 2 mm−1. The optimal end frequency decreased from 0.3/mfp′ to 0.16/mfp′ with μs′ ranging from 0.4 mm−1 to 3 mm−1, while the optimal start frequency remained at 0 mm−1. A two-step parameter estimation method was proposed based on the optimized frequency parameters, which improved estimation accuracies by 37.5% and 9.8% for μa and μs′, respectively, compared with the conventional one-step method. Experimental validations with seven liquid optical phantoms showed that the optimized algorithm resulted in the mean absolute errors of 15.4%, 7.6%, 5.0% for μa and 16.4%, 18.0%, 18.3% for μs′ at the wavelengths of 675 nm, 700 nm, and 715 nm, respectively. Hence, implementation of the optimized parameter estimation method should be considered in order to improve the measurement of optical properties of biological materials when using spatial frequency domain imaging technique.

Automatic Detection of Nonstationary Multiple Oscillations by an Improved Wavelet Packet Transform

An automatic detection technique of nonstationary multiple oscillations in industrial control loops is proposed. The technique is based on an improved wavelet packet transform (WPT) which integrates Shannon Entropy, with a proposed non-Gaussianity test and quasi-intrinsic mode function index to automatically generate the wavelet packet decomposition structure. An optimal frequency resolution of the process data for oscillation detection purposes is obtained in a form of a binary tree. Comparative studies on simulated examples as well as industrial cases indicate that the proposed approach is robust to the nonstationary features of oscillations.

Joint Spatial-Spectral Feature Space Clustering for Speech Activity Detection from ECoG Signals

Brain-machine interfaces for speech restoration have been extensively studied for more than two decades. The success of such a system will depend in part on selecting the best brain recording sites and signal features corresponding to speech production. The purpose of this study was to detect speech activity automatically from electrocorticographic signals based on joint spatial-frequency clustering of the ECoG feature space. For this study, the ECoG signals were recorded while a subject performed two different syllable repetition tasks. We found that the optimal frequency resolution to detect speech activity from ECoG signals was 8Hz, achieving 98.8% accuracy by employing support vector machines as a classifier. We also defined the cortical areas that held the most information about the discrimination of speech and nonspeech time intervals. Additionally, the results shed light on the distinct cortical areas associated with the two syllables repetition tasks and may contribute to the development of portable ECoG-based communication.

Improved spectrum estimation from digitized time series on a logarithmic frequency axis

We present a practical technique for spectrum and spectral density estimation from long time series by Fourier transforms. We apply Welch’s popular technique of “averaging over modified periodograms” which uses Fourier transforms of fixed length with time-domain windows and overlap. Our technique retains the basic properties of this method, but computes the optimal frequency resolution individually for each Fourier frequency on a logarithmic frequency axis, thus yielding results that are more useful than those of the standard techniques.

The mammalian cochlear map is optimally warped.

The form of the mammalian cochlear frequency-position map has been well described by Greenwood and empirical values found for its coefficients for a number of species. The apical portion of the mammalian map is spatially compressed relative to the base, and this nonuniformity in the representation of frequency is evidently consistent across species. However, an evolutionary reason for this consistency, encompassing critical band behavior with respect to position, is conspicuously missing. Likewise, the length of the cochlea in any mammal, including echolocating species, is related to body size, but attempts to explain the length in terms of frequency limits, range, or resolution have no general explanation. New insight stems from a hypothesis in which the map curvature may be appreciated as an adaptation for optimal frequency resolution over the auditory range. It is demonstrated numerically that the mammalian curve may be considered a member of a family of curves which vary in their degree of warp. The "warp factor" found to be common across mammals is an optimal trade-off between four conflicting constraints: (1) enhancing high-frequency resolution; (2) setting a lower bound on loss of existing low-frequency resolution; (3) minimizing map nonuniformity; and (4) keeping the whole map smooth, thereby avoiding reflections.

Sound source segregation using the inter-channel differences in both intensity and phase

A method is proposed for segregating sounds from multiple sources recorded on a two-channel system. It uses the differences in both intensity and phase, between the channels. In addition, the relation between frequency resolution and sound quality of the proposed method is examined. In the tests, three pairs of mixed sounds were used, namely, male speech with female speech, two kinds of female speech (the same person with different phrases), and a cock’s call with female speech. Frequency resolution was varied between 5 and 80 Hz. In the subjective test, subjects listened to the original sound plus sounds segregated at five different resolutions: 5, 10, 20, 40, and 80 Hz. Then they estimated the quality of these sounds on a five-point scale. Moreover, the correlation coefficient between the original sound and each segregated sound is discussed. In the case of a human voice, it is determined that the optimal frequency resolution is 10 or 20 Hz. The interesting fact is that the optimal frequency resolution was not the highest. Compared with the human voice, lower frequency resolution may be enough to segregate a cock’s call.