Computing Deblurred Time-Frequency Distributions Using Artificial Neural Networks

Abstract

This paper presents an effective correlation vectored taxonomy algorithm to compute highly concentrated time-frequency distributions (TFDs) using localized neural networks (LNNs). Spectrograms and pre-processed Wigner-Ville distributions of known signals are vectorized and clustered as per the elbow criterion to constitute the training data for multiple artificial neural networks. The best trained networks become part of the LNNs. Test TFDs of unknown signals are then processed through the algorithm and presented to the LNNs. Experimental results demonstrate that appropriately vectored and clustered data once processed through the LNNs produce high resolution TFDs. Examples are presented to show the effectiveness of the proposed algorithm via analysis based on entropy and visual interpretation.

Time–frequency features for pattern recognition using high-resolution TFDs: A tutorial review

“…The trained neural networks are then applied to the Spectrogram to estimate the true highresolution TFD. The disadvantage of this technique is that it results in a discontinuous TFD [49].…”

“…For example, these techniques have been applied to estimate an original high-resolution TFD from the Spectrogram, which can be interpreted as a blurred version of the true (t, f ) image [34]. Another method reduces the blurring in the Spectrogram by using artificial neural networks [49]. This technique estimates the inverse of the blurring function using neural networks, which are trained using a precomputed Spectrogram and the WVD.…”

A Class of Highly Concentrated Time-Frequency Distributions Based on the Ambiguity Domain Representation and Complex-Lag Moment

“…The signal's terms in (8) are obtained by using the analytic extension of x(t). In the case of multicomponent signals it can be calculated as follows:…”

“…Thus, for the analysis of signals with varying instantaneous frequency, higher-order distributions are used [3][4][5][6][7]. Also, some interesting methods based on neural networks [8,9], sparsity constraint of energy distribution [10], and autoregressive moving-average models [11] have been recently introduced to improve the resolution in the time-frequency domain.…”

Nonlinear Poisson regression using neural networks: a simulation study

“…This kind of neural network has also been successfully applied in clinical outcome prediction of myocardial infarction, mortality, surgical decision making on traumatic brain injury patients, recovery from surgery, pediatric, genecology, head trauma, and transplantation [7][8][9][10][11][12][13][14][15][16].…”