Short-time Frequency Research Articles

Phase-Aware Speech Enhancement Based on Deep Neural Networks

Short-time frequency transform (STFT) is fundamental in speech processing. Because of the difficulty of processing highly unstructured STFT phase, most speech-processing algorithms only operate with STFT magnitude, leaving the STFT phase far from explored. However, with the recent development of deep neural network (DNN) based speech processing, e.g., speech enhancement and recognition, phase processing is becoming more important than ever before as a new growing point of DNN-based methods. In this paper, we propose a phase-aware speech enhancement algorithm based on DNN. Specifically, in the training stage, when incorporating phase as a target, our core idea is to transform an unstructured phase spectrogram to its derivative along the time axis, i.e., instantaneous frequency deviation (IFD), which has a similar structure with its corresponding magnitude spectrogram. We further propose to optimize both IFD and magnitude jointly in a multiobjective learning framework. In the test stage, we propose a postprocessing method to recover the phase spectrogram from the estimated IFD. Experimental results demonstrate the effectiveness of the proposed method.

Learning Time-Frequency Analysis in Wireless Sensor Networks

Time-frequency analysis is one of essential signal processing tool for wireless sensor signal in Internet of Things (IoT), but its traditional processing approaches, such as short time-frequency transformation (STFT) and discrete wavelet transformation (DWT), are challenged by the limitation of capability to self-learn from unknown environments and adjust parameters adaptively. To address this problem, we propose to build up the deep learning network to learn time-frequency analysis to instead of traditional STFT and DWT approaches. With using typical neural network layers to remodel STFT and DWT operations, the proposed models consider the efficiency on both training and processing procedures and show their parameter adaptability and capability of deep feature extraction from sensor signals. Moreover, we demonstrate how to integrate learning time-frequency analysis networks into practical IoT applications, signal detection in noisy environment, and classifying of various modulated wireless sensor signal, by which their performance are further evaluated in terms of computation complexity and efficiency.

RETRACTED ARTICLE: Multi-label algorithm based on rough set of fractal dimension attribute

To make fractal endpoint detection algorithm, to maintain good performance and to deal with noise with higher irregularity than speech, fractal endpoint detection algorithm based on frequency domain was proposed in the paper. The frequency domain represented energy distribution and signal, and the speech harmonic component had very strong periodicity and regularity in the frequency domain. Thus, method of extracting fractal dimension after converting to short-time frequency domain had better robustness. Analysis means were introduced based on short-time frequency domain fractal against the existing fractal algorithm. Its stability was due to the frequency domain represented frequency domain energy distribution of signal and the speech signal energy mainly focused on harmonic. Thus, solving fractal dimension in short-time frequency domain could weaken the impact of different types of noises. Compared with time-domain fractal, the threshold value of short-time frequency domain fractal was more stable and the judgment criterion direction was fixed, smaller than the represented speech fragment of threshold value. Frequency domain was used for representing the signal energy distribution characteristics and the strong periodicity and regularity of speech harmonic component so as to extract fractal dimension and distinguish speech and noise. Thus, the fractal dimension extraction method after converting to short-time frequency domain proposed in the paper had better robustness. Not only is it applicable to irregular white noise, but also applicable to noises with stronger time-domain periodicity and regularity including tank noise.

Blind Source Separation of Moving Sources Using Sparsity-Based Source Detection and Tracking

Sparsity-based blind source separation (BSS) algorithms in the short time–frequency (TF) domain have received a lot of attention due to their versatility and noise reduction capabilities. In most of these algorithms, the estimation of the BSS filters relies on the accurate association of each time–frequency bin to the dominant source at that bin. The TF bin associations are then used to estimate the statistics of the source signals, and BSS is achieved by optimal spatial filters computed using the estimated statistics. The main objective of this paper is to apply such a framework to scenarios with an unknown number of moving sources. While state-of-the-art approaches employ online clustering algorithms to solve the problem for moving sources, we propose an approximate Bayesian tracker and perform the association of each TF bin to the dominant source using the tracker's measurement-to-source association probabilities. Therefore, the choice of the underlying narrowband models and measurements for the tracker as well as the resulting tracking algorithm constitute the main contributions of this paper. The TF bin associations obtained from the tracker are then used to estimate the statistics of the source signals. The performance of the resulting BSS filters is compared to the performance of state-of-the-art sparsity-based and independent vector analysis-based BSS algorithms. Our proposed approach targets scenarios with at least two spatially separated microphone arrays, with known microphone positions and relative orientations. The framework also allows for efficient management of a time-varying number of sources.

Wireless Health Monitoring System for Rotor Eccentricity Faults Detection in Induction Machine

Condition monitoring and fault detection of induction machines become an important area of research. Many techniques have been applied in this field including vibration, thermal, chemical and acoustic emission monitoring, but Motor Current Signature Analysis monitoring techniques usually are applied to detect the various classes of induction machine faults such as rotor, short winding, eccentricity and bearing fault etc. This paper presents a wireless system detection for rotor eccentricity faults in induction machine based on LabVIEW platform. Moreover, it is demonstrated that eccentricity fault generates a series of low frequency components in the form of sidebands around the fundamental frequency and its harmonic. In addition, the amplitudes of those components increase in proportion to the load and fault severity. The Power Spectral Density techniques and Short Time Frequency Transform spectrogram of current signals are used to detect the presence of those fault signatures.

Learning to short-time Fourier transform in spectrum sensing

The future wireless communication will come up with a strict requirement on high spectral efficiency, developing novel algorithms for spectrum sensing with deep sensing capability will be more challenging. However, traditional expert feature-based spectrum sensing algorithms are lack of sufficient capability of self-learning and adaptability to unknown environments and complex cognitive tasks. To address this problem, we propose to build up a deep learning network to learn short time-frequency transformation (STFT), a basic entity of traditional spectrum sensing algorithms. Spectrum sensing based on the learning to STFT network is supposed to automatically extract features for communication signals and makes decisions for complex cognitive tasks meanwhile. The feasibility and performances of the designed learning network are verified by classifying signal modulation types in deep spectrum sensing applications.

Diagnosis of insomnia sleep disorder using short time frequency analysis of PSD approach applied on EEG signal using channel ROC-LOC

Insomnia is a sleep disorder in which the subject encounters problems in sleeping. The aim of this study is to identify insomnia events from normal or effected person using time frequency analysis of PSD approach applied on EEG signals using channel ROC-LOC. In this research article, attributes and waveform of EEG signals of Human being are examined. The aim of this study is to draw the result in the form of signal spectral analysis of the changes in the domain of different stages of sleep. The analysis and calculation is performed in all stages of sleep of PSD of each EEG segment. Results indicate the possibility of recognizing insomnia events based on delta, theta, alpha and beta segments of EEG signals.

Parameter estimation for fm signals in two stages: non uniform fast fourier transform (NUFFT) and short-time frequency transform

This study deals with estimating the parameters of the received signal of the poly phase in three orders. The process of estimation occurs in two stages: in the first stage, the high orders of the signal are estimated highly accurately using the Non Uniform FFT. The received signal passes through the componential match filter which has been estimated highly accurately. As it passes through this filter, the order of the ploy phase decreases. In the second stage, the components of the low order of the ploy phase signal are estimated highly accurately using the Short Time Frequency Transform (STFT) method. Finally, the findings of the stimulation in these two stages indicated the efficiency of the presented algorithm for the received signal parameters including the initial frequency, the chirp rate, and the change chirp rate.

Diagnosis of Nocturnal Frontal Lobe Epilepsy (NFLE) Sleep Disorder Using Short Time Frequency Analysis of PSD Approach Applied on EEG Signal

Diagnosis of Nocturnal Frontal Lobe Epilepsy (NFLE) Sleep Disorder Using Short Time Frequency Analysis of PSD Approach Applied on EEG Signal

A mode shape-based damage detection approach using laser measurement from a vehicle crossing a simply supported bridge

This paper presents a novel algorithm for bridge damage detection based on the mode shapes estimated from a passing vehicle. The bridge response at the moving coordinate is measured from an instrumented vehicle with laser vibrometers and accelerometers. A modified version of the Short Time Frequency Domain Decomposition method is applied to the measured responses. The bridge mode shape is estimated with high resolution, which is appropriate for damage detection. A damage index based on mode shape squares is used to detect the presence and location of the damage. A numerical case study of a half-car model passing over a bridge is described in this paper, which validates the performance of the proposed approach. Several damage scenarios are considered including different locations and severities. It is shown that the presence and location of the damage can be detected with acceptable accuracy when the vehicle is moving very slowly. In addition, the performance of the method using higher vehicle speeds is investigated and shows that the approach works well for speeds up to 8 m/s. The sensitivity of the algorithm to measurement noise is also studied by adding several levels of noise to the responses measured on the vehicle. Copyright © 2016 John Wiley & Sons, Ltd.

Detection of Rapid Eye Movement Disorder with the Help of Short Time Frequency Applied on EEG Signal

Detection of Rapid Eye Movement Disorder with the Help of Short Time Frequency Applied on EEG Signal

Design of a Novel SiC MOSFET Structure for EV Inverter Efficiency Improvement

Inverters for electric vehicle motor drive systems are essential in converting the battery's direct current into alternating current. Si(Silicon) IGBT that is commonly used in inverter modules have large Vce,sat and turn-off time due to p+ drain and tail current. Therefore, inverter modules consist of Si IGBT with relatively low efficiency. If we can use MOSFETs instead of IGBT in inverter modules, it is possible to achieve high efficiency because of short turn-off time and high operating frequency. Yet also has a problem; Si MOSFETs has large on-resistance compared to Si IGBTs. In this study, SiC(Silicon Carbide) was used to make MOSFETs instead of Si. Futhermore, an accumulation channel concept is adapted to a SiC trench MOSFET, namely Trench ACCUFET. Compared with conventional SiC trench MOSFETs, the novel SiC trench ACCUFET structure has not only lower on-resistance but also high breakdown voltage as shown by the simulation results. We fabricated the Trench ACCUFET for verification, and described improvements that is to be made.

Identification of bridge mode shapes using Short Time Frequency Domain Decomposition of the responses measured in a passing vehicle

This paper processes the signals from accelerometers mounted on a vehicle travelling over a bridge. Short Time Frequency Domain Decomposition (STFDD) is used to estimate bridge mode shapes from the dynamic response of the vehicle. In Frequency Domain Decomposition (FDD), several segments are defined on the bridge and the measurement is performed using two instrumented axles. Here, the FDD method is employed in a multi-stage procedure applied to the bridge segments in sequence. A rescaling process is used to construct the global mode shape vector. The performance of the proposed method is validated using numerical case studies. In other indirect bridge identification methods, the road profile may excite the vehicle, making it difficult to detect the bridge modes. This is addressed using two concepts: applying external excitation to the bridge and subtracting signals in the axles of successive trailers towed by the vehicle. The results obtained from the numerical investigation demonstrate that the proposed method can estimate the bridge mode shapes with acceptable accuracy. Sensitivity of the method to added white noise is also investigated.

Estimation of Subband Speech Correlations for Noise Reduction via MVDR Processing

Recently, it has been proposed to use the minimum-variance distortionless-response (MVDR) approach in single-channel speech enhancement in the short-time frequency domain. By applying optimal FIR filters to each subband signal, these filters reduce additive noise components with less speech distortion compared to conventional approaches. An important ingredient to these filters is the temporal correlation of the speech signals. We derive algorithms to provide a blind estimation of this quantity based on a maximum-likelihood and maximum a-posteriori estimation. To derive proper models for the inter-frame correlation of the speech and noise signals, we investigate their statistics on a large dataset. If the speech correlation is properly estimated, the previously derived subband filters discussed in this work show significantly less speech distortion compared to conventional noise reduction algorithms. Therefore, the focus of the experimental parts of this work lies on the quality and intelligibility of the processed signals. To evaluate the performance of the subband filters in combination with the clean speech inter-frame correlation estimators, we predict the speech quality and intelligibility by objective measures.

Measuring temporal, spectral and spatial changes in electrophysiological brain network connectivity

The topic of functional connectivity in neuroimaging is expanding rapidly and many studies now focus on coupling between spatially separate brain regions. These studies show that a relatively small number of large scale networks exist within the brain, and that healthy function of these networks is disrupted in many clinical populations. To date, the vast majority of studies probing connectivity employ techniques that compute time averaged correlation over several minutes, and between specific pre-defined brain locations. However, increasing evidence suggests that functional connectivity is non-stationary in time. Further, electrophysiological measurements show that connectivity is dependent on the frequency band of neural oscillations. It is also conceivable that networks exhibit a degree of spatial inhomogeneity, i.e. the large scale networks that we observe may result from the time average of multiple transiently synchronised sub-networks, each with their own spatial signature. This means that the next generation of neuroimaging tools to compute functional connectivity must account for spatial inhomogeneity, spectral non-uniformity and temporal non-stationarity. Here, we present a means to achieve this via application of windowed canonical correlation analysis (CCA) to source space projected MEG data. We describe the generation of time–frequency connectivity plots, showing the temporal and spectral distribution of coupling between brain regions. Moreover, CCA over voxels provides a means to assess spatial non-uniformity within short time–frequency windows. The feasibility of this technique is demonstrated in simulation and in a resting state MEG experiment where we elucidate multiple distinct spatio-temporal-spectral modes of covariation between the left and right sensorimotor areas.

Plasma Bursts in Deep Penetration Laser Welding

We present an experimental study of the deep penetration laser welding process which aims to analyze the plasma plume oscillations on a short time scale. Using the high-speed camera we show that the plasma comes out of the keyhole in the form of short bursts rather than the continuous flow. We detect these bursts as the short-time intensity oscillations of light emissions coming from the plasma plume. We determine the period of bursts using the statistical signal processing methods and the short-time frequency analysis. Finally, we compare the characteristics of plasma bursts and the geometry of resulting welds carried out on a 2 kW Yb:YAG laser welding machine for the steel workpiece and various welding parameters settings.

A New Method for Phase Difference Estimation Based on Time-Varying Signal Model

A new signal processing method for phase difference estimation was proposed based on time-varying signal model, whose frequency, amplitude and phase are time-varying. And then be applied Coriolis mass flowmeter signal. First, a bandpass filtering FIR filter was applied to filter the sensor output signal in order to improve SNR. Then, the signal frequency could be calculated based on short-time frequency estimation. Finally, by short window intercepting, the DTFT algorithm with negative frequency contribution was introduced to calculate the real-time phase difference between two enhanced signals. With the frequency and the phase difference obtained, the time interval of two signals was calculated. Simulation results show that the algorithms studied are efficient. Furthermore, the computation of algorithms studied is simple so that it can be applied to real-time signal processing for Coriolis mass flowmeter.

Harmonic Fractals in the Brain: Transient Tuning and Synchronic Coordination in the Quasi-Chaotic Background of Ongoing Neural EEG Activity

Despite the long-time that EEG has been used in neurophysiological characterization of the brain functioning and the diverse methods that has been used to enhance its predictive power, frequency resolution and time dependent description of EEG features, it remain at the level of gross depictions. By mean of collapsing time, through doing a power spectrum analysis, the focus is put on amplitude differences among wide EEG (delta, theta, alpha, beta and gamma) frequency bands or, without any further consideration of the mid- or long-term time course evolution, of the ever changing neural processes of the brain. To try to solve this, we took an exploratory approach that consider both weaknesses of the problem by taking into account, simultaneously, the synchrony of the time course of the EEG signal and the harmonic tuning among single narrow short-time frequency bands along the whole spectrum from 0,5 to 60Hz. To characterize the neural phenomena in these terms we cleaned and selectively filtered and transformed the EEG signal into music, and used the so-transformed EEG recording to find harmonic fractal elements in it during the execution of a number of exploratory basal and experimental tests in humans. We sustained our search on the fact that EEG signal can be described as a particular type of quasi-predictable noise, also called pink or fractal noise. We looked for detection of specific and transient synchronized and wide tuned elements of EEG brain activity characterized by high self-similarity or high values of statistical autocorrelation expressed in regular shaped associated plots. We found and characterized harmonic fractals coming from EEG of 5 subjects during the execution of three exploratory experimental conditions: basal resting state with eyes open and closed; a cognitive challenge consisting in the execution of an abbreviated version of the Raven visual intelligence test; and during four minutes playing an action video game. Results showed the presence of characteristic harmonic fractals in all the conditions tested and revealed intra and inter-individual differences that can eventually be used to characterize particular brain activity states related to specific sensory, or cognitive, processes or may be indicating individual differences in the way to process information, in particular situations, developmental stage, illness or neuropsychological disorders.

Correlation between the Keyhole Depth and the Frequency Characteristics of Light Emissions in Laser Welding

Abstract This paper presents a study of the frequency characteristics of the light emissions in laser beam welding. It is based on short-time frequency analysis of the light intensity oscillations and it results in the identification of the parameter directly correlated to the actual depth of penetration. Applicability of this parameter for characterization of the welds is demonstrated for a variety of welding parameters settings usually used in industrial practice. In addition, the method of short-time frequency analysis is used to detect the eigenfrequencies of keyhole oscillations which are hardly observable by other methods.

Spectral Analysis for Current and Temperature Measurements in Power Cables

This research aims to detect spectral properties under thermal and current variations for power cables. Therefore, spectral diversities are exposed under current unbalances and different load conditions through the spectral analysis techniques. Also, huge load variations are easily detected from the current signals in the time-frequency plane using the short-time frequency analysis. Hence, this study presents the determination of the frequency characteristics and spectral similarities between the phase currents and thermal variations.