Time-domain Samples Research Articles

Generative adversarial learning for improved data efficiency in underwater target classification

In the realms of the ocean, it becomes a formidable task to detect and classify the passive acoustic targets from the convoluted acoustic mixture confronted by the sonar frontend. Though the advances in deep learning driven by enormity of data and computational infrastructure have resulted in a tremendous leap in performance across various domains, passive sonar target recognition still remains an elusive task for the acoustic as well as signal processing communities. Various channel related artifacts together with the inherent difficulty in obtaining annotated data limit the target records required for training these massive supervised networks so as to yield an optimal performance. This demands models that can generalize well beyond the often sparse training instances. In order to address this issue, generative frameworks can be utilized to model the causal attributes of the target signature so that the network becomes tolerant to the distortions induced by the ambient noise and channel artifacts. This paper exploits the generative modelling capability of an Auxiliary Classifier Generative Adversarial Network (ACGAN) to construct a data-efficient underwater target classifier. These class-conditioned frameworks based on unsupervised representation learning can model the true data distribution using the latent attributes of the training data. In order to make the causal factors of variation more explicit, the raw time domain samples are transformed into joint time–frequency representations using filterbanks initialized at different perceptual scales. Experimental evaluation of the proposed system on target instances collected from diverse locations of the Indian Ocean yields promising results in terms of data efficiency, class confidence and classification accuracy.

Dual Attention Network for Pitch Estimation of Monophonic Music

The task of pitch estimation is an essential step in many audio signal processing applications. In this paper, we propose a data-driven pitch estimation network, the Dual Attention Network (DA-Net), which processes directly on the time-domain samples of monophonic music. DA-Net includes six Dual Attention Modules (DA-Modules), and each of them includes two kinds of attention: element-wise and channel-wise attention. DA-Net is to perform element attention and channel attention operations on convolution features, which reflects the idea of "symmetry". DA-Modules can model the semantic interdependencies between element-wise and channel-wise features. In the DA-Module, the element-wise attention mechanism is realized by a Convolutional Gated Linear Unit (ConvGLU), and the channel-wise attention mechanism is realized by a Squeeze-and-Excitation (SE) block. We explored three kinds of combination modes (serial mode, parallel mode, and tightly coupled mode) of the element-wise attention and channel-wise attention. Element-wise attention selectively emphasizes useful features by re-weighting the features at all positions. Channel-wise attention can learn to use global information to selectively emphasize the informative feature maps and suppress the less useful ones. Therefore, DA-Net adaptively integrates the local features with their global dependencies. The outputs of DA-Net are fed into a fully connected layer to generate a 360-dimensional vector corresponding to 360 pitches. We trained the proposed network on the iKala and MDB-stem-synth datasets, respectively. According to the experimental results, our proposed dual attention network with tightly coupled mode achieved the best performance.

Signal Subspace Reconstruction for DOA Detection Using Quantum-Behaved Particle Swarm Optimization

Spatial spectrum estimation, also known as direction of arrival (DOA) detection, is a popular issue in many fields, including remote sensing, radar, communication, sonar, seismic exploration, radio astronomy, and biomedical engineering. MUltiple SIgnal Classification (MUSIC) and Estimation Signal Parameter via Rotational Invariance Technique (ESPRIT), which are well-known for their high-resolution capability for detecting DOA, are two examples of an eigen-subspace algorithm. However, missed detection and estimation accuracy reduction often occur due to the low signal-to-noise ratio (SNR) and snapshot deficiency (small time-domain samples of the observed signal), especially for sources with different SNRs. To avoid the above problems, in this study, we develop a DOA detection approach through signal subspace reconstruction using Quantum-Behaved Particle Swarm Optimization (QPSO). In the developed scheme, according to received data, a noise subspace is established through performing an eigen-decomposition operation on a sampling covariance matrix. Then, a collection of angles randomly selected from the observation space are used to build a potential signal subspace on the basis of the steering matrix of the array. Afterwards, making use of the fact that the signal space is orthogonal to the noise subspace, a cost function, which contains the desired DOA information, is designed. Thus, the problem of capturing the DOA information can be transformed into the optimization of the already constructed cost function. In this respect, the DOA finding of multiple signal sources—that is, the multi-objective optimization problem—can be regarded as a single objective optimization problem, which can effectively reduce the probability of missed detection of the signals. Subsequently, the QPSO is employed to determine an optimal signal subspace by minimizing the orthogonality error so as to obtain the DOA. Ultimately, the performance of DOA detection is improved. An explicit analysis and derivation of the developed scheme are provided. The results of computer simulation show that the proposed scheme has superior estimation performance when detecting signals with very different SNR levels and small snapshots.

THE PROBABITY STATISTICS OF TRAIN DERAILMENT COEFFICIENTS BASED ON THE PSEUDO-EXCITATION METHOD

To explore the random characteristics of the derailment coefficient safety index of high-speed railway trains, the random vibration model of a vehicle bridge is established by multi-body dynamics theory and finite element method based on the pseudo excitation method (PEM) and extreme value theory. A vertical wheel/rail corresponding and lateral linear creeping wheel-rail relationship is used to build the dynamic equation of the train-bridge coupling system. The track irregularity is transformed into the superposition of harmonic loads at a series of frequencies by the pseudo excitation method so that the non-stationary problem is changed into a deterministic time history problem. The power spectra of the wheel-rail force are obtained by the separate iteration method. Based on the modulation function, the harmonic superposition method is used to add the wheel rail force at each time to obtain the wheel rail force time history sample. The probability distribution function of the maximum derailment coefficient is explored by the extreme value theory and its maximum value is obtained. The stationary characteristics of the wheel rail force power spectrum under different defection span ratios are discussed. Taking a 24.6-meter simply supported concrete beam as an example, we fitted the extreme value distribution of the derailment coefficient under various vehicle speeds. The results show that the extreme value distribution can well fit the extreme value of the derailment coefficient under different speeds. The higher the speed, the stronger the discrete type of the extreme value distribution of derailment coefficient, and the larger the derailment coefficient value corresponding to 99.73% confidence level, which is greater than the maximum of time-domain samples. In the analysis cases, the bridge deformation has little influence on the stationary characteristics of the power spectra of the wheel rail force.

On-chip sampling of optical fields with attosecond resolution

Time-domain sampling of arbitrary electric fields with sub-cycle resolution enables a complete time-frequency analysis of a system's response to electromagnetic illumination. This provides access to dynamic information that is not provided by absorption spectra alone, and has recently been shown through measurements in the infrared that time-domain optical-field sampling offers significant improvements with regard to molecular sensitivity and limits of detection compared to traditional spectroscopic methods. Despite the many scientific and technological motivations, time-domain, optical-field sampling systems operating in the visible to near-infrared spectral regions are seldom accessible, requiring large driving pulse energies, and large laser amplifier systems, bulky apparatuses, and vacuum environments. Here, we demonstrate an all-on-chip, optoelectronic device capable of sampling arbitrary, low-energy, near-infrared waveforms under ambient conditions. Our solid-state integrated detector uses optical-field-driven electron emission from resonant nanoantennas to achieve petahertz-level switching speeds by generating on-chip attosecond electron bursts. These bursts are used to probe the electric field of weak optical transients. We demonstrated our devices by sampling the electric field of a ~5 fJ, broadband near-infrared ultrafast laser pulse using a ~50 pJ near-infrared driving pulse. Our sampling measurements recovered the weak optical transient as well as localized plasmonic dynamics of the emitting nanoantennas $in~situ$. This field-sampling device--with its compact footprint and low pulse-energy requirements--offers opportunities in a variety of applications, including: broadband time-domain spectroscopy in the molecular fingerprint region, time-domain analysis of nonlinear phenomena, and detailed studies of strong-field light-matter interactions.

KU-HAR: An open dataset for heterogeneous human activity recognition

• Describes the content of the KU-HAR dataset, and the associated data collection procedure in details. • Provides a comparison with other contemporary HAR datasets. • Presents some HAR classification results involving the KU-HAR dataset. In Artificial Intelligence, Human Activity Recognition (HAR) refers to the capability of machines to identify various activities performed by the users. The knowledge acquired from these recognition systems is integrated into many applications where the associated device uses it to identify actions or gestures and performs predefined tasks in response. HAR requires a large quantity of meticulously collected action data of diverse nature to fuel its learning algorithms. This paper aims to introduce a new set of HAR data collected from 90 participants who are 18 to 34 years old. The constructed dataset is named KU-HAR, and it contains 1945 raw activity samples that belong to 18 different classes. We used built-in smartphone sensors (accelerometer and gyroscope) to collect these HAR samples. Apart from the original (raw) time-domain samples, the dataset contains 20,750 subsamples (extracted from them) provided separately, each containing 3 seconds of data of the corresponding activity. Some classification results have been provided as well to propound the quality of the proposed dataset. The acquired results show that Random Forest (RF), an ensemble learning algorithm, can classify the subsamples with almost 90% accuracy. This dataset will enable smartphones and other smart devices to identify new activities and help researchers to design more delicate models based on practical HAR data.

Impact of short-time fourier transform parameters on the accuracy of EMI spectra estimates in the 2-150 kHz supraharmonic interval

Methods based on time-domain sampling and successive tapering and Fourier analysis, used in the “harmonics” frequency interval up to 2.4kHz, cannot be automatically extended to the 2-150 kHz interval, in particular considering the emissions of Switched-Mode Power Supplies (SMPSs) in a modern smart grid or low-voltage distribution scenario. Emissions are substantially impulsive and oscillatory superposed to the mains signal, possibly featuring multiple switching fundamentals and their harmonics with unstable or purposely changing fundamental frequency. Fourier analysis (namely Discrete Fourier Transform) is commonly used to characterize the spectrum of emissions from time-domain data, but the influence of its settings should be verified for accuracy and systematic errors, so to establish the uncertainty of spectral estimates. After an initial theoretical and practical discussion of characteristics, objectives and weak points, a verification is carried out with the help of several measurements performed in different operating conditions on a set of six different SMPSs, evaluating systematic error and uncertainty, and defining a set of recommendations.

Flexible and Efficient DSP-Assisted Subcarrier Multiplexing for an Analog Mobile Fronthaul

The digital formation of an analog subcarrier multiplex employing in combination both a technique using pre-IFFT frequency-domain samples and one using post-IFFT time-domain samples is proposed and demonstrated. This combined technique enables a compromise for sampling rate requirements, while maintaining low complexity and good performance.

Robust wideband adaptive beamforming based on covariance matrix reconstruction in the spatial‐frequency domain

Without the response variation constraint, the main-lobe responses of wideband beamforming usually vary with the frequencies of interest, and the beam pointing may deviate from the incident angle of the desired signal. If the interference-plus-noise covariance matrix (INCM) is directly replaced by the sample covariance matrix, the desired signal in the sample data may result in self-null problem, which can lead to the degradation of the performance of wideband beamformers. Through frequency division, a novel wideband adaptive beamforming is proposed by reconstructing the INCM and solving the convex optimisation problem. Using fast Fourier transform (FFT), the time-domain samples are transformed into the frequency field, and the covariance matrix of each subband can be obtained from the frequency-domain data. In the spatial-frequency domain, the INCM and signal-plus-noise covariance matrix can be respectively reconstructed by integrating the stacked steering vector and the Capon spatial spectrum over the corresponding angular region. Then, we establish the optimisation criteria, that is, minimising the output interference-plus-noise power, minimising the main-lobe spatial response variation, and constraining the response deviation of the desired signal. A novel convex optimisation model can be established, and then the weight coefficients for time-domain wideband beamforming can be obtained by solving the optimisation problem. The simulation and experimental results demonstrate that the proposed beamformer can provide better performance compared with other tested beamformers and maintain good robustness in various applications.

OSNR monitoring based on a low-bandwidth coherent receiver and LSTM classifier.

Optical signal-to-noise ratio (OSNR) monitoring is one of the core tasks of advanced optical performance monitoring (OPM) technology, which plays an essential role in future intelligent optical communication networks. In contrast to many regression-based methods, we convert the continuous OSNR monitoring into a classification problem by restricting the outputs of the neural network-based classifier to discrete OSNR intervals. We also use a low-bandwidth coherent receiver for obtaining the time domain samples and a long short-term memory (LSTM) neural network as the chromatic dispersion-resistant classifier. The proposed scheme is cost efficient and compatible with our previously proposed multi-purpose OPM platform. Both simulation and experimental verification show that the proposed OSNR monitoring technique achieves high classification accuracy and robustness with low computational complexity.

Joint Two-Dimensional Deception Countering ISAR via Frequency Diverse Array

Deception against inverse synthetic aperture radar (ISAR) has been a topic of active research in electronic warfare. However, most of the studies mainly focus on producing one-dimensional deception, which is inadequate for two-dimensional imaging system like ISAR. Thus, in this letter, a joint two-dimensional ISAR deception based on frequency diverse array (FDA) and interrupted sampling is proposed. Compared with conventional phased-array (PA) and single-channel methods, FDA permits more degrees of freedom in producing range deception. Combined with interrupted sampling in the slow time domain, the proposed method is anticipated to produce a group of false targets in both range and cross-range directions. The benefit is easy to control the number and distribution of false targets with high efficiency and less computation burden. Simulation results verify the effectiveness of the proposed method.

Multi-Scale Sample Entropy-Based Energy Moment Features Applied to Fault Classification

Rolling element bearings are crucial, of high failure rate and easy damage parts of rotating machinery, and significantly affect safe and reliable production processes. Much more attention has been focused on fault diagnosis of rolling element bearings in recent years. This article presents a novel feature extraction scheme for the classification of multiple bearing faults. Multi-Scale Sample Entropy (M-SSampEn) is combined with Energy Moment (EM) to construct a time-domain Multi-Scale Sample Entropy-based Energy Moment (M-SSampEn-EM) feature extractor. The classifier model for the proposed fault classification system has been built using the Least Square Support Vector Machine (LS-SVM). The M-SSampEn-EM feature extractor is used to capture two-dimensional representative eigenvectors from multiple fault classes' bearing vibration data. Its separability performance is ensured by optimizing the feature-extraction parameters, including the Embedded Dimension and the Tolerance, etc. The LS-SVM classifier is also compared with two Neural Network (NN)-based classifiers, i.e., Radial Basis Function NN (RBFNN) and Probabilistic NN (PNN), to show it better generalization performance on bearing fault classification. The experimental study verifies the excellent capacity of the proposed approach in bearing fault classification.

Spectrally Sparse Signal Recovery via Hankel Matrix Completion With Prior Information

This paper studies the problem of reconstructing spectrally sparse signals from a small random subset of time domain samples via low-rank Hankel matrix completion with the aid of prior information. By leveraging the low-rank structure of spectrally sparse signals in the lifting domain and the similarity between the signals and their prior information, we propose a convex method to recover the undersampled spectrally sparse signals. The proposed approach integrates the inner product of the desired signal and its prior information in the lift domain into vanilla Hankel matrix completion, which maximizes the correlation between the signals and their prior information. Theoretical analysis indicates that when the prior information is reliable, the proposed method has a better performance than vanilla Hankel matrix completion, which reduces the number of measurements by a logarithmic factor. We also develop an ADMM algorithm to solve the corresponding optimization problem. Numerical results are provided to verify the performance of proposed method and corresponding algorithm.

Time Domain Sampling of the Radial Functions in Spherical Harmonics Expansions

Spherical harmonics representations are widely adopted in applications such as analysis, manipulation, and synthesis of wave fields that are either captured or simulated. Recent studies have shown that, for broadband fields, a time domain representation of spherical harmonics expansions can benefit from computational efficiency and favorable transient properties. For practical usage, an accurate discrete-time modeling of spherical harmonics expansions is indispensable. The main challenge is to model the so called radial functions which describe the radial and frequency dependencies. In homogeneous cases, they are commonly realized as finite impulse response filters where the coefficients are obtained by sampling the continuous-time representations. This article investigates the temporal and spectral properties of the resulting discrete-time radial functions. The spectral distortions caused by aliasing are evaluated both analytically and numerically, revealing the influence of the distance from the expansion center, sampling frequency, and fractional sample delay. It is also demonstrated how the aliasing can be reduced by employing a recently introduced band limitation method.

Projected Wirtinger Gradient Descent for Digital Waves Reconstruction

This work attempts to recover digital signals from a few stochastic samples in time domain. The target signal is the linear combination of one-dimensional complex sine components with R different but continuous frequencies. These frequencies control the continuous values in the domain of normalized frequency [0, 1), contrary to the previous research into compressed sensing. To recover the target signal, the problem was transformed into the completion of a low-rank structured matrix, drawing on the linear property of the Hankel matrix. Based on the completion of the structured matrix, the authors put forward a feasible-point algorithm, analyzed its convergence, and speeded up the convergence with the fast iterative shrinkage-thresholding (FIST) algorithm. The initial algorithm and the speed up strategy were proved effective through repeated numerical simulations. The research results shed new lights on the signal recovery in various fields.

Error Analysis for Status Update From Sensors With Temporally and Spatially Correlated Observations

This paper studies the status update performance in wireless sensor networks when status, describing the physical reality that is being sensed, is temporally and spatially correlated. The status is modeled as a time-varying Gauss-Markov Random Field (GMRF), whereby the estimation error of status update at the fusion center is analyzed. The transmission latency introduced by wireless networks is modeled as exponentially distributed random variables. We extend the existing queuing analysis results for Age of Information (AoI) with uncorrelated sources to GMRF in the considered scenario. Closed-form expressions of average remote estimation error are obtained for both one- and two-dimensional GMRFs assuming the exponential time-correlation function, both First-Come First-Served (FCFS) and Last-Come First-Served (LCFS) service disciplines, and a single wireless link. The analytical results are then extended to scenarios wherein multi-packet reception, i.e., multiple concurrent wireless links, is enabled; the difficulty of analyzing obsolete updates in this case is addressed leveraging a reasonable approximation validated by theoretical analysis in the regime where the number of sensors is far more than that of wireless links. Monte-Carlo simulation results are also presented which agree with our theoretical analysis. Based on the results, optimal time and spatial domain sampling rates (e.g., sensor density) can be obtained, providing helpful guidance to wireless sensor deployment.

Direct Detection of Pilot Carrier-Assisted DMT Signals With Pre-Phase Compensation and Imaginary Noise Suppression

We propose and experimentally demonstrate a cost-effective pilot carrier assisted direct detection scheme based on discrete multitone (DMT) modulation format. The cost of the proposed scheme can be substantially reduced by two ways. First, we use distributed feedback (DFB) laser with large linewidth (~10 MHz) as the pilot carrier. Secondly, we design a low-complexity pre-phase compensation (PPC) algorithm combined with imaginary noise suppression (INS), in order to mitigate the effect of phase noise and consequently enhance the signal to noise ratio (SNR). Different from the conventional channel equalization schemes for the DMT signal, we pre-compensate the phase noise in time domain before fast Fourier transform (FFT) demodulation at the receiver side. The PPC algorithm is implemented with an average operation, instead of complex multipliers. Since DMT signal is a real-valued signal, the imaginary part of time-domain samples after the PPC process is only noise, which can be removed for the SNR enhancement. Simulation results indicates that 3-dB SNR improvement can be obtained, which agrees well with the theoretical analysis. Meanwhile, our experimental results show that the combination of PPC and INS can significantly improve the laser linewidth tolerance of transmission system, leading to the possible use of DFB laser, instead of external cavity laser (ECL), as the pilot carrier. Finally, we successfully demonstrate 4×50-Gb/s DMT signal transmission over 400-km standard single mode fiber (SSMF) by using DFB lasers as signal and pilot carriers, respectively, with sufficient margin to reach 20% forward error correction (FEC) threshold.

Sparsity-inducing frequency-domain adaptive line enhancer for unmanned underwater vehicle sonar

The radiated lines from underwater targets are a significant feature for passive sonar detection. Since the spatial and time processing gains are limited in an unmanned underwater vehicle (UUV) sonar, an adaptive line enhancer (ALE) is usually employed as a preprocessing step to enhance the signal-to-noise ratio (SNR) of the lines. However, the original time-domain ALE based on the least mean square (LMS) algorithm suffers from the weight noise in the adaptation, which limits the SNR gain severely. Inspired by the frequency-domain sparsity of the lines, a sparsity-inducing ALE is developed to break through this limitation. The proposed ALE is implemented in the frequency domain and a reweighted L1-norm sparse penalty is incorporated into the cost function of the adaptation. By means of the sparse penalty, a zero-attraction term is added in the adaptation, which results in the suppression of the weight noise. The SNR gain is thus enhanced. In the proposed ALE, the sliding discrete Fourier transform (SDFT) technique is utilized to transfer the time-domain input samples to the frequency domain, and the computational complexity of the proposed ALE is thus comparable to that of the original one, which meets the energy efficiency requirement of UUV. Simulation results demonstrate that the SNR gain of the proposed ALE is over 9 dB higher than that of the original ALE with an input SNR of −25 dB. Experimental data processing also verifies the superiority of the proposed ALE.

Performance and Low-Frequency Noise of 22-nm FDSOI Down to 4.2 K for Cryogenic Applications

This work presents the performance and low-frequency noise (LFN) of 22-nm fully-depleted silicon-on-insulator (FDSOI) CMOS technology. The experimental measurements and the analysis are performed as a function of temperature for the first time, focusing on cryogenic applications, down to 4.2 K. The back bias impact on device performance is evaluated. The results reveal that the threshold voltage tuning is found to be temperature independent, allowing extra drain current improvement. This is particularly interesting for short channel devices, whose drain current gain with temperature lowering is expected to be smaller in comparison with long channel MOSFETs. LFN is characterized by means of time-domain current sampling measurements. Moderate and strong inversion regimes are investigated. The carrier number with correlated mobility fluctuations model can well describe the 1/f noise behavior down to 4.2 K. The physical origin behind the drain current noise-to-signal power augmentation with temperature lowering could be mainly attributed to the normalized transconductance improvement.

An Interference Suppression Method for Multistatic Radar Based on Noise Subspace Projection

The mainlobe jamming suppression technique is one of the key techniques of radar electronic counter-countermeasure (ECCM). In the view of multistatic radar, the interference signals received with different view angles would be highly coherent, on the contrary, the target echoes would be independent due to the radar cross section (RCS) of target would randomly fluctuate as the variation of view angle. As a consequence, After the interference signals have been aligned, a steering vector can be found of interference signals, the energy of the interference signals will be concentrated in a single-rank subspace whereas the energy of the target echo signals will be scattered throughout the whole signal space of the multistatic radar. By exploiting this phenomenon, a new mainlobe interference suppression method for multistatic radar is proposed. In this method, the projection matrix is trained based on the time domain sampling data. The interference signals received by multiple node radars with different view angles can be effectively suppressed through the noise subspace projection operation. Then, a general likelihood ratio test detector used for the residual signals after projection operation is derived. The results of simulations indicate that the proposed method can effectively suppress various kinds of mainlobe jamming and improve the detection performance. Furthermore, this method is adaptive to the variation of the interference pattern and the geometric configuration of the node array.