Implementation of Hybrid Deep Reinforcement Learning Technique for Speech Signal Classification

Classification of speech signals is a vital part of speech signal processing systems.With the advent of speech coding and synthesis,the classification of the speech signal is made accurate and faster.Conventional methods are considered inaccurate due to the uncertainty and diversity of speech signals in the case of real speech signal classification.In this paper,we use efficient speech signal classification using a series of neural network classifiers with reinforcement learning operations.Prior classification of speech signals,the study extracts the essential features from the speech signal using Cepstral Analysis.The features are extracted by converting the speech waveform to a parametric representation to obtain a relatively minimized data rate.Hence to improve the precision of classification,Generative Adversarial Networks are used and it tends to classify the speech signal after the extraction of features from the speech signal using the cepstral coefficient.The classifiers are trained with these features initially and the best classifier is chosen to perform the task of classification on new datasets.The validation of testing sets is evaluated using RL that provides feedback to Classifiers.Finally,at the user interface,the signals are played by decoding the signal after being retrieved from the classifier back based on the input query.The results are evaluated in the form of accuracy,recall,precision,f-measure,and error rate,where generative adversarial network attains an increased accuracy rate than other methods:Multi-Layer Perceptron,Recurrent Neural Networks,Deep belief Networks,and Convolutional Neural Networks.

Generative Adversarial Network-Based Electromagnetic Signal Classification: A Semi- Supervised Learning Framework

Generative adversarial network(GAN)has achieved great success in many fields such as computer vision,speech processing,and natural language processing,because of its powerful capabilities for generating realistic samples.In this paper,we introduce GAN into the field of electromagnetic signal classification(ESC).ESC plays an important role in both military and civilian domains.However,in many specific scenarios,we can’t obtain enough labeled data,which cause failure of deep learning methods because they are easy to fall into over-fitting.Fortunately,semi-supervised learning(SSL)can leverage the large amount of unlabeled data to enhance the classification performance of classifiers,especially in scenarios with limited amount of labeled data.We present an SSL framework by incorporating GAN,which can directly process the raw in-phase and quadrature(IQ)signal data.According to the characteristics of the electromagnetic signal,we propose a weighted loss function,leading to an effective classifier to realize the end-to-end classification of the electromagnetic signal.We validate the proposed method on both public RML2016.04c dataset and real-world Aircraft Communications Addressing and Reporting System(ACARS)signal dataset.Extensive experimental results show that the proposed framework obtains a significant increase in classification accuracy compared with the state-of-the-art studies.

Signal classification method based on data mining formulti-mode radar

For the multi-mode radar working in the modern electronicbattlefield, different working states of one single radar areprone to being classified as multiple emitters when adoptingtraditional classification methods to process intercepted signals,which has a negative effect on signal classification. A classificationmethod based on spatial data mining is presented to address theabove challenge. Inspired by the idea of spatial data mining, theclassification method applies nuclear field to depicting the distributioninformation of pulse samples in feature space, and digs out thehidden cluster information by analyzing distribution characteristics.In addition, a membership-degree criterion to quantify the correlationamong all classes is established, which ensures classificationaccuracy of signal samples. Numerical experiments show that thepresented method can effectively prevent different working statesof multi-mode emitter from being classified as several emitters,and achieve higher classification accuracy.

Ultrasonic signal classification based on ambiguity plane feature

Ambiguity function （AF） is proposed to represent ultrasonic signal to resolve the preprocessing problem of different center frequencies and different arriving times among ultrasonic signals for feature extraction, as well as offer time-frequency features for signal classification. Moreover, Karhunen-Loeve （K-L） transform is considered to extract signal features from ambiguity plane, and then the features are presented to probabilistic neural network （PNN） for signal classification. Experimental results show that ambiguity function eliminates the difference of center frequency and arriving time existing in ultrasonic signals, and ambiguity plane features extracted by K-L transform describe the signal of different classes effectively in a reduced dimensional space. Classification result suggests that the ambiguity plane features obtain better performance than the features extracted by wavelet transform （WT）.

Data association based on target signal classification information

In most of the passive tracking systems, only the target kinematical information is used in the measurement-to-track association, which results in error tracking in a multitarget environment, where the targets are too close to each other. To enhance the tracking accuracy, the target signal classification information （TSCI） should be used to improve the data association. The TSCI is integrated in the data association process using the JPDA （joint probabilistic data association）. The use of the TSCI in the data association can improve discrimination by yielding a purer track and preserving continuity. To verify the validity of the application of TSCI, two simulation experiments are done on an air target-tracing problem, that is, one using the TSCI and the other not using the TSCI. The final comparison shows that the use of the TSCI can effectively improve tracking accuracy.

Signal classification system using global-local feature extraction algorithm

A continuous wavelet transform(CWT)and globallocal feature(GLF)extraction-based signal classificationalgorithm is proposed to improve the signal classification accuracy.First,the CWT is utilized to generate the timefrequency scalogram.Then,the GLF extraction method is proposed to extract features from the time-frequency scalogram.Finally,a classification method based on the support vector machine(SVM)is proposed to classify the extracted features.Experimental results show that the extended binary phase shift keying(EBPSK)bit error rate(BER)of the proposed classification algorithm is1.3x10_5under the environment of additional white Gaussian noise with the signal-to-noise ratio of-3dB,which is24times lower than that of the SVM-based signal classification method.Meanwhile,the BER using the GLF extraction method is13times lower than the one using the global feature extraction method and24times lower than the one using the local feature extraction method.

Few-shot electromagnetic signal classification:A data union augmentation method

Deep learning has been fully verified and accepted in the field of electromagnetic signal classification. However, in many specific scenarios, such as radio resource management for aircraft communications, labeled data are difficult to obtain, which makes the best deep learning methods at present seem almost powerless, because these methods need a large amount of labeled data for training. When the training dataset is small, it is highly possible to fall into overfitting, which causes performance degradation of the deep neural network. For few-shot electromagnetic signal classification, data augmentation is one of the most intuitive countermeasures. In this work, a generative adversarial network based on the data augmentation method is proposed to achieve better classification performance for electromagnetic signals. Based on the similarity principle, a screening mechanism is established to obtain high-quality generated signals. Then, a data union augmentation algorithm is designed by introducing spatiotemporally flipped shapes of the signal. To verify the effectiveness of the proposed data augmentation algorithm, experiments are conducted on the RADIOML 2016.04C dataset and real-world ACARS dataset. The experimental results show that the proposed method significantly improves the performance of few-shot electromagnetic signal classification.

Meta-Heuristic Optimized Hybrid Wavelet Features for Arrhythmia Classification

The non-invasive evaluation of the heart through EectroCardioG-raphy(ECG)has played a key role in detecting heart disease.The analysis of ECG signals requires years of learning and experience to interpret and extract useful information from them.Thus,a computerized system is needed to classify ECG signals with more accurate results effectively.Abnormal heart rhythms are called arrhythmias and cause sudden cardiac deaths.In this work,a Computerized Abnormal Heart Rhythms Detection(CAHRD)system is developed using ECG signals.It consists of four stages;preprocessing,feature extraction,feature optimization and classifier.At first,Pan and Tompkins algorithm is employed to detect the envelope of Q,R and S waves in the preprocessing stage.It uses a recursive filter to eliminate muscle noise,T-wave interference and baseline wander.As the analysis of ECG signal in the spatial domain does not provide a complete description of the signal,the feature extraction involves using frequency contents obtained from multiple wavelet filters;bi-orthogonal,Symlet and Daubechies at different resolution levels in the feature extraction stage.Then,Black Widow Optimization(BWO)is applied to optimize the hybrid wavelet features in the feature optimization stage.Finally,a kernel based Support Vector Machine(SVM)is employed to classify heartbeats into five classes.In SVM,Radial Basis Function(RBF),polynomial and linear kernels are used.A total of∼15000 ECG signals are obtained from the Massachusetts Institute of Technology-Beth Israel Hospital(MIT-BIH)arrhythmia database for performance evaluation of the proposed CAHRD system.Results show that the proposed CAHRD system proved to be a powerful tool for ECG analysis.It correctly classifies five classes of heartbeats with 99.91%accuracy using an RBF kernel with 2nd level wavelet coefficients.The CAHRD system achieves an improvement of∼6%over random projections with the ensemble SVM approach and∼2%over morphological and ECG segment based features with the RBF classifier.

Modified MUSIC estimation for correlated signals with compressive sampling arrays

This paper addresses the issue of the direction of arrival （DOA） estimation under the compressive sampling （CS） framework. A novel approach, modified multiple signal classification （MMUSIC） based on the CS array （CSA-MMUSIC）, is proposed to resolve the DOA estimation of correlated signals and two closely adjacent signals. By using two random CS matrices, a large size array is compressed into a small size array, which effectively reduces the number of the front end circuit. The theoretical analysis demonstrates that the proposed approach has the advantages of low computational complexity and hardware structure compared to other MMUSIC approaches. Simulation results show that CSAMMUSIC can possess similar angular resolution as MMUSIC.

Fast BSC-based algorithm for near-field signal localization via uniform circular array

In this paper,we propose a beam space coversion(BSC)-based approach to achieve a single near-field signal local-ization under uniform circular array(UCA).By employing the centro-symmetric geometry of UCA,we apply BSC to extract the two-dimensional(2-D)angles of near-field signal in the Van-dermonde form,which allows for azimuth and elevation angle estimation by utilizing the improved estimation of signal para-meters via rotational invariance techniques(ESPRIT)algorithm.By substituting the calculated 2-D angles into the direction vec-tor of near-field signal,the range parameter can be conse-quently obtained by the 1-D multiple signal classification(MU-SIC)method.Simulations demonstrate that the proposed al-gorithm can achieve a single near-field signal localization,which can provide satisfactory performance and reduce computational complexity.

Classification of Epileptic Electroencephalograms Using Time-Frequency and Back Propagation Methods

Today,electroencephalography is used to measure brain activity by creating signals that are viewed on a monitor.These signals are frequently used to obtain information about brain neurons and may detect disorders that affect the brain,such as epilepsy.Electroencephalogram(EEG)signals are however prone to artefacts.These artefacts must be removed to obtain accurate and meaningful signals.Currently,computer-aided systems have been used for this purpose.These systems provide high computing power,problem-specific development,and other advantages.In this study,a new clinical decision support system was developed for individuals to detect epileptic seizures using EEG signals.Comprehensive classification results were obtained for the extracted filtered features from the time-frequency domain.The classification accuracies of the time-frequency features obtained from discrete continuous transform(DCT),fractional Fourier transform(FrFT),and Hilbert transform(HT)are compared.Artificial neural networks(ANN)were applied,and back propagation(BP)was used as a learning method.Many studies in the literature describe a single BP algorithm.In contrast,we looked at several BP algorithms including gradient descent with momentum(GDM),scaled conjugate gradient(SCG),and gradient descent with adaptive learning rate(GDA).The most successful algorithm was tested using simulations made on three separate datasets(DCT_EEG,FrFT_EEG,and HT_EEG)that make up the input data.The HT algorithm was the most successful EEG feature extractor in terms of classification accuracy rates in each EEG dataset and had the highest referred accuracy rates of the algorithms.As a result,HT_EEG gives the highest accuracy for all algorithms,and the highest accuracy of 87.38%was produced by the SCG algorithm.

Robust signal recognition algorithm based on machine learning in heterogeneous networks

There are various heterogeneous networks for terminals to deliver a better quality of service. Signal system recognition and classification contribute a lot to the process. However, in low signal to noise ratio（SNR） circumstances or under time-varying multipath channels, the majority of the existing algorithms for signal recognition are already facing limitations. In this series, we present a robust signal recognition method based upon the original and latest updated version of the extreme learning machine（ELM） to help users to switch between networks. The ELM utilizes signal characteristics to distinguish systems. The superiority of this algorithm lies in the random choices of hidden nodes and in the fact that it determines the output weights analytically, which result in lower complexity. Theoretically, the algorithm tends to offer a good generalization performance at an extremely fast speed of learning. Moreover, we implement the GSM/WCDMA/LTE models in the Matlab environment by using the Simulink tools. The simulations reveal that the signals can be recognized successfully to achieve a 95% accuracy in a low SNR（0 dB） environment in the time-varying multipath Rayleigh fading channel.

Application of Local Wave Time-Frequency Spectrum and Neural Networks to Fault Classification in Rotating Machine

A new method of fault analysis and detection by signal classification inrotating machines is presented. The Local Wave time-frequency spectrum which is a new method forprocessing a non-stationary signal is used to produce the representation of the signal. This methodallows the decomposition of one-dimensional signals into intrinsic mode functions (IMFs) usingempirical mode decomposition and the calculation of a meaningful multi-component instantaneousfrequency. Applied to fault signals , it provides new time-frequency attributes. Then the momentsand margins of the time-frequency spectrum are calculated as the feature vectors. The probabilisticneural network is used to classify different fault modes. The accuracy and robustness of theproposed methods is investigated on signals obtained during the different fault modes (early rub,loose, misalignment of the rotor).

Blind channel estimation for multiple antenna OFDM system subject to unknown carrier frequency offset

The problem of channel estimation for multiple an- tenna orthogonal frequency division multiplexing （OFDM） systems subject to unknown carrier frequency offset （CFO） is addressed. Multiple signal classification （MUSIC）-Iike algorithm, which generally has been used for direction estimation or frequency estimation, is used for channel estimation in multiple antenna OFDM systems. A reduced dimensional （RD）-MUSIC based algorithm for channel estimation is proposed in multiple antenna OFDM systems with unknown CFO. The Cramer-Rao bound （CRB） of channel estimation in multiple antenna OFDM systems with unknown CFO is derived. The proposed algorithm has a superior performance of channel estimation compared with the Capon method and the least squares method.

FAST MUSIC SPECTRUM PEAK SEARCH VIA METROPOLIS-HASTINGS SAMPLER

A fast MUltiple SIgnal Classification (MUSIC) spectrum peak search algorithm is devised, which regards the power of the MUSIC spectrum function as target distribution up to a constant of proportionality, and uses Metropolis-Hastings (MH) sampler, one of the most popular Markov Chain Monte Carlo (MCMC) techniques, to sample from it. The proposed method reduces greatly the tremendous computation and storage costs in conventional MUSIC techniques i.e., about two and four orders of magnitude in computation and storage costs under the conditions of the experiment in the paper respectively.

ISAR cross-range scaling based on the MUSIC technique

Cross-range scaling plays an important role in the inverse synthetic aperture radar(ISAR) imaging. Many of the published cross-range scaling algorithms are based on the fast Fourier transformation(FFT). However, the FFT technique is resolution limited, so that the FFT-based algorithms will fail in the rotation velocity(RV) estimation of the slow rotation target. In this paper,we propose an accurate cross-range scaling algorithm based on the multiple signal classification(MUSIC) method. We first select some range bins with the mono-component linear frequency modulated(LFM) signal model. Then, we dechirp the signal of each selected range bin into the form of sinusoidal signal, and utilize the super-resolution MUSIC technique to accurately estimate the frequency. After processing all the range bins, a linear relationship related to the RV can be obtained. Eventually, the ISAR image can be scaled. The proposal can precisely estimate the small RV of the slow rotation target with low computational complexity. Furthermore, the proposal can also be used in the case of cross-range scaling for the sparse aperture data. Experimental results with the simulated and raw data validate the superiority of the novel method.

Efficient hybrid method for time reversal superresolution imaging

An efficient hybrid time reversal（TR） imaging method based on signal subspace and noise subspace is proposed for electromagnetic superresolution detecting and imaging. First, the locations of targets are estimated by the transmitting-mode decomposition of the TR operator（DORT） method employing the signal subspace. Then, the TR multiple signal classification（TR-MUSIC）method employing the noise subspace is used in the estimated target area to get the superresolution imaging of targets. Two examples with homogeneous and inhomogeneous background mediums are considered, respectively. The results show that the proposed hybrid method has advantages in CPU time and memory cost because of the combination of rough and fine imaging.

Polarization quaternion DOA estimation based on vector MISC array

This paper examines the direction of arrival(DOA)estimation for polarized signals impinging on a sparse vector sensor array which is based on the maximum interelement spacing constraint(MISC).The vector array effectively utilizes the polarization domain information of incident signals,and the quaternion model is adopted for signals polarization characteristic maintenance and computational burden reduction.The features of MISC arrays are crucial to the mutual coupling effects reduction and higher degrees of freedom(DOFs).The quaternion data model based on vector MISC arrays is established,which extends the scalar MISC array into the vector MISC array.Based on the model,a quaternion multiple signal classification(MUSIC)algorithm based on vector MISC arrays is proposed for DOA estimation.The algorithm combines the advantages of the quaternion model and the vector MISC array to enhance the DOA estimation performance.Analytical simulations are performed to certify the capability of the algorithm.

Computationally efficient direction finding using polynomial rooting with reduced-order and real-valued computations

The root multiple signal classification（root-MUSIC） algorithm is one of the most important techniques for direction of arrival（DOA） estimation. Using a uniform linear array（ULA） composed of M sensors, this method usually estimates L signal DOAs by finding roots that lie closest to the unit circle of a（2M-1）-order polynomial, where L 〈 M. A novel efficient root-MUSIC-based method for direction estimation is presented, in which the order of polynomial is efficiently reduced to 2L. Compared with the unitary root-MUSIC（U-root-MUSIC） approach which involves real-valued computations only in the subspace decomposition stage, both tasks of subspace decomposition and polynomial rooting are implemented with real-valued computations in the new technique,which hence shows a significant efficiency advantage over most state-of-the-art techniques. Numerical simulations are conducted to verify the correctness and efficiency of the new estimator.

A Time-Frequency Associated MUSIC Algorithm Research on Human Target Detection by Through-Wall Radar

In this paper,a time-frequency associated multiple signal classification(MUSIC)al-gorithm which is suitable for through-wall detection is proposed.The technology of detecting hu-man targets by through-wall radar can be used to monitor the status and the location information of human targets behind the wall.However,the detection is out of order when classical MUSIC al-gorithm is applied to estimate the direction of arrival.In order to solve the problem,a time-fre-quency associated MUSIC algorithm suitable for through-wall detection and based on S-band stepped frequency continuous wave(SFCW)radar is researched.By associating inverse fast Fouri-er transform(IFFT)algorithm with MUSIC algorithm,the power enhancement of the target sig-nal is completed according to the distance calculation results in the time domain.Then convert the signal to the frequency domain for direction of arrival(DOA)estimation.The simulations of two-dimensional human target detection in free space and the processing of measured data are com-pleted.By comparing the processing results of the two algorithms on the measured data,accuracy of DOA estimation of proposed algorithm is more than 75%,which is 50%higher than classical MUSIC algorithm.It is verified that the distance and angle of human target can be effectively de-tected via proposed algorithm.