Automatic modulation recognition for DVB-S2 using pairwise support vector machines

In this paper, a support vector machine (SVM) pairwise coupling algorithm is developed for classification of satellite communications signals used in second generation of digital video broadcasting via satellite (DVB-S2) standard. DVB-S2 standard adaptively uses one of QPSK, 8PSK, 16APSK, and 32APSK modulations. The proposed method extracts fourth and sixth order cumulants as features from the received signal. The features are given to a SVM pairwise coupling algorithm in which there is one binary SVM for each pair of modulation types. Finally the algorithm selects the modulation type chosen by the maximal number of pairwise SVMs as final decision. SVMs are trained by samples from different modulation types corrupted by Gaussian noise. The simulation results show that the proposed method allows higher recognition rates in comparison with previous methods, especially at low SNRs.

The paper presents a new algorithm for modulation recognition of digital communication signals based on higher order cumulants (HOCs) and support vector machines (SVM). The fourth and sixth order cumulants of the received signal are used as the input classification feature vector to the SVM. The SVM maps the input vectors nonlinearly into a high dimensional feature space, constructs the optimum separating hyperplane in that space and makes the non-separable data separable. This method is robust to Gaussian noise and constellation rotation due to the initial phase of the signal and avoids the overfitting and local minimum in a neural network. The high performance and robustness of the algorithms are proved by computer simulation.

Support Vector Machines (SVM) and Independent Component Analysis (ICA) are two powerful and relatively recent techniques. SVMs are classifiers which have demonstrated high generalization capabilities in many different tasks, including the object recognition problem. ICA is a feature extraction technique which can be considered a generalization of Principal Component Analysis (PCA). ICA has been mainly used on the problem of blind signal separation. In this paper we combine these two techniques for the face recognition problem. Experiments were made on two different face databases, achieving very high recognition rates. As the results using the combination PCA/SVM were not very far from those obtained with ICA/SVM, our experiments suggest that SVMs are relatively insensitive to the representation space. Thus as the training time for ICA is much larger than that of PCA, this result indicates that the best practical combination is PCA with SVM.KeywordsSupport Vector MachineFace RecognitionFace ImageIndependent Component AnalysisFace DatabaseThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Self-organizing feature maps are topologically ordered. One develops realistic cortical structures when given approximations of the visual environment as input, and are an effective way to model the development of face recognition abilities. Support vector machines (SVMs) are classifiers, which have demonstrated high generalization capabilities. In this paper, we combine these two techniques for face recognition problem. Experiments were made on two different face databases, achieving very high recognition rates with relative low classification cost. As the results using the combination SOM/SVM were not very far from only with SVM, but the classifier cost of SOM/SVM is one-tenth of with SVM.

Automatic modulation classification based on high order cumulants and hierarchical polynomial classifiers

Support vector machines (SVM) has achieved great success in multi-class classification. However, with the increase in dimension, the irrelevant or redundant features may degrade the generalization performances of the SVM classifiers, which make dimensionality reduction (DR) become indispensable for high-dimensional data. At present, most of the DR algorithms reduce all data points to the same dimension for multi-class datasets, or search the local latent dimension for each class, but they neglect the fact that different class pairs also have different local latent dimensions. In this paper, we propose an adaptive class pairwise dimensionality reduction algorithm (ACPDR) to improve the generalization performances of the multi-class SVM classifiers. In the proposed algorithm, on the one hand, different class pairs are reduced to different dimensions; on the other hand, a tabu strategy is adopted to select adaptively a suitable embedding dimension. Five popular DR algorithms are employed in our experiment, and the numerical results on some benchmark multi-class datasets show that compared with the traditional DR algorithms, the proposed ACPDR can improve the generalization performances of the multi-class SVM classifiers, and also verify that it is reasonable to consider the different class pairs have different local dimensions.

This paper is based on Extended Symbol OFDM (ES-OFDM) where symbols are extended in time. This way ES-OFDM can operate at low SNR. Each doubling of the symbol length improves the SNR performance by 3 dB in case of a coherent receiver. One of the basic questions is how to synchronize to signals far below the noise floor. An algorithm is presented which is based on the transmission of pilot symbols. At the receiver, the received signal is cross correlated with the known pilot symbol and the maximum magnitude is determined. The position of the maximum value within the cross correlation function indicates the time difference between transmitter and receiver. The performance of the algorithm in case of an Additive White Gaussian Noise (AWGN) channel, is assessed based on a theoretical approximation of the probability of correct detection of the time difference. The theoretical approximation matches with simulation results and shows that synchronization can be achieved for low (negative) SNRs.

In order to pursue rapid development of the new generation of wireless communication systems and elevate their security and efficiency, this paper proposes a novel scheme for automatic dual determination of modulation types and signal to noise ratios (SNR) for next generations of wireless communication systems, fifth-generation (5G) and beyond. The proposed scheme adopts unique signatures depicted in two-dimensional asynchronously sampled in-phase-quadrature amplitudes' histograms (2D-ASIQHs)-based images and applies the support vector machines (SVMs) tool. Along with the estimation of the instantaneous SNR values over 0-35 dB range, the determination of nine modulation types that belong to different modulation categories i.e., phase-shift keying (Binary-PSK, Quadrature-PSK, and 8-PSK), amplitude-shift keying (2-ASK and 4-ASK) and quadrature-amplitude modulation (4-QAM, 16-QAM, 32-QAM, and 64-QAM) could be achieved by this scheme. The application of this scheme has been simulated using a channel model that is impaired by additive white Gaussian noise (AWGN) and Rayleigh fading, covering a broad range of SNRs of 0-35 dB. The performance of this dual-determination scheme shows high modulation recognition accuracy and low mean SNR estimation error. Therefore, it can be a better alternative for designers of next generation wireless communication systems.

The voice disorders occurring due to the problems in the voice producing organs cause some changes in the intensity or tone of the voice. It is difficult to identify the diseased voice by the reason of its variable and different nature. One of the most popular voice disorder reasons is cyst which is located on the vocal cords. The aim of this study is to detect the vocal cyst problem by using acoustic voices data which were recorded from healthy people and patient with cyst diagnoses subjects by using high order statistics and support vector machines (SVMs) classifier. In this study, two experimental procedures were implemented for two different voice samples. In the first , /a/ vowel and the second, the Turkish word of “aydinlik” (mean in English “bright”) were investigated with skewness and kurtosis parameters which are third and fourth order cumulants (spectral moments), respectively. The obtained features values for healthy and cyst subjects were used as the SVMs’ inputs for classification . The experimental results show that the test accuracies of SVMs were found as 94.89% and 91.11% for /a/ vowel and “aydinlik” word, respectively. It is concluded from experimental studies that skewness provides more meaningful results than kurtosis in relation to distinguish into two voice groups as healthy and cyst . Additionally, it is assessed that the “aydinlik” is affective word for the pathological and normal acoustic voice discrimination as good as /a/vowel.

This paper proposes a novel scheme combining support vector machines (SVM) and a residual-based method for wind turbine fault detection and isolation (FDI). SVMs with radius basis function kernels are used for detecting and identifying sensor stuck and offset faults, where binary codes of fault types are used as the outputs of the SVMs to minimize the number of SVMs being used. The same output of a SVM may correspond to different types of faults and the final decision is made by all SVMs instead of one SVM. Moreover, a residual-based fault detection method using a time-variant threshold is developed to identify the abrupt change and scaling faults. Monte Carlo simulations are carried out in MATLAB to test the effectiveness and robustness of the proposed FDI methods using a wind turbine FDI benchmark model. Results show that the proposed methods can always detect the faults successfully within the required time limits.

Fault diagnosis of multivariable dynamic systems is studied by combining nonlinear spectrum feature with support vector machine. In order to resolve the problem of large calculated amount of solving nonlinear spectrum, a frequency domain variable step size normalized LMS adaptive algorithm is proposed based on the one-dimensional nonlinear output frequency response function (NOFRF). The step size is updated in real time according to the spectrum estimation error and the previous step size. After obtaining nonlinear spectrum data, kernel principal component analysis is used to compress data and extract spectrum feature. In order to improve fault recognition precision, a multi-feature fusion SVM fault classifier is established based on different frequency domain scales. Every sub-classifier is constructed by the spectrum feature of each order, and the diagnosis result can be obtained by weighed fusion of all sub-classifiers. Consider the difference of classification reliability for input features, sub-classifier weight is obtained using the distance between input and SVM separating hyperplane. Simulation experiments indicate that the proposed fault diagnosis method has good real-time performance and high recognition rate, so it can meet the requirements of online diagnosis of multivariable dynamic system.

A DC series arc fault is harmful and not easy to distinguish, so its detection technology needs to be continuously improved. This research establishes a DC series fault arc generation and experiment platform. Firstly, this research basically analyzes and summarizes the variations of line current and supply output voltage caused by an arc fault. Then, a DC series arc fault detection method, based on comprehensively using line current and supply output voltage signals, is proposed. In order to extract the characteristic frequency band of the line current, the Local mean decomposition (LMD) is employed which has a better decomposition effect. Then, fuzzy entropy and harmonic power of the corresponding frequency-band signals are calculated to reflect the variation of the line current complexity. Moreover, the standard deviation of the supply output voltage is selected as another characteristic for identifying arc faults. Finally, the support vector machines (SVM) is employed to classify the extracted features. Experiment data and simulation results illustrate that the hybrid method not only has a higher recognition rate, but also can avoid misjudgment and invalid operation of the program to a certain extent.

The reliable operation of brain-computer interfaces (BCIs) based on spontaneous electroencephalogram (EEG) signals requires an accurate classification and recognition of multichannel EEG. The design of EEG representations and classifiers for BCI are open research questions whose difficulty stems from the need to extract complex spatial and temporal patterns from noisy multidimensional time series obtained from EEG measurements. This paper proposes a Genetic algorithm (GA) and Support Vector Machine (SVM) hybrid approach to accomplish this EEG classification task for potential BCI applications. An Oddball stimulus program and evoked event-related coherence program were designed to evaluate our method. The present study systematically evaluates the performance of the one channel pair event-related coherence feature set for EEG signal classification of auditory task. A GA approach for feature selection is presented which used to reduce the dimension of event-related coherence feature parameters. With the base classifiers of SVM, classification experiments are carried out upon real EEG recordings. Experimental results suggest the feasibilities of the new feature set, and we also derive some valuable conclusions on the performance of the EEG signal classification methods. The high recognition rates and the method's procedural and computational simplicity make it a particularly promising method for achieving real-time BCI system based on evoked potential event-related coherence in the future.

Dense false target jamming can affect radar detection performance severely. A method of dense false targets jamming recognition based on time‐frequency atomic decomposition theory and support vector machine (SVM) is proposed to solve the difficulty of dense false targets jamming identification. According to the feature of ambiguity function of dense false targets jamming signal, a Gabor sub‐dictionary which has adaptive variation with signal is designed. The signal is expanded into the corresponding Gabor time‐frequency dictionary by sparse decomposition. After Gabor atomic time‐frequency parameters are extracted as individual feature vectors, SVM is utilised for classification and recognition. The experimental results show that the extracted Gabor atomic time‐frequency parameters can effectively represent the essential features of target and dense false targets jamming, respectively, and this method has a high recognition rate.

In this study, a new handwritten character recognition system able to work for a large-scale data set, faster and have a high recognition rates, is developed. For this purpose this study is implemented within four stages which are pre-processing, feature selection, classification and speedup. Two-dimensional wavelet-based method and the Sobel gradient method are used for the feature extraction process. The classification process is performed with Support Vector Machines (SVMs) using one-against-all method in the classification phase. In the speedup phase of the study, shared memory architecture was used which one of the parallel programming architectures. In this aspect, serial codes of SVM were analyzed and the code blocks which were suitable for parallel working are collimated and then the codes run with a multi-core computer.