Search

Automatic modulation recognition (AMR) detects the modulation scheme of the received signals for further signal processing without needing prior information, and provides the essential function when such information is missing. Recent breakthroughs in deep learning (DL) have laid the foundation for developing high-performance DL-AMR approaches for communications systems. Comparing with traditional modulation detection methods, DL-AMR approaches have achieved promising performance including high recognition accuracy and low false alarms due to the strong feature extraction and classification abilities of deep neural networks. Despite the promising potential, DL-AMR approaches also bring concerns to complexity and explainability, which affect the practical deployment in wireless communications systems. This paper aims to present a review of the current DL-AMR research, with a focus on appropriate DL models and benchmark datasets. We further provide comprehensive experiments to compare the state of the art models for single-input-single-output (SISO) systems from both accuracy and complexity perspectives, and propose to apply DL-AMR in the new multiple-input-multiple-output (MIMO) scenario with precoding. Finally, existing challenges and possible future research directions are discussed. [ABSTRACT FROM AUTHOR]

Nowadays, speaker verification systems begin to perform very well under normal speech conditions due to the plethora of neutrally-phonated speech data available, which are used to train such systems. Nevertheless, the use of vocal effort modes other than normal severely degrades performance because of vocal effort mismatch. In this paper, in which we consider whispered, normal and shouted speech production modes, we first study how vocal effort mismatch negatively affects speaker verification performance. Then, in order to mitigate this issue, we describe a series of techniques for score calibration and speaker embedding compensation relying on logistic regression-based vocal effort mode detection. To test the validity of all of these methodologies, speaker verification experiments using a modern x-vector-based speaker verification system are carried out. Experimental results show that we can achieve, when combining score calibration and embedding compensation relying upon vocal effort mode detection, up to 19% and 52% equal error rate (EER) relative improvements under the shouted-normal and whispered-normal scenarios, respectively, in comparison with a system applying neither calibration nor compensation. Compared to our previous work [1] , we obtain a 7.3% relative improvement in terms of EER when adding score calibration in shouted-normal All vs. All condition. • A theoretical study on the vocal effort modes "whispered", "normal" and "shouted" to better understand their main characteristics. • Experiments demonstrate that shouted-normal and whispered-normal vocal effort mismatch drastically decreases speaker verification performance. • Shouted and whispered speech detectors based on logistic regression models that are trained with x-vectors. • Different speaker embedding compensation techniques based on Gaussian mixture models (i.e., MEMLIN, RATZ and SPLICE) in the x-vector domain. • A score calibration technique based on logistic regression models to obtain speaker verification scores robust against vocal-effort variability. [ABSTRACT FROM AUTHOR]

Option pricing in the financial derivatives market is worth studying. When the volatility and the rate of return are not fixed in Blacks-Scholes (BS) equation, the analytical form of the solution is difficult to be obtained. In this paper, we propose deep learning based on blocked residual connection method (DLBR) for the numerical solving of BS equations. The validity period and inventory price are the inputs of the network, while the option price is the solution. Neurons on the multiple hidden layers are processed in blocks and each block is connected to the residuals. The boundary conditions are constructed as a part of the numerical solution and the loss function is minimized to obtain the optimal network parameters. The DLBR method has strong applicability and overcomes the shortcomings of extreme learning machines. Three experiment calculated the numerical solution of the BS equation about European options and promoted the option pricing model. Compared with the existing algorithms such as finite element method and physics-informed neural networks (PINN) method, the obtained numerical solution has higher precision and smaller error, which shows the feasibility and superiority of the proposed method. [ABSTRACT FROM AUTHOR]

With the rise of fully convolutional networks (FCN), in recent years, more and more models of salient object detection based on deep learning have used or modified FCN to perform pixel-level saliency prediction. However, many existing algorithms lack accurate saliency edges due to insufficient use of edge information. In this paper, we solve this problem by focusing on the complementarity between salient object information and salient edge information. Because salient edge features contain more details, the boundary of salient object will be more accurate when fused with edge information. Salient object features can help edge features to better locate the foreground object and filter out background interference. Accordingly, we propose a dual-information progressive optimization network (DIPONet) for salient object detection. In the first step, we extract salient edge features from the low-level of VGG or ResNet backbone and salient object features from the subsequent layers separately, and send them to group optimization fusion modules (GOFMs) which aim to achieve self-optimization of these two kinds of features. In the second step, we design a model named dual-stream information optimization (DSIO) module based on the logical relationship between salient object information and salient edge information to jointly optimize these two kinds of features in a progressive fusion way. Benefiting from the rich edge information in salient edge features and location information in salient object features, the fused features can help detecting salient objects with more accurate boundaries. Excessive experimental results conducted on five benchmark datasets show that our proposed model surpasses the state-of-the-art models and can produce accurate salient object predictions with sharp details. [ABSTRACT FROM AUTHOR]

Nowadays it is common to see image forgeries on almost every media in both professional and personal contexts. From visual retouches to deliberate fake scenes, the technology used to create image forgeries gets easier to use for all users. At the same time, different techniques to assess the authenticity of the content of an image have appeared by taking advantage of the deep learning paradigm. In this paper we propose two initialization approaches for Convolutional Neural Networks (CNNs) used for the detection of image manipulation operations. We focus on the variance stability for the output of a convolutional filter in CNN. Our first proposal is a scaling approach for first-layer convolutional kernels which can cope well with filters generated by different algorithms. Our second proposal is a random high-pass filter initialization approach for CNN's first convolutional layer. The first proposal explicitly computes simple statistical properties of the input signal, while the second approach incorporates the consideration of input statistics in the filter derivation without the need of carrying out explicit computation on the input. Experimental results show the utility of both approaches with improved performance in different image manipulation detection problems and on different CNN architectures. • Initializations of convolutional neural networks for image forensics. • Data-dependent and data-independent initialization approaches. • Mathematical derivations by considering natural image statistics. • Performance improvement of image manipulation detection. [ABSTRACT FROM AUTHOR]

The vulnerabilities of a Biometric Authentication System (BAS) to spoof attacks have gained substantial attention over the years. Iris recognition, one of the most reliable and accurate BASs is vulnerable to various kinds of presentation attacks such as by printed iris or contact lenses. The presentation attack detection (PAD) becomes even more challenging when an attacker imposes variations by carrying out multiple spoofing attacks. To address this challenge, an end-to-end Deep Supervised Class Encoding (DSCE) approach for iris-PAD is proposed in this paper. This is to deal with three main attacks by (i) printed iris images, (ii) contact lenses, and (iii) synthetically generated iris images. DSCE is an autoencoder based supervised feature learning approach that exploits the class information, and minimizes the reconstruction and classification errors simultaneously during the training phase. DSCE is employed to design an iris-PAD framework termed as DeepI, to perceive counterfeit access to an iris-BAS. Experimental results on different benchmark databases show that DSCE based DeepI outperforms the current state-of-the-art iris-PAD methods. Also, in cross-database training-testing settings, the proposed approach manifests a promising generalization capability. [ABSTRACT FROM AUTHOR]

Primary user (PU) signal detection or classification is a critical component of cognitive radio (CR) related wireless communication applications. In CR, the PU detection methods are mostly based on statistical models, and their detection performance heavily relies on the accuracy of assumed models. In this paper, we design a novel detector, dubbed as PU-Net, that dynamically learns the PU activity patterns in a cognitive 5G smart city, where a network of unmanned aerial vehicles (UAVs) is deployed as flying base stations to serve the Internet-of-Things (IoT) users. Unlike the traditional schemes, the PU-Net is free from signal-noise model assumptions and is leveraged through deep residual learning integrated with atrous spatial pyramid pooling (ASPP) to sense the PU's transmitted signal patterns in the network. The PU-Net detects and classifies the active and idle PU states by exploiting the multilevel spatial-temporal features in the signal and noise frames. The proposed model is trained using locally synthesized Rayleigh channel-impaired data with large variability of modulated signals and different noise floor regimes. Additionally, the PU-Net model is blind-tested and evaluated on real-world over-the-air signals and with variable-length frames and varying channel effects at secondary users (SUs). With extensive experiments, it is shown that PU-Net outperforms other benchmark detectors, obtaining an accuracy of 0.9974, with 0.9978 recall and 0.9970 precision in detecting and classifying the PU transmitted signal patterns. Correspondingly, the proposed PU-Net can be adopted for IoT/UAV-assisted communication systems in optimizing spectrum efficiency and resolving the coexistence issues in 5G and beyond networks. The PU-Net system model for 5G smart city with UAVs cognitive network, acting as aerial base stations to serve the IoT users. The cyan and yellow ellipses denote the network architecture with and without infrastructure, respectively. • A DL-based detection of PU transmitted signals (PU-Net) in cognitive radio network. • It dynamically learns the PU activity patterns in IoT-UAV enabled 5G smart city. • It is free from statistical model assumptions leveraging deep learning with ASPP. • PU-Net is extensively evaluated using synthetic & real-world over-the-air signals. • ROC performance curves are simulated and compared with benchmark detectors. [ABSTRACT FROM AUTHOR]

This paper proposes a combined Convolutional Neural Network (CNN) scheme appropriate for automatic modulation recognition. The suggested network includes a deep network involving multiscale convolutions with an attention module for recognizing 24 modulation modes and a shallow one network for recognizing two modulation modes, aiming to reduce the misjudgment caused by the deeper network. We evaluate our proposed network on the DeepSig dataset achieving an accuracy of 98.7% that exceeds the performance of current state-of-the-art (SOTA) methods exploiting this dataset. The network achieved almost 100% accuracy on modulation modes of the order less than or equal to 64 and under high SNR. The accuracy attained by our method on high-order modulation modes exceeds the one of ResNet, which currently is the SOTA method. [ABSTRACT FROM AUTHOR]

Most real-life signals exhibit non-stationary characteristics. Processing of such signals separately in the time-domain or in the frequency-domain does not provide sufficient information as their spectral properties change over time. Traditional methods such as the Fourier transform (FT) perform a transformation from time-domain to frequency-domain allowing a suitable spectral analysis but looses the spatial/temporal information of the signal components. Hence, it is not easy to observe a direct relationship between the time and frequency characteristics of the signal. This makes it difficult to extract useful information by using only time- or frequency-domain analysis for further processing purposes. To overcome this problem, joint time–frequency (TF) methods are developed and applied to the analysis and representation of non-stationary signals. In addition to revealing a time-dependent energy distribution information, TF methods have successfully been utilized in the estimation of some parameters related to the analyzed signals. In this paper, we briefly summarize the existing methods and present several state-of-the-art applications of TF methods in the classification of biomedical signals. We also point out some future perspectives for the processing of non-stationary signals in the joint TF domain. [ABSTRACT FROM AUTHOR]

The field of digital histopathology has seen incredible growth in recent years. Digital pathology is becoming a relevant tool in healthcare, industrial and research sectors to reduce the saturation of pathology departments and improve the productivity of pathologists by increasing diagnostic accuracy and reducing turnaround times. Artificial Intelligence (AI) algorithms may be used for the identification of relevant regions, extraction of features from a histological image and overall classification of images into specific classes. The combination of digital histopathology imaging and AI therefore presents a significant opportunity for the support of the pathologists' tasks and opens up a whole new world of computational analysis. In this paper, we have analysed the present, the challenges and the future of the computational pathology discussing the different existing strategies to overcome its main limitations and ensure the computational pathology acceptance. The lack of labelled data, which is the possibly largest challenge for all medical AI applications, is even more pronounced in computational pathology because of the multi-gigapixel nature of the images and high data heterogeneity. We consider the future of the computational pathology is the combination of weak label strategies with active learning and crowdsourcing scenarios since it would remove some of the workload from clinical experts and manual annotation obtaining clinically satisfactory performance with minimal annotation effort. In addition, we believe areas such as explainable AI, data fusion and secure role-based data sharing will be receiving increasing research attention in computational pathology in the close future. • Computational pathology is the automatic analysis of histological images. • The present, challenges and future of computational pathology are analyzed. • Challenges: multi-gigapixel images, data heterogeneity and lack of labelled data. • Future directions: explainable AI, data fusion and secure role-based data sharing. • Combining weak label strategies, active learning and crowdsourcing is promising. [ABSTRACT FROM AUTHOR]

• A new signal detection method for MIMO-OFDM-IM is addressed. • The combination of neural network training and variational inference is studied. • The performance of the proposed algorithm outperforms the traditional ones. This paper proposes a deep learning (DL) based method for signal detection in multiple-input multiple-output orthogonal frequency division multiplexing with index modulation (MIMO-OFDM-IM) systems using variational autoencoder (VAE). A framework of variational optimization is constructed and the learning deep neural network structure with fully connected layers is also designed. The network is trained offline by exploiting the dataset generated by simulations, and then the well-trained model is applied for signal detection in MIMO-OFDM-IM system. The derivation on the loss function for network training is presented from the perspective of variational inference. Also, a regularization parameter is introduced in order to achieve better performance. Numerical results are provided and the performance demonstrates the merits of the proposed method by comparison with the traditional detection algorithms. Finally, complexity measured by the runtime of the program utilizing existing detectors and the proposed method are also presented. The bit error rate (BER) performance which is close to the existing hand-crafted detectors can be obtained with lower runtime using the proposed DL network. [ABSTRACT FROM AUTHOR]

In recent years, elliptic partial differential equations (PDEs) have been challenging in numerical mathematics, engineering, and physics. In this paper, we proposed a novel approach called Multilayer Neural Network for Partial Differential Equations (MLNPDE) based on deep learning algorithms by exploring the elliptic PDE family under certain conditions such as initial conditions and boundary conditions to govern approximation solutions of PDEs. In the proposed model, we include a multilayer neural network using a densely connected network. Moreover, the approximation solution underlying PDEs can be expressed in two terms: the first term satisfies the boundary conditions, and the second term is a function of the unknown parameters that were estimated by our proposed model. Furthermore, we derive a normal form of the boundary conditions on the first term of the approximation solution for two-dimensional elliptic partial differential equation problems with an arbitrary domain. The numerical results show that our method achieved tremendous state-of-the-art performance in accuracy and efficiency compared to the classical methods. [ABSTRACT FROM AUTHOR]

Fault prediction technology provides a way to reduce the loss caused by equipment failure. Currently, many efforts of failure prediction pay more attention to the fault trend and the remaining useful life. This paper devotes to predicting a specific failure mode of device in the future. We propose a two-step prediction method with two sub-models to predict the fault mode. The first model, named regression model, combines Attention-based-LSTM (Long Short-Term Memory) and Random Forest (RF) to predict time series trends, where wavelet packet analysis is adopted to extract elementary features of time series and form multivariable time series. Based on the status data forecasted by the first model, the second model, named classification model, combines Attention-based-LSTM and Extra-Tree (ET) to classify fault mode. The classified fault will occur in the device at some point in the future. In the regression model, Attention-based-LSTM performs deep level feature extraction of vibration sensor signals, and the result is fed to the last layer (i.e., an RF layer), which takes responsibility for predicting the future sequence. The future sequence is then provided to the classification model, where Attention-based-LSTM is used to extract implicit features, and ET classifier is used to accept these features and figure out the failure phenomenon. Our approach for predicting failure modes includes the severity and type of fault, which are rarely investigated synthetically. The experimental results of bearing data show that the sub-models can achieve high prediction accuracy and our method can predict the future failure mode with precision. [ABSTRACT FROM AUTHOR]

Deep learning has been actively utilized for speech enhancement. However, deep learning-based speech enhancement usually produces over-smoothed speech, resulting in speech distortion and degraded intelligibility. In this paper, we propose the exaggeration of the training target so that the dynamic range of the enhanced speech becomes more similar to that of the clean speech. Target exaggeration can be implemented in two ways. The first approach is to exaggerate the target feature in the cost function of a deep learning-based speech enhancement system. This method can be implemented without additional parameters or computation, but can only be applied to schemes working in the time-frequency domain with the mean-square error cost function. The second approach is to introduce an additional deep neural network (DNN) that estimates the residual error in the output of a deep learning-based speech enhancement. This requires more computation, but can be applied even to time-domain approaches. To evaluate the performance of the proposed target exaggeration, it is applied to a feed-forward DNN- and long short-term memory (LSTM)-based speech enhancement scheme in the time-frequency domain, and the convolutional time-domain audio separation network (Conv-TasNet)-based speech enhancement scheme in the time domain. Experimental results showed that the proposed method improved the quality of speech produced by the deep learning-based speech enhancement system in terms of the perceptual evaluation of speech quality (PESQ) scores and outperformed other approaches, including global variance equalization and a perceptually optimized speech denoising autoencoder, to alleviate the over-smoothing problem. [ABSTRACT FROM AUTHOR]

Blind identification of channel codes is becoming increasingly important in signal interception and intelligent communication systems. However, most existing channel codes recognition algorithms extract features manually, which makes them highly demanding in real-world application. Thus, efficiently identifying channel codes is difficult using present technologies. This paper presents a deep residual network-based deep learning (DL) approach on the blind identification of convolutional code parameters for a given soft-decision sequence. The proposed method can blindly identify the convolutional codes without the need for the prior information about its coding parameters, and it achieves over 88% of recognition accuracy for 17 forms of convolutional codes when SNR exceeds or equals zero. Furthermore, we investigate factors affecting the accuracy of channel codes recognition including input length, model depth and data type. A comparison of the recognition accuracy between the proposed algorithm, log-likelihood ratio (LLR)-based traditional blind identification algorithm, and DL-based algorithm are then made. Experiment results show that deep residual network-based approaches could provide significant improvements over the traditional algorithm or existing DL-based algorithms in the blind identification of convolutional codes. [ABSTRACT FROM AUTHOR]