Direction-of-arrival (DOA) estimation is a vital research topic in array signal processing, with extensive applications many fields. In recent years, deep learning has been applied to DOA improve the performance. However, most existing learning-based methods extract information from covariance matrix (CM) input. this paper, we introduce novel scheme that utilizes raw in-phase (I) and quadrature (Q) components of as We formulate problem single-label classification multi-label based on number...

Generative adversarial network (GAN) has achieved great success in many fields such as computer vision, speech processing, and natural language because of its powerful capabilities for generating realistic samples. In this paper, we introduce GAN into the field electromagnetic signal classification (ESC). ESC plays an important role both military civilian domains. However, specific scenarios, can't obtain enough labeled data, which cause failure deep learning methods they are easy to fall...

Aiming at the problem of node identification in wireless networks, a method based on deep learning is proposed, which starts with tiny features nodes radiofrequency layer. Firstly, order to cut down computational complexity, Principal Component Analysis used reduce dimension sample data. Secondly, convolution neural network containing two hidden layers designed extract local preprocessed Stochastic gradient descent optimize parameters, and Softmax Model determine output label. Finally,...

Automatic modulation recognition (AMR) of radio signal is an important research topic in the area non-cooperative communication and cognitive radio. Recently deep learning (DL) techniques enable significant progress AMR. However, adversarial machine cause threats attacks DL-based In this paper, we aim to make AMR model robust, accurate lightweight, thus propose a multi-distillation mechanism for robust training models, namely Adversarial Multi-Distillation (AMD). framework AMD, by knowledge...

Source number estimation plays an important role in successful blind signal separation. At present, the application of machine learning allows processing signals without time-consuming and complex work manual feature extraction. However, convolutional neural network (CNN) for has some problems, such as incomplete extraction high resource consumption. In this paper, a lightweight source (LSNEN), which can achieve robust mixed at low SNR (signal-to-noise ratio), is studied. Compared with other...

Deep learning is an important support for the development of cognitive communication in Internet Things (IoT). convolution neural networks have powerful functional expression and feature extraction capabilities. Mallat et al. proposed a convolutional network model with strict mathematical theory excellent ability 2012, i.e., wavelet scattering networks, which widely used audio image classification. In this paper, we improve propose construct weight-variable by combining deep organically. We...

Recently, backdoor attacks posed a new security threat against radio signals modulation models. The attacked model performs well on benign samples, whereas abnormally samples. Current usually inject static trigger into which can be easily detected with patterns and large perturbation. In this paper, inspired by the Generative Adversarial Network (GAN), we proposed attack based adaptive (AT-GAN) to learn of trigger, generate different for specific signal. AT-GAN is composed Adaptive Backdoor...

A novel air quality index (AQI) forecasting method based on support vector machine (SVM) and moments is proposed in this paper. By using it, the AQI value of a sky image can be forecasted. In order to improve accuracy stability, color moments, correlogram wavelet features are used extract features, which transfers input from space feature space. We train SVM with these extracted features. Then trained utilized achieve for new images. The experimental results that has good results, testing...

The continuous change of communication frequency brings difficulties to the reconnaissance and prediction non-cooperative communication. core this process is frequency-hopping (FH) sequence with pseudo-random characteristics, which controls carrier hopping. However, FH always generated by a certain model kind time regularity. Long Short-Term Memory (LSTM) neural network in deep learning has been proved have strong ability solve series problems. Therefore, paper, we establish LSTM implement...

The medium access control (MAC) protocol identification is of great application value in cognitive radio. In order to realize the MAC with high accuracy and avoid manual feature extraction, we convert sampled data into form spectrogram. Then, a graphical scheme which combines convolutional neural network (CNN) spectrogram proposed. simulation includes CNN method support vector machine (SVM) method. It suggested that has better performance.

In this paper, we investigate the modulation recognition method based on locality preserved projection (LPP) in AWGN channels. Feature extraction is precondition of signal recognition. Based analyzing characteristic time and frequency domain, seven feature parameters with fine classification information are selected. order to wipe off relativity among different features, keep important identity for simultaneously, need search a best subspace which can be apart very well. LPP builds graph...

Radio modulation classification has always been an important technology in the field of communications. The difficulty incremental learning radio is that new tasks will lead to catastrophic forgetting old tasks. In this paper, we propose a sample memory and recall framework for classification. For data with different signal-to-noise ratios, use partial strategy by selecting appropriate samples memorizing. We compare performance our proposed method three baselines through large number...

In cognitive radio networks, providing accurate recognition of the primary user's signal is great important for designing spectrum access strategies. Source number estimation has served as a key fundamental technique to facilitate in mixed received signals scenario. this paper, we propose deep learning based source method under single-channel conditions. The architecture network first designed. Then, complex are reconstructed in-phase and quadrature (IQ) data order adapt convolutional neural...

A wireless communications system usually consists of a transmitter which transmits the information and receiver recovers original from received distorted signal. Deep learning (DL) has been used to improve performance in complicated channel environments state-of-the-art (SOTA) achieved. However, its robustness not investigated. In order evaluate DL-based recovery models under adversarial circumstances, we investigate attacks on SOTA model, i.e., DeepReceiver. We formulate problem as an...

Automatic Modulation Classification (AMC), which based on deep learning has been extensively researched and implemented in wireless communication systems. Universal adversarial perturbation refers to a single that can cause most samples be misclassified by models. In this brief, we aim achieve imperceptible universal attacks AMC models, thus an perturbations (imperceptible UAPs) framework was proposed. Specifically, loss functions were separately designed for target signals non-target...

In recent years, the wave of artificial intelligence technology, which is guided by deep learning, becoming more and widely applied to all fields society. Among them, cross collision between art has attracted great attention in related research fields. The migration image artistic style based on learning become one active topics. this paper, a simple effective method presented for migration. That is, firstly, we specify an input as original (it also called content image); at same time,...

In view of the individual recognition problem communication emitter, this paper, starting with subtle characteristics emitter in signal layer, proposes a method based on deep learning. First, framework learning is established, and convolution neural network containing two hidden layers designed to extract local features through operations. Secondly, stochastic gradient descent used optimize parameters, soft max model determine output label. Finally, effectiveness verified by experiments.