A dual input dual spliced network with data augmentation for robust modulation recognition in communication countermeasures.

Modulation recognition accuracy is crucial in communication reconnaissance. In this study, we propose DiDbKDT, a dual-input dual-branch network architecture for cross-SNR modulation recognition. The model leverages knowledge distillation and integrates IQ-AP data as inputs. Considering the differences in data features under different SNR scenarios and combining the respective features of IQ and AP data, we divide the SNR interval into two segments and use the IQ and AP data as the inputs of the low-SNR and high-SNR networks respectively. To balance the computational complexity and recognition rate, in the DiDbKDT network model, we utilize the knowledge distillation technique. We deploy the Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) in the low-SNR and high-SNR scenarios respectively as teacher models to quickly learn the internal modulation features of IQ and AP data. In the student model, we adopt the Transformer to further mine the global features of the data and the correlation relationships between different features, thus achieving more accurate modulation recognition and classification. We also introduce a data augmentation technique based on the channelization scenario to further improve the robustness of the model. Experimental results on six types of datasets, namely RML2016.10a/b/c, RML2018.01a, HisarMod, and MiGouMod, demonstrate that the proposed DiDbKDT outperforms the existing Automatic Modulation Recognition (AMR) methods and has higher parameter efficiency.

Recently, microwave imaging (MWI) based on randomized fields has shown promising potential in practical scenes. However, the off-grid error and low signal-to-noise ratio (SNR) measurements remain a challenge for the imaging process. In this paper, we proposed a compressed sensing (CS) based off-grid calibration method for MWI with random illuminations under low SNR environment. A noisy mismatched imaging model with off-grid error is firstly established. Under the framework of CS, the off-grid error is corrected by genetic algorithm (GA) and singular value decomposition (SVD) is conducted to enhance the SNR of echo signal. In the inversion, the OMP algorithm is used due to its high computational efficiency. At last, the effectiveness of the proposed method is verified based on the numerical data. Compared to the existing algorithms, the proposed approach is able to achieve satisfactory results in both off-grid calibration and noise suppression. Besides, the imaging performance versus the number of measurements, SNR, off-grid level and sparsity is discussed. The proposed method provides a new clue for off-grid MWI in low SNR scenarios.

The performance of high resolution subspace-based algorithms are particularly sensitive to the prior information of the source number, the Signal-to-Noise Ratio (SNR) and the snapshot. Although the existing direction-of-arrival (DOA) estimation methods without estimating the source number could eliminate the awful impact brought by incorrect source number estimation, yet its performance would get deteriorated by small snapshots and low SNR. Methods which exploit noise and signal subspaces information simultaneously, such as SSMUSIC, could provide a high resolution performance in such nonideal circumstances. However, its performance would degrade severely when the prior information of the source number is incorrect. To provide a DOA estimation method without estimating the number of source, which has a high resolution performance in small sample and low SNR scenario, using all information spreads in eigenvalues and eigenvectors, this paper reconstructs a new spatial spectrum which is very similar to the SSMUSIC algorithm. In order to enhance the robustness of the new method, we provide an empirical method to modify the eigenvalues to prohibit the spreading of noise eigenvalues caused by snapshot deflcient and low SNR. To verify the validity of the new method, comparisons with other algorithms are made in computer simulations and the measured data test.

To improve the accuracy and reliability of modulation recognition at low signal to noise ratio (SNR) and few knowledge of signal parameter, the novel method based on the cyclic spectral feature and support vector machine(SVM) is presented. In the process of novel algorithms, the cyclic spectral analysis is used to realize the feature extract of the modulated signals, and the Eigenface method is used to reduce the amount of spectral coherence feature. Then, a new scheme of classification based on support vector machine is presented to classify the modulation signal. The experiment shows that the modulation classification accuracy of presented method is significantly improved at low SNR environment.

As the electromagnetic environment in battlefields is more and more complex, automatic modulation recognition for radar signals is becoming vital and challenging. Traditional methods are more likely to cause lower recognition accuracy with higher computational complexity in low signal-to-noise ratio (SNR). Feature redundancy especially for handcrafted features is one of the shortcomings of deep-learning-based methods. In this article, a novel end-to-end sequence-based network that consists of a shallow convolutional neural network, a bidirectional long short-term memory (Bi-LSTM) network strengthening with a self-attention mechanism, and a dense neural network is constructed to recognize eight kinds of intrapulse modulations of radar signals. The autocorrelation functions of received radar signals are first calculated as autocorrelation features. Then, these features are employed as inputs of the proposed network which owns significant sequence processing advantages and adaptive selection ability of features. Finally, the proposed network outputs prediction modulations directly. The simulation results verify the robustness and effectiveness of autocorrelation features. And the proposed network achieves about 61.25% accuracy at −20 dB and more than 95% accuracy at −10 dB. Compared with four state-of-the-art networks, the proposed network has better recognition performance especially at low SNRs with much lower computational complexity. Results on measured signals also demonstrate that the proposed network outperforms these four networks.

Automatic modulation recognition (AMR) stands as a crucial core technology within the realm of signal processing and perception, playing a significant part in harsh electromagnetic environments. The time–frequency image (TFI) of communication signals can manifest modulation characteristics and serve as a foundation for signal modulation recognition and classification. However, under the influence of the electromagnetic environment, communication signals are exposed to varying degrees of interference, which poses a challenge to the recognition of modulation types. Taking into account the effects of interference and channel fading, this paper introduces a communication signal modulation recognition algorithm based on deep learning (DL) and time–frequency analysis. This approach employs short-time Fourier transform (STFT) to generate time–frequency diagrams from time-domain signals. Subsequently, it binarizes the image and feeds it as input data to the neural network. Our research presents a composite deep convolutional neural network (CNN) architecture known as the composite dense-residual neural network (CDRNN). This architecture focuses on enhancing the feature extraction and identification, aiming to achieve accurate recognition of modulation types in harsh electromagnetic environments. Finally, simulation results validate that the proposed deep learning algorithm holds remarkable advantages in boosting the accuracy of modulation type recognition with better adaptability. The algorithm shows better performance even in harsh electromagnetic environments. When the signal-to-noise ratio (SNR) is 18 dB, the recognition accuracy can reach 92.1%.

The increasingly complex electromagnetic environment and limited spectrum resources pose serious challenges for the recognition of communication signal. The technique of manually feature extraction no longer meets the actual needs. In order to solve the problem of high error rate of communication signal recognition under low Signal-to-Noise Ratio (SNR), we propose a recognition technique that can recognize eight types of modulations. First, we propose Fusion Deep Belief Network (FDBN) that extracts shape features while extracting the essential features of the signal by self-training. Furthermore, considering the problem that the features extracted by FDBN at low SNR are susceptible to noise, we explore the denoising technique based on Variational Mode Decomposition (VMD). The simulation results show that when the SNR is 0dB, the overall recognition rate of the proposed algorithm reaches 98.78%. The proposed recognition algorithm can also be used to recognize the modulation of radar signal.

The Doppler effect causes frequency offset and spectrum spread of communication signals, which increases the difficulty of feature extraction when identifying the modulation mode in aeronautical wireless channels. To improve the signal modulation recognition capability of the single monitoring node in Wireless Sensor Networks (WSN), a method based on the bidirectional convolution gated recurrent deep neural network (Bi-CGDNN) with an attention mechanism is proposed in this paper. We design the framework with different types of network layers, realizing frequency offset suppression and feature mapping. In addition, quadrature sampling signals and instantaneous feature statistics are collected to generate the dataset. Thus, the waveform features and phase-frequency information of the signals can be used for combinatorial analysis. In particular, the attention mechanism is introduced to optimize the framework for better feature perception and higher recognition rates. Experimental results show that our method achieves great recognition effects at different maximum Doppler frequencies. The recognition rate reaches 98.1% at the signal-to-noise ratio (SNRs) of 8dB with the Doppler frequency of 200Hz.

This paper addresses the problem of the opportunistic spectrum access in Cognitive Radio. Indeed, most spectrum sensing algorithms suffer from a high computational cost to achieve the detection process. They need a prior knowledge of signal characteristics and present a bad performance in low Signal to Noise Ratio (SNR) environment. The choice of the optimal detection threshold is another issue for these spectrum sensing algorithms. To overcome the limits of spectrum detectors, we propose in this paper, a blind detection method based on the cyclostationary features of communication signals. Our detector evaluates the level of hidden periodicity contained in the observed signal to make decision on the state of a bandwidth. In order to reduce the computational cost, we take advantage of the FFT Accumulation Method to estimate the cyclic spectrum of the observed signal. Then, we generate the Cyclic Domain Profile of the cyclic spectrum which allows us to evaluate the level of the hidden periodicity in the signal. This level of periodicity is quantified through the crest factor of Cyclic Domain Profile, which represents the decision statistic of the proposed detector. We have established the analytic expression of the optimal threshold of the detection and the probability of detection to evaluate the performance of the proposed detector. Simulation results show that the proposed detector is able to detect the presence of a communication signal on a bandwidth in a very low SNR scenario.

Automated ship detection process has been an essential need for modern Radar system to perform automatic target tracking. This automated process is more commonly found in pulse radars with high rate of Signal to Noise Ratio (SNR), not in Frequency Modulated Continuous Wave (FMCW) radars with very low rate of SNR. The process of automated ship detection with image enhancement and feature extraction in FMCW radars will be elaborated in this paper. The process of image enhancement is designed to split target from the noise and enhance the image of the target with very low SNR. The output of this process will be classified into several groups by utilizing object geographical, circularity and solidity data. From this process, it clearly shows that with image enhancement, the radar detecting capability in average increases by 390%, 140%, 112% and 62% for radar image within the radii of 2 Nautical Mile (NM), 4 NM, 10 NM and 20 NM. With image classification by geographical data and feature extraction, the ship images will be significantly distinguished from clutter with the accuracy of 90%.

Iris recognition is one of the most reliable biometric technologies. The performance of an iris recognition system can be undermined by poor quality images and result in high false reject rates (FRR) and failure to enroll (FTE) rates. The selection of the features subset and the classification has become an important issue in the field of iris recognition. In this paper, a wavelet-based quality measure for iris images is proposed. The proposed method includes three modules: image preprocessing, feature extraction and recognition modules. The feature extraction module adopts the wavelet transform as the discriminating features. Similarity between two iris images is estimated using Euclidean distance measures. Features extracted using higher level wavelet decompositions are shown to yield better clustering and higher success rate in recognition.

Aiming at the problems of network complexity, large computation amount and high hardware platform requirements in the current process of applying deep learning to realize digital signal modulation recognition, this paper proposes a method of using signal constellation diagram modulation recognition in the improved MobileNetV3 lightweight neural network, and introduces the cross-layer structure of Resnet into the MobileNetV3 network. The phenomenon of vanishing gradient caused by decreasing weight with increasing network layers is solved. Firstly, the received MPSK and MQAM signals are converted into constellation diagrams, which are extracted from gray images and enhanced. The image data set of constellation diagram is constructed, which is used to train the lightweight neural network weight of MobileNetV3, and then the constellation diagram is recognized. MobileNetV3 is based on deep convolution separable and Network Architecture Search (NAS) technology, which greatly reduces the number of parameters and training time on the premise of ensuring the recognition accuracy. For the modulation recognition of simple signals, the lightweight neural Network can effectively simplify the Network structure caused to reduce hardware requirements. The simulation results show that the modulated signals (BPSK, QPSK, 8PSK, 16QAM, 64QAM) can achieve the recognition rate of 99.76%. Compared with the traditional network using deep learning to realize the modulation recognition, the number of network parameters and the computational cost are significantly reduced.

The acoustic vector sensor (AVS) array is a powerful tool for direction of arrival (DOA) estimation of underwater targets. However, traditional DOA estimation algorithms generally suffer from low signal-to-noise ratio (SNR) as well as snapshot deficiency. A novel signal self-nulling based DOA estimation method is proposed. Firstly, achieve the crude estimation of the desired DOA using the traditional algorithms. Secondly, set the observation angular interval around the crudely estimated DOA. Thirdly, make the observation direction vary in the observation angular interval, and for each imaginary DOA, calculate the amplitude response of the minimum variance distortionless response (MVDR) beamformer. Finally, the pseudo spatial spectrum is achieved. Computer simulations verify that the proposed method is efficient in DOA estimation, especially in low SNR and insufficient snapshot data scenarios.

The aim of this study is to quantify the gap between the recognition performance of human listeners and an automatic speech recognition (ASR) system with special focus on intrinsic variations of speech, such as speaking rate and effort, altered pitch, and the presence of dialect and accent. Second, it is investigated if the most common ASR features contain all information required to recognize speech in noisy environments by using resynthesized ASR features in listening experiments. For the phoneme recognition task, the ASR system achieved the human performance level only when the signal-to-noise ratio (SNR) was increased by 15 dB, which is an estimate for the human-machine gap in terms of the SNR. The major part of this gap is attributed to the feature extraction stage, since human listeners achieve comparable recognition scores when the SNR difference between unaltered and resynthesized utterances is 10 dB. Intrinsic variabilities result in strong increases of error rates, both in human speech recognition (HSR) and ASR (with a relative increase of up to 120%). An analysis of phoneme duration and recognition rates indicates that human listeners are better able to identify temporal cues than the machine at low SNRs, which suggests incorporating information about the temporal dynamics of speech into ASR systems.

In order to solve the problems of low recognition accuracy and high computation complexity of radar signal processing under low signal-to-noise ratio (SNR), this paper studies a neural network based on 1-D convolution for radar intra-pulse modulation recognition. First, we generate seven different types of radar continuous FM signals and phase coded signals (LFM, NLFM, Frank, P1, P2, P3, P4). Then, the three-dimensional convolutional network suitable for image processing is improved into a one-dimensional convolutional network which will process the generated radar signal in an end-to-end approach. The simulation results show that the recognition algorithm using one-dimensional (1-D) MobileNetV2 network can effectively extract the essential characteristics of radar signal under the condition of low SNR, avoid the huge calculation caused by using time-frequency method, and achieve over 85% recognition accuracy at −6dB.