Abstract:Vanadium dioxide is a kind of material with reversible phase transition from insulating state to metallic state. It is widely used in optical devices and information technology. In this paper, the phase transition process of silicon-based vanadium dioxide is studied and analyzed by terahertz spectroscopy and array imaging technology. Firstly, the transmission spectra and reflection spectra of the whole sample in the 2.5~20.0 THz region are obtained by using the Fourier transform spectrometer. The analysis shows that the temperature range of phase transition of silicon-based vanadium dioxide is from 334 K to 341 K, and the corresponding temperature difference is 7 K. It is obtained that the transmittance of the sample to 4.3 THz radiation changes more than 40% and the reflectivity changes close to 30% after phase transition; then, a set of 4.3 THz array imaging system is utilized to measure the THz images of the whole sample before and after phase transition. When the material changes from metal state to insulating state, the transmittance of 4.3 THz light increases from 6.7% to 50.7%, and the transmittance change is 44%, which is near to the results of Fourier transform spectra at 4.3 THz. The above research results provide good experimental data support for the transmission modulation and reflection modulation of silicon-based vanadium dioxide for electromagnetic radiation above 2.5 THz.

Abstract:A near-field mmwave 3D imaging algorithm is presented. Firstly, the target is imaged in 2D, and then a joint BM3D adaptive filtering sliding window algorithm based on local information is proposed, which can effectively preserve image information and suppress clutter to achieve 3D imaging reconstruction. The system platform, signal echo model, backward projection and holographic imaging algorithm and imaging results, image filtering, and 3D imaging reconstruction are introduced respectively. The obtained results validate the feasibility and accuracy of the experiments, and the proposed method is of broad application in many scenarios.

Abstract:A dangerous item segmentation algorithm for passive terahertz imaging security inspection is proposed in response to the difficulty and low precision of dangerous item recognition in the passive terahertz imaging. First of all, the hypothesis of the dangerous item local structural difference and the hypothesis of the local luminance difference are made to locate the Region Of Interest(ROI) where dangerous items might exist in terahertz images. Meanwhile, the shallow convolutional network containing a few feature channels and nerve cells is chosen for super-resolution processing of images in ROI regions. The images are input into the U-net to obtain high-quality and clearly-outlined partitioned images of dangerous items. Finally, an experiment is conducted to verify the improvement of the detection accuracy of the proposed algorithm in comparison with traditional partitioning algorithms. This is conducive to improving the recognition rate of dangerous items by the passive terahertz imaging security inspection system.

Abstract:With the development of terahertz detection technology, it becomes more and more important to accurately measure the relative spectral response of terahertz detector. The measuring principle of the relative spectral response of terahertz detector is analyzed, and a relative spectral response measurement system of terahertz detector is set up, the uncertainty of the measurement system is verified by using terahertz detector. According to the analysis of experimental data, the uncertainty of measurement is 9.2% in the range of 1 THz~10 THz, which can meet the requirements of relative spectral response measurement of terahertz detectors.

Abstract:For few-shot electromagnetic signal classification, data augmentation is the most intuitive strategy. In this paper, Generative Adversarial Network(GAN) is employed to generate fake signal samples. The coarse-grained and fine-grained screening mechanisms are designed to screen the generated fake signals. The generated signals with poor quality are removed and the effective expansion of training dataset is realized. In order to verify the effectiveness of the proposed data augmentation algorithm, sufficient experiments are conducted on the RADIOML 2016.04C dataset. Experimental results show that the proposed method can improve the accuracy of few-shot electromagnetic signal classification effectively.

Abstract:Benchmark datasets are crucial for many data-dependent scientific studies and technology applications. Academic and industry communities have closely collaborated to release abundant datasets in many fields. However, there is still a lack of high-quality benchmark datasets in some specific domains. This paper introduces two open-source benchmark datasets, namely, the Insider Threat Dataset (ITD-2018) and the Indoor Crowd Movement Trajectory Dataset(ICMTD-2019). The two datasets are produced by program-driven synthetic data generation methods and are presented with well-defined scenarios, carefully-designed behavior models, rich data patterns, and vivid storylines. The two datasets were used in the ChinaVis Data Challenge. This paper aims to promote the two datasets for the development of the research and technology in relevant domains.

Abstract:The increasingly crowded and complex airspace environment makes it necessary to determine the initiation of the true target track. Most existing research on radar target track initiation only considers one of real-time or initiation rate, and it is difficult to complete fast and accurate track initiation in a strong clutter environment. In this paper, a track initiation algorithm is proposed based on Deep Learning and Temporal-Spatial(DLTS) characteristics of radar measurement suitable for strong clutter environment. The algorithm first selects the candidate set from the radar measurement combinations, and next extracts the temporal change vector and spatial distribution vector of the measurement combination, and uses them as the input of the One-dimensional Convolutional Neural Network(1DCNN) and Gated Recurrent Unit(GRU) hybrid model to obtain the time dimensional characteristics and space dimensional characteristics of the measurement combination, then merge the two to get the temporal-spatial characteristics. Finally, the true and false tracks are classified with the temporal-spatial characteristics processed by self-attention, and the track initiation is completed. The simulations show that DLTS algorithm can effectively improve the performance of the true and false track initiation rate when the time loss is similar to that of the logic method in the strong clutter environment.

Abstract:In recent years, the research on risk measurement and control methods based on big data analysis model has become more and more important, and the backtesting analysis on risk measurement tools can guarantee the effectiveness of the techniques used in actual data analysis. Marginal Expected Shortfall(MES) is an important tool to measure the marginal contribution of individual institutions to systemic risk, and the backtest methodologies for MES is also worthy to focus on. In this paper, the backtest method of ES in C. Acerb et al. is extended to the two-dimensional case and two backtest methodologies are proposed for MES. The results of simulation show that these two new statistics are more powerful than the statistics used in D.Banulescu et al. under situations that the difference between the null hypothesis and the alternative hypothesis is relatively small. The results of empirical analysis also support that these two new statistics proposed in this paper accept the null hypothesis more cautiously under the same prediction model. This method can give some reference for model algorithm backtesing under big data.

Abstract:A processing process of open-set specific emitter identification is built in order to achieve accurate control of urban frequency equipment. The core lies in the effective interval filtering of fingerprint features and the open set recognition model based on the deep self-encoder. By visualizing deep network activation using Class Activation Mapping(Grad-CAM), the section of signal contributing more to neural network activation can be determined, and then interval filtering for the signal can be performed without losing too much fingerprint information. On the other hand, an open-set specific emitter identification model is established based on semi-supervised adversarial autoencoders, achieving effective monitoring and identification of unknown emitters that may occur in the spectrum. Experiments show that Grad-CAM can filter out the most advantageous part of the extracted signal fingerprint, and the proposed model can achieve high-precision open set recognition without degrading the closed set recognition rate.

Abstract:The increasingly complex electromagnetic environment imposes high requirements for battlefield target detection. Accurate, quick and complete multi-radar track correlation has become an urgent problem with the continuous development of multi-radar fusion systems. Most of the existing research on track correlation only considers the latest target track points reported by radar, while ignoring the previous track information. In addition, the solution to the asynchronous track problem of most track correlation algorithms is time registration. It not only increases the computational cost of the algorithm, but also magnifies the error contained in the track information. Therefore, time registration is difficult to be applied to the current complex electromagnetic environment. In this paper, a Track-to-Track Correlation algorithm based on Siamese Network(TTCSN) is proposed, which is suitable for asynchronous track correlation and does not need time registration. A pair of tracks are sent into the feature extraction network, and TTCSN learns the hidden features of input tracks. Then the similarity of hidden feature vectors are calculated by TTCSN to get the similarity vector which is fed into the classifier to distinguish that the input tracks are correlated or not. The experimental results show that TTCSN algorithm can effectively solve the problem of asynchronous track correlation.

Abstract:Aiming at the problems of long signal sequence and poor feature robustness in Feature Engineering in individual recognition, the technology based on deep neural network is studied. Drawing lessons from Convolutional, Long Short-Term Memory, Fully Connected Deep Neural Networks(CLDNN) in speech recognition, the local amplitude features of the signal are extracted through convolution neural network and the global time-domain features of the signal are extracted through long-term and short-term memory network. A fully connected network is utilized to map the feature to the device label. Under the line of sight channel, the data of eight Lora modulated wireless data transmission stations are collected, and the Gaussian white noise is added to the simulation test. The simulation shows that when the Signal-to-Noise Ratio is low(0 dB) , the accuracy of the model can reach nearly 95% under the signal sequence length of 2 048 points. In addition, this model needs fewer parameters compared with VGG16 model. The proposed model has a certain application prospect in the deployment of Internet of things devices.

Abstract:The radio frequency fingerprints are inherent features of the device hardware, and will not change with the transmitted signal, therefore they are often used in communication anti-spoofing. In this paper, the neural network is adopted to process the original signal samples obtained by the receiver, including I/Q sequence, amplitude/phase, binary image of constellation diagram and color density diagram of constellation diagram to achieve anti-deception effect. When the signal-to-interference and noise ratio is in the range of -30 dB to 30 dB, the signal recognition accuracy can reach up to 99.93%. Being different from the existing literature, the method can be adapted to the scenes with different signal-to-interference and noise ratios. This research shows that the proposed method is feasible to achieve anti-spoofing in a complex communication environment where spoofing signals and legal signals coexist.

Abstract:A communication signal modulation recognition method based on Transformer model is proposed. In the data preparation stage, a Different Symbol Rate Modulation Recognition(DSRMR) data set is constructed. In the data preprocessing stage, a method of I/Q data enhancement is proposed to meet the quantitative and diverse requirements of model training, and to enhance the generalization ability of the model. In the model construction stage, the method of slice serialization is introduced into the modulation recognition Transformer model, and it is employed to optimize the input problem of the Transformer neural network model. Experimental results prove that the communication signal modulation recognition method based on the Transformer model can obtain high-precision in signal automatic modulation recognition.

Abstract:In order to solve the adaptive problem of generating jamming waveforms in response to enemy communication signals in Unmanned Aerial Vehicle(UAV) swarm communication confrontation, a cognitive jamming waveform design method is proposed based on Spectrally Modulated Spectrally Encoded(SMSE) model. By deploying six kinds of different waveform design parameters in the SMSE framework, the waveforms with a specific spectrum structure are synthesized in the frequency domain to generate cognitive jamming waveforms with corresponding functions to deal with communication signals under different parameters in communication countermeasures. Simulation experiments show that the theoretical model can generate the suppressive jamming waveforms: single-tone jamming, multi-tone jamming, wideband interference, and narrowband interference. It can also achieve modulation deceptive jamming for the modulation methods of BPSK, QPSK, 8PSK, and 16QAM signals. Compared with the theoretical curve, the experimental results verify the theoretical feasibility of SMSE model to generate cognitive interference waveform. SMSE model can realize the integration of suppressive jamming and spoofing jamming.

Abstract:Signal modulation identification technology has important applications in both civilian and military fields. In the current information battlefield, due to the increasing number of information radiation sources such as various radars, communications, navigation, and electronic warfare weapons, the modulation forms are becoming more and more diverse, and the signal density is increasing, which makes the electromagnetic environment of war increasingly complicated, therefore the traditional signal modulation identification technology has been unable to adapt. A robust feature extraction, fusion and recognition technology of complex communication modulation signals is put forward, and a deep learning-based AlexNet network and complex neural network are proposed. Multimodal information in the statistical graph domain and signal I/Q waveform domain is fused for signal modulation identification. The simulation results show that the recognition accuracy of the proposed method is higher than that of the single-modal recognition method and the method without the multi-modal collaborative fusion framework under different Signal-to-Noise Ratios(SNRs).

Abstract:Spectrum data contains a large number of signals in radio monitoring. Accurate extraction of these signals is conducive to mastering the spectrum usage of the whole frequency band. Due to the interference of noise, several energy values of frequency points in the signal spectrum bandwidth will be lower than the detection threshold, then the traditional threshold detection algorithm will misestimate the signal as multiple signals and generate multiple false adjacent signal intervals, resulting in a decline of the spectrum signal extraction accuracy. To tackle this problem, an algorithm of spectrum signal extraction for estimating the signal number adaptively is proposed according to the characteristics of false adjacent signal intervals. The new method can estimate the number of electromagnetic signals in the spectrum monitoring data, and extract the corresponding signal and spectrum information accurately and automatically. The experimental result shows that the new method is adaptive, strongly robust and accurate, effectively improves the accuracy of spectrum signal extraction. It can provide basic electromagnetic signal data for supporting the identification of electromagnetic environment in military and civil spectrum monitoring.

Abstract:In the context of Cognitive Radio(CR), dynamic spectrum access has become a key approach to improve the spectrum utilization in wireless network. In this paper, using fine-grained spectrum measurement data collected from Global System for Mobile Communications-Railway(GSM-R), a data-driven deep-learning method is proposed to model the spectrum pattern and a framework is developed for dynamic spectrum access. A deep spectrum generative model is adopted to guide the spectrum allocation. A deep network that combines recurrent series model and background feature embeddings is developed to model and predict the short-term spectrum occupancy, then a strategy is proposed for dynamic channel access. Furthermore, a frequency-hopping system is implemented with Software Defined Radio(SDR) platform and it is integrated with the proposed strategy. The throughput capacity of this system is evaluated with real-world historical spectrum data. It is shown that the proposed method and system can enhance the ability of opportunistic communication and utilize the spectrum resource efficiently. The proposed spectrum access framework and the implementation with SDR are of great generality, so that they can be easily integrated into systems with different scenarios and frequency spans.