Abstract:Cognitive electronic warfare is usually defined as a form of electronic warfare that is based on electronic warfare equipment with cognitive performance and focuses on autonomous interactive electromagnetic environment learning capability and dynamic intelligent confrontation task processing capability. Since it was first proposed, it has attracted extensive attention from researchers and scholars at home and abroad for its advantages of accurate perception, strong reasoning and fast decision-making. With the continuous emergence of new concepts, technologies and applications of artificial intelligence, cognitive electronic warfare has stepped into a brand new stage of development. In order to capture its future development direction, the connotation of the concept of cognitive electronic warfare was summarized and enriched from the perspective of artificial intelligence, the development of cognitive electronic warfare and typical foreign projects were sorted out, the framework and architecture of cognitive electronic warfare system was built, a comprehensive and systematic review of the key technologies of cognitive electronic warfare was conducted from the aspects of perception, judgment, decision-making, etc., and the challenges and development trends of cognitive electronic warfare were summarized.

Abstract:In order to solve the problems of high dimension of reconstructed feature vector, low computational efficiency and poor robustness of existing phase space based individual recognition methods, SEI(specific emitter identification) framework based on reconstructed attractors was proposed from the perspective of nonlinear dynamics. Within the proposed framework, a novel SEI technology based on Isomap(isometric mapping) was developed. The technology used Isomap to reconstruct the emitter attractor from phase space, which can describe the dynamic characteristics of the emitter system in a lower dimension and reflect the “fingerprint” characteristics of the emitter individual. Experiments show that the proposed method can achieve higher accuracy, higher efficiency and better robustness.

Abstract:To address the problem of mitigating asynchronous non-stationary interference in single-channel conditions, a data-driven sparse component analysis method was proposed. The aim of this method is to recover the desired signal from the received mixed signals. This method used the powerful modeling ability of deep convolutional neural network to model the complex mapping between the input and output data, and realized the adaptive selection of sparse domain of target signals, the adaptive learning of sparse representation of target signals in sparse domain, and the automatic recovery of target signals. Unlike the previous interference mitigation algorithms, the proposed method completed the “end-to-end” signal waveform recovery in the time domain, and had no prior requirement for aliasing observation, which was more universal than the existing methods. Simulation experiments verified the effectiveness of the proposed interference mitigation method under different environmental noise and interference signal strength and generalization test conditions, and the interference mitigation performance is significantly better than the existing algorithms.

Abstract:Deep learning is easily attacked by adversarial examples. Taking communication modulation recognition as an example, adding adversarial perturbations to the transmitted signal can effectively prevent non-cooperative users from utilizing the deep learning method to recognize the modulation of the signal. Thus, adversarial perturbations can help enhance communication security. To address the problem that the existing adversarial attack techniques are difficult to meet the adaptive and real-time requirements, the universal adversarial perturbation applicable to the whole dataset was obtained by the principal component analysis of the adversarial perturbation generated by a small part of the data extracted from the dataset. The computation of the universal adversarial perturbation can be carried out under offline conditions and then added to the signal to be transmitted in real time, which can satisfy the real-time requirements of communication and realize the purpose of reducing the accuracy of non-cooperative party modulation recognition. Experimental results show that the proposed method has better deception performance relative to the baseline method.

Abstract:Existing deep learning based convolutional code recognition methods still have shortcomings such as large parameter sizes and weak recognition performance. Aiming at this problem, a convolutional code recognition method based on joint learning of matrix transform features and code sequences was proposed. The received codeword sequence was arranged into a matrix form, and the soft information was used to eliminate the codewords with low reliability. Then, a new matrix transformation algorithm was used to obtain the feature matrix. During the recognition process, the original matrix of code words and the matrix of features were fed into a network model with a joint learning capability for multimodal data. The feature extraction fusion and convolution code recognition were completed in the neural network. Simulation results show that the recognition performance of the proposed method is significantly better than the existing recognition methods based on deep learning, especially for high bit rate convolutional codes. When the rate is low, the proposed method is also better than traditional methods. When the signal-to-noise ratio reaches 5 dB, the recognition rate of 25 convolutional codes with different parameters can reach 100%.

Abstract:GNSS/INS ultra-tight integrated navigation system has become a research hotspot in the field of integrated navigation due to its high positioning accuracy, excellent dynamic performance and strong anti-jamming ability. The principle of GNSS/INS ultra-tight integration was introduced, and the advantages and features of ultra-tight integration mode relative to other integration modes were compared based on the analysis of the technical principle. The domestic and international research status was introduced, represented by ultra-tight integration under high dynamics and MIMU/GNSS ultra-tight integration. The key technologies such as fault-tolerant control technology, neural network assistance, and multi-sensor assisted ultra-tight integration were summarized, and the prospect of GNSS/INS ultra-tight integration towards low cost, high precision and strong stability was prospected.

Abstract:Filter order is the core parameter affecting the time-domain adaptive anti-jamming performance and the computational complexity of satellite navigation receivers. In order to solve the problem that the current order selection is heavily dependent on engineering experience and the impact of the analysis is insufficient, the impact of the filter order on the time-domain adaptive anti-jamming performance of navigation receivers was analyzed, providing theoretical research for low-complexity time-domain anti-jamming. The analysis was oriented to different interference environments. The filter amplitude-frequency response and the signal carrier-to-noise ratio were used as evaluation indicators. The simulation experiments and practical tests verified the traditional LMS(least mean square) algorithm and the improved LMS algorithm. Finally, an adaptive design method for optimal filter order based on digital filter was proposed. Experimental data analysis shows that the anti-interference performance of the time-domain adaptive filter can be effectively improved by appropriately increasing the filter length. In practical applications, the filter order can be optimized by constraining the carrier-to-noise ratio loss.

Abstract:Due to the low transmitting power and large transmission loss, the satellite signal reaching the GNSS receiver is extremely weak, and easy to be interfered. In order to deal with the threat of electromagnetic interference, taking measures at both the system and user level would be a good choice. A model for analyzing the interference suppression capability of anti-jamming receiver with null-steering antenna was established, and the improvement effect of signal power enhancement on the interference suppression performance was quantitatively analyzed. Results show that, the interference mitigation capability of the receiver can be improved by 3～4 dB per 10 dB signal power enhancement. However, from the perspective of improving the receiver′s reception performance(carrier to noise ratio, ranging and positioning accuracy, etc.) under non limiting conditions, when the signal power is increased by 15～20 dB, the reception performance is optimal. The research result can guide the optimization of the signal power enhancement and anti-jamming design of receivers.

Abstract:The spinning vehicles are typical applications of GNSS (global navigation satellite system) receivers. When the GNSS carrier spins, the rotation will introduce higher order dynamics making the traditional tracking loop out of lock. Coupling with the INS (inertial navigation system) can effectively compensate the high order dynamic of signal carrier phase. In this case, a design of ultra-tight coupling GNSS tracking loop, which used the solutions of INS to aid the tracking of GNSS signal, was proposed. Besides, the relationship among the aiding rate of INS, the angular rate of spinning vehicle and the error of signal carrier phase was also analyzed. The simulation results show that the proposed tracking loop structure can effectively deal with the problem of signal tracking in the spinning vehicle, and significantly enhance the precision of position compared to the single GNSS navigation results.

Abstract:A GNSS multipath channel sparse fitting scheme based on K-medoids clustering was proposed to tackle the problem of a large amount of simulation computation and hardware resource overhead for GNSS channels, which is inconvenient for real-time performance evaluation and practical engineering applications. The equivalent reduced CIR (channel impulse response) parameters were extracted using a sparse fitting method based on K-medoids clustering CIR parameters extraction, and the channel simulation was realized using a sparse tapped-delay-line structure. The proposed method sparsely fits the original GNSS multipath channel model under the constraint of retaining multipath error by employing tapped-delay-line structure filter with fewer taps, which can decrease the complexity of simulation without requiring huge hardware resources. Simulation results show that the proposed scheme is effective by sparse fitting the CIR parameters generated by the reference channel model.

Abstract:In view of the contradiction that the GPU processing efficiency was limited under the short update interval and the traditional tracking loop was not robust under the high dynamic scenario under the long update interval, a carrier phase tracking algorithm with long update period for high dynamic scenes was proposed. A low complexity LFM signal parameter estimation algorithm was designed to estimate Doppler and its rate of change at the initial stage of tracking, thereby eliminating most signal dynamics, and during the tracking process, a 4-order Kalman filter was used to precisely track the residual signal phase and dynamic. It is verified by simulation that under the 200 ms update interval, fast and stable tracking of the carrier phase can be realized in sinusoidal motion scenes with Doppler primary/secondary change rates of 800 Hz/s and 64 Hz/s², respectively. The convergence can be achieved with only one update, and the tracking sensitivity is as low as 23 dB-Hz; the phase tracking accuracy is far better than the traditional third-order phase locked loop.

Abstract:A low-complexity method based on pulse blanking for fast frequency-sweep interference mitigation in satellite navigation receivers was proposed. Different from traditional methods based on time-frequency analysis, this method converted the continuous wave interference into pulse interference through a low-pass filter, and then pulse detection and blanking were used to mitigate the interference. The interference was blanked as a pulse when it located in the pass-band of the low-pass filter. Otherwise, it was suppressed as an out-band interference when it located in the stop-band of the low-pass filter. Theoretical analysis and experimental results show that the computation complexity of this method is reduced by an order of magnitude compared with that of traditional method, and it can achieve similar interference suppression performance.

Abstract:Penetrating imaging radar technology uses the characteristics that the electromagnetic wave can penetrate the medium, to get high-resolution image of discontinuities in the non-metallic medium. In order to realize millimeter high plane resolution, high detection efficiency and high portability of the radar system, a high plane resolution penetration imaging radar system was designed. Continuous wave system and fast scanning spatial sampling scheme were adopted to ensure miniaturization and high imaging performance. An integrated radar RF front end was realized. Data processing methods such as autofocus imaging processing under the condition of unknown parameters were proposed. A prototype of penetrating imaging radar system, which weighs only 2.5 kg and can be operated by one person and one hand, was developed. The experimental tests of imaging resolution and penetration ability were carried out to verify the feasibility and effectiveness of the scheme.

Abstract:In the actual tracking scenarios of the CMIMO (collocated multiple-input multiple-output) radar, the high dynamic RCS(radar cross section) fluctuation characteristic is not utilized effectively and thus it will lead to low tracking accuracy or even missing tracking. To solve this problem, a CMIMO radar power resource adaptive allocation method based on the high dynamic RCS fluctuation characteristic was proposed. Note that the target RCS was sensitive to the observing angle and the actual observing angle could be obtained dynamically via the prediction of target kinetic state, thus the polarization mode could be optimized during different tracking frames. Thereafter, the tracking posterior Cramer-Rao bound which included radar transmitting power and RCS was derived and it could see as the object function to be optimized. Finally, the internal penalty function method was implemented to tackle the aforementioned optimization problem and it achieved the optimized power allocation with high dynamic RCS. Simulation results validate that compared with the traditional RCS model allocation method, the proposed method fully utilizes the dynamic RCS fluctuation characteristics to achieve the effective allocation and it solves the mismatched problem between the allocation scheme and the actual tracking scenarios, which improves the multi-target tracking performance of the CMIMO radar.

Abstract:A modulation recognition algorithm based on signal amplitude distribution and higher-order spectral characteristics was presented. The algorithm mainly utilized the amplitude distribution of orthogonal frequency division multiplexing subcarrier signal after orthogonal demodulation by inverse Fourier transform, realized multiplexing phase shift key and orthogonal amplitude modulation recognition by histogram statistics, and the modulation order in the multiplexing phase shift key class was recognized by multi-spectral analysis. Compared with the classical modulation recognition algorithm based on high-order cumulant, it has better adaptability to carrier frequency residual deviation, and the recognition rate is improved under the condition of carrier frequency offset. Compared with the cyclostationary method, it has better signal-to-noise ratio adaptability. The simulation results show the effectiveness of this method. Under the same recognition rate, the adaptability is improved. The simulation results show the effectiveness of this method.

Abstract:Aiming at the problem of the performance degradation of the traditional tracking algorithm when confronted with glint noise, a high performance filtering method named as IMM-CKF was proposed by integrating the CKF(cubature Kalman filter) into the framework of IMM(interacting multiple model). In the proposed algorithm, the target state was modeled as Gaussian distribution, the glint noise was modeled as Gaussian mixture distribution, and the occurrence probability of the glint noise was modeled as the first-order Markov process. An IMM framework was then used to implement model-matched filtering for each Gaussian component. To further mitigate the impact of nonlinear observation condition on tracking accuracy, the CKF was utilized as Gaussian approximation filter to realize recursive prediction and update of the target state. Simulation results show that the proposed method not only has higher tracking accuracy than traditional algorithms such as Gaussian sum filter and particle filter, but also has better real time ability. Additionally, the IMM-CKF can effectively estimate the existence of glint noise.

Abstract:To extend the existing real-valued BCS(Bayesian compressive sensing) framework to the complex-valued one, a CMBCS-MLP(complex multitask Bayesian compressive sensing algorithm using modified Laplace priors) was developed to eliminate the impact of measurement noise variance, and a fast algorithm based on sequential operations was further derived. It is demonstrated by numerical examples that the developed CMBCS-MLP algorithm is more accurate and robust than the existing algorithms in the complex sparse signal reconstructions.

Abstract:To resolve the problem of high hardware consumption and difficulty in engineering implementation in digital Stretch processing with large decimation multiples, a new method based on cascading segmented FFT (fast Fourier transform) processing was proposed. The data sequence after digital mixing is cascading segmented and reorganized, and then the digital Stretch processing was realized by using the less points FFT operation. The experimental results based on the measured data prove the effectiveness of the presented strategy. The resource consumption analysis demonstrates that the proposed method can implement digital Stretch more hardware-efficiently compared with the conventional method.

Abstract:The distortion in the imaging of space-based optical cameras for actual in-orbit earth observations needs to be suppressed by geometric correction. At present, the small size and high frame rate sequential images obtained by mainstream area array cameras for ground observation are difficult to meet the requirements of traditional geometric correction methods for calculating the number of control points and spatial distribution of frame-by-frame solution, and the computational complexity is huge. To address this issue, a geometric correction method by using equivalent bias angle sparse measurement for sequential observation image of area-array sensor was proposed. The problem of parameters solution for each frame converted to the problem of equivalent bias angle recovery under the time-domain sparse measurement. The requirement for the number and spatial distribution of control points in a single frame image were reduced by using the time-frequency information of equivalent bias angle signal. Meanwhile, the real image data from the area-array camera of Gaofen-4 satellite was used to verify the validity and low computation of the proposed method.

Abstract:Aiming at the difficulty of mainstream methods to support the multilevel real-time visualization of large-scale geographic vector linestring data, a multilevel real-time visualization technique for large-scale geographic vector linestring data was proposed. An adaptive visualization model for multilevel tile rendering was established, and a PQR (pixel-quad-R) tree spatial index and an adaptive visualization algorithm based on PQR-tree were designed to support the data organization and visualization of the model, respectively. Experiments on billion-scale datasets show that the technique can calculate visualization results at any zoom level within 0.57 s. Meanwhile, its visualization time is significantly less than mainstream methods. When the data scale increases sharply, the technology still has good visualization performance at each zoom level, and the lowest visualization rate exceeds 100 tiles/s, which is much better than mainstream methods. The technique can support multilevel real-time visualization of large-scale geographic vector linestring data in the single machine, and has a good application prospect in the field of exploration and analysis of spatial big data.

Abstract:ROP(return oriented programming) attack is an important method for network attackers to break through the protection of operating system and realize vulnerability attacks, and ROP chain is the main component of ROP attack. In order to detect the ROP chain in network traffic, an intelligent detection method that can automatically extract the characteristics of ROP chain and has good generalization performance was proposed. The sequential extraction method was adopted to divide the measured network traffic into multiple sequences, one-dimensional traffic data was converted into two-dimensional feature vectors by using sliding window and numerical quantization, and the detection of ROP chain was realized based on the convolution neural network model. Different from the existing static detection methods, the proposed method did not rely on the context information of the program memory address, was simple to implement, easy to deploy, and had excellent detection performance. The experimental results show that the highest accuracy rate of the model is 99.4%, the false negative rate is 0.6%, the false positive rate is 0.4%, the time cost is within 0.1 s, and the false negative rate for the real ROP attack traffic is 0.2%.

Abstract:With the widespread application of computer three-dimensional vision, point cloud processing algorithms based on deep learning have attracted a lot of research in recent years, however, the time and storage consuming characteristics have greatly restricted its deployment and application on the mobile terminal devices. Based on the general idea of improving the loss function, a new point cloud deep model compression framework was proposed, and the knowledge distillation method was introduced into the binary quantization model. At the same time, considering the speciality of the point cloud aggregation operation, an auxiliary loss item was introduced. The improved loss function includes three parts:prediction loss, distillation loss and auxiliary loss. The experimental results show that, compared with the existing algorithms, the proposed algorithm can obtain higher accuracy, meanwhile, the application to current mainstream point cloud deep network models can also achieve good scalability.

Abstract:Memristor based neuromorphic computing architecture has achieved good results in image classification, speech recognition and other fields, but when the memristor array has the problem of low yield, the performance declines significantly. A method combining memristive neuromorphic computing based calibration method with in-situ training was proposed, which increased the accuracy of multiplicative accumulation calculation by using the calibration method and reduced the training error by using the in-situ training method. In order to verify the performance of the proposed method, a multi-layer perceptron architecture was used for simulation. From the simulation results, the accuracy of the neural network is improved obviously (nearly 40%). Experimental results show that compared with the single calibration method, the precision of the network trained by the proposed method is improved by about 30%, and the precision of the network trained by the proposed method is improved by 0.29% when compared with other mainstream methods.

Abstract:UR-SB (unrecoverable-sensitive bits), which cannot be corrected by regular refresh, will cause long-term interruption of on-orbit service of satellite load. Thus, the impact of UR-SB needs to be evaluated and improved deeply by the fault injection tests. However, the proportion of UR-SB is extremely low. If the traditional bit-by-bit upset fault injection method is adopted, the fault injection tests would take too long time, and the efficiency is extremely low. A fault injection acceleration algorithm for static random access memory-based field programmable gate array based on whole frame upset was proposed, which can quickly filter out the configuration frames without UR-SB through whole frame upset fault injection. By taking dichotomy on the configuration frames with UR-SB, the precise positioning process of UR-SB can be speeded up effectively. Taking the commonly used XQR2V3000 device as an example, the simulation results indicate that the test efficiency can be improved by 207 times under the poor conditions, and the real data experimental results of the signal generation load by our group are increased by 949 times. These results demonstrate the validity of the acceleration algorithm proposed in this article.

Abstract:In view of the problems such as inefficient data loading, insufficient utilization of multiply-accumulates resources, complex output buffering and addressing logic in existing neural network accelerators when processing sparse neural networks, a high-efficiency data loading and output buffering strategy for sparse convolutional computing was proposed. It performed an all-to-all multiply-accumulates operation on the non-zero input feature map data and the non-zero weights belonging to the same input channel, which reduces the difficulty of non-zero data pairing and improves the utilization of multiply-accumulates resources. By using input stationary calculation and intensive cyclic loading of input feature map data, it significantly reduced the number of data off-chip fetches. It optimized the output buffer design and solved the problems of address access contention and storage congestion during output buffering in existing solutions. Experimental results show that, when compare to fine-grained systolic accelerator with similar architectures, the process element area of the proposed architecture is decreased by 21.45%; the data loading speed is increased by 117.71% on average; the average utilization of multiplier is increased by 11.25%, reaching 89%.

Abstract:To address the problems of inefficient data loading and readout and poor flexibility of array collaboration in existing multi-memristor-array, a highly efficient and flexible multi-array interconnection architecture was proposed. The data loading strategy of the architecture supports data reuse in multiple weight mapping modes, reducing the need for off-chip data access; the readout network supports flexible combination of multiple processing units to achieve different scales of convolutional operations, as well as fast accumulation and readout of computation results, thus improving chip flexibility and overall computing power. Simulation experiments performed on the NeuroSim platform with running VGG-8 networks indicate a 146% increase in processing speed than that of the MAX² neural network accelerator, with only a 6% increase in area overhead.

Abstract:In order to solve the problem of lightweight and high-reliability anti-irradiation hardening for BRAM(block random access memory) in static random access memory-based field programmable gate array under the strict limitation of space-borne resources, BRAM anti-irradiation hardening design method based on time-sharing refreshing and location constraint was proposed. The time-sharing refreshing of the BRAM was realized through monitoring time slot of algorithm execution, and location constraint was added to effectively reduce the probability of simultaneous anomalies of two modules under the design of triple modular redundancy, effectively improving the reliability of BRAM radiation resistance with less resource consumption. The results of heavy ion acceleration test show that after adopting the time-sharing refreshing and location constraint hardening methods, the single event function interruption cross section of a certain type of satellite payload in our laboratory has decreased by about 81.63%. The BRAM anomaly on navigation satellites has been reduced from 3 stars, which occurred 3 times in 2 years, to 25 stars that have not occurred in 2 years, and the reliability of radiation resistance has been greatly improved.