Abstract:We propose the concept of rate entropy function to define generalized rate distortion function from the decoder side by replacing traditional distortion constraint with decoder uncertainty. Although rate entropy function is defined as a constrained variational problem of mutual information， the closed solution can be found by constructing a special solution. We propose four methods for constructing special solutions， such as entropy invariant criterion， independent error criterion， regeneration criterion and weak regeneration criterion. Accordingly， closed expressions of rate entropy function for the current common probability distributions are derived， including uniform distribution， vector Gaussian distribution and probability distributions with regeneration and weak regeneration. Entropy distortion and entropy power distortion are generalizations of mean square distortion （second-order statistic） and absolute value distortion （first-order statistic）， which are more general. The concept of rate entropy function solves rate distortion function problem of currently known common sources， enriches and develops Shannon’s theory of rate distortion function， and has important theoretical significance and application value in source coding.

Abstract:Non-continuous orthogonal frequency division multiplexing （NC-OFDM） technology is an application of cognitive radio （CR）. In NC-OFDM communication systems， cognitive users sense the spectrum holes in authorized frequency to reduce the waste of spectrum resources. In order to ensure the communication quality of authorized users， cognitive users need to exit immediately when authorized users access， which makes the real-time requirements of NC-OFDM systems higher than those of OFDM systems. Traditional resource allocation algorithms designed for OFDM systems cannot meet real-time requirements. Considering the spectrum features of NC-OFDM systems， two fast resource allocation algorithms are proposed： Firstly， the power threshold is calculated based on the bit error rate and gain-to-noise ratio， and a single judgment is used to select the sub-channels used by the cognitive users， which reduces the dimensionality of the solution space； Secondly， the water-level is calculated directly and the power and bit allocation are initialized. Finally， based on dichotomy， the remaining resource of sorted channels （method 1） or unsorted channels （method 2） is allocated. The theoretical analyses and simulations show that the proposed algorithms have the same （method 1） resource allocation results or the similar resource allocation results （method 2） as the optimal algorithm. The calculation complexity of the proposed methods is lower， which significantly speeds up resource allocation.

Abstract:Unitary matrix transformation is a commonly used real-valued method， which can effectively reduce computational complexity. However， the dimension of the observation matrix is doubled in the existing downlink channel estimation method for massive Multiple input multiple output （MIMO） systems based on unitary matrix transformation. Without dimensional compression， the goal of reducing computational complexity is difficult to achieve. Although the orthogonality of signal space and noise space can compress the dimension， the signal space can only be approximately calculated， leading to performance loss. To improve channel estimation performance， the signal space matrix is regarded as a variable and adaptively adjusted in the process. Since the signal space matrix and sparse signal matrix are highly coupled， the traditional Bayesian inference is not applicable. Therefore， the column-independent variational Bayesian inference (VBI） factorization is adopted to decouple the signal space matrix and sparse signal matrix successfully. Simulation results show that this method can significantly improve the channel estimation performance.

Abstract:The three-dimensional （3D） spectrum situation is a significant method to handle the issue of making full use of spectrum resources in the ground-air-space integrated information network. It characterizes the time-space-frequency distribution of the power spectrum density in 3D electromagnetic space. Hence， the communication system can implement a variety of applications such as spectrum prediction， spectrum decision， and spectrum management and control in a targeted manner. However， due to user deployment and other factors， the actual 3D spectrum situation is often discrete and incomplete. Therefore， this paper proposes a 3D spectrum situation completion algorithm based on generative adversarial networks （GANs）. Moreover， this paper proposes an improved GANs structure and data processing methods to decrease the completion error and the training time of the algorithm. Simulation results indicate that the proposed algorithm can effectively complete the 3D spectrum situation and outperform the conventional interpolation method in terms of completion accuracy.

Abstract:Aiming at the problem that the narrow beam cannot be matched in real time due to the relative motion of the transmitter and receiver in the millimeter-wave （mmW） communication system of unmanned aerial vehicles （UAVs）， a three-dimensional （3D） beam tracking method based on unscented Kalman filter （UKF） is proposed. Firstly， the elevation and the azimuth angle of the beam at the receiver and the transmitter are regarded as the state vector of the system， and the sampling point set is obtained by the unscented transform （UT）. Then， the state prediction value and the measurement prediction value are calculated according to the sampling point set. On this basis， the state vector is updated according to the calculated Kalman gain to obtain the optimal estimate of the state vector. Simulations show that the proposed method can improve the beam tracking accuracy of UAV in 3D dynamic environment.

Abstract:Low density parity check （LDPC） code hopping can improve the security and reliability of information transmission through error control coding based on the hopping check matrix at the physical layer. Quasi-cyclic low density parity check code （QC-LDPC） is widely used because of its good error correction performance and easy engineering implementation. This paper proposes a QC-LDPC code hopping design method. Firstly two kinds of subgroups in finite field are used to design the hop-basis matrix， and then the check matrix of the hashed basis matrix is masked by the protograph-based external information transfer （PEXIT）， so that the hop matrix has a unified architecture and a fast coding structure， which is easy to implement in engineering. Simulation and analysis show that the designed QC-LDPC codehopping has a huge code hopping set and good error-correcting performance. The number of LDPC codes in a code hopping set can be up to 10³⁴. With the increase of code length， the number of LDPC codes in a code hopping set increases exponentially， and the average performance of LDPC codes is comparable to that of many protocols. It can be used to improve the reliability and security of the communication system.

Abstract:As traditional UAV trajectory planning algorithms generate redundant search points and poor real-time path planning in emergent threat scenarios， a dynamic trajectory planning algorithm for UAVs based on multi-factor Dubins path is proposed. First， the algorithm uses the traditional Dubins path to find effective path extension points based on the UAV’s own performance constraints and the location of the sudden threat area. Sencond， a path extension point evaluation function is established based on the path length and threat combined with heuristic search ideas. Finally， the path extension point is selected through the path evaluation function， and a better path is planned. The simulation results show that when the algorithm is used for trajectory planning in sudden threat scenarios， the path length is shorter and the path extension points are fewer， and it conforms to the change of the heading angle of the drone during the actual flight， which can effectively ensure the safety of the drone and real-time trajectory planning.

Abstract:As the data compression methods based on sparse representation （SR） have the problems of low compression ratio and reconstruction accuracy， a dual-iteration concentrated dictionary learning algorithm is proposed. This algorithm maps high-dimensional signals to low-dimensional feature spaces. If features of the high-dimensional original signal are retained by the low-dimensional feature space， higher accuracy will be achieved when the high-dimensional signal is reconstructed from the low-dimensional feature space. To keep more principal components of high-dimensional dictionaries in low-dimensional dictionaries， a new transformation algorithm named ? transformation is proposed. It can improve the energy concentration of the high-dimensional dictionary. Further， aiming at the coupling relationship between the high-dimensional dictionary and the low-dimensional dictionary， a dual-iteration training method is established to improve the energy concentration and the expressive ability of the dictionary. Experiments show that， compared with the traditional algorithms， the convergence speed of the proposed algorithm is improved by more than three times. In addition， a higher compression ratio and a higher quality reconstructed signal are obtained.

Abstract:Generalized spatial modulation （GSM） technology transmits information by selecting part of the transmitting antenna. This system has a low transmission rate due to some antennas remaining silent. To improve the transmission rate of the system， all the transmitting antennas of the GSM are activated. The system still can hardly recover the transmitted signal at the receiving end. In this paper， a two-mode generalized spatial modulation system is proposed， the information bits sent in each time slot of the system are divided into index bits and constellation modulation bits， and all transmitting antennas are divided into two groups by the index bit. Two distinguishable constellation modulation symbols are transmitted simultaneously on two groups of antennas. The maximum likelihood （ML） algorithm is used to detect and demodulate the transmission symbol at the receiver. The proposed system achieves the balance between transmission efficiency and system performance. Analysis and simulation show that when the bit error rate is 10^-3， the dual-mode generalized spatial modulation system is compared with the GSM and the multiple input multiple output （MIMO） system with same total transmit power and transmission rate. The former adopts low-order modulation to maximize the Euclidean distance between constellation points， which can obtain a gain of about 4 dB， while the latter can improve the performance of detection algorithm through power index at the receiver end， which can obtain a gain of about 2 dB.

Abstract:In massive MIMO systems， the existing Gauss-Seide （GS） algorithm has lower complexity than the minimum mean-square error （MMSE） algorithm， while the detection performance is worse. This paper proposes a Jacobi-improved Gauss-Seide （JA-IGS） detection algorithm suitable for the uplink detection of massive MIMO systems. The algorithm first optimizes the iterative initial solution by introducing a Jacobi （JA） pre-iterator. Then the traditional GS is optimized linearly. With the increase of lower complexity， the detection performance and convergence speed are significantly improved. Simulation results show that compared with traditional GS and JA detection algorithms， the algorithm has a lower Bit error ratio （BER） and higher computational efficiency.

Abstract:Aiming at the problems of expensive hardware costs and poor real-time performance for the channel fading twin technology in real communication scenarios， based on the greedy coordinate rotation digital computer （CORDIC） algorithm and sum of frequency modulated model， a discrete non-stationary complex fading channel emulation scheme is developed. Furthermore， large scale complex exponential computations are realized in the field programmable gate array （FPGA） hardware platform. By introducing the domain folding technique， the greedy angle recording unit and a full parallel pipeline structure， the hardware resource consumption and real-time performance are significantly improved. In addition， the hardware resource can be further optimized by adopting the compact architecture with time-division and multi-rate scheme. Compared with the traditional look up table （LUT） method， hardware resource consumption is greatly reduced from 17.89% to 6.71%. Meanwhile， the hardware latency is reduced by 65.625% than classic CORDIC algorithm. The hardware measurement results show that the statistical properties of channel output， i.e.， the probability density function provides a good agreement to the theoretical ones.

Abstract:To solve the problem of low recognition rate of interrupted-sampling and repeater jamming， a method of jamming recognition based on short-time fractional Fourier transform（STFRFT） was proposed. In order to reduce the complexity of feature extraction in time-frequency domain， the time-frequency domain was reduced to fractional Fourier transform（FRFT） domain and time domain based on maximum search and slice after STFRFT was used to interrupted-sampling and repeater jamming. Feature parameters contained the number of spectrum peaks in FRFT domain， and the number of pulses in time domain after dimensionality reduction. For three typical kinds of interrupted-sampling and repeater jamming， simulation result shows that a satisfactory recognition effect is achieved when the jamming-to-noise ratio is greater than -4 dB.

Abstract:In order to ensure the safe transmission of the time reversal-orthogonal frequency division multiplex （TR-OFDM） system， this paper proposes a secure transmission algorithm based on quaternion encryption. The scheme is mainly divided into three steps. In the first step， the transmitter and the legitimate receiver use the estimated channel to obtain the quaternion required in the process of encrypted transmission， that is， the key. In the second step， the transmitting end takes the bit sequence to be transmitted three-dimensional mapping， then uses the quaternion to rotate and encrypt the three-dimensional constellation points， and finally modulates them into OFDM symbols and transmits them after time inversion processing. In the third step， the legitimate user uses the quaternion to decrypt and demodulate to obtain the transmitted data. The eavesdropper cannot obtain the transmission information because they do not know the key. Therefore， the proposed scheme ensures the security of system data transmission. The simulation results show that the proposed algorithm can keep the bit error rate of eavesdropping users at about 0.5. Under the same SNR， legal users can achieve a lower bit error rate than traditional two-dimensional modulation. Compared with the artificial noise scheme， the proposed algorithm will not affect the bit error rate of legitimate users.

Abstract:In pipeline nondestructive testing， the traditional gray-scale visualization method of magnetic flux leakage （MFL） data has the disadvantages of large display delay and low identification degree for the defect curve view and the gray-scale view， respectively. To solve these problems， an self-adaptive sliding window based gray feature enhancement method for MFL data is proposed. Firstly， according to the characteristics of MFL data， the unequal classification label is established and the down-sampling ratio is set to realize the down-sampling display of the original MFL data. Then， according to the preprocessed MFL data， the self-adaptive sliding window and gray value compensation algorithm are designed to realize the local gray mapping of MFL data. Finally， an self-adaptive gray mapping method is designed based on MFL data classification label to get a clear gray view of MFL data. Some comparative experiments are conducted to verify the advance and effectiveness of the proposed method.

Abstract:An improved DV-Hop algorithm based on hop-count calculation， hop-distance estimation and coordinate estimation algorithm is proposed. Firstly， the hop number is optimized by the multi communication radius. Then the average hop distance is modified by the weighting method， and then the received signal strength indicator （RSSI） ranging technology is used to define the average hop distance error to achieve the second correction. Finally， there is a need to eliminate the heteroscedasticity of the error term and further reduce the error by the weighted least square method. The simulation results indicate that the improved algorithm can reduce the positioning error while maintaining a small time complexity.

Abstract:Emotion plays an important role in singing. At present， vocal music courses in universities lack effective training for emotion mobilization. Due to the authenticity and immersive of virtual reality （VR） technology， this article applies it to the emotional induction stage in vocal music teaching. To verify the effectiveness of VR technology for emotion induction， different types of electroencephalogram （EEG） features are extracted， and then emotions are classified into two scenarios： emotional self-imagination and VR-induced. The accuracy of emotion classification， self-evaluation score of emotion and self-evaluation score of vocal music are compared to explore the influence of VR on participants’ emotion mobilization from both subjective and objective aspects. Experimental results show that compared with traditional self-imagination， VR technology can greatly induce the emotions of participants and enhance the singing performance， thus providing a new auxiliary method for vocal singing teaching.

Abstract:In this paper， we mainly apply the breast cancer data from University of Wisconsin System to predict breast cancer using penalized logistic regression. Firstly， the ten indicators related to breast cancer are selected as the predictor variables. Then， logistic regression， the LASSO penalized logistic regression， the L2 penalized logistic regression and the elastic net penalized logistic regression are used as the four classifiers. 75% of the data set is used as the training set to build models. Finally， 25% test set， a confusion matrix and a ROC curve are used to evaluate their prediction accuracy. The results show that the LASSO penalized logistic regression performs best， whose prediction accuracy reaches 97.18%. The prediction performance of the elastic net penalized logistic regression changes with the increase of α， especially when α=0.9， the corresponding prediction accuracy is 97.18%， as good as that of LASSO penalized logistic regression. The L2 penalized logistic regression ranks the third and logistic regression performs the worst in prediction performance. Therefore， for the diagnosis of breast tumors， doctors can apply the LASSO penalized logistic regression and the elastic net penalized logistic regression to improve the diagnostic accuracy.

Abstract:Due to the fuzzy property of image segmentation， this paper proposes a segmentation model for non-uniform gray and high-noise images. The model is based on the fuzzy energy functional which combines with the regional and edge information， and uses the coefficient of variation as the local regional statistics， thus avoiding the interference of noise on the segmentation and extracting the image information well. Regional energy balances the importance of the target and the background， and drives the initial contour toward the target boundary. The edge energy regularizes the pseudo-level set function to maintain the smoothness of the curve evolution. To find the minimum value of the energy functional， the difference between the old and new energy functional is calculated directly so as to update the pseudo level set. The segmentation results of synthetic and real images with high noise， mixed noise and uneven intensity show that the model has a good segmentation effect.

Abstract:Facing unrestricted open vocabularies and diverse scenes， present semantic segmentation methods have the problems of insufficient segmentation， insufficient semantic information extraction and long convergence time. Therefore， this paper proposes a semantic segmentation method that combines U-Net improvement model and superpixel optimization. The U-Net improvement model combines the atrous spatial pyramid pooling （ASPP） and the Xception structure. Firstly， the dilated convolutions （DC） is added to the branch network of the ASPP module to form the serial-parallel structure of the module itself， thus enhancing the image feature extraction capability. And the attention channels are added to the Xception module and a large convolution kernel is used to reconstruct the Xception module， thus reducing the amount of data parameters and increasing the convergence rate. On the basis of the above improvements， the image is then subjected to the super pixel segmentation processing. Finally， conditional random fields are used to impose global constraints on the segmentation results to further optimize the semantic information of pixels. The proposed method is verified on the PASCAL VOC 2012 test set and compared with mainstream networks such as DeepLab V3. Experimental results show that the performance accuracy of the proposed method is increased by 2.4%， which proves the effectiveness of the proposed method in adapting to diverse scenes and dealing with the fine semantic segmentation.

Abstract:This paper proposes a lightweight and efficient flame detection algorithm of videos. The flame detection algorithm is based on the convolutional neural network of deep learning. Considering the motion characteristics of the flame in the continuous video frame， this paper evaluates the flame detection results by the motion object detection and removes the false positive results caused by stationary objects or lights. The motion object detection algorithm based on the Gaussian mixture model is highly efficient. In addition， we collect and label a set of fire detection dataset （FDD）， including 2 487 flame images and 15 fire videos under various scenarios with different flammable materials. In conclusion， the proposed algorithm obtains 98.94% accuracy on FDD test videos.