Abstract:For the problem that spectrum resources is increasingly scare in electromagnetic spectrum space， the radio frequency machine learning （RFML） is purposed to design special machine learning models by introducing domain knowledge. It has the advantages of fast， few sample or even zero sample， interpretability and high performance. The state-of-the-art RFML in wireless communication is analyzed from the five layers， which are physical layer， data link layer， network layer， transmission layer and application layer. Moreover， based on the existing achievements， four RFML frameworks （serial/parallel/coupled/feedback dual-driven framework） are summarized by the interaction relationship of the data-driven model and the knowledge-driven model. Finally， the key challenges and open issues are identified and elaborated to facilitate the RFML research and practical applications.

Abstract:With the increasing scarcity of spectrum resources， the increasing severity of radio regulations， and the increasing fierceness of electromagnetic warfare， the research of the electromagnetic spectrum situational has to transit from spectrum perception to spectrum cognition. This paper reviews the current research status of the theories and methods for spectrum situational cognition from four aspects： The mathematical models and fundamental mechanisms for a multi-layer spectrum situational cognition system， the extraction and fusion of cross-region multi-dimensional electromagnetic features， the efficient completion and prediction for spectrum situation， and the precise inference and intention judgment for spectrum-related behaviors. To address the complex multi-domain multi-dimensional electromagnetic spectrum environment and diversified tasks， it is critical to develop a framework of spectrum situational cognition from “data to semantics” “semantics to behaviors” “a single region to cross regions” “a single layer to multi layers”. This framework builds up foundations， both theoretically and technically， for efficient spectrum sharing in space-terrestrial integrated information networks， efficient radio regulations， and advantages in the electromagnetic spectrum warfare.

Abstract:To improve the localization accuracy of the subspace data fusion （SDF） applied in distributed multi-array under the influence of unknown mutual coupling， we propose a reduced mutual coupling dimension subspace data fusion （RMCD-SDF） approach in this paper. Firstly， we introduce the mutual coupling error model into the SDF approach to make it adapt to the scenario where the antenna array is affected by the unknown mutual coupling error. Furthermore， in order to reduce the ultra-high computational complexity caused by searching all unknown parameters simultaneously， we introduce the reduced-dimension idea and construct the spectral function of RMCD-SDF. Simulation results show that the RMCD-SDF approach has advantageous localization performance when arrays are affected by unknown mutual coupling. The RMCD-SDF approach has similar computational complexity but higher localization accuracy than existing algorithms. When the signal-to-noise ratio （SNR） is 10 dB， the root means square error of the proposed approach is 8.67 dB lower than the classical SDF algorithm.

Abstract:With the improvement of modern radar signal quality and resolution， radar systems can capture more detail on targets. Objects imaged by inverse synthetic aperture radar （ISAR） may have components that rotate at high speeds relative to the whole， such as propeller blades of aircraft. The micro-Doppler effects created by these components may seriously interfere with the imaging results. Therefore， this paper proposes a method to eliminate the micro-Doppler effect in ISAR based on the time recursive iterative adaptive （TRIAA） and compressive sensing. The method uses TRIAA technology to analyze the time-frequency characteristics of signals， removes the micro-Doppler effect from the time-frequency map， and uses the compressive sensing sparse reconstruction technology to accurately restore the effective signal. Experimental results of simulation and measured data verify the effectiveness of the proposed method.

Abstract:Due to the rapid development of 5G cellular network， its coverage will be increasingly better， thus cellular network localization is a very promising technical object for research. This paper is inspired by the fingerprint localization method in wireless localization. Under the premise that the time cost of data collection is similar， a high-speed， high-precision and low-occupancy localization method is accomplished by using the emerging deep learning technology instead of the heavy fingerprint library application and distance calculation in the localization process of fingerprint localization. In this method， a convolutional neural network is built， and the training set is constructed by selecting the appropriate input data format based on the amount of features， such as received signal intensity indication， phase and direction of arrival， of the 5G antenna signal. The trained convolutional neural network can replace the huge fingerprint library in fingerprint localization， which is very beneficial to achieve localization directly in 5G mobile devices. In addition， although convolutional neural networks consume a lot of time during the training process， the classification and localization performed after the training is completed with high speed， which can guarantee the real-time implementation of localization. The trained convolutional neural network in this paper takes up less than 0.5 MB of space for weights and biases， and is able to achieve a localization accuracy rate of 95% and an average localization accuracy of 0.1 m in the real-world environment.

Abstract:Waveform design for detection and jamming integration is one of the key technologies to improve radar radio frequency stealth performance. This paper proposes two integrated waveform composite coding schemes， and Wigner distribution is utilized to analyze the jamming performance of the designed waveform. Firstly， the design scheme and signal model of frequency coding， phase coding and two composite coding waveforms are given based on chaos theory. Then， the detection performance is described by ambiguity function and the jamming performance is described by information distance based on Wigner distribution. Simulation results show that the designed integrated waveform can effectively improve the detection and jamming performances， and there exists a tradeoff relationship between the above-mentioned performances.

Abstract:In order to improve the estimation accuracy and convergence speed of millimeter-wave multiple-input multiple-output （MIMO） cascaded channel assisted by intelligent reflective surface （IRS）， the conventional bilinear alternating least squares （BALS） algorithm is improved to ω-BALS algorithm with relaxation factor and regularized T-BALS， to speeds up the convergence speed and stability based on parallel factor （PARAFAC） decomposition. When one of the numbers of array antennas on the base station， IRS element or user side is large， an improved （Singular value decomposition，svd）-BALS algorithm is proposed. The algorithm reconstructs the mode-n matrices by compressing it into a low-dimensional core tensor via singular value decomposition. Simulation results show that the normalized mean squared error performance of the algorithm is improved and the convergence speed is accelerated.

Abstract:In wireless channel modeling and simulation， it is of great significance to realize a high-efficiency and high-accuracy wireless channel prediction method. Aiming at this request， a wireless channel prediction method based on multi-population genetic algorithm-back propagation （MPGA-BP） neural network is proposed. This method optimizes the structure parameters of BP neural network by improving the genetic algorithm， thereby improving the problem of poor prediction accuracy of the BP neural network and greatly improving the prediction performance of the BP neural network. In this paper， the theoretical value of ray tracing algorithm is combined with BP neural network to realize a more efficient wireless channel prediction method. By comparing the prediction errors of the genetic algorithm （GA）-BP neural network model and the MPGA-BP neural network model， it is found that the prediction results of the MPGA-BP neural network model are better than the GA-BP neural network model， which proves that the proposed wireless channel prediction method has good accuracy. Therefore， the wireless channel prediction can be performed more efficiently.

Abstract:Since a single feature is not enough to comprehensively represent the subtle feature differences and thus limit the recognition rate for specific identification of communication emitter， a method of specific identification of communication emitter based on feature fusion is proposed. Firstly， the short-time Fourier transform and bispectrum transform are applied to the original signal to extract time-frequency features and bispectrum features. Secondly， the wavelet fusion technology is integrated to carry out feature fusion. Finally， the residual neural network is used to mine the hidden deep features of the signal to complete classification and recognition. Experimental results show that compared with the single feature method， the recognition effect of the short wave communication signal transmitted by analog signal source after feature fusion has higher recognition accuracy， and it has better recognition effect under the condition of low signal-to-noise ratio（SNR）.

Abstract:The optimization of resource allocation in wireless communication networks can be described as a mixed integer nonlinear programming （MINLP） problem. It is essentially a non-convex NP hard problem. In order to reduce the computational complexity and ensure the optimal performance of the allocation scheme， a binary whale optimization algorithm （WOA） is proposed to allocate wireless resources. Based on the original WOA position update is carried out based on the switch between values 1 and 0. The current position changes are determined by the probability calculated by the humpback spiral movement. Meanwhile， different transfer functions are used to map the continuous search space to discrete actions， and the penalty method and the optimization constraint processing are introduced. Two cases of resource allocation in wireless networks are analyzed in the experiment： The power allocation problem with maximum confidentiality and the mobile edge computing migration. The results show that the proposed method has excellent system performance and obtains similar effects to other methods， but its convergence speed is faster and its complexity is lower.

Abstract:Unmanned aerial vehicles （UAVs）assisted simultaneous wireless information and power transfer （SWIPT） can be used to improve energy efficiency of Internet of Things （IoT）. It can replenish energy for ground devices in IoT to complete the task of information-receiving uninterruptedly. In the face of the limited energy of UAVs and the demand for improving energy efficiency， this paper studies an energy-efficient UAV-ground communication optimization problem. We jointly optimize UAV transmit power and power splitting ratio， and design UAV trajectory and ground devices wake-up time allocation， in which UAV propulsion energy consumption and energy demand of ground devices are considered comprehensively. The formulated energy-efficiency maximization problem is a non-convex， fractional and mixed integer programming problem. To solve this problem， this paper proposes an alternate iterative optimization algorithm based on the successive convex approximation （SCA） and the classical Dinkelbach method. Finally， simulation results verify the effectiveness and superiority of the proposed algorithm.

Abstract:Targeting the low efficiency and privacy leakage of intra-city delivery of confidential documents， a intra-city path planning algorithm based on block-chain is proposed. It adaptively generates the shortest path to protect location privacy in real time. With the consensus mechanism and smart contract algorithm of block-chain， the distributed site is selected by route planning with homomorphic encryption. The vehicle can encrypt and decrypt the next site information by using its own context attribute， and be equipped with anti-impersonation. This algorithm also solves the problem of mutual distrust among vehicles， sites and deliveries. Finally， through experiments， the impact of the homomorphic calculation results of smart contracts， the number of different contextual attributes， and the number of different sites on the calculation cost of path planning is analyzed. The results show that the algorithm of the intracity delivery system has the capabilities of confidentiality， integrity and anti-tampering and can ensure high-delivery efficiency.

Abstract:In the cyber-physical system （CPS）， the traditional multi-source heterogeneous data integration model is difficult to map the conceptual layer between heterogeneous systems through middleware， which has the problems of low transmission performance and difficult system expansion. Due to the challenges above， a CPS-oriented heterogeneous data integration model is proposed. The data object model is designed to realize the high-level concept mapping between physical and simulation systems. The monitoring and control metadata are defined and the incremental or full field updates for different data types are used to reduce network load. A communication model of system is designed based on the Protobuf protocol to improve the system expansion capability. Based on the data interaction model and high level architecture （HLA）/data distribution service （DDS） system middleware， a CPS prototype system is implemented， which verifies the usability of the model and compares the compression performance of the message.

Abstract:In the radio frequency identification（RFID）communication system， when multiple tags conflict， the conflicting signals can be separated and then decoded to improve the communication efficiency， and the signal separation usually depends on clustering. However， the traditional methods cannot consider the time complexity and clustering accuracy. In this paper， a clustering method of maximum posterior probability estimation is proposed. The peak value is quickly found out by the Monte Carlo method that is used as the clustering center to complete signal separation. In the experiment， the simulation data and the measured data of software radio are used to test the proposed algorithm， and the results show that the proposed algorithm has a higher clustering accuracy and a lower time complexity under high signal noise ratio（SNR）. It is embedded in dynamic frame ALOHA system， and the throughput can reach 0.55， higher than that of the pure dynamic frame ALOHA.

Abstract:Indoor positioning in dynamic environment is easy to be interfered by human’s random actions and obstacles. The time-varying signal strength and the instability of data acquisition have a great impact on the positioning algorithm. To solve the problem， this paper proposes an online sequential extreme learning machine algorithm based on particle swarm optimization named PSO-OS-ELM. The algorithm inherits the characteristics of the on-line sequential extreme learning machine （OS-ELM） algorithm， such as low data acquisition cost， fast adaptation to environmental changes， fast convergence speed and high positioning accuracy. At the same time， the particle swarm optimization （PSO） is used to solve the singular value problem and instability problem in the OS-ELM algorithm. The PSO-OS-ELM algorithm， the OS-ELM algorithm and the weighted K-nearest neighbor（WKNN） algorithm are compared. The experimental results show that in the dynamic indoor environment， in terms of algorithm stability， the PSO-OS-ELM algorithm has smaller and stable positioning error， and is better than other algorithms. Compared with other algorithms， the average positioning error is reduced by about 15%. Compared with the traditional localization algorithm， the WKNN algorithm reduces the time consumption by about 55%.

Abstract:In the speech mode， the OpenSMILE toolbox is used to extract low-level acoustic features from the speech signal. Transformer Encoder is richer to excavate deep features from low level acoustic features and fuses them so as to obtain more useful emotional representation. In the text mode， considering the association between pause and emotion， the speech and text are aligned to obtain the pause information and the pause information is added to the transcript text by pause encoding. The utterance-level lexical features are obtained by the improved DC-BERT model. Then，acoustic features and lexical features are fused and the bi-directional long short-term memory based on attention neural network （BiLSTM-ATT） is used for emotion classification. Finally， this paper compares the effects of three different attention mechanisms integrated into BiLSTM on emotion recognition （local attention， self-attention and multi-headed attention），and local attention is found to be the most effective. In the experiments on IEMOCAP dataset， the method proposed in this paper achieves 78.7% in weighted accuracy for four emotion categories， which is better than the baseline system.

Abstract:The stability of sparse reconstruction algorithms in compressed sensing can be raised by reducing the mutual coherence value of the equivalent dictionary， i.e.， the product of the measurement matrix and the sparsifying dictionary. While， the existing optimal design methods do not consider how to improve the efficiency of signal reconstruction when reducing the mutual coherence value. To overcome the problem， a constrained smooth optimization problem about measurement matrix is proposed， in which the first constraint requires the mutual coherence value of the equivalent dictionary to be as small as possible， and the second one uses the L₁ norm to facilitate the sparsity of the measurement matrix. Then， a convergent alternating projection algorithm is used to solve it. The simulation results on natural images show that compared with the equivalent dictionaries obtained by several existing optimal design methods， the proposed method greatly raises the sparsity of measurement matrix and improves the signal recovery accuracy.

Abstract:In order to improve the visual perception ability of unmanned surface vehicle（USV） in harsh navigation scene， a polarization image fusion method of visible light for water navigation scene is proposed based on hue， saturation， value（HSV） color space. The fusion rules for different regions are formulated in accordance with the polarization characteristics of the water navigation scene. And based on the HSV color space， the color information of the original scene is fused， which is tested that realizes the semantic segmentation of the harsh navigation scene image. The most striking result is that the pixel accuracy（PA） value in the flare scene is 0.768 2. And the experimental results indicate that the proposed method can enhance image contrast， highlight edge contour information， and stably obtain feature information with strong contrast as well as better target characteristics in harsh navigation scene， which improves the USV’s performance in harsh navigation scene to a certain extent.

Abstract:To prevent complication caused by moyamoya disease from threatening patients’ lives， timely and effective diagnosis of moyamoya disease is needed. An improved Faster RCNN algorithm for moyamoya disease detection is presented. Firstly， the digital subtraction angiography （DSA） image of internal carotid artery is extracted and enhanced. The ratio of training set， verification set and test set is 6∶2∶2. ResNet101 network is used as the feature extraction network to avoid blurring or loss of vascular features in the process of convolution and pooling. Combined with region proposal network （RPN）， the location of moyamoya disease focus is located. Then replace ROI pooling in Faster RCNN model with ROI Align for feature mapping to avoid the error impact caused by quantization. The average precision （AP） is used as the evaluation index of the detection performance of the algorithm. The AP of normal samples and moyamoya disease samples are 99.23% and 89.39%， respectively. Experimental results show that the proposed method can realize the rapid and effective detection of moyamoya disease. It can accurately detect the location of moyamoya disease lesions in the complex vascular network， and provide some technical support for the auxiliary diagnosis of moyamoya disease.

Abstract:Brain-controlled technology based on brain-computer interface （BCI） has developed rapidly and made great progress. However， the existing research mostly adopts the single-person brain-controlled manner， which has the problems of poor execution efficiency and low degree of controllability， making it difficult to meet the needs of complex manipulation tasks. To address this problem， this study adopts a time-frequency-phase hybrid encoding method， and designs a collaborative strategy. A two-person collaborative brain-controlled robotic arm system with 108 instructions has been developed， enabling two people to write Chinese characters simultaneously one stroke by one stroke. The average online accuracy of the eight subjects is 87.92%， and the corresponding average online information-transfer rate （ITR） is 66.00 b/min. This system extends the BCI information interaction manner， and preliminarily verifies the feasibility and effectiveness of collaborative BCI manipulation of robotic arm. It provides technical support for collaborative BCI.