The air-space-ground integrated network has been regarded as one of the enable technologies in the future to provide consistent communication experience all over the world.Based on the integration of the air,space and ground network,the area with no communication coverage can be eliminated and interconnection between remote areas can be strengthened.However,the current air-space-ground integrated network has only achieved the preliminary application in some special scenarios,and there are still many challenges to achieve the goal of eliminating the blind area and strengthening the connection.This paper first outlines the structure of the future air-space-ground integrated network.Then,four major challenges encountered in the network integration process are analyzed from the perspectives of hardware,protocols,node deployment and service assurance,including the difficulty of network function evolution raised by dedicated hardware,the low performance of current satellite communication protocol,the deployment of the flying base station in a broad 3D area and the differentiate quality guarantee of various services.Finally,four typical solutions to overcoming the above challenges are summarized to provide reference for the future air-space-ground integrated network,including light-weighted network function virtualization in the satellite,5G based satellite communication system,deployment of the flying base station based on horizontal and vertical decoupling and end-to-end integrated network slicing.

Co-channel interference (CCI) is a key factor that impedes the improvement of a wireless communication system’s performance.Existing signal processing based multi-user transmission schemes avoid CCI by adjusting the interfering signals.However,they may degrade the performance of data transmissions carried in the adjusted signals.By exploiting interactions among wireless signals,a virtual signal decomposition based multiple access (VSDMA) method is proposed.By employing a coordinate transmitter (Tx) and letting it send a coordinate signal to interact with signal components originating from multiple interfering Txs at the common receiver (Rx),the Rx can recover multiple orthogonal desired signals without interference.The proposed method can not only avoid the degradation of desired transmission’s quality incurred by adjusting the transmitted signal at the Tx but also circumvent the problem of the desired signal’s power loss at the common Rx while it suppresses interference.Simulation results have shown that compared to existing schemes,given a signal-to-noise ratio greater than 0dB,our method can improve the spectral efficiency (SE) and power normalized SE of the multi-user communication system by over 10%.

To address the non-convergence problem of the traditional Gauss-Newton iterative method in time difference of arrival (TDOA) and frequency difference of arrival (FDOA) location due to the inaccurate iterative initial value,a Gauss-Newton iterative algorithm based on the constrained weighted least square (CWLS) is proposed.First,the nonlinear positioning equation in the positioning problem is transformed into a set of pseudo-linear equations about the target position and velocity.Initial values of the target position and velocity are estimated step by step.In order to realize the accurate estimation of the initial value,the equality constraint relationship between the target position and the auxiliary variable are relaxed to the second-order cone programming (SOCP) condition.The stochastic robust least square (SRLS) is introduced to construct a new linear relation.When the weighted least square solution does not meet the SOCP condition,the semi-definite programming (SDP) is used to solve the estimated solution of the target position.The target velocity is solved by the obtained target position.After obtaining the initial values of the target parameters,the Gaussian Newton iterative equations for the target position and velocity in the TDOA-FDOA localization system are established.Target parameters are solved by the Gaussian Newton iterative process,which does not require the introduction of auxiliary parameters and can directly obtain the target parameters.Simulation experiments show that the proposed algorithm has a good localization effect on both near-field and far-field targets,and that its robustness and high localization accuracy are better than those of the existing classical two-stage weighted algorithm.At the same time,simulation results show the necessity of optimizing the initial values when the Newton iterative equations are used.

In medium access control (MAC) method design,the effective management of co-channel interference (CCI) among multiple concurrent transmissions is critical to enhancing a system’s spectrum efficiency (SE).The main idea of existing interference management (IM) methods is to make multiple signals orthogonal to each other during their transmissions.However,in the multi-user multiple input multiple output (MU-MIMO) system,the numbers of transmitters and receivers and their antenna settings highly restrict the number of orthogonal concurrent data transmissions that the system can support.To remedy such deficiency,this paper proposes a relaxed orthogonal precoder based multiple access (ROPMA) method for MU-MIMO downlink transmission.By introducing an orthogonal amplitude tuning function to adjust the precoder’s amplitude in a symbol duration,and then using the adjusted precoder for signal processing and transmission,the requirement of strict orthogonality of multi-user signals in the spatial domain can be relaxed to the orthogonality in the sense of correlation.Accordingly,the receiver employs correlation based desired signal detection;by exploiting the orthogonality of amplitude tuning functions,the influence of the CCI can be eliminated at the decision time.Our simulation results have shown that the proposed method can significantly increase the number of users that the system can accommodate and improve the system’s SE without consuming the extra spectrum resource.

In the processing of vehicle-target detection in synthetic aperture radar (SAR) images,the contours of vehicles not only provide their position but also represent their condition,which is a key to SAR image understanding.But the multiplicative speckle noise in SAR images interferes with the border positioning of vehicles,resulting in difficulties for vehicle-target detection.To solve this problem,the present paper proposes an attention-mechanism-based neural network for pixel level vehicle detection,which consists of a target filtering module,a target locating module and a contour refining module.The target filtering module contains a lightweight feature extraction network with a channel-attention and self-attention mechanism to enhance feature expression.This module can decrease the effect of the speckle on features to select images containing the target quickly and precisely,and provide the output stable location heat map for the next module.The target locating module uses the foreground-background cross-attention mechanism to refine the coarse-scale features in accordance with the location heat map and refine the target location.Furthermore,the module adopts the fine-scale information to improve the details of the target contour.The contour refining module eliminates the contour uncertain points caused by upsampling and speckle noise to obtain accurate contour pixel confidence.For testing this network,a target image dataset and a large-scene image dataset are built with the pixel-level vehicle annotation of the dataset labeled by ourselves.The result of testing indicates that the network has a good pixel-level detection performance and can detect vehicle targets in large SAR images rapidly and accurately.

Existing weakly-supervised salient object detection methods often suffer from problems such as false positive,low recall rate,and unclear edges.To address the above issues,a weakly-supervised salient object detection with the multi-scale progressive network is proposed,which divides salient object detection into three sub-tasks:object localization,saliency region improvement and edge refinement.First,the input image is sampled into three images of different scales,which are respectively fed into the three stages of the multi-scale progressive network for learning.Second,in order to better locate the salient objects,a nested shift multi-layer perceptron is proposed in the object localization stage,which can balance the global feature and local feature extraction ability of the network.Finally,according to the characteristic that the structure of saliency maps is not affected by scale changes,a multi-scale self-supervision module and an object consistency loss are designed to build a self-supervision mechanism,so that the network can output a saliency map with complete regions and sharp edges.The proposed method is tested on five datasets,and outperforms the recent weakly-supervised methods in both quantitative and qualitative comparisons,and can reach 89% of the performance of the related fully-supervised methods on the F-measure index.Experimental results show that the proposed algorithm can generate saliency maps with complete saliency regions and sharp edges,and has good robustness.

In practical underwater wireless optical communication (UWOC) system,turbulence and receiver noise are two important factors that limit the system performance.In this paper,with shot and thermal noise,the average bit error rate (ABER) performance of orthogonal frequency division multiplexing (OFDM) assisted UWOC system based on the avalanche photodiodes (APD) over the weak oceanic turbulence condition has been investigated.To effectively characterize the impacts of absorption,scattering and misalignment as well as underwater optical turbulence,a composite fading model has been adopted in this work where beam spread function (BSF) is utilized.After that,the analytical expressions of ABER in shot-noise-limited and thermal-noise-limited conditions are derived,respectively.Under shot-noise-limited condition,in this paper we analyze the ABER performance affected by various parameters,specifically,receiver noise temperature,average optical signal intensity and APD gain.Similarly,under thermal-noise-limited condition with partial coherent beams (PCB) being adopted,the specific impact of underwater optical turbulence on the reliability of the considered UWOC system is also illustrated.Numerical results reveal that,for the internal factors of the photodetector,the receiver temperature and average optical signal intensity will affect the ABER performance of the system.Furthermore,choosing an appropriate APD gain can obtain a better ABER performance of the UWOC system.Moreover,PCB and the underwater turbulence will affect the ABER performance.

An unmanned Aerial Vehicle(UAV) relay broadcast communication system can provide a high quality service and multiple business assistances for users.UAV trajectory optimization is one of the critical technical problems of the UAV relay communication system.The trajectory optimization methods are investigated to improve the link transmission performance of the UAV relay broadcast communication system with the decode-and-forward protocol.First,the UAV relay broadcast communication system model is established.The calculation formula for the interruption probability of a user link and that for the average interruption probability of the communication system are given.After that,a trajectory optimization method based on the criterion of minimizing the maximum outage probability for all users is proposed,and the mathematical expression for the accurate ergodic capacity of the UAV relay broadcast communication system is theoretically derived.Finally,the influence of system characteristic parameters such as service area radius,maximum turning angle,path loss factor and transmitting power on the optimal track,system interruption performance and system capacity performance of the relay UAV is verified by computer simulation.Meanwhile,the system interruption performance and system capacity performance of two relay forwarding protocols are compared under the same optimization criteria.Simulation results show that the outage performance and capacity performance of the decode-and-forward protocol are significantly improved compared with the amplify-and-forward protocol,which verifies the effectiveness of the proposed trajectory optimization criterion.

In the process of high-speed railway train operation,in order to maintain uninterrupted communication,the train needs to constantly carry out handover with the base station.As a key technology of LTE-R communication,handover is crucial to ensuring traffic safety.Aiming at the low success rate of handover in the next generation high-speed railway LTE-R wireless communication system due to the fixed hysteresis threshold parameters,a high-speed railway handover algorithm based on the LSTM recurrent neural network is proposed.First,by using the memory characteristics of the LSTM neural network and the temporal and spatial correlation characteristics of high-speed railway handover overlapping area signals,a deep learning network for dynamic prediction of handover hysteresis threshold parameters based on the LSTM recurrent neural network is constructed.Second,through the proposed LSTM deep learning model,the handover hysteresis parameters are trained offline and predicted online to obtain the handover threshold value at the future time,which realizes the adaptive prediction of handover hysteresis parameters during high-speed train driving,and overcomes the disadvantage of fixed hysteresis threshold parameters.Finally,simulation results show that the proposed method can effectively improve the handover success rate and reduce the impact of the ping-pong handover rate compared with the traditional A3 algorithm and other comparison algorithms.Research results provide a certain theoretical reference for high-speed railway traffic safety and LTE-R evolution.

In order to improve the LO leakage caused by the DC offset of the analog baseband signal in the direct conversion transmitter,and the drift of the low-pass filter due to process deviation,temperature change,and chip aging,a method of design and implementation of an automatic frequency-calibrated low-pass filter with DC offset cancellation is proposed.In the SPI(Serial Peripheral Interface) control mode,the filter can configure the cut-off frequency from 324 kHz to 648 kHz through the adjustable capacitor array.In the adaptive tuning mode,a digital-analog hybrid calibration structure based on the latched comparator is proposed.By calibrating the RC time constant of the chip,the cut-off frequency is 373.8~393.3 kHz under different PVT conditions,and the tuning accuracy relative to the nominal value of 383 kHz is -2.4%~2.7%.The chip uses 0.18 CMOS technology for tape-out verification,and the tuning circuit area is 0.045 mm²,which is only 2.2% of the main filter area.Under the 1.8V supply voltage,the power consumption of the whole filter is 6.08 mW,and the input reference noise is 38.49 nV/(Hz)^1/2.

A position observation strategy considering system delay compensation is presented to avoid the position estimation error caused by analog and digital delay of the permanent magnet synchronous motor drive.First,a typical sliding-mode observer is established and the accessibility of sliding mode motion is analyzed.Second,the influence of analog circuits on the stator current is studied,and the transfer functions of analog circuits are derived.The total delay phase caused by the sampling circuit on the stator current is calculated.Third,the current error from digital implementation of the sliding-mode observer is analyzed.Fourth,a direct signal compensation method used for eliminating the influences of the current error is presented.The proposed strategy lowers the detection error of the sliding-mode observer and the transient process time in closed-loop control.Finally,the proposed strategy is experimentally verified to confirm its merits on system delay effect elimination and system control performance improvement.

An X/Ku-band bi-directional True-time delay is presented in 0.25 μm GaAs pHEMT E/D technology.The bi-directional operation is realized by adding two-way selecting switches to the new trombone structure,and it has the advantage of low insertion loss variation among different delay states.Fourth-order and second-order inductive coupled all-pass networks are adopted to form two delay lines.The delay differences are controlled by selecting the delay path through the bi-directional active switches.The true-time delay operates with the bandwidth of 6~18 GHz,and it realizes a 3-bit delay with a minimal delay step of 15 ps and a maximal delay of 106 ps.Simulation results show insertion loss of 8.1~15 dB and the loss variation with delay is ±2 dB.The group delay Root-Mean-Square error less than 10 ps and return loss more than 15 dB are implemented with the chip size of 1.91 mm².The direct current consumption is 110 mW and the input P1dB is more than 7 dBm.

Network diagnosis is one of the most exciting topics in graph theory and network science,which affects the reliability and security of multiprocessor systems.With the rapid growth of the scale of the multiprocessor system,the applicability of the global fault diagnosis mode of the system is reduced.Accordingly,the local fault diagnosis benefits from the lower requirements for the network topology,which can process the network in blocks,greatly improving the diagnosis efficiency,having a stronger applicability,and becoming a new research direction.Under the latest Symmetric PMC(SPMC) model,this paper studies the relevant properties of network node diagnosis(local diagnosis),proposes a new topology structure(extended tree structure),obtains the judgment conditions for diagnosable nodes,the relationship between nodes diagnosis and system diagnosis,gives the judgment theorem whether nodes are poor on the extended tree structure,and gives out the detailed proof.According to this theorem,one novel network local fault diagnosis algorithm ST₂_B-FDA with the expanded tree structure is proposed.To validate the effectiveness of this algorithm,this paper applies the Hypercube network for simulation.The time complexity of the algorithm is O(NlogN),which is much lower than that of some traditional fault diagnosis algorithms.This algorithm can effectively reduce the diagnosis cost and greatly improve the diagnosis efficiency.In addition,the proposed algorithm is simple in principle,easy to implement and apply,and can also be used as one of the diagnosis methods for large-scale regular network systems.

In the autonomous driving scenario,aiming at the problem of limited memory and insufficient computing power when the semantic segmentation model is deployed in vehicle hardware devices,it is necessary to design a semantic segmentation model that can balance efficiency and accuracy.In this paper,a single-branch network structure is used to design a lightweight multi-scale bidirectional attention network.To achieve efficient feature extraction,a lightweight convolutional unit is designed to form the feature extraction backbone of the network.In order to better locate and segment objects with large scale differences in road scenes,a multi-scale bidirectional attention module is proposed which has a global multi-scale receptive field and encodes channel attention in one direction while preserving spatial location information in the other.Based on this attention module,a skip attention connection module and a feature attention fusion module are designed,so that the output features have both detailed information and semantic information.On the Cityscapes dataset,the model in this paper achieves an MIoU(Mean Intersection over Union) of 71.86% with a parameter size of 0.9M,and achieves an inference speed of 88FPS(Frames Per Second) under a single RTX2080Ti GPU.The test results on public datasets show that the model achieves a high segmentation accuracy and is suitable for deployment and application under in-vehicle hardware,which is of certain practical value.

The low light environment affects the image capture equipment,resulting in low contrast,low brightness,and difficulty in distinguishing objects.In order to improve image quality,a method for enhancement of the multi-scale low-light image by combining an attention guidance is proposed.First,a dense residual network is constructed as a multi-scale feature extractor to extract feature maps at different scales in low light images,and the extracted feature maps are fused by using a modified RefineNet,which makes full use of the feature information in the image.Meanwhile,an interpretable attention mechanism is designed to generate an attention graph based on the results of edge detection.Then by combining a loss function the network is guided through training.The purpose is to enhance edge detail information hidden in the dark without increasing the network’s inference burden.Finally,experiments are completed on synthetic images and SID(See-in-the-Dark) datasets,with the results showing that the proposed method can effectively improve brightness and contrast,restore image edge details as well as improve subjective visual effects.Compared to the contrast algorithm,the PSNR and SSIM are improved by at least 0.79dB and 0.119 on average,respectively.

Target sentiment analysis aims to analyze the sentiment tendency corresponding to different targets in the review text.At present,graph neural network based methods use the dependency syntactic tree to incorporate dependency syntactic relations.On the one hand,these methods mostly ignore the fact that dependency relations lack distinction.On the other hand,without considering the dependency relations provided by the dependency syntactic tree,there is a lack of relations between target and sentiment words.Therefore,a dual graph attention network(DGAT) model is proposed.First,the model uses a bidirectional long short-term memory network to obtain word node representation with semantic information,and then constructs a syntactic graph attention network based on the word node representation according to the dependency syntactic tree,so as to distinguish the importance of dependency syntactic relations,more effectively establish the relation between target and sentiment words,and obtain a more accurate representation of target sentiment features.At the same time,according to the undirected complete graph of sentences,a global graph attention network is used to mine lacking relations between target and sentiment words,so as to further improve the performance of the model.Experimental results show that compared with existing models,the DGAT model has a better accuracy and macro-average F1 value on different datasets.

Non-intrusive electrical appliance identification is the key technology to realize customer-side intelligent sensing in the ubiquitous power Internet of Things.Aiming at the problem that the calculation complexity of the existing non-intrusive electrical appliance identification system is too high and is not conducive to industrialization,a low-cost,low-power consumption,high-precision non-intrusive electrical appliance identification system and the algorithm based on the Long Short-Term Memory(LSTM) network are proposed,and they can be applied to an embedded microcontroller.First,the current on the bus is collected synchronously.Second,the proposed multi-parameter detection method is used to judge the switching event of the electrical appliance.Third,the LSTM network is used to process the data before and after the switching time point,and the type of electrical appliance which is switching is obtained.Finally,the current types and quantities of electrical appliances are judged by the cumulative sum.Simulation and measurement results show that only a small amount of data training for a single electric appliance is needed,and that the recognition accuracy of combined electric appliances can be up to 99.6% in the proposed embedded microcontroller system,in which the power consumption is less than 1.5 watts.The proposed algorithm can be further applied to the statistics of the switching time point,service time and total power consumption of each electrical appliance,which provides refined user power consumption information for the smart grid.and provides an important reference for the energy management and optimization in the smart grid.

With the continuous advancement of oil and gas exploration in China,seismic exploration is facing great challenges.Affected by the complex exploration environment,acquisition method,detector sensitivity and other factors,the obtained seismic data are often mixed with a large amount of random noise,resulting in the decrease in fidelity,signal-to-noise ratio (SNR) and resolution of subsequent seismic data processing,and the accuracy and reliability of geological interpretation are ultimately affected.In order to break through the limitations of traditional seismic data processing problems,a spatially adaptive directional total variation (SADTV) regularization model for random noise suppression of seismic data is proposed.First,aiming at the problem that the seismic reflection events have the directivity of spatial variation and the poor noise resistance of dip angle calculation,a point by point estimation formula for the spatially varying dip angle based on the gradient structure tensor (GST) is proposed to obtain the direction information on events;Then,the denoising model of SADTV seismic data is established,and the Majorization-Minimization (MM) algorithm for solving the model is derived.Finally,the parameter selection method of the model is discussed,and the denoising results of synthetic and real seismic data are compared with those by similar methods.Experimental results show that the proposed model can not only improve the vertical resolution of the seismic profile and the lateral continuity of the seismic event,but also retain more geological feature information while improving the signal-to-noise ratio.

With the development of urbanization,traffic prediction plays an important role in the application of traffic planning and urban management.However,in the task of traffic prediction,it is still a great challenge to capture the highly nonlinear and complex spatio-temporal dependencies of traffic data.In order to effectively capture the time dynamics and global spatial correlation of traffic data and satisfy both long-term and short-term prediction tasks,an attention based spatial-temporal graph neural network for traffic prediction is designed.First,the attention mechanism is introduced to adjust the importance of adjacent roads and non-adjacent roads,which is beneficial to integrating global spatial information.Then,the spatial-temporal correlations are captured by graph convolutional networks and gated linear units with extended causal convolution.Experimental results on two real data sets PeMSD7(M) and PEMS-BAY show that the network model can improve the accuracy of both long and short-term traffic prediction.

Focusing on the simultaneous authentication of drugs and instructions in the medical system,a rapid dual-tag authentication scheme(CP-LRDP) is proposed,which introduces a cloud server and PUF to ensure the scalability of the RFID system and the unclonability of tags.Aiming at the problem of sequential dual-tag authentication with a low efficiency in a traditional RFID system,a dual-tag response merging process is proposed.For the system error authentication problem caused by the PUF,the optimal authentication threshold of the PUF response is calculated to reduce the authentication error rate of the system.To solve the untrusted problem of the cloud server,three ultra-lightweight bitstream functions are proposed to implement two encryption mechanisms for protecting the forward channel from the threat of cloud server privacy leakage.Security analysis shows that the CP-LRDP not only satisfies the tag anonymity and untraceability,but also can effectively resist cloning attacks,desynchronization attacks,replay attacks and other malicious attacks.In addition,BAN logic analysis and the AVISPA tool are used to further verify the security of the protocol.Compared with recent authentication protocols,the CP-LRDP with the shortest server search time not only meets various security properties,but also realizes achieving rapid dual-tag authentication with resource costs similar to those of single-tag,which is suitable for resource-constrained large-scale dual-tag authentication scenarios.

key-nets,as the first optical homomorphic encryption scheme,is used toprotect the privacy of images used for machine learning.However,in the case of the vision sensor being obtained illegally,the author obtained the key used to encrypt the image in the key-nets scheme by solving the system of linear equations.In view of the security risks in this scheme and the difficulty of machine learning model training,this paper proposes a homomorphic encryption scheme that can use different generalized random matrices for each encryption without changing the original convolutional network structure,and further use the Diffie-Hellman key exchange protocol,which improves the security of the encryption key-nets and also improves the security of the convolutional network matching the vision sensor.Through the analysis of the feasibility of the scheme,privacy parameters,forward security,backward security,etc.,it is proved that the improved scheme can still protect the image information even if the attacker illegally obtains the visual sensor.

The chameleon hash algorithm that can realize data editing provides a content controllable method for blockchain.However,once a participant obtains permission to edit the data,he can rewrite anything without being punished for his malicious behavior.At present,most redactable blockchain schemes can only punish malicious users without considering the malicious tampering of the blockchain by editors,which cannot realize the fairness of both users and editors at the same time.A fair and redactable blockchain scheme that supports malicious punishment is proposed,which ensures the fairness of the redactable blockchain scheme by effectively restricting the rights of editors and punishing their malicious behaviors.In the proposed scheme,the chameleon hash with a short-term trapdoor is used,so that editors and users must cooperate to complete the editing of the blockchain,which effectively restricts the rights of editors.Based on the secret sharing and signature schemes,a punishment mechanism for malicious editors is proposed to punish the malicious tampering behavior of editors and resist the malicious reports of users.Theoretical and experimental analysis shows that the proposed scheme not only realizes the collision-resistance and semantic security of blockchain data,as well as the fairness of users and editors in the data editing process,but also has a lower computational cost than existing schemes.In addition,the advantages of the proposed scheme become more obvious with the increase in the number of editing times,so it is of more practical application value.

Non-Orthogonal Multiple Access (NOMA)-Intelligent Reflection Surface (IRS) systems can improve user access capability through joint "transmit-reflect" beamforming between multiple antenna transmitters and IRS.For the user fairness problem of multi-user NOMA-IRS systems,the minimum received signal-to-Interference plus Noise Ratio (SINR) of users should be maximized in a practical scenario with restricted RF links,so as to guarantee the communication quality of each user without any difference.To this end,a maximum-minimum SINR fractional planning problem is constructed by clustering users according to the number of RF links,and the transmit beam vector of this problem is highly coupled with the reflected array element matrix at SINR.Therefore,a semi-definite relaxation and arithmetic-geometric mean-based algorithm is proposed to maximize the minimum channel capacity in each cluster by alternately optimizing the transmit beam vector and the reflected array element matrix in each iteration.Furthermore,a dichotomous search algorithm is used to solve the intra-cluster power allocation problem to enhance the user minimum SINR.Simulation results show that,compared with the zero-forcing scheme,this scheme can improve the minimum SINR among users with lower computational complexity,thus improving the communication quality of each user.Different from the maximum ratio transmission scheme,the minimum SINR of users in this scheme does not saturate with the increase of transmit power at base stations,thus enabling the steady growth of communication quality of each user.

In order to improve the efficiency of network transmission and obtain a better low-latency performance,caching technology appeared.Unlike traditional caching technology,coded caching enables a single broadcast transmission from the server to simultaneously satisfy different demands of users by creating multicast opportunities with a global caching gain obtained.A coded caching network with parallel transmission is considered in which the server can broadcast messages to all users and users can also send messages to each other.An uncoded prefetching coded caching scheme is proposed which is composed of three phases:the pre-caching phase,the allocation phase and the delivery phase,where the optimal delivery time is obtained by pre-allocating different workloads to the server and users.It is shown that the proposed scheme with parallel transmission has a better performance compared with either server-multicast transmission alone or transmission within a D2D network alone.Also,after considering the channel capability gap between the two different channels,the proposed scheme obtains a better performance than when channel transmission capability is ignored.Finally,the proposed cache and delivery scheme with parallel transmission in the case of uncoded prefetching is proved to be optimal when the users’ cache resources are sufficient and the server broadcast channel and the D2D network transmission channel have the same channel capacity.

Existing low-light dehazing algorithms are affected by the low and uneven illumination of the hazy images with their dehazed images often suffering from loss of details and color distortion.To address the above problems,a low-light image dehazing network with aggregated context-aware attention (ACANet) is proposed.First,an intra-layer context-aware attention module is introduced to identify and highlight significant features at the same scale from the channel dimension and the spatial dimension,respectively,so that the network can break through the constraints of the local field of view,and extract image texture information more efficiently.Second,an inter-layer context-aware attention module is introduced to efficiently fuse multi-scale features and the advanced features are mapped to the signal subspace through projection operations in order to further enhance the reconstruction of image details.Finally,the CIEDE2000 color shift loss function is adopted to constrain the image hue by CIELAB color space and jointly optimize the network together with L2 loss so as to enable the network to learn image colors accurately and solve the severe color shift problem.Both quantitative and qualitative experimental results on several datasets demonstrate that the proposed ACANet outperforms existing dehazing methods.Specifically,the ACANet improves the PSNR of dehazed images by 8.8% compared to the baseline network,and enhances the image visibility with richer details and more natural color.

In order to verify the influence of different water quality conditions on underwater optical communication,the Monte Carlo simulation algorithm is used to simulate the discrete impulse response of the underwater optical channel.Combined with the Orthogonal Frequency Division (OFDM) and strong anti-fading ability,the underwater optical communication system is built based on the OFDM.With known channel state information at the receiver and the transmitter,the 16 Quadrature Amplitude Modulation is used,and the Monte Carlo construction algorithm is applied to construct the polar codes.The CRC-Aided Successive Cancellation List polarization decoding algorithm is used to design the polar codes in OFDM-based underwater optical communication.The influence of parameters such as code length on the performance of polar codes under different water quality conditions is verified by experiments.It is also proved that compared with the low-density parity check(LDPC) codes of the same code length,the polar codes get a performance gain of 0.2dB~0.6dB at the high signal noise ratio.As the water quality environment worsens,its progressive performance becomes more obvious,and there will be no error leveling problem.In addition,the polar codes have a simpler coding structure,the decoding complexity is not much different from that of LDPC codes,and multiple iterations are not required in decoding.Since polar codes have a lower coding and decoding complexity than other coding schemes,polar codes have strong competitiveness and application potential in underwater optical communication scenarios.

In order to verify the mechanism of GEO SAR imaging with a long synthetic aperture time,the equivalence verification is executed by satellite-to-ground bistatic configuration,which is composed of the Beidou 3 IGSO navigation satellite and the ground-stationary receiver.According to the signal characteristics of navigation satellites and the time-frequency synchronization problem of the satellite-to-ground bistatic SAR,based on the echo data a time-frequency synchronization error estimation method of satellite-ground bistatic SAR echo data is proposed to estimate the delay error of navigation satellite ranging code and correct the error.First,the one-dimensional direct wave signal and the reflected wave signal are divided into two dimensions according to the pulse repetition frequency of the navigation satellite,which preserves the integrity of the whole collected signal.The correct peak position sequence under non-ideal sampling environment is obtained by the matching filter on the range of the direct wave signal,and then the peak position sequence is used to complete the time synchronization error compensation of the direct wave signal and the reflected wave signal,and so with the correction of the local ranging code,which solves the problem of the mismatch between the envelope distribution of the distance pulse pressure signal after time synchronization and the theorectical model in the satellite-ground bistatic SAR imaging of the Beidou-3 IGSO navigation satellite.Second,the corrected local ranging code is used to obtain the direct wave peak phase vector by matching filtering the direct wave signal.Finally,the peak phase vector is used to perform frequency error compensation,followed by the bistatic SAR imaging processing of the reflected wave signal.The processing results of the measured data verify the effectiveness of the proposed method.

In this paper,we investigate the covert communications in uplink nonorthogonal multiple access(NOMA) systems with the random power allocation and weighted fractional Fourier transform(WFRFT) scheme to confront a two-phase detector.In the NOMA system,a covert user(CU) and a reliable user(RU) transmit messages to Bob in the presence of a warden (Willie) who tries to detect the CU's transmission behavior.A two-phase detector,i.e.,energy detection and similarity detection phases,is designed to improve the detection performance.The similarity detection phase provides a priori probability for the energy detection phase and reduces the detection error probability.In addition,corresponding to the RU and CU,a random power allocation and WFRFT are proposed to cover the CU's transmissions.The expected minimum detection error probability(EMDEP) and connection outage probabilities(COPs) at the RU and CU are derived in closed-form expressions for the proposed scheme.To optimize the power allocation of the RU,the maximum expected covert rate(ECR) is analyzed under covertness and reliability constraints.Numerical results show that the proposed two-phase detector has a lower EMDEP,and that the random power allocation and WFRFT scheme improve the covertness performance.

Aiming at the problem of physical layer security transmission when there is no direct transmission link between transmitter and desired user,a physical layer transmission scheme based on the reconfigurable intelligent surface (RIS) is proposed.This dissertation first establishes an RIS array antenna system model.The secure communication of the wireless system is completed through the array antenna direct transmission link and the RIS reflection link.Furthermore,in view of the situation that eavesdropers passively receive information and cannot determine the specific location of eavesdropers,the dissertation optimizes the transmitting beamforming vector and the RIS coefficient matrix to maximize artificial noise interference power while ensuring reliable signal reception for the desired user.The non-convex quadratic problem is transformed into an equivalent convex problem by using auxiliary variables and semidefinite relaxation methods.The transmitting beamforming vector and reflection coefficient matrix of the intelligent reflector are optimized jointly,which can restrain the eavesdropper from receiving information and ensure the reliable and secure communication of expected users.Finally,simulation results show that the secure transmission scheme based on the RIS improves the reliability and security of information transmission.

In order to solve the problem of energy shortage,energy harvesting technology has been proposed and widely used in several typical communication systems.The author studies the problem of how to effectively select remote radio frequency units to realize downlink communication in a distributed base-station system with energy harvesting capability.First,an energy harvesting distributed base-station system model is established,which does not rely on external energy sources such as power grids,and only consists of a baseband processing subsystem,energy subsystem,and remote radio frequency subsystem.Second,based on the model,a joint optimization problem covering beamforming,energy sharing,and power allocation is formed with the goal of maximizing the information transmission rate.Since the energy harvesting situation is uncontrollable,two different energy-sharing strategies are proposed,on the basis of which the problem is analyzed mathematically,the optimal power allocation strategy of the system remote radio frequency end is deduced,and then the remote radio frequency end selection algorithm for the system model is summarized.Finally,Monte Carlo simulation is carried out based on the model and algorithm in this paper,and compared with the literature algorithm.Simulation results show that the algorithm proposed in this paper has a good performance in terms of average channel capacity and energy efficiency,and helps to save system power consumption and resources.

To address the problems of difficult identification of zone attribution of fingerprint points and misjudgment of neighboring zone matching accompanying the traditional spatial division method in fingerprint localization,a spatial fuzzy division method applicable to zone center identification and transition dual domain discrimination is proposed.By using the difference degree between inter-class distance and intra-class distance of reference points to measure the ambiguity of sub-region boundaries,we ensure the optimization of the localization cost of experimental scenes while taking into account the advantage of spatial overlap division,so as to alleviate the negative effect of absolute discrimination between sub-regions and improve the generalization ability of localization matching.In the position estimation stage,the distance metric in the signal domain between the reference point and the point to be located is transformed into a dimensionless ranking under the same source difference by considering the received signal fluctuation difference between the neighborhoods of the reference point,and the similarity between the point to be located and the reference point is indirectly mapped with the corrected multi-source ranking equalization result;in addition,the introduction of the spatial density reachable search strong correlation reference point set,combined with the signal domain and spatial domain iterative constraint reference points,to achieve dynamic selection and clustering effect of the target nearest neighbor set,so as to effectively overcome the interference of environmental changes and signal fluctuations,and improve the environmental adaptability of the localization method.After the evaluation of the localization performance by the measured data under the road it is shown that the proposed algorithm outperforms similar zoning algorithms in localization accuracy by 4.7%~11.8%,and that the average localization error can be best reduced by 0.422m in comparison with the global matching method.

As the core computing platform of the convolution neural network,general-purpose graphics processor(GPGPU),its performance of processing two-dimensional and three-dimensional convolution determines the application of the neural network in real-time target recognition and detection.However,limited by inherent cache system design,the current GPGPU architecture cannot achieve efficient acceleration of 2D and 3D convolution computing.Aiming at this problem,a dynamic L1Dcache bypassing design for this problem is proposed.First,we define a new data structure that can dynamically reflect the cache access characteristics of an instruction,and then defines a memory-access-feature record table based on this information,in order to record the execution status of different memory accesses.Second,the warp scheduling strategy with the priority thread block is adopted,which can speed up the sampling of the memory access state.Next,the L1Dcache bypassing decision of memory accesses under different PCs is obtained due to the sampling results.Finally,the L1Dcache bypassing of some low-locality data accesses is completed.As a result,the L1Dcache space is reserved for data with high locality and the memory access stall cycle of 2D and 3D convolution is reduced.In addition,the memory access efficiency of 2D and 3D convolution has been improved.Compared with the original design,experimental results show that the L1Dcache bypassing design brings 2.16% performance improvements in 2D convolution and 19.79% in 3D convolution.Experiments prove the effectiveness and practicality of this design.

In target tracking,the realization of 3D extended target tracking usually requires a large number of measurement data from multiple angles,and the measurement obtained by a single sensor can not meet the requirements of 3D shape estimation in either quantity or integrity.Aiming at the problem of poor shape tracking performance of existing 3D extended target tracking algorithms under a low measurement rate,a three-dimensional extended target algorithm based on the B-spline Poisson Multi-Bernoulli Mixture (B-Spline-PMBM) filter is proposed.First,wavelet clustering is used to process the 3D spatial measurement data obtained by multiple sensors to obtain the measurement cluster,which can extract effective information and ensure the efficiency of the algorithm.Then,the control matrix is obtained by dividing the measurement cluster.The control matrix is realized based on the B-spline control point principle,so it can represent the parameters of the complex three-dimensional shape.The shape of the 3D extended target is obtained by fitting the B-spline surface with the control matrix.Finally,the B-spline is integrated into the PMBM filter,which is extended to 3D target tracking to predict and update the motion state and shape parameters of the extended target.Simulation and real point cloud data set verify that the proposed algorithm can achieve a good tracking effect on the motion state and the extended shape of the three-dimensional extended target,and can realize the estimation of the irregular three-dimensional shape.

Block selection is a key strategy to improve the performance of pixel value ordering based reversible data hiding.By preferentially embedding data into blocks with a smaller fluctuation at first,better imperceptibility is able to be achieved.However,the accuracy of existing methods to calculate the fluctuation value is limited by the block size,shape and the selected predictor,so it is particularly important to design a calculation method that can be used under different block sizes,shapes and types of predictors.For large blocks or those irregular in shape,if the spatial position correlation of pixels used in the expansion is weak,even though the corresponding block is small in fluctuation,invalid shifting is still able to be introduced in the actual embedding process.To deal with this problem,we first propose an improved definition of fluctuation to consider the consistency of context pixels in horizontal,vertical,and bi-diagonal directions simultaneously.Once the consistency of adjacent pixels in the local neighborhood is determined,we further calculate the overall consistency of neighboring pixels in each direction,which improves the accuracy of block selection.Second,we consider the Chebyshev distance between the maximum value and the second maximum value,and that between the minimum value and the second minimum value,and reduce the invalid shift by subtracting the maximum or minimum pixel expansion.Experimental results show that our proposed scheme is able to achieve a better imperceptibility.

A provable secure consensus mechanism is proposed.The consensus mechanism consists of a committee agreement protocol and a transaction confirmation algorithm.Nodes with a strong initiative and more stakes are elected from consensus nodes through the committee agreement protocol,which form a dynamic and iterative committee to parallelly handle Normal Net Transactions (NNTs) generated by users in the blockchain.The transaction confirmation algorithm is based on a Directed Acyclic Graph (DAG) which is constructed from Chain Transactions (CTs) sent by committee members.The number of times that an NNT is confirmed by committee members is counted based on the direct and indirect references of a DAG.When an NNT is confirmed by at least two-thirds of the committee members,the transaction is welded in the chain.Under some accepted assumptions,the consensus mechanism is proven to have properties of consistency and termination.Further,a preliminarily blockchain system is built based on the consensus mechanism with the performance of the system tested.The test result is consistent with the theoretical estimation.When 16 committee members are deployed and the transaction batch is 10⁶,the transaction throughput of the system is about 17000 transactions per second;Compared with the current HoneyBadger BFT consensus protocol,under the same configuration,the transaction throughput of the HoneyBadger BFT consensus protocol is about 2600 transactions per second,which is about 1/6 that of the system.

An App traffic identification scheme based on machine learning under ShadowSocksR (SSR) proxy is proposed with the purpose being to identify from which APP the ShadowSocksR proxy traffic generated by the smartphone originates.The scheme consists of three steps:traffic preprocessing,feature extraction and model construction.First,the packet set corresponding to the ShadowSocksR traffic generated by smartphones is divided into fine-grained stream data groups according to the arrival time interval,source and destination IP address and port,and then the stream data groups containing fewer packets are further filtered with the purpose being to filter out noise traffic generated by the background App or smart phone operating system that interferes with traffic identification.Then,from the filtered flow data grouping set,the statistical features and distribution features of packet length,time statistical features,packet frequency features,packet filtering ratio features,and the combined features of the front and rear streams are extracted to form a feature matrix,which is input into the machine learning algorithm.An app traffic identification model for the ShadowSocksR traffic that needs to be identified is obtained,and after the feature matrix is obtained through the same processing steps,the flow identification results can be obtained by inputting the App traffic identification model.Experimental results show that the traffic identification method can reach an accuracy rate of more than 97% for App traffic identification under ShadowSocksR proxy.

Various location-based big data services not only provide users with convenience but also lead to privacy leakage risks.The local differential privacy model avoids the dependence on trusted third-party data collection platforms and enables users to process and protect sensitive information according to their personal needs.Therefore,it is more suitable for location privacy protection scenarios.In view of the complex encoding mechanism and low availability of the current local differential privacy location protection methods,a local differential privacy location protection method based on the Hilbert encoding is proposed.The user side performs random response perturbation on the Hilbert code of the grid where he is located according to the local differential privacy model,so as to realize the privacy protection of his original location.The server side collects a large number of users’ disturbed location codes and performs the Hilbert decoding,in order to determine the grid location of users and realize the statistical analysis of distribution density of users.Experiments on actual location datasets prove that the proposed method can provide a better location data availability and operational efficiency on the basis of realizing local differential privacy protection of users’ location.

The ring signature is a special digital signature that can provide unconditional anonymous protection for signers,and a traceable ring signature is a variant of the ring signature,which aims to prevent signers from abusing the anonymity of the ring signature,that is,the anonymity provided by the traceable ring signature for signers is not unconditional,which will lead to the identity of signers being disclosed under certain behaviors of the signer.The traceable ring signature plays an important role in an electronic voting system and an electronic cash system.Aiming at the present situation that traceable ring signature schemes on lattice are based on the PKI system and have a complex burden of digital certificate management,this paper combines identity-based cryptography with the traceable ring signature on lattice and proposes the first identity-based traceable ring signature scheme on the lattice.Different from the previous traceable ring signature schemes,the proposed scheme is constructed according to the framework of Baum et al.’s linkable ring signature scheme on lattice and based on the techniques of preimage sampling and reject sampling,etc.,thus avoiding the use of cumbersome zero-knowledge proofs.Under the random oracle model,it is proved that the proposed scheme can meet the tag-linkability,anonymity and exculpability,and that the security can be reduced to SIS and ISIS problems.In addition,compared with the related schemes,the proposed scheme also has some advantages in time overhead and storage overhead.

Vehicular ad-hoc networks (VANETs) have received substantial attention on account of great convenience to modern transportation systems.In VANETs,the authentication of the vehicular access control and the privacy of the messages are two crucial criteria.At the same time,verification efficiency is still critical due to the limited bandwidth and high mobility characteristics of vehicles.Aggregate signcryption schemes can effectively solve the above issues.However,some of the state-of-art schemes based on the Schnorr signature are unable to resist two types of signature forgery attacks due to incorrect hash binding.In addition,two vehicles can maliciously exchange their signcryption information which can be verified successfully.A new certificateless aggregate signcryption scheme for VANETs is presented.Secret key preimage protection technology is used to prevent signature forgery attacks and hash collision resistance is utilized to resist coalition attack.The confidentiality and unforgeability of the scheme are proved under the random oracle model.Furthermore,in comparison with the state-of-art schemes,the proposed scheme which requires 6n+1 point multiplication operations during the whole authentication process enhances security without increasing computational overhead.Performance analysis shows that the scheme is suitable for VANETs.

Most of the existing distributed learning schemes solve the problem of malicious nodes by adding a disciplinary mechanism to the protocol.This method is based on two assumptions:1.Participants give up the assumption of malicious behavior to maximize their own interests,and the calculation results can be verified only after the event occurs,which is not suitable for some scenarios requiring immediate verification;2.It is based on the assumption of a trusted third party.However,in practice,the credibility of the third party cannot be fully guaranteed.Using the trust mechanism of the blockchain,this paper proposes an anti malicious node scheme based on the smart contract,which realizes the whole process of model training in machine learning through the smart contract to ensure that the machine learning model is not damaged by malicious nodes.This scheme takes the distributed machine learning model based on secure multi-party computing as the research model,and uses the smart contract of the blockchain to realize the data sharing,verification and training process.All participants can only execute according to the specified protocol,converting all participants into semi sincere participants;At the same time,in order to solve the privacy problems brought by the open and transparent characteristics of the blockchain,ring signature is used to hide the data address of participants and protect the identity of participants.Experiments show that this scheme has great advantages in resisting malicious nodes compared with the traditional distributed machine learning model based on secure multi-party computing.