Edge computing(EC)is an emerging computing paradigm that enables application providers to serve users by allocating them to nearby edge servers,thereby reducing content delivery latency.How-ever,matching users with the suitable edge servers to accomplish specific optimization goals is challeng-ing.Most existing solutions ignore the impact of device heterogeneity and network complexity on the allo-cation policy.In this paper,we formally model the device heterogeneity and interference-aware user allo-cation(Hi-UA)problem and propose the bidirectional sorting greedy(DSort)algorithm and heuristic ant colony system(HAS)algorithm to improve the user allocation rate while reducing the average task laten-cy.Extensive experiments are conducted to assess the performance of the DSort and HAS methods using real user allocation datasets.The results demonstrate that the proposed methods effectively solve the Hi-UA problem and outperform the other three representative methods.
One broadband and high efficiency Class F power amplifier (PA) with a feedback circuit structure is proposed. Positive feedback is used to peak the gain and efficiency of the PA at high band. As a result, the PA is designed to have flat gain and efficiency performance over a wide bandwidth. Simulation results show that the PA has a power added efficiency (PAE) of 68.1%-72.6% and a gain of 10.6-11.4 dB in the frequency range of 1.6-2.6 GHz.
With the popularization and application of information interaction,intelligent decision-making and situational awareness in the field of UAVs,swarm warfare as a new mode of warfare has attracted much attention.Starting from the basic patterns of swarm operations,the force comparison and applica-tion characteristics of swarm operations in Syrian war,Yemen war,Naqqa conflict are analyzed.The de-velopment trend of swarm operations from the aspects of scale,informationization,intelligence and diver-sity are summarized.So as to provide reference for the development of swarm operations.
A group delay optimization method is proposed to solve the problem that the frequency phasor of ⅡR digital filters in the passband delays different time quantities.For the cascaded IIR filter system,the end of each filter stage is connected to an all-pass filter.Zero and pole analysis is performed on the designed all-pass filter,and stability of the all-pass filter and entire digital filter cascade is evaluated by analyzing position of the zero and pole in the z-plane.In the simulation and experimental part,8-order Butterworth low-pass filter in the passband range of 0～150 Hz is adopted,and sampling rate is 48 kHz.Besides,the first-order all-pass filter is adopted in the group delay optimization.Simulation results show that the change amount of group delay is optimized by 49.58％on original system.The proposed algo-rithm has the advantages of less computation and convenient implementation in comparison with other ma-instream methods.
2022 is the year of adjustment for the global semiconductor industry.Some Countries have car-ried out special investments to improve the flexibility and autonomy of their own supply chains.Techno-logical progress and process improvements have created opportunities for the development of thesemicon-ductor.Semiconductors have also become the core of technological competition.Sanctions and coopera-tion have become the new normal;new materials and new architectures have brought new opportunities for development in the post-Moore era.
The existing DGA domain name detection method has some problems,such as high detection time,low detection accuracy and poor detection effect of word-based DGA domain name.In this study,it is found that classifying domain names according to typical features before extracting more detailed fea-tures has a certain positive effect on the accuracy of the model,and multi-class parallel can reduce the detection time.In addition,targeted processing can be carried out for word-based DGA domain names that are difficult to detect.Therefore,this paper proposes a three-way heterogeneous parallel DGA do-main name detection model based on Word ninja segmentation technology.First,the domain name is di-vided into three categories,and then the detection model structure is built for each category.For charac-ter-level domain names,the effective classification of domain names is carried out by manually extracting features.For root-affix domain names,FastTest is used to extract the features between subwords,charac-ters and contexts,and then embed them as word vectors.For word-level domain names,Word2Vec is used to understand and deal with the meaning of words and the relationship between words.Finally,the proposed method is compared with the detection results of current popular methods,heterogeneous paral-lel model and single model.The experimental results prove the necessity of advance classification and the effectiveness of multi-parallel model.
Skip Connections usually correspond to image boundaries or discontinuity regions,which may have certain misalignment or inconsistency in different images.The edge alignment problem in the image fusion process can be improved by skip connections,and the fusion result can better preserve the detail information of the image,making the fused image clearer and more detailed,and reducing the occurrence of the edge discontinuity problem.In order to improve the spectral information and spatial details of the fused image without affecting the degree of loss,to improve the fusion accuracy,to reduce the information loss and to improve the network performance.In this paper,we propose a skip connections deep convolu-tional adversarial network for remote sensing image fusion,which introduces skip connections,explicitly uses residual blocks to form"short-circuit"connections,which helps the gradient back-propagation,pre-vents the training difficulty of the deep network,and reduces the loss of information and improves the per-formance of the network.The spectral information and spatial details of the fused images are improved to enhance the fusion accuracy.
Aiming at the problem that the strong pulse signal causes the"masking effeact"to affect the weak signal detection and the nonlinear distortion of the strong pulse signal causes the large pulse parame-ter measurement error,an adaptive pulse detection method under nonlinear conditions is proposed,which includes the adaptive threshold and the pulse start-stop decision method.In the adaptive threshold part,a stationary pulse difference model is proposed based on the noise distribution characteristics,and the a-daptive threshold is improved to reduce the"masking effect".In the part of pulse start-stop decision,the maximum difference decision method is adopted,and the pulse start and end point is determined at the maximum difference operation to reduce the pulse width measurement error.Simulation and FPGA meas-urement show that the improved adaptive threshold is only about 2 dB higher than the noise,and the"masking effect"is obviously reduced.Under the condition of single channel 51.2 MHz sampling rate,the average measurement error of nonlinear distorted pulse signal decreases from 0.2 μs to 0.02 µs.
Sentiment classification technology is widely used in many fields such as public opinion evalu-ation and commodity evaluation,which is of great research significance in the field of Natural Language Processing.In the current social network text,users not only use text to convey emotions,but also have strong emotional color in their text with emoji elements.Traditional sentiment analysis models are prone to ignoring emoji elements,resulting in the model not being able to accurately judge text emotions.This ar-ticle combines the BERT pre-training model with LSTM model,using weibo_senti_100k dataset with emo-ji elements to implement a sentiment binary classification model for Weibo comments.The BERT-LSTM model utilizes the BERT embedding layer to segment the preprocessed sentences and convert them into dynamic word vectors.It combines the LSTM model to extract features of text and emoji elements,and fi-nally predicts the emotional polarity of the comment text.The experimental verification of the importance of emoji elements and the effectiveness of the BERT-LSTM model for emotion classification.It showed that considering both text and emoji elements improved the classification accuracy of the model by 20％compared to pure text.The BERT-LSTM model has the accuracy of 98.31％and the F1 value of 98.28％,showing significant advantages over traditional machine learning models and other deep learning models in the final results.
A X-band radar transceiver module is designed in the form of hybrid microwave integrated cir-cuit(HMIC).The transceiver component is divided into three modules:frequency synthesizer,power amplifier and internal calibration source,and receiving link.The signal interconnection between each module is achieved through external coaxial cables.By fully utilizing the function of digital attenuator and adjusting attenuation of the amplification link,the goal of adjusting the transmission output power is achieved.Physical testing shows that within the frequency range of 10.5～12 GHz,the transmitting and output signal power of the transceiver module is adjustable in four levels:10 W,5 W,2 W,and 1 W,with spur suppression ≥ 65 dBc,receiving channel gain of 27.5 dB,and noise coefficient ≤4.8 dB.This component has the characteristics of adjustable output signal power,low noise figure,and high spu-rious suppression,which meets the requirements of project indicators.
In this paper,we propose an angle estimation algorithm based on tensor Vandermonde factor matrix reconstruction in multiple-input multiple-output(MIMO)radar with spatially colored noise.To suppress the colored noise,the proposed algorithm first achieves denoising of the received signal by the cross-correlation of the match filter outputs for different pulses.Then,using the priori information of the factor matrices,a Vandermonde constrained fourth-order tensor canonical decomposition/parallel factor analysis(CANDECOMP/PARAFAC)model is constructed.Next,we develop an iterative scheme based on constrained alternating least squares(ALS),in which the Vandermonde structure of the factor matrix is fully utilized in the alternating iteration process and the factor matrix estimates are obtained by con-structing Toeplitz matrices and performing Vandermonde decompositions.Finally,the target angles can be estimated by the least square fitting method from the estimation of factor matrices.Simulations results demonstrate that the proposed algorithm effectively improves the accuracy of angle estimation in the pres-ence of spatially colored noise for MIMO radar.
This paper proposes a recognition method based on cyclic spectrum integration features to ad-dress the low recognition rate of radar signals with composite modulation using traditional signal recogni-tion methods.Firstly,the Fourier accumulation method for computing the cyclic spectrum is transformed into the autocorrelation domain,reducing interference from noise.Secondly,by introducing the method of axial integration,the 2D cyclic spectrum matrix is reduced to a 1D feature vector while retaining the main information and eliminating redundant data.Finally,radar signal recognition is achieved using support vector machines.Experimental results show that this method guarantees an accuracy rate of over 90％for the recognition of seven different modulations of radar signals when the signal-to-noise ratio is not less than-2 dB.Furthermore,compared to other recognition methods,this method significantly improves the recognition rate of radar signals under low signal-to-noise ratio conditions,especially for phase modulation signals.This method extends the temporal features to the autocorrelation domain,making it easy to imple-ment and providing new insights into signal recognition methods.
A conceive of situation information processing system in NIC SoS(Networking Information-Centric System of Systems)is proposed.The system receives heterogeneous situation information from va-rious special networks.Then it translates the heterogeneous situation information into the system's inter-nal situation data with the unified format,and constructs the unified situation pictures in real time.It u-ses cloud storage and cloud computing to store and analyze the global situation data,and uses edge stor-age and edge computing to store and analyze the local situation data,displays and replays the situation in real time,reconstructs the situation in 3D.Last,some problems related to the system are thought.
Microblog short text usually contains rich emotional information. It is a hot research topic in network data mining to grasp the dynamics of network public opinion through microblog emotion analysis. In order to improve the effect of Chinese microblog sentiment analysis, this paper first uses word embedding technology to quantify microblog short text from high dimension to low dimension vector space; Then, the global features of microblog data are extracted through BiGRU, and the Attention mechanism is introduced to obtain important features to build a Chinese microblog emotion analysis model. The feasibility and superiority of the model were verified with the public data set released by SMP2020. The accuracy, recall and F1 values of the model reached 78.65%, 78.57% and 78.41% respectively. The experimental results show that the feature vector of BiGRU combined with attention mechanism contains more rich emotion information of short text of microblog, which can effectively improve the performance of sentiment analysis of Chinese microblog. The experimental results show that the feature vector of Bi-GRU combined with the attention mechanism contains richer semantic information of the text, which can effectively improve the performance of emotion recognition of online public opinions.
In order to solve the maneuverability forecast problem of unmanned surface vessel(USV),a identification method based on genetic algorithms(GA)is used to obtain the parameters of second-order nonlinear maneuver response model.Firstly,the zigzag simulation manipulation experiment is carried out on the simulation platform based by Runge-Kutta method.Then,the identification model is designed based on the difference method.Constructing fitness criterion function of identification model,the initial population gradually converges to the optimal solution after several generations of selection,crossover,and mutation,and the identification result can be calculated by convergence value.The identification re-sults are superior to the Particle Swarm Optimization(PSO)algorithm in terms of accuracy and conver-gence,with a maximum identification error rate of 4.19％in simulation experiments.Simulation experi-ments have verified the effectiveness and generalization of the identification results.GA is an effective high-precision identification algorithm,and the identification accuracy is better than PSO algorithm.The identification results can effectively predict the maneuverability of USV.
雷达抗干扰性能受到多种因素影响,为高效、全面地检验雷达抗干扰能力,在试验设计时需对干扰因子影响程度进行排序,获取干扰因子及干扰量值组合方案.本文利用正交设计的方法对传统层次分析法加以改进,运用满意度函数表征某干扰因子特定干扰量值下的干扰效果,结合映射函数计算得到干扰因子及干扰量值组合方案.该方法可在试验前对众多干扰因子进行筛选,获取不同组合下的干扰效果估值,可为雷达抗干扰试验设计提供参考,进而对雷达抗干扰性能摸边探底提供支撑.
区域交通流量预测是智慧交通系统的一项重要功能.联邦学习可以支持多位置服务提供商(Location Service Provider,LSP)的联合训练,使得训练数据集可以更加全面地覆盖整个区域的交通流量,提高预测准确率.但是,当前基于联邦学习的区域交通流量预测方案存在车辆数据去重、训练节点背叛以及隐私泄露等问题.为此,构建了基于联邦学习的隐私保护区域交通流量预测(Privacy-Preserving Regional Traffic Flow Prediction based on Federated Learning,PPRTFP-FL)模型.模型采用中心部署架构,由联邦中央服务器协调各个LSP联合完成模型的训练,并对全局模型进行梯度聚合与模型更新;采用交叉评价加权聚合的策略来防御部分不可信节点对全局模型的恶意攻击,提升了全局模型的鲁棒性;预测阶段使用同态加密聚合算法,各LSP在不泄露自身运营数据的情况下实现了更准确的流量预测.利用相关数据集进行测试,测试结果表明当训练数据集覆盖区域流量充分的情况下,本模型相比本地模型的预测准确率有明显的提升.对模型进行不同比例的恶意节点攻击实验,由实验结果可知,系统在存在恶意节点情况(当恶意节点数量小于总节点数量50％时)下仍具备较好的防御效果.
传统公网5G信道模型一般针对城市、街区等场景进行建模,在变电站等电力工业场景中适应性不足.为支撑5G在电网的规模化应用,需要开展对特定电力场景的信道测量与建模以指导电力5G无线网络规划和系统级仿真.本文在500 kV变电站开展5G频段的实地信道测量,分析了时延功率谱、路径损耗等信道特性,并在此基础上提出了一种引入天线高度矫正因子的路损模型修正方法,以便更准确地对天线高度与路损之间的关系进行建模,同时对比分析与5G标准路损模型之间的差异.结果表明,该模型能更好地适应电力应用场景,可为通信性能评估提供参考与指导,并为电力通信设备的覆盖范围预测提供理论依据.
针对多 目标复杂场景下装备协同引导信息处理流程复杂,计算量大,难以保证引导数据率要求的问题,本文提出了 一种服务/应用分离的分布式并行计算软件架构.服务层软件侧重于引导消息生成及分发,在服务访问方面采用微服务架构以及主备机冗余设计均衡系统负载、提高系统稳定性,在信息处理方面采用多线程技术配合openMP并行计算库提高计算效率.应用层软件侧重于引导信息显示及控制,基于Qt实现了多进程插件式集成框架,提高了拓展性和安全性.本文验证了饱和攻击场景下不同软件架构引导信息生成的指标,本文提出的软件架构在平均耗时、数据率达成率、丢帧率、重帧率等指标上大为改善,具备很高的实用价值.
针对无人艇在执行作战任务过程中,需要临机调整局部航路以实现紧急避障的需求,提出一种基于多目标粒子群-人工势场法的无人艇局部航路规划方法,首先针对传统人工势场法极小值问题,提出一种目标平移法进行改进,能够引导无人艇跳出局部极小值区域,并采用多 目标粒子群算法对改进后的人工势场法参数进行优化,设计出基于航路危险系数、路径长度系数、路径转向角系数的多 目标代价函数,实现了人工势场法参数快速、自主、最优调整.最后,通过不同工况下的仿真试验,验证了所提算法的有效性.
