算力供给的代际异构性与供应链安全需求,促使异构算力成为AI基础设施的新趋势。然而,在异构混合训练场景中,基于融合以太网的RDMA版本2(RDMA over converged Ethernet version 2,RoCEv2)方案存在负载均衡与拥塞控制缺陷,在模型训练的并...算力供给的代际异构性与供应链安全需求,促使异构算力成为AI基础设施的新趋势。然而,在异构混合训练场景中,基于融合以太网的RDMA版本2(RDMA over converged Ethernet version 2,RoCEv2)方案存在负载均衡与拥塞控制缺陷,在模型训练的并行通信中性能欠佳;而现有高性能同构智算网络方案因设备异构与集合通信库(collective communication library,CCL)闭源难以部署。为此,提出了面向异构算力场景的高性能智算网络解决方案——智能控制以太网(intelligent control Ethernet,ICE)。该方案基于RoCEv2协议体系,在避免对设备、CCL进行深度定制的前提下,将异构通信库信息采集、集中控制器与端侧自主控制相结合,实现全局最优路径规划及全局主动拥塞控制,显著提升异构并行通信性能。真实物理环境实验表明,ICE可提升集合通信性能最高达47%。ICE为异构智算网络建设提供了开创性、易部署的解决方案。展开更多
第五代移动通信网络5G以融合网络为目标,其标准不仅覆盖公共通信网络,也同时应用于下一代垂直行业网络。传统垂直行业网络是以工业自动化和控制系统为主的运营/操作技术(Operational Technology,OT)网络,OT网络采取安全域划分方式,将大...第五代移动通信网络5G以融合网络为目标,其标准不仅覆盖公共通信网络,也同时应用于下一代垂直行业网络。传统垂直行业网络是以工业自动化和控制系统为主的运营/操作技术(Operational Technology,OT)网络,OT网络采取安全域划分方式,将大规模复杂系统分为不同安全子区域,在边界处部署专用安全设备/系统进行安全防护。目前实践中多采用网闸等设备以硬隔离方式阻断恶意流量,带来的问题是严重影响正常业务的通过。依托5G的网络功能虚拟化(Network Function Virtualization,NFV)技术和软件定义网络(Software Defined Network,SDN),本文提出了一种面向5G网络的动态安全边界防护机制。该机制构建虚拟化的边界网络安全功能资源池和边界安全服务规则库,对到达边界的业务流量进行防护等级分析,并根据规则库中的规则动态生成边界安全服务功能链。机制还具备对边界服务功能链进行优化部署的能力,通过建模和启发式算法实现满足业务防护等级需求和最小化处理时延的多目标优化部署策略。基于本机制,我们设计并提出轨交行业5G专网动态安全边界防护机制实例,旨在为工程实践服务。最后,我们搭建了基于Mininet+Ryu仿真平台,模拟轨交行业5G示范网络中的安全域组成和边界安全能力,并对机制进行实验验证,结果表明,该机制能够有效地动态生成边界服务功能链并且达到控制不同防护等级业务流量通过的目标。展开更多
In this paper,the authors study the fractional Calderon type commutator T_(Ω,α)^(A)and its maximal operator M_(Ω,α)^(A)with kernels having some kinds of Log-type Dini-condition and obtain the compactness on Morrey...In this paper,the authors study the fractional Calderon type commutator T_(Ω,α)^(A)and its maximal operator M_(Ω,α)^(A)with kernels having some kinds of Log-type Dini-condition and obtain the compactness on Morrey spaces L^(p,λ)(R^(n)).展开更多
To improve the bit error rate(BER)performance of multi-user signal detection in satelliteterrestrial downlink non-orthogonal multiple access(NOMA)systems,an iterative signal detection algorithm based on soft interfere...To improve the bit error rate(BER)performance of multi-user signal detection in satelliteterrestrial downlink non-orthogonal multiple access(NOMA)systems,an iterative signal detection algorithm based on soft interference cancellation with optimal power allocation is proposed.Given that power allocation has a significant impact on BER performance,the optimal power allocation is obtained by minimizing the average BER of NOMA users.According to the allocated powers,successive interference cancellation(SIC)between NOMA users is performed in descending power order.For each user,an iterative soft interference cancellation is performed,and soft symbol probabilities are calculated for soft decision.To improve detection accuracy and without increasing the complexity,the aforementioned algorithm is optimized by adding minimum mean square error(MMSE)signal estimation before detection,and in each iteration soft symbol probabilities are utilized for soft-decision of the current user and also for the update of soft interference of the previous user.Simulation results illustrate that the optimized algorithm i.e.MMSE-IDBSIC significantly outperforms joint multi-user detection and SIC detection by 7.57dB and 8.03dB in terms of BER performance.展开更多
Dynamic adsorption processes of reaction intermediates for alkaline hydrogen evolution(HER)catalysts are still confusing to understand.Here,we report a series of A-site ordered quadruple perovskite ruthenium-based ele...Dynamic adsorption processes of reaction intermediates for alkaline hydrogen evolution(HER)catalysts are still confusing to understand.Here,we report a series of A-site ordered quadruple perovskite ruthenium-based electrocatalysts ACu_(3)Ru_(4)O_(12)(A=Na,Ca,Nd,and La),with the target sample SrCu_(3)Ru_(4)O_(12)exhibiting a very low overpotential(46 mV@10 mA·cm^(-2))and excellent catalytic stability with little decays after 48-h durability test.Precise tuning A-site cations can change the average valence state of Cu and Ru,thus the plot of HER activity versus the average Ru valence number shows a volcano-type relationship.Density functional theory indicates that the Ru 4d orbitals of SrCu3Ru4O12possesses the most suitable d-band center position among the five samples,which might be the key parameter to determine the catalytic performance.Our work provides further insight into the discovering advanced,efficient hydrogen evolution catalysts through designing precise descriptor.展开更多
Images are the most important carrier of human information. Moreover, how to safely transmit digital imagesthrough public channels has become an urgent problem. In this paper, we propose a novel image encryptionalgori...Images are the most important carrier of human information. Moreover, how to safely transmit digital imagesthrough public channels has become an urgent problem. In this paper, we propose a novel image encryptionalgorithm, called chaotic compressive sensing (CS) encryption (CCSE), which can not only improve the efficiencyof image transmission but also introduce the high security of the chaotic system. Specifically, the proposed CCSEcan fully leverage the advantages of the Chebyshev chaotic system and CS, enabling it to withstand various attacks,such as differential attacks, and exhibit robustness. First, we use a sparse trans-form to sparse the plaintext imageand then use theArnold transformto perturb the image pixels. After that,we elaborate aChebyshev Toeplitz chaoticsensing matrix for CCSE. By using this Toeplitz matrix, the perturbed image is compressed and sampled to reducethe transmission bandwidth and the amount of data. Finally, a bilateral diffusion operator and a chaotic encryptionoperator are used to perturb and expand the image pixels to change the pixel position and value of the compressedimage, and ultimately obtain an encrypted image. Experimental results show that our method can be resistant tovarious attacks, such as the statistical attack and noise attack, and can outperform its current competitors.展开更多
文摘算力供给的代际异构性与供应链安全需求,促使异构算力成为AI基础设施的新趋势。然而,在异构混合训练场景中,基于融合以太网的RDMA版本2(RDMA over converged Ethernet version 2,RoCEv2)方案存在负载均衡与拥塞控制缺陷,在模型训练的并行通信中性能欠佳;而现有高性能同构智算网络方案因设备异构与集合通信库(collective communication library,CCL)闭源难以部署。为此,提出了面向异构算力场景的高性能智算网络解决方案——智能控制以太网(intelligent control Ethernet,ICE)。该方案基于RoCEv2协议体系,在避免对设备、CCL进行深度定制的前提下,将异构通信库信息采集、集中控制器与端侧自主控制相结合,实现全局最优路径规划及全局主动拥塞控制,显著提升异构并行通信性能。真实物理环境实验表明,ICE可提升集合通信性能最高达47%。ICE为异构智算网络建设提供了开创性、易部署的解决方案。
文摘第五代移动通信网络5G以融合网络为目标,其标准不仅覆盖公共通信网络,也同时应用于下一代垂直行业网络。传统垂直行业网络是以工业自动化和控制系统为主的运营/操作技术(Operational Technology,OT)网络,OT网络采取安全域划分方式,将大规模复杂系统分为不同安全子区域,在边界处部署专用安全设备/系统进行安全防护。目前实践中多采用网闸等设备以硬隔离方式阻断恶意流量,带来的问题是严重影响正常业务的通过。依托5G的网络功能虚拟化(Network Function Virtualization,NFV)技术和软件定义网络(Software Defined Network,SDN),本文提出了一种面向5G网络的动态安全边界防护机制。该机制构建虚拟化的边界网络安全功能资源池和边界安全服务规则库,对到达边界的业务流量进行防护等级分析,并根据规则库中的规则动态生成边界安全服务功能链。机制还具备对边界服务功能链进行优化部署的能力,通过建模和启发式算法实现满足业务防护等级需求和最小化处理时延的多目标优化部署策略。基于本机制,我们设计并提出轨交行业5G专网动态安全边界防护机制实例,旨在为工程实践服务。最后,我们搭建了基于Mininet+Ryu仿真平台,模拟轨交行业5G示范网络中的安全域组成和边界安全能力,并对机制进行实验验证,结果表明,该机制能够有效地动态生成边界服务功能链并且达到控制不同防护等级业务流量通过的目标。
文摘In this paper,the authors study the fractional Calderon type commutator T_(Ω,α)^(A)and its maximal operator M_(Ω,α)^(A)with kernels having some kinds of Log-type Dini-condition and obtain the compactness on Morrey spaces L^(p,λ)(R^(n)).
基金supported by the National Key Research and Development Program of China(No.2021YFB2900602)the National Natural Science Foundation of China(No.61875230).
文摘To improve the bit error rate(BER)performance of multi-user signal detection in satelliteterrestrial downlink non-orthogonal multiple access(NOMA)systems,an iterative signal detection algorithm based on soft interference cancellation with optimal power allocation is proposed.Given that power allocation has a significant impact on BER performance,the optimal power allocation is obtained by minimizing the average BER of NOMA users.According to the allocated powers,successive interference cancellation(SIC)between NOMA users is performed in descending power order.For each user,an iterative soft interference cancellation is performed,and soft symbol probabilities are calculated for soft decision.To improve detection accuracy and without increasing the complexity,the aforementioned algorithm is optimized by adding minimum mean square error(MMSE)signal estimation before detection,and in each iteration soft symbol probabilities are utilized for soft-decision of the current user and also for the update of soft interference of the previous user.Simulation results illustrate that the optimized algorithm i.e.MMSE-IDBSIC significantly outperforms joint multi-user detection and SIC detection by 7.57dB and 8.03dB in terms of BER performance.
基金Project supported financially by the National Key Research and Development Program of China(Grant No.2023YFA1406000)the National Natural Science Foundation of China(Grant Nos.22171283 and 12474002)+3 种基金the Fundamental Research Funds for the Central Universities(Grant Nos.2023ZCJH03 and 2021XD-A041)the Fund of State Key Laboratory of Information Photonics and Optical Communications(Beijing University of Posts and Telecommunications,China)the Teaching Reform Projects at BUPT(Grant No.2022CXCYB03)the BUPT Excellent Ph.D.Students Foundation(Grant No.CX2023108)。
文摘Dynamic adsorption processes of reaction intermediates for alkaline hydrogen evolution(HER)catalysts are still confusing to understand.Here,we report a series of A-site ordered quadruple perovskite ruthenium-based electrocatalysts ACu_(3)Ru_(4)O_(12)(A=Na,Ca,Nd,and La),with the target sample SrCu_(3)Ru_(4)O_(12)exhibiting a very low overpotential(46 mV@10 mA·cm^(-2))and excellent catalytic stability with little decays after 48-h durability test.Precise tuning A-site cations can change the average valence state of Cu and Ru,thus the plot of HER activity versus the average Ru valence number shows a volcano-type relationship.Density functional theory indicates that the Ru 4d orbitals of SrCu3Ru4O12possesses the most suitable d-band center position among the five samples,which might be the key parameter to determine the catalytic performance.Our work provides further insight into the discovering advanced,efficient hydrogen evolution catalysts through designing precise descriptor.
基金the National Natural Science Foundation of China(Nos.62002028,62102040 and 62202066).
文摘Images are the most important carrier of human information. Moreover, how to safely transmit digital imagesthrough public channels has become an urgent problem. In this paper, we propose a novel image encryptionalgorithm, called chaotic compressive sensing (CS) encryption (CCSE), which can not only improve the efficiencyof image transmission but also introduce the high security of the chaotic system. Specifically, the proposed CCSEcan fully leverage the advantages of the Chebyshev chaotic system and CS, enabling it to withstand various attacks,such as differential attacks, and exhibit robustness. First, we use a sparse trans-form to sparse the plaintext imageand then use theArnold transformto perturb the image pixels. After that,we elaborate aChebyshev Toeplitz chaoticsensing matrix for CCSE. By using this Toeplitz matrix, the perturbed image is compressed and sampled to reducethe transmission bandwidth and the amount of data. Finally, a bilateral diffusion operator and a chaotic encryptionoperator are used to perturb and expand the image pixels to change the pixel position and value of the compressedimage, and ultimately obtain an encrypted image. Experimental results show that our method can be resistant tovarious attacks, such as the statistical attack and noise attack, and can outperform its current competitors.
