基于SDN的铁路系统空天地融合网络架构

针对列车通信网络制式单一、地理环境差、数据安全可靠传输难以得到保证的现状,调研铁路系统各业务的通信需求,提出一种基于软件定义网络(SDN,software defined network)和网络功能虚拟化(NFV,network functions virtualization)的铁路系统空天地融合网络体系架构。通过集中控制器管理底层铁路系统的物理设施,实现应用程序与物理基础设施的解耦,讨论了体系架构的设计细节,并对架构的协议需求、功能需求以及信息流等展开综述,给出空天地融合网络与其他单一制式网络覆盖的性能对比分析,并简要讨论了目前待解决的问题及解决方向。

基于DNN卷积核分割的边缘协作推理性能分析

随着智能芯片在边缘终端设备的普及,未来大量的AI应用将部署在更靠近数据源的网络边缘。基于DNN的分割方法可以实现深度学习模型在资源受限的终端设备上训练和部署,解决边缘AI算力瓶颈问题。在传统基于工作负载的分割方案(WPM,workload based partition method)的基础上,提出基于卷积核的分割方案(KPM,kernelbasedpartitionmethod),分别从计算量、内存占用、通信开销3个方面进行推理性能的定量分析,并从推理过程灵活性、鲁棒性、隐私性角度进行定性分析。最后搭建软硬件实验平台,使用PyTorch实现Alex Net和VGG11网络进一步验证所提方案在时延和能耗方面的性能优势,相比于传统工作负载分割方案,所提卷积核分割方案在大规模计算场景下有更好的DNN推理加速效果,且具有更低的内存占用和能量消耗。

Energy Saving Cooperative Communication over Fading Channels with Relay Selection and Power Control

In order to save energy and make more efficient use of wireless channel, this article puts forward an energy saving cooperative relaying scheme which actuates the cooperative transmission only when the feedback from the destination indicates failure of the direct transmission. The proposed scheme selects the optimal relay and its corresponding transmission power in each time slot based on channel condition and residual energy with the objective of minimizing energy consumption and extending network lifetime. In the study, the finite-state Markov channel model is used to characterize the correlation structure of channel fading in wireless networks, and the procedure of relay selection and transmission power decision is formulated as a Markov decision process. Numerical and simulation results show that the proposed scheme consumes less energy and prolongs the network lifetime.

QoE-Oriented Resource Allocation for Multiuser-Multiservice Femtocell Networks

The pursuit of high data rate and assurance of quality of experience(QoE) for end users represent the main goals of future wireless communication systems.By introducing MOS(Mean Opinion Score) based assessment models for different types of applications,this paper proposed novel QoE-oriented radio resource allocation(RRA) algorithms for multiuser-multiservice femtocell networks.An optimal QoE-oriented RRA strategy is first analyzed using time-sharing method which is applicable to best effort applications.RRA algorithms based on the cross-layer architecture are then proposed for all types of applications by considering parameters extracted from different layers of networking protocols.In the proposed algorithms,a priority mechanism is employed to ensure fairness.Simulation results show that the proposed algorithms can significantly improve the overall perceived quality from the users' perspective in comparison with traditional Quality of Service(QoS)oriented algorithms.

The adaptive distributed learning based on homomorphic encryption and blockchain

The privacy and security of data are recently research hotspots and challenges.For this issue,an adaptive scheme of distributed learning based on homomorphic encryption and blockchain is proposed.Specifically,in the form of homomorphic encryption,the computing party iteratively aggregates the learning models from distributed participants,so that the privacy of both the data and model is ensured.Moreover,the aggregations are recorded and verified by blockchain,which prevents attacks from malicious nodes and guarantees the reliability of learning.For these sophisticated privacy and security technologies,the computation cost and energy consumption in both the encrypted learning and consensus reaching are analyzed,based on which a joint optimization of computation resources allocation and adaptive aggregation to minimize loss function is established with the realistic solution followed.Finally,the simulations and analysis evaluate the performance of the proposed scheme.

基于深度强化学习的滤波器剪枝方案

针对深度神经网络模型在终端设备上部署时面临计算和存储等资源不足的问题,模型剪枝是一种有效的模型压缩方案,在保证模型精度的前提下减少模型的参数量并降低计算复杂度。传统的剪枝方案对于剪枝率及剪枝标准的设置大多依据先验知识,忽略了深度模型中不同层的剪枝敏感度和参数分布差异,缺乏细粒度的优化。对此,提出了一种基于强化学习的滤波器剪枝方案,在满足目标稀疏度的基础上最小化模型剪枝后的精度损失,并采用参数化深度Q学习算法求解构建混合变量的非线性优化问题。实验结果表明,所提方案能够为深度模型每一层选择合适的剪枝标准与剪枝率,减小了模型剪枝后的精度损失。

工业物联网中基于信息熵的联邦增量学习算法与优化

面对工业生产过程中大规模、多样且随时间增长的数据和机器学习任务,该文提出一种基于信息熵的联邦增量学习(FIL)与优化方法。基于联邦框架,各本地计算节点可利用本地数据进行模型训练,并计算信息平均熵上传至服务器,以此辅助识别类增任务;全局服务器则根据本地反馈的平均熵选择参与当前轮次训练的本地节点,并判决任务是否产生增量后,进行全局模型下发与聚合更新。所提方法结合平均熵和阈值进行不同情况下的节点选择,实现低平均熵下的模型稳定学习和高平均熵下的模型增量式扩展。在此基础上,采用凸优化,在资源有限的情况下自适应地调整聚合频率和资源分配,最终实现模型的有效收敛。仿真结果表明,在不同的情景下,该文所提方法都可以加速模型收敛并提升训练精度。

Joint multi-QoS and energy saving routing for LEO satellite network

Low earth orbit(LEO) satellite network provides global coverage and supports a wide range of services. However, due to the rapid changes and energy-limitation of satellites, how to meet the demand of the quality of service(QoS) from ground traffic and prolong the lifetime of LEO satellite network is the research emphasis of the investigator. Hence, a routing algorithm which takes into account the multi-QoS requirements and satellite energy consumption(QER) of LEO satellite network is proposed. Firstly, the satellite intimacy degree(SID) and the path health degree(PHD) are introduced to obtain the path evaluation function according to the energy consumption and queue state of the satellite. Then, the distributed routing QER is established through the path evaluation function and the idea of genetic algorithm(GA), which enables each satellite to adjust traffic and realizes the network load balancing. Simulation results show that QER performs well in terms of end-to-end delay, delay jitter, and system throughput.

Adaptive transfer learning framework for dense prediction of human activity recognition

Human activity recognition(HAR)for dense prediction is proven to be of good performance,but it relies on labeling every point in time series with the high cost.In addition,the performance of HAR model will show significant degradation when tested on the sensor data with different distribution from the training data,where the training data and the test data are usually collected from different sensor locations or sensor users.Therefore,the adaptive transfer learning framework for dense prediction of HAR is introduced to implement cross-domain transfer,where the proposed multi-level unsupervised domain adaptation(MLUDA)approach combines the global domain adaptation and the specific task adaptation to adapt the source and target domain in multiple levels.The multi-connected global domain adaptation architecture is proposed for the first time,which can adapt the output layer of the encoder and the decoder in dense prediction model.After this,the specific task adaptation is proposed to ensure alignment of each class centroid in source domain and target domain by introducing the cosine distance loss and the moving average method.Experiments on three public HAR datasets demonstrate that the proposed MLUDA improves the prediction accuracy of target data by 20%compared to the source domain pre-trained model and it is more effective than the other three deep transfer learning methods with an improvement of 10%to 18%in accuracy.

Coverage analysis of cache-enabled small cell networks with stochastic geometry methods

Caching popular content in the storage of small cells is deemed as an efficient way to decrease latency, offload backhaul and satisfy user＇s demands. In order to investigate the performance of cache-enabled small cell networks, coverage probability is studied in both single-point transmission and cooperative multipoint（Co MP） transmission scenarios. Meanwhile, the caching distribution modeled as Zipf and uniform distribution are both considered. Assuming that small base stations（SBSs） are distributed as a homogeneous Poisson point process（HPPP）, the closed-form expressions of coverage probability are derived in different transmission cases. Simulation results show that Co MP transmission achieves a higher coverage probability than that of single-point transmission. Furthermore, Zipf distribution-based caching is more preferable than uniform distribution-based caching in terms of coverage probability.