本文提出了一种将特征生成和长短期记忆(long short term memory, LSTM)模型相结合的网络流量分类方法。该方法采用矩阵乘法特征生成方式,分析对比了不同特征生成方法的分类性能。通过实验比较了原数据和特征数据在分类问题上的准确性,...本文提出了一种将特征生成和长短期记忆(long short term memory, LSTM)模型相结合的网络流量分类方法。该方法采用矩阵乘法特征生成方式,分析对比了不同特征生成方法的分类性能。通过实验比较了原数据和特征数据在分类问题上的准确性,并比较了卷积神经网络(convolutional neural network, CNN)和本文方法用于网络流分类的效果。在统计特征时采用核函数,使其可以适应LSTM输入维度,获得更佳的分类效果。对真实网络流数据的实验结果表明,本文方法在细分类中的准确度可达93.9%,而在粗分类任务中可达99.2%,其性能明显优于现有其他分类方法。展开更多
流早期分类对于优化网络管理和确保服务质量(Quality of Service, QoS)至关重要。针对传统流特征在流早期分类中性能较低的问题,在现有研究基础上,提出了两种新的特征:一是通过等距分箱划分包大小等级,计算相邻到达的两个数据包的包大...流早期分类对于优化网络管理和确保服务质量(Quality of Service, QoS)至关重要。针对传统流特征在流早期分类中性能较低的问题,在现有研究基础上,提出了两种新的特征:一是通过等距分箱划分包大小等级,计算相邻到达的两个数据包的包大小等级条件频度;二是通过将包大小序列和包到达时间间隔对应相除,得到速率序列,并计算该序列的统计特征作为分类特征。同时,考虑到早期分类的实时性要求,分析了流特征计算的时间复杂性,在特征选择中优化了时间和准确性之间的平衡。另外,针对网络视频流量占比较大的情况,提出了一种层级分类结构;先使用较少的数据包进行non-video/video的二分类,再使用后续的数据包,进行non-videos和videos的细粒度分类。采用随机森林在两个实际网络数据集上进行分类性能测试,并与文献方法进行比较,验证了该方法在快速流量分类中的优越性。展开更多
在网络流分类实践中,网络运营商通常只需要知道网络流所需的服务类别(class of service,CoS),就可对网络流优先级和资源分配做出决定。为了满足用户对体验质量的需求,提出了面向服务等级的网络流多任务分类方法。该方法是直接进行面向Co...在网络流分类实践中,网络运营商通常只需要知道网络流所需的服务类别(class of service,CoS),就可对网络流优先级和资源分配做出决定。为了满足用户对体验质量的需求,提出了面向服务等级的网络流多任务分类方法。该方法是直接进行面向CoS的流分类,而不需要推断应用类型。同时提出多任务框架,利用领域知识定义宏特征组及应用合作博弈中的Shapley Value模型来合理分析特征,并用决策树分箱来解决CoS阈值划分问题。采用真实网络数据集进行实验,通过在少量标记数据的情况下,优化网络参数和调整各网络模型时间损耗和分类准确性的稳定相关系数。结果表明,该方法分类准确度(提高了12.66%)和时间消耗(减少了39.23%)性能优于现有文献方法,同时分析了多分类实验结果并给出有关建议。展开更多
In wireless multimedia communications, it is extremely difficult to derive general end-to-end capacity results because of decentralized packet scheduling and the interference between communicating nodes. In this paper...In wireless multimedia communications, it is extremely difficult to derive general end-to-end capacity results because of decentralized packet scheduling and the interference between communicating nodes. In this paper, we present a state-based channel capacity perception scheme to provide statistical Quality-of-Service (QoS) guarantees under a medium or high traffic load for IEEE 802.11 wireless multi-hop networks. The proposed scheme first perceives the state of the wireless link from the MAC retransmission information and extends this information to calculate the wireless channel capacity, particularly under a saturated traffic load, on the basis of the interference among flows and the link state in the wireless multi-hop networks. Finally, the adaptive optimal control algorithm allocates a network resource and forwards the data packet by taking into consideration the channel capacity deployments in multi-terminal or multi-hop mesh networks. Extensive computer simulations demonstrate that the proposed scheme can achieve better performance in terms of packet delivery ratio and network throughput compared to the existing capacity prediction schemes.展开更多
A particle filtering based AutoRegressive (AR) channel prediction model is presented for cognitive radio systems. Firstly, this paper introduces the particle filtering and the system model. Secondly, the AR model of o...A particle filtering based AutoRegressive (AR) channel prediction model is presented for cognitive radio systems. Firstly, this paper introduces the particle filtering and the system model. Secondly, the AR model of order p is used to approximate the flat Rayleigh fading channels; its stability is discussed, and an algorithm for solving the AR model parameters is also given. Finally, an AR channel prediction model based on particle filtering and second-order AR model is presented. Simulation results show that the performance of the proposed AR channel prediction model based on particle filtering is better than that of Kalman filtering.展开更多
文摘本文提出了一种将特征生成和长短期记忆(long short term memory, LSTM)模型相结合的网络流量分类方法。该方法采用矩阵乘法特征生成方式,分析对比了不同特征生成方法的分类性能。通过实验比较了原数据和特征数据在分类问题上的准确性,并比较了卷积神经网络(convolutional neural network, CNN)和本文方法用于网络流分类的效果。在统计特征时采用核函数,使其可以适应LSTM输入维度,获得更佳的分类效果。对真实网络流数据的实验结果表明,本文方法在细分类中的准确度可达93.9%,而在粗分类任务中可达99.2%,其性能明显优于现有其他分类方法。
文摘流早期分类对于优化网络管理和确保服务质量(Quality of Service, QoS)至关重要。针对传统流特征在流早期分类中性能较低的问题,在现有研究基础上,提出了两种新的特征:一是通过等距分箱划分包大小等级,计算相邻到达的两个数据包的包大小等级条件频度;二是通过将包大小序列和包到达时间间隔对应相除,得到速率序列,并计算该序列的统计特征作为分类特征。同时,考虑到早期分类的实时性要求,分析了流特征计算的时间复杂性,在特征选择中优化了时间和准确性之间的平衡。另外,针对网络视频流量占比较大的情况,提出了一种层级分类结构;先使用较少的数据包进行non-video/video的二分类,再使用后续的数据包,进行non-videos和videos的细粒度分类。采用随机森林在两个实际网络数据集上进行分类性能测试,并与文献方法进行比较,验证了该方法在快速流量分类中的优越性。
文摘在网络流分类实践中,网络运营商通常只需要知道网络流所需的服务类别(class of service,CoS),就可对网络流优先级和资源分配做出决定。为了满足用户对体验质量的需求,提出了面向服务等级的网络流多任务分类方法。该方法是直接进行面向CoS的流分类,而不需要推断应用类型。同时提出多任务框架,利用领域知识定义宏特征组及应用合作博弈中的Shapley Value模型来合理分析特征,并用决策树分箱来解决CoS阈值划分问题。采用真实网络数据集进行实验,通过在少量标记数据的情况下,优化网络参数和调整各网络模型时间损耗和分类准确性的稳定相关系数。结果表明,该方法分类准确度(提高了12.66%)和时间消耗(减少了39.23%)性能优于现有文献方法,同时分析了多分类实验结果并给出有关建议。
基金supported by the National Natural Science Foundation of China under Grants No.60972038,No.61001077,No.61101105 the Scientific Research Foundation for Nanjing University of Posts and Telecommunications under Grant No.NY211007+2 种基金 the Open Research Fund of National Mobile Communications Research Laboratory,Southeast University under Grant No.2011D05 Specialized Research Fund for the Doctoral Program of Higher Education under Grant No.20113223120002 University Natural Science Research Project of Jiangsu Province under Grant No.11KJB510016
文摘In wireless multimedia communications, it is extremely difficult to derive general end-to-end capacity results because of decentralized packet scheduling and the interference between communicating nodes. In this paper, we present a state-based channel capacity perception scheme to provide statistical Quality-of-Service (QoS) guarantees under a medium or high traffic load for IEEE 802.11 wireless multi-hop networks. The proposed scheme first perceives the state of the wireless link from the MAC retransmission information and extends this information to calculate the wireless channel capacity, particularly under a saturated traffic load, on the basis of the interference among flows and the link state in the wireless multi-hop networks. Finally, the adaptive optimal control algorithm allocates a network resource and forwards the data packet by taking into consideration the channel capacity deployments in multi-terminal or multi-hop mesh networks. Extensive computer simulations demonstrate that the proposed scheme can achieve better performance in terms of packet delivery ratio and network throughput compared to the existing capacity prediction schemes.
基金Supported by National Natural Science Foundation of China (No. 60972038)The Open Research Fund of Na-tional Mobile Communications Research Laboratory, Southeast University (N200911)+3 种基金The Jiangsu Province Universities Natural Science Research Key Grant Project (No. 07KJA51006)ZTE Communications Co., Ltd. (Shenzhen) Huawei Technology Co., Ltd. (Shenzhen)The Research Fund of Nanjing College of Traffic Voca-tional Technology (JY0903)
文摘A particle filtering based AutoRegressive (AR) channel prediction model is presented for cognitive radio systems. Firstly, this paper introduces the particle filtering and the system model. Secondly, the AR model of order p is used to approximate the flat Rayleigh fading channels; its stability is discussed, and an algorithm for solving the AR model parameters is also given. Finally, an AR channel prediction model based on particle filtering and second-order AR model is presented. Simulation results show that the performance of the proposed AR channel prediction model based on particle filtering is better than that of Kalman filtering.
