Short-term link quality prediction using nonparametric time series analysis

Abstract

Wireless link quality prediction (LQP) is the foundation for proactive operations and is therefore a key technique in alleviating network performance degradation. However, accurate LQP is difficult because of the dynamic nature of wireless environments. A recent study found that fluctuations in intermediate quality links often show dynamics on a sub-second granularity, making the task even more challenging. In order to leverage the intermediate links, as well as fine-tune upper-layer protocols, we propose to use nonparametric modeling in nonlinear time series analysis that predicts short-term link quality online. Unlike existing studies, we do not define any new experimental or hypothetical models, or train models using a set of training data. Functional-coefficient autoregression is employed to predict the link dynamics at high time resolutions. We apply our approach and a local linear regression-based LQP (a typical parametric modeling approach) to both NS-2 simulation and empirical packet traces. The results indicate that the proposed method has much higher prediction accuracy and convergence speed than the local linear regression-based LQP under dynamic network conditions.

摘要

链路质量预测是执行先应式机制以缓解网络性能下降的关键技术. 然而, 无线环境的动态特性导致很难准确预测链路质量变化趋势. 最新研究表明, 中间链路具有高突发性, 且波动程度为亚秒级, 这为链路质量预测带来更为严峻的挑战. 本文提出利用非线性时间序列分析中的非参数建模方法建立预测模型, 在高时间分辨率下实时预测链路质量变化趋势, 为高层协议设计提供支持. 最后, 将基于 NS-2 的仿真实验数据和实测数据作为模型输入, 分析比较本文提出的算法和基于局部线性回归的链路质量预测算法的性能. 结果表明, 本文提出的链路质量预测模型在亚秒级尺度上具有更高的预测精度, 对于链路突发情况, 能够以较快速度收敛.