Statistical Prediction-Based Traffic Model for Radiation Sensor Networks by Multi-Resolution Analysis and Machine Learning Theory

In conventional network traffic models, packets duration time is described by a heavy-tailed function. However, heavy-tailed functions cannot precisely reflect the characteristic of transmitting rate during an active period in wireless time-variant channel in sensor networks. To obtain a more comprehensive description, we propose a statistical prediction-based (SP) traffic model using Maximal Overlap Discrete Wavelet Transform (MODWT) to make multiresolution analysis of the channel property and Support Vector Machine (SVM) to dynamically capture the instantaneous variation. A learning method is also derived for adapting the accuracy of prediction. Furthermore, an analytical expression is proposed to calculate the aggregate network traffic flow. The simulation results show that the fit of reality is improved by the proposed model.