An understanding of the traffic characteristics and accurate traffic models are necessary for the improvement of the capability of wireless networks. In this paper we have analyzed the non-linear dynamical behavour of several real traffic traces collected from wireless testbeds. We have found strong evidence that the wireless traffic is chaotic from our observations. That is we found from the traffic correlation dimension, largest Lyapunov exponent and the principal components analysis, which are typical indicators of chaotic traffic. This gives us the good theoretical basis for the analysis and modeling of wireless traffic using Chaos Theory.
We study the dynamic allocation of bandwidth for video traffic in wireless networks. Our approach consists of two stages. In the first stage, we apply the FARIMA (Fractional Autoregressive Integrated Moving Average) models to forecast traffic based on online traffic measurements. In the second stage, we use the forecast results to allocate bandwidth dynamically. We evaluate our FARIMA-based scheme by comparing it with the ARIMA-based and the static schemes in terms of packet loss probability, queue length and bandwidth utilization. Through the experiments with real traffic traces, we demonstrate our approach works well for highly fluctuating traffic in WiFi.