Network State Estimation by Spectral Analysis of Passively Measured TCP Flows

This paper presents a novel method for estimating the states of upstream networks by analyzing TCP flows at key traffic aggregation points in home and enterprise environments. Utilizing a machine learning framework that incorporates frequency domain features from RTT time series, we conducted a preliminary evaluation on a simulated virtual network. The results demonstrate the method’s capability to accurately classify the network state of a single flow within a basic model. Our findings suggest the method’s potential for broader application in network state estimation.