Detecting Network State in the Presence of Varying Levels of Congestion

We consider the problem of estimating the state of computer networks which are delivering video in the presence of other interfering services. Existing methods for measuring jitter, a Quality of Delivery (QoD) measure for video, are based on statically configured IIR filters. They do not attempt to estimate the congestion in the network that caused the jitter to change. As a result, steps taken to improve QoD are frequently taken blindly. We pose the problem of estimating jitter as the problem of estimating a target source in the presence of interfering sources. To evaluate the approach we capture QoD measurements for a target video client from a six router networking test-bed where video is delivered over a substrate which is shared with varying levels of interfering sources which cause congestion. We demonstrate the performance of the new jitter estimator as part of a background congestion level detector. Numerical results based on real data show that considerable gains in congestion state classification are achieved for all congestion levels.