Proactive management of MPEG traffic in ATM networks using time sequenced RLS filters

The capability of predicting VBR traffic can significantly improve the effectiveness of numerous management tasks such as dynamic bandwidth allocation and congestion control. This work demonstrates the applicability of time-sequenced adaptive filters in linear prediction of statistically multiplexed MPEG (Motion Pictures Experts Group) VBR (variable bit rate) video traffic. Time-sequenced adaptive filters allow for the cyclostationary nature of the input by periodically changing the filter and adaptation parameters. This predictor set up is ideal for predicting multiplexed MPEG traffic which has periodically recurring statistical properties and can be considered as a concatenation of PAR (periodic autoregressive) cyclostationary processes. The viability of the approach is illustrated through computer simulations. A number of half-hour long empirical MPEG-1 traces are multiplexed and, subsequently, the aggregated traffic is predicted. The RLS (recursive least squares) algorithm is used for adaptation. The results indicate that the RLS algorithm clearly outperforms the conventional LMS (least mean square) adaptive algorithm in terms of convergence speed and steady-state mean-square prediction error, and, hence, is a more suitable candidate for such an application.