Support Vector Regression Technique for Multimedia Quality Control in Multicast Networks

Multimedia quality control has become an important research field especially with the increased QoS requirements. In this paper, we propose a novel measurement and machine learning mixed approach for real-time multimedia quality control for multicast networks. Most of current adaptive multimedia quality control schemes are based on measurements of available network resources (whether active or passive measurements) or the use of network traffic models and they are reactive in nature. The proposed scheme is based on statistical learning techniques, such as Support Vector Regression (SVR), mixed with network measurements to predict packet loss rate in the near future and thus is a pro-active scheme. The prediction scheme triggers the rate control mechanism to pre-adjust the multimedia sending rate before network conditions start deteriorating. We use packet dispersion technique to measure end-to-end queuing delay combined with video display buffer information to train the acquired dataset for prediction in real time. The performance of the proposed scheme is evaluated over a set of network simulations. The results show that it is effective and exhibits superior performance compared to traditional measurement-based schemes.