Novel Extended Kalman Filter Using Matrix-Based Levenberg-Marquardt Algorithm and Its Application for Variable Bit-Rate Video Frame-Size Prediction

Dynamic bandwidth allocation based on multimedia network-traffic prediction has been emerging as an important problem in multimedia networks. The well-known Kalman filter has been adopted for such network-traffic prediction but it is assumed that the state-transition model is linear and known a priori. Therefore, it is favorable to extend the conventional linear state-transition model to be nonlinear and dynamically estimate it. It is not trivial to estimate such a nonlinear model especially for a multimedia network supporting the 5G technology and operating in a highly mobile environment. In this work, we would like to address the aforementioned challenges by designing a new matrix-based Levenberg-Marquardt algorithm based extended Kalman filter (MLMA-EKF) to dynamically estimate the video frame-sizes in compiance with MPEG-4 specifications. Numerical results over MPEG-4 encoded movies demonstrate that our proposed novel MLMA-EKF frame-size predictor is effective for predicting the future bit rates, or video frame-sizes, in terms of normalized mean square error (NMSE).