Abstract: The traffic matrix is one of the crucial inputs in many network planning and traffic engineering tasks, but it is usually impossible to directly measure traffic matrices. So, it is an important research topic to infer traffic matrix by reasonably modeling, and incorporating the limited empirical information. If the proposed methods, Kalman Filtering method is a more efficient and accurate method than many others. However, the error covariance calculation components of the Kalman Filtering arc difficult to implement in realistic systems due to the existence of ill-conditioning problems. The authors proposed Square Root Filtering/Smoothing traffic matrix estimation(SRFsTME) algorithm to improve it, and also proposed a data pre-filtering method to reject the "bad" data with considerable noise. Simulation and actual traffic testing results show that SRFsTME algorithm is more numerical accurate and stable than Kalman Filtering.

Key words: Traffic matrix, OD traffic, Kalman filtering, Square root factorization