Error Modeling in Network Tomography by Sparse Code Shrinkage (SCS) Method

Errors in data measurements for network tomography may cause misleading estimations. This paper presents a novel technique to model these errors by using sparse code shrinkage (SCS) method. SCS is used in the field of image recognition for denoising the image data and we are the first to apply this technique for estimating error free link delays from erroneous link delay data. To make SCS adoptable in network tomography, we have made some changes in the SCS technique such as the use of Non Negative Matrix Factorization (NNMF) instead of independent component analysis (ICA) for the purpose of estimating sparsifying transformation. The estimated (denoised) link delays are compared with the original (error free) link delays based on the data obtained from a laboratory test bed. The simulation results verify the accuracy of the proposed technique.