Measurement matrix for sparse internet data based compressive sampling
Irawati I.D.a, Suksmono A.B.b, Edward I.J.M.b
a Telkom Applied Science School, Telkom University, Bandung, Indonesia
b School of Electrical and Informatics, Institut Teknologi Bandung, Bandung, Indonesia
Abstract
© 2018 IEEE.Compressive Sampling (CS) process consists of three main parts, namely sparse representation, measurement, and sparse reconstruction. The measurement matrix is used for sampling signal from the sparse signal representation process. The success of the reconstruction process is also strongly influenced by the selection of proper measurement matrix. One important factor in CS is the use of well-design measurement matrix. This paper compares the performance of measurement matrices which are Uniform, Normal, Binary, Half-normal, Log-normal, Binomial, Poisson, and Exponential matrix that applied to internet traffic data. At the right Restricted Isometric Constant (RIC), we evaluate the measurement matrix performance using Normalized Mean Square Error (NMSE) and Compression Ratio (CR). The results show that Binomial measurement matrix is superior with the smallest NMSE and the highest CR. This binomial measurement matrix is also capable of overcoming various patterns of data loss on the network.
Author keywords
Compressive sampling,Data loss,Measurement matrix,Normalized mean square error,Reconstruction process,Sparse reconstruction,Sparse representation,Sparse signal representation
Indexed keywords
Compression ratio,Compressive Sampling,Data loss,Internet traffix data,Measurement matrix
Funding details
ACKNOWLEDGMENT This research was funded in part by Telkom Foundation and LPPM ITB. We thank to our colleagues from LTRGM Laboratory who were very helpful to this research.