Design of Graph Filter Based on Transformation Matrix and Spectral Response
Su‐Ling Lee, Chien‐Cheng Tseng · 2021 International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS) · 2021
In this paper, designs of polynomial graph filters based on transformation matrix and spectral response are presented. In the literature, there are two methods to describe the given filter specifications. One is the desired transformation matrix; the other is the ideal spectral response. This paper aims to study the least-squares (LS) filter designs using these two different specifications and to determine which design gives the better performance. As a result, the transformation matrix method provides smaller approximation error, but it is a graph- dependent design. If graph topology is changed, the filter coefficients need to be re-designed. For the spectral response method, it is a graph-independent design though it has slightly larger approximation error than transformation matrix method. Finally, the real temperature data of sensor network is used to evaluate the performance of these two design methods in terms of the approximation error.