An Improved GraDe Method for Blind Separation of Graph Signals
Mohammad Hossein Sadeghi, Massoud Babaie‐Zadeh · IEEE Transactions on Signal Processing · 2023
For blind source separation (BSS) of Gaussian graph signals, an algorithm called GraDe (graph decorrelation) has been introduced in [1], [2]. In the current paper, it is shown that GraDe does not achieve a good performance for some types of graphs. This is attributed to the estimation of covariance/autocovariance matrices using signal samples, which may not be reliable. To address this weakness, an improvement based on the spectral representation of the signals is proposed, focusing on removing the impact of the outlier eigenvalues. Numerical simulations show that the proposed method outperforms the original GraDe algorithm.