Try the idea of Bai et al. for the stabilization

Bai et al. have in their paper the idea of separating small and big scales in the calculation of cgr