Synopsis
		sffxynpre < Fin.rsf > Fou.rsf verb=n sym=n opt=y fs=n Nfrac=6.0 Ntimes=5.0 pow=0.5 niter=100 n1win=n1 n2win=nx n3win=ny r1=0.5 r2=0.5 r3=0.5 mode=0 nsx=2 nsy=2 rect#=(1,1,...)

with frequency-axis smoothness constraint
with frequency-dependent smoothing

Reference: Wang et al. (2021), Non-stationary predictive filtering for seismic random noise suppression - A tutorial, Geophysics, 86(3), W21-W30.

Parameters
		float Nfrac=6.0 frequency-dependent smoothing starts from 1/Nfrac * (Nyquist frequency)   float Ntimes=5.0 Maximum smoothing radius is Ntimes*(reference smoothing radius)   bool fs=n [y/n] y, determine frequency-dependent smoothing   int mode=0 predictive filtering mode; default: non-stationary   int n1win=n1 first window length   int n2win=nx second window length   int n3win=ny second window length   int niter=100 number of iterations   int nsx=2 number of shifts in non-causal prediction filtering   int nsy=2 number of shifts in non-causal prediction filtering   bool opt=y [y/n] y, determine optimal size for efficiency   float pow=0.5 fraction parameter   float r1=0.5 first overlapping ratio   float r2=0.5 second overlapping ratio   float r3=0.5 third overlapping ratio   int rect#=(1,1,...) smoothing radius on #-th axis   bool sym=n [y/n] y, symmetric scaling for Hermitian FFT   bool verb=n [y/n] Verbosity flag