Back to the main page.

Bug 1321 - projectmom in ft_sourcedescriptives suboptimal for time domain methods?

Reported 2012-02-09 11:24:00 +0100
Modified 2018-11-12 10:56:41 +0100
Product: FieldTrip
Component: core
Version: unspecified
Hardware: PC
Operating System: Mac OS
Importance: P3 normal
Assigned to: Jan-Mathijs Schoffelen
Depends on:
See also:

Jan-Mathijs Schoffelen - 2012-02-09 11:24:30 +0100

ft_sourcedescriptives uses svdfft for the estimation of the optimal orientation, both for frequency domain and time domain source reconstruations. Internally, the csd/covariance is computed, and an svd is performed. As far as the frequency domain is concerned, the csd is correctly estimated from f*f' where f are the 'raw' fourier coefficients. Yet, it is questionable whether the same applies for the covariance, (mom*mom') where mom is the 3D-dipole moment as a function of time. The mom is not necessarily zero mean. Does this yield a sub-optimal covariance estimate? (and thus a sub-optimal estimate of the optimal dipole orientation?). Note: this non-zero mean issue also arises in the computation of the source.avg.cov (mom*mom'), and the determination of optimal orientation in ft_sourceanalysis (cfg.fixedori = 'yes'). I would suggest to adjust svdfft with an optional extra flag which specifies whether the data need to be de-meaned, or to implement a svdmom (probably cleaner, because svdfft is specifically designed for freq-domain analysis and it hacked its way into the time-domain code too).

Robert Oostenveld - 2012-02-09 15:11:52 +0100

Hi JM I just finished discussions with the FIL team on SPM12. There are some relevant plans that also impact fieldtrip, esp. on the source estimation (ft_inverse_xxx) and the post-processing of the estimations. Please postpone any actions until we have discussed this in detail (i.e next week). best Robert

Jan-Mathijs Schoffelen - 2018-10-09 12:38:12 +0200

not for now.

Jan-Mathijs Schoffelen - 2018-11-12 10:56:41 +0100

closing time.