[VoxBo] 1/f options for an ER design

[Daniel Kimberg]

Hi Malathi.  I can't claim to have the authoritative answer on this, but here are some thoughts.

Re: option 1, the stock 1/f parameters that come with VoxBo were estimated from data collected on an old 1.5T GE scanner, and you probably don't want to use them.

Re: options 2-4, even for blocked data it's probably best not to fit the 1/f to the actual data you're modeling, and with an event-related design, you probably have no chance of filtering out the task frequencies before doing the fit.  So these are probably no good as well, unless it does turn out that some or all of your runs have the task frequencies concentrated into a small number of frequencies.

Your best option would be to to have a separate run from which to estimate the noise spectrum.  I assume if you had that you wouldn't be posting.  But a decent second-best option would be to borrow the 1/f parameters from someone who's collected data on the same scanner and who has an appropriate run from which the do the noise estimate.  This is essentially what you'd be doing if you had an appropriate stock 1/f.

The last thing I should mention is that the last time I checked, the noise spectrum from HUP6 was pretty flat beyond the lowest frequencies.  You're probably not giving up too much if you filter out the lowest frequencies (exact number depending on the details, but probably 2-3), beyond which mis-specifying the noise spectrum might not be as bad as it would be if the 1/f noise were spread across the whole spectrum.  Perhaps someone who's collected data on HUP6 more recently can corroborate or refute this observation.

dan