problem on wavelet analysis, combine channel

Nathan Weisz weisz at LYON.INSERM.FR
Fri Mar 30 16:23:47 CEST 2007


hi eve,

>     Thank you so much. But I am 80 percent sure it's not a common
> memory problem.

this leaves 20% uncertainty :-)

taking jans' formula:
> -(numberoftrials)x(numberofchannelcombinations)x
> (numberoffrequencypoints)x
> (numberoftimepoints)

and assuming for simlicity you have 100 trials (... 302 channels,
right).
 > ((100*90600*27*21)*16)/c(10^9) #90600 equals to number of all
pairwise channel combinations
[1] 82.19232

1GB RAM is not much help here for such a beast (and probably the
lower limit when doing MEG-analysis), if i have no fundamental
misunderstanding here.
do you really need 'powandcsd'? why don't you first do the 'pow'-
thing, look what's interesting (perhaps some stats) and then redo the
freqanalysis with 'powandcsd' on time-frequency points of interest,
taking account of your RAM.

all the best,
nathan



More information about the fieldtrip mailing list