Dear Jing,
The issue of filter stability has little to do with precision or with
SPM. It would appear for the same combination of filter order,
sampling rate and cutoff at least in any Matlab software, but probably
for any Butterworth filter implementation in general. It is indeed
sensitive to the precise value of the cutoff.
Best,
Vladimir
On Wed, Jun 22, 2011 at 6:12 PM, <[log in to unmask]> wrote:
> Hi Vladimir,
>
> May I ask does the "filter" function of SPM also has the precision
> problems? For example, in my data, if I put "cutoff" value of "high-pass
> filter" as "0.03", "0.05" or "0.1", the "filter" generates matrix with
> "NaN"; however, if I change the "cutoff" value into "1", it indeed
> generates the reasonable filtered data. Is this means the precision for
> the high-pass filter in SPM is "1", or this is only in my data's case?
>
> Many thanks in advance!
>
> Jing
>
>
>
>> Dear Panagiotis,
>>
>> There was indeed a subtle bug, hopefully not affecting most users. SPM
>> only supports sampling rates with precision up to tenth of a Hz,
>> whereas your rate had two digits after the decimal point. Critically,
>> while there was rounding for the the actual resampling, the sampling
>> rate of the new dataset was set to the requested rather than actual
>> rounded rate. This led to drift with respect to events defined in
>> terms of time which was sufficient to create the jitter you found. The
>> problem was especially severe when the requested rate was also
>> fractional which was the case when you divided the original rate by 2.
>>
>> The bug is fixed in the attached function and the fix will be included
>> in the next update.
>>
>> Thanks for your meticulous work,
>>
>> Vladimir
>>
>> On Mon, Jun 20, 2011 at 10:56 AM, Panagiotis Tsiatsis
>> <[log in to unmask]> wrote:
>>> Hi Jing,
>>>
>>> Just a thought...
>>>
>>> Did you downsample a continuous file before epoching? I have noticed
>>> that if
>>> I downsample a continuous (CTF-MEG) file after filtering and before
>>> epoching, the triggers shift in a weird way (up to +/-10ms) even if he
>>> downsampling rate is very close to the original one... (I mean, you
>>> expect a
>>> reasonable jitter in the triggers if you go from 1000Hz to 200Hz, but
>>> not
>>> +/-10ms if you go from 1000Hz to 950Hz for example). I wanted to
>>> investigate this issue further and provide a proper report but I did not
>>> have the time to do it yet... and maybe I got something else wrong.
>>>
>>> Anywayz,
>>> Good Luck :)
>>>
>>>
>>>
>>> On 6/20/2011 11:45 AM, Jing Kan wrote:
>>>>
>>>> Dear SPMers,
>>>>
>>>> Thanks for your concern!
>>>>
>>>> I am dealing with the MEG data pre-processing. After the steps of
>>>> filter,
>>>> epoch, artifact detect, averaging and baseline checking, I realized
>>>> the
>>>> response shifted a few seconds, i.e. the strong response occurred after
>>>> the stimulus shifted into 7 ms before the stimulus.
>>>>
>>>> May I ask is anyone know the reason about this?
>>>>
>>>> Many thanks in advance!
>>>>
>>>> Jing
>>>
>>>
>>> --
>>> Panagiotis S. Tsiatsis
>>> Max Planck Institute for Biogical Cybernetics
>>> Cognitive NeuroImaging Group
>>> Tuebingen, Germany
>>>
>>
>
>
>
|