# Convolution of NIRS signal

**URL:** <https://mne.discourse.group/t/convolution-of-nirs-signal/11319>\
**Category:** Support & Discussions\
**Tags:** fnirs\
**Created:** [June 25, 2025, 1:08pm UTC](https://mne.discourse.group/t/convolution-of-nirs-signal/11319 "2025-06-25T13:08:27Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Zanchi](https://avatars.discourse-cdn.com/v4/letter/z/b19c9b/32.png) [@Zanchi](https://mne.discourse.group/u/Zanchi)\
**Post date:** [June 25, 2025, 1:08pm UTC](https://mne.discourse.group/t/convolution-of-nirs-signal/11319/1 "2025-06-25T13:08:27Z")

</div>

Hi,  
I am analyzing data related to Freezing of Gait (FoG) episodes in Parkinson’s disease. Based on video recordings, we were able to identify the precise onset and offset times of each FoG event.  
When I plot the HbO signal (without applying any HRF convolution), I notice that the end of the FoG event — as determined from video — generally does not match the point where HbO begins to return to baseline.  
To address this, I tried convolving the event timings (onset and duration for each event) using the `'spm'` HRF model during design matrix creation in `mne_nirs`. However, after convolution, the first FoG event (which originally lasted 11 seconds) appears to span ~45 seconds in the convolved regressor.  
My question is:  
When using `hrf_model='spm'` in `mne_nirs`, should I input zero-duration (impulse) events, or is it appropriate to specify the actual event durations derived from video and clinical observations?