Hi, Sri,

I've used these "sparse sampling" designs (where the stimuli are presented in silent gaps between scans) for a couple of studies, and I've always just specified the stimulus onset times (using the stim_times flag in 3dDeconvolve) and the HRF function I want to use -- that is, you don't have to tell 3dDeconvolve explicitly that you had a sparse sampling design. For instance, here's an excerpt from a 3dDeconvolve command I ran recently, where the onset times for three conditions (between, target, and within) are convolved with a gamma function to estimate the HRF.

...
-stim_times 1 stimuli/between.1D 'GAM' \
-stim_label 1 between \
-stim_times 2 stimuli/target.1D 'GAM' \
-stim_label 2 target \
-stim_times 3 stimuli/within.1D 'GAM' \
-stim_label 3 within \
...

This is also consistent with the approach used in the literature that I've read -- here are a couple of references that might be helpful:
Guediche, S., Salvata, C., & Blumstein, S. E. (2013). Temporal cortex reflects effects of sentence context on phonetic processing. Journal of Cognitive Neuroscience, 25(5), 706-718.
Myers, E. B. (2007). Dissociable effects of phonetic competition and category typicality in a phonetic categorization task: An fMRI investigation. Neuropsychologia, 45(7), 1463-1473.

Cheers,
Sahil

--
Sahil Luthra, Ph.D. (he/him/his)
Postdoctoral Researcher, Carnegie Mellon University
email: sahil.bamba.luthra@gmail.com
website: sahilluthra.net