Empty noise recordings - importance, substitute?

Discussions
1

Hi Brainstorm,

I have some data from a site at which empty noise recordings for the noise covariance matrix have not been routinely collected.

I’m using identity matrices for now; it’s sleep data so there’s no clear baseline period. I’m wondering:

  1. If there have been any formal (or informal) comparisons of analyses done with and without empty room recordings (so I can show them it’s useful :slight_smile: )? Do we know how much difference does it make to data quality? (I suppose I could try some variations with the tutorial data if not…)

2 ) How stable the noise covariance matrices are over days/weeks on the same instrument? I ask because they might have made empty room recordings 1-2 x week that I could use (not sure if recording parameters are identical yet).

  1. Are there any cases in which you might want to use a noise covariance matrix based on a baseline condition specifically to help compare conditions, or is that not a sensible idea? For example, if I made a matrix based on stage 2 sleep and used it on REM sleep, would that be helpful to reveal REM-specific activity or would it be better to use an identify matrix on both and then compare their frequency content or whatever in other ways? (I realize this likely depends on research questions, I just want to know if it makes any sense!)

Emily

2

Hi Emily,

  1. If there have been any formal (or informal) comparisons of analyses done with and without empty room recordings (so I can show them it's useful :slight_smile: )? Do we know how much difference does it make to data quality? (I suppose I could try some variations with the tutorial data if not...)

I am not aware of any formal study on this topic. The most detailed piece of information I am aware of is the Brainstorm tutorial page: http://neuroimage.usc.edu/brainstorm/Tutorials/NoiseCovariance#Variations_on_how_to_estimate_sample_noise_covariance

How stable the noise covariance matrices are over days/weeks on the same instrument? I ask because they might have made empty room recordings 1-2 x week that I could use (not sure if recording parameters are identical yet).

It depends on the system. A CTF system in a stable environment would show very little difference within a week. For an Elekta system in a noisy environment that requires sensors to be tuned everyday, it may change a lot from day to day.

  1. Are there any cases in which you might want to use a noise covariance matrix based on a baseline condition specifically to help compare conditions, or is that not a sensible idea?

I'm not sure. Maybe Sylvain, Beth, Richard or John have additional advices?

Francois