Prescribing and measuring SNR

[garrettwrong]

During our Flatiron meeting, it sounded like they had some interest in studying SNR (measurement). A couple ideas came up relating to ASPIRE.

Currently we rely on users to compute their own noise variance, and understand how that relates to SNR in the case of Simulation. I think we should provide users an easy way to create experiments with a prescribed SNR (expressed in SNR). The question I have is how that would fit in with our existing tools/filters in a neat way. (We add noise, by adding a filter, with a noise var; this is already obscure for my taste, but I guess most people would just copy pasta our examples...)

I also think for a given Simulation, we should be able to return the actual SNR as part of the Simulation API.

Our own post docs have asked me for both of the above before as well.

Another topic that came up is measurement of experimental data's SNR in our domain. This is a more complicated and interesting question.

Additionally, I have long believed generating simulations with more realistic noise would be very beneficial to ASPIRE and algo devs. Understanding and providing additional methods to generate more sophisticated/realistic noise distributions might be worth looking into. I think at least one of our researchers and Amit were looking into this relatively recently.