Reading presentations from existing tool providers I noticed that in order to detect peaks they first use bunching (average N points) and then use slope and curvature to detect peaks. I'm guessing they use bunching to get rid of high-frequency noise.
Aren't there better alternatives like removing high frequencies with FT? I'm concerned that bunching requires peak width to be known to define Bunching Factor. Which means we first need to detect peak width in some other way before using curvature. But I've seen more than 1 presentation talking about bunching, so there have to be reasons to use this approach instead of others.