Grouping of like Magnitudes

First off let me say again how much of a help this forum has been to me in my work. I'm here atleast once a day to look something up

I'm writing an algorithm to characterize a waveform of noise from an op amp and after much work I'm at a point where I can fairly accurately give the magnitude and duration of output spikes. Whats needed now is to group like magnitudes into discrete states to show at what states the part is going to.

After a week of work I really don't know how to create my states. This algorithm will be run on thousands of waves but here is a good example.

http://i.imgur.com/qt6oXrO.png

Unfortunately some of the magnitudes are fake. Anything below say 50 pA in the histogram is a mirroring of a positive spike; though not close enough that I can just check for opposite values.

I'm really at a loss where to go from here and I would love some input from you guys.

[Maxwell Roth]

Here are 2 more examples. These have 3 obvious states. One fast and one really slow http://i.imgur.com/Gp43znq.png

http://i.imgur.com/0jTPA9S.png

Answer 1

[Wayne King]

Do you have the Signal Processing Toolbox? Have you seen this function:

https://www.mathworks.com/help/signal/ref/statelevels.html

The above is for a bilevel waveform like a pulse waveform. Do you have more than two states?

[Maxwell Roth]

I have not seen that function; looks promising. The consensus on the first supplied wave is that it has at least 3 states. One at ~260, ~0, ~12. You could go further and say there is also a sate at ~15 and another at ~20 but if I can get the first 3 it'll be good enough use.

EDIT: I played with it and though it does a great job of finding the 2 largest frequencies of magnitude it don't really help past 2 states and I don't know how many states I have when I first read in the data.