I once had to do something sort of like this for a coding challenge. I ended up using the RMS Point-To-Point function to perform a moving RMS measurement. From the RMS measurement, you can detect when the change in amplitude happens.
Correction: It was the AC & DC Estimator PtByPt VI
How about peak detection? When you detect the peak has changed a substantial amount, you know the amplitude as stepped up.
Attach your VI's with the data saved as either a control with default data or a constant.
What do you mean that "peak detection no go". What have you tried?
RMS is doing averaging. The question then is how much of a window are you averaging? You are seeing the effects of the change in amplitude in your graph where the RMS value is beginning to increase. Since you are averaging over a larger window, the effect is taking some time to come into play. If you look for the point where the slope of this graph is beginning to increase, that is where you peaks have changed.
Try this. You can see in your output waveform that you have two stair steps. An easy way to detect those stair steps is with a derivative. As your slope changes from 0, you will notice a change in your derivative. I ran the Pt by Pt Derivative function on the output of the Pt by Pt RMS and was able to see two distinct spikes where your RMS is changing. You can run a threshold (just a simple >= function) or a Peak Detector to detect a change in RMS of that magnitiude. You can then run some other code based on whether this value is true or not.
This isn't a perfect solution, but it should get your closer to your goal. You will probably need to run some additional filtering to get some really clean peaks. The Peak Detector can handle some of that for you.
How much noise does your real signal have?
How many cycles (minimum) will occur at each level? How much does the frequency vary?