Need orientation from DSP expert (phase-locked loop?)

What I have is a periodic but non-sinusoidal signal (in this case, the output of a pressure sensor). From the signal, certain features are distinguishable: the features boxed in red might represent one type of event, while those in blue or green represent other events. Assume I have a model that allows me to derive information about these events from the signal and that I can sample at high resolution on an angle basis. What I'm looking for is a mathematical structure with which to pull out information specific to the regions of interest.

**broken link removed**

As an example, say I want to keep a running average of the height of the features in red. A simple implementation, given that I know the angles at which these features should occur, would be to repeatedly subtract the expected minimum-value angle (say, the left edge of each red box) from the expected maximum-value angle (right edge of each red box), add the difference to a circular buffer, and then mean the buffer. However, I believe there should be a more elegant (and possibly more effective) way of doing this.

Could a phase-locked loop be used to accomplish my goal? Would I employ a separate loop for each region of interest? If one of you can just point me in the right direction, I believe I'm capable of learning for myself all that's required to implement this.

If there's another forum in which I should post this question, please let me know.

Not quite clear if you know in advance when each of these boxed features would appear in the signal. If not, you could use a correlator (multiply with expected signals and add up, one for each expected signal) to find out which happens when. This could then trigger appropriate analysis. The value should also tell you grossly how close you are to the expected signal.

However, if the times are known in advance, you have all the samples, you can extract any reqd info from it. A PLL can help you track variations from a certain frequency, and for that the signal has to persist for a certain amount of time for the PLL to "start seeing". It doesnt seem suitable here.
- B