Hi Jim, It sounds to me like you want to high-pass filter the signal and then compare the output of the filter to a threshold value. That will work as long as the noise/general variation cannot have spikes as large as the jumps you are looking for. If that is not the case, then you could look at a bandpass filter of some kind. In addition to the typical types (like Butterworth) you might try a kind of matched filter where you convolve the signal with a step function prototype or maybe an exponential approach (i.e., a low-pass filtered step function a.k.a. RC circuit response to a step function). Sean On Fri, Dec 13, 2013 at 12:50 PM, wrote: > > All, > > I want to do some analysis on several data sets onsisting of 32768 data > points per file. > Each file consists of a signal value that has some average value, and > varies from the average value > versus time. From this information, one can determine the direction > and magnitude of each datapoint > but subtracting the average from it (or adding, whichever is > appropriate). > > The next scenario is similar, but with a slight difference. The data > from out DUT is going along > just as it did in the scenario above for some period of time. Then, > all of a sudden, there is a step > function whering the data central value drops some arbitrary magnitude, > stays there for a while, then > returns to the previous level and continues on as if nothing had > happened. > > This step may happen once, it may happen several time, or it ma not > happen at all. > > With all of that said, here is the question... > > Is there a way to determine, using mathematical methods, if this step > dunction occured in a given file > without actually graphing the file and physically looking at it? > > I have about 1000 files to check for this phenomenon, and I would like > to do some sort of mathematical > analysis on it to fine files that have signs of this step function > happening rather than graphing each > and every file, only to find most of them don't show the problem. > > What I want to do is do some dort of test on each data point and see if > here is a large step from one > point to the next, and if so, how long (in sample periods) does it take > to get to the new steady value. > Then test to see how many sample periods it stays at that new level. > And finally, how fast does it rise > back to the previous steady level. > > I've tried the usual statistical functions, but I can't seem to get a > reliable indication. > Does anyone have any idea? If so, please enlighten me. > > > 2 4 > . . . . ... V V . . . . > 1> .... . . . . . . . .... . . .. . . . . <5 > . . . . > . . > . . . > 3> . . . . > . . > > I have included ere a very primitive drawing of what I want to find. > If this drawing comes through intact, you will see an average value, > denoted by number 1 on the X axis. > > The at point 2, we see a sudden drop to a new average value at point 3. > The signal stays at this level for a period of time (Varies between > occurances), > and then goes back to the previous level at point 4, and stays at that > level for > another period of time (like at point 5). > > The step may occur again in the same file one or more times, or it may > not > happen again in that file at all. > > So if anyone has any ideas as to how to detect this mathematically, I'd > appreciate > hearing how you propose to do it. > > P.S. This is not math homework, and I am not a student wanting help > with my homework. > I am doing this as part of my employment. > > > > Thanks and Regards, > > Jim > > > -- > http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive > View/change your membership options at > http://mailman.mit.edu/mailman/listinfo/piclist > --=20 http://www.piclist.com/techref/piclist PIC/SX FAQ & list archive View/change your membership options at http://mailman.mit.edu/mailman/listinfo/piclist .