How do I determine my waveform settling time?

[lockheed_eng]

Hello,

I'm trying to determine my waveform's settling time.   I have attached a jpg of a sample waveform.  This is a DC capture of the output of a UUT when the load is stepped from 1.5A to 2A.  I need to determine programmatically when the settling time is from this step.  Unfortunately, there is a lot of ripple on the signal which interferes with calculating the settling time. 

Can anyone help with this?

[johnsold]

Reece,

It looks like your data may be a first order response (1 - e^x) plus the spike plus the high frequency signal. If the high frequency component is stable, perhaps you could extract it and then subtract it. That would leave just the transient. Fitting a first order response curve to the data and obtaining the parameters from the fitted curve can also be done. If the system has a higher order response, the fitting might still work bit the period of the oveshoot may be too close to that of the high frequency signal for good results.

How do you define "settling time" for your purposes?

Lynn

[lockheed_eng]

Thanks for your reply, Lynn

I think I understand what you're saying about subtracting out the "high freq" component, but I don't follow you concerning the "Fitting a first order response curve to the data and obtaining the parameters from the fitted curve can also be done."  How is that accomplished?

As far as what is considered settling time, that should be a "stable DC value".  Once the "ripple" is removed, a DC value +/- ~50mV for a period of 20mS would be acceptable.  Those numbers are arbitrary, tho, and can be adjusted if necessary within reason.  I think that if we were to draw a trend line from the average of the values shown on that graph, the DC component left will be considered "stable" enough.

[johnsold]

Another idea: Using a trigger (or detect the spike) and take a baseline sample set, perhaps about 5 ms of data starting 10 ms before the trigger and take a "final value" set starting 10-20 ms after the trigger. The mean values of those samples could be compared to the cycle by cycle value mean (over one cycle of the high frequency signal) to see when it settles.

There are curve fitting routines in LV. I don't know how well they would work for fitting a 1-e^x equation to the data.

Lynn