Waveform smoothing

[snorkack72]

I have an oscilloscope waveform represented by 1000 points. The middle part is a slope which I need to measure. The data points are not monotonic which makes it difficult to find two points on the slope for the measurement. I need to smooth the data enough that it will be monotonic. I have tried 5 and 7 point moving average which are not good enough. What method should be used for smoothing and where can I find equations I can implement without pages of theoretical Greek which I won't understand?

 

[BigBeachBanana]

I have an oscilloscope waveform represented by 1000 points. The middle part is a slope which I need to measure. The data points are not monotonic which makes it difficult to find two points on the slope for the measurement. I need to smooth the data enough that it will be monotonic. I have tried 5 and 7 point moving average which are not good enough. What method should be used for smoothing and where can I find equations I can implement without pages of theoretical Greek which I won't understand?

Try exponential smoothing. Excel data analysis tool has the functionality built-in. Only requires your data input and a damping factor value which is between 0 and 1. The closer to 1 the smoother it is and vice versa.

 

[Deleted member 4993]

I have an oscilloscope waveform represented by 1000 points. The middle part is a slope which I need to measure. The data points are not monotonic which makes it difficult to find two points on the slope for the measurement. I need to smooth the data enough that it will be monotonic. I have tried 5 and 7 point moving average which are not good enough. What method should be used for smoothing and where can I find equations I can implement without pages of theoretical Greek which I won't understand?

Remember that smoothing operation "looses" information. Did you try FFT?

 

[snorkack72]

I found some info. on exponential smoothing. Because each sample is multiplied by a constant less than one the smoothed samples are all going to be lower amplitude. That will make measurements on the smoothed samples incorrect. Doesn't FFT just provide info. on what frequencies make up a waveform? I don't see how that will help. I just need to average in points that fall relatively too far from the curve.

 

[Cubist]

I'd recommend that you try RANSAC to eliminate outlying data points. The following online document contains a good description

http://www.cse.psu.edu/~rtc12/CSE486/lecture15.pdf

And then perhaps follow this up with linear least square fitting on the subset of points identified by RANSAC. You might be able to find a free implementation somewhere. Python seems to have it as a library function.