Fourier Transforms?

[donal botkinParticipant]

Anything available to transform data from time-domain to frequency-domain? FFT or DFT?

[davidModerator]

Currently there is no FFT functionality in DataGraph.  A raw DFT as in complex numbers doesn’t fit into DataGraph because it would require a complex data column and operations and adding that is beyond the scope.

But that is not the only thing that could be used. Some might fit into the scope of DataGraph other into ImageTank.  So some questions:

Are you ok with the signal being being length 2^n or 3^n? FFTW is pretty pricey for a non-academic license and it is hard to do a general purpose DFT (which is why it makes sense for FFTW to not be free).  The OS however has the above prime powers already implemented.  Of course a O(N^2) method isn’t slow for smaller numbers of N.

Are you looking for the magnitude vector.  For N values that gives you the magnitude for N/2 first frequencies and really only the first few are “accurate”.  In quotes because they are accurate in the discrete sense but if you look at signal perturbations the higher frequencies are very sensitive.  For this what you really get are two columns, frequency and amplitude, and you need both if you want to draw that properly?

Do you want to use the DFT to filter a signal?  For example apply a smoothing?  So what you need is the action map->scale->inverse map.  And if so what type of scaling expression are you looking for?

Is your signal unformly spaced?  That is are you looking at this as a #-y or x-y data set?

[donal botkinParticipant]

What I’m trying to do is identify sources of variation in voltage data in the range of 5-10VDC. For example, if a device has a response to changes in ambient temperature. At present, my solution is to plot the data and use a combination of non-linear regression and eyeballs to visualize periodicity. The signal is from what is meant to be a precision source, so most of the variation has been designed out or otherwise corrected. The remaining sources of error are few in number and are on the order of a few PPM in a system that should be accurate at a couple of orders of magnitude better.

The data is an #-y in most cases although timestamp could be added to the collection, likewise for ambient temperature. What might be other sources remain in the land of “what I don’t know that I don’t know.” LED lighting in the lab and the quality of wiring and connectors are thought by some experts to be sources of noise.

[Author]

Posts