FFT producing incorrect results about jsyn

[salk31]

Hello, I was curious about this one so I compared the FFT in jsyn to the output from Apache commons and the DFT from https://www.nayuki.io/page/how-to-implement-the-discrete-fourier-transform and those two match to many decimal places while not obvious to spot a match with the jsyn FFT. Is the jsyn version doing something special or should they match? I'm up for trying to fix if any interest.

[salk31]

So only six years after you said it, I can now only see a problem with the scaling. If I effectively disable it (-1 for sign and double the output) all three implementations match. I'm not sure what the sign or the scaling is for. Looking at places it is called one just has a TODO next to it "transform(1, n, ar, ai); // TODO -1 or 1"

[philburk]

The scaling for the FFT and IFFT is kind of arbitrary.
I have seen lots of different approaches.
We want the IFFT(FFT) to return the original signal.
JSyn scales the FFT by 2/n and IFFT by 1/2. code

[philburk]

James was seeing smearing across bins, which I am not seeing. But I am seeing errors in scaling.