Undescribed Horrific Abuse, One Victim & Survivor of Many
gmkarl at gmail.com
Thu Nov 10 06:42:34 PST 2022
so i ran it in pdb. i had to fix a couple typo-y things.
the output of the sequence, when passed back into irfft, is almost as
long as the original signal, but the data is quite different.
that's the test failure that is expected.
so, it was an accomplishment to get this far, to get all those parts
in, and now we get to make a test pass, showing it works.
the failure of the output data matching the input likely relates to
both phase offset in the output and the density of the signal content
at higher frequencies than are measured.
we can reduce the high frequency content by changing how the example
data is generated, or we can sample it better by greatly increasing
the recoridng buffer size and reducing the original signal buffer size
a bigger challenge for me is how to compare two signals that have an
unknown phase offset. an alternative would be to calculate what the
phase offset is, it may be the average of all the phase offsets of all
the placed test waves. this has more steps for me to make a mistake
normally, without a gfx-visualisation inhibition, i would just plot
it. the human brain is really good at comparing phase offsets.
i'm thinking it would make sense to try taking the fft of both waves,
and compare their lower frequencies. this is simply enough, and likely
enough to work, to be worth trying.
More information about the cypherpunks