On Sat, Mar 16, 2013 at 1:50 AM, Daniel J Sebald
<address@hidden> wrote:
Below seems to be the fftfilt accuracy issue cropping up again. I seem to recall we went through all these tests and reworked them somewhat by increasing the tolerance value somewhat larger than eps (many multiples) and adding more tests to really check the fftfilt routine beyond the few degenerate cases. None of that seems to be in the fftfilt.m file. Was some changes set lost or reverted along the way? In the repository, the most recent changeset is:
http://hg.savannah.gnu.org/hgweb/octave/file/72c96de7a403/scripts/signal/fftfilt.m
Mon, 02 Jan 2012 14:25:41 -0500 (14 months ago)
It seems like the group examined this since then, sometime in early September of last year according to my old emails.
Dan
processing /usr/local/src/octave/octave-dialog/octave/scripts/signal/fftfilt.m
***** test
r = sqrt (1/2) * (1+i);
b = b*r;
assert (fftfilt (b, x ), r*[1 1 0 0 0 0 0 0 0 0] , eps);
assert (fftfilt (b, r*x), r*r*[1 1 0 0 0 0 0 0 0 0], eps);
assert (fftfilt (b, x.'), r*[1 1 0 0 0 0 0 0 0 0].', eps);
!!!!! test failed
assert (fftfilt (b, r * x),r * r * [1, 1, 0, 0, 0, 0, 0, 0, 0, 0],eps) expected
Columns 1 through 3:
0.00000 + 1.00000i 0.00000 + 1.00000i 0.00000 + 0.00000i
Columns 4 through 6:
0.00000 + 0.00000i 0.00000 + 0.00000i 0.00000 + 0.00000i
Columns 7 through 9:
0.00000 + 0.00000i 0.00000 + 0.00000i 0.00000 + 0.00000i
Column 10:
0.00000 + 0.00000i
but got
Columns 1 through 3:
0.00000 + 1.00000i 0.00000 + 1.00000i 0.00000 - 0.00000i
Columns 4 through 6:
-0.00000 + 0.00000i -0.00000 + 0.00000i 0.00000 + 0.00000i
Columns 7 through 9:
-0.00000 + 0.00000i 0.00000 - 0.00000i -0.00000 + 0.00000i
Column 10:
-0.00000 + 0.00000i
maximum absolute error 2.22478e-16 exceeds tolerance 2.22045e-16
shared variables scalar structure containing the fields:
b =
1 1
x =
1 0 0 0 0 0 0 0 0 0
r = [](0x0)