nafflib problems about at HOT 4 CLOSED

Tried the default example from calcnaff and got the same phases each time. Also tried it on some BPM data. I couldn't repeat the problem for AT2.0. nafflib.c revision date 25/7/2017.

You can use it for just real signals and the phases be also be correct (taking only the positive frequencies). However for the amplitude you have to sum the positive and negative frequency amplitudes to get it right. Or you can simply multiply by a factor of 2 if the frequency is not zero or fs/2.

from at.

Dear @carmignani,
I cannot reproduced your bug.

If you run the nafflib algorithm with a real signal, its works well for me.
This is a simple example :

t = linspace(1,2pi20,1000);

[frequency amplitude phase] = nafflib(sin(t),sin(t)0, 1, 5, 1)
*** NAFF results ***
NFS = 2
AMPL= 4.207309e-01+i 2.701499e-01 abs(AMPL)= 4.999955e-01 arg(AMPL)= 5.707992e-01 FREQ=-1.247885e-01
AMPL= 4.207355e-01+i*-2.701512e-01 abs(AMPL)= 5.000000e-01 arg(AMPL)=-5.707963e-01 FREQ= 1.247885e-01

The phase is stable and the amplitude of each harmonic is 0.5 as expected.

from at.

Nevertheless, I suspect something wrong the in the c-library, since sometimes, the function returns NaNs for valid input signal.
If I rerun again the same command with the same input argument the results is then fine.
I suspect something not properly defined with the dynamic memory allocation in the code.

I need to run the code in debugger to understand. This is quite annoying and time consuming to debug.
Any advice is very welcome.

from at.

I get exactly the same output running the example you wrote, but if I run it many times sometimes I get nans.
I also suspect there are some problems in the dynamic memory allocation in the C code.

from at.