plot fails with -Inf

I need to plot a fft of an audio file. It appears that if any of the fft bins are zero, the conversion to dB produces an -Inf. Then the plot function does draw the figure but does not plot any data.

Is there an efficent way to trap the 0 value bins in the conversion to dB? For my work -Inf PSD would be equiv to zero.

The test audio file was a code generated audio file, 100Hz Sine for 10 seconds. Since only one fft bin will have a value, all the other bins are zero.

Thanks.

%Code Fragment:

[yS, Fs] = audioread(PathFN,'double');

L = length(yS);

Y = fft(yS,L);

% Version One

Px = Y.*conj(Y); % PSD Power of each freq components

Px = Px(1:L/2);

Px = 10*log10(Px); % Convert to dB

fVals = Fs * (0:(L/2)-1)/L;

plot(fVals, Px); % Plot

[Peng Li]

log(0) = -Inf so I don't personally understand why you think -Inf PSD should be zero.

If you want to show the log(0) part in figure, you can add a small offset to Px such as Px = 10*log10(Px + eps).

[Jonathan]

Hi, thank you for the response. The problem is that the plot function seems to failing because some of the y data is -inf

[Peng Li]

did you try my solution? If you add a small number like eps to Px, there won't be any -inf.

Answer 1

[Stijn Haenen]

you can use Px(isinf(abs(Px)))=0; to set all -Inf values to 0

[Jonathan]

Hi, thank you for your response. Your suggestion does catch the -inf values. However, the plot still does not work. If I take out Px = 10*log10(Px); then it will plot. Any ideas? Thanks.

[Stijn Haenen]

did you place it after the convertion to dB?

like this:

Px = Y.*conj(Y);              % PSD Power of each freq components
Px = Px(1:L/2);
Px = 10*log10(Px);            % Convert to dB
Px(isinf(abs(Px)))=0;
fVals = Fs * (0:(L/2)-1)/L;
plot(fVals, Px);         % Plot

Does your Px also contains non real values?

[Jonathan]

Yes, it was placed there and no just real values. aside from a typo, it does plot now. The single positive is at 100Hz as expected. However, it plots also at every 40Hz a spectra that is at/about -250db for the full range of Fs/2 except where the -Inf were. I didn't think it possible to have a negative PSD. Round off/precision errors from the program that generated the audio file? Thanks.