FPGA Central

Please excuse this beginner question.

I have about a minute's worth of sound data taken by a microphone and
recorded into a file.

The sound field has one frequency of interest and some random noise (at
a lower magnitude than the random noise). My sample rate is about 6x the
frequency of interest.

I want to show the change in the frequency of interest alone.

I am thinking to perform an FFT on segments of the data, say, one
second's worth at a time, and record the peaks at the frequency, ending
up with a rough plot of time vs magnitude.

Is there a name for this kind of plot?

Also, any recommendations on whether I should overlap the segments and
whether to use a particular window function?

Thanks in advance for any suggestions,

Amanda

Amanda Robin wrote:
> Please excuse this beginner question.
>
> I have about a minute's worth of sound data taken by a microphone and
> recorded into a file.
>
> The sound field has one frequency of interest and some random noise (at
> a lower magnitude than the random noise). My sample rate is about 6x the
> frequency of interest.
>
> I want to show the change in the frequency of interest alone.
>
> I am thinking to perform an FFT on segments of the data, say, one
> second's worth at a time, and record the peaks at the frequency, ending
> up with a rough plot of time vs magnitude.
>
> Is there a name for this kind of plot?
>
> Also, any recommendations on whether I should overlap the segments and
> whether to use a particular window function?
>
> Thanks in advance for any suggestions,
>
> Amanda

If you plot all of the FFT bins as a row for each time, forming a 2D
image, it's called a spectrogram. It is common to use a data window and
a 50% overlap. If you have Matlab, type 'help spectrogram' for details.

John

John

Or specgram"....depending on the MATLAB version.

~ Snigdha L


john wrote:
> Amanda Robin wrote:
> > Please excuse this beginner question.
> >
> > I have about a minute's worth of sound data taken by a microphone and
> > recorded into a file.
> >
> > The sound field has one frequency of interest and some random noise (at
> > a lower magnitude than the random noise). My sample rate is about 6x the
> > frequency of interest.
> >
> > I want to show the change in the frequency of interest alone.
> >
> > I am thinking to perform an FFT on segments of the data, say, one
> > second's worth at a time, and record the peaks at the frequency, ending
> > up with a rough plot of time vs magnitude.
> >
> > Is there a name for this kind of plot?
> >
> > Also, any recommendations on whether I should overlap the segments and
> > whether to use a particular window function?
> >
> > Thanks in advance for any suggestions,
> >
> > Amanda
>
> If you plot all of the FFT bins as a row for each time, forming a 2D
> image, it's called a spectrogram. It is common to use a data window and
> a 50% overlap. If you have Matlab, type 'help spectrogram' for details.
>
> John
>
> John

Amanda Robin wrote:
> Please excuse this beginner question.
>
> I have about a minute's worth of sound data taken by a microphone and
> recorded into a file.
>
> The sound field has one frequency of interest and some random noise (at
> a lower magnitude than the random noise). My sample rate is about 6x the
> frequency of interest.

How much lower is the noise? Is any of it at a closely adjacent
frequency to you signal of interest?

> I want to show the change in the frequency of interest alone.

What are your resolution needs for the frequency of your signal
of interest and the changes in frequency.

> I am thinking to perform an FFT on segments of the data, say, one
> second's worth at a time, and record the peaks at the frequency, ending
> up with a rough plot of time vs magnitude.
>
> Is there a name for this kind of plot?

Spectrogram.

> Also, any recommendations on whether I should overlap the segments and
> whether to use a particular window function?

If the noise and adjacent interference are low enough, you might
try phase vocoder analysis. This can give a much higher frequency
resolution than by using a non-interpolated FFT alone. Maybe
try a simple von Hann window and a 50% overlap for starters.
Narrower windows and more overlap seems to tracks larger frequency
excursions better, wider windows and less overlap oftens shows
more frequency resolution for more stable sinusoids.


IMHO. YMMV.
--
rhn A.T nicholson d.0.t C-o-M

Ron N. wrote:
> Amanda Robin wrote:
> > Please excuse this beginner question.
> >
> > I have about a minute's worth of sound data taken by a microphone and
> > recorded into a file.
> >
> > The sound field has one frequency of interest and some random noise (at
> > a lower magnitude than the random noise). My sample rate is about 6x the
> > frequency of interest.
>
> How much lower is the noise? Is any of it at a closely adjacent
> frequency to you signal of interest?
>
> > I want to show the change in the frequency of interest alone.
>
> What are your resolution needs for the frequency of your signal
> of interest and the changes in frequency.
>
> > I am thinking to perform an FFT on segments of the data, say, one
> > second's worth at a time, and record the peaks at the frequency, ending
> > up with a rough plot of time vs magnitude.
> >
> > Is there a name for this kind of plot?
>
> Spectrogram.
>

Yeah, but technically a spectrogram is a 2-D plot (image). What the OP
described is a 1-D plot of the row peaks in a spectrogram.

John