naturerecordists
[Top] [All Lists]

Re: frequency and time resolution

Subject: Re: frequency and time resolution
From:
Date: Fri, 30 Apr 2004 14:58:12 +0200 (MEST)
The nominal frequency resolution of a digital spectrogram is computed as
follows:

resolution =3D  sample rate / FFT size
(it would be 43 Hz =3D 44100 Hz / 1024 in your set-up)

The bit-depth (8, 16 or 24 bit) does not affect the resolution.

However, the above resolution value corresponds to the =93height=94 of the
discrete pixels in the spectrogram image only. The actual physical frequenc=
y
resolution (the analysis bandwidth, as known from the ancient analog Kay
Elemetrics SonaGraph machines) is usually lower. The final bandwidth depend=
s
on the evaluation window (Hamming, Hanning, Rectangle, ... ) and whether
there is any zero padding mode activated. Assuming a Hamming window with no
additional zero-padding, the bandwidth would be about 56 Hz in your case.

Usually, a bandwidth of about 300 Hz (time constant =3D 3ms) is most
appropriate for bird sound analysis (which corresponds to the =93wide=94 se=
tting
in the old SonaGraph machines) because it provides a good compromise betwee=
n
time and frequency resolution. A narrower bandwidth would unfortunately als=
o
decrease the temporal resolution (rapid frequency modulations or short soun=
d
elements cannot be resolved).

The physical time resolution (the time constant) of a spectrogram is the
reciprocal value of the bandwidth (1/56 Hz =3D 18 ms). In digital
spectrograms, the amount of overlap between the consecutive FFT spectra may
further decrease the final temporal resolution. However, large overlaps wil=
l
never increase the final temporal resolution of a spectrogram. As in the ol=
d
analog sound spectrograph machines, the limit is still the reciprocal of th=
e
bandwidth.

Cool Edit does not provide a consistent way to adjust this overlap paramete=
r
(it selects the overlap according to the current zoom stage). So, it would
be difficult to specify the time resolution.

Therefore, I would recommend other programs for spectrogram-based
measurements (and also for exporting or printing spectrogram images). For
instance, the free Avisoft-SASLab Light software could be used instead. Tha=
t
software also indicates the actual time and frequency resolution of the
spectrograms. It is available together with the demonstration version of
Avisoft-SASLab Pro from
http://www.avisoft-saslab.com/downloa_htm (first item).

Regards,
Raimund


> Hi everyone,
>
> I was wondering if someone could help me with the following:
>
> I would like to determine the resolution of my measurements of frequency
> and
> time characterists in CoolEdit. At present, I use wave files at 44.1 kHz,
> 16
> bit and FFT of 1024.
>
> Given the sample rate, bit rate and FFT size, is there a simple formula t=
o
> determine the frequency resolution (Hz) and time resolution (ms)?
>
> Thanks and best wishes to all.
>
> George Sangster
> The Netherlands
>
>
>
>
> "Microphones are not ears,
> Loudspeakers are not birds,
> A listening room is not nature."
> Klas Strandberg
> Yahoo! Groups Links
>
>
>
>=20
>

--
NEU : GMX Internet.FreeDSL
Ab sofort DSL-Tarif ohne Grundgeb=FChr: http://www.gmx.net/dsl



________________________________________________________________________
________________________________________________________________________

<Prev in Thread] Current Thread [Next in Thread>
  • Re: frequency and time resolution, recordingbird <=
Admin

The University of NSW School of Computer and Engineering takes no responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the naturerecordists mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU