rs
At 12.12 14/10/2002 -0500, you wrote:
>Could it be that what you're hearing is the nyquist freq. foldover
>from sounds in excess of 22kHz? I know that digital recorders employ
>brickwall filters at around 20k, but maybe they're less than perfect
>and the amplitude is in excess of the filter's capabilities? Hence
>you would hear artifacts resulting from the difference of the
>original frequency minus 1/2 the sampling freq. e.g., 24kHz-22050Hz
>=3D 3500Hz. Add that to your original signal and you'd end up with some
>interesting inter-modulation distortion.
Most DAT recorders and sound boards for PCs have relatively smooth filters=
because sharp filters are more expensive and may introduce phase errors
close to the cutoff frequency.
If sampling at 48kHz the Nyquist frequency is 24 kHz and most filters begin=
to attenuate above 21-22 kHz. At the Nyquist frequency some filters
attenuate only 10-20 dB and have their zero at 28 or more kHz.
This is not dramatic while recording normal musical programs where acoustic=
energy falling above the Nyquist frequency is relatively low and the
artifacts they eventually produce fall in a frequency range where our ears=
are not too sensitive.
It may become a problem if recording synthetizers or nature sounds with
high energy contents at high frequency (dolphin whistles, some bird songs,=
some insects).
In any case, if we only consider frequencies below the sampling rate, the
frequency range contaminated by the aliasing should be calculated as
Nyquist-(F-Nyquist). This means that a 30 kHz tone could appear as an
artifact (alias) at 18 kHz =3D24-(30-24).
If the filter of your DAT completely cuts frequencies above 28 kHz, a tone=
at 30 kHz will never appear in your recording. Though, components between
Nyquist and the zero of your filter will appear as aliases with an
attenuation which will depend on the attenuation of the filter above
Nyquist. If the filters attenuates 26 dB at 26 kHz, then a tone at that
frequency will appear at 22 kHz attenuated by 26 dB.
To check the quality of your filters you simply need a signal generator
sweeping at least up to the sampling rate. Generate a sweep from, say, 10
kHz to the sampling rate of your device. Record the signal at -6 dB. Then
analyze your recording with a spectrum analyzer: you'll see the peak of
your signal moving upward across the spectrum... when the freq will reach
21-22 kHz you'll see your peak attenuate. Then the peak will reach the
Nyquist freq. Measure the height of the peak just before it reaches
Nyquist. Then you'll see the peak moving downward... you're seeing the
aliasing! A peak at 22 kHz will be the alias of the input signal at 26 kHz,=
at 21 kHz it will be the alias of the original signal at 27 kHz and so on=
(if you're playing with a DAT). Follow the peak until it disappear... that=
is the zero of the filter. To plot the attenuation curve of your filter you=
simply have to measure the height of the peak every, say, 500 Hz. Then for=
every point measured on the aliased components you must compute the real
original frequency.
By using that curve, you can asses the frequency and level of the aliases
you may get for every input signal... if your vocalizing animals produce
sounds (including harmonics) which may produce aliasing, either udible or
visible on the spectrograms, it may be the case to switch to a higher
sampling rate!
I hope these notes may help in facing correctly with the aliasing problems.=
..
Gianni
--------------------------------------------------------------
Gianni Pavan
Email
Centro Interdisciplinare di Bioacustica e Ricerche Ambientali
Universita' degli Studi di Pavia
Via Taramelli 24, 27100 PAVIA, ITALIA
Tel/Fax +39-0382-525234 Laboratori
Tel/Fax +39-0382-526208 Segreteria
Web http://www.unipv.it/cibra
________________________________________________________________________
________________________________________________________________________
>From Tue Mar 8 18:22:49 2005
Message: 1
Date: Tue, 15 Oct 2002 09:27:13 -0500
From: Rob Danielson <>
Subject: Re: Digital over mod? /grasshopper distortion
Paul and Walt wrote:
>Paul Dickinson wrote:
>> Could it be that what you're hearing is the nyquist freq. foldover
>> from sounds in excess of 22kHz? I know that digital recorders employ
>> brickwall filters at around 20k, but maybe they're less than perfect
>> and the amplitude is in excess of the filter's capabilities? Hence
>> you would hear artifacts resulting from the difference of the
>> original frequency minus 1/2 the sampling freq. e.g., 24kHz-22050Hz
>> =3D 3500Hz. Add that to your original signal and you'd end up with some
>> interesting inter-modulation distortion.
>
>What I'm finding is a perfectly natural sound, it's out there and the
>mic records it. It's not any kind of digital error. The real world just
>does not play by our sound recording rules.
>
>It is a clear beat frequency between the calls of two or more frogs in
>my case. It's also possible to get this sort of thing if you have a
>strong multipath signal. ie. things out in the environment are
>reflecting signal at you out of phase. I've also heard this from insects
>as well. In both cases I can hear it by ear at the original site.
>
>I've not seen any foldover problems in my gear, even when you overdrive
>digital, it's easily explained by the fact that sampling is discrete
>intervals. When you clip, it's not a single new frequency that's
>produced, it's a whole series in even steps.
>
>We did see some foldover when we were discussing slowing sounds by
>forcing them to play at different sample rates. At least with one model
>of iMac. I don't think we sorted out exactly how it was doing it. Could
>have been a programming error rather than hardware.
>
>Walt
>
Paul and Walt--
All plausible explanations. Now that I think about it, could be even
tweeter irregularities-- responding to such concentrated, hi energy
levels and/or dithering or other side effect from eq/gain boosting.
Many factors seem able to point to chances for peculiarities. Thanks.
Will help my listening.
Best--
Rob D.
________________________________________________________________________
________________________________________________________________________
|