[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [ARSCLIST] Listening Test - was: Sampling Theory (was Fred Layn's post on the Studer list re: Quantegy)
See comments below preceeded by "EJ>".
-----Original Message-----
From: Association for Recorded Sound Discussion List
[mailto:ARSCLIST@xxxxxxx]On Behalf Of Dave Bradley
Sent: Wednesday, January 19, 2005 5:29 PM
To: ARSCLIST@xxxxxxxxxxxx
Subject: Re: [ARSCLIST] Listening Test - was: Sampling Theory (was Fred
Layn's post on the Studer list re: Quantegy)
Hi Eric,
I will reply to your private e-mail shortly, but for now the public one.
>I know an analog function generator does not produce music,
>but it does produce a very repeatable signal for basic
>listening tests, and as it is analog, it contains all the
>harmonics that can affect timbre.
Just because it is "analog" does not mean it contains harmonics. If you are
playing it through a loudspeaker and testing that output with a microphone,
then there may be harmonics, but if you are simply generating a tone with
the generator and it's direct wired into your testing equipment, then there
won't be harmonics unless the device is designed to create harmonics
intentionally. A sine wave isn't a sine wave if there are harmonics mixed
with it.
EJ> Thank you - you make a good point. A "perfect" analog
EJ> oscillator such as for testing communication equipment is quite
EJ> expensive - I don't have one. Most (all?) audio gear (amps,
EJ> preamps, even signal generators) has some small (or not so
EJ> small) harmonic distortion. We're talking real world signals,
EJ> which is what I wanted to test with. Dollars to donuts, if you
EJ> connect your analog waveform generator to a spectrum analyzer,
EJ> you will see harmonics.
EJ>
EJ> I do have a dedicated audio oscillator, though, and
EJ> measured the first harmonic at -60 dB. Whereas the function
EJ> generator that I used for my test has a first harmonic at
EJ> -32 dB (ie. more harmonic distortion than the oscillator).
EJ>
EJ> You are absolutely correct that a distortionless sine wave
EJ> would have no harmonics. But I think you'd be hard pressed to
EJ> find such a perfect sine wave in the real world of music. For
EJ> that reason, I feel the function generator with its harmonic
EJ> distortion still makes for a valid test, if not a more valid
EJ> test because of the harmonics.
>At 10 kHz, the sine and triangle were not clearly differentiated
>by their sound, but rather by their intensity or volume level.
>The triangle being slightly louder than the sine. The square
>still had a distinct timbre to it that was different from the
>sine and triangle.
Which disproves the issue of a 10 kHz square wave with 44.1 kHz sampling at
16-bits that you had originally written about. For the record, after
receiving some clarification from Eric about the choice of the 10 kHz
frequency after writing to him about a test I did, I tried the test with a
20 kHz sine wave and a 20 kHz square wave (I didn't try this particular
test with a triangular wave). The 20 kHz square wave couldn't be
reproduced by my pro ADC / DAC (RME-Audio PAD 96) at those settings, but
could be at 96 kHz / 24-bit. The sine wave was fine at the lower digital
resolution.
EJ> The 10 kHz square wave through a 44.1 ADC-DAC looks far more
EJ> like a sine wave than a square wave. It's not a perfect sine
EJ> wave, but if you didn't have anything to compare it to, you
EJ> would swear it to be a sine wave. Images from my oscilloscope
EJ> are available to anyone interested - and those images make the
EJ> point pretty clearly that a 10 kHz square wave at 16/44.1
EJ> looks like a sine wave - at least through my system. Has
EJ> anyone else tried this?
>At 16 kHz, I could no longer differentiate the sine and triangle,
>but I could still consistently pick out the square wave by its
>intensity - but not by its timbre.
Yes, but could you hear a difference between the analog and the digital
versions of those waves that you could hear?
EJ> As I stated in my earlier post, that's a test for another day.
>To make things more interesting, we repeated the entire listening
>exercise by monitoring the ADC-DAC output sampled at 88.2 kHz, and
>the results were the same as the analog listening test! There
>may have been subtle differences in loudness and timbre between
>analog and digital, but that wasn't the goal of this test.
Actually, I would have expected that it was part of the goal of the test
since the original hypothesis posted on the list was that a 10 kHz square
wave couldn't be properly sampled at 44.1 kHz 16-bit resolution.
EJ> I am glad that I fulfilled your expectation!
>However, at 44.1 kHz I had much more difficulty differentiating the
>square wave at 10 kHz - but I still consistently could.
Did you try with frequencies slightly off from 10 kHz? Like 9,542 Hz as a
random example? Just curious if that had any impact? You could also go
above that 10 kHz if you think it's a matter of hearing limitations. You
could do 10,458 Hz instead of 9,542 and see if you could differentiate it
better or worse. The reason I propose this is that if there is an actual
frequency or type of wave form that 44.1 kHz 16-bit can't reproduce
properly, it's possible that it's mathematical and a slight variation could
clarify that. For example, a sine wave at 22,050 Hz should be able to be
sampled and reproduced by a 44.1 kHz 16-bit ADC DAC combination, but if it
couldn't, then try 22,049 and see if that is better. Nyquist theory and
all that jazz.
EJ> Time was limited, so we only tested 1, 2, 4, 8, 10, 12, and 16 kHz.
EJ> FWIW, the scale on the frequency pot of the function generator has
EJ> limited precision. So a 10 kHz setting could very well be 10,007 Hz.
EJ> As a test, I was looking for gross detectable differences as opposed
EJ> to very fine and subtle differences that may require a "golden ear".
EJ> Besides, everyone's ears are different, and one person may be more
EJ> sensitive to a specific range of frequencies than another person.
EJ> This test was only valid for my ears, my equipment, in my room.
EJ> Worth noting was that the results were very consistent with
EJ> the gross frequency changes, so I would be surprised to see different
EJ> results with small frequency changes.
>I could not tell the 10 kHz sine and triangle apart. At 16 kHz I could no
>longer differentiate the square wave from the triangle or the sine,
>whereas I could at 88.2 kHz sampling.
That may be similar to not being able to distinguish something at 15 ips
half track analog, but being able to distinguish it at 30 ips half track
analog.
>The most interesting conclusion for me from this experiment is that
>the waveform clearly impacts the EQ.
That's one way to describe the effect you witnessed. I'd think it's not
actually the "EQ" but a matter of the properties of the waveform. The
reason it's a different waveform is the same reason it sounds different.
Square waves do definitely have a different sound than a triangle wave or a
sine wave.
EJ> Agreed - it's not EQ/frequency based, but rather waveform based.
>If a non-sinusoidal wave or impulse has its waveform significantly altered
>in the ADC-DAC chain, then the perceived loudness of those frequencies
>will change as well, and hence the EQ!
Again, maybe not the EQ, but you are correct that their perceived loudness
will change, which will change the overall combined sound.
EJ> Yes, exactly!
-----------------
Diamond Productions
Specializing in analog tape & film preservation / restoration in the
digital domain.
Dave Bradley President