Re: Test report MBP built-in audio device
Re: Test report MBP built-in audio device
- Subject: Re: Test report MBP built-in audio device
- From: Brian Willoughby <email@hidden>
- Date: Wed, 27 Aug 2008 18:39:13 -0700
Mikael,
There is a very simple test for the things you are interested in, and
it is quite well established. It's called an Impulse Response. An
impulse response tells you everything about a linear time-invariant
system. You are correct that the attack portion of natural sounds
are a very important aspect of our human perception of those sounds.
You are also correct that steady-state test signals cannot measure
this. You are incorrect when you state that the distortion
characteristics of a system like the MacBook Pro built-in audio
change when your special test signal arrives - what you're seeing
there are mathematical artifacts of your algorithm. One problem is
that you're generating a sine wave and hoping to measure impulse
response.
As an example from the industry, Earthworks publishes the Impulse
Response of each microphone model (and of each individual microphone,
too) rather than simply publishing a frequency response. They are
also confirming that attack portions of recorded signals are very
important, particularly maintaining phase and other attributes.
However, there are unavoidable artifacts of FT analysis, and thus you
cannot prove that your results reflect only the system under test.
You've not stated how you produce a perfect sine wave on the sending
side, nor how you produce a perfect sine wave on the receiving side
for comparison. Another claim you've made is that you've created
very special test signals, which must obviously exist in the digital
realm for your MacBook Pro tests, and yet you also claim that you
have performed the same tests without a DAC - and I don't see how you
can guarantee that you can produce a special test signal in both the
analog and digital realms and be 100% certain that they are identical
signals. In other words, your algorithm has too many secrets for
anyone with a math background to confirm your results. You keep
promising us that your new methods are flawless, but I guess you'll
just have to accept the fact that many educated people will not
believe you. I am not trying to change your mind, just to let you
know why you are meeting with such resistance.
Brian Willoughby
Sound Consulting
On Aug 27, 2008, at 17:52, Mikael Hakman wrote:
On Wednesday, August 27, 2008 3:54 PM, James Chandler Jr wrote:
From: "Mikael Hakman" <email@hidden>
I'm not physically, electronically or even logically switching
the device on. The device is on for a long period of time and it
is feed by a continuous signal. The signal consists of silence
(zeros) up to a certain time. At that time the signal becomes
pure sine wave with phase 0 (i.e. first sine sample is sin(0),
next is sin(dt) etc). I continuously record output from the
device and I know the exact delay between my output into device
and device's output into my program. Therefore I know which of
recorded samples that corresponds to the first sine wave sample.
I start my analysis from that sample. When I say that I know the
exact delay I mean the exact number of samples, which of course
vary a little between the runs but is measured at the very
beginning of each run, earlier (before silence and sine wave) in
the test signal so to speak.
The reason I'm using such a test signal is that it is a crude
simulation of what happens when a musical instrument is played.
Because proper reproduction of signal during instrument's attack
time (first few milliseconds after you hit a string, start
blowing etc) has been shown to be the second most important
factor after harmonic content to our perception of timbre, I
decided to measure distortion in experiments that are close to
this musical reality. Perhaps such experiments could explain why
2 audio devices having the same or very close specs, may sound so
differently, one sounds right, the other doesn't.
I know very little about it, but if you for instance duplicate
this experiment, feeding your signal (silent head + sudden onset
sine wave)-- If you process your test signal thru a simple IIR
digital HiPass DC Blocker filter--
Visually examine the output, and the signal onset will not look
like a clean sine wave for quite awhile after the signal onset,
until the DC Blocker settles.
You can see the same thing with about any kind of filtering,
including the low-pass anti-alias filtering on ADC inputs and DAC
outputs (though those filters settle quicker than a DC Blocker). A
sudden onset interacts strongly with the impulse response of the
filter.
Dunno what an FFT of that first sudden-onset 'odd-looking' DC
Blocker sine cycle would show. A short un-windowed FFT on that
first wave cycle might make you think it indicates distortion.
Dunno. Never tried it.
But such linear filters usually don't make harmonic distortion,
only 'time distortion' contributed by the impulse response.
Maybe you are looking at some other artifact, and have a way of
ignoring linear filter settling artifacts?
Audio DAC outputs and ADC inputs, will almost invariably have high-
pass, DC Blocking characteristic. A short-term harmonic distortion
test should have some way of ignoring the time-distortion of such
filters' impulse responses.
That is one advantage of a windowed steady-state test-- In that
case any linear impulse responses in the system have presumably
settled, and so they will not confuse the distortion measurement?
Hello James,
My aim is to measure how far from or close to an ideal system, the
actual system is. Therefore I cannot ignore all those distortions
you propose to ignore because it is the magnitude of these very
distortions that make one system better (closer to ideal) than
another.
Without here going into the math of convolutions and transfer
functions, a system under investigation produces an output signal
(response) given an input signal. Any change to the original input,
that isn't constant time shift, isn't constant attenuation, and
isn't adding constant offset, is by the very definition a
distortion. This includes distortion produced by your IIR digital
HiPass DC Blocker filter, and LowPass, BandPass, and StopBand
filters and of course FIR dunno.
There are many components in a system that contribute to this
distortion. Imperfections in DAC/ADC algorithms, various filter
effects (whether filters are inherent to DAC/ADC construction or
not), output/input circuitry, variation of power or reference
voltages, temperature changes, other effects in semiconductors etc.
While knowledge about reasons why such distortion appear and what
components that contribute and in what proportions to the total
distortion may be interesting to system's vendor, it is not
required when you want to compare systems and assess relative
quality of audio that is (re)produced.
Both research and experience shows that steady-state distortion
measurements cannot explain some by humans perceived differences in
audio quality. To put it simply, given 2 devices with the same or
very similar specifications, one may sound right while the other
not. This ambiguity has been observed both at consumer, prosumer,
audiophile, and studio levels.
It has also been shown that proper reproduction of signal during
attack phase of an instrument's voice is second most important
factor in human perception of timbre (first being harmonic
content). Duration of this attack phase vary with instrument. For
most instruments it is in a range of few milliseconds. Therefore
I'm studying distortion produced by a system during this period of
time. This cannot be done using steady-state measurements because
during attack time there is no steady-state yet (otherwise it
wouldn't be attack anymore).
I hope this answers your questions.
Regards/Mikael
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Coreaudio-api mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden