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: "Mikael Hakman" <email@hidden>
- Date: Tue, 26 Aug 2008 04:15:10 +0200
- Organization: Datakonsulten AB
Brian,
I'm sorry if I sound a little harsh below, but you have a way to put words
in my mouth which I never said, and to put your thoughts in my head that I
never thought. You also invent relations that don't exist and argue against
what you think I think or do without knowing. In every argumentation there
must rational logics, consistency, and agreement with reality. Putting a
sequence of words without above qualities isn't god enough, and irritating.
On Monday, August 25, 2008 9:03 PM, Brian Willoughby wrote:
Mikael,
One thing which really stood out in your analysis was the claim that
"distortion measured right after the start of the signal yields [...]
0.5% distortion!" This is actually due to your tools lacking windowing,
and not an actual indication that the Apple hardware changes it's
distortion specs when your magic signal appears. Your choice to avoid
windows in your analysis may improve the graphs for a very small selected
subset of possible signals, but you're increasing the distortion inside
your tools for most other signals, and thus your measurements are not
always accurate.
Nonsense! Distortion measured under steady-state conditions is -75 dB at
1000 Hz. This is in Apple's spec and I have verified and reported this in my
report. Distortion at 1000 Hz measured right after the signal starts
(reaches the receiver) is not specified by Apple neither by codec vendor. I
have measured it to be around -46 dB or 0.5%. It is quite common for DACs
that they exhibit higher distortion at the very beginning of a signal.
Interesting issue is not whether there is higher distortion but by how much,
and how fast the device stabilizes. Use or not use of windowing function in
analysis has nothing to do with it.
Actually, using windowing function on such clean and pure test signals as
mine, can only increase the computed THD+N value. This is because a window
smears out that sharp single frequency peak to become a number of
frequencies, all located near the peak but in separate DFT "bins" and with
high amplitude. Then because THD+N computation takes RMS value of all the
bins, the resulting value will be higher, much higher. Under these
circumstances using a windowing functions is directly inappropriate. For
example, in the actual case, using Hamming window with N=4800 results in
THD+N of only -5 dB instead of those -75 dB measured by me and specified by
the vendor.
Furthermore, because there are many window functions, and you can use
various parameter values in these functions, which one should you choose to
get right and consistent THD+N values?
Windowing is used because the FT analyses a small part of a signal as if
that part repeated forever. Unless your D/A and A/D line up perfectly,
your windowless analysis is actually looking at a completely distorted
picture of the signal, and thus you'll see very high distortion as an
artifact of your custom tools, not due to any problem in the Apple audio
circuits.
Unless what? A/D doesn't know the signal is coming from another D/A, it
could be a microphone. D/A doesn't know it sends output to an A/D, it cold
be a monitor. So if I do get distorted signal starting with pure one then
they must have distorted it, and this is precisely what I want to know and
measure.
In fact, this is worst case when you start the test signal after
silence, because the FT sees a repeating waveform that is very complex -
and this totally explains the high distortion you're seeing in your
custom tools.
Worst case!? Well, about as worst as saxophone starting to play a note from
silence or near silence. About as worst as an artist playing a slow classic
composition on acoustic guitar and letting every note decay to low levels!
What are you talking about?
Because what? I send pure simple sine wave and if it becomes very complex
signal then the device must have made it so. Very complex signal from simple
one = high distortion, which is precisely what I want to know.
Digital audio is full of counter-intuitive practices. We add dithering
noise to reduce quantization noise, yet the initial reaction is "why
would we want to add noise to digital audio?" We use windowing on any
signal fed to Fourier analysis to avoid discontinuities in the signal
within the buffer, yet there are some cases where particular windowing
options produce as many problems as they cure. The challenge is to
select the right windowing function for the job, but I doubt you can
measure an unknown process accurately without any windowing at all.
Actually, now that I think about it, even if you used windowing, you'd
probably still get false readings showing high distortion at the start of
your test signal because there could still be a discontinuity in the
waveform in the middle of the buffer where most windowing functions are
not altering the signal. You need to devise a different solution for
this, and you certainly need to be careful not to report false distortion
specs for the equipment you're measuring with unconfirmed tools.
There isn't any discontinuity; my specially designed signals guarantee this.
I start analysis when and at the very first test signal sample reaches my
receiver. If there were any discontinuity thereafter, my code would throw an
exception. I just designed that different solution you are asking for and
the results are put forward in my report.
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