Re: Test report iTunes SRC
Re: Test report iTunes SRC
- Subject: Re: Test report iTunes SRC
- From: "Mikael Hakman" <email@hidden>
- Date: Sun, 7 Sep 2008 14:46:51 +0200
- Organization: Datakonsulten AB
On Friday, September 05, 2008 11:33 AM, Richard Dobson wrote:
Mikael Hakman wrote:
Using my precision tools for assessing quality of audio devices and
algorithms I investigated iTunes SRC. The test report is at
http://www.dkab.net/iTunes report.pdf.
As compared to previous MBP built-in audio device report at
http://www.dkab.net/Realtek HDA report.pdf, the new report contains 4
additional analyses and charts. These are: a comparison between test and
response signals, differences between the two, and variations of
amplitude and phase with frequency. MBP built-in device report has also
been updated to contain the new analyses.
Excellent to include the time-domain figures. Really does make it clear
eactly what you are doing. May I suggest numbering the figures? This will
make it easier to refer to them within the text, and for example to note
the similarity between the blue line in the 13th figure and the right hand
half of the IR in the first figure (which I still think should simply be
called the impulse response, and not the "single pulse response"). It is
now clear the measurement starts aligned to the centre peak of the IR (the
first half in effect being truncated); something that I do not think is
clear from the text. Since for a given input length N the output length
must be N+(L-1) where L is the length of the IR (i.e. including both pre-
and post-ringing), it is important to be sure which parts of the full
output are being truncated to obtain the matching 4800 samples.
A picture is worth a thousand words - I too think that time-domain charts
help understanding how the numbers and charts in both iTunes SRC and MBP
built-in device reports were obtained. Also, I enjoy being didactical once I
switch to my teacher role.
Yes, you are quite right, the figures should be numbered. My intention was
to do so but then I decided to use Open Office in this project (just to
learn it, not because I'm dissatisfied with Microsoft Office that I use for
years) and there were so many new details to learn (finer points of
formatting) and I was eager to get something out so that figure numbering
wasn't done. I'll fix it next time I'll be updating the reports, which
hopefully will be soon when adding IMD charts.
I agree, "Impulse" is more precise than "Single Pulse". The reason I used
the latter name is mainly that I've been working for may years in another
sub area of signal processing, and the experimentalists in that area had
their own "single" things (single dose, single injection etc) so we used
"single pulse" in order to facilitate communication with these people. I'll
try to fix it at the next update and see how it feels.
Building on your excellent observation of similarity between right hand IR
and the blue line in 13th figure, one could ask why these differences appear
only at beginning of the signal but not later on. The answer is that later
on, the changes introduced by the right hand IR part are balanced by changes
introduced by left hand IR. At the very beginning, the left hand IR "works"
on silence signal and therefore changes done by right hand IR aren't
balanced. You probably need to think in terms of convolution integral (or
sum of products in digital world) to clearly see this. This means also that
whenever there is a non steady-state signal in audio (normal case in music),
these 2 parts won't balance perfectly and short-time distortion will result.
The magnitude of this distortion will depend on actual device IR and also on
how fast the actual signal changes in time at that very moment. This kind of
distortion is not detected by traditional steady-state measurements.
This effect is measurable in both MBP built-in audio device and in iTunes
SRC algorithm. Because length of IR in both cases is under 1 millisecond
(the L in your post) then the duration of this intermittent distortion will
be under 1 millisecond. During reproduction of audio signal this will arise
every time there are fast changes in the signal, which means all the time
for normal music. Whether this effect is audible or not is another question,
of course.
In MBP built-in device there is however also another effect, 150
milliseconds variation of DC-offset. I don't think this is related to IR in
above way.
Returning to iTunes SRC, the values shown in my report are very good. I
would be very pleased if I could find a commercial DAC/ADC device with these
characteristics. Nevertheless, I still think that iTunes should have
user-controllable option allowing for automatic matching of source and
device SR by requesting correct SR from device.
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