Discussion:
Some observations about the Benchmark DAC1
darrenyeats
2013-04-25 07:28:17 UTC
Permalink
See my post here.

http://www.pinkfishmedia.net/forum/showthread.php?t=137152

Regards, Darren

Sent from my HTC Sensation Z710e using Tapatalk 2


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-04-25 15:16:50 UTC
Permalink
Post by darrenyeats
See my post here.
http://www.pinkfishmedia.net/forum/showthread.php?t=137152
Regards, Darren
Sent from my HTC Sensation Z710e using Tapatalk 2
Thanks Darren. I like the comment about overshoot handling. Will need to
have a look at this when I do my measurements. Will be very curious what
the Transporter and Touch looks like in this regard.

Also, it's interesting to note the worsening of jitter with the USB
interface clearly demonstrated. I have actually seen a similar
phenomenon with my ASUS Essence One comparing the USB with SPDIF; even
though it is supposedly better due to asynchronous protocol.


------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-04-27 12:32:09 UTC
Permalink
Archimago, you may want to look again at the PFM thread since there has
been a bit of question-and-answer! Also screenshots.
Darren


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-04-27 15:58:15 UTC
Permalink
Post by darrenyeats
Archimago, you may want to look again at the PFM thread since there has
been a bit of question-and-answer! Also screenshots.
Darren
Thanks Darren.

I've tried some preliminary tests. I'm actually not sure what Rockwell
is getting at with the overshoot handling plots. When I try to roll over
that second image (where he lowered the volume), I'm not seeing a
difference (anyone else seeing something?).

Here's what the AUNE X1 looks like with a 1.00227 kHz square wave (that
extra 2.27Hz is because of the 0.1kHz in 44.1kHz):
14753
Note the bit of eve harmonics thrown in and I didn't bother calibrating
to 0dB.

Here's what happens with an exact 1.00000 kHz square wave.
14754

The DAC1 doesn't look clean compared to his Sony SACD player output
because of the upsampling from 44.1kHz to 110kHz - a non-integer
upsampling factor; what is or is not contributed by overload is unclear
to me.

The DAC2 I think upsamples to 211kHz which again is not a clean multiple
of 44.1kHz. As such, I suspect again if you look at a 1.00227kHz signal
like on the CBS CD-1, the spectrum will still look ugly compared to the
SCD-XA777ES.

Mr. Rockwell did not mention this issue...


+-------------------------------------------------------------------+
|Filename: Standard_Exact_1kHz_AUNE.jpg |
|Download: http://forums.slimdevices.com/attachment.php?attachmentid=14754|
+-------------------------------------------------------------------+

------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-04-27 23:27:32 UTC
Permalink
Archimago,
Did both your examples use square waves with 0db peaks?

I thought the fact the DAC1 uses ASRC for jitter would mean integer
multiples are irrelevant.

Thanks,
Darren

Sent from my Nexus 10 using Tapatalk HD


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-04-28 01:08:39 UTC
Permalink
Post by darrenyeats
Archimago,
Did both your examples use square waves with 0db peaks? The screenshot's
a bit fuzzy.
I thought the fact the DAC1 uses ASRC for jitter would mean integer
multiples are irrelevant. Hmm.
Thanks,
Darren
Sent from my Nexus 10 using Tapatalk HD
Yes, both are square waves at 0dB. Unfortunately the message board
resizes the images.

Integer multiples are *essential* when looking at this type of test with
square waves. That's why I suspect that Mr. Rockwell is unaware of this
fact despite the fancy gear (I like his camera review BTW) :-).

Bottom line, the ugly looking spectrum is due to the Benchmark DAC1's
110kHz upsampling (why did they pick this and not something better like
176 or 192!?) unless proven otherwise!


------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-04-28 10:12:02 UTC
Permalink
Ok, your criticism of Rockwell's analysis makes sense now.

Returning a moment to all the raw measurements, the Benchmark spectra,
with strong spurs across the board at -60db, are still distinctly worse
than the bad one you just posted and the also-bad spectra for the Apogee
and Audioengine. Could we infer that all were non-multiples and this BM
difference is due to overshoot handling?

The much bigger difference in the Sony spectrum and your better spectrum
would still be explained by the integer multiple (as well as potentially
better overshoot handling?)
Darren



Sent from my HTC Sensation Z710e using Tapatalk 2


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-04-28 15:17:44 UTC
Permalink
Post by darrenyeats
Archimago,
Ok, I now understand your criticism of Rockwell's analysis. Good point
well demonstrated. It explains the Sony.
The Benchmark results with strong spurs across the whole spectra at
-60db, are still distinctly worse than the dirty one you just posted ...
plus Rockwell's (also dirty) spectra for the Apogee and Audioengine.
Could we infer that all were non-multiples and this BM difference is due
to poorer overshoot handling?
The much bigger difference in the Sony spectrum and your better spectrum
would still be explained by the integer multiple (as well as potentially
better overshoot handling?)
My point is, we know the DAC1 lacks DSP headroom (and the DAC2 has this
headroom) and I expect this is at least contributes to Rockwell's
particularly bad DAC1 spectra.
Darren
Sent from my HTC Sensation Z710e using Tapatalk 2
Yes, that is exactly what I was trying to get at. It looks like ASRC
caused the square wave spectrum to look distorted compared to the clean
Sony. As you said, perhaps he's suggesting that lowering the volume will
show a significant reduction in the amount of distortion but it's hard
to tell from those graphs, especially the 2nd graph in that section
since the text suggests there should have been another image if the
mouse rolled over it.

I'm also a little confused about the difference he's seeing between USB
and TosLink (first roll-over graph). I would have thought that with the
ASRC, the spectral difference would be minimal if even present. It also
doesn't look like he's using Bit-Perfect with iTunes to ensure accuracy,
furthermore I'm also a little concerned about his decision to use that
"six-year-old Belkin 7-way powered USB hub with a 10-year-old, 6-foot
USB cord" - nothing wrong if we know for sure this old setup isn't
introducing its own noise into the mix (I've seen this kind of problem
in tests).


------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-04-28 18:19:20 UTC
Permalink
The roll-over is merely to compare the two graphs more easily. So when
you roll-over the top graph you can A/B with the bottom graph. He
mentions knocking the levels but it was accidential and just 1db so it
isn't to demonstrate headroom (and indeed it isn't enough for headroom,
Benchmark suggest 3.5db is required).


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-06 06:48:26 UTC
Permalink
Had a look at the effect on the ASUS Essence One which has a "symettric
upsampling" feature. So 44.1kHz material is upsampled 8x to 352.8kHz.

No upsampling - square wave, 0dBFS, 1.00227 kHz:
14787

With 8x upsampling - square wave, 0dBFS, 1.00227 kHz:
14788

Nice :-)
Harmonics remain clean and I don't see any issue with overload - looks
like ASUS has compensated adequately for the intersample overhead for
this test. Also notice how clean the even order harmonics are compared
to the AUNE X1.


+-------------------------------------------------------------------+
|Filename: 1khz_Square_-_With_Upsampling.jpg |
|Download: http://forums.slimdevices.com/attachment.php?attachmentid=14788|
+-------------------------------------------------------------------+

------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Wombat
2013-05-06 13:13:07 UTC
Permalink
Somehow i canŽt believe that just comes froms integer/non- integer.
IsnŽt it that you get such spikes as the DAC1 shows when you ignore
dither?


------------------------------------------------------------------------
Wombat's Profile: http://forums.slimdevices.com/member.php?userid=4113
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-06 15:18:42 UTC
Permalink
Post by Wombat
Somehow i canŽt believe that just comes froms integer/non- integer.
IsnŽt it that you get such spikes as the DAC1 shows when you ignore
dither?
edit: For example the dithered/non-dithered pics of this SRC
http://shibatch.sourceforge.net/ssrc/
Maybe... I'll have a look at this more soon once I get the Teac UD-501
and try out the ASRC on that device to 192kHz. Obviously I suspect that
will looks like the DAC1.


------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-08 06:26:01 UTC
Permalink
Here's we go... Non-integer upsampling of that square wave:

No upsampling - 1.00227kHz square at 0dBFS:
14793

Asynchronous upsampling to 192kHz of same square wave:
14794

Looks kinda like the DAC1, bunch of extra harmonics courtesy of the ASRC
algorithm, not due to volume overload...

Measurements off TEAC UD-501.


+-------------------------------------------------------------------+
|Filename: Upsampled_192kHz.jpg |
|Download: http://forums.slimdevices.com/attachment.php?attachmentid=14794|
+-------------------------------------------------------------------+

------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-05-08 10:57:47 UTC
Permalink
Archimago,
Nice. Can we do a comparison with lowered amplitude though (both integer
and non-integer)? I am guessing a square wave will need headroom?
Darren


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Wombat
2013-05-08 14:24:50 UTC
Permalink
Post by Archimago
Measurements off TEAC UD-501.
And again i wonder if this is also because of insufficient dither. Can
you resample that square wave with SoX and force to disable dither to
see what happens?


------------------------------------------------------------------------
Wombat's Profile: http://forums.slimdevices.com/member.php?userid=4113
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-08 18:03:07 UTC
Permalink
Post by Wombat
And again i wonder if this is also because of insufficient dither. Can
you resample that square wave with SoX and force to disable dither to
see what happens?
The square wave was generated in native 16/44 with Audacity or Audition
(don't remember which one now).

So, basically turn the 16/44 1.00227kHz to say 24/192 in SOX with no
dither and see how that looks when played out on the DAC?


------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Wombat
2013-05-08 18:11:41 UTC
Permalink
I don't know how you create these pics. If you do it with DAC output you
should make sure no further resampling is done and for comparison it
should be nice to see both results with SoX, one with dither and one
without. Maybe this hints to something. Sorry if that makes you busy
only because i have a wild speculation.


------------------------------------------------------------------------
Wombat's Profile: http://forums.slimdevices.com/member.php?userid=4113
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-09 04:13:55 UTC
Permalink
Post by Wombat
I don't know how you create these pics. If you do it with DAC output you
should make sure no further resampling is done and for comparison it
should be nice to see both results with SoX, one with dither and one
without. Maybe this hints to something. Sorry if that makes you busy
only because i have a wild speculation.
edit: didn't see your edited post, was typing to long on my pad...
The pics are snapped from real-time FFT done from the direct RCA capture
of the DAC output through the E-MU 0404USB at 24/96. No resampling or
DSP engaged... Just spectrum analysis of what's coming in.

Tonight, as I look into this, I realized just how complicated it's
getting and how hard it would be to show the pictures unless this was
one of my long blog posts! Bottom line:
It really depends a lot on the resampling algorithm and obviously
hardware!

It looks like *both* integer resampling and dynamic headroom have a part
to play in the square wave tests at 0dB.

IN SOFTWARE: If you do integer resampling on 0dBFS like 44kHz --> 88 -->
176, Adobe Audition, and dBPowerAmp maintains the clean harmonics.
SOX for some reason is not giving me the same result! Not sure if it's
because Audition and dBPowerAmp are anticipating and upsampling with a
small amount of volume overhead adjustment.

IN HARDWARE: Using the TEAC UD-501, integer resampling is not a problem
- for example 48kHz --> 192kHz results in a nice spectrum. Likewise the
ASUS Essence One "symmetric upsampling" was not a problem. However
non-integer 44kHz --> 192kHz results in the additional harmonics I
showed last night with the Teac; presumably similar to the DAC1 pictures
from Ken Rockwell.

Now as for volume headroom - YES, this does result in cleaner harmonics
including SOX even with non-integer upsampling from what I see using
SOX, Audition, and dBPowerAmp. Therefore, Darren / Benchmark is right to
allow headroom while upsampling - avoidance from clipping is never a bad
thing!

I couldn't see a difference with dithering since this was upsampling
from 16 --> 24-bit which is logical when I tried SOX (both dithered and
undithered looked nasty going from 44 --> 192)...

In any case, as you can see, the situation looks complicated and it'd be
tough to put together images of the combinations I tested.

Bottom line, and I think the *only graph* that would be of benefit for
Benchmark DAC1 owners is one where a -3dB 1.00227kHz square wave is
passed through the DAC1 and there's demonstration of a clean 1kHz
primary with the appropriate clean 3/5/7/9...kHz harmonics to show that
the volume reduction helps. Based on my TEAC UD-501, it should look like
this:

-3dB Square:
14796

-3dB Square upsampled to 24/192 by Teac:
14797

Nice and clean with minimal even order harmonics...


+-------------------------------------------------------------------+
|Filename: -3dB_Square_Upsampled.jpg |
|Download: http://forums.slimdevices.com/attachment.php?attachmentid=14797|
+-------------------------------------------------------------------+

------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-05-09 11:43:56 UTC
Permalink
Very nicely done. Am I right to include the volume reduction makes the
ASRC/upsampling much less of an issue then?
Darren


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-09 14:46:25 UTC
Permalink
Well ferreted Archimago. Am I right to conclude the volume reduction
lets the ASRC/upsampling work with minimal impact, then?
Darren
Yes Darren,
That's what I'm able to demonstrate with the Teac's 24/192 upsampling. A
-3dB attenuation will allow the upsampling to work without adding
significant distortion on the square wave at 0dB.

Peak volume extension on these square waves produced by Adobe Audition
are around +2dB so the hardware will need at least that amount to
prevent clipping.


------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Wombat
2013-05-09 15:11:19 UTC
Permalink
Thank you! Would be interesting why SoX does so bad because it is
perfect at downsampling. I also wonder how many devices offered in the
past did really bad upsampling while called high-end audiophile. On the
other side squarewaves are not exactly music so this all is very
theoretical.


------------------------------------------------------------------------
Wombat's Profile: http://forums.slimdevices.com/member.php?userid=4113
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-09 15:22:10 UTC
Permalink
Post by Wombat
Thank you! Would be interesting why SoX does so bad because it is
perfect at downsampling. I also wonder how many devices offered in the
past did really bad upsampling while called high-end audiophile. On the
other side squarewaves are not exactly music so this all is very
theoretical.
Yeah, not sure why with SoX. Since I didn't use gain parameters to
decrease volume, it did warn me about clipping so it's doing it's
diligence to warn the user of a problem.

Yup... Square waves.


------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-05-09 16:32:31 UTC
Permalink
Would be interesting why SoX does so bad because it is perfect at
downsampling.
With down-sampling, there is nothing to do in terms of level adjustment.
With up-sampling, you might need level adjustment. But it seems Sox
doesn't do -automatic- level adjustment, hence the difference for
up-sampling and down-sampling.

But, as Archimago says, Sox does provide warnings so you can make the
level adjustment yourself (e.g. gain -3.5) when a conversion needs it.
Darren


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Wombat
2013-05-09 18:01:54 UTC
Permalink
Post by darrenyeats
With down-sampling, there is nothing to do in terms of level adjustment.
With up-sampling, you might need level adjustment. But it seems Sox
doesn't do -automatic- level adjustment, hence the difference for
up-sampling and down-sampling.
No, with downsampling there is enough clipping going on also with peaks
that can reach above 1dB often enough. I use SoX myself and while
testing how to downsample best i found that SoX does less clipping as
other resamplers, especially when used with not so steep settings.
I never liked the idea of upsampling so i donŽt know much about its
behaviour.


------------------------------------------------------------------------
Wombat's Profile: http://forums.slimdevices.com/member.php?userid=4113
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
darrenyeats
2013-05-10 14:57:24 UTC
Permalink
Post by Wombat
No, with downsampling there is enough clipping going on also with peaks
that can reach above 1dB often enough.
Often, really? Oh ... is this to do with the filter?
Darren


------------------------------------------------------------------------
darrenyeats's Profile: http://forums.slimdevices.com/member.php?userid=10799
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Wombat
2013-05-10 21:27:46 UTC
Permalink
I am no expert but the filters should be pretty similar for up and
downsampling. The tradeoffs are pretty similar also. I don't know a
recording with high samplerate to test with that has similar bad dr
numbers as the ones Archimago used for upsample testing to have some
comparison.


------------------------------------------------------------------------
Wombat's Profile: http://forums.slimdevices.com/member.php?userid=4113
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Julf
2013-05-11 07:30:41 UTC
Permalink
And this all would of course not be an issue at all if people would have
kept following the original Philips/Sony recommendations of how much
amplitude margin to leave in a digital recording (as specified in the
original CD standards)....


------------------------------------------------------------------------
Julf's Profile: http://forums.slimdevices.com/member.php?userid=42050
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
mlsstl
2013-05-11 13:09:47 UTC
Permalink
Post by Julf
And this all would of course not be an issue at all if people would have
kept following the original Philips/Sony recommendations of how much
amplitude margin to leave in a digital recording (as specified in the
original CD standards)....
That is a not uncommon issue in all human activity:

Step 1: Don't follow instructions

Step 2: Notice that things didn't turn out right

Step 3: Attempt to fix the problem by adjusting or changing everything
except the original error.

;-)


------------------------------------------------------------------------
mlsstl's Profile: http://forums.slimdevices.com/member.php?userid=9598
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Archimago
2013-05-11 16:48:01 UTC
Permalink
Post by Wombat
I am no expert but the filters should be pretty similar for up and
downsampling. The tradeoffs are pretty similar also. I don't know a
recording with high samplerate to test with that has similar bad dr
numbers as the ones Archimago used for upsample testing to have some
comparison.
Well boys I found you some music where upsampling "needs" ALOT of
headroom! Feast your eyes on probably the most horrendous LOUD, CLIPPED
"album" ever to "grace" this terrestrial sphere (and managed to get
published)... I give you Venetian Snares & Speedranch's mega-lo-fi
electronica-noise album "Making Orange Things" (2001). Check out these
DR numbers:

----------------------
foobar2000 1.1.8 / Dynamic Range Meter 1.1.1
log date: 2012-06-25 19:25:19

--------------------------------------------------------------------------------
Analyzed: Venetian Snares & Speedranch / Making Orange Things (DR0,
FLAC)
--------------------------------------------------------------------------------

DR Peak RMS Duration Track
--------------------------------------------------------------------------------
DR0 0.00 dB -1.04 dB 4:26 01-Fire Beats
DR0 0.00 dB -0.86 dB 4:42 02-We Hate Russell
DR0 -4.56 dB -3.75 dB 2:32 03-Pay Me For Sex
DR0 0.00 dB -0.44 dB 3:56 04-Cheatin'
DR0 0.00 dB -0.46 dB 3:26 05-Unborn Baby
DR0 0.00 dB -0.41 dB 4:05 06-Meta Abuse
DR0 0.00 dB -0.32 dB 3:51 07-Molly's Reach Around
DR0 0.00 dB -1.09 dB 3:33 08-Russell Hates this Track
DR0 0.00 dB -0.76 dB 3:23 09-Viva Las Vegas
DR1 0.00 dB -4.03 dB 3:39 10-Tushe Love
DR0 0.00 dB -1.61 dB 3:51 11-Halfway Up the Stairway of
Mucus
--------------------------------------------------------------------------------

Number of tracks: 11
Official DR value: DR0

Samplerate: 44100 Hz
Channels: 2
Bits per sample: 16
Bitrate: 1292 kbps
Codec: FLAC

------------------------------------

Behold the horror of track 7 - "Molly's Reach Around" - nasty in more
ways than one! RMS volume at -0.32dB. This is either an audiophile
geek's nightmare or something for Ripley's...
14802

Now let us reduce by 6dB's:
14803

Upsample with Adobe Audition 3.0 to 32/192:
14804

Peak volume now is an amazing *-0.18dB*!!!

So, I guess if you listen to this type of "music", you "need" 5.82dB of
overhead! Congrats to the band...

Of course, you would not expect anything like this with acoustic
recordings...


+-------------------------------------------------------------------+
|Filename: Upsampled.jpg |
|Download: http://forums.slimdevices.com/attachment.php?attachmentid=14804|
+-------------------------------------------------------------------+

------------------------------------------------------------------------
Archimago's Profile: http://forums.slimdevices.com/member.php?userid=2207
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Julf
2013-05-11 17:09:16 UTC
Permalink
Post by Archimago
Behold the horror of track 7 - "Molly's Reach Around"
Well, I guess someone was bound to produce the antithesis of John Cage's
4′33″...


------------------------------------------------------------------------
Julf's Profile: http://forums.slimdevices.com/member.php?userid=42050
View this thread: http://forums.slimdevices.com/showthread.php?t=98661
Loading...