[Currawong]

 
Quote:

That really blows.  It your first link, is that with or without the HD650.  It looks like it is without, but I want to confirm.  The HF rolloff is unacceptable.  -4.2dB @ 20KHz, starting at 2KHz.  Hopefully it's not a design issue, and someone just stuffed the board incorrectly (not likely, but hopeful).
 
Someone should ask AGD about this.
 
Also, if someone can reverse engineer the DAC configuration, that would be great.  What filter modes are the WM8741s put into?  Is de-emphasis enabled or disabled?

 
Most of what we hear as treble is in the 5k to 12k range. There's almost nothing to hear at 20k. MP3 compression usually filters out everything above 16k for example.
 
 


Quote:

Audio-gd's other DACs have similar rolloffs in measurement, see this thread:
 
http://www.head-fi.org/forum/thread/435290/rmaa-test-results-for-audio-gd-dac-19mk3-and-reference-1
 
It's likely a digital filter.

Very likely, set to slow roll-off, same as in most high-end DACs.
 
I would indeed ask Kingwa about the measurements, but remember that it's a $200 combined unit, not a $2000 one.  However, that stereo crosstalk measurement corresponds with my experiences of the low-end Audio-gd gear that with HD-600s one doesn't get a very wide soundstage. You need to go up a tier for that.

 

[m00t]

Going by this post here, it looks like the high frequency roll-off is deliberate. Kingwa was asked about a similar thing happening in the Audio-gd DAC-19MK3 and that post is his response.

 

[holland]

 
Quote:

Pin 4 is 0 volts, Pin 22 is 3.5V and Pin 24 (Hardware/software) is at 0 volts.
 
So yes it is hardware mode.  
 

 
Great, thanks.  The input rate is set for 192KHz and using the default linear filter mode.  From what I understand, changing Pin4 to floating and remove the pull-down resistor would offer a better filter than the default linear filter.  This would sacrifice some phase linearity for better time domain.  The roll-off is the same, but it is to eliminate pre-ringing.  The roll-off is to eliminate post-ringing.
 
I'm curious to see what the sweeps look like with an upsampled 192KHz sample rate, since oversampling is set for a 192KHz stream in the chip, so the chip is really expecting 192KHz data and the filters are set for high frequency data vs 44.1KHz.

Ayre whitepaper
 
Too bad there are no switches or jumpers to play with the filters.
 
Also note that if oversampling is disabled and the input stream to the DAC is 44.1KHz or 48KHz, the filters where the corner rate is normally around 30KHz moves to 15KHz.  This is *exactly* what the RMAA has shown.  By extension, if you provide an upsampled stream, the sound will change.

 

[tranzy]

Hi all,
My current setup is an Audinst MX1 paired with a  DT770Pro/80ohm and a Sennheiser PX100-II.
I've placed an order for the NFB12 too and from what I understand I should receive it later this month.
The problem is that I want to upgrade the headphones also. From the reviews that I read and the budget that I have, I'm leaning towards the HD650 or the DT880/600ohm. 
Which of those two is more suited for the NFB12? Because from the earliest posts from this thread, especially the ones from Supercurio, I understand that the HD650 are not good with this dac/amp and the DT880/600ohm will be more suited. But between my headphones, I like more the PX100-II sound's signature, which is somehow close to the HD650 from what I read.
So... please help me to choose the right headphones. 
Big Thanks!
 
PS: @Supercurio. I'm a big fan of your voodoo kernel and audio modding. Keep up the good work.

 

[Slaughter]

Not sure if anyone can tell you which one to get. I can tell you that being a previous 880/600 owner, I would love to try the 880 with NFB-12. I think it will add some of that missing warmth in the mids and needed bass impact. Senns really sound best with tubes in my experience.

 

[holland]

I use all my Beyer headphones with a 120-ohm adapter.  It adds warmth and tones down the high frequency a bit.  I really think the headphones were designed around the 120-ohm output jack in accordance with IEC 61938.

 

[supercurio]

Quote:

Quote:

Anyway guys, as long as I have it, I can make some other measurements if you like, just ask.

Can you make a sweep at 96KHz and 192KHz sampling rates?  I think what you see is that the WM8741 is being configured for 192KHz operation (no oversampling) and one of the soft-knee filters.  It needs to be configured for 192KHz, since the coax side accepts 192KHz.
 
I'd bet the roll-off reduces dramatically.

 
You're right, I'm running new tests at the moment.
Stay tuned.
 

 

[supercurio]

Hi again!
 
I ran new analysis with improved settings and setup (more adequate gains, only 1 computer to avoid noise sources, better cables)
 
This time, benchmark results are better, but you'll see easily the artifacts I was talking about with the additionnal spectrograms at 44k.
Roll-off is confirmed due to a digital filter applied.
 
Spoiler alert: Upsampling is not optional.
 
 
44k/24bit performance
http://supercurio.project-voodoo.org/audio/audio-gd/RMAA/audio-gd-NFB-12-44k-24bit.htm
 
96k/24bit performance
http://supercurio.project-voodoo.org/audio/audio-gd/RMAA/audio-gd-NFB-12-96k-24bit.htm
 
192k/24bit performance
http://supercurio.project-voodoo.org/audio/audio-gd/RMAA/audio-gd-NFB-12-192k-24bit.htm
 
Samplerates comparison
http://supercurio.project-voodoo.org/audio/audio-gd/RMAA/audio-gd-NFB-12-samplerates.htm
 
44k/24bit without or with upsampling
http://supercurio.project-voodoo.org/audio/audio-gd/RMAA/audio-gd-NFB-12-44k-upsampling.htm
 
udial at 44k, sampled at 96k :

 
udial upsampled at 96k, sampled at 96k:

 
RMAA test wave at 44k, sampled at 96k :

 

RMAA test wave upsampled at 96k, sampled at 96k:

 
 
PS: sox graphs are generated with: 

  for x in *.wav ;do sox $x -n spectrogram -o $x.png -x 1600 -s -t "$x" -w Kaiser -z 120; done

 

[llama_egg]

*whoosh*
 
I can honestly say that I have no idea what the heck those graphs are showing, but I'm sure I've been them on the screen during Star Trek though.

 

[supercurio]

Quote:

*whoosh*
 
I can honestly say that I have no idea what the heck those graphs are showing, but I'm sure I've been them on the screen during Star Trek though.

Almost in the midlle of the each graph you see 22050Hz. Past this frequency, everything is noise/artifacts & sound defects because the source is a 44100Hz signal.
 
And without upsampling, there's a lot of unwanted artifacts past 22050Hz.
Despite the initiative of the manufacturer to filter and reduce highs to reduce the harshness, those "mirror images" kills the purpose with high frequencies that introduce distortion, listening fatigue..
 
Even cheap HDA codecs integrated in computer mainboards don't exhibit this kind of problems those days.
If you upsample before sending to the DAC, the performance is honorable. If you use a CD player outputing 44100Hz signal, the performance is just crap.
 
Looks like someone didn't do its homework in audio testing, or my amp is misconfigured?

 

[Slaughter]

I assume these are results for the amp and not the DAC output? Also, did you run these test on both high and low gain as you should get different results?

 

[supercurio]

Quote:

I assume these are results for the amp and not the DAC output? Also, did you run these test on both high and low gain as you should get different results?

"DAC" and "Headphone" output look the same in measurements.
Low gain is usless, it only attenuate the input signal (so, you only loose in terms of performance or benchmarks)

 

[holland]

Quote:

 
Even cheap HDA codecs integrated in computer mainboards don't exhibit this kind of problems those days.
If you upsample before sending to the DAC, the performance is honorable. If you use a CD player outputing 44100Hz signal, the performance is just crap.
 
Looks like someone didn't do its homework in audio testing, or my amp is misconfigured?

It's set up incorrectly.  It's an easy fix though, setting 1 pin properly, though I would change the filter to be the minimum phase filter instead of the linear.  I'd take phase distortion over edginess any day of the week, as per the Ayre whitepaper.  If you really want to have flexibility you need to add parts.  You'd need an ASRC, but I would run the DAC at 96KHz instead of 192KHz, because the WM8741 runs better at 96KHz instead of 192KHz.  In 192KHz it runs at 128Fs instead of 256Fs.  I might build an ASRC module when I get mine in March or just set it to 44.1KHz and the minimum phase filter.
 
Nice job on running the tests.  The issue is that the DAC is told the input stream is 192KHz, and it simply is not.
 
Email KingWa the data.  I'd wait a little longer for an ASRC implementation, or at least a 2 3-way switches in the back to select the HW filter mode and the HW sampling rate.

 

[supercurio]

Interesting!
However, I don't understand why the DAC is not setup to use the SPDIF clock as source (or USB transmitter)
Based on the input source clock, it's possible to generate a master clock that would be used for the DAC, set in DSP or I2S slave right ?

 

[holland]

Quote:

I assume these are results for the amp and not the DAC output? Also, did you run these test on both high and low gain as you should get different results?

The DAC is setup to indicate the data stream is 192KHz, to configure the oversampling mode.  The data it is getting is not 192KHz, so it's algorithms are out of sync.  A full upsample to 192KHz would give the best audio quality, as the WM8741 may think of it.  This is well documented in the WM8741 spec sheet.  Personally, I would change the HW sample rate to 44.1KHz/48KHz and use the more aggressive minimum phase filter, but tastes vary.  However, setting the proper sample rate is not really optional.  The chip uses a number of algorithm changes based on it.
 
There are a number of fixes, from cheap and quick, to complex and expensive.
 
I guess this is a case of marketing for the upsampling/DI hype crowd, but forgetting that the chip has limitations.