Continuation of R2R DAC Discussion From Stax Thread

[Sechtdamon]

15 hours ago, RudeWolf said:

 

Okey, I'll PM you, thanks a lot sir. Lets talk all the things I need to know, I might need a bigger pic, Then I'll make all the PMs more readable and post it on this topic.

 

Ps: nice mice. but clean it for the love of the madcat.

Edited December 17, 2016 by Sechtdamon

[jp11801]

Andy thanks! Every once and a while I feel like the yggy has some hardness is the treble but other than that I love it. 

[kevin gilmore]

left 3 top and bottom blue resistors are the segment bits 5.11k

02C resistors and all the other blue ones in that row are 10.2k

69B resistors and the other blue ones in that row are 5.11k

130 is 13 ohms

top switches v- = ground

bottom switches vcc=ground

 

Edited December 17, 2016 by kevin gilmore

[blubliss]

4 hours ago, jp11801 said:

Andy thanks! Every once and a while I feel like the yggy has some hardness is the treble but other than that I love it. 

Yes, I feel the same way at times.  The Holo can be a little easier to listen to sometimes, but that little bit of softness may be annoying over time. For instance, on one CSN song (49 bye byes) which is not a great recording, the extra clarity of the Yggy reallly helped.

[Sechtdamon]

Okey, some more google findings about hibiki R2R module and other hibiki findings:

 

 

 

This one is huge:

 

 

 

 

This one is the classic one i suppose:

 

 

 

 

And those links are chinese hifidiy links, If anyone knows Chinese:

 

http://bbs.hifidiy.net/thread-1149252-1-1.html

 

http://bbs.hifidiy.net/thread-1192897-1-1.html

[joehpj]

green ones are prototype. red ones are final product. Just go for the red ones.

[Sechtdamon]

Good to know, but I'm asking for enlightment of their structure. Like Dr. posted about soekris 3 posts above. And some people want to check for their closer taken picture too. Thats why I'm posting all my findings.

 

Also, a website (you can check op for it) sells 3 version of it. Are they still selling prototypes, or they have 3 different product for sale?

[alvin1118]

On 12/3/2016 at 4:48 AM, Sechtdamon said:

Vinshine Audio DacR2R Ref (based on soekris module, I cannot read which one as piece of the chasis covers it) 2100 Dollars:

http://www.vinshineaudio.com/product-page/a5952b51-0edc-b170-4f94-648d4a9dec86

The power of google brought me here to head-case! I'm alvin1118, the founder of VINSHINE AUDIO. The Soekris board in the photos is Rev3 0.02% 

p/s: The currency in the webstore is SGD S$. S$1 = $0.71 approx $1480. 

[joehpj]

http://www.audio-gd.com/Pro/diy/DA01/DA01EN.htm

a-gd has released their R2R module. NO digital filter included so it seems future promising when better filter is available. Kingwa said if there were enough people want this, he would also do the digital filter module, separate DSD DAC, and also PS for the whole thing. Please notice the last few lines:"The module had the DC offset , customer want to use the couple cap, or the DAC analog filter output had the couple caps or DC serve design. For the balance DAC design, the proper design can remove the DC offset in the analog filter amp output , without couple caps or DC serve design ".

KG's unbaltobal seems to be a good chioce for the buffer or with SuSy dynalo to make a wonderful DAC/amp

Edited December 22, 2016 by joehpj

[mypasswordis]

Hm, am I missing something here? This just seems like a less complete/more raw version of Soekris DAC with lower precision resistors... the Soekris already has sign magnitude and programmable filter chip, and with FIFO/reclocking. Same 24 bit, 3.072MHz oversampling as well. I believe they solved the DC output problem already as well so it can be DC coupled without blowing up your stuff. 

Edit: Hm, maybe it's for people who want to do it themselves (DIT?)  Looking forward to see how people use these modules.

Edited December 22, 2016 by mypasswordis

[joehpj]

This module use .5% resistors in parallel to get .25%. So it's not too far from .2% Soekris used.

Edit: .1% in parallel to get .05%. Wrong information. My bad.

KG said Soekris module used wrong switch and caused some problem. The overall design was flawed. Can't find the comment now. I don't know if this is better.

Also, the original OP based buffer of Soekris sounds like shit. A separated buffer was needed.

Edited December 22, 2016 by joehpj

[Remolon]

2 hours ago, joehpj said:

This module use .5% resistors in parallel to get .25%. So it's not too far from .2% Soekris used.

 

Not exactly. Suppose we need 100R.

Paralleling two 200R: 200R + .5% = 201, in parallel 100.5R, exactly 100R + .5%, Although with less probability of the maximum error.

[joehpj]

it will lower the average error. Not in extreme cases just like you said.

[kevin gilmore]

here is the soekris output section

http://gilmore.chem.northwestern.edu/dac.pdf

notice that the top set of switches is wired to +4 and ground, and the bottom set is wired ground and -4v (this is how you get sign/magnitude without any dc on the output.

each audio-gd board is half (less than half) of a soekris single output. how you put two boards together to get rid of the dc is unclear. I see no way to wire one set of switches to ground and -vcc.

soekris is sign/magnitude, audio-gd in single board form is definitely 2's complement

switches look the same

soekris does dsd by using just the 3 segment bits

 

Edited December 22, 2016 by kevin gilmore

[joehpj]

5 hours ago, Remolon said:

Not exactly. Suppose we need 100R.

Paralleling two 200R: 200R + .5% = 201, in parallel 100.5R, exactly 100R + .5%, Although with less probability of the maximum error.

updated information. My bad.

[Dusty Chalk]

But you're still using the same math, just different numbers.  I don't think overall that's what you'd get.  You could get as good as that here and there, but your average error would still be bad.

[Craig Sawyers]

It is all down to how the tolerance is statistically distributed. You can get two different scenarios. (i) All resistors in any production batch are clustered in value or (ii) resistance values in a batch are randomly distributed.

This could only be established by measurement.

However, if (i) holds then the batch will have a constant error and if (ii) holds then the values will be normally distributed. Typically the nominal tolerance will be 3-sigma, which would mean that less than 0.27% of resistors were out of tolerance. So the majority of resistors will be significantly better than nominal tolerance.

The difficulty in getting accuracy is that you need to test a large number of resistors. If 100 resistors are tested the nominal error is 10%. To get 1% confidence you would need to test 10,000 resistors - er, no thanks at that level!

[lostinaudio]

3 hours ago, Craig Sawyers said:

It is all down to how the tolerance is statistically distributed. You can get two different scenarios. (i) All resistors in any production batch are clustered in value or (ii) resistance values in a batch are randomly distributed.

This could only be established by measurement.

However, if (i) holds then the batch will have a constant error and if (ii) holds then the values will be normally distributed. Typically the nominal tolerance will be 3-sigma, which would mean that less than 0.27% of resistors were out of tolerance. So the majority of resistors will be significantly better than nominal tolerance.

The difficulty in getting accuracy is that you need to test a large number of resistors. If 100 resistors are tested the nominal error is 10%. To get 1% confidence you would need to test 10,000 resistors - er, no thanks at that level!

Can't they laser trim the resistors now to guarantee an accuracy of 0.01% as TotalDAC? I wonder what negative effect ANY error has on the data and sound? Might well negate the benefits of the array in the first place v a chip or FPGA?

[mypasswordis]

Total dac uses 0.01% foil resistors. Best  version of Soekris uses 0.01 resistors also, but not foil I believe. These are still much worse than laser trimming on a single substrate can do, though I'm not sure the exact amounts. I would guess at least a factor of 2 (edit: i meant order of magnitude) better. What is more important than absolute linearity errors after a certain point is high monotonicity.

Edited December 22, 2016 by mypasswordis

[Craig Sawyers]

There is absolutely nothing to prevent laser trimming, or selection to any tolerance you like provided that temperature and voltage coefficients are low enough. 0.01% tolerance is 10ppm or 1 ohm error in 10k. Temperature coefficient is around 3ppm/C so to match tolerances at a 10ppm level the measurement temperature has to be constant to 1C or thereabout.

Foil resistors are superb, but are punishingly expensive.

And any SM component has to be soldered using precisely the soldering profile in the datasheet for the parts - because component stress and thermal shock will shift the resistance too.

However you look at it, it isn't trivial, is all I am saying.

[Pars]

On 12/16/2016 at 0:15 PM, mypasswordis said:

Speaking of TDA1541A, anyone want to share experiences with them? Particularly DACs with single and double crown units. The prices are insane!

Funny how audiophiles at the start of digital kept saying digital sounded worse than analog ("digititis" and whatnot) and ever since, audio companies use that as a selling point ("my design sounds more like real analog than all the other audio companies trying to make stuff that sounds like real analog!"). 30 years later, people are saying the old DAC chips sound like analog and the new ones don't. 

My Rotel RCD-855 (modded) is a TDA1541A-based player. I've also built Pedja Rogic's DAC and liked it, but didn't really use it much. Jim (morphsci) has it now. This DAC is the predecessor of the AYA DACs. I have at least a couple of spare chips as well. The 1998 Taiwan chips are supposedly the best ones (go figure).

Check https://www.audialonline.com/topics/tda1541a-and-model-s-usb-part-4-tda1541a-grades-and-series/ for more info on the grades (than I ever knew). Apparently many of the std. A grade chips measure so close (or as good or better than) that it isn't worth paying the exorbitant pricing people are charging for crown chips.

[joehpj]

IMO, if people are asking for precise resistors, and they should go to single chip like 1704 or what schiit uses. All resistors are laser trimmed and packed together to avoid different voltage drift. Discrete design can never reach the level of precision of single chip.

[Dusty Chalk]

What?  No.  If someone wants to believe that discrete sounds better, they need a product to satisfy their placebo tendencies.

[Craig Sawyers]

6 hours ago, joehpj said:

I1704 or what schiit uses. All resistors are laser trimmed and packed together to avoid different voltage drift. Discrete design can never reach the level of precision of single chip.

Unless my use of Google is in meltdown, the PCM1704 is obsolete. Is this wrong?

For discrete R2R, assuming that matching is done correctly, the whole resistor chain needs to be kept isothermal. There is little evidence that this is the case, but could be arranged to be the case.

[joehpj]

Pcm1704 has been NRND for several years. And recently got removed from the website. Audio gd still got some to make new dacs. But I think this is the last batch they could make.

Resistors and solder joints exposed in air will be influenced by humidity, oxidation. Some will drift slightly over years. I don't know if every resistors will.

Single chip w/ laser trimming would be better solution in terms of stability. IMO.

Discrete however provides better flexibility of design. And the overall performance is not only limited by dac module/chip itself.