Reference DAC Module - Discrete R-2R Sign Magnitude 24 bit 384 KHz

[ScreamingLordByron]

ok maybe its obvious to me more than to everyone that 24bit PCM1704 has no 24bit monotonicity, 20bit PCM63 has no 20bit and ofcourse 28bit DAM has no incredible (and unnessesary!) 28bit because it is not possible to get such resistors tollerance even with lasser triming.
After your detailed technical explanation ScreamingLordByron, I think no one has any doubts that thing are different then that. I agree 100%.


The problem with market ready R2R DAC chips is that it is always current output DACs (voltage output are taken from internal OPAMP like in AD1865) so to achieve full performance you need to load it with almost zero impedance - opamp, some other transistors stages more or less sound affecting.
Secondly low DAC current output level require very low noise output stage if we do not want to mask LSBs with OS intristic noise..
DAM is voltage output DAC with pritty high voltage swing which can drive many amplifiers directly.
Now, what is the FINAL OUTPUT accuracy of PCM1704 DAC after output stage whatever it is?
There are many PCM1704 implementations on the market and in DIY world, but how many of theme can be compared to 0,01% resistor tollerance TotalDAC in term of sound quality? Why some would choose this $$$ 13-14bit DAC over relatively cheap PCM1704 DAC? DAM is not TotalDAC, but both of theme are 14bit effectively.
So Andrea, maybe it not so obvious but when you compare 14bit monotonicity DAM with most of 17bit monotonicity PCM1704 from the same price level and then from few times more expensive level you will have to answer yourself why this DAC worse on paper is such better in the air.

I don't think a solid output stage is a problem at all for DACs. There are just too many killer opamps out there

But here's a question for you: If you cannot quantify the performance of the DAC you are listening to via measurements, how do you know it was built correctly? In other words, when someone builds a widget, they test it to make sure it was built correctly. If you like the widget because of the way it behaves, and yet behavior cannot be quantified, who is to say it was built correctly? There is no test to be sure, right?

Now, extend that thinking to a product that has so much variability inherent to the design (dozens and dozens of 0.01% resistors). How do you know you didn't get an (effective) 12-bit DAC instead of a 16-bit DAC, just through the roll of the dice?

I can believe the designer with golden ears got the first unit working well on the bench. But unless he is listening to each unit that goes out the door, SOME units will have manufacturing errors. How are those caught if there isn't a test that measures the magic?

 

[soekris]

ScreamingLordByron,

I know it might be too much to expect, but try reading all my post in this thread as you misses a couple of things, especially the whole point of the R-2R DAC being Sign Magnitude. Or maybe you don't get the Sign Magnitude principle.

A couple of points:

The dam1021 R-2R DAC is level linear and low distortion down to the last bit thanks to the Sign Magnitude architecture. I personally don't really care about pure harmonic distortions below 0.1% as nobody can really hear it....

But most people can actually hear details way below the noise level. Remember also that noise measurements are related to bandwidth, look at my FFT plots on the first page of this thread and then imagine a -130 db tone.... So 24 bits and maybe even more is actually useful.

The PCM1704 is probably the best single chip DAC ever made (note "single chip"), but they cost $50 each and are "not recommended for new design". They're afaik the only manufacturer that dared to show a -120 db sinus on the datasheet.

When I get around to find/build a suitable low noise preamplifier I'll do a -130 db 10 Khz sinus tone, and trust me, it will be clearly visible and look like a sinus.

And we actually test every single unit, but just with a single FFT of each channel. We do see a little better with the more precise resistors, but not a 5:1 difference, maybe 2:1.

The dam1021 is also for playing music, not for measuring applications. So I really don't care much about measurements, I only use them as one of my tools and because people ask for measurements....

 

[andrea_mori]

Don't get me wrong, I'm really glad you posted because I'm understanding more versus the previous few posts by Andrea previously which focused on SNR (which has been debunked a few times)

That said, isn't this more an industry wide problem? No one talks about DACs (or even architecture) by effective bits and accuracy. So what we know (from the first post) is that this DAC has 28-bits that (do something) but have a SNR of about 20~21 bits and right now (from your post) with an accuracy of about 14~16 bits (depending on the luck of the draw). So Andrea is saying this should be called a 14-bit DAC... but does the industry call the ESS9018 ~20-bit DAC because it's THD is about 120dB? And we're assuming here that there is audible data in all 24-bits?

Another question I have is that THD+N measurements of the chip DACs aren't directly comparable in a sense, no? You still need to add other parts to it (notably the output stage). Will that THD+N figure remain the same when other passive/active components come into the picture? The measurements in the first post include power supply + driver (i.e. it's something that can be used straight to the amp). So why are we comparing it in a vacuum?

I did never talk about SNR, I said always "THD" (or distortion), see my previous posts.

Sigma delta DAC can easily reach 24 bit accuracy, see post #2287, since they use a different approach (time instead of resistor ladder).

 

[andrea_mori]

So Andrea, maybe it not so obvious but when you compare 14bit monotonicity DAM with most of 17bit monotonicity PCM1704 from the same price level and then from few times more expensive level you will have to answer yourself why this DAC worse on paper is such better in the air.

This is your subjective opinion, respectable, but anyway your opinion.
Can you provide some technical explanation?
Which commercial or diy dac did you compare with this dac?
What kind of comparative method did you follow? Blind listening session?

 

[andrea_mori]

I mean, intuitively you MUST be getting a better DAC with 0.01% resistors versus 0.05%? Right? But why is it better? Why spend the extra money? The answer is THD. 0.01% resistors gets you an extra ~1.5 bits of resolution over 0.05% resistors.

I wonder why no one ask for a 10% resistors tolerance, they are very cheap.
One can reach 4 bit accuracy, but since accuracy doesn't matter...

 

[andrea_mori]

The PCM1704 is probably the best single chip DAC ever made...

This time I agree with you. But don't forget the TDA1541A, if you are looking for a dual dac chip.

 

[ScreamingLordByron]

ScreamingLordByron,

I know it might be too much to expect, but try reading all my post in this thread as you misses a couple of things, especially the whole point of the R-2R DAC being Sign Magnitude. Or maybe you don't get the Sign Magnitude principle.

Yes, I understand it well. But as the last plot I shared showed, it requires extremely tight matching/tracking between the upper/lower dac references, otherwise distortion creeps north very quickly. There is a reason your distortion is where it is....

The dam1021 R-2R DAC is level linear and low distortion down to the last bit thanks to the Sign Magnitude architecture. I personally don't really care about pure harmonic distortions below 0.1% as nobody can really hear it....

The concern isn't the last bit at low levels. The concern is that during an excursion ABOVE zero (so that only one DAC string is active) the error associated with bit 13 or 14 exceeds that of bit 0 due to tolerance stackup.

The distortion either comes from dac reference mismatch/tracking and/or resistor tolerance stackup. If both of these were dead on, your THD would be nil.

Regarding 0.1% for THD...fair enough, everyone has their own threshold, but....

But most people can actually hear details way below the noise level. Remember also that noise measurements are related to bandwidth, look at my FFT plots on the first page of this thread and then imagine a -130 db tone.... So 24 bits and maybe even more is actually useful.

...It's odd you argue nobody can hear moderately poor level of distortion, but then make the case that someone can hear 130 dB below their system peak...

The PCM1704 is probably the best single chip DAC ever made (note "single chip"), but they cost $50 each and are "not recommended for new design". They're afaik the only manufacturer that dared to show a -120 db sinus on the datasheet.

You sell what you got, right They made a part that kicked A$$ at low level zero crossings but was a bit behind the times in terms of thd.

When I get around to find/build a suitable low noise preamplifier I'll do a -130 db 10 Khz sinus tone, and trust me, it will be clearly visible and look like a sinus.

And then can you find a system where you can actually hear the tone? If system max is 110 dBSPL, then 130 dB down from that is well below the threshold of hearing...

And we actually test every single unit, but just with a single FFT of each channel. We do see a little better with the more precise resistors, but not a 5:1 difference, maybe 2:1.

Thanks for the data point. Can you relate that to THD % or dB?

The dam1021 is also for playing music, not for measuring applications. So I really don't care much about measurements, I only use them as one of my tools and because people ask for measurements....

Can't disagree there. My original post was to a guy that wanted to cal out the errors...I explained why it would be hard, the sources of errors, and said enjoy it for what it is...I think this is a cool project.

 

[new2hifi]

Numerical Superiority?

............Well designed output stages have vanishing small distortion. A good op amp today is -135 dB, which is much smaller than even a top of the line DAC. So, as far as line levels go, your line performance should be very, very close to your raw DAC performance level.

............Well, you are absolutely free to get a dac with such that accuracy, but is not so obvious that you prefer a 13-14 bit accuracy dac rather than a 16-17 bit dac, the PCM1704. That's an R2R dac, not a sigma-delta dac. And it's a sign magnitude DAC, not a two's complement dac. And it uses a single ladder network, not a double ladder. And network resistors and switches are imprinted in the same silicon wafer, that (typically) ensures better thermal stability.

Not so obvious.

andrea_mori & ScreamingLordByron:

I don't have this dam dac yet so whatever I am writing is not in support of it.

At least I am not on this diy forum to have numerically superior or measuring DAC. I used to have Mark Levinson 360S (SHARC DSP, FIFO, 4xPCM1704, HDCD decoder, opamp...etc). Once I heard non-oversampling, no filter TDA1541A based AMR CD-77, I knew what was wrong with Mark Levinson 360S.

I have very scientific attitude, however, time and time again I have been disappointed to realize that my ears prefer to just Listen and not Measure.

I understand that PCM1704 must be very good. But I am unable to find even a single recent diy effort (my technical skills are limited to do it on my own) towards this chip. Where is that masterfully executed dream board for DIYers? Please point me the that forum!

 

[ristar]

Can't disagree there. My original post was to a guy that wanted to cal out the errors...I explained why it would be hard, the sources of errors, and said enjoy it for what it is...I think this is a cool project.

Cheers, thank you for explaining all of that in a manner we can all understand.

One last point I'll make to sort of "close the loop" so to speak with regard to output stages for chip DACs. If the various setups in this forum (all the various DACs) are to be referenced, opamps used as output stages are often not the setup of choice? There are tube buffers... discrete buffers... transformers etc. Hence my point that it's quite likely that THD+N figures won't remain quite that high.

I guess that says something also about how much THD+N figures mean to these builders of customized DACs.

 

[Marek]

But here's a question for you: If you cannot quantify the performance of the DAC you are listening to via measurements, how do you know it was built correctly?

Now, extend that thinking to a product that has so much variability inherent to the design (dozens and dozens of 0.01% resistors). How do you know you didn't get an (effective) 12-bit DAC instead of a 16-bit DAC, just through the roll of the dice?

I didn't make any monotonicity tests of DAM nor PCM1704, AD1865, AD1853 which I used in my DIY DACs.
IMO reducing non-monotonic 24 or 28bit dac to real 12-14bit DAC isn't quite fair. What do you think what is the reason for 24bit in PCM1704 though it offers "only" 17bit. Do you think real 17bit DAC (if there exist) will perform the same as fake 24bit PCM1704 (17bit effective) in term of low level signals?

Anyway:
14 bit DAC offers 84dB dynamics
16 bit DAC offers 96dB
17 bit DAC offers 102dB

My system for sure has worse THD than even 14bit DAC:
6n30p in SRPP - VT25 (choke loaded) - GM70 - SAC OPT - DIY monitors (ETON 7-360 + fountek CD2.0)

What is dynamic of rest of your out systems? 80? 90? 100dB?

This is your subjective opinion, respectable, but anyway your opinion.
Can you provide some technical explanation?
Which commercial or diy dac did you compare with this dac?
What kind of comparative method did you follow? Blind listening session?

As I said before IMO the reason is that there is no output stage at all - there is no need to amplify low level signals - we got 2V directly from ladder, no need to do I/V conversion using "killer" OPAMPs (which kills sound).
Oversampling filter built into DAC chip makes huge difference - in DAM OS filter and room correction filters are programmable.
Maybe there are other reasons that I do not realize in this moment.

I compare DAM to my own built DACs:
- AD1853 (+AD1896 ASRC) + OPA output stage (AD797, OPA627) silver mica, OSCON nad elna ROA caps
- 4xPCM1704 balanced, 4 pcs low noise local regulators per chip, stepup transformer output LL1931
- simple NOS AD1865 balanced, transformer output LL1931

and to commercial (commercial blocks assembled by myself) : Twisted Pear Buffalo II + Trident regulators set + Ian's Isolator and XO board (2x Crystek CCHD957, I2S synchronous reclocking with POTATO FF)

At least I am not on this diy forum to have numerically superior or measuring DAC. I used to have Mark Levinson 360S (SHARC DSP, FIFO, 4xPCM1704, HDCD decoder, opamp...etc). Once I heard non-oversampling, no filter TDA1541A based AMR CD-77, I knew what was wrong with Mark Levinson 360S.

Unfortunetally this is true that in audio there is no 2+2=4. When you take two DACs both with same level of THD, same bandwidth, same noisefloor thay can still sounds very different. Why? IMO Psychoacustics - the numbers used to describe technical parameters contains very little information about how it sounds which is probably impossible to catch in numbers.
Some can prefer 17-bit true monotonicity DAC just because he knows that it is linear down to 17 bit but he can be suprised how much better can perform other one with worse *THD* - one of the many parameters than can affect SQ (like time related distortions eg jitter effect, digital filter impulse time smear, ringing etc)

 

[CFT]

Del..

 

[andrea_mori]

andrea_mori & ScreamingLordByron:

I don't have this dam dac yet so whatever I am writing is not in support of it.

At least I am not on this diy forum to have numerically superior or measuring DAC. I used to have Mark Levinson 360S (SHARC DSP, FIFO, 4xPCM1704, HDCD decoder, opamp...etc). Once I heard non-oversampling, no filter TDA1541A based AMR CD-77, I knew what was wrong with Mark Levinson 360S.

I have very scientific attitude, however, time and time again I have been disappointed to realize that my ears prefer to just Listen and not Measure.

I understand that PCM1704 must be very good. But I am unable to find even a single recent diy effort (my technical skills are limited to do it on my own) towards this chip. Where is that masterfully executed dream board for DIYers? Please point me the that forum!

I have never heard the CD-77, but as far as I read probably it's one of the best machine ever made. I own an old Naim CD3, TDA1541A based, it sounds great, with its poor 16 bit.
Firstly, in my opinion the best dac chip ever made is just the TDA1541A, not the PCM1704. Also I prefer the "system" way (like CD-77) rather than separate source and dac.
I don't know if there are diy implementations of the PCM1704, but I believe it's pretty simple, except for the IV output stage.
I'm planning to design myself a little board to accommodate the 1704 in the future (I promised this board to a friend).
Another question is the I to V converter, since it's a current output dac, that typically loves to see very low impedance at its output.
I published a IV schematic on another thread, a vacuum tube converter where the dac sees around 1 ohm at its output. There is also a solid state version where the dac sees less then 0.1 ohm. Without any IV resistor.
They could be used also for other current output dac like the TDA1541A.
I have not yet found the time to build the above circuits, but if you or someone else were interested to build a prototype, I could publish the schematics and assist the prototyping.

 

[ScreamingLordByron]

I didn't make any monotonicity tests of DAM nor PCM1704, AD1865, AD1853 which I used in my DIY DACs.
IMO reducing non-monotonic 24 or 28bit dac to real 12-14bit DAC isn't quite fair. What do you think what is the reason for 24bit in PCM1704 though it offers "only" 17bit. Do you think real 17bit DAC (if there exist) will perform the same as fake 24bit PCM1704 (17bit effective) in term of low level signals?

THD and THDN will largely depend on your linearity. Dynamic range and SNR do not (assuming harmonics aren't an issue at -60). So, it makes perfect sense to build a 24-bit converter with 17 bits of effective resolution, because you can still offer up an impressive dynamic range figure.

So, yes, there is a performance difference between a PCM1704 with just 17 bits and a PCM1704 with 24 bits and limited linearity. The latter will have better DR and SNR performance in most (all?) cases.

But they will both have comparable THD and THDN, assuming harmonics are quite a ways above the noise floor.

What is dynamic of rest of your out systems? 80? 90? 100dB?

A system that is silent (no audible noise) while playing a silent audio file isn't uncommon these days. So, perhaps 100 or 105? I have trouble hearing the noise while playing a silent audio file on my smartphone with IEMs and max volume.

Unfortunetally this is true that in audio there is no 2+2=4. When you take two DACs both with same level of THD, same bandwidth, same noisefloor thay can still sounds very different. Why? IMO Psychoacustics - the numbers used to describe technical parameters contains very little information about how it sounds which is probably impossible to catch in numbers.

I think you are either in one camp or the other: As transparent as possible OR sounding as good as possible. I want to hear what the producer heard sitting at the mixing desk. And very likely, there were countless sigma delta DACs and ADCs in the process of making the music (that I listen to) because they are transparent. The guitar player played through an effects rack laden with SD ADCs and DACs, the vocals were captured using SD ADCs, and pumped through outboard effects with more SD ADC/DAC, brought back into the PC again via ADC, etc. In other words, if music I like involved dozens of round trips through sigma delta ADCs and DACs during production, is one more on the way to my ears going to hurt? Not likely.

Transparent is easy to measure. "sounding good" is not. Some like the bass and treble jacked and the mid scooped. Some like to turn on the "stadium setting" when listening to Dark Side of the Moon. Who am I to judge? Do whatever you think makes the music sweeter.

Measuring transparency is easy. That simply means that everything we can measure is measured. That the impacts the sound the least (taking into account what we know about the human ear) wins.

Sigma Delta ADCs and DACs are more transparent than R2R. Period. Opamps are more transparent than tubes. If you want to say there's an objectionable yet immeasurable quality to sigma delta, I can't argue with that anymore than I can't argue with my brother telling me "stadium mode" on dark side of the moon sounds better. To each his own.

 

[TNT]

ScreamingLordByron,

I know it might be too much to expect, but try reading all my post in this thread as you misses a couple of things, especially the whole point of the R-2R DAC being Sign Magnitude. Or maybe you don't get the Sign Magnitude principle.

Care to comment on this?

//

 

[ScreamingLordByron]

Care to comment on this?

//

I did in 2447

 

[danny_66]

Sigma Delta ADCs and DACs are more transparent than R2R. Period. Opamps are more transparent than tubes.

You better start reading the Filter brewing thread, with different filters the transparency varies because of the time smearing.
There's more to transparency than just very good THD.

 

[Marek]

ScreamingLordByron,

The difference in SQ may be caused of how we interpret measurments.

lets get Soekris 0,05% resistor version measurments:

An externally hosted image should be here but it was not working when we last tested it.

An externally hosted image should be here but it was not working when we last tested it.

0dBFS THD+N=0,0063%
-60dBFS THD+N=0,37% <- This is too large value taking into account THD=0,027%. There is something going on with noise under 500Hz (not included in FFT graph) because almost all harmonics are under -140dB.

Reffering to MSB Technology discrete DAC:

- look at THD measurments page 6 (page 46 of magazine - pretty similar to Soekris so probably they use similar resistor tollerance level) http://www.msbtech.com/reviews/msb_platinum_review_Australian_Hifi_Magazine.pdf

than from 'About the measurments' chapter:

Signature DAC IV:

An externally hosted image should be here but it was not working when we last tested it.

0dBFS THD+N=0,002%
-60dBFS THD+N=0,015%


from TI PCM1704 datasheet (page 4): http://www.ti.com/lit/ds/symlink/pcm1704.pdf
0dBFS THD+N(24bit)=0,001% and THD+N(16bit)=0,0015%
-60dBFS THD+N(24bit)=0,6% and THD+N(16bit)=0,9%

from TI PCM1738 datasheet (page12): http://www.ti.com/lit/ds/sbas174c/sbas174c.pdf
0dBFS THD+N=0,0004%
-60dBFS THD+N=0,15%

MSB Tech interpretation of above graph:

The graph demonstrates the percentage of Total Harmonic Distortion (THD) in reference to the volume level of a recorded signal. There are three DACs on this chart, the Platinum Signature Ladder DAC, The Burr-Brown PCM 1738 (a high performance delta-sigma DAC), and another ladder DAC by Burr-Brown, the PCM 1704. The Delta-Sigma DACs do spec very well but they do not tend to sound as good as the Ladder DACs. Take a look at the horizontal 0.1% THD line of the graph. This line is the general rule of thumb for the maximum amount of distortion in a signal that is undetectable. If you follow the red line straight down from where each of these DACs cross the 0.1% THD line you can see the signal level at which the respective DACs output will have audible distortion. Playing a 1 kHz signal at ~ 48dB below 0dB (0dB being the loudest reference volume on a CD) , the PCM 1704 DACs output signal has an audible 0.1% distortion. At -57 dB the PCM 1738 will output 0.1% of the signal distorted. The MSB Signature DAC is rated at more than 78dB below 0dB for the same signal, more than 20dB quieter than the competition!

Soren, can you please post full bandwidth FFTs of all versions of DAM?

 

[nightsky]

Interesting new approach:MQA

I read an interview of J.R.Stuart with R. Harley about MQA in the new TAS. What's interesting regarding this ongoing discussion about what's the better DAC and the importance of the sheer number of bits is the change in perspective on technical parameters if the reality of human hearing gets involved.

Here are some highlights:
..the type of information received by the brain is more important than the sheer quantity of information.. ...the massive number of bits generated by 192kHz/24-bit encoding are useless baggage rather than conveyors of real musical information...
...time is more important than frequency because of the way we hear things...
Bits and sample rate are ‘comparatively meaningless numbers from the coding'. ‘Resolution is better described by as shortest event that can be carried by the system by means of which we can tell two things apart.’

 

[andrea_mori]

I read an interview of J.R.Stuart with R. Harley about MQA in the new TAS. What's interesting regarding this ongoing discussion about what's the better DAC and the importance of the sheer number of bits is the change in perspective on technical parameters if the reality of human hearing gets involved.

Here are some highlights:
..the type of information received by the brain is more important than the sheer quantity of information.. ...the massive number of bits generated by 192kHz/24-bit encoding are useless baggage rather than conveyors of real musical information...
...time is more important than frequency because of the way we hear things...
Bits and sample rate are ‘comparatively meaningless numbers from the coding'. ‘Resolution is better described by as shortest event that can be carried by the system by means of which we can tell two things apart.’

Interesting approach. Let's say this change a bit the perspective.

Firstly: resolution doesn't matter.. why not use a good 16 bit dac, such as the 1541A? IMHO, the best dac ever built.

Second: time is more important.. why not use a good master clock? at least a Crystek oscillator or something better (phase noise at least 20 db better than the SI5xx).

Take a 1541A, feed it directly by the master clock, then slave the source with the same clock.
Result: time perfect machine (or so).

 

[soekris]

Soren, can you please post full bandwidth FFTs of all versions of DAM?

Sorry, but:

1) Don't have time, especially as I view these discussions about THD as waste of time, as even MSB say, harmonic distortion below 0.1% is not really an issue, despite what all these ultra low distortion fanatics claim....
2) Don't have good enough equipment, my EMU-0404 is way too noisy, especially above 20 Khz on higher sample rates....

Somebody send me an AP-555 and I'll reconsider