Soekris DAM1021 R2R DAC Measurements

[March Audio]

Thanks Don, but what about the absolute level linearity, what is the controlling factor there?

 

[DonH56]

It's... complicated, and I don't have time to do a good job right now, sorry (waiting for a test to finish).

Absolute linearity to me means absolute step size per the IEEE ADC Standard (1241). Absolute linearity discussed here seems to be more focused on deviation at low levels. Bu fundamentally absolute linearity is how well you match the steps in a DAC. All 2^n steps for an n-bit DAC should ideally be exactly the same size under all circumstances. Any deviation from that ideal adds nonlinearity.

I'll start by just waving my hands and saying each step in a DAC must be precisely 1/2^n of full-scale for an n-bit DAC (there's actually one less step but let's ignore that). Any step deviating from that ideal step size means the DAC's output will be nonlinear. We can then break it into DC accuracy, i.e. every step measured as you slowly ramp through them, and AC accuracy, which gets more complex because various effects occur. One is that the DAC's slew rate (how fast it can change from one step to another), bandwidth, and other parameters may cause the AC signal to deviate from its DC performance. Make a perfect DC DAC, and when you send a real signal through it, it won't be quite as good because the signal gets modified by limited bandwidth and such. The other side of AC accuracy is how the DAC switches from one step to another; switches are not ideal and do not switch at the same time and without modifying the signal a little bit. Because not all switches change exactly simultaneously in a real DAC, there will be some glitches added. Furthermore, switches have their own parasitic components that add and take away from the signal as they switch. All that (among other things) causes glitches at the output that are usually related to the signal level, and that causes the output to be nonlinear since the output depends on how the switch acts for various signal levels.

The zero-crossing problem is that, as you cross zero in an R2R DAC, one-half the switches turn off, and the other half turn on. That is a lot of switching going on right where the signal is at its minimum! Large spikes result and have to settle out, and at the output of a DAC those spikes are just passed along to the rest of the system. There are usually filters but of course they may misbehave when you hit them with a large spike, and even if they don't, they will usually integrate the energy so a large, fast (short) glitch becomes a smaller, slower (longer) glitch. Things like unary cells to break up the number of switches at the zero crossing, or signed magnitude which changes the way the DAC switches at the zero crossing, are aimed at reducing those nasty glitches. And even without glitches, if the sum of half the cells does not match the sum of the other half to within an lsb (single step), then there will be a discontinuity right at the zero crossing. It is somewhat like the zero-crossing distortion in class B amplifiers and why AB is so popular. Class AB keeps everything on around the zero crossing, running in pure class A for low-level signals, and switches things off (class B) to save power as the signal gets larger (if the signal is positive, the negative side idles along at a low level until it is needed again).

Lot of hand waving and a lot of words...

HTH - Don

 

[Superdad]

That’s very clear. Thank you Don.

 

[amirm]

I could send Amir one of my DACs, there seems to be a focus on cheaper DACs here, and the cheapest Schiit R-2R one didn't do that good, so a soekris dac1321 would be a good candidate to show how good a low cost R-2R DAC can be....

I would love to measure it! Our focus on cheaper DACs is due to level of funding we have. I can only take so much food off the plate to put toward this cause. But I have tested expensive DACs like my own Exasound E32 ($3,400) and have a few more coming soon in the $2,5000 to $5,000 level. Please PM me if interested and I will give you my shipping information.

 

[soekris]

So why arent all dacs sign magnitude if the methodology eliminates linearity errors? Thats a holy grail.

Because a Sign Magnitude DAC is basically twice as large and R-2R DACs are already expensive, and it's really only for audio you need it....

R-2R DACs disappeared as Delta Sigma DACs became available at low cost, R-2R DACS chip need expensive precise process technology and laser trimming, while Delta Sigma DACs can be manufactured on simple low cost process technology. R-2R DACs are coming back now as they can be made discrete as ultra precise thin film resistor has become (relatively) inexpensive....

Burr Browns last R-2R Sign Magnitude chips are by many regarded the best sounding ever made, but are not manufactured any more, the Audiophile market was too little to sustain chip manufacturing.

Ok, Im going to plead ignorance here. Lets discuss how the dac creates the output voltage. This is for the purpose of understanding what potential sources of error there could be in the process, because at the moment you are essentially saying there are no potential sources of error. Please excuse me for being a little sceptical about that.

Just simply describe what the dac does - a voltage or current is switched across a resistor for example.

There are other sources of errors, but a well implemented R-2R Sign Magnitude DAC do not have problems at low levels.

A R-2R DAC is actually pretty simple, you take your typical 24 bits and send them directly to a R-2R network which convert them to voltage (or current). The problem with Audio is that it's centered around half the voltage, so each time it crosses signal zero, no matter the signal level, all bits change, like from 10000000xx to 01111111xx, meaning all bits matter. The Sign Magnitude fixes that by having a dedicated R-2R DAC for the positive and negative signal, so it for a full level its the same as a non Sign Magnitude, but for lower levels, and most audio have a lot of lower levels, t.ex. at -42 dB level the positive goes 00000001xx to 00000000xx while the netative goes 11111111xx to 11111110xx. All the MSB don't change, so they don't contribute linearity errors, that why distortion on a sign Magnitude DAC is relative to the actual signal level.

There are way of improving a regular R-2R DAC, Analog Devices top R-2R AD1862 add a offset, so it move the zero crossing down a little, and Analog Device top industrial ones, like the AD5791 are extremely precise and also very expensive and still have zero crossing problems, but its that far down it doesn't matter that much as long as you don't waste bits, like by using a digital volume control....

The Soekris DACs don't have any zero crossing problems at all, that's why they sound so great, even when using a digital volume control.

The Delta Sigma DACs are a completely different thing, started being one bit, but nowadays three to five bits DACs, to get the needed resolution they do some fancy modulation at high speed with all kinds of side effects, typically resulting in very nice specs with simple test signals, but fails, some miserable, when playing actual music....

 

[RayDunzl]

It is somewhat like the zero-crossing distortion in class B amplifiers

Somewhat.

Just for the record:

Zero crossing for a solid state Class B gaps a range of (about) 1.4V around 0V -- .7v (bias hole) in both directions, and may be rather audible when listening at a level of a couple of volts (normal quiet listening level).

The waveform of a Class B amp's music at a couple of volts amplitude would look something like this (below), with everything that should be there from +0.7 to -0.7 lost (flat area at 0V). A high frequency glitch would probably occur (at the speaker) at each transition from wave to flat and flat to wave. I'd expect it to sound a little buzzy.

From what I see, DAC zero crossing problem would appear to be way way way below that both in calculation, displayed waveform, and analytical measurement.

---

From MSB, on sign magnitude ladders: http://www.msbtechnology.com/faq/why-ladder-dacs/

"But most important to MUSIC rather than TEST SIGNALS, and very different from Delta Sigma DACs the MSB DAC module are most accurate with signals crossing zero, where music actually exists."

Huh?

---

Not to say I wouldn't like to hear a nice sign magnitude ladder DAC, so I could judge how deficient my inferior delta-sigma DAC is.

 

[Blumlein 88]

snippage.......
The Delta Sigma DACs are a completely different thing, started being one bit, but nowadays three to five bits DACs, to get the needed resolution they do some fancy modulation at high speed with all kinds of side effects, typically resulting in very nice specs with simple test signals, but fails, some miserable, when playing actual music....

Can you expand upon this concluding statement of yours. How do delta-sigma dacs work on test signals yet fail on music? What happens with the music, what gets distorted, how, by how much?

 

[RayDunzl]

PS: I do think this is a really nice looking board/product... Makes me want one... Good job...

 

[Blumlein 88]

@BE718 What did you decide about the Quant Asylum noise floor? Do you think it moves with signal level?

 

[Wombat]

Somewhat.

Just for the record:

Zero crossing for a solid state Class B gaps a range of (about) 1.4V around 0V -- .7v (bias hole) in both directions, and may be rather audible when listening at a level of a couple of volts (normal quiet listening level).

The waveform of a Class B amp's music at a couple of volts amplitude would look something like this (below), with everything that should be there from +0.7 to -0.7 lost (flat area at 0V). A high frequency glitch would probably occur (at the speaker) at each transition from wave to flat and flat to wave. I'd expect it to sound a little buzzy.

View attachment 10474



From what I see, DAC zero crossing problem would appear to be way way way below that both in calculation, displayed waveform, and analytical measurement.

---

From MSB, on sign magnitude ladders: http://www.msbtechnology.com/faq/why-ladder-dacs/

"But most important to MUSIC rather than TEST SIGNALS, and very different from Delta Sigma DACs the MSB DAC module are most accurate with signals crossing zero, where music actually exists."


I would hope that Class B amp designers utilise 'pre-biasing' in their designs.

See here:https://www.electronics-tutorials.ws/amplifier/amp_7.html

 

[DonH56]

Somewhat.

Just for the record:

Zero crossing for a solid state Class B gaps a range of (about) 1.4V around 0V -- .7v (bias hole) in both directions, and may be rather audible when listening at a level of a couple of volts (normal quiet listening level).

The waveform of a Class B amp's music at a couple of volts amplitude would look something like this (below), with everything that should be there from +0.7 to -0.7 lost (flat area at 0V). A high frequency glitch would probably occur (at the speaker) at each transition from wave to flat and flat to wave. I'd expect it to sound a little buzzy.

View attachment 10474



From what I see, DAC zero crossing problem would appear to be way way way below that both in calculation, displayed waveform, and analytical measurement.

---

From MSB, on sign magnitude ladders: http://www.msbtechnology.com/faq/why-ladder-dacs/

"But most important to MUSIC rather than TEST SIGNALS, and very different from Delta Sigma DACs the MSB DAC module are most accurate with signals crossing zero, where music actually exists."

Huh?

---

Not to say I wouldn't like to hear a nice sign magnitude ladder DAC, so I could judge how deficient my inferior delta-sigma DAC is.

Yeah. The "dead zone" depends on the type of devices and bias levels. Feedback pretty much eliminates the visible notch so all you see typically is a little spike at the zero crossing. Which push-pull class A amps can also have, just usually very small.

 

[March Audio]

Because a Sign Magnitude DAC is basically twice as large and R-2R DACs are already expensive, and it's really only for audio you need it....

R-2R DACs disappeared as Delta Sigma DACs became available at low cost, R-2R DACS chip need expensive precise process technology and laser trimming, while Delta Sigma DACs can be manufactured on simple low cost process technology. R-2R DACs are coming back now as they can be made discrete as ultra precise thin film resistor has become (relatively) inexpensive....

Burr Browns last R-2R Sign Magnitude chips are by many regarded the best sounding ever made, but are not manufactured any more, the Audiophile market was too little to sustain chip manufacturing.



There are other sources of errors, but a well implemented R-2R Sign Magnitude DAC do not have problems at low levels.

A R-2R DAC is actually pretty simple, you take your typical 24 bits and send them directly to a R-2R network which convert them to voltage (or current). The problem with Audio is that it's centered around half the voltage, so each time it crosses signal zero, no matter the signal level, all bits change, like from 10000000xx to 01111111xx, meaning all bits matter. The Sign Magnitude fixes that by having a dedicated R-2R DAC for the positive and negative signal, so it for a full level its the same as a non Sign Magnitude, but for lower levels, and most audio have a lot of lower levels, t.ex. at -42 dB level the positive goes 00000001xx to 00000000xx while the netative goes 11111111xx to 11111110xx. All the MSB don't change, so they don't contribute linearity errors, that why distortion on a sign Magnitude DAC is relative to the actual signal level.

There are way of improving a regular R-2R DAC, Analog Devices top R-2R AD1862 add a offset, so it move the zero crossing down a little, and Analog Device top industrial ones, like the AD5791 are extremely precise and also very expensive and still have zero crossing problems, but its that far down it doesn't matter that much as long as you don't waste bits, like by using a digital volume control....

The Soekris DACs don't have any zero crossing problems at all, that's why they sound so great, even when using a digital volume control.

The Delta Sigma DACs are a completely different thing, started being one bit, but nowadays three to five bits DACs, to get the needed resolution they do some fancy modulation at high speed with all kinds of side effects, typically resulting in very nice specs with simple test signals, but fails, some miserable, when playing actual music....

Thanks Soekris its great that you are willing to engage here. Btw, as I have stated elsewhere on the forum, I do like the sound of the dac.

So, would I be correct in saying that the harmonic distortion we see is due to resistor mismatch between the positive and negative ladders? We already know it improves withe the higher tolerence resistors.

 

[DonH56]

For that DAC, more likely just the usual ladder bow and transistor/buffer nonlinearity. Google DAC INL and DNL (integral and differential nonlinearity). There is usually a little "warp" or "bow" in the ladder from top to bottom and that adds harmonic distortion. Interestingly enough, duty cycle variation among the switches is also a source of error. The top and bottom switches don't toggle as often as the ones in the middle and that can cause thermal offsets. I wrote a program to analyze it for a BJT-based ADC many years ago. Buffers also get less linear away from the middle of their range and that also adds distortion. Etc.

As an aside, it would be interesting to compare performance among their various resistor options. One lsb at 16 bits is 0.0015% so clearly there is some trimming or segmenting going on to hit 16+ bits of linearity.

Edit: Agree it is awesome to have someone from the company contributing!

Edit 2: I was sooo tempted to spout off on delta-sigma thangs but am eagerly awaiting his response and holding my tongue. It's tough shutting up...

 

[March Audio]

@BE718 What did you decide about the Quant Asylum noise floor? Do you think it moves with signal level?

Havent drawn a conclusion yet. As you know it produced plots extremely similar to the stereophile ones but I also want to see the results Amir gets with my explorer2. Ultimately I really need to build a 1kHz notch filter to prove one way or the other. It would be interesting to see Amirs results with and without the APs notch filter.

 

[March Audio]

For that DAC, more likely just the usual ladder bow and transistor/buffer nonlinearity. Google DAC INL and DNL (integral and differential nonlinearity). There is usual a little "warp" or "bow" in the ladder from top to bottom and that adds harmonic distortion. Interestingly enough, duty cycle variation among the switches is also a source of error. The top and bottom switches don't toggle as often as the ones in the middle and that can cause thermal offsets. I wrote a program to analyze it for a BJT-based ADC many years ago. Buffers also get less linear away from the middle of their range and that also adds distortion. Etc.

As an aside, it would be interesting to compare performance among their various resistor options. One lsb at 16 bits is 0.0015% so clearly there is some trimming or segmenting going on to hit 16+ bits of linearity.

Edit: Agree it is awesome to have someone from the company contributing!

Edit 2: I was sooo tempted to spout off on delta-sigma thangs but am eagerly awaiting his response and holding my tongue. It's tough shutting up...

Thanks Don, what Im really getting at here is there any reason to think the sign magnitude design is inherently immune from low level linearity errors. Should you always expect if you input a -138dBfs signal it will output precisely that. This seems to be the implication from Soekris.

Edit, just seen your reply above. So essentially there still are multiple factors which prevent ultimate linearity. This was my understanding. My pleading ignorance was really more to encourage discussion of this subject. Sorry

 

[Blumlein 88]

Havent drawn a conclusion yet. As you know it produced plots extremely similar to the stereophile ones but I also want to see the results Amir gets with my explorer2. Ultimately I really need to build a 1kHz notch filter to prove one way or the other. It would be interesting to see Amirs results with and without the APs notch filter.

I'm tired so hopefully not missing something. Couldn't you check this by sending a signal that starts out very low and sweeps to maximum value over a few seconds? See if the noise floor follows the signal. Try that at different gain settings on the Quant Asylum.

Of course you need a DAC that doesn't have modulated noise floors for that.

Or another idea. Take a DAC with healthy output and decent SNR. Reduce the level of a max level tone via an analog pot or similar. Reducing the signal x db should reduce the noise floor by the same db until the test ADC has an equal or higher noise floor.

You could even combine both tests above for that matter. Run a low level to high level sweep. Turn down level in an analog pot, and run the sweep at lower volumes to spot where the noise in the signal begins to interact with the ADC.

 

[March Audio]

I'm tired so hopefully not missing something. Couldn't you check this by sending a signal that starts out very low and sweeps to maximum value over a few seconds? See if the noise floor follows the signal. Try that at different gain settings on the Quant Asylum.

Of course you need a DAC that doesn't have modulated noise floors for that.

Or another idea. Take a DAC with healthy output and decent SNR. Reduce the level of a max level tone via an analog pot or similar. Reducing the signal x db should reduce the noise floor by the same db until the test ADC has an equal or higher noise floor.

You could even combine both tests above for that matter. Run a low level to high level sweep. Turn down level in an analog pot, and run the sweep at lower volumes to spot where the noise in the signal begins to interact with the ADC.

Thats the problem. Is it dac or adc? The stereophile data showed the same noise floor movement with their AP as I did with the QA.

I want to build a notch filter anyway, it will improve certain measurements.

 

[RayDunzl]

I want to build a notch filter anyway, it will improve certain measurements.

For grins, what if you notch them digitally just to see if the noise floor does the same dance?

I'm assuming there are some digits to play with between the ADC and the Analyzer, I guess.

 

[soekris]

Can you expand upon this concluding statement of yours. How do delta-sigma dacs work on test signals yet fail on music? What happens with the music, what gets distorted, how, by how much?

Now we're moving into subjective statements....

I, as I sell only R-2R DACs, believe that Delta Sigma DACs have all kind of side effect, some large and clear, like idle tones, other smaller, all a result of the modulation and noise shaping process the they need to do to get something useful out of those one to five bits DACs. Those tings are hard to measure while simple signals, like a Sine, is much easier for a Delta Sigma DAC to handle and result in those nice numbers, while there isn't really any good way to measure with real music, except your ears....

While the R-2R DACs are pure, directly converting from the input PCM data to Audio output, without needing to do any tricks at all, resulting in a more natural sound. It's not without reason that some of the considered best DACs in the world are R-2R types....

 

[soekris]

Thanks Soekris its great that you are willing to engage here. Btw, as I have stated elsewhere on the forum, I do like the sound of the dac.

Thanks for the nice words. This place is interesting, seems to be at a high technical level, although it also seems to be too much focused on what you can measure....

So, would I be correct in saying that the harmonic distortion we see is due to resistor mismatch between the positive and negative ladders? We already know it improves withe the higher tolerence resistors.

Not between the positive and negative itself, that would just result in increased even harmonics.... The distortion come from general mismatch between the resistors, and will improve with more precise resistors, although the ones I use are typical much better than the max specs....