Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: DAC IV stages (Read 49912 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

DAC IV stages

How important are DAC IV stages in the design of a DAC? I've heard they can cause audible differences. Is this true? I'm not a DAC designer, but I assume there are those who understand the technical merits. Is it a big deal?

DAC IV stages

Reply #1
The I-V (current to voltage) conversion stage of a digital to analog converter is just one of several critical analog stages that can impact the performance of a DAC.  Two products may share the same D/A conversion IC, but may have very different performance.  Each of the analog stages that follow the DAC IC contribute noise and distortion to the audio.  With careful design, these artifacts can be minimized to levels that are well below audibility.  For example, Benchmark's new DAC2 converter uses a ES9018 DAC IC and achieves a 129 dB SNR A-weighted, and distortion that is better than -109 dB.  This implies that distortion is below the threshold of hearing until peak music levels reach 109 dB SPL.  Furthermore, noise would not exceed the threshold of hearing until peak music levels reach a damaging 129 dB SPL.  For all practical purposes, the noise and distortion produced by the DAC2 are inaudible.  The DAC2 probably defines the current state of the art.

However, I have measured consumer DVD players with ES9018 DAC ICs that are as much as 20 dB nosier than the DAC2.  Distortion measured about -80 dB on the DVD player.  The vast differences in performance are due to differences in the analog output stages.  The DAC2 and the DVD player share the same DAC IC, but the measured performance is very different.

Analog output stages include:

1) I-V conversion
2) Low-pass filters
3) Gain stage
4) Output buffers
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #2
Thanks for the reply John. I'm getting a few pro-DAC users citing IV stages in DAC design as some holy grail cause of significant audio differences. I'm trying to attack the argument in as many ways as I can. The stumbling block right now is the lack of published DBTs. At least I can't seem to find any positive or negative results.


DAC IV stages

Reply #3
Designing an I-V converter with a 130 dB SNR is not trivial.  Impedances need to be very low, and this puts significant demands on the opamp.  The opamp must have very low EIN and must be able to deliver high current drive while maintaining low distortion.  There are about 2 opamps that will work in this application.  Resistors must be metal film (to keep distortion low). 

One complication is that high-frequency switching noise is always present on the current outputs of sigma-delta converters.  This high-frequency noise must be bypassed with rather large NPO or COG capacitors before entering the I-V converter.

There are many opportunities to cut costs, and there are many opportunities to make mistakes when designing an I-V conversion stage.  The I-V conversion stage may be more important than the choice of conversion IC.
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #4
But it is largely irrelevant if it is competently designed? Modern DACs... should largely not have any issues in this department? I realise that is pure speculation on my part.

DAC IV stages

Reply #5
But it is largely irrelevant if it is competently designed? Modern DACs... should largely not have any issues in this department? I realise that is pure speculation on my part.


I remember a user posting here on hydrogenaudio years ago with scope traces of a DAC that didn't appear to even have a reconstruction filter (or perhaps very poorly designed).  This was on the line-out of a name-brand laptop, if I remember correctly.  I'm not familiar with the ins and outs of the I/V circuit, but I suspect it's easy to get wrong either due to poor design or more likely cost-cutting.  We can debate the audibility of the noise and distortion generated by an I/V that's merely "good enough" and not "excellent".  But practically speaking, if the analog stages of the DAC are significantly noisier than the converter itself, you've overspec'd the converter in your product.

DAC IV stages

Reply #6
Take a look at published specifications for D/A converters, DVD players, sound cards, iPod, etc.  Many only have an 80 to 90 dB SNR.  This means that the noise will be audible in an A/B/X test if the peak music levels exceed 80 to 90 dB SPL (easily achieved).

Worse yet, most D/A converters are preceded by a digital volume control.  If 20 dB of digital attenuation is dialed in (very typical), the SNR will degrade by 20 dB.  In many systems, the user will need to apply 20 dB of digital attenuation to achieve a playback level that peaks at 100 dB SPL.  Under these conditions the converter noise may reach 30 to 40 dB SPL.  It is not hard to hear the noise floor drop from 30 of 40 dB SPL to less than 0 dB SPL in an A/B/X test. 

I have an ABX tester, and it is very easy to pick out D/A converters that have high noise floors.

If one wishes to use a digital volume control, the D/A converter will need an extra 20 dB SNR to keep noise inaudible at loud playback levels.  If we want to achieve "CD" quality and use 20 dB of digital attenuation, then 116 dB SNR is required.  Under these conditions, noise may still become audible if playback peaks exceed 96 dB SPL.

D/A converters with >116 dB SNR are few and far between.  Noise alone is a dead giveaway when comparing D/A converters.
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #7
So your view is that it is a potential issue but it can be rectified with careful design. So a competently designed DAC should not pose issues significant enough to be audible? It would seem to be similar to jitter which is also often cited left, right and center. Jitter can be a potential issue if the quantities are large enough but in a good unit there is no reason to assume it should pose an issue. Am I on the right track?

DAC IV stages

Reply #8
Take a look at published specifications for D/A converters, DVD players, sound cards, iPod, etc.  Many only have an 80 to 90 dB SNR.


You're exaggerating.  Most quality players (iPods included) are very nearly quantization noise limited over line out.  80dB is quite rare and would be exceptionally bad for a name brand device.  The Apple stuff will give you 15.5 effecitve bits over headphone out with full analog volume control for not all that much money. 

D/A converters with >116 dB SNR are few and far between.  Noise alone is a dead giveaway when comparing D/A converters.


Yeah but in reality it doesn't make much difference since you never use that dynamic range.  CDs using > 80dB dynamic range are far rarer then D/A converters with < 80dB dynamic range.

DAC IV stages

Reply #9
For example, Benchmark's new DAC2 converter uses a ES9018 DAC IC and achieves a 129 dB SNR A-weighted, and distortion that is better than -109 dB.  This implies that distortion is below the threshold of hearing until peak music levels reach 109 dB SPL.


Does not mean something completely different?  Does that not mean that distortion, being 109 dB down, isn't crossing the threshold of hearing  when peak levels are 109dB up, but rather crossing the threshold of existence?  I can't hear 20dB sounds, much less 0.

Creature of habit.

DAC IV stages

Reply #10
80dB is quite rare and would be exceptionally bad for a name brand device.  The Apple stuff will give you 15.5 effecitve bits over headphone out with full analog volume control for not all that much money.


15.5 effective bits when the volume control is set at maximum, and to the best of my knowledge, none of the Apple devices have analog volume control (please correct me if I am wrong).  With 20 dB digital attenuation, 15.5 effective bits becomes 12.2 effective bits.  The problem is that significant amounts of digital attenuation are often used at normal listening levels, and the converter noise is often audible.

Yeah but in reality it doesn't make much difference since you never use that dynamic range.  CDs using > 80dB dynamic range are far rarer then D/A converters with < 80dB dynamic range.


CD and high-resolution low-noise recordings do exist, and if these are used in an ABX test, the noise floors of many converters should be audible (on the basis of calculated SPL of the noise).
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #11
For example, Benchmark's new DAC2 converter uses a ES9018 DAC IC and achieves a 129 dB SNR A-weighted, and distortion that is better than -109 dB.  This implies that distortion is below the threshold of hearing until peak music levels reach 109 dB SPL.


Does not mean something completely different?  Does that not mean that distortion, being 109 dB down, isn't crossing the threshold of hearing  when peak levels are 109dB up, but rather crossing the threshold of existence?  I can't hear 20dB sounds, much less 0.


My point was that nobody can rationally argue that the distortion is audible if it is below the threshold of hearing.

Distortion will become audible at some level significantly above the threshold of hearing.  Music masks much of the distortion, and this limits our ability to detect it.  Our ears are also very tolerant of low-order harmonic distortion as many musical sources have significant harmonic distortion.  For these reasons low-order distortion can often reach very high levels before it becomes noticeable.  However, non-harmonic distortion components are often much easier to hear, and may be detectable at relatively low levels if they are not well masked by the music. 

A published THD+N number does not tell us much about the audibility of the distortion (or lack thereof), unless the number is so low that audibility is very unlikely (as with the -109 dB performance of the DAC2).  The -109 dB THD+N number provides a good indication that the distortion should be inaudible.
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #12
Take a look at published specifications for D/A converters, DVD players, sound cards, iPod, etc.  Many only have an 80 to 90 dB SNR.  This means that the noise will be audible in an A/B/X test if the peak music levels exceed 80 to 90 dB SPL (easily achieved).


That is far from being a universal truth.  The actual answer depends on the test.

The answer given above probably presumes that we set up the system to play music at a desired high SPL, and then mute the music player. If we hear any noise even if that requires sticking our heads inside the loudspeaker, then the test is failed.  That will demonstrate the need for a DAC whose SNR is equal to the peak SPL observed during the listening phase. This should be less than 120 dB or we are talking pretty severe ear damage.

The first step towards what I think is a more reasonable experiment would be to not allow the listener to stick his head inside the speaker cone, but rather constrain himself to a normal listening position. If then we presume a typical quiet room in a residence, then we can probably still do the same basic kind of test as above and get away with a DAC whose SNR is the peak SPL minus maybe 20-30 dB.  If we cherry pick the room, maybe only 10-15 dB. 

If we back off from a very high peak SPL to a more typical and comfortable SPL that is not the one we use with visiting firemen, then we may pick up another 10 dB or more.

The above rationalizing moves as I would categorize them with some personal bias, probably gets us below 100 dB.

The opposite extreme can be obtained by doing a different experiment that is IMO no less "real world".  In this experiment, we set up the system to play music at a desired high SPL, and then wait for the music to stop playing, but continue to play the recording which would then contain the sound in the performance room that remains including the noises made by the people while they are trying to be quiet, the HVAC, etc.  This test gets us below 90 dB, and may get us down as low as 60 or 70 dB.    75 dB is IME a 90% solution. 85 dB is a 99% solution.

As a practical matter, the vast majority of people find players and systems with 90 dB dynamic range to be highly satisfying, to say the very least. 

Remember that allegedly picky listeners like vinyl, and a vinyl system with 75 dB "Needle Raised" dynamic range is about as good as that technology ever got.

DAC IV stages

Reply #13
D/A converters with >116 dB SNR are few and far between.  Noise alone is a dead giveaway when comparing D/A converters.


I guess that TI, Crystal Semiconductor, and AKM have been pulling our chains because they all have fairly inexpensive chips that do 115-120 dB dynamic range. ;-)

I've had a LynxTwo since what, 2002 that met that criteria and they are hardly rare. That isn't specsmanship because I've measured it myself. The same converters or competitive converters are now in the < $5 per channel price range.


DAC IV stages

Reply #14
For example, Benchmark's new DAC2 converter uses a ES9018 DAC IC and achieves a 129 dB SNR A-weighted, and distortion that is better than -109 dB.  This implies that distortion is below the threshold of hearing until peak music levels reach 109 dB SPL.


Does not mean something completely different?  Does that not mean that distortion, being 109 dB down, isn't crossing the threshold of hearing  when peak levels are 109dB up, but rather crossing the threshold of existence?  I can't hear 20dB sounds, much less 0.


My point was that nobody can rationally argue that the distortion is audible if it is below the threshold of hearing.


Equating the threshold of hearing with audible distortion requires a fair amount of naivete if we are leaving the synthetic land of listening to isolated pure tones, and condescending to listening to more complex sounds such as music, speech, and sound effects.  ;-)


DAC IV stages

Reply #15
Designing an I-V converter with a 130 dB SNR is not trivial.  Impedances need to be very low, and this puts significant demands on the opamp.  The opamp must have very low EIN and must be able to deliver high current drive while maintaining low distortion.  There are about 2 opamps that will work in this application.  Resistors must be metal film (to keep distortion low). 

One complication is that high-frequency switching noise is always present on the current outputs of sigma-delta converters.  This high-frequency noise must be bypassed with rather large NPO or COG capacitors before entering the I-V converter.

There are many opportunities to cut costs, and there are many opportunities to make mistakes when designing an I-V conversion stage.  The I-V conversion stage may be more important than the choice of conversion IC.


Does one have to design anything? Aren't the application notes reliable?  Can't I lift the design of the evaluation board?

DAC IV stages

Reply #16
D/A converters with >116 dB SNR are few and far between.  Noise alone is a dead giveaway when comparing D/A converters.


I guess that TI, Crystal Semiconductor, and AKM have been pulling our chains because they all have fairly inexpensive chips that do 115-120 dB dynamic range. ;-)

I've had a LynxTwo since what, 2002 that met that criteria and they are hardly rare. That isn't specsmanship because I've measured it myself. The same converters or competitive converters are now in the < $5 per channel price range.

Most commercial products do not achieve the published specifications provided by the IC manufacturers.  These specifications are an indication of the best performance that can be achieved by these parts on an ideal evaluation board, using lab power supplies.  Lots of things go wrong when manufacturers cram a bunch of parts in a small package while trying to meet cost and schedule targets.  The LynxTwo is a pro product and was not subject to the same design goals as consumer products.

The audio performance of the D/A converters in consumer devices such as TVs, DVD players, portable audio devices, and cable boxes do not come close to 115 dB SNR.  Of equal importance is the fact that nobody listens to these devices with the included digital volume control set to 100%.  Furthermore, the included digital volume controls are usually not dithered (they truncate).

A meaningful ABX test should attempt to replicate the most demanding typical use:

1) Digital volume control at typical setting
2) Loud but reasonable playback levels
3) Reasonably quiet room
4) Accurately matched levels
5) Low-noise source material

If A and B cannot be distinguised under the above conditions, then it is fair to say that neither offers a sonic advantage over the other in typical use.
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #17
Does one have to design anything? Aren't the application notes reliable?  Can't I lift the design of the evaluation board?


Great question!  The answer is no.  It is not quite that simple.

1) The evaluation boards do not have power supplies and magnetic components in close proximity to the converter and analog circuits (a power supply is a necessity in AC-powered commercial products).
2) Battery powered devices must use lower power opamps and higher impedances than those found on an evaluation board.
3) Evaluation boards use very generous amounts of PCB real-estate (often not available in a practical product).
4) Evaluation boards may use components that are outside of the budget of consumer products.
5) Evaluation board often fail to remove the high levels of common-mode distortion and noise that is created by the D/A IC.  These evaluation boards only measure well when feeding the precisely balanced inputs on audio analyzers.  Unbalanced measurements of evaluation boards are often very poor for this reason.  A differential amplifier should be added to remove the common mode components before driving an unbalanced output.  This omission is nearly universal with all of the evaluation boards I have seen.
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #18
D/A converters with >116 dB SNR are few and far between.  Noise alone is a dead giveaway when comparing D/A converters.


I guess that TI, Crystal Semiconductor, and AKM have been pulling our chains because they all have fairly inexpensive chips that do 115-120 dB dynamic range. ;-)

I've had a LynxTwo since what, 2002 that met that criteria and they are hardly rare. That isn't specsmanship because I've measured it myself. The same converters or competitive converters are now in the < $5 per channel price range.


Most commercial products do not achieve the published specifications provided by the IC manufacturers.


Given that I have measured about 100 different such devices, let's just say that my evidence says that they often come very close.

Quote
These specifications are an indication of the best performance that can be achieved by these parts on an ideal evaluation board, using lab power supplies


Lab power supplies being easily matched by IC regulators...

Quote
Lots of things go wrong when manufacturers cram a bunch of parts in a small package while trying to meet cost and schedule targets.


Not exactly. Lots of thing can go wrong is true, but that's not saying that they do always go wrong which is the clear meaning of the above sentence.

Quote
The LynxTwo is a pro product and was not subject to the same design goals as consumer products.


Again, I've tested 100s of consumer products and often the line between pro and consumer is blurred, to say the least. Time is part of the story. When the Lynxtwo first came to market the designer told me that the converter chips he used were the most expensive parts on the market and my research put them over $20 each.  Today is a decade later, the parts are still on the market as of sometime in the past year, but in small volumes they cost only a fraction as much.

Quote
The audio performance of the D/A converters in consumer devices such as TVs, DVD players, portable audio devices, and cable boxes do not come close to 115 dB SNR.


As I have pointed out with colaborating evidence, there is no practical need to achieve 115 dB as a rule. It may be fun to thrown out numbers without real-world substantiation, but that leaves one open to challenges from people who have at least one foot in the real world.

Quote
Of equal importance is the fact that nobody listens to these devices with the included digital volume control set to 100%.


Agreed.

Quote
Furthermore, the included digital volume controls are usually not dithered (they truncate).


Again, that appears to be an alleged fact with no real-world evidence to support it.


Quote
A meaningful ABX test should attempt to replicate the most demanding typical use:

1) Digital volume control at typical setting
2) Loud but reasonable playback levels
3) Reasonably quiet room
4) Accurately matched levels
5) Low-noise source material

If A and B cannot be distinguished under the above conditions, then it is fair to say that neither offers a sonic advantage over the other in typical use.


Agreed.

DAC IV stages

Reply #19
Take a look at published specifications for D/A converters, DVD players, sound cards, iPod, etc.  Many only have an 80 to 90 dB SNR.  This means that the noise will be audible in an A/B/X test if the peak music levels exceed 80 to 90 dB SPL (easily achieved).


That is far from being a universal truth.  The actual answer depends on the test.


I agree, but the test conditions may be closer to typical applications than you describe below:

The answer given above probably presumes that we set up the system to play music at a desired high SPL, and then mute the music player... 

If we back off from a very high peak SPL to a more typical and comfortable SPL that is not the one we use with visiting firemen ... probably gets us below 100 dB ...

As a practical matter, the vast majority of people find players and systems with 90 dB dynamic range to be highly satisfying, to say the very least.


Using your numbers:
100 dB SPL (slow)
90 dB SNR D/A converter

And adding the following assumptions:
10 to 20 dB music crest factor
30 to 40 dB SPL ambient (quiet room)
10 to 20 dB digital attenuation prior to D/A conversion


system output noise = (average SPL) + (crest factor) - ((D/A SNR) - (digital attenuation))

Worst case given the above assumptions:
(100 + 20) - (90 - 20) = 50 dB SPL

Worst case, the converter noise is 10 to 20 dB louder than the room noise.  This should easily be noticeable between tracks, and may be audible in quiet passages of music, even if the noise is white.

Best case given the above assumptions:
(100 + 10) - (90 - 10) = 30 dB SPL

Best case, the converter noise is equal to the room noise, or 10 dB quiter than the room noise.  It may be very hard to hear this noise between tracks IF the noise is white.

But, please notice the "IF"

The converter SNR may be limited by a single spurious tone and not by white noise.  If so, the noise will be much more noticeable.

Our ears have the ability to hear a 3 kHz tone 30 dB lower in amplitude than the surrounding noise (I encourage the skeptical reader to try this test for themselves).  The 30 dB at 3 kHz number can also be derived from masking theory.

If one of the converters under test has a noise floor that is limited by a single tone (AC hum, power supply switching frequencies, or crosstalk from other parts of the system), then this noise will be much more noticable (and much more objectionable than the above calculations would suggest).

Bottom line, it is quite reasonable to expect noticeable differences in a converter's noise floor in a moderately demanding application.  It is unreasonable to assert that all converters are good enough to be indistinguishable.
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #20
80dB is quite rare and would be exceptionally bad for a name brand device.  The Apple stuff will give you 15.5 effecitve bits over headphone out with full analog volume control for not all that much money.


15.5 effective bits when the volume control is set at maximum, and to the best of my knowledge, none of the Apple devices have analog volume control (please correct me if I am wrong).


All portable electronics have analog volume control, Apple's included.  You couldn't buy an mp3 player that had digital volume control if you wanted to.  They aren't made.

With 20 dB digital attenuation, 15.5 effective bits becomes 12.2 effective bits.  The problem is that significant amounts of digital attenuation are often used at normal listening levels, and the converter noise is often audible.


Yeah but no one does this so it doesn't matter.

Edit:  Although I don't mean to imply that analog volume control imposes no SNR hit.  I'm sure at the lowest volume the SNR on an ipod is much worse.  But at reasonable volumes its more then sufficient, which is what I was trying to say.

Yeah but in reality it doesn't make much difference since you never use that dynamic range.  CDs using > 80dB dynamic range are far rarer then D/A converters with < 80dB dynamic range.


CD and high-resolution low-noise recordings do exist, and if these are used in an ABX test, the noise floors of many converters should be audible (on the basis of calculated SPL of the noise).


Yes, but people don't buy CE to do ABX tests of specially selected samples.  They buy them to listen to music.  With enough effort I can find samples that break MP3, AAC, Vorbis, etc.  Doesn't mean I dislike those formats.  It just means I can break them if I try.

DAC IV stages

Reply #21
system output noise = (average SPL) + (crest factor) - ((D/A SNR) - (digital attenuation))


I have several doubts on what you say an definitely, that formula doesn't help.

The part that confuses me especially is why do you imply that lowering the volume digitally, increases the noise level.

What i mean is, decreasing the SNR does not increase the noise SPL.


Quiet room: 30dB SPL.
Let's take a reasonable device (you keep talking about consumer products after all), so output: 90dB SPL
Let's also say that the device has a SNR of -91dB ( 1, just to make it different than 90).

Now, decrease the volume by 20, so max is 70db SPL.
The SNR of the device also now increases from -91dB to -71dB
The difference between max and noise floor (the SNR on this playback situation) is 40db SPL.

Is the room noise floor hearable now, compared to the signal? Maybe.
Is the noise produced by the device audible? I still wonder how.


Let's take a setup with more SNR.
Let's take some in-ear phones. : 100db SPL
Now, since they are in-ear, the room noise floor is reduced, let's say 20db SPL.
Taking the same device that has SNR of -91dB

Let's attenuate by 20dB, so peak signal 80db SPL.
The SNR of the device increases from -91dB to -71dB
Difference between max and noise floor (SNR of playback situation) is now 60dB.

So... what is hearable, after all? The SNR of the device is still below the noise floor.


DAC IV stages

Reply #22
The first step towards what I think is a more reasonable experiment would be to not allow the listener to stick his head inside the speaker cone

Oh Arny, you are too practical!

I'd apply the same constraint for playback volume. Yes, you can hear quantization distortion on reverb tails at 16 bits if you crank the volume by 40 dB during the fadeout. So don't do that!

--Ethan
I believe in Truth, Justice, and the Scientific Method

DAC IV stages

Reply #23

system output noise = (average SPL) + (crest factor) - ((D/A SNR) - (digital attenuation))


I have several doubts on what you say an definitely, that formula doesn't help.

The part that confuses me especially is why do you imply that lowering the volume digitally, increases the noise level.


Decreasing the volume with the digital volume control is necessary to reduce the volume to a normal listening level.  This decreases the signal, but does not decrease the noise level.  Most systems are designed to provide "normal" listening levels at something less than full volume (often 20 dB less than full volume).  If you use 20 dB of digital gain reduction, the SNR degrades by exactly 20 dB. 

If you do the math, the noise floor of consumer D/A converters is often higher than the ambient noise level in a quiet room when audio is playing at "normal" levels.
John Siau
Vice President
Benchmark Media Systems, Inc.

DAC IV stages

Reply #24
All portable electronics have analog volume control, Apple's included.  You couldn't buy an mp3 player that had digital volume control if you wanted to.  They aren't made.


Absolutely incorrect! 

Almost all of these products all have buttons or cap-touch buttons (or sliders) that control volume up and down.  In almost all cases, the volume control is a digital multiplier.  Very few consumer products still have analog volume controls (variable resistors) in the audio path.  DSP is cheap and is already required to do MP3 decoding.
John Siau
Vice President
Benchmark Media Systems, Inc.