DAC design: the I/V stage

ack

VIP/Donor & WBF Founding Member
May 6, 2010
6,774
1,198
580
Boston, MA
WARNING: TECHNICAL CONTENT

I have to admit that, despite my love for analog, I am fascinated by digital, its challenges and immense complexities. Leaving a DAC component's analog sections aside, the rest of it is a complex maze of digital circuitry, with noise and jitter lurking all over, where slew rates are also a critical aspect (as in analog circuits), which brings me to the current/voltage (I/V) conversion stage of a DAC.

The following paper http://electronicdesign.com/analog/designing-high-speed-current-voltage-conversion is typical and written by an Analog Devices engineer in their High Speed Amplifier Group, and in it basically covers noise in the I/V stage and also states that slew rates following the DAC chip's output must be fast enough: "the op amp needs to slew at least 300V/µs. If it can’t, slew distortion will slow waveform edges and generate code-dependent jitter in the output" - I redacted the "300" because it may not be literally applicable to audio and the main point is in bold typeface, and as he says elsewhere, "the op amp used for the transimpedance amplifier needs to slew fast enough to match the DAC’s output"

Lately, I've been reading up on the Esoteric K-01x and the upcoming K1 players, and the K-01X is using a MUSES 02 bipolar op amp for the I/V conversion - and not the allegedly superior MUSES 01 which is a JFET op amp - while the Grandioso K1 will be using the new MUSES 03 JFET op amp. The MUSES 01 has a stated slew rate of 12v/usec, the 02 5V/usec, and the specs for the 03 are not available yet. Nonetheless, 5 or 12V/usec are not really fascinating numbers. Having said that, the players are claimed by Esoteric to offer intrinsic slew rates of "2000V/usec" in the analog output buffer, which is extra-ordinarily fast. Meantime, a few other manufacturers use discrete I/V conversion (of unknown slew rates), and yet others a simple resistor.

So the questions are:

1) which I/V design is superior and why
2) why wouldn't everyone use a simple resistor
3) given that the MUSES 01/02 are just similar in price, is there a benefit of one over the other; some DIYers claim the 01 is really the superior choice for this application
4) what are good slew rates for this part of the circuitry as well as the DACs themselves
5) On diyaudio.com, Wayne of Pass is quoted as saying that op amps like the MUSES "don't fit" in their discrete-circuit design philosophy (analog or digital, though Pass don't do digital), so why wouldn't everyone design a discrete I/V?

I am looking for real, honest answers, not google parrotisms - Artificial Intelligence is always apparent.

Thanks

-ack
 

Ken Newton

Well-Known Member
Dec 11, 2012
243
2
95
WARNING: TECHNICAL CONTENT

So the questions are:

1) which I/V design is superior and why

That depends on what parameters you are trying maximize, which are partly dependent on which DAC you are utilizing. Like with many things no one best answer.

2) why wouldn't everyone use a simple resistor

Because a simple resistor is not a virtual ground. Current output DACs typically produce the lowest harmonic distortion operating in to a virtual ground, which is among the main reasons why op-amps configured as feedback based transimpedance amplifiers are so commonly employed. That said, a simple resistor offers many other performance advantages in I/V application.

3) given that the MUSES 01/02 are just similar in price, is there a benefit of one over the other; some DIYers claim the 01 is really the superior choice for this application

I have no experience with the MUSES devices.

4) what are good slew rates for this part of the circuitry as well as the DACs themselves

The problem isn't so much in tracking the slew rate of the DAC output as much as it is that active I/V circuits typically feature a fixed slew rate, causing code dependent errors as the DAC output changes from sample to sample. An closely related related problem is that the DAC's unfiltered output is simply wideband, producing undesired spectral replications of the desired signal far in to the RF range. Those replications can provoke errors in an active stage, especially one employing negative feedback.

5) On diyaudio.com, Wayne of Pass is quoted as saying that op amps like the MUSES "don't fit" in their discrete-circuit design philosophy (analog or digital, though Pass don't do digital), so why wouldn't everyone design a discrete I/V?

Typically, it's due to the design time and manufacturing cost associated with discrete circuitry. Integrated circuits usually require less design time and expertise, and also lower manufacturing cost. That isn't just true for I/V circuits but also for most other types of audio circuits.

-ack

Reponse above.
 

opus112

Well-Known Member
Feb 24, 2016
462
4
148
Zhejiang
1) which I/V design is superior and why

As Ken already mentioned, the answer is 'it depends'. Firstly I'll assume superiority is determined subjectively, by listening. In which case my findings are that a suitably designed discrete stage sounds best, ahead of opamps and ahead of passive I/V. But my experience here has so far only been with one particular multibit DAC chip, with other manufacturer's devices its possible the preference order may change.

2) why wouldn't everyone use a simple resistor

For quite a few years I used a simple resistor and got great results. But then in my experiments I noticed that making the power supply to the DAC chip better (i.e. lower impedance) made the sound even better. I scratched my head for a very long time as to why this might be and eventually came to the realization that the ability of the particular DAC chip I was using to reject power supply noise was dependent on the impedance presented by the following I/V stage to its output. In EE parlance, the PSRR of the DAC is improved with reduced output compliance.

Normally the argument from objectivist designers is that output compliance should be reduced to prevent distortion caused by code-dependent output impedance changes. These certainly are an issue for the measurements but seems less so for SQ. An opamp by virtue of its huge open loop gain presents a very low static impedance to the DAC, dynamically though its performance is far less ideal.

3) given that the MUSES 01/02 are just similar in price, is there a benefit of one over the other; some DIYers claim the 01 is really the superior choice for this application

I have no experience of using these chips, they do look to be strongly marketed to audiophile designers which is something of a turn-off to me personally.:p I merely note that using a classAB output stage (present in all opamps to my knowledge) in an I/V is a poor design choice. A discrete designer has the option to bias everything in classA which is what I did when I designed my own I/V stage, the result was it sounded better.

4) what are good slew rates for this part of the circuitry as well as the DACs themselves

A designer wouldn't be concerned with just a single number like slew rate as the issue is more to do with how much time the I/V spends at the wrong value. Slew rate is part of that but more relevant is settling time and the nature of the settling transient.

5) On diyaudio.com, Wayne of Pass is quoted as saying that op amps like the MUSES "don't fit" in their discrete-circuit design philosophy (analog or digital, though Pass don't do digital), so why wouldn't everyone design a discrete I/V?

Not all designers are so comfortable with discrete design as the guys at Pass. Its a matter of what the engineer's experience is - many see no problem at all with opamp I/V as generally the numbers show up well.
 

ack

VIP/Donor & WBF Founding Member
May 6, 2010
6,774
1,198
580
Boston, MA
Thank you both for the very helpful responses! Regarding the I/V stage's slew rate and it being adequate in relation to the rest of the design, I was just reading Keith Johnson's write-up in the 4000SV player's bulletin:

A newly designed DAC summation I/V amplifier is optimised to support the higher speed and gain of the the SV output section
 

microstrip

VIP/Donor
May 30, 2010
20,806
4,698
2,790
Portugal
(...) For quite a few years I used a simple resistor and got great results. But then in my experiments I noticed that making the power supply to the DAC chip better (i.e. lower impedance) made the sound even better. I scratched my head for a very long time as to why this might be and eventually came to the realization that the ability of the particular DAC chip I was using to reject power supply noise was dependent on the impedance presented by the following I/V stage to its output. In EE parlance, the PSRR of the DAC is improved with reduced output compliance.

Normally the argument from objectivist designers is that output compliance should be reduced to prevent distortion caused by code-dependent output impedance changes. These certainly are an issue for the measurements but seems less so for SQ. An opamp by virtue of its huge open loop gain presents a very low static impedance to the DAC, dynamically though its performance is far less ideal.

Long ago I also got excellent results using a 50 ohm resistor and a passive CLC filter in a Sony X7 ESD that used BB PCM63k chips.

Theoretically you should have a low value resistor to overcome as much as possible the effects you refer, at a cost of reducing the output signal voltage. However, if we consider that the analog people get signals around 500 microvolt from moving coil cartridges for their SOTA systems, amplitude should not be is not a subjective issue! The experiment I never carried and thought a lot about - using a high quality 0,1 ohm resistor and getting the signal from its extremes in differential mode with wires to a quality MC transformer connected to a phono stage without RIAA compensation.
 

microstrip

VIP/Donor
May 30, 2010
20,806
4,698
2,790
Portugal
Thank you both for the very helpful responses! Regarding the I/V stage's slew rate and it being adequate in relation to the rest of the design, I was just reading Keith Johnson's write-up in the 4000SV player's bulletin:


Ack,

Late arrival to this thread, but IMHO you can not separate the opamp slew rate needs from the DAC output characteristics, such as setting time - e.g the PCM 63 had a settling time of 200 ns for full scale (2mA). In some cases, as DACs are not ideal, perhaps a finite controlled slew rate can be a good subjective thing, as it introduces some kind of noise filtration.
 

ack

VIP/Donor & WBF Founding Member
May 6, 2010
6,774
1,198
580
Boston, MA
Let's agree that we are not really only talking just about the slew rate - that's just what's published for the op amps I referenced and what the AD article discusses - and that we are talking about speed, aka slew rates and settling time together. With that aside, the Analog Devices article and Johnson basically say the same thing, in that the speed of the I/V stage must be adequately high. Looking at Johnson's language, which basically says the I/V speed must be able to support to the output's, I then go back to what Esoteric are claiming and doing: they claim an exceptionally high output buffer slew rate of 2000V/usec, but the I/V stage preceding it is nowhere near as fast. What gives?
 

Ken Newton

Well-Known Member
Dec 11, 2012
243
2
95
Let's agree that we are not really only talking just about the slew rate...we are talking about speed, aka slew rates and settling time together...I then go back to what Esoteric are claiming and doing: they claim an exceptionally high output buffer slew rate of 2000V/usec, but the I/V stage preceding it is nowhere near as fast. What gives?

DAC settling time is composed of three primary factors, slew rate, glitching and any DAC internal amplifier transient damping delay. Glitching is where the DAC quantizer switching causes the analog output to briefly take an incorrect value before arriving at the correct value. This is distinct from the slew rate, which is a function of the DAC's realative ability to drive current in to it's parasitic capacitance. Taken together, those three essentially define the DAC's settling time, which is to say, the period of time where the DAC's output signal has not stabilized in value.

The challenge usually presented to the I/V amplifier then is to accurately track the sharply stepped output of the DAC without further contributing error to the analog sample value. It's because the DAC's analog output rapidly steps (switches) from value to value that causes problems for the I/V amplifier. Keep in mind that it's the discretely stepping (slewing) of the output levels which produces the undesired ultrasonic images, so, unless we have an application where retaining those images is important - which we don't for audio - then we don't actually need for the I/V to accurately track the DAC's sharp output slewing. For example, there are a number of very highly regarded DACs (Audio Note, for example) that passively filter the DAC chips stepped output signal before applying the now band limited signal - with it's greatly reduced slew rate - to the I/V circuit.

The I/V circuit need only to not introduce errors in the audio band. One way to do that is by accurately tracking the fast slewing DAC output directly in a brute force manner, which is technically very challenging. If the post DAC chip circuitry is designed to not introduce slewing errors of it's own, then the fast slewing DAC output does not actually need to be directly tracked since the fast slewing discretely stepped output only carries the undesired ultrasonic images we will later reject anyhow.
 

ack

VIP/Donor & WBF Founding Member
May 6, 2010
6,774
1,198
580
Boston, MA
DAC settling time is composed of three primary factors, slew rate, glitching and any DAC internal amplifier transient damping delay. Glitching is where the DAC quantizer switching causes the analog output to briefly take an incorrect value before arriving at the correct value. This is distinct from the slew rate, which is a function of the DAC's realative ability to drive current in to it's parasitic capacitance. Taken together, those three essentially define the DAC's settling time, which is to say, the period of time where the DAC's output signal has not stabilized in value.

The challenge usually presented to the I/V amplifier then is to accurately track the sharply stepped output of the DAC without further contributing error to the analog sample value. It's because the DAC's analog output rapidly steps (switches) from value to value that causes problems for the I/V amplifier. Keep in mind that it's the discretely stepping (slewing) of the output levels which produces the undesired ultrasonic images, so, unless we have an application where retaining those images is important - which we don't for audio - then we don't actually need for the I/V to accurately track the DAC's sharp output slewing. For example, there are a number of very highly regarded DACs (Audio Note, for example) that passively filter the DAC chips stepped output signal before applying the now band limited signal - with it's greatly reduced slew rate - to the I/V circuit.

The I/V circuit need only to not introduce errors in the audio band. One way to do that is by accurately tracking the fast slewing DAC output directly in a brute force manner, which is technically very challenging. If the post DAC chip circuitry is designed to not introduce slewing errors of it's own, then the fast slewing DAC output does not actually need to be directly tracked since the fast slewing discretely stepped output only carries the undesired ultrasonic images we will later reject anyhow.

Ken, this makes a lot of sense, and I will emphasize the highlighted sentence - that's really the reason why I redacted that "300V/usec" figure from the AD article in my original post, because the article is not really covering audio frequencies, but much higher bandwidth networks, therefore, the slew rate they calculate may not really be applicable to audio, and as you also said "the fast slewing DAC output does not actually need to be directly tracked".

Which brings me to the next question: why would Esoteric (or anyone) tout a fast-slewing output buffer, if the gating factor is really the I/V stage preceding it, and whose speed was apparently chosen to track the DAC chips' speed only as "needed" (and not "as fast as possible"). By contrast, Johnson and Spectral appear to be approaching the speed of this part of the design (I/V and output) as "one", and it seems more accurate to me.
 

DonH50

Member Sponsor & WBF Technical Expert
Jun 22, 2010
3,946
305
1,670
Monument, CO
A resistor, whilst being very wideband, may reduce the DAC's intrinsic linearity because it is not a virtual ground as said earlier. Higher voltage swing at the output of some DACs will cause higher nonlinearity (distortion).

A very wideband buffer may be able to handle the very high frequency content of glitches and other switching transients without slew limiting and having its own problems settling. Hitting a low-speed device with a high-speed signal can cause settling and stability issues. If this buffer is the I'V converter and provides a good virtual ground and is followed by a filter to reduce the high-frequency content then the output driver/buffer's requirements are much less stringent.

Also as stated before, there are so many variables in DAC and buffer design, that "which is best" is meaningless without fairly in-depth technical knowledge of the circuits IMO. A 50-ohm resistor worked well for a 10-bit, 10 GS/s DAC but an active I/V converter worked much better for a 16-bit, 5 MS/s DAC. At least IME and perhaps that of nobody else. - Don
 

Ken Newton

Well-Known Member
Dec 11, 2012
243
2
95
...Which brings me to the next question: why would Esoteric (or anyone) tout a fast-slewing output buffer, if the gating factor is really the I/V stage preceding it...

I can only guess at Esoteric's specific technical reasons, if there are any. I rather suspect that it has more to do with being a figure-of-merit for marketing purposes. Sort of like how intel was always touting the clock speed of the latest pentium microprocessor. Figures-of-merit such these become a proxy for product performance in the minds of consumers, whether or not they actually produce a useful benefit.
 

ack

VIP/Donor & WBF Founding Member
May 6, 2010
6,774
1,198
580
Boston, MA
Precisely!
 

YashN

New Member
Jun 28, 2015
951
5
0
Canada
Additional interesting questions:

1. What's the effect on SQ on chaining Opamps in the output section (same slew rate or different slew rates)

2. What's the 'slew rate' of a tube output section?
 

Ric Schultz

Well-Known Member
Jun 21, 2013
227
56
333
Soquel, CA
Some facts, etc.: The Esoterics do not use an I-V converter. The AKM DACs output voltage (one volt rms per phase). The Muse op amp is used as a filter. The buffer after that is a high speed zero feedback super sounding buffer device (LME49600) .....it is super fast and it is very transparent (not quite as transparent as a single fet....but close). For more transparency I would eliminate the Muse op amp and just use a single jfet biased into class A with another fet....then a super transparent coupling cap per phase (no summing). A single pole filter is all you need between the DAC and fet. You would, however, only get one volt rms per phase. But most use a preamp....so no biggie there.
 

opus112

Well-Known Member
Feb 24, 2016
462
4
148
Zhejiang
I can only guess at Esoteric's specific technical reasons, if there are any. I rather suspect that it has more to do with being a figure-of-merit for marketing purposes. Sort of like how intel was always touting the clock speed of the latest pentium microprocessor. Figures-of-merit such these become a proxy for product performance in the minds of consumers, whether or not they actually produce a useful benefit.

I'm in agreement :) Marketing audio nowadays is supported by lists of bullet points, the more boxes ticked apparently the better the product. Intel's pursuit of the metric of clock frequency produced the P4 - the less said about that particular CPU, the better.
 

opus112

Well-Known Member
Feb 24, 2016
462
4
148
Zhejiang
Long ago I also got excellent results using a 50 ohm resistor and a passive CLC filter in a Sony X7 ESD that used BB PCM63k chips.

The PCM63 is amongst the best performers subjectively I understand. I haven't listened to one critically myself but I do have four samples from way back sitting in an antistatic box somewhere in my possession. I reckon that with the PCM63's low 670ohm output impedance it'll be strongly subject to power supply quality - an I/V resistor as low as 50ohms will make it even more sensitive. Best would be a discrete I/V stage I'd have thought.

Theoretically you should have a low value resistor to overcome as much as possible the effects you refer, at a cost of reducing the output signal voltage.

That's the conventional theory - to minimize the output compliance - but conventional theory ignores PSRR which I've found to matter more in terms of subjective sound quality.

Take as an example another chip fairly popular with DIY DAC builders, the TDA1543. It turns out this sounds best with its internal current source switched off (pin7 open circuit) and this improvement I suggest is down to improving its PSRR. Plenty of DIYers in search of the best sound run this chip at its maximum supply voltage so they can maximize the output compliance with the largest I/V resistor it can accommodate.
 

ack

VIP/Donor & WBF Founding Member
May 6, 2010
6,774
1,198
580
Boston, MA

ack

VIP/Donor & WBF Founding Member
May 6, 2010
6,774
1,198
580
Boston, MA
OK. So what type of filtering is needed then, with those DACs? Is it as Ken Newton said earlier?
 

Ric Schultz

Well-Known Member
Jun 21, 2013
227
56
333
Soquel, CA
Most DACs these days put out very little out of band noise. I use a single pole filter in all my designs, rolling off above 30K.....if in series then a resistor and then another high value resistor to ground along with a polyprop small value cap. If you use a transformer that has a natural rolloff then you need no other filtering what so ever.
 

About us

  • What’s Best Forum is THE forum for high end audio, product reviews, advice and sharing experiences on the best of everything else. This is THE place where audiophiles and audio companies discuss vintage, contemporary and new audio products, music servers, music streamers, computer audio, digital-to-analog converters, turntables, phono stages, cartridges, reel-to-reel tape machines, speakers, headphones and tube and solid-state amplification. Founded in 2010 What’s Best Forum invites intelligent and courteous people of all interests and backgrounds to describe and discuss the best of everything. From beginners to life-long hobbyists to industry professionals, we enjoy learning about new things and meeting new people, and participating in spirited debates.

Quick Navigation

User Menu

Steve Williams
Site Founder | Site Owner | Administrator
Ron Resnick
Site Co-Owner | Administrator
Julian (The Fixer)
Website Build | Marketing Managersing