Creating a better DAC

Discussion in 'Interviews and Factory Tours' started by BarryT, Mar 19, 2021.

  1. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    My apologies for the delay, medical issues have reduced my output for a while, I am ramping back up.

    Last year I had an epiphany about the nature of DAC chips and DACs in general. I tested my idea with the proto-platform I showed you earlier in this thread. Since then I have had to confront all my doubts about my perception, but I was also confronted with the reality of my listener friends. comments. The danger with this field of exploration is the Dunning-Kruger issue, self deception, especially when I can not yet offer a logical mathematical explanation on paper. In writing this I had to review all of that again, though this time with greater security as I have become able, in a limited way, to symbolically write expressions of the interactions I perceive. Am I deluding myself to feel that I am getting a handle on it? Who knows? So let me take you down a path anyway.

    I would like to offer my motto, part of my logo, which is "Feedback sucks the life out of your music."

    As I looked into switches in DACs I saw that they were all pretty standard, that is the chain of resistance, or current sources, were all 'vanilla', several variations but no secret sauce. The CMOS switch technologies used in the ICs are well established, proven, and highly predictable. The conclusion was that the different commercial DAC chip's 'quantizers' (that which converts the information as digital bits in to an analog current by gating electrical current flows) all should have the same general sonic qualities. One small possible exception to this would be discrete R2R world, so I went ahead and laid one out. In it I was able to use bigger discrete switches for each bit than could be used in the IC version of the same circuit. Lower 'on' resistance and noise, all by the simple fact that there is more chip surface to build on and it was not a mixed signal silicon substrate. [Oohh, what a seductive Mistress Discreet R2R could be, but the cost is out of hand unless you make manufacturing compromises that make it no better than ICs]

    So the digital data side of the equation looks pretty damned good to me.

    In thinking this out the logical next step in the chain is making a current that represents the bit-code into Audio. The answer of course is to sum the output of all the switches into one current, and then convert that to voltage, which is what is done. To get the most predictable and accurate current the summation must be made at a Zero-Z node. A description of a zero ohm impedance point is no voltage, no capacity, no inductance, it can have current flow. That's also one of the ways to describe the Ground Point in the circuit. Run a Current DAC into such a point and you get a 'by the book' DAC that will be as close to 'perfect' as possible where the temperature lets water be liquid. We must pay homage to Maxwell's demons

    There is a myth in the Audio World that Zero-Z is good. You may be familiar with the output of a power amplifier, that's supposed to be a Zero-Z point. The other is the output of a current DCA chip. Into passive loads or or sources, both get the best numbers due to Zero-Z. Most DAC chip manufacturers offer an internal OP-AMP based Current-to-Voltage (T-to-E) converter (an Op-Amp) to reduce parts count. The better chips let you do the conversion yourself, costs more, but is all the difference in the world. If you have faith in a particular DAC chip then go to the manufacture's site and look at the data sheet, look at the recommended circuits on the output of their DAC chip. You will see an op-amp hard at work.

    And the way you get that is through Feedback. And to quote me, "Feedback sucks the life out of music".

    DAC.jpeg

    This diagram illustrates a resistive process (switched resistors inside the DAC) on the left , and the inverting (-) input of an amplifier marked I(T) got total current.

    Currents from the DAC resistors are summed and then canceled out by an opposite current that comes through a resistor (Rf) from the output of the inverting amplifier.

    This inverting signal pin input becomes a virtual ground or a Zero-Z Node.

    The output of the amplifier is a linear voltage representation of the current flow from the DAC, it is an I-to-E converter

    This is PERFECT for the DAC - but for music this is where stuff meets the moving blades for the first time. Yes, there are further issues down the line but this is the big one.

    I need a breather, have to do some work as do you.

    Next writing I want you to see feedback as I do, it will help explain a lot of stuff that audio mythology has turned into fog-obscured shapes and lights.

    As a Texan I pay homage to Oat Willie too

    Oat Willie.jpeg
     
    • Like Like x 7
    • Epic Epic x 4
    • List
    Last edited: Apr 7, 2021
  2. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    Feedback sucks the life out of your music.

    You use feedback all time, a great is example is you picking up a coffee cup this morning. You look at the cup, move your hand to it, correct its location by what you saw, grabbed the cup until you finger pressure told you that you had it. You corrected your hand’s location first with visual feedback, and then using tactile feedback from you fingers squeezed the cup enough to hold and not break the cup. You use the past to control the present.

    If you reach for that cup without watching your hand you use feed-forward, you are guessing based on your future expectation of where the cup will be.

    Both come in either negative or positive modes. In the negative mode you reduce your movements as you get closer (subtract the errors you see) and with positive you increase the movement (add the errors you see). We generally us negative a lot and positive very little, but they both are useful. We now associate all this with cybernetics Our interest in for audio falls under the cataorey of.

    Since most things in life involve interactions of some sort, we use negative feedback a lot to get what we want to happen, positive usually leads to trouble. The social as well as physical world.

    In non-linear dynamics theory we use feedback to control gain, that is amplification done by amplifiers (duh). With negative feedback we get:

    Improved stability of gain
    Reduction in distortion
    Reduction in noise
    Increase in input impedance
    Decrease in output impedance

    The active electronic components (tubes and semiconductors) we have to work with are not linear. The input and output relationships, the gain of such devices are not simple but follow complex and somewhat conditional non-linear results.

    We use negative feedback to tame this problem. We stack the components to get more gain than we need and then take some of the output and subtract it from the input to get where we want. We make it a loop, no feedback is open looped, add feedback to close the loop. Here is a quick visualization, a graphic explanation of it all.

    feedback amps.jpg

    The essence of this is that if you subtract some of the output from the input you will reducing everything including distortion. Seems pretty straight forward , works well and predictably, but it comes with a hefty dark-side.

    First, the feedback mechanism produces a bunch of audio garbage, that is low level upper frequency crap, noise like, based on the signal being processed and generally less that the distortion so you don't notice it on a distortion analyzer, it can be noted by using a spectrum analyzer on the output of the distortion analyzer. This was first noted by Peter Baxendall (of the Baxendall Tone Control circuit fame back in the 1950), and promptly ignored.

    Second, well that is the point of all this DAC stuff and the realization that let me build a better DAC system, the center of my next entry.

    S
     
    • Like Like x 12
    • Epic Epic x 3
    • List
  3. haywood

    haywood Friend

    Pyrate
    Joined:
    Oct 22, 2015
    Likes Received:
    763
    Trophy Points:
    93
    Before you get into an explanation of how the secret sauce works... if you have something you think is unique / original then you should look into filing a patent to protect your interests. The bad news is it’d involve a lawyer.
     
  4. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    Thank you very much, you are of course, right. While I do consider the final implementation a Trade Secret, I am as of yet not dead-nuts sure that I could convince a lay jury of the originality or novelty. I have been in patent litigation in the past. I am filing a Provisional now before I open the kimono all the way. I have been delaying it because writing this monologue has forced me to confront my demons of doubt and laziness. Working under temporal pressure is entirely compelling, in fact i relish it - it forces focus like nothing else. Just for a second the whole damn thing has to make sense, the ultimate grok is also the ultimate thrill. And what is life but thrills and boredom?

    When I was young I could threaten me with speed, acceleration, flight, motion, a physical adventure and the like for a thrill. Days have changed. Surrendering physical agility and endurance makes the thrill in putting the pieces together with accuracy all the more sweet. Something about being alive.

    This would make patent #35 for me. First one was a trick to get Class-A with high efficiency.

    I appreciate and celebrate your concern and honesty, it makes the world a better place.
     
    • Like Like x 9
    • Epic Epic x 2
    • List
    Last edited: Apr 14, 2021
  5. soekris

    soekris MOT - Soekris Engineering

    Pyrate
    Joined:
    Mar 25, 2016
    Likes Received:
    954
    Trophy Points:
    93
    Location:
    Denmark
    A R-2R discrete DAC don't need to be I out, you can easily make them V out and avoid the I to V converter....
    And btw, the digital audio signal is usually transferred using a I2S bus, not I2C....
     
  6. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    Ah, yes on both, among other things switch noise in the voltage domain it harder to deal with in the voltage version of R2R. Current mode offers greater accuracy for the obvious reasons.

    My use of I2C is an historic error, they are virtually the same protocols by Phillips and I started working with the I2C when chip designing in the very early 80's in PCs. It is just habitual slang for me defining 3-wire connections, it was about serial communications between hardware elements. If it makes you feel better I will try to be more PC in the future (see, even PC has two meanings now). If you understand (which you do) that is means clock driven communications synchronicity with an androgynous master/slave relationship then it works either way. Thanks.
     
  7. sphinxvc

    sphinxvc Gear Master (retired)

    Staff Member Pyrate BWC
    Joined:
    Sep 28, 2015
    Likes Received:
    3,320
    Trophy Points:
    113
    This is a great thread, thank you for sharing this process with us.
     
  8. soekris

    soekris MOT - Soekris Engineering

    Pyrate
    Joined:
    Mar 25, 2016
    Likes Received:
    954
    Trophy Points:
    93
    Location:
    Denmark
    Actually, in my experience, voltage out don't have the same noise problems as current switching, in my DACs, I just have a small capacitor on the output of the voltage out R-2R networks....

    Sorry, but I2S and I2C, although both created by Philips, are two completely different serial busses. I2S are 3 wires, for point to point high speed Audio, I2C are 2 wires, and a lowspeed multimaster peripheral bus.
     
  9. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    Again I thank you for the pressure to get my thoughts together. To carry on with the 'Best DAC" we have begun to focus on feedback, which I claim sucks the life out of recorded music. To lay additional ground work I would like to review some realities of the analog/digital worlds. This table compares bits to volts addressing what bits do what work. Let's put this all into perspective.

    Four columns: The Number of Bits in data steam, The Number of increments or steps offered by the bits, Voltage of each step represents, How Many dB down that number is, and what percentage it is.


    Distort.jpg

    Now look at the numbers in the boxes -

    The details of what I am going to say is arguable in the exact numbers and detail but sound in terms what it describes.

    Bit # 16 is 90dB down and represents the errors associated with 0.003%, think of it as the lower limit of the distortion and the noise floor. This is probably the signal to noise ratio of your power amplifier and its ideal THD.

    Bit #20 exceeds the noise, dynamic range, and distortion measurements on any combination of audio components you can come up with that reproduces music.

    Bit #24 just happens to be the Johnson-Nyquist noise floor of the 1 Hz to 20kH band at 72 degree F. It does not get any quieter folks! Not in this universe and time.
    In the real world below this signal level there is no information, it is all noise from here down.

    Bit #30 describes the world at absolute Zero, the noise of a cold dead universe.

    So when someone brags to me about a full 32 bit decode I think of this-

    Reality Check2.jpg
    Think this over, it's a Data Point. The process is Data becomes Information creating Knowledge which results in Wisdom. For all practical purposes, correctly done 16-bit digital is pretty damn good. Clearly it's not the bits after a certain point. If you can relax your prejudices from the myths you believe the Fog will begin to lift as we shall see.
     
    • Like Like x 11
    • Epic Epic x 2
    • List
  10. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    I have spent time communicating with Soekris who has responded to this thread. He has a company that builds discrete R2R modules and user end-product DACs. From what I hear they are very good althought I am concerned about his OEM prices, that remains to be seem. One of the gents who critiques me is buying anew unit of his and the unit is in the transit. If it works out well I will be especially excited. Soekis's stuff is very impressive and as I mentioned in an earlier post discrete R2R with big switches just feels right to me. I am less comfortable with all the high frequency digital data on the same substrate as the audio processes as I find in all chip DACs.

    If it works out then I will marry the R2R output to my current-mode amplifier stage and see if I am right.

    Oh, sorry, I introduced the current-mode-node out of order. So let's standard voltage mode amplifiers. Back to business.

    My beef with feedback used in abundance to get good numbers or to create a current-to-voltage device is that even the best implementation produces audio discord. Especially that which occurs within the frequency band of our greatest aural sensitivity, that of 300 to 3000 Hz, and generally the next three octaves of 1 kHz to 10 kHz. Audio Discord - not very scientific, right?

    Here is a chart of the open and closed loop gain of a good Operational Amplifier As the math and physics for all voltage amplifiers is roughly similar this is a good place to work from. . Study this graphic for a minute. The black is the open loop gain and the red is the gain closed to 20 dB by feedback. You will see that would get you to about 100 KHerz and then the gain would no longer be flat. This Op-AMP has a 1 megaHertz Gain Product Bandwidth (GPB) which means it runs out of poop at 1 mHZ and the slope is approximately 20dB/Decade. The amount of actual feedback is 100 dB (120-20) at 1 Hz and about 17 dB (37-20) at 10 kHz.
    open and closed gain.png

    Two things strike the casual observer. The first is subtle, you might notice that the gain graph is very similar to the distortion curve (not shown) but is inverted. This is because the amount of feedback (the difference at any frequency between open and closed loop gain) is going down as the frequency goes up and the distortion is controlled by the amount of feedback.

    The second issue is more pernicious, for each frequency the amplifier is a different process, the amplifier operates differently at 100, 1,000, and 10,000 Hertz. Feedback is like a BandAid in this case. Distortion is different, damping is different, the internal current demands are different, the internal gain is different for each frequency, but it looks OK. Fundamental tones are treated differently than the harmonics. As people say today, OMG,WTF is going on here. And we are ignoring phase issues completely in this discussion.

    Thus the feedback based current-to-voltage converter as well as a buffer/filter stage(s) used to make the filters when implemented in an Op-Amp gets to 100% feedback. Listen carefully, you can hear it sucking the life out of that lovely and amazingly complex event, the rendering of complex music timbre. Thus Peter Baxandall's remarks about amplifiers from 70 years ago are explained, nothing has changed.

    So I drug you all the way to this point to hit the real barrier, feedback! The savior of mediocre design looking for great numbers that is killing the gift of the muses, the music.

    My answer is a different amplifier topology it all in current-mode rather than voltage. Here is an image of one of my current stages.
    DSCF0449.JPG
    There is no feedback, the distortion does not change with frequency nor do any of the other issues, the gain is consistent with frequency up to about 40mHz, and it sources and sinks current rather than following voltage on the input. Pretty cool, same circuit that I use in the Black AMP. Each stage module has its own power regulation, I use 4 of the these in the DAC (L+, L-, R+, R-) for balanced operation of signaling.

    I am now strapping these together with a simple Analog Devices DAC and then with the R2R DAC, the ear will tell. I will be back when I have done that with more.
     
    • Like Like x 8
    • Epic Epic x 3
    • List
  11. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    I have been amp-busy - trying to define SLAM and HEFT as electronic configurations, ultimately into create code or circuits as I bring The Black AMP into a production product. I have found an alternative for the I/O Wave digital front end, and a R-2-R board, my I/E converters are tested and done, demo boards on the way. A busy two months including medical issues, visits by family, you know the stuff. Now cleared and ramping back up.
     
  12. Beefy

    Beefy Friend

    Pyrate
    Joined:
    Apr 10, 2021
    Likes Received:
    1,738
    Trophy Points:
    93
    Location:
    Canada
    Perhaps I'm missing something. I'm not quite sure why you need a current converter from an R-2R DAC, particularly one of the Soekris modules. It is inherently a voltage output system, a Vref being divided by a resistor ladder. You can already take that output voltage directly with zero amplification and zero feedback, and pipe it directly into just about any amplifier.

    I mean sure, in a commercial product, for safety reasons, and for people trying to drive a low impedance or reactive component directly, you probably need a buffer. But why a current converter specifically?
     
  13. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    Ahh, a delightful question.

    I think you are asking 'why even bother with a current based solution in view of the ease and simplicity of a voltage system?'. A question I ask every day and always come back to the same conclusion.

    I don't have enough data to fix an answer yet.

    I do now believe that I will have enough data when I complete the two prototypes I am now building and get my listeners to evaluate.

    As I have said before, R2R is VERY seductive because of its mythology. The down sides are 1) cost, 2) parts availability, 3) parts stability.

    Those sneaky little f'ing resistors present some problems in today's practical international supply chain. Their cost is going up, I have concerns about their aging qualities and vendor QC issues, scalability in general and quality specifically. And they cost a lot. Did I mention that the cost a lot?

    In the part of my heart driven by technical facts and my experience with hard electronics I know that ICs are better for manufacturing and absolute performance, stability and consistency, availability and most of all - cost.

    I believe they can be made to sound a lot better when used differently, even better than discrete R2R. I think I know how to do that in defining the relationship between the actual R2R process in the chip. My first kludged thing was a remarkable sound built from cheap Chinese Ebay hardware and my secret-sauce I-E converter.

    What I am building now lets you switch the output of the digital input processor from a R2R DAC and a AD1955 chip and then through my current-mode amplifiers/buffers. We can compare this pair of paths with two standard products, a stand alone Scandinavian unit of very high repute and an ESS based unit in a four-way kind matrix switch/gain matcher. We then can real-time compare all four paths from USB to headphones or speakers with the same content.

    Then I think I will be able to answer this for myself and I will be glad to share with you.

    My decisions will be made on good old collective human observation. I am seeking an altered metaphysical effect from our hardware, the goal is eargasms. The measurement device is a bunch of good ears connected to good minds.

    Thank you for your question and observation thereof. If you have suggestions I would appreciate them, in the end this is a collective project made up of good minds.
     
  14. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    I see that I failed to answer your question, why current?

    I am looking to current because I have found a current-based transistor technology that we are only beginning to get experience with that has remarkable sonic proprieties while accomplishing voltage gain. I use in The Black AMP and from day one it was significant.

    True, R2R DACs can be run in either voltage or current configurations. While it would seem to be obvious voltage is the simplest and direct way to drive an amplifier, we live in a world of unintended continuances.

    You can build anything you want for home use but once you go out the front door and offer it up to the public all the rules change.

    One my not like, may even find it repugnant but all commerce comes with conditions, rules, morals, goals, expectations and regulations that have nothing to do with the technology or product, and everything to do with the sociology and psychology of the process. In the end we must make our technology fit into the process. Spend some time as a clerk at Home Depot and you will walk away asking your self how people can be so smart and yet s stupid.

    It is said that common sense is not a gift, it's a punishment. Because you have to deal with everyone who doesn't have it.

    Current-mode technology lets me build a product that will have exceptional performance values at modest cost especially in my case, a low volume bootstrap business. If I grow then costs will come down, for now costs are a driver and R2R is less favored if can get favorable equal the same bang for less buck.

    Besides, current-mode hardware just sounds better as I and others are learning. It is early in the game, be patient and have fun.
     
  15. BarryT

    BarryT MOT: Austin Audio Works

    Pyrate
    Joined:
    Feb 8, 2021
    Likes Received:
    425
    Trophy Points:
    53
    Location:
    Austin, Texas
    Home Page:
    Hello Folks, this project has not died but I have been totally engaged in the up grades for both the Black Swan phono preamp and the BlackAMP headphone amplifier after suggestions on their tours on SBAF. I am now stuffing newly arrived printed circuit boards and moving the chassis from the machine shop to the powder coaters and laser etchers. I am also building a pre-production 100 watt version of the BalckAMP.

    Doing this first production run has eaten up all my time and I had to set the DAC test platform aside for a couple of months. Current scheduling looks like late September to finish it and report back to you.

    If you have followed the BlackAMP and the Black Swan tours and the fine reception to them on the SBFA Loaner pagers please note the a limited number of production units will be available at the end of next month.

    This would be a good time to pre-order a unit for deliver then. No deposit required at this time.

    Thanks, Barry
     

Share This Page