MD x Airist x Sosolar: RDAC design mysteries

Discussion in 'Digital: DACs, USB converters, decrapifiers' started by bimmer100, Jun 5, 2018.

  1. Thad E Ginathom

    Thad E Ginathom Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    12,000
    Trophy Points:
    113
    Location:
    India
    Curious, that. You want to respond to mentions of old reports about stuff that may or may not have happened a while back, but you are completely silent on the thread topic? Not even to say that you don't have anything to say, or are not able to say anything or... anything.

    Just... this sticks out like a sore thumb.
     
  2. JustAnotherRando

    JustAnotherRando My other bike is a Ferrari

    Contributor
    Joined:
    Aug 15, 2017
    Likes Received:
    1,371
    Trophy Points:
    93
    Location:
    Hong Kong
    Well, refer to the previous posts about not saying anything that might be used against you if this turned into a legal mess. So it's understandable if Massdrop refuses to comment. Even if I don't agree with it.

    Headfi joke completely coincidental!
     
  3. Thad E Ginathom

    Thad E Ginathom Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    12,000
    Trophy Points:
    113
    Location:
    India
    There is that. But I would have preferred something like "Cannot comment at this time," or something!

    The absence of that looks bad. It's like sod the elephant in the room, let's nitpick this thing. I'm not saying this because of the specifics; company story and individual. It's a general thing. At least say can't say. Just seems rude not to.
     
  4. lehmanhill

    lehmanhill Almost "Made"

    Contributor
    Joined:
    May 3, 2018
    Likes Received:
    421
    Trophy Points:
    63
    Location:
    Michigan US
    Looking past the ethics and IP issues, another worry will be production quality. I understand that Mass Drop has provided quality products in the past, but it does appear that the RDAC is being built down to a price.

    In my days in the auto industry, the early production cars that were sent out for press reviews were built with carefully selected components, then fluffed, buffed, tuned, and tweaked to perfection. The normal customer didn't get this level of service.

    While it is entirely possible that Mass Drop will be able to provide consistent high quality in production, it seems like a wait and see attitude would be wise. I would like to see one of the big 3 SBAF reviewers have a listen to a random production RDAC, just to see if it still has the magic.
     
  5. Torq

    Torq MOT: Headphone.com

    Friend
    Joined:
    Jan 15, 2016
    Likes Received:
    8,187
    Trophy Points:
    113
    Location:
    Seattle, WA
    All "budget" level gear is "built down to a price".

    The amount one might save on a small-unit run (and ~1,000 units is a small run) on a $349 (retail) unit is unlikely to be enough to be worth the paperwork, let alone the risk. You have to get your unit count way up to make any kind of meaningful saving. Board assembly will be automated ... switching vendors would be more costly than anything you might save in their ability to use a less-precise/well controlled tool there.

    While "quality fade" happens, it's usually with recurring, large volume, orders - and typically not on the first couple of rounds of those.

    As for listening to a "random" production RDAC ... someone that's not already known to Massdrop will have to buy one for that to occur. I'm in on the drop, so I'll get a production unit, but if you're worried about "cherry picked" product going to reviewers me, or @purr1n or @Hands buying one directly won't eliminate that concern.

    Waiting for an audition before buying one is one thing, can't argue with that for any piece of gear, but otherwise this seems like an excessive amount of paranoia. They seem to keep things pretty consistent with their other "x" series products.
     
  6. lehmanhill

    lehmanhill Almost "Made"

    Contributor
    Joined:
    May 3, 2018
    Likes Received:
    421
    Trophy Points:
    63
    Location:
    Michigan US
    Fair enough. I can see that my automotive analogy doesn't apply. There, a small run is 100,000 units and a penny saved is millions of dollars earned. Clearly not the scale of audio electronics. I bow to your greater wisdom and experience.
     
  7. spwath

    spwath Hijinks master cum laudle

    Friend BWC
    Joined:
    Dec 13, 2015
    Likes Received:
    7,801
    Trophy Points:
    113
    Location:
    Madison, WI
    A penny saved on 100,000 is $1000. $1000≠millions
     
  8. rshuck

    rshuck Friend

    Friend
    Joined:
    Mar 8, 2016
    Likes Received:
    41
    Trophy Points:
    18
    Location:
    Bridgeport, WV
    What, are we supposed to be math scientists?
     
  9. zenelectro

    zenelectro Rando

    Joined:
    Jun 12, 2018
    Likes Received:
    9
    Trophy Points:
    1
    Location:
    Australia
    What Soekris is trying to say and I can explain a little bit further is as follows:

    The correct way to measure and specify noise for a DAC (or any audio device) is RMS noise within a certain bandwidth.
    Usually it is 20Hz to 20kHz or IOW a 20kHz bandwidth. With noise measurement you MUST specify the bandwidth or the
    measurement is meaningless.
    When you look at a typical FFT plot from say 20Hz to 20kHz of a DAC there will be a certain noise floor shown or what we
    can sometimes call in slang 'the grass'. If this is at the -150dB level it does not mean that the DAC has a noise floor of
    -150dB. The FFT is the measurement instrument doing a slow sweep from 20Hz to 20kHz and plotting that sweep. The actual bandwidth of that sweep is quite narrow and will determine how low the grass is.

    For example if you look at a typical 16 bit FFT noise floor, the usual FFT 'grass' should be close to -130dB. This can be seen for example on Stereophiles measurement of Benchmark DAC3, Fig 5. For RMS noise floor of about 96dB, the 'grass' is usually over 30dB lower typically depending on the test gear sweep BW measurement settings. So a full 24 bit noise floor of -144dB will show measurement FFT 'grass' close to -180dB.

    Getting back to the DAC in discussion, if the FFT measurement shows noise floor around -150dB, then the RMS noise in audio band (20kHz) will be somewhere around -120dB maybe even less.

    Now - to take this one step further, a DAC's dynamic range is always specced with a -60dB signal present. This can show noise floor modulation with signal present. One can compare the 'grass' level at 0dBFS to -60dB to no signal at all and see the change in level that is correlated to signal level.

    All this stuff sounds very technical and most people will say 'who cares, as long as it sounds good' which is fair enough. However these are all standard ways of measuring audio equipment so we can see them on a level playing field and make sense out of them rather than just propelling big numbers of hyperbole which have no real context.

    Terry
     
    Last edited by a moderator: Jun 12, 2018
  10. Thad E Ginathom

    Thad E Ginathom Friend

    Friend
    Joined:
    Sep 27, 2015
    Likes Received:
    12,000
    Trophy Points:
    113
    Location:
    India
    I leave all that stuff to people like @atomicbob
     
  11. mscott58

    mscott58 Friend

    Friend
    Joined:
    Oct 13, 2015
    Likes Received:
    2,024
    Trophy Points:
    113
    Location:
    Philly - Yo
    Dude, it's not like I'm a rocket surgeon or anything...
     
  12. Walderstorn

    Walderstorn Friend

    Friend
    Joined:
    Mar 20, 2016
    Likes Received:
    1,643
    Trophy Points:
    113
    Location:
    Europe
    Its really a shame. I was a backer and lost interest after all this. I have been avoiding MD more and more and this is just another situation that keeps me away from it.

    So it's life but I, ethically, expected more from them. If they would had come clean from the get go, even with the bs of "it took us years to develop but we outsourced a part of it and had zero knowledge of it after all" than it would've been better (barely) but MD stand on this is what really pisses me off.

    That said seems they have sold a good amount of it.
     
  13. Baten

    Baten Friend

    Friend
    Joined:
    Mar 18, 2018
    Likes Received:
    1,071
    Trophy Points:
    93
    Location:
    EU
    So I wanted to get on board with this drop but my CC company refused the transaction for some reason and I therefore missed it. Maybe I could still hop on by the time it's actually in production but I'm second guessing if I really want this DAC.

    I want an R2R though. Maybe I could wait on sosolar's version and save up some more. I like his idea's of DSD2048 upsampling, focus on low jitter and possible balanced version. Decisions..

    Edit: sorry if I come across as a noob really, this is my first R2R purchase and I'm not entirely sure what to go with (there's soekris, schiit, metrum, airist, older-gen ones like the PCM1704, ...)
     
    Last edited: Jul 4, 2018
  14. Torq

    Torq MOT: Headphone.com

    Friend
    Joined:
    Jan 15, 2016
    Likes Received:
    8,187
    Trophy Points:
    113
    Location:
    Seattle, WA
    Doing that on a discrete R-2R board doesn't make sense. You don't need an array of precise resistors for DSD, nor do you want resistor strings. In fact you really only need a high-speed switch and a capacitor. Running the DSD bit-stream through the ladder would not get you anything ... you'd either be using a single rung (pointless) or you'd be converting to PCM.

    So, the only way it makes sense to handle DSD via an R-2R DAC is to first convert it to PCM.

    Upsampling to 2048 DSD (which to do properly will require some fairly serious horsepower*) only to convert it back to, roughly, 16-bit/2819.2 kHz PCM (or 24-bit/1880 kHz) to stuff through the ladder, gets you what? An easier time in building your filters ... but with much more processing power required to do it.

    Oh, and that's going to require much faster switches than are generally used in products of this nature/level. And a lot of these types of boards use shift registers rather than actual switches, which means they, and the FPGA driving them, is going to need to be faster still.

    --
    *As a comparison, using HQPlayer with it's "best" settings for converting to just DSD512/filtering tends to need a dedicated PC with some serious grunt. I've seen people doing this with 16-core Threadrippers because they couldn't get glitch-free performance with their normal desktop machines.
     
  15. Baten

    Baten Friend

    Friend
    Joined:
    Mar 18, 2018
    Likes Received:
    1,071
    Trophy Points:
    93
    Location:
    EU
    Thank you for the insight @Torq , appreciated.

    I suppose only R2R DACs with a dedicated DSD board like the Holo Audio Spring / Cyan can do native R-2R processing of DSD?
     
  16. Torq

    Torq MOT: Headphone.com

    Friend
    Joined:
    Jan 15, 2016
    Likes Received:
    8,187
    Trophy Points:
    113
    Location:
    Seattle, WA
    Per Holo Audio, the DSD version of the Cyan does NOT use an R-2R configuration for DSD processing:


    To use a resistor ladder to convert DSD to analog either requires converting to a multi-bit format, in part or in whole (and in doing so immediately, defeats one of the biggest, claimed, benefits of DSD ... namely perfect linearity in the conversion of each pulse), or it requires only partial use of the ladder (i.e. one resistor, typically the MSB).

    --

    It's worth noting that PCM -> DSD and DSD -> PCM (or other multi-bit representations) conversions are not mathematically bit-perfect even though they should be audibly transparent. In other words, if you take a PCM file, convert it to DSD, and then convert it back to PCM you will not get the same data you put in. The same is true for DSD files converted to PCM and then back to DSD.
     
  17. dmckean44

    dmckean44 In a Sherwood S6040CP relationship

    Friend
    Joined:
    Nov 17, 2016
    Likes Received:
    1,418
    Trophy Points:
    93
    Location:
    Peoria, IL
    Is it really as simpale as a high speed switch and a capacitor for doing proper DSD conversion?
     
  18. soekris

    soekris MOT - Soekris Engineering

    Friend
    Joined:
    Mar 25, 2016
    Likes Received:
    944
    Trophy Points:
    93
    Location:
    Denmark
    Yes, DSD to Analog is in principle a clocked driver (a flipflop) plus a lowpass filter. The complications is that the clock and driver need to be very precise and the filter need to be really sharp, like a 8th order one, to keep the quantization noise down....

    I consider it easier and better to convert the DSD signal to PCM by filtering it in the digital domain and then do volume control in the digital domain....

    Actually I consider DSD to be a crap format, moving the Delta Sigma Modulator from the DAC to the Encoder, thereby fixing it at one bit.... And you can't do digital volume control on DSD signals.
     
  19. sosolar

    sosolar Hibiki DAC Designer

    Joined:
    Jun 7, 2018
    Likes Received:
    78
    Trophy Points:
    13
    Location:
    China

    Any thought about treating R2R ladder as a binary filter to decode DSD stream?
    Let's think about how DSD decoding network: Yes we do use numbers of shift registers to averaging DSD stream to get a decent 'analog value', resistor are at the same value, so the 'analog value' shoulbe be (x1*1+x2*1+x3*1....xn*1)/n. Or if you like, it is an avereging decoing network. Another way to do so is replace these same value resistors with somehow quantized from window function like han/rectangle/bessel, to form a analog filter achieving better frequency response(because we got a hardware filter here!). These are the common way to decode DSD.
    However, I have develop another aproach to decode algorithm to do this job is to shuffle DSD stream with a pattern that can force r2r ladder to be a analog filter. I name it binary filter...this is why my new product is named BDS(Binary Decoding System)
     
  20. soekris

    soekris MOT - Soekris Engineering

    Friend
    Joined:
    Mar 25, 2016
    Likes Received:
    944
    Trophy Points:
    93
    Location:
    Denmark
    As I said, I consider DSD to be a crap format, don't seems to be a good idea using resources trying to improve playback of it....
     

Share This Page