Advice for Schiit on how to improve Magni 3+ power supply

Discussion in 'How to Win Friends and Influence People' started by kris2014, Mar 31, 2020.

  1. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us
    Digging into engineering details (to distance myself from asr), it looks possible to achieve the goal. Schiit messed up power section of Magni 3+. Just a little tweak would improve LM317/337’s output performance by a good margin. LM317/337 cannot drive large capacitance load and it doesn’t benefit from it. Schiit currently uses a MLCC as Cadj and this will result in voltage resonance near 5Khz. They probably tried to mitigate the problem by adding those 1000uf in parallel with MLCC. However this would only cause LDO’s voltage regulation to slow down. Those 1000uf (2200uf in heresy) is all the amp has to rely on when LDO is out of linear region. Also, if Adj pin is not isolated from input reservoir, it would increase output impedance of the LDO.

    This is probably why transient measured by Atomicbob looked a bit weird on both magni3+ and heresy.

    Sadly i cannot design a discrete amp to join the contest.
     
  2. schiit

    schiit SchiitHead

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    9,937
    Trophy Points:
    93
    Location:
    Texas and California
    Home Page:
    Measurements (and our actual experience) does not show this. We went through plenty of little tests and tweaks on the power supply, trying to make it better (and confirming on the APx555.) What we have is optimal, in the absence of post-regulation (which would drop the voltage) or exotic regulators (which would not work within the Magni budget.)
     
  3. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us
    well, you can try removing output MLCC and substitute that 1000uf with a 220uf. Then use a tantalum or regular 22uf electrolytic cap as Cadj (resistance is needed). I bet performance will increase. Isolating Cadj, Radj with input neutral GND with a 1ohm resistor will lower output impedance a lot.

    if that 1000uf is really needed, a 1R between it and the LDO output will also help.

    MF filtering is better done before LDO rather than after LDO.
     
  4. schiit

    schiit SchiitHead

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    9,937
    Trophy Points:
    93
    Location:
    Texas and California
    Home Page:
    Like I said, we've tried a lot of things, and verified results on the APx555. What you have described has not increased measured or transient performance. What would help is post-regulation, even just a cap multiplier (which will reduce noise in discrete circuits with low PSRR), but I don't have the volts to throw away. We tried active tricks on the 317/337s as well, but they were prone to oscillation, and took more parts and added more cost.

    Not saying it can't be made better, we just haven't found a way to do it without significant additional parts or voltage loss. This is a very cost-constrained device.

    I understand the desire to tweak. Please understand our desire to produce an affordable, reliable, very high performance product, based on actual experience and measurements.
     
  5. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us

    Your current design was in a case where resonance happened. It’s not really new.

    LM 317/337 won’t benefit from a large cap downstream at all. Above 220uf noise won’t be lower and nothing helps with its higher frequency noise at all unless it’s filtered pre-regulation, as data sheet suggested.

    Low ESR caps on the output will only increase output resonance. That’s why you got oscillation when a multiplier was inserted there.

    Capacitance multiplier should be place before those chips.
     
  6. schiit

    schiit SchiitHead

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    9,937
    Trophy Points:
    93
    Location:
    Texas and California
    Home Page:
    Cap multiplier afterwards absolutely lowers noise in low PSRR circuits. 317/337 have plenty of HF noise on the output, regardless of input smoothing. I'm not talking ripple, I'm talking noise. I can show you this on the AP. We use that on several other products where we can lose the volts.

    Oscillation was caused by active noise cancellers, not by cap multipliers.

    Again, this is based on actual building, tweaking, and testing. The answers would be different on a cost-unlimited product, or one with more volts. And the answers may not be to the datasheet, but based on our own experience. Please give us the benefit of the doubt here.
     
  7. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us
    Lm317/337 has built in voltage reference. Large capacitance load and low esr load would make the device struggle keeping it nominal. It is stated so in datasheet. It is stable so it won’t easily went out of stability region but it’s transient performance will be lowered if it’s implemented that way.

    i do agree that these two LDOs aren’t the best and should be changed if it doesn’t meet requirement.

    BTW, actually what i proposed should lower cost.
     
  8. schiit

    schiit SchiitHead

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    9,937
    Trophy Points:
    93
    Location:
    Texas and California
    Home Page:
    I have a looooong list of stuff on datasheets that doesn't play out in reality. At least not the way you might think it will. This holds true for all parts, including the new exotic stuff. There's a ton of gotchas there, especially with parts claiming low noise. Build it, measure it, see how it works. That's how we'll do it. If you want to tweak afterwards, enjoy.
     
  9. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us
    i just ordered one and will perform experiments myself. We’ll see later this week.

    last week i noticed a pcb i received had a cap polarity marked backwards, as double proven by multimeter. Then it took me 6 long emails to persuade the supplier to check it. He just thought i was dumbass and was showing me all the design details... until he did the measurement that took him 10s to do.
     
  10. schiit

    schiit SchiitHead

    Pyrate
    Joined:
    Sep 28, 2015
    Likes Received:
    9,937
    Trophy Points:
    93
    Location:
    Texas and California
    Home Page:
    I don’t think you’re a dumbass. If you can do us the same courtesy, that would be appreciated. We both have perspectives here...but we also both have different constraints.
     
  11. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    How? Did you measure the final design? This is the best measuring discrete parts headphone amplifier I have ever measured. If you think you can do better design from scratch please do. I too can advise how I think Toyota can improve the weight distribution and suspension tuning of its GT86 and the specific brake and suspension parts they should use.

    Which one are you referring to? How does it look weird? What references to measurements of other gear do you have that would establish it as being weird?
     
    Last edited: Mar 31, 2020
  12. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us
    i was referring to the dc-dc power supply section only. please refer to this article down here for what i was talking about.

    https://www.tnt-audio.com/clinica/regulators2_impedance1_e.html

    This problem was discussed in length elsewhere. Large smoothing caps created the output voltage ringing and would slow the LDO’s response to changing voltage. But if LDO is out of linearity, it can help keeping output stable in a short period of time. Therefore, you can throw in a more expensive 10000uf to mitigate the issue that was created (loop: bigger caps-> issue, but even bigger caps ease the pain a bit).

    The correct logic is to use smaller and cheaper caps there.... Schiit currently used (small quantity price) 2x $1.8 USD and 2x $1 USD caps, with 4x expensive mlcc (likely 0.7 USD each in avg). Changing to smaller 22uf and 220uf electrolytes would cut the price in half with the same psu noise and faster LDO response.

    But let me reiterate, magni3+ isn’t bad in overall performance. However if you are looking to improve it, there is room.
     
    Last edited: Mar 31, 2020
  13. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    Do you have 'scope results to show that this occurs in Modi 3. Or are we talking hypothetically.
     
  14. atomicbob

    atomicbob dScope Yoda

    Pyrate BWC MZR
    Joined:
    Sep 27, 2015
    Likes Received:
    18,653
    Trophy Points:
    113
    Location:
    On planet
    Please define "a bit weird".
     
  15. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    [​IMG]
     
  16. purr1n

    purr1n Desire for betterer is endless.

    Staff Member Pyrate BWC
    Joined:
    Sep 24, 2015
    Likes Received:
    89,778
    Trophy Points:
    113
    Location:
    Padre Island CC TX
    More seriously. I don't want to come off as somewhat of a jerk, but it's SBAF custom to demonstrate and prove the results with measurements when making assertions of how a design could be improved. One could mention ideas, leave them as ideas to investigate; but without proven results, it's not cool, actually kind of rude to say assert stuff like "But let me reiterate, magni3+ isn’t bad in overall performance. However if you are looking to improve it, there is room." because it makes you sound like a pretentious butthead, especially if you are a hobbyist or dabbler with zero constraints. In other words, don't be a Kevin Gilmore, or Gilmore Lite.
     
  17. JellyRhino

    JellyRhino Facebook Friend

    Joined:
    Jan 27, 2020
    Likes Received:
    226
    Trophy Points:
    33
    Location:
    Montreal, Canada
    Time is money. You might be wasting it, or both. Get something more expensive and save money in the end? I thought this hobby was about music.
     
    Last edited: Mar 31, 2020
  18. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us
    This weekend or earlier next week. The article i provided above is a real-world example of the exact same scenario.


    I enjoy experimenting hardware as well as listening to it. So not losing for me.
     
    Last edited: Mar 31, 2020
  19. JellyRhino

    JellyRhino Facebook Friend

    Joined:
    Jan 27, 2020
    Likes Received:
    226
    Trophy Points:
    33
    Location:
    Montreal, Canada
    Last edited: Apr 1, 2020
  20. kris2014

    kris2014 Acquaintance

    Contributor Banned
    Joined:
    Feb 1, 2020
    Likes Received:
    38
    Trophy Points:
    18
    Location:
    us
    I will post measurement later and we will see.

    This specific matter is well documented in technical white papers. www.ti.com/lit/an/slva115a/slva115a.pdf

    Fig 4/5 of this paper shows what happens if LDO output caps’ ESR is too low. Fig 5 shows how ESR improves transient from a LDO output. There are a series of papers dates back to 2000 on this specific matter.

    Rough estimation of LM317 suggests output caps ESR shouldn’t be lower than ~130mOhms for better performance while MLCC implemented can be well below 10 mOhms under 20khz. That’s why small capacitance Tantalum caps (ESR in several Ohms for small ones like 1uf) were recommended as output smoothing caps.

    I simple don’t take it that datasheet doesn’t work only because someone cannot make it work. “We tried” is not a valid analysis.
     
    Last edited: Apr 1, 2020

Share This Page