DAC 2496 (AK4393) DAC KIT With CS8416+AK4393+5532

[Trileru]

Thanks for the info. I need to start looking into some of the things you've mentioned. I didn't get into clocks yet so I have some to catch up.

[Trileru]

I have compared in LTSpice LM317 and 1117-5V and 1117-adj versions. These all can be found in the sot-223 package and are compatible with the footprint on the pcb. The findings are interesting. Even though the 1117-5V + dienoiser certainly has better performance than just a single 1117-5.0, lm317 and 1117-adj versions certainly are tempting.
lm317 has a higher PSRR, then the 1117 adj version then the fixed voltage. I'm not sure how important this aspect is for the position this regulator has in the circuit. Theoretically the LM7812 already deals with the ripple.
This is the PSRR simulation results:



As you can observe 1117-adj + dienoiser has a PSRR somewhere in between 1117-5.0 +dienoiser and lm317+dienoiser.
The original single 1117-5.0 has poor PSRR compared to any of the dienoiser enhanced regulators.


Now for the output impedance results (I didn't add the simple 1117-5.0, it has around 15mOhm):





As you can see the lm317 has the lowest output impedance while the 1117-adj has the most linear one in the audio range.
Zoomed in for a better view of lm317:



The output impedance of the lm317 is around 6uOhm at 100Hz. But considering that the 1117-adj is more linear I think it would be more desirable in the position of feeding AVCC/Vref. For 1117-adj + dienoiser you'd only need an extra resistor compared to the 1117-5.0 + dienoiser mod. And conveniently enough it goes across pins 1 and 2. This would also offer the advantage of being able to better tweak the output voltage. The 1117-5.0 + dienoiser sits at around 5.3V and it's pretty close to the 5.5V max rating of the DAC.
edit: ideally you'd source legit lt1117-adj ones. I'm not sure how ams1117-adj behaves. For reference while ld1117 adj or other prefixes are around 10-30 cents each, the lt1117 (from linear technology/analog devices) is around 3.5$ each.

[Trileru]

In the end I used LT1085+dienoiser for the VACC/Vref rail, and I also replaced the ams1117-3.3V from the CS8416 VA line (spdif clock recovery) with a LM317 + dienoiser. I had to cut the tab off both to fit them. I tested the exact reg + resistors + dienoiser add-on board on a diy supply so I make sure the voltage does not exceed the max allowed on each of the two rails.

I used 100uF/63V Panasonic FC output cap for the LM317 circuit. The ESR is a bit higher to make sure the LM317 likes it.

Seems to be working fine and it sounds absolutely great!

[Markw4]

Very interesting how that works

[Trileru]

Here's a comparison for the 3.3V rail noise (with 60dB LNA), orange is simple LM317 with just the two resistors and the teal line is the LM317+dienoiser.
I still had a LT1084 I could have used but I figured the more linear output impedance in the audio range is not of use for the clock recovery rail. The lower noise is what is useful so I figured LM317 + dienoiser is good enough.

[Markw4]

One can certainly figure to estimate probably of a good result. Nothing beats checking with an experiment though. Now and then I have been rather surprised.

[Trileru]

That's why I decided to build the LNA, so I can make sure the implementation is done correctly. The dienoiser is most likely to oscillate than not, without adjusting the comp values to the particular application.
But the denoiser version should work without issues, and it offers good performance closer to the dienoiser than to the stock LM3x7/LM78xx/LM79xx implementation.

[streametch]

Hi Trileru, Thanks for sharing and test with D-Noizator mod, will try it on VACC/Vref 5v