Decided to test this circuit on my Magni 3 headphone amplifier (discrete current feedback amplifier). To do this I had to make two add-on boards. I broke the circuit like this:
View attachment 1080502
For this to work I had to completely remove the LM3x7 voltage setting resistors and replace the 120R-240R one (don't know the actual value Schiit used) with 2.7K. Also Cadj is gone.
View attachment 1080505
The actual add-on boards look like this:
View attachment 1080503
View attachment 1080504
And implemented on the PCB it looks like this:
View attachment 1080506
View attachment 1080507
View attachment 1080508
Not the cleanest job but I was in a hurry. Also I had to add the protection diodes which was a bit of a pain. Schiit omitted them.
I sensed the output on the 0805 100nF (removed) capacitor footprints that are in parallel with the output caps, which is a convenient place.
I set Vout at 16.7V instead of the stock 17V. Figured I leave some extra room. Output caps got replaced with 220uF for +Vout and 330uF for -Vout. These values seem to keep an acceptable startup current in the base of the denoiser BJT.
Actual used values:
View attachment 1080509
I used 2.2nF compensating capacitor and seems fine and stable.
For BJTs I chose the higher gain BC817-40 for Q6 as lower gain ones require a larger R15 and I couldn't be bothered. I don't have very high values resistors.
D21 is an IR LED with a Vdrop of around 1.1V. The bag of LEDs I took it out of had CQY99 written on it.
R14 50R value makes for around 7.5mA to the denoiser. C14 could have been 22uF for more LF PSRR performance but it would increase the turn-on surge current into Q6 base. Not by much but I decided to play it safe. This way I don't need the protection series (with C14) resistor.
To limit startup first swing I used 4x 1n4148 diodes in series and seems to work fine. On positive rail I get 17.1V max swing and on negative rail I get 16.9V. Both rails sit at around 16.7V nominal. Negative version needs around 1meg for R15 for same operating point.
Vdrop across the diode string is around 0.5V (denoiser side lower) after everything settled.
As far as performance goes it's extremely good! Here's some before and after measurements.
Positive rail before:
dBV scale
View attachment 1080510
Noise
View attachment 1080511
Positive rail after mod:
dBV scale
View attachment 1080512
Noise
View attachment 1080513
Negative rail is similar. Before:
dBV scale
View attachment 1080514
noise
View attachment 1080515
And after mod:
dbV scale
View attachment 1080516
and noise
View attachment 1080517
One mention is that measurements before were done without the denoiser half of the mod, only the NoNoiser way of setting Vout, which is similar to how a regular LM3x7 without Cadj performs. I had Cadj removed from an earlier mod and couldn't be bothered to install it for the before measurement. So stock Magni 3 might have a bit better performance than what I measured without Cadj.
I also measured the headphone output but sadly I completely forgot to ground the inputs or set the potentiometer, I have no idea of its position during these measurements. Still take it as a differential measurement before/after mod.
Noise for one of the channels before mod (low gain setting):
View attachment 1080518
And after mod:
View attachment 1080520
So even with BC8x7 for denoiser BJT performance is incredibly good. On high gain setting with the pot maxed out and I still can't hear the least bit of noise. It's dead quiet.
