Is this schematic "as built"? I see C38 and C39 which need to be removed.
yes as built. I tried removing these 2 caps previously and it didn't affect the circuit. adding an elect. cap made it stable even with these 2 caps.
Zero reason to add them back. They do nothing but slow the feedback which can increase the noise. Feedback is what reduces noise. Slow it and the noise is not properly mitigated. Get rid of them permanently.
I see three separate filtering circuits. I don't understand why you need them to feed the on board switchers. Filtering the input to U1 is useful, but not going to the switchers. Does Vcc go on to feed other sensitive circuits? If so, then the filters prevent noise from the switcher inputs from feeding to the other circuits. You do need the larger cap and the smaller cap to smooth the current pulses.
one for each switcher, that makes it 2 filtering circuits. and another one to deliver 12v input to output connector which goes to dreamcast and it is used to drive gd-rom drive motor... no need for it to be low noise at all, but also it shouldn't add the noise to the system as it uses the same ground. filtering it this way was the solution. this rail also is the input to the 5v reg and linear stage op-amp as you see, which need to have low noise.
You seem to want to add things without understanding how they work. I assume the dreamcast circuitry is fed through the output connector on the pin labeled "12V"? That pin also feeds the 5V regulator, so any noise coming back from that is fed into your voltage reference. Why not feed that from the Vcc input before the filter?
Remove the two filters feeding the two switchers leaving a 22uF cap and a 100nF cap. With a 3 amp load, you might want to increase the capacitance from 22 uF to 100 uF.
C * dV = I * dT, dV and dT are deltasI don't know the frequency your switcher is running, but assuming 1 MHz, that gives...
dV = I * dT / C = 3 A * 1 us / 100 uF = 0.030V So 100 uF will give 30 mV ripple going into the switcher which should be fine. 22 uF will be 136 mV ripple.
Also, the ground circulation paths should be isolated from the rest of the design if you can. A very short path between the high current/high frequency loops on a separate area of copper, like an island. Connect that area to the rest of the ground plane via a single connection. Remember the thermal breaks I was complaining about in one of your layouts? One connection like your thermal break. This prevents voltage differentials from interfering with the rest of the circuit.
filtering switchers input is not really necessary but it would be a problem if the 12v input itself is very bad noise and ripple, I thought it might affect the switchers so i added a small inductor with caps.
That is not of use. The place to add filtering is between the switcher and the linear.
It is important to route the paths around the input, through the input caps back to ground very short. This path is an important source of noise. The current through the output inductor and cap are not. The current through the inductor is pretty constant... that's what inductors do. The diodes D1 and D2 need to be as direct as possible between LX and ground.
the layout is done properly and took a lot of time. very good layout as datasheet and better.
I don't see a reason to not increase R1 from 100 ohms to 1K. Worse case draw on the output of that regulator is 2 mA, so dropping 2 volts won't harm a thing. Higher resistance gives higher attenuation of noise.
this may aid in making it more low noise, but the problem was that the circuit wasn't functioning to begin with.
You asked for advice. This is my advice. If you want to have noise in your circuit, the reference will cause it more than anything else.
Much of your original design was either adding ineffective components like the filtering before the switchers or even counter productive like the cap on the FET gate. The reference is the gold standard for the entire rest of the design. Keep that clean or the output will be hard to clean up.
Who said the output cap needs to be 220V??? That is an absurdly high voltage.
actually it is 220uF 16v, that was a typo.
This is your original design from weeks ago, right?
this is the design that was produced, yes.
and the "low_noise_psu.pdf" is a totally new design based on new TI switcher which promises <1mV of total ripple and noise. I spent some time with this to make it very good layout. please check it and see if you have a comment. the boards are not manufactured yet, only quotation.
I don't see a layout, only schematics.