Low ripple low noise small power supply design

[VEGETA]

I'm stuck in a quarantine lock-down...
Bored to death...

you guys still on that!? we have normal life but after 12:00 midnight no one is allowed.

(Everything else you selected looks fine...)

here are summery of my selections:

Mosfet: gonna get all these 3 but mainly your suggestion for the default position.

https://lcsc.com/product-detail/MOSFET_PUOLOP-PTD3006_C479048.html

No...
https://lcsc.com/product-detail/MOSFET_PUOLOP-PTD3006_C479048.html

you mean this one won't work? why? it has the required Vgs. I wanted to order 5 pieces of it, but I am gonna stick to the one you mentioned before.

gnuarm

I suggest you go from last pages backwards then see the first schematic you stumble upon. We made 4.6v for 3.3v linear stage and about 6.25v for 5.5v stage. Plus, we use Mosfets for power regulation instead of ready-to-use linear ICs.

Cap multiplier was chosen since I liked the idea of ultra low noise for retro console (Dreamcast) instead of available solutions which uses switching regulators only.

[gnuarm]

gnuarm

I suggest you go from last pages backwards then see the first schematic you stumble upon. We made 4.6v for 3.3v linear stage and about 6.25v for 5.5v stage. Plus, we use Mosfets for power regulation instead of ready-to-use linear ICs.

Cap multiplier was chosen since I liked the idea of ultra low noise for retro console (Dreamcast) instead of available solutions which uses switching regulators only.

That's ok.  What you are not addressing is what you expect the cap multiplier to do that the linear regulator won't do.  Have you built anything that you found to not be suitable?  You talk about having the cap multiplier in contrast to a switcher, but that's not the question I have.  What do you think a cap multiplier will do that a linear regulator won't? 

I found a schematic using the LM317 in place of the capacitor.  So you have dropped the idea of a capacitor multiplier and are using an NPN to boost the regulator output.  Fine, I hope someone has advised you about this potentially affecting the stability of the LM317.  So be aware of that.  I just wanted to dispel the idea that a capacitance multiplier is better at removing low frequency (audible) ripple than a properly designed linear regulator.  What you have now IS a linear regulator.  These can be found to handle your load current, so you could do the same thing in a single package without the NPN.

[VEGETA]

That's ok.  What you are not addressing is what you expect the cap multiplier to do that the linear regulator won't do.  Have you built anything that you found to not be suitable?  You talk about having the cap multiplier in contrast to a switcher, but that's not the question I have.  What do you think a cap multiplier will do that a linear regulator won't?

I found a schematic using the LM317 in place of the capacitor.  So you have dropped the idea of a capacitor multiplier and are using an NPN to boost the regulator output.  Fine, I hope someone has advised you about this potentially affecting the stability of the LM317.  So be aware of that.  I just wanted to dispel the idea that a capacitance multiplier is better at removing low frequency (audible) ripple than a properly designed linear regulator.  What you have now IS a linear regulator.  These can be found to handle your load current, so you could do the same thing in a single package without the NPN.

check the schematic in here: https://www.eevblog.com/forum/beginners/small-mosfet-for-capacitance-multiplier-in-a-small-psu/msg3331484/#msg3331484

it is the final one, but a bit more finalized in kicad.

Here is my little explanation of this:

I feed 12v from a cheap or not-so-good generic switching supply to this board. The board must filter it properly to deliver as clean as possible 5v and 3.3v (amperage different between them).

12v is taken through a 100R resistor with lots of caps (1000uF + 22uF +100nF) into the 7805 linear regulator which outputs 5v and 3.3v via resistor divider. This "reference" signal is very clean since it was taken from a resistor divider which is the cap multiplier part of this design. the cap multiplier here is taken from V_filtered which is about 10v due to the dropout... it is the signal fed to the supply of the op-amp and then will be delivered to the gate of the mosfets. We kept changing it until we reached this formula.

Getting all of that into the linear 7805 is very beneficial to attenuate low frequency noise as you stated which is a further plus along with high capacitance + inductance which takes care of high frequency noise.

the rest is straightforward, the op-amp is fed with 3.3v and 5v from the reference clean signal and outputs the necessary to make the linear regulator mosfet regulates.

the trick here is the cap multiplier at the op-amp stage.

So is this a linear regulator? yes of course, but it is powered or enhanced by capacitance multiplier + switching pre-regulator.

If you are gonna design such a thing, what are you going to do? what parts to use? me and Brian took this project to the maximum by keep changing it and enhancing it... till we wished it end xD

[BrianHG]

I'm stuck in a quarantine lock-down...
Bored to death...

you guys still on that!? we have normal life but after 12:00 midnight no one is allowed.

(Everything else you selected looks fine...)

here are summery of my selections:

Mosfet: gonna get all these 3 but mainly your suggestion for the default position.

https://lcsc.com/product-detail/MOSFET_PUOLOP-PTD3006_C479048.html

No...
https://lcsc.com/product-detail/MOSFET_PUOLOP-PTD3006_C479048.html

you mean this one won't work? why? it has the required Vgs. I wanted to order 5 pieces of it, but I am gonna stick to the one you mentioned before.

At room temp, the Vgs is 4v at 3 amps.  That's twice as high as the other mosfets.  We wanted as low as 3v.
If you sell this device into a cold temperature area, like 10 degrees C, that Vgs would even increase a bit.
Also, the graphs we are provided in the data sheets do have an error margin associated with them.  This is one area where bipolar have much tighter voltage base emitter voltages.

However, all the other chosen mosfets Vgs at around 2-2.5v are fine.

[VEGETA]

I'm stuck in a quarantine lock-down...
Bored to death...

you guys still on that!? we have normal life but after 12:00 midnight no one is allowed.

(Everything else you selected looks fine...)

here are summery of my selections:

Mosfet: gonna get all these 3 but mainly your suggestion for the default position.

https://lcsc.com/product-detail/MOSFET_PUOLOP-PTD3006_C479048.html

No...
https://lcsc.com/product-detail/MOSFET_PUOLOP-PTD3006_C479048.html

you mean this one won't work? why? it has the required Vgs. I wanted to order 5 pieces of it, but I am gonna stick to the one you mentioned before.

At room temp, the Vgs is 4v at 3 amps.  That's twice as high as the other mosfets.  We wanted as low as 3v.
If you sell this device into a cold temperature area, like 10 degrees C, that Vgs would even increase a bit.
Also, the graphs we are provided in the data sheets do have an error margin associated with them.  This is one area where bipolar have much tighter voltage base emitter voltages.

However, all the other chosen mosfets Vgs at around 2-2.5v are fine.

Ah ok. I will request the stuff soon.

anyway, I can in the meantime (next 2 weeks so I can get home to have the boards)... I can connect 12v from my switching lab power supply (I DIY it, on my channel) on the 12v input... as well as, about 6.25v for the input of the 5v stage and similar for 3.3v rail. I guess this can be sufficient since I don't have the switchers now. What do you think?

the expected result from this is to see 3.3v and 5v on output rails.

[VEGETA]

I have ordered all the parts from mouser and LCSC.

Only thing remaining is the heatsink which is locally made. And the thermal tape needed which I still don't know which part to get.

[gnuarm]

That's ok.  What you are not addressing is what you expect the cap multiplier to do that the linear regulator won't do.  Have you built anything that you found to not be suitable?  You talk about having the cap multiplier in contrast to a switcher, but that's not the question I have.  What do you think a cap multiplier will do that a linear regulator won't?

I found a schematic using the LM317 in place of the capacitor.  So you have dropped the idea of a capacitor multiplier and are using an NPN to boost the regulator output.  Fine, I hope someone has advised you about this potentially affecting the stability of the LM317.  So be aware of that.  I just wanted to dispel the idea that a capacitance multiplier is better at removing low frequency (audible) ripple than a properly designed linear regulator.  What you have now IS a linear regulator.  These can be found to handle your load current, so you could do the same thing in a single package without the NPN.

check the schematic in here: https://www.eevblog.com/forum/beginners/small-mosfet-for-capacitance-multiplier-in-a-small-psu/msg3331484/#msg3331484

it is the final one, but a bit more finalized in kicad.

Here is my little explanation of this:

I feed 12v from a cheap or not-so-good generic switching supply to this board. The board must filter it properly to deliver as clean as possible 5v and 3.3v (amperage different between them).

12v is taken through a 100R resistor with lots of caps (1000uF + 22uF +100nF) into the 7805 linear regulator which outputs 5v and 3.3v via resistor divider. This "reference" signal is very clean since it was taken from a resistor divider which is the cap multiplier part of this design. the cap multiplier here is taken from V_filtered which is about 10v due to the dropout... it is the signal fed to the supply of the op-amp and then will be delivered to the gate of the mosfets. We kept changing it until we reached this formula.

Getting all of that into the linear 7805 is very beneficial to attenuate low frequency noise as you stated which is a further plus along with high capacitance + inductance which takes care of high frequency noise.

the rest is straightforward, the op-amp is fed with 3.3v and 5v from the reference clean signal and outputs the necessary to make the linear regulator mosfet regulates.

the trick here is the cap multiplier at the op-amp stage.

So is this a linear regulator? yes of course, but it is powered or enhanced by capacitance multiplier + switching pre-regulator.

If you are gonna design such a thing, what are you going to do? what parts to use? me and Brian took this project to the maximum by keep changing it and enhancing it... till we wished it end xD

I see your schematic and I don't think you understand what a capacitor multiplier is.  The linear part of the circuit is just a home brewed linear regulator.  There is no capacitor multiplier.  What you are calling a capacitor multiplier is just a linear feedback control circuit with a capacitor on an internal node... that may well make the circuit unstable and oscillate.  I don't know this cap will cause instability, that would require an analysis of the circuit.  It would be as good to pick a good, low noise linear regulator and use it.  Your circuit is fine, but you need to understand it and design it properly.  Rather than using a regulator as your voltage reference you can get better results with a low noise voltage reference. 

BTW, R19 should be 6.7k to match the impedance of the V_3.3 voltage divider to minimize offset voltage.

[VEGETA]

I see your schematic and I don't think you understand what a capacitor multiplier is.  The linear part of the circuit is just a home brewed linear regulator.  There is no capacitor multiplier.  What you are calling a capacitor multiplier is just a linear feedback control circuit with a capacitor on an internal node... that may well make the circuit unstable and oscillate.  I don't know this cap will cause instability, that would require an analysis of the circuit.  It would be as good to pick a good, low noise linear regulator and use it.  Your circuit is fine, but you need to understand it and design it properly.  Rather than using a regulator as your voltage reference you can get better results with a low noise voltage reference.

BTW, R19 should be 6.7k to match the impedance of the V_3.3 voltage divider to minimize offset voltage.

I know the linear part of the circuit is a linear regulator, that is what we wanted to do.

at first we went with a traditional capacitance multiplier circuit, then we kept changing it until it became like this. I don't remember why and when it happened since that was a long time ago.

I think the effect of cap multiplier is still there, not as textbook circuit though. instead of connecting the R+C circuit of the cap multiplier circuit directly into the gate of the mosfet, we put a linear regulator just before that to further aid in low noise and stability and perhaps the price since I was using 2 linear regulators instead. it will also isolate the high freq noise due to the big emi filter before the 7805. this means all of that ripple and noise will never reach the linear part to begin with, so the linear part has to deal with low freq. noise.

thus the mosfets needed more dropout room due to this effect which is why I bumped it up a bit and it is therefore not a low dropout regulator = need a heatsink.

maybe "capacitance multiplier" alone is not good enough to describe our circuit. but still provide the same or better results


maybe Brian can put his opinion on this too.

but why you think this circuit will not work?

[BrianHG]

Vegeta's title: Small MOSFET for capacitance multiplier in a small PSU was there because he wanted to get rid of the ripple from a switching supply.

The ripple on the output of a switching supply begins at 200KHz with signals in the 2MHz switching range depending on load.

Linear regulators pass these signals right through as their PSRR is designed to have a high rejection down below 1KHz.  In fact, some linear regulators enhance the ripple to the output.

Using a basic BJT capacitance multiplier can work since the emitter-follower configuration can reject high frequency modulation on the power being fed into the collector, however, the circuit looses output regulation as the base has nothing more than an RC filter with a load.

My original recommendation was to use a brand new ultra-low ripple, ultra low EMI switcher IC made by TI all on it's own, however, it's availability was only by direct order from TI.

     ---Vegeta needed MORE POWER---

The choice of using an opamp & 78L05 as a reference was price and availability.  Powering those from 12v through a 100ohm, 1000uf RC filter and a load current on them of ~2ma was chosen so that a complete cut and replacement of power within 1/10th of a second wouldn't nudge the output let alone a 200KHz to 2MHz signal making it through that first series 100uh to 1000uf, the even more 100ohm, 1000uf+10uf+0.1uf cap to GND RC filter stage making sure even if the switchers injected noise into the +12 could not get through that initial filter stage.

The mosfets are wired in a source follower configuration with the drain target voltage at 1v above the desired output since the switcher's output ripple according to the data sheet can dip ~0.5v under transient loads.  Having the op-amp powered close to 12v means the driving of the gate voltage will always be high enough to allow the source to drive the desired voltage out.  The only transmittance between the switcher and the linear output side is the internal capacitance between drain & gate & source.  This circuit is no longer a capacitance multiplier though tying the gate to a reference voltage in a source follower configuration where that gate does not have any noise coming from the switcher is keeping close to the theme, except now, with an additional negative feedback & reference, we can assure a degree of precision voltage on the output even with a wide shift in current load.

[gnuarm]

I know the linear part of the circuit is a linear regulator, that is what we wanted to do.

at first we went with a traditional capacitance multiplier circuit, then we kept changing it until it became like this. I don't remember why and when it happened since that was a long time ago.

What I am trying to tell you is that this circuit is no longer a capacitive multiplier... period!  So you can forget that term.  It is just a linear regulator following a switcher.  The linear regulator has frequency limitations because of the opamp and the capacitor you have on the gate imposes more frequency limitations exacerbating the problem.  The only high frequency filtering you will see is due to the capacitors on switcher outputs and the linear outputs.  C38 and C39 are on the control loop in a way that is detrimental to performance. 

I think the effect of cap multiplier is still there, not as textbook circuit though. instead of connecting the R+C circuit of the cap multiplier circuit directly into the gate of the mosfet, we put a linear regulator just before that to further aid in low noise and stability and perhaps the price since I was using 2 linear regulators instead. it will also isolate the high freq noise due to the big emi filter before the 7805. this means all of that ripple and noise will never reach the linear part to begin with, so the linear part has to deal with low freq. noise.

What "big EMI filter"  I see a couple of 22uF caps.  That's a good start, but it won't reduce the EMI so much.  Looking at your PCB artwork it looks like you are running the ground and power on traces rather than planes.  These traces seem to be forming long loops which create EMI.  Switching regulators require careful attention to detail.  Get a data sheet for a Linear Technology part and read every detail.  The best way to eliminate noise is to not generate it.

thus the mosfets needed more dropout room due to this effect which is why I bumped it up a bit and it is therefore not a low dropout regulator = need a heatsink.

maybe "capacitance multiplier" alone is not good enough to describe our circuit. but still provide the same or better results


maybe Brian can put his opinion on this too.

but why you think this circuit will not work?

I am very worried about the impact of adding a capacitor to the control loop in your linear regulator.  As I've said several times, you will get a much better result if you just drop the idea of designing your own linear regulator and pick a chip that has a low noise specification.  I can't think why you feel professionally design voltage regulators won't do a good job. 

I don't mean to be rude, but you don't actually understand how to design a linear regulator.  The capacitor you think is helping to filter the output is actually preventing the regulator from working to remove noise.  The only thing in your favor is that the corner frequency of the RC is about 1.6 MHz.   So lower frequency noise will still be regulated.  Since this is above the LM358 bandwidth it may not actually be harming anything, but it clearly isn't helping anything as you seem to think it does. 

Consider how the control loop works.  If the output droops a bit it feeds back to the inverting input of the op amp.  That pushes the op amp output up a bit which drives the gate higher which pulls the FET source (the circuit output) up to the proper level again.  That's good.  Add the RC filter between the op amp and the gate and you get a delay between the op amp responding and the FET being driven higher.  This allows the output more time to drift.  Not good.  In the worst case the delay is long enough that the op amp has had time to drive a significantly larger amount higher on its output.  Then when the the circuit output finally changes and drives the op amp back it has now over driven the FET and the op amp output is further delayed in correcting that... producing an oscillation.  It is not unusual for designers to take special precautions on high speed circuits to minimize pF of stray capacitance so it doesn't cause oscillations.  They certainly do not add any capacitance.

If you want to produce a good power supply design, focus on implementing a good layout and add a section of LC filtering to the output of the switcher.  That will cut the high frequency noise significantly.  There should not be measurable low frequency noise after a proper linear regulator.  Also read the linear regulator data sheet carefully as they often have specifications on the input and output capacitors to prevent oscillation. 

[gnuarm]

The mosfets are wired in a source follower configuration with the drain target voltage at 1v above the desired output since the switcher's output ripple according to the data sheet can dip ~0.5v under transient loads.  Having the op-amp powered close to 12v means the driving of the gate voltage will always be high enough to allow the source to drive the desired voltage out.  The only transmittance between the switcher and the linear output side is the internal capacitance between drain & gate & source.  This circuit is no longer a capacitance multiplier though tying the gate to a reference voltage in a source follower configuration where that gate does not have any noise coming from the switcher is keeping close to the theme, except now, with an additional negative feedback & reference, we can assure a degree of precision voltage on the output even with a wide shift in current load.

Yes, I was able to sus out most of that from reading the schematic.  I didn't at first understand what you were saying about the input filtering, but now I get that you are isolating the noise from the op amp supplies.  They don't show up in the schematic I have, but I can see that would be a concern. 

The linear regulator is a basic configuration that is available in chip form, separate power sources for the internal circuits and the regulated output.  This is typically so the control circuit can work with a higher voltage to manage the control circuitry and use a lower voltage for the power path to minimize losses just as you are doing. 

You mention that the switching regulator creates noise starting at 200 kHz and that somehow the use of a FET will help to stop that from reaching the output.  That is the same circuit as used in many commercial regulators.  I'm not clear on how your circuit will improve on those designs.  The internal FET is just a FET.  It is the internal op amp that limits the performance frequency range, same as in your design.  Adding the capacitor can be detrimental to the operation of the feedback circuit.  Are you suggesting that it acts as a voltage divider for the high frequency noise that couples from the drain to the gate? 

Have you looked at any of the chip regulators that separate the two power connections for different voltages?  Do you have any expectations on how this circuit will work compared to those?  How do you expect to keep the control loop stable?

[VEGETA]

What I am trying to tell you is that this circuit is no longer a capacitive multiplier... period!

correct, but we can't change the name of the thread. also, we started wanting this to be a cap multiplier, then kept slowly changing it until it became like this. now we cannot change the name of the thread so there is that.

The linear regulator has frequency limitations because of the opamp and the capacitor you have on the gate imposes more frequency limitations exacerbating the problem.

the small capacitor (1nf) on mosfet gate plus the resistor is to ensure stability, especially the resistor.

What "big EMI filter"  I see a couple of 22uF caps.

by that I meant the filtering on 12v rail + the filtering before 7805 (op-amp supply). add to them the filtering before each switcher ic.

22uf is on op-amp positive input.

Looking at your PCB artwork it looks like you are running the ground and power on traces rather than planes.

maybe that is not the final layout since we stalled for months until we settled on on, literally tens of times we changed it. I can send the project to you on private if you want to see the final stuff.

I am very worried about the impact of adding a capacitor to the control loop in your linear regulator.

the 1nf one?

As I've said several times, you will get a much better result if you just drop the idea of designing your own linear regulator and pick a chip that has a low noise specification.

I already manufactured the boards and should be ready to be finalized soon when the rest of components arrive.

I actually started doing just that and I think I still have the schematic. it was like this: Switching regulator -> LC filter -> cap multiplier -> linear regulator with caps on the output. However the cap multiplier part got us where we are now until we changed the entire thing.

Consider how the control loop works.

well, we didn't design this to be like a bench power supply or anything with vastly changing loads and so on. the load pretty much won't change as it is a retro gaming console with only one thing to do continuously.

Add the RC filter between the op amp and the gate and you get a delay between the op amp responding and the FET being driven higher.

adding that 100R is necessary for op-amp and mosfet stability, and the mosfet gate itself is capacitive. we only put 1nF to ensure more stability as I used to suffer from such particular circuits oscillating due to no cap on that place.... but that is on a linear bench psu on ltspice.

you suspect that it could output the wrong voltage if a fast change happens?

if this is the only issue in the design and it caused a problem, i can simply remove that 1nF from the board very easily.

If you want to produce a good power supply design, focus on implementing a good layout and add a section of LC filtering to the output of the switcher.  That will cut the high frequency noise significantly.  There should not be measurable low frequency noise after a proper linear regulator.  Also read the linear regulator data sheet carefully as they often have specifications on the input and output capacitors to prevent oscillation.

trust me, we spent months on layout from datasheets and app notes... down to mm positioning of traces.

by "proper linear regulator" you mean those expensive ICs? I could buy 3~5$ one from Analog Devices but then it will cost a lot per board unlike the opamp+mosfet which is less than 0.5$ combined. I would need 2 of them for 2 rails which will be like 6-10$ per board just for 2 ICs.

I've been searching of ready-to-use switching and linear ICs for very low noise but they are all expensive as hell. I wanted to manufacture this on JLCPCB SMT service as much as possible too.

what output noise and ripple do you expect from this? how many micro\milli volts?
do you think it may not work at all? I paid a lot for these and will be disastrous if they didn't work.

[gnuarm]

What I am trying to tell you is that this circuit is no longer a capacitive multiplier... period!

correct, but we can't change the name of the thread. also, we started wanting this to be a cap multiplier, then kept slowly changing it until it became like this. now we cannot change the name of the thread so there is that.

I am responding to your repeated use of the term capacitor multiplier in conversation, not the title of the thread.  There is no capacitive multiplication going on in this circuit.

The linear regulator has frequency limitations because of the opamp and the capacitor you have on the gate imposes more frequency limitations exacerbating the problem.

the small capacitor (1nf) on mosfet gate plus the resistor is to ensure stability, especially the resistor.

If you had given any other reason that might have flown.  But "stability" is the one thing I can guarantee that cap will not provide.  It has the opposite effect in fact.  As I have observed, the corner frequency of the RC is above the unity gain of the op amp, so it may not add to the instability of the circuit, but it is assured to not improve stability.

What "big EMI filter"  I see a couple of 22uF caps.

by that I meant the filtering on 12v rail + the filtering before 7805 (op-amp supply). add to them the filtering before each switcher ic.

Yeah, the schematic I see does not identify a 12V rail, but I assume you are talking about the one labeled Vcc?  You might do better using a single filter on the input to the rest of the circuitry rather than two circuits on the two inputs to the two regulators. 

22uf is on op-amp positive input.

And on the output of the switchers on the schematic I see.

Looking at your PCB artwork it looks like you are running the ground and power on traces rather than planes.

maybe that is not the final layout since we stalled for months until we settled on on, literally tens of times we changed it. I can send the project to you on private if you want to see the final stuff.

No, that's ok.  I'm already involved in a project that is not going well.  A team is designing a ventilator and the coordination is very poor.  Rather than having set requirements we just work on stuff until we think something works.  It is going so badly I might leave.  I design a circuit and it gets tossed out.  I redesign it and it gets tossed out.  All because of poor planning. 

I'm not looking to pick up any more projects.

I am very worried about the impact of adding a capacitor to the control loop in your linear regulator.

the 1nf one?

Yes

As I've said several times, you will get a much better result if you just drop the idea of designing your own linear regulator and pick a chip that has a low noise specification.

I already manufactured the boards and should be ready to be finalized soon when the rest of components arrive.

I actually started doing just that and I think I still have the schematic. it was like this: Switching regulator -> LC filter -> cap multiplier -> linear regulator with caps on the output. However the cap multiplier part got us where we are now until we changed the entire thing.

Yeah, that might actually work a little bit better than the circuit you have now as the cap multiplier will have a higher frequency than the linear.  But the higher frequencies are responsive to filtering.  So it is not important to use an active device to remove the switching noise.

Consider how the control loop works.

well, we didn't design this to be like a bench power supply or anything with vastly changing loads and so on. the load pretty much won't change as it is a retro gaming console with only one thing to do continuously.

Add the RC filter between the op amp and the gate and you get a delay between the op amp responding and the FET being driven higher.

adding that 100R is necessary for op-amp and mosfet stability, and the mosfet gate itself is capacitive. we only put 1nF to ensure more stability as I used to suffer from such particular circuits oscillating due to no cap on that place.... but that is on a linear bench psu on ltspice.

you suspect that it could output the wrong voltage if a fast change happens?

if this is the only issue in the design and it caused a problem, i can simply remove that 1nF from the board very easily.

If you want to produce a good power supply design, focus on implementing a good layout and add a section of LC filtering to the output of the switcher.  That will cut the high frequency noise significantly.  There should not be measurable low frequency noise after a proper linear regulator.  Also read the linear regulator data sheet carefully as they often have specifications on the input and output capacitors to prevent oscillation.

trust me, we spent months on layout from datasheets and app notes... down to mm positioning of traces.

by "proper linear regulator" you mean those expensive ICs? I could buy 3~5$ one from Analog Devices but then it will cost a lot per board unlike the opamp+mosfet which is less than 0.5$ combined. I would need 2 of them for 2 rails which will be like 6-10$ per board just for 2 ICs.

Aren't you using chips for the switcher?  I don't typically use $5 regulators.  They can be found as cheap as $0.50.  Units with a split voltage are a subset, so you would need to check to see what they cost. 

I've been searching of ready-to-use switching and linear ICs for very low noise but they are all expensive as hell. I wanted to manufacture this on JLCPCB SMT service as much as possible too.

I don't know what your total costs are, but if the devices are "expensive as hell" then I suppose you should not use them.  But you should do a stability analysis on your circuit to see if it will work.

what output noise and ripple do you expect from this? how many micro\milli volts?
do you think it may not work at all? I paid a lot for these and will be disastrous if they didn't work.

I don't have a clue.  I would have simulated the design if I had specific design goals for noise and stability.  Have you heard of LTspice?  What is the expensive part?  I thought the rest of the design was not expensive???   If you have a model for the FET in your circuit you can model the effect of the capacitor on the gate. 

[BrianHG]

I am very worried about the impact of adding a capacitor to the control loop in your linear regulator.

the 1nf one?

Yes

This is a tune-able point.  I would prefer no cap and lowering the series resistor value below 100 ohms.  However, I only recently found out Vegeta doesn't have a scope yet.  For now, I would say 'DO NOT INSERT' the C38 & C39 1nf caps.

Can you suggest a linear regulator which rejects noise on the power source at 20KHz through 2MHz?
Also, has a 0.5v dropout with a 3.5 amps load continuous.
And costs less than 30 cents each?

Other part values on the PCB was Vegeta's attempt at resistor part value consolidation.  I could not convince him to use real values and single resistors at single points.

[VEGETA]

Yeah, the schematic I see does not identify a 12V rail, but I assume you are talking about the one labeled Vcc?  You might do better using a single filter on the input to the rest of the circuitry rather than two circuits on the two inputs to the two regulators. 

I sent the pdf to you on private. Yes vcc is 12v.

Aren't you using chips for the switcher?  I don't typically use $5 regulators.  They can be found as cheap as $0.50.  Units with a split voltage are a subset, so you would need to check to see what they cost. 

switchers are basic chips here not special. what do you suggest for linear regulator that is cheap? assuming low ripple low noise design that we went with,

Have you heard of LTspice? 

yes and I even made video about it, but I am not too skilled with it especially making a custom model.

I didn't think this circuit needs much simulation since it is kinda basic. load is nearly constant with no special requirements.

This is a tune-able point.  I would prefer no cap and lowering the series resistor value to 100 ohms or less.  However, I only recently found out Vegeta doesn't have a scope yet.

My concern is that the design works or not. if removing the cap is what it takes, then this is easy to do. if i could sell this product then I can afford getting the scope I need. this is the plan

Other part values on the PCB was Vegeta's attempt at resistor part value consolidation.  I could not convince him to use real values and single resistors at single points.

real values?

resistor consolidation doesn't affect the circuit. what resistor value do you think affects the circuit in wrong way.?

[gnuarm]

I am very worried about the impact of adding a capacitor to the control loop in your linear regulator.

the 1nf one?

Yes

This is a tune-able point.  I would prefer no cap and lowering the series resistor value below 100 ohms.  However, I only recently found out Vegeta doesn't have a scope yet.  For now, I would say 'DO NOT INSERT' the C38 & C39 1nf caps.

Can you suggest a linear regulator which rejects noise on the power source at 20KHz through 2MHz?
Also, has a 0.5v dropout with a 3.5 amps load continuous.
And costs less than 30 cents each?

Other part values on the PCB was Vegeta's attempt at resistor part value consolidation.  I could not convince him to use real values and single resistors at single points.

I can't give you a part number.  But upwards of 100 KHz it is not hard to filter noise with passives.  Most linear regulators only start to roll off in rejection towards 100 kHz.  Since there is no spec on the goal there is no way to pick one over another or even to design a passive filter. 

I got into this because Vegeta kept talking about the capacitor multiplier which is not even in the circuit anymore.  The linear regulator shown here has unity gain at 1 MHz, so is not usefully attenuating input noise until the same frequencies as most linear regulators.  If you want to get better regulation at higher frequencies a high frequency op amp is required.  It will also require careful design for stability.  I believe this is greatly facilitated using LTspice, but the learning curve is steep. 

It is an interesting idea to mitigate the gate-drain and gate-source capacitance by adding capacitance from the gate to ground.  But that way lies madness with the tradeoff as it impacts the feedback control loop making it ring or potentially oscillate.  Maybe you can find a happy medium where it helps. 

[BrianHG]

This is a tune-able point.  I would prefer no cap and lowering the series resistor value to 100 ohms or less.  However, I only recently found out Vegeta doesn't have a scope yet.

My concern is that the design works or not. if removing the cap is what it takes, then this is easy to do. if i could sell this product then I can afford getting the scope I need. this is the plan

Other part values on the PCB was Vegeta's attempt at resistor part value consolidation.  I could not convince him to use real values and single resistors at single points.

real values?

resistor consolidation doesn't affect the circuit. what resistor value do you think affects the circuit in wrong way.?

Best not to insert the cap.  The component was placed there in case you needed to add one, you wouldn't have to do a hand-wire job.  The lower the capacitance, the faster the response from the opamp.

As for the resistors, as an example, you place a numerous 10k in series to avoid placing a single 20k.  This took up PCB area, but somehow, we still managed to fit everything.

[VEGETA]

Best not to insert the cap.  The component was placed there in case you needed to add one, you wouldn't have to do a hand-wire job.  The lower the capacitance, the faster the response from the opamp.

the cap is already soldered on the board during assembly as you know. I could remove it if it affects performance, by hand.

As for the resistors, as an example, you place a numerous 10k in series to avoid placing a single 20k.  This took up PCB area, but somehow, we still managed to fit everything.

I admit I am a bit stiff for consolidating parts especially resistors 

However, board size is the only problem in this, and as you said, we managed to fit it.

The linear regulator shown here has unity gain at 1 MHz, so is not usefully attenuating input noise until the same frequencies as most linear regulators.

how did you know its unity gain at 1 MHZ? from op-amp?

you asked about design goals, here is what I have: low noise low ripple to never affect analog video and audio by any means. the problem is analog video\audio signals are about 1v p-p so any noise and ripple will affect them. so the lower noise\ripple the better.

on a side note: i managed to edit the topic title!! now gnuarm can rest xD. However, website thread link didn't change. It is ok, since no one looks at it but rather look at thread name.

[VEGETA]

on another side note, I found an analog oscilloscope locally: https://jo.opensooq.com/ar/search/139119964/

it is Texio cs-4125a, looks like a Kenwood. 20Mhz dual channel with very clean and nice condition. However, I read that its sensitivity is 1mV so would it be able to measure noise and ripple beyond 1mV? is it gonna benefit me here?

Price is 100 JOD (currently negotiating) which is about 140$. Do you think it is worth it?

[BrianHG]

on another side note, I found an analog oscilloscope locally: https://jo.opensooq.com/ar/search/139119964/

it is Texio cs-4125a, looks like a Kenwood. 20Mhz dual channel with very clean and nice condition. However, I read that its sensitivity is 1mV so would it be able to measure noise and ripple beyond 1mV? is it gonna benefit me here?

Price is 100 JOD (currently negotiating) which is about 140$. Do you think it is worth it?

It looks nice, but 140$us for a dual channel 20MHz CRO...
I do not think 1mv of noise on the power line can have any effect on the Dreamcast's output.  Remember, the worst case noise coming out of the AOZ switcher you are using is in the neighborhood of 400mv pulse 100us wide (~10-50Khz MHz region) (see bottom left hand graph on page 6 of AOZ data sheet.) during a 2 amp swing in output load.  (We engineered this module to deal with this worst case pulse which can happen on the 3.3v bus during a change in CPU processing load.)  A scope with 1mv sensitivity is 400 times more sensitive than that.

In the video signal, if it were a ~0.7v signal, 1mv would represent ~1/700th a luminance level.  Or, greater than 256 shades of grey which is what you have with a 24bit color image.  That is if 100% of your supply voltage noise entered into the video output signal, which it does not since the video output isn't the full 5 or 3.3 volts.  Even our circuit as well as the original Dreamcast power supply has more noise than 1mv on the output unless you slap a 10 farad capacitor on the output.

Hunt a little more before commiting.  Search and ask around here on eeveblog.
In your situation, I would offer $50 US.

[BrianHG]

https://www.amazon.com/OWON-Oscilloscope-2-Channel-Oscilloscopes-Bandwidth/dp/B07VPNFVMN/ref=sr_1_1

[VEGETA]

well, 400mV seems like a lot is not it? what about normal operation at 2.8A of 3.3v, what ripple and noise is expected?

I plan to get DS1054Z for sure, but right now I cannot do it because it is gonna take a lot of money. Shipping is not cheap and itself is not cheap. thus I saw locally offered items by chance and found this one. Still negotiating with him though.

So I could get this one, then wait and get the rigol one.

we'll see what happens

EDIT:

on AOZ page 6, full load operation continuous load seems to have very little ripple at 50mV\div. the graph is not so clear but that seems like about 5mV maximum, this is without linear stage or filters.

[VEGETA]

I've been searching for better parts in case I wanted to make v2 of this board, and other project I have in mind. I found this:

MIC94325: https://www.microchip.com/wwwproducts/en/MIC94325
it is very low power, not suitable for this application but certainly for my other project (RGB to component\composite\s-video converter). It claims "Active noise rejection over a wide frequency band ...    >50dB from 10Hz to 5MHz at 500mA load" by using their ripple blocker technology. so powering it from USB port 5v then stepping it down to 3.6 via simple buck converter delivers this performance. dunno if this is correct or not but its specs seems great.

the other interesting part is this:

TPS62913: https://www.ti.com/product/TPS62913

low voltage output but high current, suitable to this project. still very new part with only samples are available now. it claims this:

•  Low output 1/f noise < 20 μVRMS (100 Hz to 100kHz)
•  Low output voltage ripple < 10 μVRMS after ferritebead
•  High PSRR of > 65 dB (up to 100 kHz)

cuz it uses a ferrite beat or small inductor in its control loop but says 100hz to 100khz low noise while itself working on 1mhz or 2.2mhz range.

what do you think?

on a side note, I could make a version 1.1 of this board which includes removing op-amp 1nf + adding pi passive filters (1nf+100nf then 2.2uH then 1nf+100nF) after switchers and 1nf+100nf after linear regulators. I made another PCB with those made and wired them. will it help much?

[BrianHG]

I though 3 amp isn't enough.

You still need that Murata ferrite bead, you should probably use their same power inductor and you will need to match their second PCB layout exactly to achieve their specs.

Notice they gone through charting the high frequency ripple but make no mention of some of the basics, like transient load response.

[VEGETA]

I though 3 amp isn't enough.

You still need that Murata ferrite bead, you should probably use their same power inductor and you will need to match their second PCB layout exactly to achieve their specs.

Notice they gone through charting the high frequency ripple but make no mention of some of the basics, like transient load response.

3 amps is full load yes, did I say anything else? TPS62913 supports 3 amps as I see. I am not talking about that MIC part, that is just to see your opinion on it as a regulator for small load project.

is there any other switcher that could deliver very low noise and ripple that I can play with later on? something that can deliver < 1mV p-p ripple and noise with or without a simple linear reg.

I offered 40JDs (57$) for that analog scope, could go to 50JD (71$) tops. I will get it for sure, golden chance for me really.

I've sent v1.1 schematic to you, not final but yeah .. did it in 15 mins or so. gonna start working on it once i verify the boards... gonna take time