Low ripple low noise small power supply design

[VEGETA]

I am using IRLIZ44N mosfets.

right now I have 2 boards, one with op-amp and 2 mosfets installed which is used on tests mentioned previously. and another one without opamp which I removed.

I will use the no opamp board for wiper test and the board with opamp for testing caps on source and gate.


as for your previous question on requirements, i wanted low noise low ripple psu for Sega Dreamcast. I would be happy with less than 10mV total ripple+noise. much happier with lower than 1mV but I guess it is kinda hard with my current skills.... until I found about TPS62913 but this IC still in pre-production.

So I am trying my best to make current circuit and boards work. worst situation is that I would redesign the board from scratch. then I would be using different and more straightforward and easy approach, like using those TI switchers I mentioned in previous post here.

[VEGETA]

I have tested putting 100nf then 1nf on source to gate... didn't work. sound increased and output voltage increased to 5.8v and 4.3 volts.

HOWEVER, I put 1000uF caps directly on 3.3v and 5v rails on the output connector... and it worked. sound disappeared and putting the massive load of 1.6 regulated well -> 3.3v on 3.3v rail and about 4.950v on 5v rail. It was increasing but I took it off due to heat.

same test on 3.6 ohms load resulted in 3.340v and 5.004v. on both trials... 0 sound happened. I could hear my air conditioner out of the sudden silence that happened.

any explanation of this?

EDIT:

I tried up to 100uF and it works, however 3.3 and 6.8uf didn't work.


no load test:

3.3v: 3.360v
5v: 5.042v
both 3.3v opamp inputs are 3.360 and 3.361.


1.6 ohm:
3.3v: 3.263 (increasing slowly, disconnected due to heat)
5v:4.925 (decreasing slowly, disconnected due to heat)
3.3v opamp ref: 3.325
3.3v opamp fb: 3.325
5v opamp ref: 5.020 or 5.019
5v opamp fb: 5.020

[gnuarm]

I think this indicates the previous readings were disrupted by a large AC component.  The circuit has been working all along, but the oscillations corrupted the measurements.  The gate voltage was probably messed up because of the rectification by the input diode.  Once you get your oscilloscope you will probably be able to see large excursions on the output of the op amp when you remove the cap.

Reminds me of one of the early researchers in liquefying gasses had been running the device for some time but could not see anything in the collection flask.  Finally he tried shining a light into the flask and found it was half full.  The optical density of the liquid was so close to the gas that the meniscus was hard to see! 

[gnuarm]

I'm actually very surprised you don't have a large cap on the output anyway.  Caps are the first and easiest way to clean up ripple after all.  The schematic shows 22 uF on the output.  Is that ceramic?  I would add a 470 uF electrolytic cap. 

[VEGETA]

I'm actually very surprised you don't have a large cap on the output anyway.  Caps are the first and easiest way to clean up ripple after all.  The schematic shows 22 uF on the output.  Is that ceramic?  I would add a 470 uF electrolytic cap.

well, you are correct. I assumed it would be enough to have 22uf ceramic despite knowing the elec. cap is easy to add and good to circuit.

maybe now I can buy elec. caps and solder them in a neat way on bottom side to make the circuit work and the product as intended. next version, I will try adding it in schematic and PCB layout near the pin. 100uf was enough but 470uf is better while having the same smt footprint.

so, can you figure out the oscillations source? I am annoying the scope guys with phone calls daily now...!

[BrianHG]

so, can you figure out the oscillations source? I am annoying the scope guys with phone calls daily now...!

It must be the opamp feeding the mosfet with it's hard turn-on 'knee', overshooting the output voltage, then relaxing down below the set output voltage.  The buzz was probably the 22uf ceramic output caps as that hard turn-on knee will make an audible sound do to the microphonic effect ceramic caps have.  The 1000uf cap prevents that turn-on & turn-off effect allowing the opamp to settle on a fixed voltage.  Without the cap, increasing the output load also has a similar effect as the output wont shoot past the set voltage as easily as a 33ma load being driven by a 50amp mosfet.

Though the gate voltage will change with mosfet temperature, the output should stay stable unless the opamp & 78L05 are also warming up.  However, this voltage change should only be a few 10s of mv, maybe 0.05v with a cheap opamp.  This output voltage drift can be removed with a high quality, or even temperature compensated opamp.

[gnuarm]

Continuing oscillations (as opposed to a temporary ringing) happen for one reason, the Barkhausen criteria is met.  That criteria is the loop gain is 1 and the phase change is 0.  Your circuit has an inversion at the op amp input (180 degrees), so another 180 degrees is being shifted in the RC of the capacitance.  Each RC circuit can shift up to 90 degrees.  Then a bit of delay in the amp adds the rest. 

Is C39 removed?  That part has lots of potential for causing oscillations.

I'm not sure how adding capacitance to the output fixes this, but regulators often have requirements for minimum and maximum output capacitance.  They also indicate a minimum ESR.  I'm no expert on control loop stability, so I can't explain the details of why your circuit gets so emotional.  That's also why I use off the shelf regulators. 

You might try a small cap across R19.  I often see this in regulator circuits for stability purposes.  Try your 1 nF cap there or even smaller, 100 pF is not too small.  This capacitor shifts the phase in the other direction and also improves the response time of the feedback loop.

[VEGETA]

Continuing oscillations (as opposed to a temporary ringing) happen for one reason, the Barkhausen criteria is met.  That criteria is the loop gain is 1 and the phase change is 0.  Your circuit has an inversion at the op amp input (180 degrees), so another 180 degrees is being shifted in the RC of the capacitance.  Each RC circuit can shift up to 90 degrees.  Then a bit of delay in the amp adds the rest. 

Is C39 removed?  That part has lots of potential for causing oscillations.

I'm not sure how adding capacitance to the output fixes this, but regulators often have requirements for minimum and maximum output capacitance.  They also indicate a minimum ESR.  I'm no expert on control loop stability, so I can't explain the details of why your circuit gets so emotional.  That's also why I use off the shelf regulators. 

You might try a small cap across R19.  I often see this in regulator circuits for stability purposes.  Try your 1 nF cap there or even smaller, 100 pF is not too small.  This capacitor shifts the phase in the other direction and also improves the response time of the feedback loop.

C39 and C38 are removed. even when removed, without elec. caps, the circuit still have the same behavior. adding the elec. caps was the answer.

now I need to figure how to add them to the boards in a neat way. 

also, to modify the PCB and add them as smt parts on bottom side. I guess there is some room near the mosfets and along the way to output connector. putting small elec caps can be done there... 100uF is guaranteed but gonna search lcsc for bigger value caps with small sizes.


I will try adding a small 1nf cap around r19 to see if i can get away with it and without elec caps.

[VEGETA]

I was searching for high quality switchers do to another version, as a side project and found these:

https://www.ti.com/product/TPS62913 -> promises noise\ripple in the uV range. still pre-production, TI told me release date is still not confirmed and fat. approx price is 1.8-2.5$ @ 100 quantity. ~1.65$ @ 250 quantity.

https://www.ti.com/product/LM61460 -> active part. output noise\ripple is <5 mV for FPWM mode, and about 10 mV for auto (if I read the graphs properly). Price is 3.3$ @ 100 quantity, <3$ @ 250 quantity.

https://www.analog.com/en/products/LT8652S.html -> active and new part with Silent Switcher 2 technology. Promises 10 mV ripple\noise at full load without putting extra caps\inductors. It has 2 outputs so it can be used as one chip solution for 3.3v and 5v. Price is 5.8-6.5$ @ 100 quantity. So maybe 6$ for one device plus passives and output connectors... probably cheaper than multi part solution? I kinda like this solution if I could enhance its output to lower than 5mV as a side project for now (no 125 boards hhhh).

what do you think of these picks if you were to design it again?

Note: for now, I am sticking with current design and will enhance it by adding the required caps to the boards. this is for not-so-seen future xD

[VEGETA]

I realized I made a stupid choice of getting 3.96mm pitch 2-position connectors instead of the standard for fans of 2.54mm.

I guess this won't be fixable on current boards... damn. more to fix on next boards version...

[gnuarm]

So I spent a little time simulating an idea and here is what I came up with.  This circuit uses transistors in place of the op amp since that is limiting the performance.  Even then, the limitation seems to be in driving the capacitance of the FET gate.  The FET in this simulation is fairly low, 9 nC.  They measure the charge required rather than the actual capacitance since that is the more useful way to view it.  By adding 8 - 0603, 1 uF, low ESR, ceramic caps I got the output ripple down to 7 mVpp with a 100 mVpp ripple at the input (from the switcher).  I think that is pretty good.

To utilize the caps to their best advantage will require good layout.  I also added a 100 uF electrolytic cap, but it does little for the ripple really.  The ESR on that type of cap is large.  0603 is a good size for ceramic caps.  Smaller is better for high frequency, but this is not so high as to require 0402 or smaller caps.  You can find 0805 size with lower ESR which might let you use fewer of them.  Since they are larger it may be a wash in terms of board space. 

The resistors are sized to use 26 mA in the control circuit.  Using less current slows it down for more ripple.  The resistors can be smaller (higher current) for lower ripple, but a point of diminishing returns will be eventually reached. 

I just read about the fan connector.  That can be fixed with an adapter cable easily enough.

[VEGETA]

putting a ceramic 22nf cap around r19 didn't help. also around d3 and d4 (opamp inputs) didn't help. the only way it could be stable is just on the output.

I tested that 100uf is the minimum it works with, I will be  buying lots of them to hand solder them. However, making them fit neatly is tricky.

looks like you didn't check the 3 switchers I posted above.

if I ever get a scope I will test your circuit.


EDIT: I found that 220uF 16v elec. caps are the perfect physical size. they are shorter than 100uF 16v but a bit wider which is not important. I guess 220uF gonna be ok.

[gnuarm]

No, I didn't read the data sheets on the three candidates.  I think you said you had to wait for two of them as they are not in production.  All three would require a board spin.  What would you like me to look at exactly? 

If 100 uF is the minimum that makes the circuit stable then you definitely want to use the 220 uF part.  But you need to test with the parts you will use and get a data sheet to see what the ESR is.  The ESR is likely more important than the value of capacitance. 

There are many form factors for capacitors.  I'm sure you can get both parts in the same height.  What is important to a capacitor is volume.  More capacitance or more voltage means more volume if all else is the same.  There is no reason to let the form factor dictate the electrical value of the capacitor.  What height are you using?  Our project is using some large value caps that are pretty compact, 8.3 mmsq x 10 mm tall, 1,000 uF, 25V.  Since you can use a smaller value and voltage you should be able to find a much smaller package.  Have you used Digikey's search facility?

What is the frequency of the switchers you have designed in?

[gnuarm]

Did you look into the many USB attached digital oscopes available and reviewed on this site?  I think many of them will do the job you need with a price around $100.

[VEGETA]

No, I didn't read the data sheets on the three candidates.  I think you said you had to wait for two of them as they are not in production.  All three would require a board spin.  What would you like me to look at exactly? 

yes that would require board respin as an updated version, let's call it version 2. which is not top priority right now. I wanted your opinion about using these as the main part in v2 of the board which is gonna be without linear stage. will they be good enough to reach <10mV ripple\noise? perhaps even lower than that?

TPS62913 is the only one which is still in pre-production, the others are active. LM61460 seems good but only 1 output at relatively high price, while LT8652S has 2 outputs in one package with price of 6.5$ @ 100 quantity, very pricey but gonna be enough to do everything by itself plus passives. do you know better switchers in terms of output ripple?

If 100 uF is the minimum that makes the circuit stable then you definitely want to use the 220 uF part.  But you need to test with the parts you will use and get a data sheet to see what the ESR is.  The ESR is likely more important than the value of capacitance.

There are many form factors for capacitors.  I'm sure you can get both parts in the same height.  What is important to a capacitor is volume.  More capacitance or more voltage means more volume if all else is the same.  There is no reason to let the form factor dictate the electrical value of the capacitor.  What height are you using?  Our project is using some large value caps that are pretty compact, 8.3 mmsq x 10 mm tall, 1,000 uF, 25V.  Since you can use a smaller value and voltage you should be able to find a much smaller package.  Have you used Digikey's search facility?   

the ones I will get are 220uF 16v elec. caps which are leaded not smt, and from local suppliers. will never get them from outside. I have small quantity here and they are tiny, like 7mm in length which is suitable to the board since I am gonna hand solder them and situate them where it fits.

However, for the enhanced version of this current design (call it v1.1), I will include smt elec. caps on board for sure while removing 1nf opamp to mosfet gate caps. footprint won't be a problem since we need 6.3v rated caps which are generally small. I guess we can even get 470uf with low footprint, gonna get down to do it sometime in near future.

Also, I have included a pi filter (1uH inductor with ceramic caps before and after it) for both switchers output just before linear stage. its placement in layout is not perfect but probably better than before.

What is the frequency of the switchers you have designed in?

we went for 2 MHz.

Did you look into the many USB attached digital oscopes available and reviewed on this site?  I think many of them will do the job you need with a price around $100.

I know some Hantek and Owon ones. getting them to me in Jordan will have shipping costs and customs.

I saw this one on Amazon for quite good price. gonna take 41$ shipping and customs fee. However it doesn't support AC coupling!

this one seems better but it will cost 150$ to get here.

I will wait to see what happens with that DOS1102 used scope for now.

[BrianHG]

yes that would require board respin as an updated version, let's call it version 2. which is not top priority right now. I wanted your opinion about using these as the main part in v2 of the board which is gonna be without linear stage. will they be good enough to reach <10mV ripple\noise? perhaps even lower than that?

TPS62913 is the only one which is still in pre-production, the others are active. LM61460 seems good but only 1 output at relatively high price, while LT8652S has 2 outputs in one package with price of 6.5$ @ 100 quantity, very pricey but gonna be enough to do everything by itself plus passives. do you know better switchers in terms of output ripple?

Take a careful look at the TPS62913 datasheet's example PCB.  The example PCB, figure 10-2 on page 29 with the second LC filter was needed to achieve the super low < 10uV ripple noise.  That PCB is 44mm wide.  If you want to replicate that low noise, you will need to make a PCB at least 44mm plus your final output connectors, say 55mm wide.  Now I am not saying you need a perfect replication, but, if you want what's shown to you in the datasheet, you need that 'void/space' between the switcher and second inductor L, at least 3/4 the width I said.  To prevent power dips during transient current loads, especially on the 3.3v rail, you will want some heavy-duty uF caps on the output as well as the source 12v supplying the system.

Noting that it is a TI part, I believe you can achieve a good clean 3 amps with the part if you use the same primary L1 inductor.

Following the rules, you should expect that the 12v to 3.3v 7mv ripple on figure 6.11 should vanish like what's seen on figure 6.13, though the switcher performs with half the noise when going from 5v to 3.3v.

Going back on this thread, I posted another low EMI switcher from TI like the LM61460, but it was a 4 amp version.  Just another part you might look into.  Make sure you configure either switcher in FPWM mode so that the output stays perfectly regulated even at currents below 500ma at the cost of some efficiency when drawing low current outputs.  Since you aren't making a battery powered device, the FPWM mode is fine.

[BrianHG]

this one seems better but it will cost 150$ to get here.

I will wait to see what happens with that DOS1102 used scope for now.

The USB isolation is a good protective measure, all round, not just for your PC, but the device you are testing.

[VEGETA]

Take a careful look at the TPS62913 datasheet's example PCB.  The example PCB, figure 10-2 on page 29 with the second LC filter was needed to achieve the super low < 10uV ripple noise.  That PCB is 44mm wide.  If you want to replicate that low noise, you will need to make a PCB at least 44mm plus your final output connectors, say 55mm wide.  Now I am not saying you need a perfect replication, but, if you want what's shown to you in the datasheet, you need that 'void/space' between the switcher and second inductor L, at least 3/4 the width I said.  To prevent power dips during transient current loads, especially on the 3.3v rail, you will want some heavy-duty uF caps on the output as well as the source 12v supplying the system.

The 2nd filter is necessary for the performance, it is just a ferrite bead with some caps and they take the feedback after the ferrite bead.

I didn't find where they mention I need that space, but rather the layout shape has to be followed. I would need 2 of that switcher, one for 3.3v and another for 5v, all will be in 50x50mm board since now a lot of components (mosfets + 7805 + opamp) won't be there which makes room.

Following the rules, you should expect that the 12v to 3.3v 7mv ripple on figure 6.11 should vanish like what's seen on figure 6.13, though the switcher performs with half the noise when going from 5v to 3.3v.

yes, the performance is all about the 2nd filter. it achieves <500 uV total ripple and noise which is marvelous. the only hard part about it is the layout which in of it self not so hard given time. remember this is gonna be the only critical component in the board.

Going back on this thread, I posted another low EMI switcher from TI like the LM61460, but it was a 4 amp version.  Just another part you might look into.  Make sure you configure either switcher in FPWM mode so that the output stays perfectly regulated even at currents below 500ma at the cost of some efficiency when drawing low current outputs.  Since you aren't making a battery powered device, the FPWM mode is fine.

you mean LM61440? about 2.27$ @ 250 quantity, so 4.6$ for 2 of them, assuming removing linear stage and other stuff, this is not so expensive. it shows 20mv p-p for 100mA auto mode which is not good but the 4A load is about 3.5 mV auto mode. they don't show the output ripple\noise for fpwm mode though. I assume if we used fpwm mode and added other stages of L-C filters plus big bulk caps we would achieve < 1mV??

[Vovk_Z]

I'm actually very surprised you don't have a large cap on the output anyway.  Caps are the first and easiest way to clean up ripple after all.  The schematic shows 22 uF on the output.  Is that ceramic?  I would add a 470 uF electrolytic cap.

Using quite a large (470 uF) cap at the output to maintain stability - is not a smart way. It is rather the opposite. (I'm sory).

[VEGETA]

I'm actually very surprised you don't have a large cap on the output anyway.  Caps are the first and easiest way to clean up ripple after all.  The schematic shows 22 uF on the output.  Is that ceramic?  I would add a 470 uF electrolytic cap.

Using quite a large (470 uF) cap at the output to maintain stability - is not a smart way. It is rather the opposite. (I'm sory).

well, putting a large cap 100uf or more was the solution to the problem and made the circuit stable. I will use 220uF.

what is you suggestion then?

[VEGETA]

I've been playing around and got some modification to the board:

- 1nF caps of opamp output removed.
- R19 is now 6.8K.
- added elect. caps (220uF) to linear output + 2 more ceramics (100nF + 1nF).
- added LC filter after switchers which are 2.2uF with 100nF + 1nF caps before and after it. I've put them in the layout as good as I could, maybe not perfect.
- adjusted the footprint of J1, J4 and J5 to be 2.54mm since this is the standard which made a lot of room.

here are some pics: https://slow.pics/c/9loK3z3G

[BrianHG]

I've been playing around and got some modification to the board:

- 1nF caps of opamp output removed.
- R19 is now 6.8K.
- added elect. caps (220uF) to linear output + 2 more ceramics (100nF + 1nF).
- added LC filter after switchers which are 2.2uF with 100nF + 1nF caps before and after it. I've put them in the layout as good as I could, maybe not perfect.
- adjusted the footprint of J1, J4 and J5 to be 2.54mm since this is the standard which made a lot of room.

here are some pics: https://slow.pics/c/9loK3z3G

Yes the 220uf are needed, however, the way you had to wire the LC filter due to location and space may make things worse or have no effect.  Notice on the example TI PCB how straight and fat the GND and filter cap traces are for their LC filter and how deliberately separated from the switcher it is positioned.  It is laid out that way on purpose.  I would wait until you can scope the current board to see if the extra LC filter is needed.

As for R19 and R18, I cannot say what the best value should be without scoping the PCB.  They were put in to prevent the input of the opamp from receiving the 'grunt' of a high current power signal trace, not as some sort of gain value.  I would make those 2 resistors as small as possible yet not be too heavy current if the opamp input happens to be diode clamped during power-up or power-down.  I would have used 1k for those 2, but, you were 'part consolidating'...

[VEGETA]

the way you had to wire the LC filter due to location and space may make things worse or have no effect.

i am open to suggestions as i told you they placing is not perfect. however, only the one on top is kinda bad as it is above Vcc trace, the other one is good. for the one on top, i can let the power trace go down a bit to make room for making ground trace as thick.. gonna eat a bit of the mosfets pour but still ok.

I can even relocate all of them to be directly above L5 which makes power and gnd signals short and thick.

moving the other LC filter to the right will make it above 5v and 3.3v rails... not good.

As for R19 and R18,

changing resistor values is easy. I can choose any other value used in the design if it suites, or at worst case, get a new value. no big deal.

[VEGETA]

oscilloscope wise.. getting my dream DS1054Z will cost me 461$ from amazon but without customs which may get 100$ more... while from a proxy service ubuy will cost me totals including customs 421 JOD which is 594$ not so huge but not so little. Isolated Owon will cost 150$. the DOS1102 guy still didn't respond and I am getting frustrated.

The only think available locally is this: https://mikroelectron.com/Product/PICOSCOPE-2204A-MADE-IN-UK/ which is 100 JD (145$ or so) for just 10 Mhz with 10mv per div.

[gnuarm]

I've been playing around and got some modification to the board:

- 1nF caps of opamp output removed.
- R19 is now 6.8K.
- added elect. caps (220uF) to linear output + 2 more ceramics (100nF + 1nF).
- added LC filter after switchers which are 2.2uF with 100nF + 1nF caps before and after it. I've put them in the layout as good as I could, maybe not perfect.
- adjusted the footprint of J1, J4 and J5 to be 2.54mm since this is the standard which made a lot of room.

here are some pics: https://slow.pics/c/9loK3z3G

Yes the 220uf are needed, however, the way you had to wire the LC filter due to location and space may make things worse or have no effect.  Notice on the example TI PCB how straight and fat the GND and filter cap traces are for their LC filter and how deliberately separated from the switcher it is positioned.  It is laid out that way on purpose.  I would wait until you can scope the current board to see if the extra LC filter is needed.

As for R19 and R18, I cannot say what the best value should be without scoping the PCB.  They were put in to prevent the input of the opamp from receiving the 'grunt' of a high current power signal trace, not as some sort of gain value.  I would make those 2 resistors as small as possible yet not be too heavy current if the opamp input happens to be diode clamped during power-up or power-down.  I would have used 1k for those 2, but, you were 'part consolidating'...

The only reason to use R18 and R19 at all is because of the bias current of the op amp inputs.  Ideally this resistance is as low as possible, but it should match between the + and - inputs.  This creates a balanced drop for both inputs minimizing offset.  V5 runs through a 3.3K resistor to the + input of U4A so R18 is 3.3K.  V3.3 is from a divider that produces a 6.7K Thevenin equivalent so R19 should be 6.7K.  Of course these values can be adjusted up or down as long as the + and - input resistances match.  The schematic is an image rather than a PDF so I can't search it to see where V_5 goes and how much current is drawn, but I expect the total load is light and the resistor chain for creating V_3.3 can draw a lot more current without problem.  Or a proper voltage reference could be used and resistor dividers used for feedback, but I think it has been said that the DC accuracy is not at all important.