Low ripple low noise small power supply design

[VEGETA]

Dear EEVBlog members...

I am making a small PSU for the Dreamcast, it is linear with pre-regulator switching supply fed by 12v laptop\cheap power supply.

I made the design and it is final but it occurs to my thought using a capacitance multiplier to eliminate the ripple more, despite using 1000uf elec caps (+ other 10uf ceramic caps before and after linear ldo).

I assume these caps are good enough but capacitance multiplier seems to enhance it more as Dave showed, and low ripple low noise is important for me.

I searched and found this mosfet: PMV16XN -> https://datasheet.lcsc.com/szlcsc/Nexperia-PMV16XNR_C110920.pdf

which is low space and small size which is important to me.

can this withstand 1-3 amps? notice that it won't regulate anything but rather will be put before the linear regulator and Vgs will be 6v for one rail (5v one) and 8v for another (3.3v one) which is enough i guess.

So I figured it would be just the Rds on dropout voltage: 3 amps * 3 amps * 0.02 R = 0.18 watts.

So are my calculations correct?


thanks for help!

[Vovk_Z]

Voltage drop of a capacitance multiplier is related to Vgs but not to Rds. Rds is not a parameter for linear mode. So in linear mode a Mosfet transistor will dissipate up to 4V × 3 A = 12 W.
That needs TO220 transistor with large enough heatsink.

[VEGETA]

Voltage drop of a capacitance multiplier is related to Vgs but not to Rds. Rds is not a parameter for linear mode. So in linear mode a Mosfet transistor will dissipate up to 4V × 3 A = 12 W.
That needs TO220 transistor with large enough heatsink.

But this is a logic level mosfet where 4.5v Vgs is fully on, thus won't operate in linear mode right?

I will feed it Vgs of more than 4.5v so I guess it could be fully on. However, I thought that this is how load switches are, not capacitance multipliers. So I wanted to verify more.

Voltage drop of teh cap. mulitplier is not so important to me but rather the heat. I can adjust the switching regulator to compensate for any drop but wouldn't imagine it could be this big despite using > 4.5v on gate.

All this comes from the thought that 1000uF elec. cap + 5 of 10uf ceramics aren't enough to eliminate the ripple.

[Vovk_Z]

Ok, you may use logic level mostet too, of cause, but still 1..2 VDC voltage drop multyplied by 1-3 Ampere is 1..6 W. SOT23 case wan't help to dissipate such amount of power.

[BrianHG]

I am curious, why have you abandoned BJT?
Mosfets have a gradual turn on curve/knee making them poor at output regulation.
A BJT has a very sharp knee and turns on at around a 0.7v drop only.

Using an NPN transistor with a beta of ~100m under load in emitter-follower setup, tying the a 100 ohm resistor to V+ then to the base while having a 100uf cap at the base to GND would be equivilant to having something like a 10,000uf cap in circuit.  Say you place a 1000uf cap, this isn't a really large number, but, you would get something like a 0.1 farad cap output at the emitter, and unlike the mosfet, the transient load capabilities will be regulated in the mv with a change in load in the amps, not a half a volt like a mosfet.

Adding a pull-down resistor on the base, or a series diode in line with the 100ohm resistor would only give your filter an additional 0.7v drop/separation/headroom from the source VCC's ripple improving regulation in the case where you may need to clean a large 1v ripple.  2 diodes in series to clean a 1.5v ripple.  You have very precise control here unlike mosfets with their wide Vgs.

Note that with a voltage drop of 2v, 3 amps is still 6 watts of heat you will need to dissipate and I suspect you will need at least a TO-225 package transistor with a little heatsink, or a TO-220 to be safe.

Operating that tiny mosfet in the linear region, at 3 amps, there will be an around 2v drop from drain to source, meaning that sot-23 will need to dissipate 6 watts of heat.  At room temp, that device is only rated to dissipate 0.51watts (P_tot in table 5) at room temp.  When it heats up to 100 degrees C, (and it will with even @ 1 amp load), that drops to 40% according to the graph on Fig.1, meaning 0.2 watts.  Do I need to say 'up in a puff of smoke'?

[BrianHG]

The minimum transistor I would use is an MJE200.
Very good h_FE up to 3 amps.  IE expect 70x you capacitor figure up at a full load of 3 amps, 100x and below 2 amps.

BD437 also looks good.

[BrianHG]

For a mosfet in your application, in a linear voltage follower mode, the choice I would make would be the 'MCU90N02' by Micro Commercial Components.  With a 3 amp load, the Vgs will be around 1.6v.

It's in a DPAK case and I would still have good copper flood on my PCB to dissipate that heat from linear operation, or a tiny heatsink glued onto the device package will give you enough headroom.

Vishay's 'SQA410EJ' is in a much smaller package 'PowerPAK SC-70' and has better Vgs performance on par with your SOT-23 device, around 1.2v @ 3 amp.  However, make sure your PCB has a proper good plated surface area to cool the device and you will just make it.

[langwadt]

why would a Dreamcast need "low ripple low noise" ?

[VEGETA]

why would a Dreamcast need "low ripple low noise" ?

I am making a replacement PSU as I mentioned in the beginning of the topic.

Giving it noisy power may affect its output since this is analog video and audio we are dealing with. People came up with DreamPSU which uses switching regulators only which is noisy. Original PSU had several problem and it didn't age very well according to many many people especially when you hook the GDEmu device which makes the 12v no longer used... then 12v regulator is floating -> starts generating heat and so on... you could put 1k resistor as minimum load on it but this is not a professional solution right? still people report the heat and stuff from it.

My solution is this: a little PSU (50mm x 50mm) which takes 12v input from a laptop or chinese or whatever power plug as input, and outputs a cleaner voltage due to the fact that I used a switching pre-regulator with linear post-regulator with about 0.5v in between to save power and heat. this way I can provide a cheap device, low noise (as low as possible) and won't generate heat.

Then I knew or remembered the cap. multipler thing and thought it may enhance the performance more... however, the board is packed xD and no place on it. even sot-223 may not fit anymore. Since I will be using jlcpcb service to make the boards, at least for now.

I put 1000uf elec. + 5 of 10 uf ceramics at the output of each rail hoping it is enough, is it?

I am curious, why have you abandoned BJT?

I didn't, I just looked at dave's video without much thinking about it. My problem is with the heat, I don't want heatsinks at all and thought that maybe I could make a cap. multiplier to do the job without much heat and drop voltage which using very small components.

Here are some pictures of the design: https://slow.pics/c/iwFVoT81

[BrianHG]

I am curious, why have you abandoned BJT?

I didn't, I just looked at dave's video without much thinking about it. My problem is with the heat, I don't want heatsinks at all and thought that maybe I could make a cap. multiplier to do the job without much heat and drop voltage which using very small components.

Here are some pictures of the design: https://slow.pics/c/iwFVoT81

I can see how looking at an SOT23 with 2amps and think, hey, this device wont give off heat like a BJT.
The Virshay 'SQA410EJ' I listed is actually just about identical to you NXP PMV16XN, it only has slightly higher Rds ON figure of 28mOhm  VS 16mOhm, however where the Virshay wins is at room temp, it can dissipate 13watts, 4.5 watts at 125 degrees through it's package where NXP can only do 0.5watts, or, 1.2watts on a PCB with 6cm square (Well just read Table 5 in the datasheet) which goes down to 40%, IE 0.50 watts @ 125 degrees with PCB heat-sink.  Again, in NXP data sheet, see figure 1/2, the power de-rating curve.

For your project, you have a ripple being generated by a cheap PSU switching supply.  I assume that there is no modulating current in the load.

Feeding a mosfet gate with a cap multiplier, you can get away with 1uf or 10uf with a good series resistor like 1k instead of 100 ohm like the BJT.

Adding a 1k parallel resistor to a 12v to load a minimum 12ma isn't a big problem when the supply is capable of delivering over an amp.  I would just make sure it is a 1/4watt resistor.

As for switching noise, there do exist quality switching regulator designs which make really cleaner and tighter voltage outputs, though, this usually means more coils, or, higher frequency switching ICs.

Remember, a capacitance multiplier doesn't regulate, it only smooths out noise by slightly lowering voltage.
And that output will raise and lower depending on output load by a bit, IE there is no true voltage regulation at the output.

If I were to make this project, I would concentrate on filtering the 12v supply input and having a clean switcher for the 3 output voltages.  You have a really tight PCB and if you want 'Analog' grade voltage regulation, you might not be able to achieve that in that truly cramped space for 3amp, 3 outputs.  To get analog quality supply, this also means taking care about GND loops and power tracing locations and how the layout of the caping is located on your PCB, otherwise, the ringing EMI from switching supplies will go through the PCB circuitry as if it were an antenna designed to tune into those nasty spikes.

Now, if Q1,Q2,Q3,Q4 on your PCB are the NXP mosfets which are your capacitance multipliers, and that's all the space you got to release their heat, you are in trouble.

Also, if you have 12v input, from a dirty supply, only expect 10v output after the capacitance multiplier, maybe 11v at loads under an amp.

[BrianHG]

Have you though about the new modern low noise step down switching converters?

There are new 2MHz ICs which will give you a ~15mv P-P noise with 100ma load, and that shrinks down to ~3mv P-P when driving 4 amps.  This is so small it's even difficult for many scopes to measure.  And since it's up at 2MHz, all you would need is a second series output choke inductor or ferrite bead with a 1uf cap the erase that residue oscillation.  If you have any length of wire between your PCB and Dreamcast, ~10mv at 2Mhz wouldn't even make it to the Dreamcast as it's onboard caps would filter out such a weak high frequency noise over any length of wire, especially if you run that wire through a ferrite core.

The 0 load output regulation has also vastly improved on these new ICs.
At no load, you would expect a 5v output to be at 5.07v, dropping down to 5.00v at 500ma staying flat all the way up to 4 amps.  Even 4 amp linear regulators operate in this ballpark.

Take a look at TI's LM62440 here: https://www.ti.com/store/ti/en/p/product/?p=LM62440APPQRJRRQ1&HQS=OCB-tistore-invf-storeinv-invf-store-findchips-wwe

Using 2 of those on your PCB for 3.3 & 5v with a good choke filter to pass through the 12V input to output would simplify your PCB to 2 regulators and a few extra inductor chokes.

Or you can have a 15-24v source supply and use 3 switchers to get all 3 regulated outputs.

(Warning: to achieve the datasheet's noise levels, you need to follow the example PCB layout as close as you can.  Making changes, unless you know what you are doing may increase that ~5mv ripple (assuming moderate loads).)

[VEGETA]

For your project, you have a ripple being generated by a cheap PSU switching supply.  I assume that there is no modulating current in the load.

the 12v rail is a direct connection to the cheap psu which is not so critical.

However, the 3.3v and 5v are the important ones. I wanted a switching pre-regulator followed by linear post-regulator. I insist on linear being the final one.

When I return home I will give you the names of the ICs chosen.

Now, if Q1,Q2,Q3,Q4 on your PCB are the NXP mosfets which are your capacitance multipliers, and that's all the space you got to release their heat, you are in trouble.

No, those in the bottom are for minimum load for each rail... they get about 12 ma or so to solve the problem of floating supply rail and ensure stability.

Cap. multiplier doesn't exist in this design yet.

Have you though about the new modern low noise step down switching converters?

I will give you the name of my chosen IC when I return and oped KiCAD. I think it is AOZ1284, but I will verify soon. it is configurable from 200khz to 2mhz.

According to it's datasheet, the output ripple is about 5mv or so (rough reading from plot). This followed by a linear regulator + many caps... I think it would be enough? gotta check the kicad project!

Or you can have a 15-24v source supply and use 3 switchers to get all 3 regulated outputs.

this could be another solution but people are used to readily available 12v psus. I could make another version later on but I need to make this one works as intended.

The 12v rail is used only to drive the GD-ROM device which is maybe just the motors and so on to drive the assembly... nothing serious, so yeah... 12v could be noisy here. However, 3.3v and 5v are different, they are for the board itself which contains the output voltage.

as you know analog voltage is 0.7v p-p which is highly sensitive to noise.

[BrianHG]

For your project, you have a ripple being generated by a cheap PSU switching supply.  I assume that there is no modulating current in the load.

the 12v rail is a direct connection to the cheap psu which is not so critical.

However, the 3.3v and 5v are the important ones. I wanted a switching pre-regulator followed by linear post-regulator. I insist on linear being the final one.

When I return home I will give you the names of the ICs chosen.

Now, if Q1,Q2,Q3,Q4 on your PCB are the NXP mosfets which are your capacitance multipliers, and that's all the space you got to release their heat, you are in trouble.

No, those in the bottom are for minimum load for each rail... they get about 12 ma or so to solve the problem of floating supply rail and ensure stability.

Cap. multiplier doesn't exist in this design yet.

Have you though about the new modern low noise step down switching converters?

I will give you the name of my chosen IC when I return and oped KiCAD. I think it is AOZ1284, but I will verify soon. it is configurable from 200khz to 2mhz.

According to it's datasheet, the output ripple is about 5mv or so (rough reading from plot). This followed by a linear regulator + many caps... I think it would be enough? gotta check the kicad project!

Or you can have a 15-24v source supply and use 3 switchers to get all 3 regulated outputs.

this could be another solution but people are used to readily available 12v psus. I could make another version later on but I need to make this one works as intended.

The 12v rail is used only to drive the GD-ROM device which is maybe just the motors and so on to drive the assembly... nothing serious, so yeah... 12v could be noisy here. However, 3.3v and 5v are different, they are for the board itself which contains the output voltage.

as you know analog voltage is 0.7v p-p which is highly sensitive to noise.

If you are talking about audio, 2Mhz is outside the audio band.
(TI-Switcher, ~2amp load) In analog video, 0.003v P-P ripple @ 2.1Mhz may be seen if it is injected right into the video line, however, I doubt this could happen unless the there are no caps on the Dreamcast.
(AOZ1284 - ~2amp load) You are getting 0.01v P-P ripple at 600Khz.  A little more noise, however, this regulator also injects a nasty 0.2v P-P into your 12V supply coming in.  This is harder to get rid of especially at the lower 600Khz frequency.

If your problem is not these ripples, but load transient regulation, then this is a different problem and a mosfet capacitance multiplier will actually makes things much worse.  Even BJT will be much cleaner, and if your load is at a minimum say 250ma, going up to 3amp, a darlington transistor would near 0 high frequency ripple, with the cost that there will be a 1.7v drop from collector to emiter with the TI switcher, though, your 100uf cap with a 1k series resistor will look something like a 1 farad cap on the power supply line, better than a lead acid battery.  You are at a point where you might as well use no switchers and 2 darlingtons with zener diodes on the bases to GND creating your own super responsive linear regulators.  Though, powerup will take something like half a second as the caps on the base slowly charge to the zener diode point.

Example:
(TI Switcher, 2amp to 4amp transient on 5v)  A 0.1v positive or negative spike for 40us before it completely settles.  This correction is outside the audio band, you would not be able to hear it.

(AOZ1284 -  2amp to 4amp transient on 5v)  A 0.4v positive or negative spike for 100us before it completely settles.  (You can hear this as tin-type washy bit error sounding noise with heavy processing if these load changes are due to CPU/GPU current processing load as it will be in the 10Khz band.)

(Voltage drop across the PMV16XN when going from 1amp to 3 amp load) - A ~0.4v, not a spike, but a continuous drop which stays there until the load is released.  This will be heard throughout the audio band as there is no output feedback correction whatsoever.

The low noise TI switcher beats both the AOZ1284 and PMV16XN mosfet capacitance multiplier hands down.

Normal speed LDOs take up to 30us to settle from a 2 amp transient load at up to 0.3v spike.  The low noise TI switcher actually competes here where is has less ripple and is slightly slower at 40us to completely settle.

High Speed Linear LDO regulators like the LM323 can win here in that with a 2 amp load swing, or 2v Vin swing, they will deviate 0.2v, but, for only 1us, correcting their output in the 1Mhz band.

Now without knowing the setup and the source of the Sega Dreamcast noise, as it should have noise suppression on it's analog components, however, older gaming console may have not had such quality audio components, the low noise TI switcher will get you damn close to an linear supply.  And, if the original Dreamcast supply used old analog regulators like the 7805 series/similar, the TI switcher should outperform them.

The AOZ1284 switcher has transients and responses due to load change within the audio band.  This you may want to double regulate, however, just using a mosfet capacitance multiplier will reveal even more lower frequency noise since it's output drifts with current load.

I remember with really loud volume on my old Amiga 1000, there was ugly noises as I moved the mouse and when it's blitter processor rendered animation.  The +/-12v going into the op-amps weren't the problem.  Even the 5v coming out of the PSU wasn't too bad.  It was the routing to the sound IC (Paula) itself on the Amiga motherboard which had a noisy 5V since back in the day, it was a 2 layer PCB.  The only fix was t remove the IC, sit it on a breadboard IC socket with it's own 5v VCC supply separated from the rest of the computer motherboard, with GND referenced to that IC socket's GND and a dedicated 7805 for that 1 IC, I was able to erase all processing noise.  No capping or improvements to the main power supply could fix this problem since that noise was introduced on the motherboard itself.

[VEGETA]

Update:

I am using AOZ1284 with 20KOhm frequency set resistor... that means 2 MHz switching frequency.

I take the 12v into the AOZ1284 then output 3.8v for the 3.3v LDO and 5.5v for the 5v LDO (input of LDOs have 3x 10 uF ceramic caps). The LDOs are LM39302 (Chinese clones, cheap) which are followed by 1000uf elec. cap (10mm x 10mm one) + 5x 10 uF ceramic caps. The inductors are 10uH big size.

You guys mentioned 2MHz devices, well, here is one used... also followed by linear LDO with good amount of caps.

How much ripple do you think we get?

[BrianHG]

Did you check the 'LM39302' datasheet for the 'Line Transient response, power supply VS ripple rejection frequency range?

The 'lemon' range for those LDOs is 200Khz to 2Mhz.  Your 2MHz switcher ripple and transient load noise will basically only be lowered by only 15db at best if your PCB is perfect with a middle thick GND plane.  Combining a well chosen ferite bead with a good 1-10uf LOW-ESR cap would actually do a better job than adding a linear regulator, cost less and make no heat not to mention taking less PCB space.

Since you are combining technologies in a small footprint, not using a 4 layer board with a dedicated reference GND middle layer and a power GND layer poured in the right places on top, you will need to experiment as switching supply EMI noise tends to penetrate linear regulators.

The 'AOZ1284' wasn't designed or specified to be low-EMI radiating and it does place a lot of ripple back onto your +12v meaning 2 of them will make your 12v look nasty without isolation from each other.

The only true way to avoid all these issues it to purely go all linear, but, it seems if your 12v PSU may be a low quality switcher which you might not have control over, it's EMI might go right through the linear regulators if it has spikes in the 200Khz region.

If I were you, before spending money on a PCB, I would first make simple linear power supply (hand wired or quality linear bench supply) with >10000uf caps (paralleled with 1000uf & 100uf, the smaller ones closer to the Dreamcast) on the 3 lines powering the Dreamcast to make sure that the noise you are trying to eliminate will be gone, otherwise, this may be an exercise in futility.

As for the capacitance multipliers, for this app, I would scrap the linear regulators and just use darlington transistors at the output of the switcher if I had to take that route.

If the above 100000uf linear supply test performed the way I like, these would be my 2 strategies:

A) I would  just go with the high quality TI switcher, operate it in it's 'FPWM' mode, if you analyzed it's datasheet, you would have noticed a perfect voltage output at any load from 0 amps to full 4 amps.  Then filter it's output with an smd ferite bead & an ultra-low ERS 47uf cap & it would be as clean as any linear regulator.

B) I would go with the AOZ1284, tune the output voltage by ~2v, then feed an emitter follower darlington transistor buffer (1k to base, 47uf to GND) with a 200 ohm pulldown on the 5v output and 100ohm pulldown on the 3.3v output (1/4 watt) to cancel out the darlington's internal resistors and keep them switched on.

[langwadt]

seems like a discussion of whether an anti aircraft gun or an surface to air missile is the most appropriate for  killing a fly

[BrianHG]

If I were you, before spending money on a PCB, I would first make simple linear power supply (hand wired or quality linear bench supply) ...

I think testing a hand made 'nuclear bomb' first with on hand parts may be wise to make sure attempting such a supply will be effective.  Otherwise, if his Dreamcast still makes noise he doesn't like, the problem may lie elsewhere and no matter what kind of PSU he makes will solve the problem.

For example, if it's interference in the audio, no PSU may solve the problem, yet, adding a thick copper GND to a particular IC on the Dreamcast with a good 220uf ultra-low ESR cap to that same IC's VCC pin without anything else may completely erase the problem.  Maybe in combination with lifting that IC's VCC pin and placing an SMD ferite bead.

[Doctorandus_P]

I did not read all the post here.
For a capacitance multiplier, a BJT is usually a better / easier option than a MOSfet.

If it's for in between the pre-regulator and the linear regulator, then seriously consider adding an extra inductor, to make it a LC filter
Electronics are having trouble with reacting fast enough tho high frequency stuff, while inductors only get better at higher frequencies.

Not all inductors are equal though, and it's not just inductance and current saturation rating. For good HF filtering, it's got to keep its properties into MHz range. Low capacitive inter winding coupling probably also helps.

It's just speculation, but using a straight ferrite antenna core, and using regular plastic insulated wire around it may be an excellent choice The thick plastic insulation introduces distance, and therefore lower capacitance between windings. Straight rod inductors are used as filters in for example PC power supplies. I do not know how their properties relate to toroidal or potted cores though.

Another Idea I've thought of, but never experimented with is to put a wire through a stack of "carosserie" washers. This would not work well as an inductor, but the eddy currents may be excellent at dissipating the HF noise as heat. A piece of thick walled aluminum or copper pipe may be even better at this.

[Zero999]

Be careful with LC filters, they can ring and generate higher voltages, at certain frequencies. It's often desirable to add some resistance, in parallel wih the inductor, or in series with the capacitor. If possible, choose the core type for the inductor so it's lossy enough at the resonant frequency of your filter, to make it criticaly/over-damped.

[VEGETA]

My design is just 2 layers now and I thought it could be good enough.

I don't have high quality test gear, only a DIY switching psu with no oscilloscope. I intend to buy 1054Z in December when I got enough money and stuff. I really prefer a cheap used one if it is gonna save a lot of money.

I could make 10 of these boards now and send to to people to test it, if you don't mind having one when they arrive.

The original stock PSU is using linear design with transformer, but I don't know if it makes any switching or so before the linear stage. I got an idea before this, which is to use one of those meanwell AC-DC module (https://www.meanwell.com/productPdf.aspx?i=683#1) as a source instead of the cheap 12v psu but it is very pricey and heavy to ship here to Jordan, also it has about 150mv p-p ripple which is worse that cheap psus right?. The complete unit should sell for about 40$.

Also, maybe I forgot to tell, but my goal too is to exclusively use JLCPCB SMT service to make everything which is why I spent much time picking parts suitable to it... and it worked. All the items can be assembled by that service which also explains why no component exists on the bottom of the board.

B) I would go with the AOZ1284, tune the output voltage by ~2v, then feed an emitter follower darlington transistor buffer (1k to base, 47uf to GND) with a 200 ohm pulldown on the 5v output and 100ohm pulldown on the 3.3v output (1/4 watt) to cancel out the darlington's internal resistors and keep them switched on.

this would need a heatsink right? which is not suitable to this thing... I don't want heatsink, thus I made only 0.5v difference between linear and switching stages.

[BrianHG]

B) I would go with the AOZ1284, tune the output voltage by ~2v, then feed an emitter follower darlington transistor buffer (1k to base, 47uf to GND) with a 200 ohm pulldown on the 5v output and 100ohm pulldown on the 3.3v output (1/4 watt) to cancel out the darlington's internal resistors and keep them switched on.

this would need a heatsink right? which is not suitable to this thing... I don't want heatsink, thus I made only 0.5v difference between linear and switching stages.

You had linear regulators, correct?
Did they have heatsinks?
If you use a transistor in place of a linear regulator, the heat given off will be identical.

Take a look at the NPN MJB44H11T4-A.  If you were to use this transistor, it will multiply you capacitor capacitance and divide it's ESR by at least 200 all the way up to 3 amps.  It will also reject frequencies above 10Mhz as it is a single BJT.

If you wanted to multiply that capacitance by 40000, I would add a MMBT3904 in front of the base creating your own Darlington, with a emitter to base series resistor of ~10 ohm and a pulldown resistor on the 3904's emitter to gnd of ~470ohm.  This will keep that speed where as with a true darlington like the Fairchild/On-Semiconductor's 'FJB102TM' will give you 4k, but only at 2 amp load.  It has a weaker gain at lower currents because of the internal load resistors and their setup means the transistor will only reject frequencies above the 1MHz range.

If you do not want a capacitance multiplier, you should just use the Diode's Inc 'AZ1084CD-ADJTRG1' linear adjustable 5 amp regulator.  Study the datasheet's load transient response.  It is what you are looking for and it only needs a quality 10uf cap at the output.  Do not use Linear's LM1084 as it is over 10x slower and has 5x the voltage spikes on the graph.  It also happens to be the cheapest of the 5 amp linear regulators.  Though for rejecting RF, you cannot beat the MJB44H11T4-A / MMBT3904 combo.

If you studied the graphs on the 'AOZ1284' and the TI's 'LM62440-Q1', you would have noticed that the 'AOZ1284' is a piece of crap.  2 MHz isn't the important factor here.  Even running TI's  'LM62440-Q1' at 400KHz according to the charts completely obliterates the 'AOZ1284', IE TI's figure 58(400Khz) & 60(2Mhz) on page 40 VS aosmd's figure on the bottom left of page 6.   Only Diode's Inc linear regulator 'AZ1084C' puts TI's switcher to shame.

[BrianHG]

Give me a day on finding you a suitable PSU.  You shouldn't be paying more than 6$, 15 at most for 12v, 30 watts.

Quality noiseless switchers, with high mains isolation & UL/CE approvals already exist at the 25$-35$ price point from reputable sources like TDK & TrIIad which I've used in Hi-Fi audio & video equipment with studio grade performance.

However, it's the size as these would need to be in an external box and they have a metal frame.
Though, you already get 12v,5v,3.3v all in 1 package.

[VEGETA]

You had linear regulators, correct?
Did they have heatsinks?
If you use a transistor in place of a linear regulator, the heat given off will be identical.

the linear regulator will have 0.5v dropout only, so assuming 3A max current (which, typically would be only 1A or so), this will be 1.5W total. This transistor will get 2v dropout at least right? so this wouldn't be enough without heatsink.

If you do not want a capacitance multiplier, you should just use the Diode's Inc 'AZ1084CD-ADJTRG1' linear adjustable 5 amp regulator.

so this instead of my linear regulator (LM39302) will make a big difference? it is cheap and available at JLCPCB assembly service. however, why this one is significantly better than LM39302 here despite having the same input ripple?

Though for rejecting RF, you cannot beat the MJB44H11T4-A / MMBT3904 combo.

if I could find a way to put these before the linear regulator and still getaway without heatsink, I will look into it.

If you studied the graphs on the 'AOZ1284' and the TI's 'LM62440-Q1', you would have noticed that the 'AOZ1284' is a piece of crap.  2 MHz isn't the important factor here.  Even running TI's  'LM62440-Q1' at 400KHz according to the charts completely obliterates the 'AOZ1284', IE TI's figure 58(400Khz) & 60(2Mhz) on page 40 VS aosmd's figure on the bottom left of page 6.   Only Diode's Inc linear regulator 'AZ1084C' puts TI's switcher to shame.

Where exactly in the datasheet can I find these info? I mean to know which one is better at noise rejection and final output noise quantity... etc?

[VEGETA]

I have found these:

ON Semiconductor MJD44H11T4G.

MMBT3904

they are cheap and most importantly available at JLCPCB assembly service which is critical to me as mentioned. However the transistor is no -A but rather -G. the A is for automotive, does this really matter?

also the MJD44H11T4G has a junction-to-ambient temperature of 71.4 degrees per watt.. meaning if it has 2v of drop voltage at maximum current it will be 6 watts. 6 watts = around 450 degrees! but if we reduce the drop voltage to 1v this will be about 240 degrees which is still too much.

So if we wanted to use this approach then heatsink is a must?? this DPAK package can use heatsinks like this: https://www.fischerelektronik.de/fileadmin/fischertemplates/images/SMD_Bauteile/bild1.gif right?

if I wanted cheaper ones I could get a piece of Aluminum cut at a square shape then screw it to the pad. this could work but requires more money and labor. no way i can order these from outside! plus I don't think I can solder them with that thermal mass they have.

can we make this work without heatsinks? I am really ready to adjust the design altogether  if we can do it without heatsink.

My assumption is this:

12v source -> AOZ1284 -> Cap. multiplier -> LM39302 for 3.3v and 5v.

how much ripple and noise are expected after this?

[BrianHG]

I have found these:

ON Semiconductor MJD44H11T4G.

MMBT3904

they are cheap and most importantly available at JLCPCB assembly service which is critical to me as mentioned. However the transistor is no -A but rather -G. the A is for automotive, does this really matter?

also the MJD44H11T4G has a junction-to-ambient temperature of 71.4 degrees per watt.. meaning if it has 2v of drop voltage at maximum current it will be 6 watts. 6 watts = around 450 degrees! but if we reduce the drop voltage to 1v this will be about 240 degrees which is still too much.

So if we wanted to use this approach then heatsink is a must?? this DPAK package can use heatsinks like this: https://www.fischerelektronik.de/fileadmin/fischertemplates/images/SMD_Bauteile/bild1.gif right?

if I wanted cheaper ones I could get a piece of Aluminum cut at a square shape then screw it to the pad. this could work but requires more money and labor. no way i can order these from outside! plus I don't think I can solder them with that thermal mass they have.

can we make this work without heatsinks? I am really ready to adjust the design altogether  if we can do it without heatsink.

My assumption is this:

12v source -> AOZ1284 -> Cap. multiplier -> LM39302 for 3.3v and 5v.

how much ripple and noise are expected after this?

Do you have the Junction to ambient temperature of the 'LM39302'?  If it is in the same package, it is not much different.
Also, because of those 0.5v spikes coming from the AOZ1284, and the regulator dropout of ~500mv at top load, with a little regulation safe zone, you would still be powering the LM39302 with ~+1.3v, 3.75 watts of heat.

The transistor derating you are reading is the transistor not even mounted on a PCB, complete open air.  Even a PCB alone will drain away heat.

Why do you need the 'LM39302' when your cap multiplier becomes a linear regulator just by adding the right zener diode between the MMBT3904's base and GND in parallel with your say 10uf cap.

Ok, scrap the MMBT3904 and LM39302.

Just get the cheapest adjustable linear 100ma regulators (LM317 in SMD) which can go to at least 18v input and down to 3.3v out.
You will use the 100ma regulator's output to feed the base of the MJD44H11T4G, multiplying that supplied current by the transistor's current gain curve which could drive ~ 5amps, however, the sweet spot is at the 3amp mark where the transistor's gain is clearly above 100.

The trick to preventing your 12V supply's ripple from reaching the 2x 50ma linear regulators is use a 1/2watt 100 ohm resistor from 12v to the regulator's Vin, and at that Vin, have a good 10uf 25v cap to GND (The GND trace by the output connector).  This 1 resistor and cap can power both regulators simultaneously, or, if you want super separation, use 1 resistor and cap for each LM317.  (Remember, if the MJD44H11T4G is driving a full 3amp load, the regulator powering it's base is driving ~15ma + a minimum pulldown resistor = ~20ma total max.)  While each AOZ1284, remember it makes spikes up to 0.5v, plus you want a little headroom, should power the collector with 1.3v more than the output voltage.  ~4.6v for the 3.3v output and ~6.3v for the 5v output.

     However, the 2 problems with this circuit is output regulation unless you try something I never had.  Tune the LM317 resistors for the desired voltage, however, place the voltage output feedback divider resistor on the emitter output of the transistor instead of the output of the LM317 directly.  This may regulate/correct the temperature drift and load change on the output of the transistor which may have introduced a ~0.3v variance over temp and load conditions if the LM317 took it's feedback from it's output pin.  But you need to make sure that there is no power-up overshoot spike in this case.  You may wire your PCB to operate in both modes since it will only be the placement of 1 resistor or the other.

     The second problem is that there is no true over-current protection.  You will be relying on the maximum current of the AOZ1284 to limit the output power.

With this, since the 2 linear regulators and transistors should be close and share the GND on your output power connectors, and you can move the switchers and V+ traces which feed the collectors further away from everything else, you would do fine.

This wiring configuration relies on the MJD44H11T4G collector absorbing and not passing through all the switching noise to it's emitter.  For this, the output will need a minimum load with a small cap.  Otherwise down at 0ma, some ripple may make it through as the transistor's internal capacitance will transmit some signal through as a slight DC error offset on the output.  This shouldn't be bad as a single transistor like this can operate above 85MHz.

If you knew the individual currents for both 3.3v and 5v, a single switcher at 6.3v may be enough to feed both MJD44H11T4G, though, the 3.3v one may get really hot is all the current is on the 3.3v supply.  Same with the 5v, if it is only an ~1 amp load, you may get away with feeding it more voltage on it's collector.


It's too bad the collector is the tab and not the emitter.  The collector is where you are getting all the noise from the switching supply source and a hunk of metal heatsink and fat PCB power trace may act as an antenna.

As for your illustrated heatsink, for ~5watts of heat, that will be fine, though placing the heatsink under the PCB right under the transistor with stitched vias and a rectangular copper pour on both sides will get rid of the heat more effectively.  Just gluing/taping that heatsink right ontop of the transistor's plastic case may be enough.

If you want to truly maximum protect your transistor, a TO-220 version vertically mounted with a screwed on heatsink would be best, however, for ~5 watts, I don't think you need to go that route.


More powerful transistors exist than the 'MJB44H11T4-A', I just picked the cheapest which would meet your needs.