Linear PSU with modern components, will it matter ?

[BravoV]

Seeing lots of threads on building linear lab psu, and most are based on old proven design, like 723 and etc.

I've been wondering, if using current/latest technology advancements at today components on building it, will it make any significant differences ? And at which parts of PS performance that stand out ?

New components I'm talking here are at all fronts, say like starting like much improved op-amps, better performance as using modern process on the semiconductor or new material at discrete components e.g.: BJT, mosfet, better temp-co or polymer cap, cheap & fast MCU and etc, that practically not existed decades ago.

Ignore the cost part in this discussion please.

Love to hear from experienced EEs thought/idea/insight on this matter.

[Kleinstein]

The main part of a linear power supply does not need modern parts.  The usual way to do it is using reference OPs and power transistors (either BJT or MOSFETs).  The modern OPs might give a little better precision.  Newer DACs and ADCs makes it more practical to use digital set points and digital display instead of the old way ten turn pot and analog meter. However good ADCs where already available in the 1980s.

Today one might chose to use a SMPS instead of the classical heavy iron transformer. This also make pre-regulation more practical.  However it still depends on the power level and required noise.

The LM723 was never intendent for a lab supply. The more lab supply like chip was the long obsolete MC1466.

An advantage one has today is that one can use a simulation to optimize a circuit that can be still quite conventional like build 30 years ago. So less real world tweaks needed.

[tszaboo]

Well, for one, transistor packaging has come a long way, so while a an IRF540 in TO220 has something like 1.5K/W temperature coefficient, a modern FET in the same package could be 0.5K/W which means you need les transistors if you have the cooling for it. You have higher precision ADCs and DACs, so faster control loop can be built around it, and more accurate voltage.
In the end, you get something like a Keithley 2280

[David Hess]

I would say that error amplifiers peaked not long after the OP-07 and similar parts.  Once  you get to 100nV/C of offset drift, chopper stabilized amplifiers do not help much in a large design because external thermocouple effects.  Even in the 1970s there were parts almost that good.  I have made some incredibly accurate power supplies with parts like the LT1007 and OP-27 taking advantage of their low input noise and high open loop gain.

Ring emitter bipolar transistors for low noise and fast response have been around almost as long.  Power MOSFETs are faster if you drive them hard enough but also higher noise.

There are more precision references available now but I think the availability of precision resistors has had a greater effect.  I think there was a better selection of good potentiometers in the past though.

So to sum up my thoughts, I think the design and implementation has been more important than the parts since the 1970s.

[David Hess]

Faster opamps may introduce instabilities. You need to slow it down for stability. But low noise opamps are very appreciated.

The good news is that slowing them down does not increase their noise.  Trying to lower the noise too much results in such a low impedance for the feedback network that self heating ruins the gain stability if not the DC stability; this could matter for a programmable power supply.

Polymer caps/MLCCs may also introduce instabilities. You need some feedback trick to introduce more output ripple into the loop for stability.

Power supplies designed for the best constant current performance minimize output capacitance which means like a 0.22 microfarad film capacitor per amp with a series resistor instead of the more typical 47 to 100 microfarads per amp.

But the trick I like now is to use high side current sensing with the current shunt on the output where it contributes to the output capacitor's ESR if AC feedback is taken before the current shunt.

There is another trick I have not tried where the error amplifier is compensated with *less than* 90 degress of phase shift increasing the phase margin so that a zero ESR output capacitor does not cause oscillation.  I read about this in connection with ATE drivers which have to deal with unpredictable loads.

[coppercone2]

there is a case to be made but it needs to be evaluated carefully about the semiconductor physical sizes and depositions and stuff in terms of reliability and overload survivability and possibly other factors like thermal cycling related fractures. It's kind of a black hole and the information is difficult to get and you get all sorts of complications like possibly higher quality during the manufacturing phase in fabs (for instance I think they used to pay the techs alot more),and stuff might be hidden behind NDA. And I think there used to be a bit less emphasis on cost competition with China.

I mean I don't know how deep you are going with this but you might want to take a look in the art of electronics chapters about mosfets and BJT to see the difference (a modern design tendency would be to use the MOSFET).

As for the foil capacitors I have been recently interested by the manufacturing process of how the foil is connected to the lead (its some sort of spray process I have been meaning to learn more about). I get a bit of a uneasy feeling about it but I think its nothing.

You might also be very suprized by the manufacturing processes you use to make SMD ceramic caps, they use tape based processes IIRC. It's very strange and I think KEMET or one of them had like a 5 minute long video on youtube on how they are manufactured. It just came off (to me) as some weird shit thats difficult to control quality in.



My suggestion to you is to find manufacturing data and videos if possible on the processes they use to make the parts and think about it yourself in a side by side comparison to see if anything there makes you uneasy.

But keep in mind it might look bad but sometimes you are really wrong. When I saw how gorilla glass is made I thought NO WAY that is how they make something so kick ass.

[coppercone2]

I don't put too much faith in MTBF but it is a extremely good starting point.

[coppercone2]

also see the recent thread someone made with good discussion about diodes and transformers being used together to form unregulated power supplies.

There was a good discussion directly about your interest about how changes occurred to the normally used diode. In sort there was a manufacturing process of old that produced parts that had particular features that were beyond the datasheet documentation, the part continued to be used and manufactured for many years with the same data sheet spec but the manufacturing process changes over the decades caused undocumented parameters to change (which were useful in engineering).

read this whole thread
https://www.eevblog.com/forum/projects/film-cap-across-bridge-rectifier-in-xxi-century/?topicseen

[digsys]

After many years designing linear P/Supplies, mostly Intrinsic / mining etc, this was my goto design for several years -
(obviously not the full design, but main excerpts) - 0-36VDC op, 0-20A+ op: MAX loss - Bridge and regulator = 0.8V (best min was 0.6V, but I left headroom).
NO fan, very minimal heatsink ! ~95% efficiency (not including transformer) .. not bad for linear :-)  220-260V ACin
Original design had a multi-tap transformer (lots of taps), then played with magnetic feedback, resonant circuits etc .. but a completely linear circuit !
One day, I'll see if Daves capacitance multiplier works :-) ... oddly enough it has a micro, but just for the front end :-)
So, can Linear P/Supplies get better ?? Shit yeah !!

[coppercone2]

What happens to noise when you start messing with magnetic feedback paths?

I am very excited about your design methodology.

[digsys]

  What happens to noise when you start messing with magnetic feedback paths? ...

Funnily enough, there is no noise .. as such :-)  What you get is AC distortion in V and I ie it is not a pure sine wave anymore BUT it is still an AC waveform,
so you don't have to worry about switching "spikes", harmonics etc  It is doable ie a reasonably clean AC V and I, but you need to know a lot about core
magnetization / leakage / etc etc Actually, even though we got it pretty good, we abandoned it for other ideas :-)  .. much simpler
Magnetic feedback is used in many systems - google "Transformer magnetic feedback" Images for lots of ideas ... gl

[coppercone2]

I wonder if LT  or AD ever wanted to make a LT4320 version that will run directly off mains

I think its a bad idea, but I always wanted to experiment using it rather then a diode bridge.

I found the part before I had a SA and RF gens/amps then I forgot about it for a while now it might be time to do some comparisons with various DC powered precision circuits I have.

[BravoV]

After many years designing linear P/Supplies, mostly Intrinsic / mining etc, this was my goto design for several years -
(obviously not the full design, but main excerpts) - 0-36VDC op, 0-20A+ op: MAX loss - Bridge and regulator = 0.8V (best min was 0.6V, but I left headroom).
NO fan, very minimal heatsink ! ~95% efficiency (not including transformer) .. not bad for linear :-)  220-260V ACin
Original design had a multi-tap transformer (lots of taps), then played with magnetic feedback, resonant circuits etc .. but a completely linear circuit !
One day, I'll see if Daves capacitance multiplier works :-) ... oddly enough it has a micro, but just for the front end :-)
So, can Linear P/Supplies get better ?? Shit yeah !!

Digsys, thanks, any chance you share the whole schematic ? Including the tap switcher if you mind.

Sorry, is that with adjustable current too ?

I can see you're still using that venerable LT4320 there. 

[digsys]

  ... any chance you share the whole schematic ? Including the tap switcher if you mind ... is that with adjustable current too ?

Tricky situation. It was 1 of many contract designs ... then got sacked after several years :-) (long story) .... never paid for R+D or other "trades" ...
With all that, I don't want to risk any more animosity ... circuit operation theory, though, is common knowledge -
Run a small dual 5-8VAC, say 50mA transformer to power logic, REF 0V, and an isolated 6-12V high side FET driver rail
Use a small micro, with A/Ds to monitor AC V and I, and DC V and I. Then depending on Vop, switch in the respective winding, using the bridge type FETs.
Have this pre-regulator on a separate PCB, with it's OWN storage caps. Then, IF you want to get fancy, "active feed" the MAIN storage cap that supplies the
regulator PCB. At low current, it doesn't matter if Vin_Reg is a bit higher, as power dissipation will be low. Keeping the Vreg cap at an ideal, low ripple voltage
is the "tricky" part :-) .. but then, working that out, is the challenge :-)  In one 10A design, I believe I got ~97% efficiency .. not bad for linear (-Trx).

I can see you're still using that venerable LT4320 there.   

Ahhh yeah, didn't I send you a couple boards way back? Did you ever use them? Useful? It's a DAMN fine IC !!

[BravoV]

  ... any chance you share the whole schematic ? Including the tap switcher if you mind ... is that with adjustable current too ?

Tricky situation. It was 1 of many contract designs ... then got sacked after several years :-) (long story) .... never paid for R+D or other "trades" ...
With all that, I don't want to risk any more animosity ... circuit operation theory, though, is common knowledge -
Run a small dual 5-8VAC, say 50mA transformer to power logic, REF 0V, and an isolated 6-12V high side FET driver rail
Use a small micro, with A/Ds to monitor AC V and I, and DC V and I. Then depending on Vop, switch in the respective winding, using the bridge type FETs.
Have this pre-regulator on a separate PCB, with it's OWN storage caps. Then, IF you want to get fancy, "active feed" the MAIN storage cap that supplies the
regulator PCB. At low current, it doesn't matter if Vin_Reg is a bit higher, as power dissipation will be low. Keeping the Vreg cap at an ideal, low ripple voltage
is the "tricky" part :-) .. but then, working that out, is the challenge :-)  In one 10A design, I believe I got ~97% efficiency .. not bad for linear (-Trx).

Understandable, infact thanks for introducing that Micrel chip.

How its perform, say with switcher pre-regulator instead of using step-down transformer ? Did you ever try this combination ?

I can see you're still using that venerable LT4320 there. 
Ahhh yeah, didn't I send you a couple boards way back? Did you ever use them? Useful? It's a DAMN fine IC !!

Yep, used it at few projects, they perform flawlessly ! 

[digsys]

How its perform, say with switcher pre-regulator instead of using step-down transformer ? Did you ever try this combination ?

Ahhh there's the problem ... NO switchmode allowed in intrinsic / mining areas AT ALL !! period ! Devices must be 100% safe from arcing / HV etc
(ok as close as possible). If you DID want to try, It's be the same cost / effort as getting something approved for planes.

[coppercone2]

How do you deal with inductance ? Like a high power rail that is long? Snubbers allowed or do you need to do something else?

[coppercone2]

AVE video teardown of a mining product

[digsys]

How do you deal with inductance ? Like a high power rail that is long? Snubbers allowed or do you need to do something else?

Luckily, I only do the stuff inside the box :-)  (usually explosion proof w/ no pluggable connections). That's a whole new messy world that the specialist
electricians have to deal with :-)  The only specialist one I knew well, croaked this year, so I can't ask him .. sadly.

[BravoV]

As power switching supply is matured (cmiiw), when it comes to use it as pre-regulator, instead of big transformer, with modern components, does it matter on the suppressing the switching noise at the linear front end, say to make it as comparable as step down transformer based linear PSU ?

[coppercone2]

if your switcher is making RF then it makes sense to do so before the LDO because the RF gets into the LDO and it ends up mixing and rectifying and messing with its control system and also becoming higher in frequency and more difficult to deal with.

I think it might manifest itself as a low frequency ripple / wandering offset voltage, especially with high impedance set LDO like a LT3080.

[Kleinstein]

For the really high frequency ripple of a switched mode pre-regulator it is a good idea to have some LC filtering before the linear stage. At least for the higher frequencies this is relatively simple, so that the LDO only has to deal with the base frequency at a reduced level.  At high power it might be worth to have the switched mode part in a multi-phase circuit to reduce the ripple and keep the frequency high.

For the linear stage it depends on the type of circuit how good they suppress higher frequencies. Ready made LDOs like the LT3080 may not be the best solution here.

[BravoV]

For the really high frequency ripple of a switched mode pre-regulator it is a good idea to have some LC filtering before the linear stage. At least for the higher frequencies this is relatively simple, so that the LDO only has to deal with the base frequency at a reduced level.  At high power it might be worth to have the switched mode part in a multi-phase circuit to reduce the ripple and keep the frequency high.

Cmiiw, the LC filtering is good at high freq spikes right ?

What interest me is you mentioned "base frequency" , with the current linear regulator, even designed from ground up with discrete components, does it matter at what frequency its the easiest to suppress ?

We're talking switching freq range from > 100Khz up to few MHz, and how about sub 100KHz ? Lower is better ?

For the linear stage it depends on the type of circuit how good they suppress higher frequencies. Ready made LDOs like the LT3080 may not be the best solution here.

Any hint or direction which or what is the ideal circuit (and/or incl. chip) especially as the front end for the switcher ?

[exe]

Lower is better ?

I'd say slower rise is better (for noise, not for efficiency). If we slow down frequency we can also decrease switching speed => less RF noise. The downside is the inductor needs to be bigger. It may also have worse parasitics, e.g., capacitance and series resistance. This may also affect size of the board and pcb parasitics.

[BravoV]

Lower is better ?

I'd say slower rise is better (for noise, not for efficiency). If we slow down frequency we can also decrease switching speed => less RF noise. The downside is the inductor needs to be bigger. It may also have worse parasitics, e.g., capacitance and series resistance. This may also affect size of the board and pcb parasitics.

As the scope of this discussion is for bench top psu, I think size and efficiency are not big concern.

So what are you saying is the old switcher that work at sub 100K Hz switching actually are better in our case here ?

[Kleinstein]

Fast switching and high PWM frequency would cause more RF "noise". However the higher frequency also makes filtering more efficient. It takes more care with the layout but filtering can do the job.  For the final performance the frequency of the SMPS stage does not have so much influence. It takes larger inductors and capacitors at lower frequency - so the higher frequency can be attractive for economic reasons.

In principle the lower frequency SMPS could be higher efficiency, but at a high price. So in the practical implementation a higher frequency SMPS with modern switches can be higher efficiency.

For high power modern parts make a multi phase SMPS circuit more practical. This can be more important than the frequency.

[exe]

So what are you saying is the old switcher that work at sub 100K Hz switching actually are better in our case here ?

Not really  as
1) modern designs can have less noise just because SMPS techology also advanced last years (e.g., smaller components, often integrated, better layouts => smaller loop area, less noise).
2) old design might not optimized for noise performance.

Ultimately, the only way to evaluate a design is to measure it (imho). And it's often not trivial, esp. if you are interested in the whole spectrum, not just lower 20MHz noise. That's because noise goes well into hundreds of MHz, a typical scope with passive probes simply not good-enough to do such measurements. Because of this most noise measurements done by mere mortlas are garbage. I've seen even one dude on youtube showing off his aliexpress SMPS. He limited scope bandwidth to 100kHz or something.

Anyway, as a base for a new design, I'd take a relevant app note or demo board. I've seen a few, even those claiming noise well below 1mV (but for limited power output and not sure if it was p-p or rms noise). BTW, look what I found: https://www.analog.com/media/en/technical-documentation/application-notes/an101f.pdf . An app note discussing post filters. Only looked at plots, seems to be relevant.

Also, most applications are not noise-sensite (as long as noise is within 10-30mV p-p). Although, for my lab I built a linear one just to be sure it's suitable for all the possible case I might have.

But if efficiency is not a problem, why not take a linear PSU blackdog's pre-regulator? It's proved design, work well for me. BTW, even with linear PSU it's possible to screw the noise. Like, it can get injected through mains or something.

PS one more thing, it seems a high-power and/or high voltage LDOs for postfiltering don't exists. That's my conclusion after reading hundreds of datasheets and many month of searching for components when I was designing a lab psu with buck converter + ldo post-filtering. More beefy regulators have less ripple rejection in 10kHz+ region. I'd say an LDO does nothing to filter noise that's above its control loop frequency. All the filtering about 100-200kHz is because of input/output caps, imho. Some datasheets even mention this explicitly and show how PSRR is affected by different output caps (different size / esr (esl?)).

PS for some ideas how to design a low-noise PSU one can have a look at modern high-freq scopes. They have tens of local regulators. Like, R&S. Just keep in mind right component selection and proper layout a critical for success. Just taking the same schematic will probably not work.

PPS at freq above few MHz "normal" 0805 capacitors become inefficient due to ESL. This can be seen in a datasheet showing impedance vs freq plot (murata, avx and kemet provide this). Paralleling caps of different size will make it even worse due to, uhm, parasitic resonanse? I can find a video explaining this, if needed.