Output capacitance of higher end bench power supplies

[markus-k]

Hi everyone,

I've been wondering what the output capacitance of higher end bench power supplies like the Keysight E36313A or the R&S ones Dave has is. I couldn't really find anything about what range they are in. I have a Siglent SP-D3303X, which measures ~900µF on each output, and I've seen similar values in Rigol teardowns. That seems quite high to me, especially since it greatly impacts it's current limiting performance to the extend that I have accidentally popped parts with it before despite having set a fairly low current. Shorting the outputs, or just plugging in a device with a big input capcitor at higher voltages always results in quite some sparking. My old HP/Agilent E3611A measures ~230µF output capacitance, better, but I would've expected something in the double digit µF-range maximum.

I understand that some output capacitance is required for stability, but my dad always told me that you'd only need a very small capacitor (hundreds of nF) if the supply is properly designed. So that makes me wonder if they are just badly designed or if it's just impossible to build a 100W+ supply without such big capacitors on the output?

[KaneTW]

My HMC8043 has 32u capacitance on the output (measured by measuring the discharge time from 25V to 0V into a 1k resistor). Multimeter measurements don't really work here (it showed wildly changing numbers).

[maginnovision]

They're older but my R&S NGPV(120W) supply has 500pf output capacitance and you can optionally switch in 47uF. My NGPE(800W) also seems to be fairly low effectively, it has a lot of filtering but will also sink current. I haven't measured it though. This picture is the first filter, there are two but the second has no capacitors. There are quite a few of capacitors here but the current monitoring is after that in the circuit and it has remote sense for the voltage at the output.

[nctnico]

I understand that some output capacitance is required for stability, but my dad always told me that you'd only need a very small capacitor (hundreds of nF) if the supply is properly designed.

I doubt that. You need a decent amount of output capacitance for stability.

IMHO your primary problem is connecting loads while the PSU output is switched on. Better use a power supply which has an output enable switch (or modify it to switch the voltage control line between zero V and the potmeter's wiper) AND teach yourself to attach loads with the output switched off.

[Fungus]

Surely the current limiter has to be after the capacitor, it simply can't work otherwise (current limiting).

[Kleinstein]

Surely the current limiter has to be after the capacitor, it simply can't work otherwise (current limiting).

The output capacitor is by definition at the output - so no extra current limiter after that.

I have not build a lab supply for quite some time, but did some simulations / plans for this. I found a minimum capacitance somewhere in the 1 µF range with some 10 µF as a more realistic value. So it may be possible to reduce the capacitance if really wanted. It makes the output design more tricky, but not too much. It is not so much the power that is relevant, more like the current level. There usually is an upper frequency limit at some 100 KHz - 1 MHz where the capacitor is needed to take over the output impedance.

There are 2 type of regulators: one is with a more collector / drain side output with a more current controlled output stage. These need some output capacitance, depending on the speed.
The others are with a more emitter-follower type output. These can get away without a physical present capacitor. However there still tends to be a large current spike on a sudden short, as the slew rate is limited. So one kind of gets a simulated capacitance or similar or even larger size.

Besides an actual capacitance, there can be a delay until the current limit really sets in. For some use this is actually good, so that current spikes from a DUT would not engage the current limit earlier than wanted.

[exe]

They're older but my R&S NGPV(120W) supply has 500pf output capacitance and you can optionally switch in 47uF.

Wow, this is the first time I see this feature. I'm actually designing a power supply with switchable output capacitors, and I thought it's a  very unique feature of it.

[Fungus]

Besides an actual capacitance, there can be a delay until the current limit really sets in. For some use this is actually good, so that current spikes from a DUT would not engage the current limit earlier than wanted.

So ... devices are allowed to draw more current than the current limit? 

[maginnovision]

Exactly Fungus. Most power supplies are perfectly fine exceeding the current limit for some time and by some amount. Especially general purpose supplies.

[MadTux]

You can reliably zap small transistors and (zener) diodes with output capacitance by setting a power supply (HP 361xA) to a few mA and directly connecting the leads to it, like if you want to measure unknown zener diode when reverse engineering a PCB.

So always use a resistor in series for that or use a source meter (I ❤️ Keithley 236/237  )

[KaneTW]

Use the right tool for the right job

[David Hess]

22 to 100 microfarads per amp is typical for a good design but it is possible to get down to below 0.1 microfarads per amp with care which considerably improves the constant current performance.

[MegaVolt]

I observed a large output capacitor in sources such as:
Agilent E3645A
Rigol dp811
They issue a large spark without any problems with any current limit set.

[J-R]

The 20V 5A HP 66332A appears to have 0.047uF on the output at all times in Fast mode and with Normal it adds in 100uF when the output is enabled.

[markus-k]

IMHO your primary problem is connecting loads while the PSU output is switched on. [...] AND teach yourself to attach loads with the output switched off.

Yes, I agree. It's a little inconvenient at times and lazy me just leaves the output switched on then. But the big capacitor would still affect the current limit if something shorts out later on.

But apparently many supplies do offer way lower output capacitors, which is good to know.

[bsdphk]

The HP66332 is an example of a so-called "agile" supply.

When you use a PSU for (automated) testing, it very often involves rapid changes in voltage and huge output capacitance makes that hard, so instead that kind of supply always have remote sensing, often with surprisingly high bandwidth, which is why you should always use twisted-pair for sense wires.

Some supplies even have what HP calls a "Down Programmer", which is essentially a shunt transistor to make it possible to rapidly reduce the output voltage, irespective of any built in and external capacitance, some of them even take it as far as using it for two-quadrant operation, so you can use the PSU as an electronic load too.

I can highly recommend the schematic of the HP6626A family of supplies, if you like to "study the masters" and if you are really into this, comparing them to the HP6624A family schematics is a good education in marginal noise reduction.

When you do not need agility adding more capacitance does wonders for the noise, and is a lot cheaper than what HP had to do in the 6626A to keep the noise down.

But if you have the space for it, a pre-owned HP6626A for a few hundred bucks is a *really* good deal, the 14 bit precision also makes it a competent source for characterizing components etc.

For reference, the HP6626 has 0.022µF+4.7µF+0.047µF output capacitance on port 1&2 (=25W) and 0.022µF+4.7µF+0.1µF on port 3&4 (=50W)

[thm_w]

My HMC8043 has 32u capacitance on the output (measured by measuring the discharge time from 25V to 0V into a 1k resistor). Multimeter measurements don't really work here (it showed wildly changing numbers).

It will be slightly higher as supply will usually have some resistive load or constant current load on the output to pull it down (1-5mA). But probably not by much, 47uF?
Seems to be: https://goughlui.com/2019/07/28/rs-hmc8043-psu-review-in-depth-ch6-teardown/

[exe]

The HP66332 is an example of a so-called "agile" supply.

Oh, wow, this supply really looks interesting on the inside. The service manual is huge, I'm still going through it (I'm yet to see other manuals you suggested). It looks like it has very minimal output capacitance. I wonder if it is stable with big output capacitance. Probably, yes, but it's not mentioned anywhere in the manual.

Now a practical question. What makes a power supply fast to respond? Like, I want to design one, how should I approach the problem?

Here is my check list so far:
1) do not drive bjts too hard, they tend to slow down. They are only fast at a fraction of their max current. Also higher Vce makes it working faster, they tend to be slow in near saturation.
2) having multiple bjts in parallel may make it faster
3) for small currents it's better to use "smaller" (in terms of current handling) bjts. I actually want to have two channels on my supply: one for big currents, and another one 100mA max, but with greater transient performance and precision.
4) current shunt together with output cap makes a low-pass filter. So, to have less phase lag, it should be smaller.

Any other suggestions?

[Kleinstein]

The output capacitor takes over the output impedance for the higher frequency end. So one should have a fast regulator loop, so that the regulator can generate a suitable low Z even up to higher than normal frequencies.

The output stages also tend to become slower when at relatively low current. So one may consider a low current and higher current part in parallel to also get high speed at low current. It can also help to have the output work as a push pull output - getting a minimum current and being able to go down fast as well. This can also avoid working close to the limit, where anti windup can kick in.
Some kind of anti wind-up can help to speed up the transition between CC and CV modes, but this part can also be tricky.

For the power devices there are relatively fast BJTs made for audio. With a suitable gate driver MOSFETs are also quite fast.

Much capacitance at the output can make the regulator react slow. For stability one likely needs some capacitance with loss (series resistance). It can help to have not more than needed for stability - more usually only makes things slow.

[exe]

The output capacitor takes over the output impedance for the higher frequency end. So one should have a fast regulator loop, so that the regulator can generate a suitable low Z even up to higher than normal frequencies.

The output stages also tend to become slower when at relatively low current. So one may consider a low current and higher current part in parallel to also get high speed at low current. It can also help to have the output work as a push pull output - getting a minimum current and being able to go down fast as well. This can also avoid working close to the limit, where anti windup can kick in.
Some kind of anti wind-up can help to speed up the transition between CC and CV modes, but this part can also be tricky.

For the power devices there are relatively fast BJTs made for audio. With a suitable gate driver MOSFETs are also quite fast.

Much capacitance at the output can make the regulator react slow. For stability one likely needs some capacitance with loss (series resistance). It can help to have not more than needed for stability - more usually only makes things slow.

Thank you very much for elaborate answer.

As for slow down at low output currents, I currently addressed it with a current sink on output. So, for 0-100mA output, I have a current sink of 20mA, this way output current changes only 20-120mA, which helps with stability. This approach is not without downsides, one is, the current sink needs to be precise, or current measurements will be off. And if there are multiple current ranges, the current sink should switch ranges too. It's still attractive because I can, say, "auto-zero" current offset. I also haven't figured out how to make CC and CV loops co-exist without oscillation (and I have piles of documents yet to read on opamp compensation and stability).

As of proposed design approaches (push-pull and parallel stages), I currently don't have skills to design a system with push-pull output stage, or with parallel stages. I'm afraid push-pull is too complicated, as it probably needs the current shunt to be on the ground. The good news is, this way it feels like it's easy to get a two-quadrant power supply. Or even four-quadrant? (I'll read AoE).

Parallel stages... do you have an example of a device that does it? If not, can you please explain how it works? Is it like two power supplies, with one (the big one) having slightly lower voltage output that will take the load if small-current supply will go in CC mode under the load?

PS BJTs I consider are TP2955/TP3055 for high-power output, BD139/BD140 for low-power output.

[Kleinstein]

For the general setup there are two ways to build a linear supply: one is with a current mode output stage, where the regulator sets the current. This usually works with an auxiliary supply for the actual regulator circuit. This version usually needs some output capacitance - though the values can be low (e.g. 100 nF per amp of max current). A push pull version is also possible and when done right with the current measuring shunt not effected by the amplifier internal cross current.  The push pull version can use the auxiliary supply for the negative side to get a limited 4 quadrant operation (limited to a few volts and lower current for the negative side).

The other version is using a more voltage controlling output stage with low output resistance, e.g. much like a class AB audio amplifier. This system does not need a true capacitor at the output. However the power stage tends to kind of simulate a capacitor. So it is not clear which version is lower "measured" capacitance. This system is kind of limited to lower voltages and less flexible.

The TIP2955/TIP3055 are more like normal slow BJTs. For fast response I would consider 2SC5200 or similar transistors made for high end audio. For a smaller version D44H / D45H could be an option.

I had plans for such a supply (floating regulator type), and the simulations looked reasonable good. Not to go too far off topic I would more like start a new thread on a fast linear regulator design. Just normal slow CC CV cross over is relatively easy with a kind of diode minimum circuit. The anti - windup is slightly more tricky, as it is nonlinear and may lead to extra local oscillations. One kind of needs to address windup because the control loop would be only conditionally stable and thus needs to handle large signal stability separate.

[bson]

Depends on what you're doing.  If you're testing designs you want as little capacitance as possible in the supply - otherwise you won't know if you have enough on your board.  Verifying your supply capacitance becomes impossible with a load of it in the bench PSU.  On the other hand, if you're a student learning and powering trivial experiments, then it's nice not to have to always provision enough capacitance to avoid excess supply ripple on transient loads.  (Especially since there's bound to be insufficient damping and decoupling to begin with.)

So, horses for courses.  Ideally there is a large capacitor and a switch to remove it from circuit, but I've never actually seen this (if I designed a supply it would have this).  That failing I prefer to have none as it's pretty easy to stick one between the terminals, while removing an internal capacitor... better buy a different supply.

[bsdphk]

"Now a practical question. What makes a power supply fast to respond? Like, I want to design one, how should I approach the problem?"

The way I have started to think about it:  Voltage Reference + D/A Converter + (weird) Amplifier stage.

[Kleinstein]

What makes a power supply fast to react is having a fast control loop and not too much extra capacitance at the output. A little bit of capacitance at the output is likely needed, but more will mainly slow down the response. Usually the main control loop is still in the analog world. The limiting factors on how fast the control can be can be things like the power transistors and also parasitic inductance: for a low aimed for output impedance inductance will become important as the time constant is L/R.

A good starting point for a PS design is a simulation. I can not include all details, but it is a good start and does not guarantee stability, but its's kind of first test to pass, as it is slow and hard to do some of the tests for the extremes in real world.

[nctnico]

A good test for any power supply is to test it with a pure current sink. For example the collector of a standard BJT (without any control loop!).