Switching Lab supply design

[Alakis]

Hello,

I am in the process of designing a programmable 0-24V, 0-3A lab supply. I am planning to drive it from an existing 12-48V DC supply. I would like it to be a small form factor so I am thinking about a switch mode topology. I am a little bit confused on how to pick the inductor sizing for such a wide parameter range. Are there any simple to understand reference designs I could take inspiration from? Ideally they would have some sort of explanation attached to them as well.

What I have planned so far:

1) Input buck-boost converter taking input voltage and converting it somewhere near 24V (maybe a little more to adjust for possible drop out)
2) CV part of the circuit, which I still haven't worked out well. I would take a reference voltage as input to convert the 24V to some desired level.
3) CC part of the circuit, similar to the CV part, just controlling the max current.
4) Current and voltage feedback would be done by some simple opamp circuits. For that, I am planning to add a split rail SMPS circuit. (something like here: https://www.digikey.at/de/articles/design-tips-for-generating-split-rail-power-supplies)
5) I am planning on using an INA219 or some other IC to read the output voltage and current, this would be separate from the CV and CC mode monitoring, as I don't want to rely on MCU response time for controlling the output.
6) MCU to communicate with PC via USB, display output values and control reference voltage DACs.

I am mostly confused and need guidance on the CC and CV parts. As I said before, switch mode component values change a lot depending on the output voltage and current. The reference designs always seem to be some crappy LM317 circuits, which work better as heaters than power supplies I would like to design something more efficient and without giant heatsinks.


Thank you in advance!

Edit:
Update on the project.

I've decided to add a pre-tracking regulator switching supply that supplies a voltage just above the target output voltage (similar schematic to this TI guide:https://www.ti.com/lit/ta/sszt535/sszt535.pdf?ts=1707321871821&ref_url=https%253A%252F%252Fwww.ti.com%252Fdocument-viewer%252Flit%252Fhtml%252FSSZT535 ). That way the heating on the linear stage in minimal and I still get all the benefits of a linear regulator output.

The way the block diagram is now there are plenty of examples to achieve the functionality I want. I will update this post with schematics as I progress. But for now the general topology seems clear. Thank you all for the ideas!

P.S. The supply is meant to be used inside an FCT test fixture, thus I am using MCU control to set the output parameters. I need to pay extra attention to the limited space and the temperature build-up inside the test rig.

[jonpaul]

Lab CV/CC should be linear NOT SMPS.

Many app notes and existing designs in manual see HP, Power Designs, Lambda lab PSU.

We like the vintage HP metered supplies.

A dual ot triple is best for bench use as many designs may need for instance +5, +15, -15.

Jon

[Alakis]

The examples I found include a bunch of transistors at the output for CV/CC. That does not seem like the optimal solution when optimizing for space, even more so when you take into account the big heatsink. I was wondering about the possibility of dynamically adjusting the buck-boost voltage to be close to the output voltage to improve efficiency. But then I run into the same issue of adjusting the switching part of the circuit over a wide voltage range. Also there is the issue of adjusting both the switching and the linear part in sync as to avoid magic smoke. Changing one before the other creates a big power loss on the linear regulator. I am wondering if there are any industry standard solutions to this problem?

[pqass]

See attached HP E364xA schematic (page 209 from here; see link for high resolution image).
It uses a combination of transformer tap switching via SCRs/opto-TRIACs and PWM-ing (Q508 and center-top circuit) the main bulk caps (C173,C174; 2*12000uF) prior to the linear pass FET (Q109, center-right parts).
It's not an issue if the linear pass FET (Q109) sees a large input-output difference if it only lasts for a fraction of second.

[dobsonr741]

Take a look at CN0508 from Analog. https://www.digikey.com/en/product-highlight/a/analog-devices/cn0508-benchtop-power-supply

You can buy or build yourself. Exactly your specs.

[jbb]

I suggest you have a look at the EEZ BB3 supplies; they have a switch mode pre regulator and then a linear constant voltage / constant current stage.

https://www.envox.eu/dc-power-modules/dcp405-power-module/

[JohanH]

Take a look at CN0508 from Analog. https://www.digikey.com/en/product-highlight/a/analog-devices/cn0508-benchtop-power-supply

You can buy or build yourself. Exactly your specs.

So they made a variant with digital control. Here is their older analog version:

https://www.analog.com/en/resources/evaluation-hardware-and-software/evaluation-boards-kits/dc2132a.html#eb-overview

It has been discussed on this forum (search for DC2132A).

[selcuk]

I built a lab supply for myself years ago. But It was linear to eliminate switching noise from the supply voltage. So if you think to build a linear one there was an article on Elektor magazine named "Digital Benchtop Power Supply" on 2001. You can search it on the Internet. It was built around LT1491 opamp and that was controlled by a MCU. I replaced the MCU with the one I usually used, wrote the firmware and made it.

[Kleinstein]

A switched mode regulator is limited in speed and this makes it difficult to get a fast regulation. For a lab supply it is thus still common to use linear regulation or a mix with SPS preregulator and a final linear stage.

The linear design is the easier way, though it need large heat sinks. I would suggest starting with a linear version first, possibly with a reduced power.

[jonpaul]

OP: Your application eg an RF TX will be less susceptible to SMPS noise CM, DM noise etc than a high end audio power amp.

We found lab PSU must be linear except for VERY high power, and seldom is more that 30V 1A needed.

Finally the trend to all digital and software control is slow, clumsy and much less practical than a simple knob control and analog meters.

Just my observations,

Jon

[David Hess]

Supporting a higher or lower input voltage adds complexity.  Methods that come to mind:

1. 4 transistor synchronous buck-boost converter, like you suggest.
2. Push-pull or bridge synchronous buck converter with a step-up transformer.
3. Boost converter followed by synchronous buck converter.

Take a look at CN0508 from Analog. https://www.digikey.com/en/product-highlight/a/analog-devices/cn0508-benchtop-power-supply

You can buy or build yourself. Exactly your specs.

OP also wants to support a 12 to 48 volt input range for an output of 0 to 24 volts.  The power supply from Analog Devices only operates as a buck converter.

[dobsonr741]

I am nostalgic for liner power supplies. However, I realized the boards I bring up will almost always have a local regulator. The realities, even for audio, is that the mains power insulation will be done by a switching regulator.

If an extreme low noise metrology situation happens, it is battery powered.

So, on my bench I have an Extech 382260 and small PTN78000 based negative rail converter, if I need a split rail and the target board does not generate it itself.

[T3sl4co1l]

SMPS can be quite nimble; the fundamental limitation is the output filter network.

I don't see it's any problem for a "lab supply", as most of those have some big stinking 470uF or whatever on the output (seemingly, whether the control circuit / amplifier needs it or not), and that sets the transient short-circuit current (CV to CC transition), or recovery time / slew rate (CC to CV).

If the total capacitance of the filter is less than a typical linear unit's, then it's all up to the controls.

The impedance also matters:

Suppose the SMPS is a buck type with fixed duty cycle (freeze the control loop, for the thought experiment).  Suppose it's also in forced CCM (synchronous inverter).  The LC filter hanging off the inverter, has a Thevenin short-circuit at its input port (the inverter has very low impedance), the inverter ripple can be ignored for purposes of output dynamics (superposition applies), and the inverter can be seen as a Thevenin DC source proportional to inverter supply voltage times duty cycle.  (You would almost always have inverter switch Rds(on) << filter Zo, so that the short-circuit condition applies.)

Given this setup: the series inductor is shorted to AC ground on one side, and the shunt capacitor is already to ground, therefore we draw an output-referred equivalent circuit, where the filter is a parallel resonant tank to ground, with characteristic impedance Zo = sqrt(L/C) and resonant frequency Fo = 1 / (2 pi sqrt(LC)).

If we apply a step current load to this tank, we get a corresponding ringing waveform, with peak voltage Vpk = Ipk Zo.  The peak appears 1/4 wave or 1/(4 Fo) after the step, and repeats (+/-) thereafter.

Preferably, we would dampen this resonance so that repetitive current pulses, or generally frequencies near Fo, do not cause resonant gain.  The capacitor should probably be electrolytic with ESR = Zo, in which case the peak voltage increases as there is an immediate voltage step in response to the current step, Vstep = Istep * ESR, which adds with the ringing waveform.  (The Vstep decays exponentially as it's shunted by the inductor.)  We can also use an ideal C, and shunt it with a "bulk C" with some loss; the resulting L + (C1 || (R+C2)) circuit can have a Q factor below 1 when C2 > 3 C1 or so, and R = Zo.

Notice what's happened through this process: for an input step current, some maximum peak voltage change results.  This describes the transient impedance of the power supply, the self-regulation (change in Vo from change in Io).  (As opposed to source regulation, which would be Vo vs. change in Vin or Iin.  Or as opposed to Vo vs. set point, which is the gain or transfer function of the control.  It's important to understand there are at least these three variables determining actual instantaneous Vo, and a power supply must take account of all of them.)

If we want a power supply that's maximally stable (output voltage/current ripple being the "least surprising" at any given setting), given some range of CV and CC operation say zero to max in each parameter, then we should simply set Zo = V_CV / I_CC.

That only leaves one free parameter, the Fo of the filter, which is determined by Fsw and acceptable output ripple, and the control loop response.  (Control loop dominant pole must be much less than Fo for a voltage-mode type control, but current mode can in fact be above.  I would most likely choose average current mode control for a project like this.)

As for ripple and noise, high-frequency noise is easily handled by filtering and shielding.  I say easily, of course it's easy for me to say so, but to be perfectly transparent, it takes many years to develop a complete understanding of how to do this, and this goes beyond the scope of this thread/post.  More to say that -- a solution exists, and it can be made arbitrarily good, given no other restrictions (like size or cost).

Ripple is the harder case, as the lowest-frequency content (Fsw or potentially subharmonics) directly conflicts with the goal of control loop bandwidth (in V or I, or any combination thereof, say if you want to mimic some Zo, not just CC or CV -- which really are just the special cases Zo --> 0 or infty).  The same is true of a linear supply's control of course, but which is limited by device characteristics (pass transistor and driver speed?) rather than a network hanging off it.

Ripple can be handled in many ways.  For example, a phase interleaved converter can cancel it out, so that each stage is running at Fsw yet Fsw is nulled in the output, with the dominant peak pushed to Fsw*N (for N stages interleaved).  A linear stage can be employed in series or shunt, to stabilize the output voltage, current or impedance, in ways that a passive network cannot do; including feedforward from Fsw to null the ripple.  A post-reg stage can be used to optimize transient response, allowing fast current or voltage limiting for example.

Tim

[nctnico]

SMPS can be quite nimble; the fundamental limitation is the output filter network.

I don't see it's any problem for a "lab supply", as most of those have some big stinking 470uF or whatever on the output (seemingly, whether the control circuit / amplifier needs it or not), and that sets the transient short-circuit current (CV to CC transition), or recovery time / slew rate (CC to CV).

There are two problems: first one is that switchers typically push noise into the outputs which is what you don't want when measuring a circuit. And no, this is not easy to filter. Been there, done that (especially where it comes to switchers which are fed by mains as you typically get noise through the grounding). Secondly the transient behaviour may not be very good. The switching lab PSUs from Delta from the late 90's where horrible in that respect.

[T3sl4co1l]

There are two problems: first one is that switchers typically push noise into the outputs which is what you don't want when measuring a circuit. And no, this is not easy to filter. Been there, done that (especially where it comes to switchers which are fed by mains as you typically get noise through the grounding). Secondly the transient behaviour may not be very good. The switching lab PSUs from Delta from the late 90's where horrible in that respect.

Thanks, I appreciate how you read my post in its complete splendor.

Tim

[moffy]

SMPS can be quite nimble; the fundamental limitation is the output filter network.

I don't see it's any problem for a "lab supply", as most of those have some big stinking 470uF or whatever on the output (seemingly, whether the control circuit / amplifier needs it or not), and that sets the transient short-circuit current (CV to CC transition), or recovery time / slew rate (CC to CV).

There are two problems: first one is that switchers typically push noise into the outputs which is what you don't want when measuring a circuit. And no, this is not easy to filter. Been there, done that (especially where it comes to switchers which are fed by mains as you typically get noise through the grounding). Secondly the transient behaviour may not be very good. The switching lab PSUs from Delta from the late 90's where horrible in that respect.

There is a middle ground of resonant power supplies. Switchers are hard to filter if they have very fast edges, which also produce large current spikes, the noise gets everywhere and is very hard to filter. The resonant supplies, of which T3sl4col1 has a lovely discrete version, of course get around that. Not being that experienced with them I am not sure if they have the output range of a standard PWM supply but maybe it could be followed by a linear regulator for greater control and lower noise, using the resonant switcher as a preregulator.

P.S. I am really not sure if resonant supplies are suitable for variable rather than fixed only supplies.

[David Hess]

There is a middle ground of resonant power supplies. Switchers are hard to filter if they have very fast edges, which also produce large current spikes, the noise gets everywhere and is very hard to filter.

Linear Technology has some switching regulator controllers intended for low noise power supplies that have adjustable voltage and current slew rate limits.

[T3sl4co1l]

Not even resonant; though that helps.  The key is to keep the high frequencies boxed away, with the only connections penetrating it with feedthru filters.

In a relatively open design like your typical commercial PSU, this can be challenging; it's no accident they measure output ripple in 20MHz BW with bypass, and measure EMI with output mostly floating.  Even if the output isn't a problem by itself, higher-frequency content can backfeed up the line, and, some CM-DM mode conversion later, you've got spikes on the oscilloscope probe or whatever.  That's the usual way problems arise, or of course anything depending on that, like aliased impulse in an acquisition system, DC offsets via RF rectification, spurious tones in a radio receiver, you name it.

But all the same, you can stick that inside a box, add feedthrus to all connections (all, that's mains, output, and any aux power or status or control signals you're using!), and one wire can be common to the box if it be acceptable (usually input earth or output ground, often both).

The main thing is, it's a bother to do, and when compact or cheap design is required, it's necessary to eliminate the noise at the source, and use not just a resonant converter but perhaps a particularly-soft-switching one.

Elephant in the room: so much test equipment has SMPS.  Both my scope and spec do.  I built a couple SW-band radio projects with an SMPS.  An electric vehicle is a portable SMPS especiale, and still manages to receive broadcast radio perfectly fine.  All of these examples are guarded jealously by solid metal boxes, plus a little attention in design and testing.

Tim

[jonpaul]

Further to the issues re SMPS noise, resonant PSU, we have some experience:

In general soft switching topologies have much less noise and EMI than hard switching.

Higher freq trend means smaller L/C filters and less impact on the suseptible devices.

Control methods must be choosen in light of the rquried regulation eg a 0-100% contro of Vout or I out will be impossible for some menthoids and switch topologies.

All of this and the modest Pout of most lab power supplies means they are usually linear.

See the many papers and seminars on soft switching, resonant mode and high frequency switching power.

The SMPS><Linear hybrid is more  complex  but still used in most FIXED outot supplies eg for instruments, scopes, some PCs.

Any engineer always seeks the simplest solutuion to a design problem.

The obvious choice is linear for 0-100% control and CC/CV lab supply.

Just my experience,

Jon

[coromonadalix]

and you double thread  ... not nice

[TimNJ]

I guess you also need to define what is “good enough” for “lab” use.

Obviously there are switch-mode supplies out there which serve a lot of people very well, but may not work as well for the most sensitive applications/experiments.

If you’re planning on designing super low noise stuff, perhaps battery powered, then a switch-mode supply could perhaps introduce issues. Otherwise, everything runs on switch-mode adapters anyway, so it’s kind of more realistic? Of course, noise, ripple, EMI, etc. will not be the same between different supplies, but may be better than designing and testing in an unrealistic “silo”. But, that’s just one perspective; others may want their test equipment to minimize the variables in a particular test setup, as to not suspect the test equipment for some issue detected in their design.

[Simon]

So you want to do a lab supply and you are stuck on how to select the inductor. I would suggest you progress in steps. First design a SMPS then think about current limiting, then think about what you will do with that inductor value if you want a wide voltage and current range, look carefully at the low current scenario, that is what will blow your circuit up not the high current scenario, for that you just use parts that are chunky enough.

If you search for switch mode power supply calculator and TI you should come upon a spreadsheet that gives you a good start for calculations. Don't just use it to get values, play with it, see the effect of changing frequencies, voltages, currents. If you want low noise then you want to either stay out of discontinuous conduction or you need to work out how to filter the resulting noise out.

Get LTSpice and use a voltage source outputting a square wave voltage to approximate your power switch and rectification diode or synchronous rectification transistor, now that you have abstracted the practicalities of generating PWM power start playing with inductors, capacitors and different values of load. Make sure you tweak about the parasitic's of the output capacitor and the inductor and learn the effects they cause. You will see twice your input voltage appear at the output at startup, understand why and think about how to deal with that if you are making your own control system. If you use a chip then understand that actually there is a lot of clever stuff going on that keeps you out of trouble most of the time.

[ejeffrey]

There are two problems: first one is that switchers typically push noise into the outputs which is what you don't want when measuring a circuit. And no, this is not easy to filter. Been there, done that (especially where it comes to switchers which are fed by mains as you typically get noise through the grounding). Secondly the transient behaviour may not be very good. The switching lab PSUs from Delta from the late 90's where horrible in that respect.

"May" is the key word there.  It's down to design choices.  The VRM on your motherboard that supplies the CPU Vcore is a switching converter and it has load transient response that will run circles around your bench supply regardless of its design.  Load transient response is one of the key design constraints of a VRM and they do it very well.  They do it by using high switching frequency, many interleaved phases, and lots of capacitors.

For a general purpose bench supply you need to balance a lot of factors, and it's definitely easier to just use a linear supply but it's not true that switching can't have good transient response or low noise.

[Zero999]

Preferably, we would dampen this resonance so that repetitive current pulses, or generally frequencies near Fo, do not cause resonant gain.  The capacitor should probably be electrolytic with ESR = Zo, in which case the peak voltage increases as there is an immediate voltage step in response to the current step, Vstep = Istep * ESR, which adds with the ringing waveform.  (The Vstep decays exponentially as it's shunted by the inductor.)  We can also use an ideal C, and shunt it with a "bulk C" with some loss; the resulting L + (C1 || (R+C2)) circuit can have a Q factor below 1 when C2 > 3 C1 or so, and R = Zo.

Just one question: how do you design the filter so it can drive a load with a very low AC impedance and remain stable? The user might connect it to something with big fat low ESR capacitors, via thick, short, low impedance cables. Hoc can you be sure this won't mess it up?

[T3sl4co1l]

Preferably, we would dampen this resonance so that repetitive current pulses, or generally frequencies near Fo, do not cause resonant gain.  The capacitor should probably be electrolytic with ESR = Zo, in which case the peak voltage increases as there is an immediate voltage step in response to the current step, Vstep = Istep * ESR, which adds with the ringing waveform.  (The Vstep decays exponentially as it's shunted by the inductor.)  We can also use an ideal C, and shunt it with a "bulk C" with some loss; the resulting L + (C1 || (R+C2)) circuit can have a Q factor below 1 when C2 > 3 C1 or so, and R = Zo.

Just one question: how do you design the filter so it can drive a load with a very low AC impedance and remain stable? The user might connect it to something with big fat low ESR capacitors, via thick, short, low impedance cables. Hoc can you be sure this won't mess it up?

A couple of things:
1. Those cables (or bus bars for that matter) have ESL.  They might be thick, they might be short, but that's still a few 100 nH from say 10s of cm.  Not much, but at low impedances (we're talking 10s mΩ ESR in electrolytics say, and single or less mΩ for ceramics in quantity), that does the job.  So, the ESR can still provide damping value even with very low impedance or shorted loads.
2. The control is designed for critical to over damping, so that the output impedance is flat and low, and doesn't resonate with reactive loads.
3. Under short circuit conditions, current mode control (it's current mode, right--?) simply controls inductor current into the load regardless of its impedance, simple enough.
4. And, you can design a filter network for additional filtering with OC/SC damping, either by RC damping an intermediate node (the middle C in a CLCLC filter), or damping the L (probably not with ESR because of DC losses, but L || R or L || (R+L) can be used).  Or making a constant-resistance filter.

Obviously, this varies with scheme, but resonant schemes can be operated in a current-mode way (not necessarily that you'll see any random controller offering it..), where "current mode" most likely relates to the magnitude of current during a cycle moreso than instantaneous or cycle-averaged current as for a square-pulse converter.  The current out the rectifier then is just whatever's pushed into the transformer or through the network, and so output DC current corresponds.

Tim