how does blackdog's PSU work?

[David Hess]

Topics such as frequency compensation or choosing the pass element are quite rarely discussed (at least I didn't find much info, yet alone measurements).

Don't forget that a power supply is just - in essence - a DC coupled power amplifier. So, texts covering power amplifiers go over much of the same ground. If you want a discussion of pass element choice that's a place to look for it. Just make sure you stick to reliable sources, otherwise you'll find yourself going down the audiophool rabbit hole.

This is especially the case because common voltage feedback operational amplifiers duplicate the topology used in audio power amplifiers with a low voltage gain differential input stage, high voltage gain level shifter, and unity gain push-pull output stage.  Remove the upper or lower half of the output stage and a DC coupled audio power amplifier becomes a regulated power supply although some power supplies keep the other half of the output stage for better control or 2 quadrant operation.

The transistors blackdog used in his design are ring or distributed emitter transistors with low Ft droop and hfe droop intended for audio power amplifiers and work great for high performance regulators.  Similar old but now inexpensive parts include the TO-220 D44/D45 transistors which were sometimes found in high performance point of load linear regulators for CPUs and low noise voltage regulator designs.

[blackdog]

Hi,

It is true what David says about the transistors that I used.
And I also tested with the D45H, and that went not wel, kaboem! 

I tested with 4x D45H, this transistor is useful with a 30V 2,5-Ampere power supply, but only if a pre-regulator is used or say transformer tabs to lower the dissipation.
The SOA characteristic is not so good of these transistors, in the Motorola datasheet you can see what the SOA is at 70C.
At 30V Vce max 0.6-Ampere..., so you need a good heatsink.
My preference is for the transistors that are in the schematic.

Kind regards,
Bram

[David Hess]

All of the usual warning about SOA (safe operating area) apply to the D44H/D45H series just like any other transistor.  In addition, the high Ft and hfe of ring emitter parts makes them more susceptible to destructive local oscillations if care is not used.  Just like with an audio power amplifier, the equivalent of a series RC Zobel network is often included from the output to ground.

The audio guys wanted larger parts like the 2SA1943 to get costs down but that is just good for us if we can repurpose them for high performance power supplies.  I tend to prefer smaller parts though because even if I have to use more of them in parallel, cooling is easier because the case to heat sink thermal resistance is distributed over a larger area and a lower total thermal resistance is possible.  On the other hand, this does make for more parts and layout complexity and an increased possibility of local oscillation if care is not used.

I have a list of the various On Semiconductor ring emitter parts intended for audio amplifiers and a quick check indicates that the D44H/D45H has about a 50% better ratio of power per dollar.  It would not surprise me however if they are also counterfeited more than the less common larger transistors.

I agree with T3sl4co1l though that these fast transistors are of marginal benefit in a general purpose bench power supply simply because it defeats the purpose of having excess high frequency performance at the end of a pair of test leads.  This does not apply for a point of load regulator and special purpose applications where the load is close.

[T3sl4co1l]

Relevant example: in a recent design for a power supply, I wrote into the specification that the supply will have a maximum equivalent inductance characteristic.  The purpose of this is twofold: 1. the customer had been testing with external power, on leads of so-and-so length, therefore I can't do any worse than what is already known to work, if I meet this; and 2. the supply will have effective inductance due to control loop roll-off and output filtering, and this spec gives me a suitable (finite, nonzero) target to design to.

Tim

[blackdog]

Hi guys,

I am aware that with a power supply without external sensing, I have little control over what happens after the wiring at the load.
Earlier I have made a comment about that.

And once again I want to point out that this power supply is for controlling precision circuits, as I indicated earlier.

Something like in the development of reference ovens, I want to have everything powerd from +15V.
So the oven heating and the reference, connected via short and twisted wires.
This happens in this way, both parts of the circuit go with separate wires to the power supply.
These parts are not electrically connected, except on the terminals of the power supply.

The heater can be a linear one, slow PWM(Arduino PWM) or a normal PWM circuit.
Normally i remove the fast edges on the pwm circuit, to get the noise down and it is easier on the power supply control loop.
And of course there is a lot of attention on decoupling.

A remark can be, that i can test with two power supplies, so i have no connection at all.
But that would not be the reality of the finished project.

Furthermore, it is not too difficult to bring out the sense wires with some extra security components to get better control at the load for short 4-wiring applications.

About the smaller power transistors, i did test this, but for 30V 5-Amps 4 of them where not enough to make it reliable.
I like the D45H D44H transistors, but nut at this power level, and again price is not a issue 

Kind regards,
Bram

[exe]

Concerning discussion "how fast bench PSU should be", I'd say I want a faster one because I use them to supply, say, esp32 which has built-in WIFI. So, with a poor power supply it will reboot while transmitting. To some extend this can be fixed with more output capacitance. Although, I didn't do any measurements, so, may be, just putting a smaller cap closer to the load could help.

Another topic interesting to me is how to parallel caps. Say, my power supply has output cap, and my load has an input cap (they may resonate). So, may be low-esr output caps is a bad idea? But I'm too far from understanding these matters.

[blackdog]

Hi,

My opinion
1e
A LAB power supply must be nice to his load, realy taking care of it. 

2e
This implies, among other things, that he has the smalest possible capacitor at the output and has a rapidly responding current limitation.

3e
No power On/Off and enable switch abberations.

4e
Low Ri and low noise.

5e
Stable with a wide range of loads.
But, the design i'am showing here is not for large inductive loads, a 2 of 4 quadrant powersupply should be used then.

6e
There is no "one size fits all" powersupply.

7e
As stable as posible with high capacitve loads, normally not a problem because the wiring resistance to the load.

Just an example of which I think the designer has not understood how to do it right.
My Rigol DP832 power supply, 2x30V at 3A and 1x 5V 3A.
On the output banana jacks there is 470uF!
That tells me that they have not had good control over the loop stability during development.
In fault situation the energie in the capacitors can kill your load...
I handle this rule of tumb, use about 50uF for every Ampere output current (for linear LAB power supply's)
Something else with this power supply, the sense wires were bundled together with the live wires, so stupid ...
A simple different twist fixt the problem, see this topic: https://www.circuitsonline.net/forum/view/130740#highlight=dp832

The better you design the loop control(phase margin), the lower the output capacitor can be. (within a certain range)

My design has two capacitors and resistors on the output so this wound not ring by itself (ofcouse i tested this :-) )
The serie resistors will almost certainly kill the Q of these components.

Kind regards,
Bram

[technogeeky]

Concerning discussion "how fast bench PSU should be", I'd say I want a faster one because I use them to supply, say, esp32 which has built-in WIFI. So, with a poor power supply it will reboot while transmitting. To some extend this can be fixed with more output capacitance. Although, I didn't do any measurements, so, may be, just putting a smaller cap closer to the load could help.

Another topic interesting to me is how to parallel caps. Say, my power supply has output cap, and my load has an input cap (they may resonate). So, may be low-esr output caps is a bad idea? But I'm too far from understanding these matters.

This is where "high speed" power supplies like the Keithley 2304 come into play. There are HP models for these as well. These are also known as "mobile communications" power supplies. There are also similar models that are known as "battery simulators".  Dave has done a teardown of one of these^[1]. The primary difference between a high speed power supply and a battery simulator is that the battery simulator has circuitry to control the output impedance to a selected value.

These power supplies usually have very little output capacitance. I am not 100% sure, but I believe in general these power supplies have a dual push-pull topology. They are essentially the positive voltage half of a SMU (source measure unit).

One way you can determine if the supply you are looking at is a high speed supply or not, even without reading the detailed specification, is looking for transient response in the specification. In the end (page 111) of the Keithley 2304a manual,  you can see that they specify the transient response time to be less than 50 (resp. 40) microseconds. This may or may not seem fast to you, but keep in mind for these high speed power supplies this is a 1000% change in load. Most supplies will only specify for a 100% change in load. The numbers listed in the manual are the worst case. In practice, I've measured  times as low as 5 to 10 microseconds for a 1000% change in load.

In any case, the overall point of these power supplies is to be able to simulate/emulate the battery or power source for a mobile communications device (anything with a transmitter, really). It can source current (normal usage) and sink current (emulating a battery being charged). In such situations, you often use a small amount of power normally until the wireless radio transmits. At this point, the current usage can increase by a huge amount, but only for a small amount of time. So the overall picture to the power supply is a small quiescent current with a pulse train superimposed on it. All of this could be done using an oscilloscope, current probe, and a battery bank.

Note that these are not linear power supplies. They have extremely good noise characteristics for switching power supplies (almost as good as linear, which is an impressive feat).


^[1]: battery simulator .

[David Hess]

I am aware that with a power supply without external sensing, I have little control over what happens after the wiring at the load.

Whether external sensing is used only matters at low frequencies.  The contradiction is designing for fast loop response to provide low output impedance at high frequencies when the inductance in the external leads prevent the control loop from even seeing the load at high frequencies.

Furthermore, it is not too difficult to bring out the sense wires with some extra security components to get better control at the load for short 4-wiring applications.

I have done this before down to the microvolt level at load currents of amps and it works fine but capacitive decoupling at the load is what lowers the impedance at high frequencies.  The high frequency performance of the power supply becomes less relevant at the leads become longer.

When using remote sense, be sure to limit the difference between the force and sense lines at the power supply with diodes or resistors to prevent the output from running away damaging the load if a sense line becomes disconnected.

Concerning discussion "how fast bench PSU should be", I'd say I want a faster one because I use them to supply, say, esp32 which has built-in WIFI. So, with a poor power supply it will reboot while transmitting. To some extend this can be fixed with more output capacitance. Although, I didn't do any measurements, so, may be, just putting a smaller cap closer to the load could help.

Usually fast closed loop designs with high frequency transistors are used to provide a low AC output impedance while using only a minimum of output capacitance improving the constant current performance.  A standard power supply might have 47uF per amp of output capacitance to provide a low AC impedance while a high frequency design can get by with less than 0.1uF and a carefully controlled ESR to preserve stability.

Another topic interesting to me is how to parallel caps. Say, my power supply has output cap, and my load has an input cap (they may resonate). So, may be low-esr output caps is a bad idea? But I'm too far from understanding these matters.

The power supply's output capacitance forms part of the frequency compensation and if it has an ESR which is too low, excessive phase shift can lower phase margin compromising transient response or even causing oscillation.  So the output capacitor may have a very controlled ESR to prevent problems.  The bypass capacitors at the load are much smaller and located remotely where the series resistance is greater so low ESR does not matter for them.

This can be seen in the application circuits for regulators like the 78xx and 79xx series where a small low ESR ceramic bypass capacitor is shown at the input if the input bulk capacitor is located remotely but the output shows *only* a small high ESR tantalum or aluminum electrolytic capacitor.  These regulators are well behaved and will not care about a 0.1uF ceramic capacitor at their output but this capacitor does nothing useful except perhaps keep RF out.  For power supply bypassing purposes, it needs to be placed at the load instead.

[blackdog]

Hi David,

I talkt about remote sensing on the Dutch forum, and explaned that if you want a fast reacting power supply, you need to build it like a HF amplifier 
Twisting high current cables, twisting your sense wires, keep the buffer capacitor close to the power stage, the output connectors close to the power transistors etc.

And of course the development of a fast power supply means that the wiring to your load is becoming increasingly important to keep the Ri low.

I did some testing with my HP 6632B with a 1-meter long connection cable with remote sensing.
Of course, capacitors were needed at the terminals for some good dynamic behavior, but you cant go to high.
There is a lot of usefull information about sense wires in the PDF of the KeySight 6632B.
Long cables together with sense wiring can almost never result in a fast power supply due to the induction of the wiring.
You cant win 

Actually, we should have a separate section on this forum, which contains all the basics for designing a power supply for various applications where we can put our joint knowledge together.
There is so much misunderstanding about the design of power supplies, look at the many compensated to dead designs on the Internet...

Kind regards,
Bram

[Kleinstein]

It depends on the application on how fast the current limit should engage. Sometimes you want a very fast current limit, essentially limited by the output capacitor and sometimes one wants a current limit that is a bit slow and can deliver more current for a few ms. In the slow case one can usually tolerate more capacitance at the output.

The simple single pole loop results in a constant output inductance for the regulator. However this inductance can be close to ideal and thus cause ringing together with a low loss load capacitor. For small caps a high loss output capacitor (e.g. the extra series resistors) helps to dampen the possible resonance, but this does not work anymore with large caps. It's only the lead resistance that can save the day. With external sense wires one may loose this resistance - so adding external sense wires can get tricky.

For the sense wires one should also consider the case of the load wires interrupted and thus delivering power over the protective diodes / resistors.

For very low output capacitance a push pull output stage with some standing current helps as this essentially the only way to provide a low output impedance even with low load current. At low current one way output stages often get slow and thus might get unstable there.

[exe]

In practice, I've measured  times as low as 5 to 10 microseconds for a 1000% change in load.

To me, that's just a matter of pre-loading . Like, the "useful" load changes from 1mA to 1000mA. So, it's like 1000x change, right? But if I put an internal load sinking 200mA, the actual change for the PSU would be just 5x. So, this let's getting away with slower power supply. At least that's what I get in a simulator.

Of course this wastes power, that's why I want this current to be settable.

But this also opens another big question to me: how PSU behavior changes with the load. I think this should also affect stability.

[blackdog]

Hi exe,

You can use my design and scale it down is you only need say 1,5 Ampere.
The output capacitor can then be scaled down also, so the peak current wil also scale down, the power supply is fast enough.

But maybe a nother design is better for you, one moment...
Look at this, it is a first setup...


The original schematic you wil find in app note 16 from LT, is uses en LT118 and a LT1010.
My version is absolute not finished!
It need extensive testing and probably the zobel network can be smaller.
The first tests look promising, verry fast acting current limiting.
The easy way for more current in the orriginal design is to use say 5x LT1010 TO220.
It is easy to parallel them, the have a internal 7 Ohm output resistor and that maks them perfect for capacitiv loads.
Read the datasheet and app notes on the LT1010.

Do not use a high voltage power supply if your requirements for only 12V output voltage.
This wil keeps the dissipation down and easyer to build.

Kind regards,
Bram

[Kleinstein]

With a simple output stage the load current can effect stability. Usually the transistors get slow at low current. There is an additional change with the Sizlaki stage. With MOSFETs the effect could also be quite significant.

So coss-over to instability usually either starts at low current because the output stage gets to slow or at high current because the output stage is fastest there.  An internal constant current load is only a partial solution, as there can be times when this current is used to discharge an output capacitor when recovering from overshoot.

There is also a difference between power supplies with a more current controlled output stage and those with an emitter-follower type. The emitter-follower type can be more sensitive to changing current. To get away from the low current problem, a push pull output stage can help.

[Zero999]

Hi Hero999,

Thanks for the warning
But i already now that, i dril out the holes around the 230V powerline connections before i do the real testing, this is only the reference section build.

Then please make that clear in future, especially as this is the beginners section.

[blackdog]

Hi Hero999,

You are correct, i dit not think of that... 
And I would like to add that if you test such a printed circuit board as I showed, you will isolate all mains connections!
Also the live connection on the bottom of the board.

When I test, I place a piece of plastic on the bottom so that I can not touch the live parts.
And on the top side i use electrical tape to isolate the live parts.

If you place a line transformer on the China boards, dril at least one row of holes around all the live connections.
Make sure that all burrs from drilling are removed on both sides and clean the board with alcohol.

So be carefull!

Kind regards,
Bram

[exe]

Dear blackdog, what's the purpose of D13 and R39 on http://www.bramcam.nl/NA/NA-01-PSU/NA-PSU-50.GIF ? As I understand, D13 protects BJT's base from being pulled too low. Is this correct?

R39 remove leakage currents from the base? So, I expect some current flowing through it. Can I connect it before the shunt resistor so this current is accounted? (although, I guess, this won't make a big difference).

[blackdog]

Hi exe, 

You are correct, for the diode and the resistor!
It is a good tip.

I will place R39 in front of the 0.1 Ohm current sense resistance to lower the minimum current.

Kind regards,
Blackdog

[AG7CK]

Quote: "The Harrison design i use (from the late 1950's) is really beautiful."

Agree. Harrison manuals should be obligatory reading for anyone who is "building my own power supply".

I find it amazing that none of the posters here seem to recognize that the topology you use is the hallmark of HP power supplies from the Harrison period until "modern" Agilent power supplies (and probably also Keysight-branded smaller bench supplies up until today [no schematics]).

Anyone wanting to understand your "version" should start by reading Harrison Laboratories / HP Harrison / HP / Agilent manuals with theory of operation and schematics. There are dozens of them on the web. It hardly matters where you start because they are almost all based on the same theme: floating reference, dedicated bias supply and a current controlled linear power section.

Those that are historically inclined can start here http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1962-07.pdf .

[exe]

Thanks a lot, haven't read it yet, but looks like exactly what I need.