The State of Regulators for Linear Power Supplies - A Rant (!)

[robert67]

Hi, since this will be kind of a rant, I put this into the general chat ...

I've build my first linear power supply about 35 years ago with a LM317 (adjustable voltage only). And a few weeks ago I gutted an old, defective LM723 based power supply (a 1983 EA-3013). So I thought it was high time to build another one (having now an empty case with a big transformer in it). Nothing fancy though: 0-30V / 0-5A (the specs of the one I just gutted). Now I searched the internet high and low for a ...

• good enough (don't care if it regulates down to a few 100mV / a few dozen mA only, about a little ripple on the output, or if it goes down a few dozen mV when the load goes from 0A to 5A)
• simple (meaning low component/pin count)
• one chip solution (yes, you will need in addition one or two power transistors to handle the 5A)

And what did I found:

• "LM723 still does its job" (yes, with an additional Op-Amp for the current limiting, and a driver transistor for the power transistors)
• "I throw in the L200" (nice, but I still need an additional Op-Amp for the current limiting)
  
• "LM614" (4 Op-Amps and a adjustable voltage reference - nice, but why aren't the Op-Amps rail-to-rail, why? And yes, still needs a driver for the power transistors. And I've to figure out all the caps for compensation!)
• "LM10, LM432, TSM101, TSM106, TSM107,  TSM103, NCP4300, TL103 ..." (basically two Op-Amps with a voltage reference, and the Op-Amps and voltage references internally somehow wired - why bother!)

So my conclusion so far is I could go with the LM317 I've used 35 years ago, throw in an Op-Amp for the current limiting, and be happy (there is even a reference design for that).

Any thoughts?

[Messtechniker]

So my conclusion so far is I could go with the LM317 I've used 35 years ago, throw in an Op-Amp for the current limiting, and be happy (there is even a reference design for that).

Any thoughts?

Yep. Simply the way to go even today when wanting to keep things simple.

[The Soulman]

Yes linear regulators haven't changed much in the last years, luckily there are plenty of options from the old days still in production.
The reason why IC manufacturers don't develop new linear regulators is simple: there is no demand.
The name of the game to day is low cost, for any traditional power supply the mayor costs are the transformer
and the large heat-sink.
The arrival of (cheap) fast mosfets and cheap controllers meant that today switch-mode power supply's can be cheaply made omitting the transformer and large heat-sink.

How cheap? Look at:

https://www.aliexpress.com/wholesale?ltype=wholesale&d=y&origin=y&isViewCP=y&catId=0&initiative_id=SB_20170529101729&SearchText=power+supply+module&blanktest=0&tc=af

[NiHaoMike]

I thought the 723 had built in current limiting?

[David Hess]

I thought the 723 had built in current limiting?

The 723 has built in non-adjustable peak current limiting.  You can set the current limit by selecting the shunt resistance but it is intended for fault protection and not an adjustable current output.

[robert67]

Yep. Simply the way to go even today when wanting to keep things simple.

Thanks. I was feeling like I'm getting to old 

But maybe I'll go 21st century at least and use the L200 (introduced around 2000?). I like the L200 reference design - using the additional Op Amp as a variable gain amplifier for the voltage drop across the shunt resistor - better than the LM317 one - using the OP Amp as a comparator with the low side of both input voltage dividers floating around depending on the voltage settings. Though the LM317 reference design gives you a "current limit" LED for free.

[robert67]

Yes linear regulators haven't changed much in the last years, luckily there are plenty of options from the old days still in production.
The reason why IC manufacturers don't develop new linear regulators is simple: there is no demand.
The name of the game to day is low cost, for any traditional power supply the mayor costs are the transformer
and the large heat-sink.
The arrival of (cheap) fast mosfets and cheap controllers meant that today switch-mode power supply's can be cheaply made omitting the transformer and large heat-sink.

How cheap? Look at:

https://www.aliexpress.com/wholesale?ltype=wholesale&d=y&origin=y&isViewCP=y&catId=0&initiative_id=SB_20170529101729&SearchText=power+supply+module&blanktest=0&tc=af

Yeah, nobody like heavy bulky power supplies any more. But they have so many uses:

• You can use them to flatten some crumbled schematics (which you spilled coffee on)
• You can use them to hold a door open when there is a draft
• And there is nothing more soothing than the 50Hz (in some regions of this planet 60Hz) hum when you max out a transformer

[robert67]

I thought the 723 had built in current limiting?

The 723 has built in non-adjustable peak current limiting.  You can set the current limit by selecting the shunt resistance but it is intended for fault protection and not an adjustable current output.

Yes, just using a shunt resistor and the internal current limiting is "limited". It relies on the voltage dropped across the shunt resistor exceeding the base-emitter voltage of a transistor, which drifts between 670mV at 25°C and 450mV at 150°C. And that's not "good enough" for current regulation.

[Ian.M]

I posted a LTspice sim of the LM723 based Philips PE 1535-00 bench PSU earlier this month.
https://www.eevblog.com/forum/projects/how-do-make-a-current-limiting-knob-using-lm723-for-linear-reg-power-supplies/msg1212923/#msg1212923
It does have fully adjustable current limiting but its pretty crude.  However it might be a good starting point for figuring out what's actually needed to add on adjustable precision current limiting by pulling down the COMP pin ,without running into major instability issues. 

[Kleinstein]

The typical linear lab supply circuit is more or less based on 2 OPs, a reference and than a power stage. So there is not that much to gain making a special chip with the 2 OPs and the reference. Lab supplies also are made for different power ranges, thus a special chip for all sizes would be kind of difficult. This is especially true for modern version with a SMPS part before the linear regulator.

The L200 is way older than from 2000, more like 1970, not much newer than the LM317.

The LM317 is made as a voltage regulator for an internal supply to a given load. So it is not a good choice to build a bench top supply of lab supply - it was never made for this. A lab supply needs a different kind of frequency compensation, more tolerant to a difficult load. Especially adding a kind of after-though current limiting is a good way to build a poor lab supply.

It is a good thing to have the control part and power part separated - so don't complain about an extra external driver needed in many LM723 designs.

[jolshefsky]

So my conclusion so far is I could go with the LM317 I've used 35 years ago, throw in an Op-Amp for the current limiting, and be happy (there is even a reference design for that).

Any thoughts?

Thank your lucky stars that the you can use a reliable old design without being mocked that it doesn't have an internal lithium battery for round-the-clock WiFi "Internet of Things" connectivity.

[schmitt trigger]

In the past, I also built several LM723 power supplies, including a switchmode one (which appeared in an Unitrode app note).
Thus, I'm also a bit nostalgic about it. Actually, quite nostalgic. 

The 723 is similar to the 555 in which its ultra-simple architecture allows to be understood easily by experienced and beginner users alike. And which allows the circuit to be applied in many different ways..
That is the reason that both are still being used, and a Google search will turn up +200k hits.

The major issue it has is that it has no thermal protection for the output transistors.

Of course, being a linear regulator, it is also bulky and inefficient. And as other posters have correctly indicated, nowadays a complete SMPS will be cheaper, smaller and far more efficient.

[robert67]

The L200 is way older than from 2000, more like 1970, not much newer than the LM317.

Thanks, I didn't know that. The ST datasheet is of January 2000 without mentioning previous revisions - so I just assumed it's from2000.

The LM317 is made as a voltage regulator for an internal supply to a given load. So it is not a good choice to build a bench top supply of lab supply - it was never made for this. A lab supply needs a different kind of frequency compensation, more tolerant to a difficult load. Especially adding a kind of after-though current limiting is a good way to build a poor lab supply.

I read that statement several times on different boards. What I don't understand is why there is a "Constant Voltage/Constant Current Regulator" reference design for the LM317 (for the LM150 too) in the datasheets then.

It is a good thing to have the control part and power part separated - so don't complain about an extra external driver needed in many LM723 designs.

Oh, but I'll keep complaining  You see, my design goal is a "simple" (low pin/component count) and "good enough" (see start of the thread) power supply. So I take advantage of any product from any vendor that can lower my pin/component count, and I will accept a certain loss in quality for that. And I will continue to throw a fit until TI, ST or whoever announces a chip that only needs two pots attached to build my power supply 

[robert67]

Thank your lucky stars that the you can use a reliable old design without being mocked that it doesn't have an internal lithium battery for round-the-clock WiFi "Internet of Things" connectivity.

And I do, I definitively do, several times a day! I fear the time will come when I can no longer switch my soldering station on, when the internet is down. Or solder parts that are not in the (of course cloud based) database of the soldering station. Or I need different soldering stations for parts from different manufactures, because not every soldering station vendor has an agreement with every parts vendor. The world is evil     

[David Hess]

The LM317 is made as a voltage regulator for an internal supply to a given load. So it is not a good choice to build a bench top supply of lab supply - it was never made for this. A lab supply needs a different kind of frequency compensation, more tolerant to a difficult load. Especially adding a kind of after-though current limiting is a good way to build a poor lab supply.

I read that statement several times on different boards. What I don't understand is why there is a "Constant Voltage/Constant Current Regulator" reference design for the LM317 (for the LM150 too) in the datasheets then.

I agree with Kleinstein and others that integrated regulators are just not designed for lab power supply applications.  There is a market there but it is occupied by discrete designs and except for extra simplicity which is not required, there is no advantage to a more integrated design and some disadvantages.  Lab power supplies are just not all that complicated.

Where I disagree is that integrated regulators do have a part in simplifying a lab power supply design.  An LM317 or similar is an easy way to drive the output stage and provide SOA protection, thermal protection, and short circuit current limiting.  An LM395 would be even better for this but they cost a lot more.

A design using an LM317 like that is not going to be quite as high performance (it will need more output capacitance) but I consider the trade off worth it.

It is a good thing to have the control part and power part separated - so don't complain about an extra external driver needed in many LM723 designs.

Oh, but I'll keep complaining  You see, my design goal is a "simple" (low pin/component count) and "good enough" (see start of the thread) power supply. So I take advantage of any product from any vendor that can lower my pin/component count, and I will accept a certain loss in quality for that. And I will continue to throw a fit until TI, ST or whoever announces a chip that only needs two pots attached to build my power supply 

There are some parts like that but they will require external power pass elements to meet your voltage and current requirements; the LT3081 comes close.

One problem with a fully integrated implementation is that the performance of the reference and error amplifier are limited by temperature swings from the power element.

[floobydust]

I've used LM317, LT338A for years in bench power supplies.
But now I need adjustable current-limiting, and less than 1.25V min. output.

Looking around, lazy me purchased the $6 eBay 0-30V 2mA-3A PSU kit and found it somewhat sucks, as this guy's blog mentions.

I am left to designing from scratch, using a few op-amps and overkill on the pass transistors to avoid them failing.
But I need reliable, low-drift, low-noise PSU. Other linear bench supplies are copies of 1970's designs (hail LM741!) with a few bugs.
Modern MCU controlled like Rigol 832 aren't good enough, they have so much overshoot and EMI I can't use them for low-level analog design. But fancy LCD.

[David Hess]

There is not much performance improvement to be had by just using an operational amplifier better than a 741 in a high current power supply so if you want precision and low noise, I would use a separate low and high current design if your application will allow it.  Other factors like reference quality (1/f noise and drift), feedback network temperature coefficient (especially in current shunts), and the temperature coefficient and resistance of copper in the output leads need to be considered.

Further within the bandwidth of the control loop, the low frequency output noise of an integrated regulator used as a pass element will be reduced by the control loop.  High frequency noise is filtered by the output capacitance.  So at least where it counts, a low noise output stage by itself does not deliver the improvement of noise implied by just the difference in output stage noise.

The schematic below gives a general idea of what is required for high stability.  Kelvin sensing, a separate low noise and low drift reference, and a precision operational amplifier are combined to control what would otherwise be a noisy and drifty integrated regulator.  I have used this same circuit to build high current blow out proof fixed power supplies with microvolt stability under line and load changes.

In a general purpose power supply, the reference and amplifier would be powered by a separate supply instead of bootstrapped from the output but the single point common is still required for precision.  Add a current shunt to the output and a separate single operational amplifier control loop for current limiting and you have duplicated some of those old power supply designs from the 1970s but with more precision.  They had better operational amplifiers than the 741 like the 308 (and lots of others) and could have included the design features I mentioned above but did not bother because the added precision is not generally needed in a high current power supply.

[robert67]

There are some parts like that but they will require external power pass elements to meet your voltage and current requirements; the LT3081 comes close.

Thanks for the LT3081 hint.

I had a look at the datasheet, and it (as well as the LT3086) come really close to what I want (simple and good enough). Two pots as well as some resistors/capacitors and you're done. You don't even need an external shunt resistor for the current regulation. The only drawback of these is that you can't beef them up using a power transistor (at least I couldn't figure out how), instead they are designed to be used in parallel if you need more current.

They even have demo board for a real good lab power supply using the LT3081 (http://cds.linear.com/docs/en/lt-journal/LTJournal-V24N2-02-df-BenchSupply-Szolusha.pdf). Switch mode pre-regulator and all. But of course they override all internal current sources with external precision ones to avoid temperature drift.

[BravoV]

They even have demo board for a real good lab power supply using the LT3081 (http://cds.linear.com/docs/en/lt-journal/LTJournal-V24N2-02-df-BenchSupply-Szolusha.pdf). Switch mode pre-regulator and all. But of course they override all internal current sources with external precision ones to avoid temperature drift.

We have a review posted by fellow member DutchGert, check this thread ...

  -> Linear Technology DC2132A CV&CC Adj. Bench Power Supply Board Review

[ZeTeX]

Here is a thread about LT3081 PSU:
http://www.diyaudio.com/forums/equipment-tools/263471-triple-outputs-160w-lab-psu-epsux3-version-2-a.html
I'm having a hard time setting the current limit digitally, if you can have a go at LTspice and see if you can get the current limit to controlled digitally.

[Zero999]

Here is a thread about LT3081 PSU:
http://www.diyaudio.com/forums/equipment-tools/263471-triple-outputs-160w-lab-psu-epsux3-version-2-a.html
I'm having a hard time setting the current limit digitally, if you can have a go at LTspice and see if you can get the current limit to controlled digitally.

I don't see the issue. Use an error amplifier with its reference set by a DAC, to look at the current on the monitoring pin and pull the set pin down, if it exceeds the reference.

For me, one of the most annoying things with the LT3081 is the requirement for a minimum load current. If the adjustment range is small, then a resistor can be used but this is not feasible in an application which can work down to 0V, as the resistor would need to be close to zero and would draw an enormous current at higher voltage settings. The only solution is a current sink. To really get down to 0V you need a negative power supply but a simple current mirror can get close to 0V without one.

[ZeTeX]

Here is a thread about LT3081 PSU:
http://www.diyaudio.com/forums/equipment-tools/263471-triple-outputs-160w-lab-psu-epsux3-version-2-a.html
I'm having a hard time setting the current limit digitally if you can have a go at LTspice and see if you can get the current limit to controlled digitally.

I don't see the issue. Use an error amplifier with its reference set by a DAC, to look at the current on the monitoring pin and pull the set pin down, if it exceeds the reference.

For me, one of the most annoying things with the LT3081 is the requirement for a minimum load current. If the adjustment range is small, then a resistor can be used but this is not feasible in an application which can work down to 0V, as the resistor would need to be close to zero and would draw an enormous current at higher voltage settings. The only solution is a current sink. To really get down to 0V you need a negative power supply but a simple current mirror can get close to 0V without one.

And get oscillations and instability, you are adding another external op amp that controls the current, another op amp that you will have to compensate and basically the same as taking LM317 and adding current limiting. Not an option.
"Frex" from the other forum found a solution, but I don't understand how it works.
EDIT - by looking at the schematic he seemed like he is having a current source into the Ilimit pin, so I tried that and it works:

[Zero999]

Here is a thread about LT3081 PSU:
http://www.diyaudio.com/forums/equipment-tools/263471-triple-outputs-160w-lab-psu-epsux3-version-2-a.html
I'm having a hard time setting the current limit digitally if you can have a go at LTspice and see if you can get the current limit to controlled digitally.

I don't see the issue. Use an error amplifier with its reference set by a DAC, to look at the current on the monitoring pin and pull the set pin down, if it exceeds the reference.

For me, one of the most annoying things with the LT3081 is the requirement for a minimum load current. If the adjustment range is small, then a resistor can be used but this is not feasible in an application which can work down to 0V, as the resistor would need to be close to zero and would draw an enormous current at higher voltage settings. The only solution is a current sink. To really get down to 0V you need a negative power supply but a simple current mirror can get close to 0V without one.

And get oscillations and instability, you are adding another external op amp that controls the current, another op amp that you will have to compensate and basically the same as taking LM317 and adding current limiting. Not an option.

Well no, it's not the same as taking the LM317 and adding current limiting, as you no longer need a separate current sense resistor and high side differential amplifier. The LT3081 does all of that for you.

As far as the instability is concerned, yes you'll need phase compensation, but you need that with any linear regulator design. I don't see how it's any more difficult than using the LM723.

"Frex" from the other forum found a solution, but I don't understand how it works.
EDIT - by looking at the schematic he seemed like he is having a current source into the Ilimit pin, so I tried that and it works:

That also looks like a bit of a PITA. The current source needs to have a low drop-out voltage and level shifting will be required, if the reference is on the negative rail.

[ZeTeX]

Here is a thread about LT3081 PSU:
http://www.diyaudio.com/forums/equipment-tools/263471-triple-outputs-160w-lab-psu-epsux3-version-2-a.html
I'm having a hard time setting the current limit digitally if you can have a go at LTspice and see if you can get the current limit to controlled digitally.

I don't see the issue. Use an error amplifier with its reference set by a DAC, to look at the current on the monitoring pin and pull the set pin down, if it exceeds the reference.

For me, one of the most annoying things with the LT3081 is the requirement for a minimum load current. If the adjustment range is small, then a resistor can be used but this is not feasible in an application which can work down to 0V, as the resistor would need to be close to zero and would draw an enormous current at higher voltage settings. The only solution is a current sink. To really get down to 0V you need a negative power supply but a simple current mirror can get close to 0V without one.

And get oscillations and instability, you are adding another external op amp that controls the current, another op amp that you will have to compensate and basically the same as taking LM317 and adding current limiting. Not an option.

Well no, it's not the same as taking the LM317 and adding current limiting, as you no longer need a separate current sense resistor and high side differential amplifier. The LT3081 does all of that for you.

As far as the instability is concerned, yes you'll need phase compensation, but you need that with any linear regulator design. I don't see how it's any more difficult than using the LM723.

"Frex" from the other forum found a solution, but I don't understand how it works.
EDIT - by looking at the schematic he seemed like he is having a current source into the Ilimit pin, so I tried that and it works:

That also looks like a bit of a PITA. The current source needs to have a low drop-out voltage and level shifting will be required, if the reference is on the negative rail.

if you need phase compensation like you said its the same thing as building PSU from scratch. I'm also quite sure that getting a stable current limit this way will be very hard if possible, you can't just look inside the regulator, and in this case, there is not point on using the LT3081, unless you want crappy current limiting e.g. LM723.
OP wants a simple IC that you connect 2 pots, and have it all stable fast and working. I don't know how stable over different loads the LT3081 is, it might be good, it might not. If it was that good and easy like it looks in simulation, wouldn't any manufacturer use it? it must be cheaper than having a discrete everything.

[David Hess]

Just recently I have run across old Motorola and National integrated power supply controllers but they are long discontinued.