LM723 power supply

[mike_mike]

Hello! It looks like the power supply from the attached schematic is oscillating (please have a look at the oscilloscope screenshots). I didn't install the capacitors at the output of the psu (1000uF and 100nF).
I tried by increasing and decreasing the pin 13 to pin 4 capacitor (by using 10nF or 470pF) but it didn't help.
Also, by adding capacitance on the output (100nF non polarized in total) it makes the oscillation worse (bmp_188_006).
The screenshots were took by using 1nF capacitor from pin 13 to pin 4. For test, I used a 10R load and 27V output voltage set by the 10k pot. The 10R load have been switched on and off by a NE555 circuit using a N-fet. The oscilloscope is a 20MHz dual channel from OWON.

I would like to know where should I start to check in order to damp the oscillations ?

I used: T1N4 = BT151, 2N2904 = BC327-40, BA723 = LM723CN

[mike_mike]

I made some modifications by adding ceramic capacitors into the circuit as follows:
1.  100nF in parallel with the 33V zener diode
2.  100nF from pin 5 of the 723 to gnd
3.  10nF in parallel with the 10k (voltage) potentiometer
4.  100nF in parallel with the output.

Now the waveform at the output looks like the attached screenshot.
I wanted to ask if the small oscillating damped waveform (marked with red rectangle) it is safe or it needs to be eliminated ?

[Kleinstein]

The ringing shown in the Red box is safe. However it depends how much and under what conditions (load). It is very hard (nearly impossible) to avoid ringing under all conditions - so for unfavorabel conditions some ringing like shown is normal. For more favorable this usually means that there is no, or much less ringing.

[mike_mike]

I found that when the 10nF ceramic capacitor is connected to the 10k (voltage) potentiometer, then when applying a load (even 120mA) then the output voltage drops by about 2-3 V (measured using the multimeter). For example, if I set the output voltage to 27V, when connecting the 120mA resistive load, then the output voltage drops to about 25V and it never come back to 27V, only when I remove the load.
It is something to be expected or it should be corrected by modifying some components values ?

[EPAIII]

The schematic you posted is difficult to read for two reasons: first it's resolution is low and second the person who drew it did not know how. That second reason does not inspire much confidence in my mind.

From what I can see, it is very complicated for a 723 power supply. I have built 723 supplies just by following the suggested circuits shown in the TI datasheet. They were much simpler and they worked just fine. If this is an attempt to breadboard a power supply I suggest that you follow the TI datasheet recommendations. Oh, and be sure to use a name brand 723 so you know it is a good part. 

https://www.ti.com/lit/ds/symlink/ua723.pdf

[Kleinstein]

The 723 circuits are often a bit difficult to read, as the chips functions are not obvious from the pins.
The circuit here uses an additional auxiliary negative supply, that complicates things.

Normally the regulation should be quite good. At the 27 V level, one may however be above the voltage the circuit can supply. The darglinton emittter follower type ouput stage adds to the voltage drop and the  negative supply limits the maximum input voltage. With some -6 V negative supply one may get dangrously close to the maximum ratings of the LM723 quite fast. So I would not consider this circuit to be a good starting point.

[Avelino Sampaio]

Mike

Expand your knowledge with this project by our friend Bram.

https://www.eevblog.com/forum/projects/good-quality-lm723-lab-power-supply/

[tggzzz]

Hello! It looks like the power supply from the attached schematic is oscillating (please have a look at the oscilloscope screenshots). I didn't install the capacitors at the output of the psu (1000uF and 100nF).

What's your construction technique? A photo would help.

If it is on a solderless breadboard, you may find you are debugging its parasitic resistance, inductance, capacitance. Consider an alternative construction technique such as manhattan, FFI: https://entertaininghacks.wordpress.com/2020/07/22/prototyping-circuits-easy-cheap-fast-reliable-techniques/

[Hydrawerk]

Hello, this 723 IC was introduced probably in 1972, that is 50 years ago. Is it ever going to be discontinued?? Or is it going to be produced almost forever because it is still popular?
https://www.ti.com/product/LM723?keyMatch=LM723&tisearch=search-everything&usecase=GPN
https://www.ti.com/product/UA723?keyMatch=UA723&tisearch=search-everything&usecase=GPN
They say that the product status is ACTIVE.
It is on stock.
https://octopart.com/search?q=lm723&currency=USD&specs=0
https://octopart.com/search?q=ua723&currency=USD&specs=0
Why is LM723 so much more expensive than UA723? Is it because LM723 is military grade with operating temperature range -55°C to 150°C?

[Swake]

It's status is similar to the 555. No reason to change that unless you've got something better.

Google this: ua723 vs lm723 and it comes up with this from the ti support site:
LM723C and UA723 are exact same devices. The LM version was designed by National Semiconductor and UA is TI's device . All of the technical specs are similar.  The only difference is in package types. LM723 is only available in PDIP(14 through hole) and TO100 (for extended temperature range). UA723 is available in 14PDIP, 14SO and SOIC packages.

[RoGeorge]

I didn't install the capacitors at the output of the psu (1000uF and 100nF).

Why not?  Does it still oscillates when adding something between 100uF...1000uF at the output?

[magic]

There is a whole thread here about reverse engineering the various 723 variants:
https://www.eevblog.com/forum/projects/lm723-die-pictures/

Google this: ua723 vs lm723 and it comes up with this from the ti support site:
LM723C and UA723 are exact same devices. The LM version was designed by National Semiconductor and UA is TI's device .

TI customer support drones are the same idiots who recommended somebody that he burns-in his UA723 chips by shorting the output to ground and running them into thermal shutdown.

Pro tip: the only thermal shutdown mechanism available in 723 chips involves release of magic smoke

All of the technical specs are similar.  The only difference is in package types.

National version uses a bandgap voltage reference instead of a crappy-ass, 1960s technology Zener diode. This reference doesn't suffer from so-called "zener walkout", which is a significant early drift of reference voltage (and the reason why some people tried to bake TI's parts to stabilize them). Generally, the LM723 was found to have superior long term stability by one experimenter who bothered to test it.

On the other hand, the bandgap version has much higher noise before filtering and this is reflected in the datasheet specs, see "Output voltage noise" with C_REF=0, but those losers apparently couldn't even be assed to compare the datasheets.

LM723 is only available in PDIP(14 through hole) and TO100 (for extended temperature range). UA723 is available in 14PDIP, 14SO and SOIC packages.

AFAIK, LM723 is no longer available in DIP. It seems they only make them in metal cans, for some last remaining aerospace or similar customers who are stuck using 723 due to the cost of re-certification of a new design and they are stuck with the NatSemi version due to aforementioned long term stability issues. So they are paying for them through their noses (look up the pricing) and TI beancounters are happy. Otherwise, this part is no longer available, you can only buy the TI version which uses a Zener diode, like the original.

[iMo]

Could the schematics even work?
Is the pin 7 GND of the 723 grounded properly? Should it not go to -8.2V?
Not sure about the R5 value (is that 27k 29k or 2k7 ??), though..

[Kleinstein]

The difference amplifier inputs won't work below ground (pin7). Noramlly it need some 1 V of headroom. The exact required minimum can be a bit different between versions.
So there is something wrong with the ground or resistor values. It could be 15 K instead of 1.5 K.  R5 should not make that much difference, mainly setting the lower voltage limit.

[iMo]

This shows at least some life..

X axis is the wiper's position from 0 to 1..

[David Hess]

That looks like a parasitic oscillation in the output transistors, possibly from layout.  Modern 2N3055s are more than 10 times faster than the 2N3055s from 1972.

LM723 is only available in PDIP(14 through hole) and TO100 (for extended temperature range). UA723 is available in 14PDIP, 14SO and SOIC packages.

AFAIK, LM723 is no longer available in DIP. It seems they only make them in metal cans, for some last remaining aerospace or similar customers who are stuck using 723 due to the cost of re-certification of a new design and they are stuck with the NatSemi version due to aforementioned long term stability issues. So they are paying for them through their noses (look up the pricing) and TI beancounters are happy. Otherwise, this part is no longer available, you can only buy the TI version which uses a Zener diode, like the original.

Mouser shows the TI 723 available in PDIP, TO-100, and SOIC.

[Kleinstein]

I would not expect the LM723 to be used in large quantities anymore. So existing stocks of the DIP version could last quite some time.

For the stability of the output stage a faster power transistor is usually not a real problem. It often makes things even easier as normally the power transistor should not be the speed limiting element. Many circuits can have a problem with to slow a transistor. So an old 3055 (e.g. the 3055H) could cause oscillations.

The LM723 is a chip made for internal power supplies to provide power to a fixed, relatively well defined load. It is not designed to be used as a lab power supply. The difference is that the internal supply needs to be stable with the given, fixed load. A lab supply should be stable over a relatively wide range of loads. This seamingly small difference is quite a step, from a relatively easy compensation to a relatively demanding design.
The other point is that the current limit in the 723 is rather crude, for protection purpose, not really a viable adjustable current regulation. So I would consider the LM723 OK for a fixed supply if one can not get away with the 3 pin regulator (e.g. need better stability, custom current limit), but a poor starting point for a lab supply.

[RoGeorge]

Here the 723 was the de-facto chip for home made lab power supplies for at least 20-30 years, until the switching power supply became the norm.

There were industrial models too, not only hobby grade.  Adjustable voltage/current was casual, some were having compensation inputs, too.  Lab power supplies were either with discreet transistors or with 723.  There was not much choice of ICs, and the performance was very good on all the range of voltage/currents/loads.  I remember 723 as quite sturdy.  That was when all was analog, with phisical knobs and buttons (one potentiometer for coarse, and one for fine adjustments, so 2 for voltage 2 for current  ), before digital and before SCPI instruments became the norm.

[iMo]

I built my first lab psu some 45y ago, my father brought me the metal version of the 723, a large 24V transformer and a large metal transistor on a large heatsink. All mounter on a piece of wood, not in a box but open, with an "ON" bulb wired on the transformer, the current limit shunt was hand wound, with a simple front panel with the posts, a pot and a small 30mA meter showing volts. I was using it for all my experiments, it never failed.. 

[Hydrawerk]

The LM723 is a chip made for internal power supplies to provide power to a fixed, relatively well defined load. It is not designed to be used as a lab power supply.

OK, so should we rather use a circuit made of opamps and transitors? Like these?
http://electronics-diy.com/adjustable-lab-power-supply.php
Or this chinese DIY kit... https://www.qsl.net/z33t/dc_0-30v_0-3A_eng.html
that is probably inspired by this "Zdroj G400" published in Czech republic in 1996. http://paja-trb.cz/konstrukce/zdroj/zdroj_G400.pdf

[Hydrawerk]

https://www.tek.com/en/datasheet/ps280
Tektronix PS280 (probably a rebadged GW Instek from 1991?) uses LM723, some opamps and transistors.
https://w140.com/tekwiki/images/e/e5/070-8356-00.pdf

[Kleinstein]

There was the MC1466 as a higher perfromance alternative for a lab supply, but this one is long obsolete. For a lab supply it makes more sense to use OP-amps and a seprate reference. Chances are a LM358 and TL431 could be more useful than the LM723.  After all the LM723 is not much more than a moderately good reference and 1 difference amplifier that is not even single supply.
It can be worth using a more precise amplifier for the current regulation - lower noise and less drift from the amplifier may allow to save on the shunt's power rating.

The floating regulator (like many older HP supplies) with a 2nd tranformer to supply the regular is working quite well and many of the cheap Chinese supplies are based on this concept.  The main problem with these is not the circuit concept, but too high a maximum current leading to thermal problems and low reliability. The 2nd transformer (or extra winding from a custom transformer) is extra effort, but it helps to get low drop regulation and also working with output voltages higher than the OP-amps supply. So the circuit concept works both for low (e.g. 5 V) and relatively high (e.g. 50 V) voltage versions.

For adjusting the voltage it is usually better to set the set voltage to the regulator and keep the feedback divider constant. In the old days (also with the circuit at the start of the thread) it was popular to adjust the divider ratio: this may help a little to get lower noise, but it changes the loop gain with the voltage setting and for this reason is usually not a good idea. In addition it gets hard to get all the way down to 0.

[iMo]

Here is a rather long thread where a lot of folks here spent some time with elaborating a PSU
https://www.eevblog.com/forum/beginners/lm324-power-supply-with-variable-voltage-and-current/
Have a look on that - there is a lot of info..

[David Hess]

For the stability of the output stage a faster power transistor is usually not a real problem. It often makes things even easier as normally the power transistor should not be the speed limiting element. Many circuits can have a problem with to slow a transistor. So an old 3055 (e.g. the 3055H) could cause oscillations.

The waveforms shown are consistent with a local oscillation and not a loop oscillation, and we do not know what the physical layout was.  Power designs with parallel output transistors often have them mounted remotely with long wiring.

There was the MC1466 as a higher perfromance alternative for a lab supply, but this one is long obsolete. For a lab supply it makes more sense to use OP-amps and a seprate reference. Chances are a LM358 and TL431 could be more useful than the LM723.  After all the LM723 is not much more than a moderately good reference and 1 difference amplifier that is not even single supply.

The thing which is difficult to replicate with operational amplifiers is the transconductance output of the 723 error amplifier.  With a current output, there is no slew rate limitation when switching between voltage and current modes, which is what the 1466 did.  In practice of course it is not much of a problem with careful design, but it was easier with a 723 or 1466.

Some operational amplifiers can in theory provide a transconductance output through their external compensation pin, and I have seen it used as a clamp in some power supply designs, but this is limited to selected old parts like the 301A and maybe 308 derivatives.

That leaves using a discrete differential pair to use as the error amplifier which will be more expensive than a 723, so 723 it is if you want that kind of performance.

[floobydust]

I found OP's circuit clipping original article 1990/5-6 Colectia Tehnium Romanian electronics magazine. I think the circuit is flawed in too many ways and not worth pursuing. Many of these old magazine circuit collections were like that- typos or bodged designs. Who runs 44V into a PSU with max. 32V output unless you like the heat. Who uses 1,000uF output capacitor unless you like the sparks flying...
Old linear open-frame power supplies for many years used the LM723, I was surprised to see one using instead a TL431 and a few more discretes. The '723 seems to have fallen out of favour.

[iMo]

..Some operational amplifiers can in theory provide a transconductance output through their external compensation pin, and I have seen it used as a clamp in some power supply designs, but this is limited to selected old parts like the 301A and maybe 308 derivatives.
That leaves using a discrete differential pair to use as the error amplifier which will be more expensive than a 723, so 723 it is if you want that kind of performance.

You mentioned several times in past a version with a "pair of 723" wired via their compensation pins. Is that schematics somewhere available?

[David Hess]

You mentioned several times in past a version with a "pair of 723" wired via their compensation pins. Is that schematics somewhere available?

The datasheet for the MC1466, actually MC1466L, shows how it is done with the current outputs of the transconductance error amplifiers tied together into one output with diodes.  Many switching regulator controllers that have two dedicated transconductance error amplifiers have them tied together in the same way, but not as many details are published.  This is the only example with a pair of 723s that I have found online, and it does not do this:

http://www.ve2ums.ca/chasse/Serge/Atelier/Projets/Membres/VE2EMM/alimentation_ang.htm

The design below shows how the compensation pin of an operational amplifier can be used as a clamp input for fast current limiting response.

[iMo]

Is there an estimation how faster is the clamping through the compensation pin against a traditional path through the modern opamp internally compensated?

[Kleinstein]

In the circuit the clamping part makes sure the current limiting amplifier starts not far (e.g. some 0.6 V) from engaging. So it takes relatively little time to engage the current limit. For the details it would still need a simulation. The circuit uses a kind of voltage settling output stage and is this still limited by the amplifiers slew rate. The use of the compensation pin is only an elegant way to  keep the dead time small. One still has delayed responst in case of a dead short that need the voltage to drop a lot.

For a circuit with classical compensated OP-amps it still depends a lot on the circuit. One can get a silimar limiting in the off state. If the respeonse of than faster or slower depends on the ampifiers and circuit. There are many different options on how to implement the combination of voltage and current regulation. One interesting way is the system used with some SMUs:  the current and voltage regulator parts are combines with some diode min / max circuits and the frequency compensation / loop speed is only set after that with some PI element. This gives kind of similar effects as combining the current signals in the MC1466L or with 2 x LM723 linked with the compensation pins and using a common capacitor for compensation.
Combining the difference amplifiers currents is elegant, but not that unique that similar response can not be obtaines with normal op-amps.

[David Hess]

In the circuit the clamping part makes sure the current limiting amplifier starts not far (e.g. some 0.6 V) from engaging. So it takes relatively little time to engage the current limit.

Exactly, the clamping prevents windup of the integrator which is responsible for the frequency compensation.  Nothing prevents clamping both operational amplifiers so that the response is faster for both switching to constant current and constant voltage.

The above applies when operational amplifiers are used.  If operational transconductance amplifiers are used instead, like with the 723 or MC1466L, then the frequency compensation happens *after* the diodes which combine the outputs so switching between modes is practically instantaneous.

In theory the compensation pin can be used as a transconductance output for operational amplifiers like the 301A, however I have never seen this done.  If I have some time, I may do some experiments with it next year.

[iMo]

Here is the TI Nat Semi LB-28 PSU from above in the LTSpice, you may play with.. Modded the LM395 special transistors..

[iMo]

V2 with fixed errors. Some changes in fb capacitors, 25V/3A max, it somehow works, but it oscillates with various settings and with capacitive loads. Needs some elaboration, sure..

[David Hess]

V2 with fixed errors. Some changes in fb capacitors, 25V/3A max, it somehow works, but it oscillates with various settings and with capacitive loads. Needs some elaboration, sure..

The LM395s are awfully fast, in the figure of merit sense, compared to a common power transistor and have much higher current gain decoupling the load from the error amplifier.  Ring emitter transistors like the D44H11 series or audio parts like the MJL4281A are more suitable.

The frequency compensation for this design requires more care, and I would probably make changes to increase the phase margin to handle difficult loads better while keeping the low output capacitance.

[iMo]

Tried with D44H8 (I've found the model in my LTSpice) and it works it seems, with smallest fb capacitancies, into a capacitive load as well..

[David Hess]

I do not know why they used a 308 in the original design, instead of another 301A, but an OP27 is much faster than a 308 and will require external frequency compensation changes.  The lower noise of the OP27 is not a benefit because even a low noise reference will determine the output noise.  My guess is that at the time, the 308 had better precision than a common operational amplifier and was more easily available without a high price.

The LT1008, LT1012, and LT1097 are superior replacements for the 308 with similar dynamic performance, but a cheap OP07 should be almost as good.  The LT1008 is an exact replacement including compensation, while the LT1012 and LT1097 have intriguing support for overcompensation.

There are tons of suitable modern operational amplifiers though, except for replacing the 301A if clamping is used.

I like using this design as an example because it shows several things:

1. High side current sensing combined with an error amplifier.  Note that this creates an output current error from the pass element base current, which was acceptable with the LM395 but probably not with bipolar output transistors.  This could be corrected with a JFET or depletion mode MOSFET driver, or a MOSFET output.

2. Use of external compensation for clamping, which unfortunately is limited to the 301A, but at least they are still in production.  The 301A also has a common mode input range which includes the positive supply.

3. Active pull-down of the output.

4. Compensation to allow for a minimum of output capacitance.

[iMo]

I used to use the OP27 in the simulation as an example only (faster than the OP07).
Perhaps we have to start a new thread with that design and try it in situ. Also it would be great to identify parts which are readily obtainable and working fine with that concept. The clamping via pin 8 is the interesting option, thus the opamp should support that.
PS:  also I think there is an error in the original scheme - the 1k2 resistor should be wired in series with the LED diode, not in parallel..

[Kleinstein]

The LED part is a bit strange. Besides the series / parallel question with the resistor the LED current would add to the output current. Another problem is that with transients the LED may see quite some reverse voltage that may exceed the usual 5 V limit.

The active sink part has some positive effects, but it can also upset some Op-amps (especially faster ones)  with the control via the neg supply pin.

There are plenty of OP-amps suitable for the voltage control - the main point here is a high permissible supply voltage. Depending on the supply (e.g. a version without a negative supply) single supply operation may be desirable. The MC33171/2 would be a possible canditate in this respect.

There are not many OP-amps with external compensation similar to the LM301.
The LT1008 has external compensation and improved accuracy but also at a high price and it does not work near the positive supply. So it is still not a direct replacement.
The NE5534 also has external compensation (not sure if usualbe in the same way), but as well not working to the possive supply).

[David Hess]

The NE5534 also has external compensation (not sure if usualbe in the same way), but as well not working to the possive supply).

I checked the schematic and I think it can be.  The compensation connection comes off of the Vbe multiplier which is current driven.

The LED part is a bit strange. Besides the series / parallel question with the resistor the LED current would add to the output current. Another problem is that with transients the LED may see quite some reverse voltage that may exceed the usual 5 V limit.

The output from the 308 cannot pull negative very far because it will drive the output stage of the 301A through the clamp connection pulling the base of the transistor low through D3.