uA723 voltage regulator viable in 21st century?

[Wolfgang]

Hi,

thanks for the tip. I will try this one out !

Wolfgang

[Wolfgang]

Hi,

apart from being in cents range for 100pcs, the LM723 has survived a lot of competitors that provided better data.
One of these was the SG3532 regulator, which offered the following advantages:
- lower voltage due to a 2.5V reference
- high error amp gain
- precision current limiting
- overtemp protection
- better tempco
...

There were some others, too, like the  LAS1000 and LAS1100. The funny thing is that all of them have been discontinued. The key point was
that the LM723 price/performance was hard to beat, and the LM723 was good enough for most applications, so the niche for the competitors
was not large enough.

Today, there (supposedly) are chips that outperform the 723, but they do not even come close to the price.
BTW: Which chips today can compete and who builds them ? Just curious ...

So - maybe its lasts for some more decades to come. The 723 has already survived its inventor by 27 years now !

Regards
  Wolfgang

[bd139]

The 723 has already survived its inventor by 27 years now !

Only because he drank so much

[David Hess]

There are a couple of things which I think the 723 should have included like thermal protection and a second transconductance amplifier for precision current limiting but its price for performance cannot be beat.

The 709 on the other hand was a terrible operational amplifier but contrast it to devices like the 301A and 741 which are still excellent.

[CopperCone]

What kind if noise performance can you get with uVrms with 100KHz bandwidth limit and 10Hz bandwidth limit?

OK, I found it on some audio forum

It says the 723 is roughly 2uV RMS

Most linear regulators out there that are considered low noise will be between 100-0.5uV RMS, with the LT3042 being 0.8uV rms or so, which is the lowest noise manufactured part I found with a rating.

As for negative rails, for precision parts you will get between like 100uV rms to 20uV rms I think. They don't do quite so well.

I like LT3080 and LT3015 (neg), which have 20uVrms/60uV rms respectively. 



************
Can the uA723 be used to make a negative rail by any chance? If it's only 2.5uV rms for a negative rail that would be an order of magnitude improvement over other devices.

And how are it's specifications for PSRR and tempco compared to modern references? If they are particularly bad it would require its own temperature stabilization as not to effect a system. The tempco should not have much effect on AC circuits though.

150ppm/C - 723 (worse case), ripple rejection not bad either. Best case is 30ppm.

For comparison, the LT3080 -
0.5% change in bias current over 75C,
0.006%/C, 60ppm. Unknown if this is typical, max, random sample, etc. Just extrapolated from a graph. Has a curve around 0 so if you freeze it then it will reverse.


I think I ignored this part because it is old. It seems decent. For home made instrumentation I'm not sure I give a shit if a linear regulator melts or has current limiting. Maybe hard to make use of it for powering modern op-amps.

It is only 100mA though.


****
I think that parts like the LT3042 which have a lower noise are more designed for PSRR at MHz then low noise, for use in biasing an RF transistor or MMIC directly from a rail. Modern op-amps have typically a high PSRR so you can kinda get away with feeding them garbage, so I am not sure how much benefit lowering the rail noise has... This I think changes if you use them to bias transistors (i.e. differential matched transistor input stage for op-amp). Not sure how PSRR works out for something like a MMIC, usually I just see the voltage connected to the output by a resistor, but it does have VDD pins I think.

I would like to hear an explanation about the corporate decisions and market demands that lead to the current product market that we have today when it comes to precision linear regulators.

[Wolfgang]

Hi,

just for curiosity, please give me a few examples of the new linear regulators than can rival 723 performance.

Thanks
   Wolfgang

[CopperCone]

I don't know the real specs of the 723. I just found 2.5uV RMS for unknown bandwidth in some audio forum.

The LT3042 has graphs and stuff and it is rated at 0.8uV RMS for a wider bandwidth I think.

Everything else is at least 10uV rms IIRC. But thats just noise.

[Wolfgang]

The noise specsof the LT chip are not bad, but its not a replaceable part because it has much less voltage tolerance and dissipation.
What I would be looking for is a 40V part with better specs than the 723. The SG3532, LAS1000 and LAS1100 are gone, ist there something else I missed ?

Best regards, thanks
   Wolfgang

[David Hess]

What I would be looking for is a 40V part with better specs than the 723. The SG3532, LAS1000 and LAS1100 are gone, is there something else I missed ?

If you want better specifications then use an external buffered reference and precision low noise operational amplifier to drive the pass element of your choice.

The idea of buffering the reference is to low pass filter it.  A 723 by itself incidentally makes a pretty good buffered reference except for its tolerance.  Zener references have lower inherent noise than bandgap references when their higher output voltage is taken into account; multiplying a bandgap reference up to the working voltage multiplies its noise as well.

The pass element can be a three terminal voltage regulator to take advantage of its integrated protection.  Placing it inside the feedback loop of a low noise operational amplifier lowers its noise.

Noise measurements of different parts and circuits really need to be made under controlled and specified conditions.  There is more to it than just noise bandwidth.

[chris_leyson]

One of Ulrich Rohdes books gave me the idea of using an LM723 as a low noise regulator. I'm pretty sure it was "Communications Receivers, principles and design" 2nd Edition but I could be wrong. Found something similar in Microwave and Wireless Synthesizers: Theory and Design, fig. 2-20. Got a copy somewhere I will have to go dig it out. Found the original harware for 20 years ago and I didn't bypass the non-inverting input or use an external PNP pass transistor and the noise spectral density was still better than any of the othe "low noise" regulators at the time. I think that when I measured the output noise spectral density I used a pair of AD797's AC coupled to give me 40dB gain and an HP3585. Sadly don't have the 3585 anymore.

[Wolfgang]

Hi.

I agree that regulator noise is a difficult topic. The german paper I mentioned in one of my last posts did a very comprehensive study on several regulator chips
plus different loading and filtering environments.

IMHO, the critical points for standard chip regulators are:
- the reference must be filtered
- If a pass transistor needs to be used, it should be a PNP bipolar type (this adds less noise due to the fact that the base spreading resistance of a PNP is less than an NPN,
  causing less noise there).
- in case the regulator has no spectacular PSRR (as the LM723), its own supply should be cleaned up (preregulator)
- To clean up rests of noise and ripple, active noise killer circuits (like the shunt designs by Charles Wenzel) can help.

When observing all this, sub-millivolt RMS noise levels are possible up to more than 100kHz.

If you need definetely more than that, there are no single-chip possibilities anymore. Then the way to go would be:
- An ultra low noise heated reference (LM399, LTZ1000).
- No mechanical stress on the reference, buffer and error amp. Make PCB cuts to make sure.
- No sockets
- An ultra low noise reference buffer (LT1028)
- Only ultra stable and low noise metal FOIL resistors
- No electrolytic caps in the control path (due to leakage currents causing noise)
- A super high gain error amp (LT1028), with almost no power dissipated inside the chip
- A PNP bipolar driver and pass transistor with low noise specs (very good audio types)
- No ceramic caps due to piezoelectricity, nonlinearity and noise. Only NP0 or good quality foil allowed !
- No current limiting inside the control loop. Do this outside, by a dedicated circuit.
- Use a preregulator (this could be well an LM723) for the sensitive part of the circuit.
- No unshielded connections to the load. Use shielded, coaxial, low thermal EMF and low triboelectricity cables to the load and for sensing.
- In extreme cases, do not use a PCB, but single ceramic standoffs.
- Stalinistic packaging and shielding is a must.

If you make no mistakes, microvolt noise levels are possible. The number of pitfalls is enormous, however.
And an 8 1/2 digit Fluke or Keysight multimeter is neccessary, too ...

Regards
   Wolfgang

[CopperCone]

i am debating if stalin would produce a good package or not.

[Wolfgang]

Stalinistic is probably a European term for "Paranoic", leaving no room to electricity to escape ...

[CopperCone]

im still debating if stalin would make a good package, i think his solution would be marginal because he would be afraid that he might be misusing precious party resources that could be better used to crush capitalism.

maybe he would slowly scrape away the walls till he got to a level thats acceptable. not sure.

[bd139]

Stalin would tell everyone the manufacture of packages would empower the proletariat in the name of Lenin yet at the same time Siberian gulag workers would be grinding the packages out of rocks with their finger nails under threat of execution to meet the package quota.

[Wolfgang]

... well, he considered himself an expert of fencing in people (it was called "Iron Curtain" at the time), so the term is sometimes misused for a watertight packaging style. From an economical or rational point, his actions never made  a lot of sense ...

What is serious, however, is the importance of perfect EMI shielding for extreme performance power supply circuitry. Shielded transformers, a lot of space between stray fields and possible receptive circuitry, enclosures with feed-thru caps, ... are a must. Even fans must be viewed with caution ! Another issue is mechanical vibration.

So, whatever term you prefer, make sure that nothing you dont want can get into your circuitry. Have you ever tried placing your mobile close to a DUT when it just receives a call or is dialling out ? You might be surprised ...

Best regards
   Wolfgang

[CopperCone]

people like to use plastic lol

The only PSU I have seen with RF feedthroughs is the small power supply from my 1980's spectrum analyzer. Each of the rails had a RF feedthrough, and the unit fan had a big series torrodial inductor on it.

For what you want, you either need to build it with a special chassis that has air cooled heat sinks that go to the outside, formed as part of the shield (like some microwave amplifiers I have) or you want to put RF-Honeycomb on your air vents

Example:
http://www.raymondemc.ca/products/products6a.htm

The fan itself may be a problem because they are made of plastic typically and have their own circuitry. You might have to get a metal shielded motor, figure out its proper grounding and isolation, then use a belt to drive a fan, this way you can control variables, or figure out how to turn it into a fan yourself.

I have absolutely no clue as to the status of some kind of low electrical noise and emission fans.

[Wolfgang]

Agreed.

Probably the best way is a "no holes" design with a milled chassis where the cooling fins on the outer side of the RF-tight enclosure. Fans are a problem; if you want to be sure they dont create hassle you cold detach them from the fooling fins via an air duct  loosely filled with iron wool of a few cm AND use a honeycomb against RFI. I have seen techniques like this in Rohde and Schwarz equipment from the 80ies.

This shielding and precision instrument business is a bag of tricks more than science ...

[CopperCone]

its not really a bag of tricks as much as you may need to do complex time consuming RF/parasitic simulation to really optimize it, and most people don't know how

But your gonna get a good baseline using certain design techniques. You still need to know how to do relevant calculations on filters, shielding thickness, impedances, etc

I don't like the term 'bag of tricks' because then some wise ass is going to get the idea that its not really engineering effort that goes into it, then some ones boss is going to be acting like it should be easy, fast and that you don't need a good understanding of electronics to do it right. You can still do rough calculations to guide a design using first order models of parasitics and stuff, and its not super popular so you probably want to test along side of designing to make sure your going in the right direction.

[BravoV]

Just found my LM723 stash, almost 3 decades old, given by my mentor many-many moons ago. 

[schmitt trigger]

Produced in 1989! Almost 30 years old.
Because of their metal can and frit seal, I am sure they are as good as new.

You are sitting in a stash of 24 karat gold, my friend.

[Wolfgang]

What I experienced is that there is a limit to what can be successfully simulated. The higher the precision gets, the larger the number of dirt effects will get that do play a role in your design. It is very instructive to see how really precise (7 1/2 and 8 1/2) digit multimeters are built.
Just have a look there at component selection, PCB design , thermal management, aging, drift, shielding, EMF, thermoelectricity, triboelectric effects, hysteresis, ...

I have never seen a model that takes all these effects into account. For certain effects, only heuristic models are available (Arrhenius, ...).

It is probably fruitless to build a complete, successful model of a very high accuracy ultra high precision circuitry taking ALL relevant effects into account. Some effects can be handled heuristically, without a complete theory behind it, and the heuristics can still be useful.

Conclusion: A well handled bag of tricks can make your precision elecectronics design life a lot more easy. If a consistent theory is available, you dont need the tricks anymore, but at cutting edge technology we are quite far from this situation.

So, the search for exact models and deep understanding is fine, but pragmatic solutions are OK too, as long no established and general theory is available.

It is like the old Aristotelian saying : The Perfect prevents the Possible.

Enough of philosophy, get back to your benches, measure, research, model and create better a better understanding. In the meantime, some (plausible) tricks are OK. The final result is what counts.

Regards
   Wolfgang

[Zero999]

There are a couple of things which I think the 723 should have included like thermal protection and a second transconductance amplifier for precision current limiting but its price for performance cannot be beat.

The 709 on the other hand was a terrible operational amplifier but contrast it to devices like the 301A and 741 which are still excellent.

A lower reference voltage and an error amplifier with a common mode range down to 0V, would also be nice, but probably would have pushed the price up too high back then.

[David Hess]

There are a couple of things which I think the 723 should have included like thermal protection and a second transconductance amplifier for precision current limiting but its price for performance cannot be beat.

The 709 on the other hand was a terrible operational amplifier but contrast it to devices like the 301A and 741 which are still excellent.

A lower reference voltage and an error amplifier with a common mode range down to 0V, would also be nice, but probably would have pushed the price up too high back then.

A lower reference voltage would have meant a more complex bandgap reference.

A PNP input gain cell with transconductance reduction like the 324/358 would have been nice not only for operation down to ground but because PNP transistors on an NPN process support a high differential input voltage but that had not been invented yet.  It was common in later switching and linear voltage regulator controllers.  An NPN version would have allowed high side current sensing but adding these sorts of features can get out of hand quickly.

It would be nice if there was a standard precision operational transconductance amplifier for these types of applications.

[Giuss]

I think it is a very good chip: stable, reliable, easy to use and cheap.
 I have used it in some projects and I have no reasons to search modern chips to replace it