The simplistic Salas low voltage shunt regulator

[anatech]

Hi Salas,
I'm not about to go doubting that your readings from your Fluke are out. I used to calibrate Fluke meters and they hold their calibration extremely well. Especially after they went to closed case calibration. I'd expect the same thing from HP / Agilent / Keysight or Kiethley. All excellent meters.

I must either be lucky, or have ordered a tighter tolerance part. I don't know why though. All that is needed is a stable low noise reference, not one that is 6.900 VDC. Mine were within 100 mV as I recall.

I Like Scott's idea with the AD581 chips. I have yet to score an LMZ1000, but I have everything else that is close. I haven't put anything together yet.

-Chris

 

[Salas]

Could be the ambient and the dissipation between us. Our room temp in C and voltages across, currents passing as we measured. DZ is the 100ppm max cheaper one. They also used to offer better ppm more expensive versions to get. Its not about reverse breakdown tolerance as you said. Its not a Vref chip. Its a low impedance low noise Zener equivalent. Those are the specs they guard.

 

[anatech]

Hi Salas,
True enough. I use them as shunt regulators (zeners), but not normal zeners due to noise. Aren't these also temperature compensated to some degree? Anyway, I've come to think of them as low noise "references" rather than zeners. I even store them with my references. Zener diodes are things I use for less than critical applications without regard to noise.

What's really nice is that you don't have to cook these to get low noise. Of course you are the noise expert in this conversation.

-Chris

 

[Salas]

They are current amplified and compensated subsurface Zeners. IC Zeners. No true Vref gives 0.65V nominal voltage tolerance. They can be truly seen as low noise stability references in my view. What you need in an instrument's reference is not to drift. Its per sample absolute value is something you measure and can trim for in the circuit. Then you can guarantee ppm.

 

[Salas]

P.S. Linear Technology lists down to 10ppm LM329A. But they spec Vz 6.7 to 7.2 for all ppm versions. Its all about ΔVz/ΔTemp.

 

[anatech]

Hi Salas,

What you need in an instrument's reference is not to drift. Its per sample absolute value is something you measure and can trim for in the circuit. Then you can guarantee ppm.

Yes, I have a large number of new voltage references on hand for instrumentation, which I work with. The LM329 is merely a stable voltage source that has low noise that I use for audio. I feed all these devices with a current source, fixing the current. I do the same with the LM329 as that is a lot quieter than a resistor from a regulated output.

I was too lazy to call up the data sheet for the LM329, but it is excellent for use as a low noise reference for audio. In this application the actual voltage is not that critical, not even drift is a problem really. I use things like the LM399 or LH0070-0H for critical applications, and I have some REF5020 and ADR440 family devices along with just about every other Vref out there. I use an HP 3457A (71/2 digits from the buss) and HP 34401A for monitoring these devices. I can't afford an Agilent or Keysight 3458A, nor can I justify that expense.

Anyway. once I'm using one, I do have the data sheet out and can talk much more intelligently on that device.

Its all about ΔVz/ΔTemp.

Well, clearly. Subsurface, or buried zeners are about the lowest noise you will find because you don't have surface defects to worry about. They also tend to be more stable in operation. But the temperature dependence is the reason why references like the LM399 have an oven built in and come with their own insulated package. But I would never use something like that for audio. Slow drift wouldn't normally be a problem for audio circuits. I think the noise from a buried zener is lower than a band gap device, but I should maybe check on that. If they were lower noise, I imagine you would have gone that route.

Best, Chris

 

[bcarso]

voltage/current references

Hi Salas,

Yes, I have a large number of new voltage references on hand for instrumentation, which I work with. The LM329 is merely a stable voltage source that has low noise that I use for audio. I feed all these devices with a current source, fixing the current. I do the same with the LM329 as that is a lot quieter than a resistor from a regulated output.

I was too lazy to call up the data sheet for the LM329, but it is excellent for use as a low noise reference for audio. In this application the actual voltage is not that critical, not even drift is a problem really. I use things like the LM399 or LH0070-0H for critical applications, and I have some REF5020 and ADR440 family devices along with just about every other Vref out there. I use an HP 3457A (71/2 digits from the buss) and HP 34401A for monitoring these devices. I can't afford an Agilent or Keysight 3458A, nor can I justify that expense.

Anyway. once I'm using one, I do have the data sheet out and can talk much more intelligently on that device.

Well, clearly. Subsurface, or buried zeners are about the lowest noise you will find because you don't have surface defects to worry about. They also tend to be more stable in operation. But the temperature dependence is the reason why references like the LM399 have an oven built in and come with their own insulated package. But I would never use something like that for audio. Slow drift wouldn't normally be a problem for audio circuits. I think the noise from a buried zener is lower than a band gap device, but I should maybe check on that. If they were lower noise, I imagine you would have gone that route.

Best, Chris

Although no one I know of does so, with sufficient amounts of low rbb' bipolars one could contrive very low noise band gap references. I've been tempted to do this as an exercise as I have a few bags of the now-unobtainium 2SC3329 and 2SA1316.

But for not-too-critical temp drift, finding JFETs with a bias as an I source that results in a ~zero tempco of drain current, and using that current to develop a voltage somewhere, results in about the lowest-noise reference voltages I've ever seen. Singly- or multiply-cascoded the power supply sensitivity gets very small, but you have to adjust the bias for the effects of the cascoding device tempco. All in all it is very laborious, and then the developed voltage usually needs to be buffered. Of course the works can be isolated and temp-stabilized. One is tempted to say if you need this for an audio circuit you're doing something wrong, but it is there if needed. It's how I developed a general system reference in a complex spectrometer circa 1976, at that time using some ADI JFETs that had very low 1/f noise for the day (so old that they have long since recycled the part numbers for other uses).

In the case of these shunt regulators and other related circuits, and with the use as described by Preamp of both polarities, it's a natural to use such I sources to develop the reference voltage, again with low noise gain. I'm not going to add to the fray but one can see the way to a few more parts and removing the tempco of the input amplifying device by using a differential pair, biased from the opposite rail, a current mirror, and thus push-pull dirve of the DMOS shunt device. I simulated a few versions early this AM and it looks like they avoid start-up problems and should have exemplary noise performance.

 

[bcarso]

The other thing to remember is that averaging the noise from a plurality of references always works. For fun I stacked a bunch of bog-standard 431s where I needed a fairly large reference voltage for a series regulator, and results were as expected, with an overall improvement in the "signal-to-noise ratio" of the resulting voltage going as the square root of N. The surprise was how they all played nice together --- I was rather expecting oscillations. Thereafter I filtered a good deal as well.

 

[Salas]

Hi Salas,

Yes, I have a large number of new voltage references on hand for instrumentation, which I work with. The LM329 is merely a stable voltage source that has low noise that I use for audio. I feed all these devices with a current source, fixing the current. I do the same with the LM329 as that is a lot quieter than a resistor from a regulated output.

I was too lazy to call up the data sheet for the LM329, but it is excellent for use as a low noise reference for audio. In this application the actual voltage is not that critical, not even drift is a problem really. I use things like the LM399 or LH0070-0H for critical applications, and I have some REF5020 and ADR440 family devices along with just about every other Vref out there. I use an HP 3457A (71/2 digits from the buss) and HP 34401A for monitoring these devices. I can't afford an Agilent or Keysight 3458A, nor can I justify that expense.

Anyway. once I'm using one, I do have the data sheet out and can talk much more intelligently on that device.

Well, clearly. Subsurface, or buried zeners are about the lowest noise you will find because you don't have surface defects to worry about. They also tend to be more stable in operation. But the temperature dependence is the reason why references like the LM399 have an oven built in and come with their own insulated package. But I would never use something like that for audio. Slow drift wouldn't normally be a problem for audio circuits. I think the noise from a buried zener is lower than a band gap device, but I should maybe check on that. If they were lower noise, I imagine you would have gone that route.

Best, Chris

Precision is not the first thing to look for in regulators targeting audio applications IMHO too. Its good to have but not a priority. Take an LM329 and a generic red Led to your HP SA for instance and you may find that some Led in your random collection could show bit less nVrtHz and less harmonic noise than the LM329 especially when the Led or a string of those match the JFET's CCS impedance.

 

[Salas]

The other thing to remember is that averaging the noise from a plurality of references always works. For fun I stacked a bunch of bog-standard 431s where I needed a fairly large reference voltage for a series regulator, and results were as expected, with an overall improvement in the "signal-to-noise ratio" of the resulting voltage going as the square root of N. The surprise was how they all played nice together --- I was rather expecting oscillations. Thereafter I filtered a good deal as well.

Fortunately random noise is uncorrelated power so it adds not like voltage. Don't always assume its all uncorrelated though if there is way of external interference mixed in. That one will be correlated.

 

[anatech]

Hi Salas,

Although no one I know of does so, with sufficient amounts of low rbb' bipolars one could contrive very low noise band gap references. I've been tempted to do this as an exercise as I have a few bags of the now-unobtainium 2SC3329 and 2SA1316.

Have you ever tried this with a transistor array? CA3046 or similar? I built a few. The IC collection of transistors keeps their temperatures very close, and they are also matched pretty closely. It would be much easier than what you have probably been up to.

But for not-too-critical temp drift, finding JFETs with a bias as an I source that results in a ~zero tempco of drain current,

Time consuming. I have played with this but have yet to actually use a JFET this way successfully.

One is tempted to say if you need this for an audio circuit you're doing something wrong,

Yes, I almost posted that comment myself. There is a thread on a Nelson Pass inspired JFET preamp that a member found he could drastically reduce distortion by carefully setting the supply voltages to specific voltages. I think it would have been slightly different depending on the parts (even of the same type).

In the case of these shunt regulators and other related circuits, and with the use as described by Preamp of both polarities, it's a natural to use such I sources to develop the reference voltage, again with low noise gain. I'm not going to add to the fray but one can see the way to a few more parts and removing the tempco of the input amplifying device by using a differential pair, biased from the opposite rail, a current mirror, and thus push-pull dirve of the DMOS shunt device. I simulated a few versions early this AM and it looks like they avoid start-up problems and should have exemplary noise performance.

That actually sounds like a worthwhile project. The negative rail is normally the most sensitive for noise when it comes to Op Amps. You should develop this further.

-Chris

 

[anatech]

Hi Brad,

The other thing to remember is that averaging the noise from a plurality of references always works.

It sure does. I'll be trying this with some of the new voltage references.

I stacked a bunch of bog-standard 431s where I needed a fairly large reference voltage for a series regulator, and results were as expected, with an overall improvement in the "signal-to-noise ratio" of the resulting voltage going as the square root of N.

I hadn't considered stacking the 431's. I have a hundred of the silly things and hadn't considered stacking them. I briefly thought of hanging them in parallel using one set of resistors to set the voltage, but was very concerned about them oscillating.

Tell me, how was the temperature coefficient of your stack? I have similar problems and stacked zeners have one heck of a tempco. I need a 275 V regulated voltage, and the gain from the error amp really amplifies the zener drift. My current set up has about 100 V of zener with thermistors in contact with the zener's body and the drift is now "okay" but not great. A simple 270V stack of zener diodes has less thermal drift. I'm seeing between 1 and 2 volts now.

Thereafter I filtered a good deal as well.

Given that the 431 is an active device with gain, a solid idea!

-Chris

 

[anatech]

Hi Salas,

Take an LM329 and a generic red Led to your HP SA for instance and you may find that some Led in your random collection could show bit less nVrtHz and less harmonic noise than the LM329 especially when the Led or a string of those match the JFET's CCS impedance.

Interesting idea. I normally use a red LED and a transistor as the core of a CCS. I haven't gone for the best, low noise source going. It's just a big improvement over a resistor most often. The little assembly slides in where the resistor used to be, plus a contact to another voltage or common, so one flying lead on top and the regulator you worked on just became much, much better. Bang for the dollar is high with this.

-Chris

 

[bcarso]

Hi Salas,

Have you ever tried this with a transistor array? CA3046 or similar? I built a few. The IC collection of transistors keeps their temperatures very close, and they are also matched pretty closely. It would be much easier than what you have probably been up to.


-Chris

Alas the CA3046 from most vendors has horrendous popcorn noise. SGMicro (then SGS) made one for a little while with much higher beta and lower noise, although their datasheet was just a copy of the ancient RCA one and didn't show this substantial improvement. I stumbled upon it and have a few, but they discontinued them, so I can hardly specify them for anything. In those days I used that array for a host of things, including a band gap reference and an input diff pair, using other transistors as heaters and thermal sensors. Worked o.k. but did nothing for the high noise.

Today we have the decent but expensive THAT Corp. chips, which I wish had lower rbb' and higher beta. I spoke to them about this and one told me that they'd tried, but that the breakdown voltages had varied all over the place.

 

[anatech]

Hi Brad,
Some of those old RCA numbers are still available from Intersil (I think) as surface mount parts. I was able to find some of those, but my interest was more to the RF mixers. I only bought a couple as playthings.

There were a bunch of really cool parts as arrays that died due to lack or interest. I only discovered them while searching for current sources. They had some current mirrors that I'm sure would have been used had people only known about them. Too bad that many of those parts died without ever being known.

-Chris

 

[bcarso]

Hi Brad,

It sure does. I'll be trying this with some of the new voltage references.

I hadn't considered stacking the 431's. I have a hundred of the silly things and hadn't considered stacking them. I briefly thought of hanging them in parallel using one set of resistors to set the voltage, but was very concerned about them oscillating.

Tell me, how was the temperature coefficient of your stack? I have similar problems and stacked zeners have one heck of a tempco. I need a 275 V regulated voltage, and the gain from the error amp really amplifies the zener drift. My current set up has about 100 V of zener with thermistors in contact with the zener's body and the drift is now "okay" but not great. A simple 270V stack of zener diodes has less thermal drift. I'm seeing between 1 and 2 volts now.

Given that the 431 is an active device with gain, a solid idea!

-Chris

I was advised by one in diyaudio to use the Sipex (now part of Exar) version, the SPX2431, for lower noise. I haven't as yet. I believe they are SM only. A comparison of datasheets with TI indicates that the recommender spoke sooth.

Tempco of the stack of ten TL431 to produce 25V was afaik the same as a single part as a percentage of total volts. If drift was Gaussian-distributed one would expect it to improve along with the percentage noise, but usually you'll have a batch effect so that won't help --- that is, the drifts will be correlated and displaced from the datasheet mean value.

Of course the impedance of the stack is N times the individual device's, so in that sense it's not a completely free lunch. But with no direct cap bypassing them, I saw no signs of instability, and as mentioned R-C filtered to the reference input of the regulator. The Z of the stack at 2 ohms or so is still useful, although I pre-regulated with another conventional zener. Indications are another 431 with gain would have worked better as the prereg.

Zeners are lousy for temp stability unless around 4.7V, and unfortunately that voltage is usually quite noisy. The temp compensated reference zeners of old used a voltage with a positive tempco and compensated with a standard forward-biased diode. They had to be operated very close to the reference current to deliver the stated performance. PITA.

 

[bcarso]

Hi Brad,
Some of those old RCA numbers are still available from Intersil (I think) as surface mount parts. I was able to find some of those, but my interest was more to the RF mixers. I only bought a couple as playthings.

There were a bunch of really cool parts as arrays that died due to lack or interest. I only discovered them while searching for current sources. They had some current mirrors that I'm sure would have been used had people only known about them. Too bad that many of those parts died without ever being known.

-Chris

The Intersils I believe closely follow the original RCA parts and although nice they exist, are really nasty-noisy. But for some applications o.k., and the parts are fast --- about f sub t of 500MHz if memory serves.

RCA did a nice business with that part, advertising it iirc as the Economy Five. My father, who was in the switching system business, dressed up his old boards with them in lieu of discrete parts. He felt the discretes looked too old-fashioned.

Some of the RCA parts included PNPs as well, but I believe in keeping with the processes of the day were laterals and not much to write home about. EDIT: I finally remembered the number of the cheap one, without guaranteed matching: CA3086. I still have dozens of them.

 

[bcarso]

JFETs as reference current sources

Time consuming. I have played with this but have yet to actually use a JFET this way successfully.

-Chris

If one were to get serious about using JFETs as reference current sources, you would do well to have something to servo between a couple of temperatures, and adjust the source resistor until the two temps produced a null result. A thermoelectric heat pump would work well, used in a loop with a temp sensor.

When I have resorted to this determination, I simply bring a soldering iron close to the package, observe the change, readjust the 10-turn pot, and let it cool. It takes a while. The canonical operating point according to Cobbold for the zero tempco is about 630 mV more turned on than the cutoff Vgs. This could be a useful presort.

In the days of few-volt Vth for FETs, the zero tempco point was usually still not that close to zero, i.e., Idss. But short-channel devices like the 2SK170 actually have such a low Vth that the zero tempco point occurs close to zero, and for a bunch of BL parts I have, actually requires a slight forward bias on the gate! I wasn't able to get V parts, but some of them are probably zero tempco at zero volts. However, these are not the best devices for high drain impedance.

I have an ammo pack of an obsolete medium-channel-length NJFET, the 2SK381. Paralleling a few of them (no matching required) yields a composite part with useful net current for a not-too-large composite source resistor. This cascoded with a J111 is quite useful.

Simulations with instead an SK170 and J111 developing about 15V across a bypassed R (not necessarily zero tempco) indicate that the noise contribution of the reference thus will still dominate the shunt regulator noise for everyday moderate-rbb' bipolars in the input of the error amp --- but not by much. If the error amp diff pair is made from super-low rbb' parts run at highish collector currents (say ~7mA) the reference strongly dominates. But the predicted noise voltage at the shunt output is very very low. It's almost worth building and measuring just to confirm, although I would use some of my other JFETs.

Beyond this one gets into battery territory and the work done to measure them at NIST.

 

[anatech]

Hi Brad,
Thanks! A ton of useful information and confirmation of what I've heard and found on my own. I may try to get a-hold of some SPX2431 parts. I have been trying to transition my work to surface mount where it makes sense. The hit you take on heat dissipation is tough to swallow. You have to really look at the dissipation of existing designs, plus the lack of airflow that comes with sm technology. This can force some of the standing currents below where you would like them.

Tempco of the stack of ten TL431 to produce 25V was afaik the same as a single part as a percentage of total volts.

That's what I found with Zener diodes too. The tempco was strongly negative, more than I expected. Should have figured the drop out via calculator first. The TL431 was a part I forgot about completely, even though I had tons of them. I just didn't consider them for a higher voltage application. Duh!

The temp compensated reference zeners of old used a voltage with a positive tempco and compensated with a standard forward-biased diode. They had to be operated very close to the reference current to deliver the stated performance.

Yes, they were a little higher breakdown voltage. I simply got in the habit of running them at their test current - which made them warm. A real PITA for sure.

-Chris

 

[anatech]

Hi Brad,
Thanks! A ton of useful information and confirmation of what I've heard and found on my own. I may try to get a-hold of some SPX2431 parts. I have been trying to transition my work to surface mount where it makes sense. The hit you take on heat dissipation is tough to swallow. You have to really look at the dissipation of existing designs, plus the lack of airflow that comes with sm technology. This can force some of the standing currents below where you would like them.

Tempco of the stack of ten TL431 to produce 25V was afaik the same as a single part as a percentage of total volts.

That's what I found with Zener diodes too. The tempco was strongly negative, more than I expected. Should have figured the drop out via calculator first. The TL431 was a part I forgot about completely, even though I had tons of them. I just didn't consider them for a higher voltage application. Duh!

The temp compensated reference zeners of old used a voltage with a positive tempco and compensated with a standard forward-biased diode. They had to be operated very close to the reference current to deliver the stated performance.

Yes, they were a little higher breakdown voltage. I simply got in the habit of running them at their test current - which made them warm. A real PITA for sure.

-Chris