LM399 based 10 V reference

[iMo]

A practical advantage of a fast opamp buffer (with built in current limit) would be a faster response to some spikes coming from the DMM. As has been discussed here, people claim there are such pulses coming off the DMM's inputs (I've seen a picture in the "Sampling with 3458A" book, like 15nA and couple of us long coming off the 3458A, but it could be anything).

What are the practical advantages (no theory please!) of having these discrete transistor output buffers over using a buffer opamp?

Just wondering if I should change my design...

[iMo]

Some notes to my "MOD" above:
1. the resistor in the heater could be 10-33ohm, I would definitely put there something with TESLA's clones, indeed
2. C27 100pF - what is the role of that cap there? The simulation shows no impact on the stability, afaik
3. I would put a 100-1k resistor into the Q8 base, thus there will be no "direct path" from 15V to 10V output
4. I would go with a higher Vcc (when possible), like 18 to 24V.

[David Hess]

What are the practical advantages (no theory please!) of having these discrete transistor output buffers over using a buffer opamp?

A discrete buffer can be less expensive and provide a lower open loop output impedance, although that is not the case with this example.

[argintviu]

Thanks for all the answers and suggestions.

I'll add the RC damper, lower C16 a bit and add a buffer for the raw 7 volts as well. It's always nice to be able to push out some milliamps. And then the output can also be protected with a TVS.

It's sad to hear about the MAB399 heater problems. I have 2 pieces, both have been aged for ~1000 hours on a current limited power supply. It was set for about 100mA and that might have protected them on each start. Inrush limiting as suggested by iMo is a must then. Would be interesting to see what failed inside of a reference with dead heater. I don't think this part saw much use because it arrived too late (1990 judging by the date code), when the soviet economy was already down the drain. No use means not much testing, so it could be flawed .

It might be wise to buy a LM399 when found at a decent price then, to avoid the nightmare of the MAB stopping working after thousands of hours of aging. What would be better, a vintage NatSemi or a new LT?

Regarding the 100pF C27 cap, it's not included in similar circuits. I think it speeds up the response of Q5 in case of a sudden short.

As for the input voltage, I already bought the 18 VAC transformer some time ago so it can't be much higher, but it could probably do 18V DC out.

[mawyatt]

It's sad to hear about the MAB399 heater problems. I have 2 pieces, both have been aged for ~1000 hours on a current limited power supply. It was set for about 100mA and that might have protected them on each start. Inrush limiting as suggested by iMo is a must then. Would be interesting to see what failed inside of a reference with dead heater. I don't think this part saw much use because it arrived too late (1990 judging by the date code), when the soviet economy was already down the drain. No use means not much testing, so it could be flawed .

I'm sure you are aware but others may not, a typical Power Supply usually has significant output capacitance, usually the better quality ones this is minimized. When the active supply is connected to the DUT in CC mode, the supply has achieved an initial voltage (Voltage Limit) across the output capacitors before connection. If this is set much higher than the DUT requires (a dynamic low impedance) significant energy discharge into the DUT can occur, from prior work (EMP sensitivity) the amount of energy required to damage a typical IC is on the order of just 10mJ on a single IC pin.

Seasoned folks usually set the PS Voltage Limit just slightly above the nominal required DUT voltage, and why quality accurate Power Supplies are desirable for development work.

Best,

[iMo]

..It was set for about 100mA and that might have protected them on each start.
..I don't think this part saw much use because it arrived too late (1990 judging by the date code), when the soviet economy was already down the drain. No use means not much testing, so it could be flawed ..

I had none current limit and lost perhaps 5 pieces out of a dozen.
The part was in the development in late 80ties at Czechoslovakian TESLA, came into production perhaps around 1990, and soon gone with the decline of TESLA corp, where the microelectronics divisions were later acquired by Motorola/OnSemi, afaik.

There were Chineese and Soviets clones as well - https://www.eevblog.com/forum/projects/tesla-mab399-mac199-voltage-references/

Noopy did the die shot of the MAC199 I sent him too - https://www.richis-lab.de/REF02a.htm

TiN messed with the MAB399 a lot - https://www.eevblog.com/forum/metrology/tesla-mac199mae299mab399-and-mac01-teardown-and-tests/

[2N3055]

With LM399 PSU current limit does not matter because heater is driven by current limited circuit internally... Sometimes external current limit was used to limit the spike for various reasons, but it is absolutely unnecessary for LM399 survival.

Maybe TESLA 399 clones had something different in that part of circuit. And so these need some current limiting...
Thanks iMo for the info!

P.S. as a kid I had TESLA bass guitar amplifier. It had just average sound, but was built like tank.. Undestructable ...

[iMo]

Btw I compared the noise (0.1-10Hz) of the MAC199s (one of them with broken heater) with the LM399A/Hs and they were the same..
https://www.eevblog.com/forum/metrology/low-cost-lm399-noise-indicator/msg5055250/#msg5055250

PS: My bet all MAB/MACx99 refs available today at ebay or in various eshops were already carefully "sift through" (mind they are 30+ years out of production), also the markings on their hats might be replaced/forged. All chips/dies inside are "MAC199" afaik, and the "MAB/MAC/MAE" versions were just stamped onto the easily replaceable plastic hat after the binning during the testing..

[argintviu]

Interesting point mawyatt. I must admit I didn't think of this before messing with the references. In my setup, there was the K7200 diy PSU (100uF output cap), ~1.5m of wires and a small board with 2 refs that also had a 10uF capacitor and a sliding switch. I tried to recreate the possible surge in LTSpice. Oddly enough the current pulse on the heater pins is bigger with the 10uF cap than without it. The pulse current seems to largely depend on how fast the switch is closed. If the rise time is faster than 1 ms, the chip would be in trouble (transients get up to a few amps).

Thanks for the information about the clones iMo. Those Chinese and Russian ones must be really rare, haven't even found pictures around.

P.S. as a kid I had TESLA bass guitar amplifier. It had just average sound, but was built like tank.. Undestructable ...

I am very fond of TESLA, had a B100 stereo reel to reel player, it sounded great.

PS: My bet all MAB/MACx99 refs available today at ebay or in various eshops were already carefully "sift through" (mind they are 30+ years out of production), also the markings on their hats might be replaced/forged. All chips are "MAC199" afaik, and the "MAB/MAE" versions were just stamped onto the easily replaceable plastic hat after the binning during the testing..

Who knows... The probability for forgery might be low though, since they are cheap and not that popular. There's still one shop in the Czech Republic selling them for 1,18 EUR / piece, but they sell only locally. The MA* dies are indeed the same. In the factory I think they just sorted for tempco, so there's not a big difference between the MAB and MAC versions regarding noise / other parameters.

[edit]
Found a few meters that used them, the Metra M1T 390 (4.5 digits) and M1T 380 (5.5 / 6.5 digits, quoted as the most advanced meter produced by Metra Blansko). Some pictures are available here: https://forum.elektrolab.eu/resources/metra-m1t-380.629/. I found the schematic for it and in the part list they wrote LM199 but in reality they used MAB399 (picture 8 in the link above). In any case, I was curious to see if they limited the surge current somehow, but nope, the heater is simply connected to 15V (generated from a MA7815) and that's it. The zener is supplied from 20V (generated by a MAA723) via a 12K resistor, so 1mA current.



[edit2]
On a second look, there is actually a current limiter for the 15V rail (that powers the heater of the reference) in the M1T 380. At first I thought it was a capacitance multiplier, but it actually limits the current to about 375 mA before the input of the MA7815. This may not help the reference too much since I doubt the heater surge is bigger than 150-200 mA on a cold start. I will add a current limit in my circuit as well at the 15V regulator.

[iMo]

Your Metra link says the M1T380 meter was produced a) with "standard stability" - with the TKZD13/D zener (we discussed here the samples), and b) with "enhanced stability" - with the MAB399.
Now, the big question is how many b)s they actually produced (as Metra Blansko stopped its production in 1991 - in the same time as the MAB399 arrived)..
https://www.eevblog.com/forum/reviews/czechoslovakian-measure-equipment-collection/msg85691/#msg85691
PS: the meter in your link (the teardown pictures) is using the TKZD13 zener. Frankly, I doubt there were any Metra meters with the 399 ever sold..

[argintviu]

Interesting read, seems like the parts were notoriously unreliable. I bet not many of those multimeters were made, possibly a few tens at most. The other components like the CPUs were probably in high demand at that time, slowing down production.

In any case, the high failure of your 199/399s, the cracked dies found in MAA723s and MetraCollector's testimony make me think that some eastern components are better suited as a collectible than for actual use. I'll search for some proper LM399s, just to be on the safe side. If this project gets completed, I'll make a new thread and present it there, so this one stays clean.

[Edit]
Regarding the picture, see "D1639.jpg". The MAB is in the middle towards top right.

[iMo]

I got three Metra 4.5-6d meters in my hands (long time back) for free. All not working. Spent a week elaborating them. At the end of the day I removed the ww resistors, some transistors and analog chips for my collection, the reed relays, and the TKZD zeners. The rest went into the shredder..
PS: ..and the D1791.jpg is with the zener..
But nice to see MAB399 in an actual application 

[argintviu]

Still playing with the circuits in this thread. Went back and read all the posts that I missed, but it made things worse because now I can't decide which output buffer to use, opamp (e.g. NE5532) or current limited emitter follower .

There is another thing that I'd like to do before starting the build, a way to know if the heater is operating correctly. From what I gather, the LM399 gets from ambient temperature to ~90C in a few seconds. I imagine that the die is subjected to a big amount of thermal stress and that the heater could sometimes break, especially with some parts (like iMo said above). If the reference is placed inside a case, under thermal isolation, it would be hard to know if the heater is working or not.

A "passive" way to implement this would be to place a NTC close to the pins of the 399, on the PCB island, and then have a comparator circuit made of a jellybean op-amp light a LED if the heat gets over a threshold. But I don't think that there is enough heat at the legs to work reliably, especially on a hot summer day when the ambient temperature might trigger the comparator. Maybe there's a better way to do this?

[dietert1]

As the heater gets so much hotter than ambient temperature, heater power is roughly independent of ambient temperature. So inserting a 1 Ohm resistor into the heater supply and looking at the heater current with a dual comparator should be enough to assert correct heater operation.

Regards, Dieter

[iMo]

Also do not put anything with low thermal resistance close to the 399 pins as it will increase the thermal flow (the "something" will cool down the reference). It is a recommendation of ADI afaik (and I saw the effect - an increase of stddev - in-situ here).

[IconicPCB]

The "something " could be a small incandescent bulb akin to the wien bridge nonlinear gain element.

[argintviu]

Thanks for the ideas. The shunt + comparator might be the best way. And a bigger value resistor might also work as an inrush limiter. I'll follow the advice and not put stuff around the legs then.

[argintviu]

The heater current monitor proved to be a bit more difficult than expected because one needs to work close to the 15V rail and a simple comparator + shunt won't work. Luckily there is a nice high side current sense IC (ZXCT1009) that makes life easier, so here's a try at a monitoring circuit. It will light up an LED when the current through the heater is smaller than 5mA.

[iMo]

The ZXCT may indicate to you the temperature inside the box - the heater current changes with the surrounding temperature.

[dietert1]

MAX4373/MAX4374/MAX4375 supervisors may be interesting, too. They include the high side shunt amplifier, a reference and two comparators. Max input voltage is 28 V.

Regards, Dieter

[David Hess]

OPA2186 took my atention due to extremely low price. I'm reluctant to use it blindly since it seems different from other TI modern Az opamps. The max voltage is 24V. All other TI AZ OpAmp members have 36V. Is it different architecture or different Fab process? - I will order few and evaluate them.

I am in the process of updating my preferred operational amplifiers list and I have used the low power OPA187 recently.

I suspect that the OPA186 has a lower supply voltage because its input stage is different from the other parts in the series to support rail-to-rail input operation.  My guess based on the offset voltage versus common mode voltage graph and input bias current is that both PMOS and NMOS stages are used in parallel and a lower gate-to-source voltage rating of one limits the supply voltage.

[dietert1]

Thanks, i was looking for a zero drift opamp with input rail to rail and couldn't find one. All i found excluded the upper 1.5 to 3 V of common mode voltage or limited operation voltage to 5 V.
The OPA186 datasheet figure 6.7 seems to indicate some anomaly at the lower 3 V of the input range. Maybe that disappears at lower supply voltage. Anyway now there is a choice..
The f = 0.1 Hz to 10 Hz input noise spec 125 nVRMS for 90 uA supply current is amazing.

Regards, Dieter

[macaba]

I am in the process of updating my preferred operational amplifiers list and I have used the low power OPA187 recently.

I would like to hear more about your list when you are ready. I would suggest adding LTC2057 to the evaluation list - I recently did some evaluation on a group of zero drift opamps, measuring the input bias current over CM voltage range, whilst varying the R/C of input/feedback. LTC2057 was the only device to have both: 1. Nearly constant input bias over CM voltage range, 2. Maintain that constant input bias over a variety of R/C conditions. It was quite astonishing to see how badly behaved other parts were under various conditions and LTC2057 would always have a flat clean line on the charts.

[David Hess]

Thanks, i was looking for a zero drift opamp with input rail to rail and couldn't find one. All i found excluded the upper 1.5 to 3 V of common mode voltage or limited operation voltage to 5 V.
The OPA186 datasheet figure 6.7 seems to indicate some anomaly at the lower 3 V of the input range. Maybe that disappears at lower supply voltage. Anyway now there is a choice..

Also the input bias current get significantly worse starting at the same low common mode input voltage.  For operating near the negative supply, there are better parts.

Maxim had a chopper stabilized part which used an internal charge pump to get clean rail-to-rail input operation.

The f = 0.1 Hz to 10 Hz input noise spec 125 nVRMS for 90 uA supply current is amazing.

I am having trouble replicating the 0.1 to 10 Hz noise specifications in the recent Texas Instruments datasheets from their input voltage noise spectral density graphs and specifications, and Texas Instruments has a history of lying on their noise specifications, so I would not trust them blindly.  I would hate to build a dedicated 0.1 to 10 Hz testing jig, but I may have to.  It is not that difficult to do, but I currently lack a suitable sampling multimeter to make the measurement.

That 125 nVrms specification should be about 0.75 uVpp so it is by far the noisiest part in that series of chopper stabilized parts, and I suspect that is why they used nVrms instead of nVpp like the others.

I would like to hear more about your list when you are ready. I would suggest adding LTC2057 to the evaluation list - I recently did some evaluation on a group of zero drift opamps, measuring the input bias current over CM voltage range, whilst varying the R/C of input/feedback. LTC2057 was the only device to have both: 1. Nearly constant input bias over CM voltage range, 2. Maintain that constant input bias over a variety of R/C conditions. It was quite astonishing to see how badly behaved other parts were under various conditions and LTC2057 would always have a flat clean line on the charts.

I will add the LTC2057 at least for comparison purposes.  I am trying to get away from parts supplied by Analog Devices because their prices have increased so much.

I was unpacking my electronics stuff after moving and reorganizing everything, cleaning parts cabinets, etc., and decided to go through my prototyping parts in detail because so many are either no longer produced, have outrageous prices, or now have better alternatives.

[argintviu]

Back with a (stupid) question. What is the difference between using a PI bifilar common mode filter on the output vs. using separate cores for each wire like in older voltage standards (e.g. Cropico ESC1)?