LM399 based 10 V reference

[Kleinstein]

The common mode choke would mainly act on the common mode signal, e.g. to filter a so perfectly isolating DCDC converter and reduce an injected signal from there. The output regulation is much less effected (the CM mode chokes still have some differential mode inductance) and can still be fast (e.g. with somewhat transisent loads).
With separate inductors the filtering would also effect the output drive / regulation speed. Transient loads may lead to more LC ringing and slow regulation. This could be an issue with separate drive an sense terminals.

[dietert1]

Common mode filters are usually made with high inductance cores that are less useful for making chokes as they easily saturate. So a typical inductance for chokes like those in the image is 10 to 100 uH, while a typical common mode choke is at 5 or 10 mH.
I am sometimes using Würth 744866104 common mode chokes with 100 mH. Those have a stray inductance of about 1 mH.

Regards, Dieter

[iMo]

..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..

Provided all three coils are perfectly identical (incl the winding direction) there is none difference.
As on your above shot the proper way would be to use single toroidal core with trifilar winding. Also the return path wire (ie gnd) must be included (the number of wires/signals in the single CM choke is unlimited, ie you may pass entire flat cable with say 100 different signals and their returns through a single CM choke).

[argintviu]

Ah, that's good to know. So I can put the 10V, 7V and GND signals through a single choke instead of using 2 (i.e. 10V and GND on one core and 7V and GND on another). Helps save a few parts.

[Kleinstein]

..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..

Provided all three coils are perfectly identical (incl the winding direction) there is none difference.

3 separate coils are different. They could have the same effect on the common mode signal, but the effect on a differential signal is quite different, with much more inductance seen to the differential signal.
Separate coils (especially if not identical) also tend to have stronger coupling from the CM to the differential signal.

[Birb]

Hi, here's an attempt at designing a lm399 based Vref. Note that the OP077 will probably be replaced with something else and is currently a placeholder.
Also, the input voltage is 15V, which gets regulated by an LM7812, giving sufficient headroom for a lower noise.
Lastly, the entire reference section should be heated by a small PTC heater to around 50C. It should be good enough to keep the resistors more stable (Using weird BWL resistors - they seem to be ok, but I don't trust the tempco).
Though I worry that the buffer may not be sufficient?
Any feedback is appreciated, thanks a lot.

[dietert1]

R2 and R3 are exchanged.
R4 should be about 3K instead of 10K.
U3 needs a heatsink and is mounted outside the oven. A higher supply voltage is preferred, like 15 or 18 V. 15 Vac will give you about 22 V from the rectifier. Depends on the heater power.
U4 should be mounted close to the PTC.
10 V tuning is better with fixed resistors (as far as possible).
Concerning the output buffer it depends on what will be the application of this reference.

Regards, Dieter

[Kleinstein]

An additional oven around the LM399 is tricky, as the reference starts with a relatively high power and the thermal insulation thus can not be that good. So one would end up with a rather high heater power.
The LM399 itself hardly need the temperature regulation and the gain stage it depends. Same series resistors may show some compensation of the there TC. Precision use normally is also done with not so extreme room temperature, as the other instruments would also be effect from the temperature.

[argintviu]

@Birb, you could try to ovenize just the resistors / opamp and simply thermally isolate the 399 (top and underside). Kind of how branadic did with the QH40 crystal heater on top of the gain resistors here.

[Birb]

Thanks for the suggestions!
I've revised the schematic like so, and added a jumper for a small NTC temperature sensor if needed.
In addition, I added a low pass filter on the output of the LM399 (Is this worthwhile?), and also removed the trimmer stuff. (One lm399 is like ~7.001V which makes it close enough to 7V that I don't need to do much trimming)
The original idea was to ovenize just the resistors  (and possibly the op-amp) with a PTC heater, then encapsulate everything in some insulative foam.
The LM399 will be placed separately.
Unfortunately I do not have another 0.1% ~3k resistor to spare, so I'll use 3 0.1% 1k instead. (Originally I wanted to parallelize 3 10k, but I don't have that)
For the buffer, I think it doesn't need to be high current? I only intend for it to be a voltage reference for testing purposes.
As for the rectifier, I've decided to use a discrete module, which seems to take AC 18V and turns it into DC 24V. I may possibly need to find a small enough transformer to put into a case.
A few questions:
How can I know the current flowing through the zener diode based on the resistance?
Also, why should U4 be located near the PTC?

[mawyatt]

Thanks for the suggestions!
I've revised the schematic like so, and added a jumper for a small NTC temperature sensor if needed.
In addition, I added a low pass filter on the output of the LM399 (Is this worthwhile?), and also removed the trimmer stuff. (One lm399 is like ~7.001V which makes it close enough to 7V that I don't need to do much trimming)
The original idea was to ovenize just the resistors  (and possibly the op-amp) with a PTC heater, then encapsulate everything in some insulative foam.
The LM399 will be placed separately.
Unfortunately I do not have another 0.1% ~3k resistor to spare, so I'll use 3 0.1% 1k instead. (Originally I wanted to parallelize 3 10k, but I don't have that)
For the buffer, I think it doesn't need to be high current? I only intend for it to be a voltage reference for testing purposes.
As for the rectifier, I've decided to use a discrete module, which seems to take AC 18V and turns it into DC 24V. I may possibly need to find a small enough transformer to put into a case.
A few questions:
How can I know the current flowing through the zener diode based on the resistance?
Also, why should U4 be located near the PTC?

Check the left end of the three 1K resistors, that end should be connected to the LM399 and not thru the 5.6K resistor.

Also using a 12V regulator might be too low a voltage for the Op-Amp, suggest 15V minimum since you have 24V available.

Best,

[dietert1]

I proposed to put U4 close to the PTC as it sees roughly the same current, yet at a higher voltage = 12 V - 5 V = 7 V. So its heating power is always 1.4 times the heating power of the PTC. Together they will be like a PTC operating at 12 V with 2.4 times the power.
At 24 V in  and 12 V out U3 will about replicate that heat plus the LM399 heater power.

Regards, Dieter

[iMo]

..it has no sense to mess with 3x 1u in that filter.. 100nF is enough for some high freq components.. R5 could be smaller like 1k..

For the buffer, I think it doesn't need to be high current? I only intend for it to be a voltage reference for testing purposes.

"Testing purposes" - that is a pretty good reason to protect the opamp's output (overvoltage/overcurrent).. Your opamp's output current is 2mA now, with an ADR1399 it will be 4.5mA, the strategy is to have the opamp less loaded, so an output buffer is to recommend in your case.

[ivo]

One neat thing I realised from Birb's schematic is that actually the common E24 values 3.3k and 7.5k actually make a pretty good divider for the nominal LM399 voltage. It actually works out to 6.944V! 3.3k then also gives just under 1mA reference current as well. Or two 3.3k's in parallel would give 1.8mA which should still be fine, even possibly preferable. Only thing is that the divider is also drawing 1.6mA from the opamp as well.

Watch out that LM399's 'nominal' current source is 1-2mA in most designs (although the datasheet does indeed note up to 10mA) whereas the ADR1399 datasheet has bumped the recommended current up to 3mA (up to 10). In truth, for a design trying to accomodate either, I'd guess either of them would run fine at 3mA.

One way to solve things cheap is just a dual opamp. One is doing all the heavy lifting with the 3 or 4 resistors to do the voltage boost and zener current, the other you just wire as a unity buffer from the 10V point to actually get your reference from. Although the package as a whole constantly doing 3-4mA isn't idyllic as voltnuts would prefer, it probably still works fine... (testing required...)

The possible reason I'd see to use 3x1uF is to get down to a true 10hz filter. 5.6k/3u gives a 9.5Hz 3dB cutoff; and film caps start getting a lot bigger after 1uF. Is a full 10Hz low pass really an improvement worth going for? Probably would need careful measurement to pass judgement if it noticeably helps over something more modest.

[Kleinstein]

How much filtering is sensible depends. Some tests may reactor to the noise in the 10-30 Hz range (e.g. DMMs in 1 PLC AZ mode). It is OK to include the space for the slightly larger film capacitor.

If a separte driver is used, this could be a transistor (the current to the diver and ref. drive is aready quite a bit) or a separate OP-amp. A dual OP-amp makes little sense as the heat would still be in the same case and many precision amplifiers (e.g. OP07) don't even come as duals.

With the LM399 the resistors to provide the reference current are not that super critical. The drift of this resistor is attenuated something like 1000 - 10000 times with the rather low output impedance. The main part to worry about is the divider for the 7 to 10 V step and than the question if one also wants a 7 V output with or withput an extra buffer.

[iMo]

..
Watch out that LM399's 'nominal' current source is 1-2mA in most designs (although the datasheet does indeed note up to 10mA) whereas the ADR1399 datasheet has bumped the recommended current up to 3mA (up to 10). In truth, for a design trying to accomodate either, I'd guess either of them would run fine at 3mA.
..

It does not matter much how large the "outside" 399 zener current is like, simply because there is none single zener inside - but rather a lot of other stuff around the zener as well. In 399 the zener current is set internally to 250uA, in 1399 they finally increased the current and set it to perhaps 1-2mA or something like that.
Anything above these internal settings is shunted down by the circuitry around the zener. Whether you set 1mA or 10mA with 399 the zener current will always be fixed 250uA and with 1399 it will be fixed inside too.

[Andreas]

Hello,

long term stability is specified at 1 mA +/- 0.1%

with best regards

Andreas

[Birb]

Thanks for the suggestions.
I've added a transistor buffer, as well as a current limiter. This should prevent the outputs from getting shorted.
Currently, the op-amp is still  a placeholder, until I actually route the AD706.

I also intend to heat the op amp as well. Is this advisable?

[iMo]

There is a notorious output buffer/current_limiter (aprox 30mA max) you may find at least 100x here  which may help you. Your wiring is rather odd.

[Birb]

Actually, the major difference is that I am using two op amps, one for the amplifciation, and one for the buffering. The transistor buffer is largely the same as the second one, save for the zener diode, and the edits to make it work with one op amp.
Is it actually better to use one or would using two be better? Thanks

[iMo]

The single opamp plus the tansistors is better in your case as your first opamp feeds the 399's (or 1399's) zener directly which adds 8mW of heat to it. Better to move the heat off the first opamp, let heat up the transistor.
Also with 2 opamps their drifts may add up in an unwanted way.