ADR1399 reference

[iMo]

From the TP-xx DS they look like China clones of the OP07..27 series (and my bet when you order OPxx on Ali you get them like a sanded_off/rebadged TPxx). Do you have some practical experience/measurements with those opamps??

The actual device utilizes TP27-SR  op-amps at all position.

It is a chopper amp, isn't it?

Cheers

Alex

The TP27-SR datasheet does not explicitly mention 'chopper' or 'zero-drift' architecture, but its specifications clearly indicate that one of these topologies is being used. In any case, their real-world performance is an excellent fit for this circuit and will not become a bottleneck. Among the more common op-amps, the OPA189 would be my first choice to consider.

[bsw_m]

From the TP-xx DS they look like China clones of the OP07..27 series (and my bet when you order OPxx on Ali you get them like a sanded_off/rebadged TPxx). Do you have some practical experience/measurements with those opamps??

To put it briefly, I haven’t tested every single datasheet parameter, but my bench control of the key specifications — specifically peak-to-peak noise, input offset voltage, dynamic thermal response, input bias current, and unity-gain bandwidth — showed that the tested samples strictly align with their specified limits and are an excellent fit for this topology. Despite the part numbers implying a connection to the OP07/OP27 series, these op-amps are definitely not clones of those legacy bipolar designs.

As for AliExpress, I honestly have no idea what is being sold there as I do not source any components from that platform. Consequently, I cannot comment on any counterfeit or rebadged op-amps and ICs found in those listings.

[bsw_m]

Interesting approach. I like the idea too.
I try to figure out how the circuit works but i did not succeeded
e.g U8A seems on input to be gnd and another tied to zenner positive side let say +7V . Is this the intention?
do you vant to have ADR1399 Pin 1 close to GND and ADR Pin2 close = -7V

Yes, your assumption in the last sentence is absolutely correct. That is exactly how the circuit is intended to work. In this topology, the traditional biasing is inverted:ADR1399 Pin 1 (Cathode) is actively held at a dynamic virtual ground (0V) by the U8A bootstrap loop.ADR1399 Pin 2 (Anode) is driven down to a stable about -7V relative to the system GND. The primary objective of this inverted configuration is the complete isolation of the ADR1399 Zener from the system ground (GND). This stops any return currents or ground loops from affecting the reference core. As a nice additional bonus, keeping the cathode at 0V eliminates voltage differentials across that part of the PCB layout, effectively reducing surface leakage currents to zero.

[exe]

On the schematic, there is an internal diode between heater's -15V and zener's anode. As I understand, zener's anode is at about -7V. Wouldn't the built-in diode start conducting?

[bsw_m]

On the schematic, there is an internal diode between heater's -15V and zener's anode. As I understand, zener's anode is at about -7V. Wouldn't the built-in diode start conducting?

The internal substrate diode will be deeply reverse-biased by this voltage differential (Anode -15V, Cathode about -7V), so it cannot conduct under any circumstances.

[miro123]

Interesting approach. I like the idea too.
I try to figure out how the circuit works but i did not succeeded
e.g U8A seems on input to be gnd and another tied to zenner positive side let say +7V . Is this the intention?
do you vant to have ADR1399 Pin 1 close to GND and ADR Pin2 close = -7V

Yes, your assumption in the last sentence is absolutely correct. That is exactly how the circuit is intended to work. In this topology, the traditional biasing is inverted:ADR1399 Pin 1 (Cathode) is actively held at a dynamic virtual ground (0V) by the U8A bootstrap loop.ADR1399 Pin 2 (Anode) is driven down to a stable about -7V relative to the system GND. The primary objective of this inverted configuration is the complete isolation of the ADR1399 Zener from the system ground (GND). This stops any return currents or ground loops from affecting the reference core. As a nice additional bonus, keeping the cathode at 0V eliminates voltage differentials across that part of the PCB layout, effectively reducing surface leakage currents to zero.

ave attached the circuit that I built in the past. I am still in the process of evaluating and tuning the component values. In the next revision, I will apply your idea of using a virtual GND at the Zener.

From my experience, adding a ×1000 amplifier between GND and VGND has several purposes, and I will use it in future designs. Besides using it for validation, I use this output to calculate the confidence interval of the voltage-transfer measurement. It also allows you to immediately see whether you have EMI issues or distortion introduced by the measurement equipment.

I have one question for the forum members: How do you define the confidence interval for a voltage-transfer measurement? How do you determine how the voltage standard interacts with your measurement setup?

[exe]

Omg, I have an unexpected jump in voltage. It seems it occurred when I connected my hotair gun to power (with power cable lying near TEN399), and that caused ~5uV jump up. I just disconnected the gun, let's see tomorrow if I get the voltage back to the previous value.

When I'm not in the lab, the voltage plot is very predictable. But just me walking around and using my computer causes quite some disturbance. 

[TUMEMBER]

When I'm not in the lab, the voltage plot is very predictable. But just me walking around and using my computer causes quite some disturbance. 

It's nothing unpredictable; you're a low-frequency spatial field, also a "90 pF capacitor." It's good if you're literally naked, then most of the charge flows "to the ground." If you're wearing "plastic rags," you're an active generator of static electricity—this affects the environment. Some species of fish "identify their potential prey" this way. Note that this happens in water.
  A Faraday cage is recommended for such delicate measurements. The object being measured and the measuring instrument should be located there, separated from the interfering environment.
Gnathonemus petersii
https://atlas-zwierzat.pl/ryby-ktore-potrafia-wykrywac-pole-elektryczne/

[Alex Nikitin]

Spooky stuff  . Here are the latest jumps from my K34970A (LM399) , measuring 10V output from Fluke 731B overnight, in parallel to K34465AM (ADR1399) and HIOKI DM7275 (LTZ1000),  (10h in 0.5s intervals, NPLC10 on all three meters). The reference in the K34970A has quieten a lot since I received it, almost no jumps visible, however in this particular measurements not only two three jumps occurred over 10h, but the voltage noise level distinctly changed with the level reading jump. No change on other two meters, obviously. On the graphs are the complete set of data, including F731B temperature (it has a TC about -1.5 ppm/C) and all three meters measurements, plus two three zoomed parts, clearly showing that the jump is faster than 0.5s sampling period and that the p-p noise doubles to 2ppm after a jump up in the reference voltage ("down" on the measured value) and drops back to 1ppm p-p after a jump "back".

Cheers

Alex

P.S. - and before someone asks, there were no obvious external events, and I was not in the lab at the time, the first jump was from 2:15am to 2:48am, the second is ~15seconds long at 4:43am and the third is from 6:50am to 7:08am . The negative spike at 8:38am (Zoom4 graph) is a wire disturbance, not visible on the K34970A because of the way the lot wired (a separate screened wire directly from Fluke) but clearly visible at the same spot on the Hioki and K34465AM..

[iMo]

Yep, and when you will measure your 34465AM with the ADR1399 inside for a longer period you will get a jump too.. Below last 5 days of the 1399 with none jump..
The last jump detected was on 11th Aug, 3:30UTC..

PS: I work in the lab 0.7m off the meter setup daily, switching the big PC and lamps on/off couple of times a day. No jumps or spikes induced by that activity in the graph visible. You may see it on the temperature fluctuations when somebody is present in the room.. The last large T jump is when I opened the window as it is nice cold outside this morning.
 
Sampling @ 100PLC, the TC of the battery is -34ppm/C..

[Alex Nikitin]

Yep, and when you will measure your 34465AM with the ADR1399 inside for a longer period you will get a jump too.. Below last 5 days of the 1399 with none jump..

Should I show you how my K34465A originally jumped as I've received it, a year and half ago? See attached  . I am pretty sure it might and most likely will jump at some point in the future, however so far it didn't do it for 18 months straight. As I've said, I have seen ADR1399 jump, just not in this unit, working 24x7 and logging most of the time.

Cheers

Alex

[exe]

So, disconnecting hotair gun from mains didn't change anything.

I wiggled wires a little bit and moved them and voltage dropped down almost to the previous value. So, two theories I have:
- bad contact somewhere
- EMI

I'll experiment if shielding changes anything.

[exe]

I added a shield from aluminum foil, connected to DMM's negative terminal (as, I believe, suggested in manual). The moment of attachment is what, perhaps, caused that spike in the middle. I don't see any changes. Maaay be slightly less noise. Let's see if there is any longer-term impact.

(Ignore the bottom plot, it was an attempt to do temperature correction)

I asked AI to calculate some stats, but I didn't verify correctness, so take it with a grain of salt.

Code: [Select]

                        before         after        change
spread (σ)         0.741 µV 0.506 µV −32 %
sample-to-sample noise 0.518 µV 0.383 µV −26 %
peak-to-peak         6.13 µV         2.64 µV         −57 %


[TUMEMBER]

Graphically, there are fewer "long spikes" visible on the right side of the chart, which is also visible in the peak-to-peak parameter.

There's progress in at least one parameter. If this "value" persists, you've at least partially eliminated (limited) the external interference.

[exe]

Definitely shielding helped. I include three plots, for last 1h, 24h and 168h. On 168h plot one can see when the shield was added (August 20).

[Alex Nikitin]

Some more "strange" results from my newly acquired Agilent 34970A. As I've reported already the original LM399 reference in it is somewhat "jumpy", not terribly so, a few times in a day, but annoying nevertheless and could be a problem in some situations. SO I've decided to upgrade the reference as I have few ADR1399 gathering dust. As I have little experience with the 34970A and none with 34401A, I've decided to approach the upgrade carefully, read the thread on 34401A and devised a script to measure the noise as much as practically sensible, logging 60 seconds of data at NPLC 1, 10, 100 and with the input voltages 0V , +/-1V and +/-10V (as appropriate for the range measured) from Fluke 731B with a polarity switch. The script calculates max/min/p-p in Volts and ppm, SDev in V in ppm for each run. Run this script on Hioki, 34465AM and untouched 34970A with reasonable results. It takes just under an hour to run the script as it needs a manual switching of the voltage and polarity.

So far so good. I opened the 344970A, replaced R409 with 1K0 MMELF (to provide required 3mA for the ADR1399) added the RC network, replaced the reference, warmed the unit up and run the tests. The results were, let's say, nothing short of a complete disaster. I did not expect much improvement in the noise and at NPLC10 / 100 the noise stayed roughly the same. At NPLC1 however the noise went up, and not by a bit, but about three times, both p-p and SDev. WTF?!  OK, removed the RC - stilll the same. Put the old reference back - all nice and quiet, even with 3mA current. So, perhaps something wrong with the ADR1399. OK, the second one goes in with the same horrible results. The table with the results is attached (SDev in ppm of range, the red number is the result of the LM399 jump).

In a bit of a panic I've looked through my stash of various suitable voltage references and found old used LM199A (833) and old NOS LM129A from National. OK, let's try both of these. The results are on the graphs (1V input, 1V range, 1NPLC, 300 samples each run, 2ppm/div (2uV) vertical scale. I am building an extender cable for the reference as it is very cumbersome to replace it in the unit and then to wait an hour for it to warm up! So far it looks like the old reference stays  !

Cheers

Alex

[Kleinstein]

When the reference votlage changes, one may want a new adjustment of the µC internal ADC scale. Ideally the µC inernal ADC would also scale with the reference voltage, but there may be some offsets. A not perfect adjustment there could cause some extra error for the 1 PLC case. I don't remember which CAL part also does the adjustment of the µC internal ADC.

The ADR1399 without the RC is likely no working well - if at all it would be adding a 2nd RC element (e.g. 10 µF + 10 ohms) for even more damping.

The 100 PLC may want a few more than 60 s worth of data to calculation the noise. Just some 15 readings is a bit low for that.

[mawyatt]

Alex,

Interesting!

Recall that HPAK used select references, that had been burned in and selected. So maybe not so surprising that an HPAK select LM399 with many years of "aging" might have less noise than a new random pair of ADR1399s.

Best

[Alex Nikitin]

When the reference votlage changes, one may want a new adjustment of the µC internal ADC scale. Ideally the µC inernal ADC would also scale with the reference voltage, but there may be some offsets. A not perfect adjustment there could cause some extra error for the 1 PLC case. I don't remember which CAL part also does the adjustment of the µC internal ADC.

The ADR1399 without the RC is likely no working well - if at all it would be adding a 2nd RC element (e.g. 10 µF + 10 ohms) for even more damping.

The 100 PLC may want a few more than 60 s worth of data to calculation the noise. Just some 15 readings is a bit low for that.

I've recalibrated DC ranges twice fully during these experiments, once for the first ADR1399, second back to the original reference at the standard current (put the resistor back as well for that time). No substantial difference in the 1NPLC noise was visible after each calibration, so to play with various chips I didn't do more calibrations (as it takes considerable time). I will play with this suggestion however if my idea of an extender cable works... .

Cheers

Alex

P.S. added  the results for the first ADR1399, taken after a full DCV calibration.

[bsw_m]

replaced R409 with 1K0 MMELF (to provide required 3mA for the ADR1399)

But using a 1 kΩ resistor for R409 will not give 3 mA in this circuit.

We also need to take the R441/R442 divider into account. With these resistor values, the approximate current left for the Zener will be around 2.8 mA, which is somewhat low for the ADR1399.

A more suitable value for R409 would be 920 Ω.

[Alex Nikitin]

But using a 1 kΩ resistor for R409 will not give 3 mA in this circuit.

We also need to take the R441/R442 divider into account. With these resistor values, the approximate current left for the Zener will be around 2.8 mA, which is somewhat low for the ADR1399.

A more suitable value for R409 would be 920 Ω.

 I've tried with 4mA as well, using 2x1K47 in parallel, with the same result.

Cheers

Alex

P.S. if we are to believe AD, even 2.5mA should be enough for a fully warmed up reference.

[iMo]

I've been running the 1399 in my 34401P with 820ohm (10ppm/C) resistor.
The 1399s are usually by aprox 70-100mV lower than the 399s.

Now - with "RC" you talking the "snubber" I assume, I also experimented with "RC low-pass" in front of the AD706 there (finally I decided not to have the RC there).

Without the snubber I saw oscillation like "telegraphing" with 1399 as well as with MAC199 in my "noise meter" in past, so I would place the snubber there for sure. I have not seen an effect of the snubber with the 399.

Another thing I observed in past is following - any wiring too close or direct to the 1399's solder joints at the bottom pcb side increases the noise fluctuations. Therefore I wired the 1u MLCC (it should be 25V one - mind the decrease of its capacitance with voltage there) via piece of manganin wire with aprox 5-6ohm resistance.

Btw., in the last 3-4 months I opened/closed my 34401A box (incl. removing the inner shield with the glued temp sensor) aprox 120 times.. 
So be warned.. 

PS: the "typical 30-40 seconds" in our discussions about sliding/rolling standard deviation mean the "rolling window size in seconds". So you have to measure the stuff much longer than 60secs. I would rather go with 20-30secs window for 34470, btw. The sliding std mean you buffer the last 30secs of the incoming samples and you calculate the std over the entire window upon each new incoming sample.
Or make the ADEV.

From my experience the replacement of a 399 with a good 1399 - the 2uVpp one in my case (my 4 samples had 1.7-2.5uVpp noise in 0.1-10Hz thus the spread is large) - the decrease in the noise is almost immediately visible at the incoming numbers - the std @10PLC went down from 1.9-2.0uV to say 1.4-1.5uV, the replacement of the OP27 with OPA140 in the integrator pushed it down to 0.8-1.2uV.
At 100PLC you have to see something like 500nV std as the mean in the sliding std (with OPA140).

Always do the measurements with 10V input (10V range).

That all measurements were against my ADR1001#1_10V which itself is rather noisy.

[mawyatt]

Would be fascinating to get to the solid state physics details of these buried Zeners compared to the older versions. Makes one wonder how the processes changed over time/company, the diffusion profiles and such. One might expect the LM329 and LM399 back from National to be the same National process, and how that changed with LT and now ADI, and how the newer designs like ADR1399 compare.
 
Don't think ADI is going to reveal much tho, for sure TI will be watching

Best

[iMo]

..Don't think ADI is going to reveal much tho, for sure TI will be watching

The product manager of the 1399 said at the Metrology meeting (2022?) the 1399 is the same 399 die/layout with small changes in the resistor values here and there (internally it has got 8x higher Iz therefore lower noise - 399 has got 250uA, the 1399 2mA, afaik).

Did @Noopy make a die shot of the 1399, btw.??
I have one 1399 here which died - when Noopy would need one..

[mawyatt]

..Don't think ADI is going to reveal much tho, for sure TI will be watching

The product manager of the 1399 said at the Metrology meeting (2022?) the 1399 is the same 399 die/layout with small changes in the resistor values here and there (internally it has got 8x higher Iz therefore lower noise - 399 has got 250uA, the 1399 2mA, afaik).

Did @Noopy make a die shot of the 1399, btw.??
I have one 1399 here which died - when Noopy would need one..

The changes in resistor values likely are to also accommodate the process changes in addition to slight circuit changes (the higher Iz for example). What would be more interesting from our perspective would be the actual process details, including diffusion profiles.

Precision stable analog IC circuits like these references are highly dependent on the IC process particulars, especially the diffusion details, no so much on the node feature size tho. Digital is much less dependent, altho modern nanometer level feature size digital becomes more dependent on process details. Getting nanometer CMOS to even work is absolutely amazing to us, and doing so with a 100 billion devices... just unbelievable

Best