The LTFLU (aka SZA263) reference zener diode circuit

[BU508A]

Here I have some pictures and the schematics of my crude prototype which I had with me at the Metrology Meeting.
And when I was going to do some measurements together with HighVoltage, it failed. Reason: for any unknown reason
the B-E diode of the 2N2219A became high-impedance. No idea why. I changed the transistor and everything
is nice and dandy since then.

Ok, as promised, here are some pictures.

Let's start with the schematics, made with a high sophisticated CAD system 



This is the Bopla plastic case opened. On the top right there is one of two
Makita 18V LiIon batteries, used for the power supply. In the socket,
there is a 15V LDO from TI.



This is the on/off switching, realized with a latching relay and some reed sensors.
Also some filtering with Wima capacitors on the left.



This is the ground plane, made of 1mm copper.



Inner case opened and overview:



Detail view of the LT1028 and the air wiring (no need of guardening the lines)



Detail view of the LTFLU-1 in it's  socket (coming from Fischer electronic, they are also doing cases)
The yellow resistors were made by forum fellow Edwin G. Pettis. Thanks again for the really fast delivery.



Detail view of the outgoing binding posts:



And here I've assembled the reference prototype with my resistor decade (top left),
the GW121 measuring the air temperature and the DMM 7510



It is continously drifting upwards. I think, the main reasons for this is:
- too short power on time, so it did not found its thermal balance
- I did not set up the tempco point of the LTFLU-1, this is the next thing on my list.
- my bizarre resistance decade box is probably not the best in respect of thermal drift



These are the settings used for the measure:



More to come, but don't know, when.
Next step is getting the 0.something ppm tempco point.

Will let you know about my (probably slow) progress.

Best regards,

Andreas

Edit: some typos

[Kleinstein]

The LT1028 is not the right OP to choose here. The signal from the reference is high impedance (a little less than R5). So a more suitable choice would be the normal OP07, or maybe ADA4077. The bias and maybe also the low frequency current noise of the LT1028 are a problem. There is no need for a super low noise OP, as the SZA263 reference should be the noise setting part, not the OP. The transistor inside the reference provides a gain of some 100 and this way effectively reduces the noise and drift of the OP.

In addition the LT1028 is tricky with unity gain.

The other odd point in the circuit is the large capacitor value for C3, especially compared to C2. With this choice a low noise OP makes some sense, though not the LT1028. C3 together with R5 sets the cross over from where the OP is responsible for the noise.

[BU508A]

So Andreas left some LTFLUs here on Metrology Meeting 2019 for x-ray analysis.
Since I couldn't resist, yes I've asked Andreas before, I dirty hacked together some components on a breadboard/veroboard.

The diagnosis is absolutely clear: you are infected with the LTFLU-flu. There is no remedey nor escape. Get along with it. 

[BU508A]

The LT1028 is not the right OP to choose here. The signal from the reference is high impedance (a little less than R5). So a more suitable choice would be the normal OP07, or maybe ADA4077. The bias and maybe also the low frequency current noise of the LT1028 are a problem. There is no need for a super low noise OP, as the SZA263 reference should be the noise setting part, not the OP. The transistor inside the reference provides a gain of some 100 and this way effectively reduces the noise and drift of the OP.

In addition the LT1028 is tricky with unity gain.

The other odd point in the circuit is the large capacitor value for C3, especially compared to C2. With this choice a low noise OP makes some sense, though not the LT1028. C3 together with R5 sets the cross over from where the OP is responsible for the noise.

Yes, I am aware, that the LT1028 is not the right choice here. The reason, why I've used it: I want to to some noise measurements. But, when I had assembled it,
it dawned to me, that using this OP here is not necessary: If I want to do some noise measurements, I should do this directly at the collector of the transistor.   

In the next version I'll use probably the LT1008 or LT1012 which are, as per datasheet from LT, the recommended replacements for the originally used LM308A.
It is a prototype and I'm playing around with it.

C3 comes directly from the original Fluke schematic for the 3330B. I have one of the original reference modules here, where I can do some measurements for comparison.

See attached schematic from the 3330B reference board.

Edit: some typos

[Kleinstein]

To measure the noise of the reference it is OK to measure the 10 V - no problem there, just some filtering from C3 included.  The position for the filtering is in deed not that bad, as the relatively high impedance is there already and some capacitance may be needed for stability anyway. For noise tests one may consider reducing C3 (down to maybe 100 nF), to see the unfiltered noise of the reference.

The LT1028 is not only not needed, it is rather noisy in the form of too much current noise. Drift of the bias current could also effect the stability. A good choice for some 35 KOhms source impedance would be something like ADA4077, OP177, OP07 and similar.  As the OP is mainly relevant for the higher frequency noise one could also use some low noise JFET type like OPA141 or similar. The LT1008 / LT1012 are more on the other side with very low current noise, but quite some voltage noise.
Below the filter frequency the transistor supports the OP - so the low frequency noise and drift of the OP are not that important. The transistor gain may be just enough so that even the noisy LT1028 may not be noticeable.

Looking directly at the transistor collector would measure the noise before the amplification of the OP and at a relatively high impedance node. I don't think this would be a good idea. If at all it might make sense to look at the base of the transistor to check for long term drift and TC from the reference only, without the scaling resistors.

[mimmus78]

What is function of D3 D4?

[Kleinstein]

D3 and D4 have some effect on the temperature effect. Other circuits can work without it or with only one diode. This may effect the second order TC.

[BU508A]

D3 and D4 have some effect on the temperature effect. Other circuits can work without it or with only one diode. This may effect the second order TC.

That's right. This is from the Fluke 731B and as you can see, it has only a 1.27kOhm resistor and no diodes.

[BU508A]

The transistor gain may be just enough so that even the noisy LT1028 may not be noticeable.

What I do not understand is: why are you saying, the LT1028 is noisy?
Looking at the datasheet, it's specs are very good in respect to noise. As far as I know, there aren't much other OpAmps around which can beat the LT1028 in that discipline, imho.

*scratching head*

[The Soulman]

The transistor gain may be just enough so that even the noisy LT1028 may not be noticeable.

What I do not understand is: why are you saying, the LT1028 is noisy?
Looking at the datasheet, it's specs are very good in respect to noise. As far as I know, there aren't much other OpAmps around which can beat the LT1028 in that discipline, imho.

*scratching head*

You have to remember: datasheets are written by marketeers not engineers.
The zener must be treated as a high impedance source, so current noise specs come into play.

[chuckb]

What I do not understand is: why are you saying, the LT1028 is noisy?
Looking at the datasheet, it's specs are very good in respect to noise. As far as I know, there aren't much other OpAmps around which can beat the LT1028 in that discipline, imho.

*scratching head*

A very good zener reference chip like this will have a noise density on the order of 50-100nV / rt Hz. The R5 collector resistor (39.5k) will add 28nV / rt Hz to the Zener noise. These random noise sources add together via RSS. So using an opamp with a 1nV / rt Hz noise density will not help reduce any noise. All of the opamps with low VOLTAGE noise will have a very high CURRENT noise. In this case it's 10pa / rt Hz. On three LT1028A opamps that I tested the current noise was 15-20pa / rt Hz at 10Hz. When the 10pa / rt Hz interacts with the R5 impedance of 39.5K it will generate an additional 400nV / rt Hz of noise. The current noise of the LT1028A will generate 4-8 times the voltage noise of the zener.

I believe the Fluke 732A voltage standards use a LM308 in this application. The LM308 will add a total of about 50nV / rt Hz to the output noise. The newer LT1008 will add about 20nV / rt Hz to the Zener noise, so it will not be noticeable.

I don't know if the voltage gain of the LTFLU internal transistor will suppress the LT1028A opamp current noise.

Every opamp has an optimum source impedance for lowest added noise. See the attached Design note from Linear Tech for the math.

[dietert1]

Isn't C3 meant to be a lowpass filter taking away the LT1028 input current noise from R5 and the zener noise away from the amplifier input? With 39K and 4u7 it works above 0,8 Hz.
If i had a 1nV/sqrt(Hz) amplifier around i would just do some measurements with different values of C3. Maybe it needs to increase a little.
I would also do some tests with an extra miller capacitor for the LTFLU transistor in order to keep it quiet.

Regards, Dieter

[BU508A]

@soulman and @chuckb

Thank you for your explanations. Now some things makes more sense to me and I hopefully have my lesson learned.   

Also a big Thank you to branadic who did some black x-ray magic and scanned the LTFLU-1 devices and one SZA263.

Here are the pictures he sent me:



LTLFU-1 ACH:



Zoomfactor 5:



next piece LTFLU-1 ACH



and another one



and the last one of the LTFLU-1-ACH



This ones are LTFLU-1 CH (without the "A", which is a newer version)



Zoomfactor 5 of the LTFLU-1 CH



And here, for comparision, some pictures of the SZA263

This one is coming from this message https://www.eevblog.com/forum/metrology/the-ltflu-(aka-sza263)-reference-zener-diode-circuit/msg911832/#msg911832



X-Ray picture made by branadic:




I did not expect, that the SZA and the LTFLU are that different.

Thanks again for doing the pictures. 

Regards,

Andreas

[BU508A]

D3 and D4 have some effect on the temperature effect. Other circuits can work without it or with only one diode. This may effect the second order TC.

That's right. This is from the Fluke 731B and as you can see, it has only a 1.27kOhm resistor and no diodes.

Here is an additional reference circuit, this time from the Fluke 332D (with an oven):

The A2 is a LM301A

[Kleinstein]

I did a quick simulation of the circuit in LTspice:

The model for the Zener is essentially noise less, so this important noise source is missing. To get some noise for the OP I added a resistor (r11) in series to the input.

One can still see that the cross over is not at the frequency one would expect from just R5*C3 (0.8 Hz) but at a much higher frequency of some 80 Hz.  This is due to the extra gain for the Zener signal.
So it is possible to get some filtering from C3, but it is not as effective as one might expect from the relatively high impedance node.

C3 is also not really effective in filtering the current noise from the OP. This noise has quite some 1/f part that is not effectively filtered.

p.S: the slightly more effective position for filtering would be a miller cap, from collector to base of the transistor. Here the impedance of the divider is added.

[dietert1]

Isn't 80 Hz the cutoff frequency of R2 and C2? Obviously R2C2 does not match R5 C3, and it does not need to since impedance and noise are lower anyway. BTW in the proposed circuit opamp inputs are not at +/- 7 V but at 8.7 V or so.
What happens if you make the zener the dominant noise source? Can you give the zener the 7 Ohms differential resistance (maybe a series resistor) that was determined before and feed noise with a 1K plus capacitor?

Regards, Dieter

[Kleinstein]

Having the cross over at about where R2C2 is, is more coincidence. The cross over frequency is set be R5*C3 divided by the voltage gain of the transistor stage that is at around 100, maybe a little less if the voltage at the OP inputs is higher and thus less voltage at R5.

I have checked how noise (or a signal) from the zener propagates: it looks like a simple 1st order low pass with 63 Hz cross over. The frequency does not change with C2, but it does change with R2 and thus the DC voltage over R5. A smaller R2 reduces the gain of the transistor and thus cross over frequency.

The impedance of R2,R3,C2 has an odd effect: with a fixed C2 the noise goes down (not relevant in the sum) when R2 and R3 get higher values (at least without the current noise from the OP).  C2 seem to be chosen just large enough that in the relevant frequency range R2 and R3 don't contribute much to the noise. With a BJT based OP it would make sense to go here up to the 40 K range so one has a matched impedance on both sides of the OP.

P.S. the zener model has a differential resistance of some 19 Ohms.

I also checked the effect of a current signal injected at the OPs input: It has an effect like seeing a resistance of some 700 Ohms at the lowe frequencies, which would be the 40 K divided by the transistor gain. So some 15 pA/Sqrt(Hz) from the LT1028 would result in some 10 nV/sqrt(Hz).  So this is likely still less than the zener noise at 10 Hz.

[dietert1]

Sometimes the dynamic behaviour of those regulators is surprising.

BTW in the meantime i put the recommended ADA4077 into the reference circuit of one of our HP 3456A and made some changes to get the extra bandwidth. I am using it with a snubber on the -12 V reference output that consists of a 4,7uF MKS parallel with 1 Ohm plus 100 uF. Then the OpAmp sees a 1 Ohm resistive load up to a time scale of about 10 or 20 usec and stability is almost guaranteed. Maximum output current of the Opamp is about +/- 10 mA and enough to control output voltage across the 1 Ohm resistor. The snubber buffers fast current spikes from the ADC and residual noise remains well below 100 or 200 uV (looking at it with a scope). Since the ADA4077 shows an offset voltage of 50 uV i will also try a ADA4522.

I am not sure whether a 100 uF electrolytic cap is ideal for C3 in the proposed circuit.

Regards, Dieter

[Kleinstein]

For the filtering cap it really depends if filtering of the reference is needed 7 wanted.
Instead of C3 one could use a miller cap - this give a cross over frequency at 3 times lower, as there is the additional impedance from the divider at the base. Still filtering would be at relatively high frequencies only.

The main reason I see for filtering would be if some AZ measurements are used (either from an external meter or if the reference us used with an ADC). The filter would kind of bridge the pause caused by the zero reading. So it would need at least some 20 ms (maybe longer, depending on the meter). Currently C3*R5 /70 is at about 3 ms. So the 4.7 µF cap does not really help much. There is probably not much room for changing R5, as this is more like one point to adjust the TC. Even with 5 µF as a miller cap one would only be at some 10 ms (e.g. C * (R5/70 + R1||R7)).

In principle an electrolytic cap could work. Due to the transistor gain the effective resistance is at some 700 Ohms, both for the cross over frequency and the leakage / bias related error.  With some 100 µF would would be at some 60 ms and thus just enough to cover 1 to maybe 3 PLC auto zero.  Besides leakage, there can be also some DA (hard to tell apart from the experimental side) with an electrolytic cap. So it may take quite some time (e.g. hours) to stabilize after turn on.

[TiN]

All this talk couldn't let me sleep... 

UPDATED 6/20 version:

[branadic]

A few words to the x-ray images shown above.
With some imagination you can see the rectangular die as a slight shadow on the squared metallic piece of the package. You can also see the die attach, but you can't see any bond wires as aluminium bond wires are transparent to x-ray (other then bond wires made of gold). Comparing this images with the pictures of the already cracked open parts in previous posts indicates, that there is at least the same thing inside. Finally only electrical characterisation can show, if they behave like a LTFLU even though the package marker is at the wrong place.

-branadic-

[Kleinstein]

@TiN: the circuit shown looks really odd in some points.
Like with the LTZ1000, there is little need for the OP used with the LTFLU/SZA263  voltage loop to be an AZ type.
The extra resistor between the divider and base of the reference does not really make much sense with such a small cap at the transistor.

For the switched capacitor circuite C221 directly in series with the switch does not make so much sense. It looks a little like a capacitive inverter, but not really.

[iMo]

@TiN: the max LTC1050 voltage is 16V.. (abs max 18V).

[TiN]

Yup, you guys right. Haste is never good in circuit design.

Pin 3 from DG444 missing AGND connection, it is the inverter there. Power is valid note too. As of opamp choice, have to admit, I'm copycating Fluke ref circuit, where LTC1150 is also chosen. There is also little PCB oven around the ref, aka K7510 design. I have only two salvaged SZA263 chips to try this , not even LTFLU. Ordering bigger board, so this will be just a test vehicle on free area, not a standalone proper ref.

[TiN]

Reuploaded updated schematics version.
* Fixed inverter missing ground connection
* Fixed U128 power to meet rated spec
* Cleanup