Puzzling LTZ1000 circuit behavior

[Galaxyrise]

Maybe this isn't a beginner problem, but I still feel like a beginner, so I'm putting this here.  I'll link to it  from the main ltz1000 thread.

I have a test circuit with a bunch of test points and trim pots on it so I can better understand the parameters of the LTZ and its reference circuit.  It's undergone a few bodge changes, but I think the attached schematic represents its current state of affairs.  When doing the experiment below, I was logging both ltz collector currents, I(R14), V(Pin4 - Pin7), V(Pin6-Pin7). 

One of my experiments was to adjust R4 so that pin6 was high enough that the heater would never turn on (it comes on around 603mV, so I had it set at 613mV) and then raise pin 6 voltage some more (to 626mV)  I was expecting a very slight increase in the corresponding collector current and that was pretty much it.  But what I got was a shift in pin 4 of about 500uV and a corresponding change in current (which is what I noticed first.)  I considered a change in self heating, but since Q2 current is going up by so little with a small voltage drop, I would have expected the shift to be orders of magnitude less. As it stands, I think total power is going down, not up!  I would have expected for of a "time constant" shape to the curve, too.

I built up a mock LTZ1000 which is discrete 2n3904's, a zener, and two heater resistors.  I replaced the real LTZ with the mock LTZ to see if it was something about the LTZ or about my circuit.  It behaved differently, but still in a way I don't understand.  Instead of the near-instant, slope-matching response of the real LTZ, the mock LTZ appeared to very slowly warm up a little whenever I changed pin 6, regardless of which direction I changed it.  This made me think the temp loop must oscillate on a change.  So I put my scope from pin 1 to pin 2, but didn't observe any oscillation >= 1mV during a change in pin 6.  When using the mock LTZ, there's some dozens of centimeters of extra wiring (cat5) between the test points and the components, but I don't expect this kind of result from that.

I've been at this a couple days, and I'm out of theories and things to try.  Any ideas?

Changes from the datasheet schematic are: (the intentional ones, anyway, hehe.)

• Addition of C11, C13, C14 to the temperature loop from Andreas' earlier schematics.  (he has subsequently found C14 to be a problem with higher speed opamps)
• LT1112 instead of LT1013
• Addition of R13 across the heater transistor to ensure some heater current even when the transistor is off, keeping pin 1, pin 2, and pin 4 in their proper potential order
• Addition of a second diode between pin 2 and pin 7.  Because I'm running so cold, vbe pin 4 is very close to a 4148 Vf and I wanted more headroom.
• Diode between pin 7 and LT1112 V- to account for its voltage ranges and to allow me to buffer pin 7 (with a 2057 or something) later
• Addition of R14 before the LTZ inputs to limit current in the event I messed something up.  9mA through R1 would raise pin4 over pin2 even with the extra diode.
• R10 and R11 are left over from when there was a diode from that voltage divider to the ltz inputs to ensure bootstrap.  But that diode is gone so I don't expect R10 and R11 to do anything but warm up the board a little.

[Andreas]

Hello,

only guessing (without deeper analysis):

perhaps the headroom of the power supply is too short.
7V + 2 Diodes + up to 4V for the LT1112 under high current (2kOhms) condition.

Did you try to remove the 400K resistor already?

With best regards

Andreas

Edit: forgot R14 which will consume another 2-2.5V.

[Galaxyrise]

While the schematic says 12V, it's being supplied with 15V.  Forgot to adjust that part of the schematic.  15V is still really close to the list of things you were adding up, though...  I'll try again with higher V+.  A key difference between the mock LTZ and the real LTZ is the mock ltz has a lower zener voltage.

I haven't tried removing the 400k yet because I wasn't observing anything happening at pin1.  But I'll give it a go after trying the above and report back.

Thanks

[Kleinstein]

The comon mode voltage with both OPs is about U_BE of the transistors, so something around 0.5 V.  Only during startup lower values apply. The OPs are not very critical for precission, as the transistors already give quite some amplification - so not real need to look for something better than the LT1013. However this makes stability critical in the circuits.

The temperature regulator of the mock-up vesion may very well need adjustments, to make is stable, as the thermal system is much slower. The critical time constant is R7*C1. So much larger values for C1 might be needed to get a similar response. As it is difficult to make this without too much leakage, I would not make a direkt copy for the thermal system. Also the thermal regulation circuit of the LTZ1000 is not really good: the good part is the low effort low noise temperature sensor - but not the regulation part itself, this a simple PI regulation, not even linearearized to compensate for the heater having a square law.

The temperature regulation and the voltage stabilization are essentially independent, esspecially for the mock-up version.
The resistor R9 coupling the two parts is a rather crude approximate corretion for resitdual TK - I would try is without first, especially in the Mockup version or a second diode added at the low side of the heater.



 

[Galaxyrise]

The amplifiers used with an LTZ must have a common-mode range that includes ground.  IIRC, the LT1112 does not meet this criteria.  Try changing it with an LT1013 to see if your circuit starts working...

I have V- of the LT1112 offset from pin7 by a diode drop, addressing that issue.   Maybe it's still too close to the bottom of the range for precision operation? I could add another diode.  I do have a 1013 on hand, but unfortunately, changing the opamp would be a real chore because of how the current one is bodged in. 

It's not that the circuit isn't working; it holds a steady state within my ability to measure it.  But it's exhibiting a dynamic behavior that I don't understand.

[Galaxyrise]

perhaps the headroom of the power supply is too short.

Did you try to remove the 400K resistor already?

I tried 18V which did not help.

I removed the 400k, this also did not help.

I gave the LTZ more time between adjustments, and observed the same behavior that I did with the mock LTZ in addition to the spike, so the real LTZ has two strange behaviors.

And then I figured out one of them.  There really is a temperature shift when I adjust the trim pot: Me.  I wasn't expecting my temporary presence to affect the temperature that strongly, but I was able to reproduce the gradual shift and recovery just by sitting down next to it long enough to adjust the trim pot and then leaving again (but without actually touching it.)  So there's one noobish problem understood.

But the real LTZ still exhibits the immediate and inverse shift of pin4 when I adjust pin6.  Tomorrow, I will try changing the test so I can adjust pin6 remotely.

[Alex Nikitin]

Just as a suggestion - in the "real" LTZ both emitters have the same connecting pin 4 and possibly some interconnects shared on the chip. Perhaps these have enough resistance to explain the  shift you observe, as the collector current change of the temperature sensing transistor would produce a voltage drop applied to the emitter of the voltage sensing transistor? Imagine there is a small resistor in series with pin 4 on your schematics. You can add that resistor to your "dummy" LTZ circuit and check its behaviour.

Cheers

Alex

[Kleinstein]

In the temperature regulation part the additional capacitors C13 and C14 likely cause more trouble than help. C14 directly between the inputs of the OP amplifies noise of the OP and can cause instabilitly. C13 adds delay to the regulation loop just in the middle of the proprotional range and might upset the temperature loop with the real LTZ1000, especially at higher heater currents. I don't think C13 is a good idea, at more than about 500 pF.

Also the divider to set the temperature needs to come from the stable voltage, not directly from the OPs output.

The voltage might get a problem with 550 Ohm (like in the plan) in series - it makes the output even more sensitive to capacitive loading. A series resistor is good only if there is additional capacative feedback from before it.

Due to the extra gain from the transistor the voltage loop is somewhat prone to oscillation. So you have to check it with the scope. The original circuit (with capacitive loading) and also the modifications Andreas showed in the other thread is rather close to oscillation. This is the reason why it is critical to use AZ OPs there, as they tend to have extra phase shift in the low frequency range (e.g 1-10 kHz).

[d-smes]

You show C3 as 2nF.  I believe there is a mistake in the 7V Positive Reference Circuit in the LTZ1000 data sheet as the other example circuits show this to be 22nF.  App notes AN-86 (Figure 12) and AN-42 (Figure 67) also show a recommended value of 22nF.   I doubt this explains your problem, but since C3 provides compensation to the regulation loop, 22nF might be worth a try.

[branadic]

You show C3 as 2nF.  I believe there is a mistake in the 7V Positive Reference Circuit in the LTZ1000 data sheet as the other example circuits show this to be 22nF.  App notes AN-86 (Figure 12) and AN-42 (Figure 67) also show a recommended value of 22nF.   I doubt this explains your problem, but since C3 provides compensation to the regulation loop, 22nF might be worth a try.

Yes, it's a mistake in the latest datasheets, with the nice and fancy LT logo. If you compare older datasheets, such as this you will find, that 22n (0,022µ for our imperial guys) is the correct value.

[Kleinstein]

2 nF is really on the low side. It may still work with a kind OP. In the simultion it costs about 20 degree of phase margin at 1 MHz - more at lower frequency. So it could be a problem, especially with slow OP like the LT1112 and possibly extra phase shift vom capacitive loading.

[Galaxyrise]

The temperature regulator of the mock-up vesion may very well need adjustments, to make is stable, as the thermal system is much slower. (...)
The resistor R9 coupling the two parts is a rather crude approximate corretion for resitdual TK - I would try is without first, especially in the Mockup version or a second diode added at the low side of the heater.

The mock-up was initially made just to verify I had the circuit basics correct and the ltz pin voltages would not exceed their datasheet specifications.  I wasn't expecting it to achieve temperature stability, I just wanted to make sure I didn't have the opamp inputs reversed and that sort of thing; so a slow oscillation was success.  For this test, since pin6 is set unachievably high, so the heater is always "off" and the poor time constant tuning shouldn't matter.  And indeed, now that I am controlling pin6 remotely, the mock LTZ pin4 is unaffected by changes to pin6.

(The circuit does have R9 removed now.)  What is "residual TK"?

Just as a suggestion - in the "real" LTZ both emitters have the same connecting pin 4 and possibly some interconnects shared on the chip. Perhaps these have enough resistance to explain the  shift you observe, as the collector current change of the temperature sensing transistor would produce a voltage drop applied to the emitter of the voltage sensing transistor?

I don't think a resistance between the emitters and pin7 test point is the answer.  Check my logic:  First, I believe a voltage there would cause pin 4 to go up when Q2 current goes up.  I'm observing pin4 go down when pin 6 goes up and vice versa.  (double-checked in spice) Second, Q2 current (both base and collector currents summed) should be going up by some microamps I think.  For it to shift pin 4 by several hundred microvolts would take a fair amount of resistance, more than I expect from the interconnects. (Though if the interconnects do have 10s of ohms resistance, it would explain why the LTZ is worse than the foot of cat5 wire connecting my mock unit.)

You show C3 as 2nF.  I believe there is a mistake in the 7V Positive Reference Circuit in the LTZ1000 data sheet as the other example circuits show this to be 22nF.  App notes AN-86 (Figure 12) and AN-42 (Figure 67) also show a recommended value of 22nF.   I doubt this explains your problem, but since C3 provides compensation to the regulation loop, 22nF might be worth a try.

Oh wow, good spot! Thank you for reviewing the schematic that closely. 2nF did seem awfully small but I didn't remember anyone complaining about it before. I will see what I can put in parallel with C3 and give it another try.  Will definitely include that fix in future revs.

Attached is the result of my latest run with the real LTZ, controlling pin6 remotely so I don't introduce extra temperature fluctations with my body.  (Programmable power supply connected to Pin6 via 10k resistor and to Pin7.) I then subtracted out most of the ambient temperature drift to make the effect in question easier to see.  Blue is the left axis and red is the right axis.  The curve after the jump in the red line makes sense as self heating changes due to zener current changes, but the jump remains unexplained.

Thanks for all the inputs, I really appreciate it!

[Galaxyrise]

Same behavior with C3 changed to 22nF.

[Galaxyrise]

We've talked about the increase in current from an increase in pin6, but I realized raising pin6 has another effect as well: it lowers the voltage at pin8.  If I turned the test around so I was adjusting the voltage at pin 8 more directly, it would also help eliminate the current theory because now current would go up when pin 8 goes up.

So I rigged a FET+resistor in parallel with R3 and then used that to adjust R3 remotely.  Pin4 changes by about 1/26 the voltage change at pin8.  I wasn't measuring pin8 before, but this is about the factor I would expect from the 0.5mV shift I was getting before, and has the correct sign.

Thus I think pin8 is the influence in my setup. Running the device as I am, with pin8 around 50-60mV, is certainly not how it's meant to operate in steady state.

[Andreas]

Hello,

pin 8 and pin 6 should always be the same voltage (in steady state)
otherwise the heater regulation loop does not work correctly.

Are you shure that collector + emitter of Q1 are not interchanged?
Is there any significant (more than 0.1-0.2V) voltage drop across R8?

With best regards

Andreas

[Kleinstein]

With pin 8 at less than about 100 mV, the transistor is in saturation and the heater should be all the way on. This might be due to a temperature setpoint that is way to high. This also means the volatge at pin 6 depends on the temperature and is not mainly set by the resistors.   This could be because R4/VR4 are still from the wrong (to high) voltage.

[Andreas]

Hello,

just some measurements from LTZ#3 (with LT1013A)

I put a 22K resistor in series to J6 Pin 11 to my PWM generator (on the open Drain output).
So total resistance is 24.7K to ground in parallel to the 1K resistor giving around 9 deg C setpoint jumps.

the schematic for LTZ#3 is here:
https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=189655

CHA (blue)   PWM open drain
CHB (red)    temperature setpoint (R5)
CHC (green)  current (R1)
CHD (yellow) J6 pin 12 (BC639 base)

behaviour at base around the same with 80mV difference
This is explained by the same temperature of the base emitter junction.
Dynamic differences are due to actual temperature following setpoint.

heating around 37ms
cooling dead time around 280 ms
heating after dead time around  41 ms

With best regards

Andreas

Edit: voltage across R8: 109mV (so within the expected 0.1-0.2V).

[Kleinstein]

Interesting curves to see how fast the temperature control works. The response looks quite good for such a simple controller. The curve shows a little of overshoot for the temperature.

One can also see that control is faster, when the heating power is hight - this is likely from the nonlinear heater curve.

Still there is quite some noise visible in the green curve. Is this a problem of the zenerdiode, the temperature controll or just a problem in the measurement / scope ?

[Andreas]

Hello,

cannot really tell what it is.

The measurement is done in +/-1V range with 12 Bit physical ADC and mathematical resolution enhancement to 16 Bits.
The "noise" is about 500uV so 1 digit of the physical resolution.
With the resolution enhancement there seems to be a around 1.6 kHz hum.
But since the measurement is not done within the cookies box it can be anything.
(e.g. interference from the Laptop).

With best regards

Andreas

[Galaxyrise]

pin 8 and pin 6 should always be the same voltage (in steady state)

With pin 8 at less than about 100 mV, the transistor is in saturation and the heater should be all the way on. (...) This could be because R4/VR4 are still from the wrong (to high) voltage.

Sorry if I wasn't clear enough in the opening post, but I have intentionally set VR4 such that Pin6 is too high, leaving Q2 in saturation, the opamp output low, and the heater permanently "off"  (not on; the heater is on when pin8 > pin6.) 

This was so I can put the LTZ through external temperature changes of a known magnitude and measure the response of pin4 with the goal of turning pin4 voltage into an absolute temperature. Then I could adjust VR4 so that the temperature control would put pin4 where I wanted it, and then compute the resistor ratio to get the die temp where I wanted it with much more precision than the +-10C in the datasheet.

The bigger picture is I want it to set it at 30C, and put it in a box chilled to 10C, so that the temperature cycle it experiences is small (+-10C from a 20C ambient.)  Since 30C is pretty far away from the nominal setpoint, I wanted to measure actual values for things.

just some measurements from LTZ#3 (with LT1013A)

Awesome, thank you for doing that!  That first image looks like what I'd expect, with the positive tempco of the reference causing output voltage to chase the temperature setpoint over a span of some seconds (no instant jump.)  I haven't looked to see if I'm getting the same instantaneous jump at pin3 that I am at pin4. 

But I don't understand what I see in the second one.  In the zoom overview, the setpoint goes up, output drops, is flat for awhile, then slowly rises up.  What's causing that delay and rise at the end?

[Andreas]

Hello,

But I don't understand what I see in the second one.  In the zoom overview, the setpoint goes up, output drops, is flat for awhile, then slowly rises up.  What's causing that delay and rise at the end?

the delay (dead time) after increasing the setpoint = reducing the temperature value
is needed to cool down the LTZ.
At end of the cool down phase the heater power increases slowly until stable working point is reached.

Sorry if I wasn't clear enough in the opening post, but I have intentionally set VR4 such that Pin6 is too high, leaving Q2 in saturation, the opamp output low, and the heater permanently "off"  (not on; the heater is on when pin8 > pin6.) 

You cannot expect any useful measurements when the regulation loop is not working.
Especially as there are some OPs which have input protection diodes between the + and - input.
(The LT1013 has not, the LTC2057 has around 4-6V zeners, others have only 1-2 diode drops as input limiting).
The LT1112 has only 1 diode drop limiter at the input. (See datasheet page 12 Q9+Q10)

with best regards

Andreas

[Kleinstein]

@ Andreas: The dead time in the context of control theory is rather short - looks like somthing in the 10 ms or slightly less range. Hard to see from the curves as the heater is ramping up / down at a similar rate.
The time you get where the regulator is out of control because the heater is either fully on or off is a different thing. This is mainly the limited power of the heater.

The 1.6 kHz "noise" is the more obvious part - its , much stronger in the high phase of teh PWM signal - so thats something coupled in from the PWM part. I was more thinking about the smaller lower frequency part. But much of this looks like 60 Hz (strange to have this and not 50 Hz). So not a big deal.

Though the heater goes all the way into saturation with what looks like a good way back, though there is some overshoot - which is just not directly visable because the PWM signal is too fast to see it. The steady state heating level for the higher temperature still has to go higher than the other one - so what looks like a final value can't be right, it will overshoot if more time is given. Still having a little overshoot is usally a good regulator setting. So it seem to be resonably tuned to the thermal system. Though it might be still interesting to see how the more original control circuit (no c13/C14) would perform - I personly would definitely leave out C14 and likely make C13 smaler by a factor of 10 or so.

@Galaxyrise:
Using the voltage at pin4 to measure temperature makes sense. At least when in regulation the current is to the transistor is constant  (set by R2). So it should behave like a normal diode: linear towards 1.12 V at 0 K - so about the usual 2 mV/K. The small change in the voltage at pin 4 thus corosponds to a tiny bit of temperature increase. The reason fot the voltage at pin 6 to go up could be just the pot (VR4) drifting. This voltage is more set by the divider and not so much responding to temperature changes.

Using a 10 C enviroment can be rather tricky unless you have a very dry enviroment, as relative humidity goes up leakage current can go up. Its aktually a good idea to have the critical circuit part at at least 10 K above the normal enviroment - never below unless humidity is taken care of. So its more like 30 C for the circuit and 45-50 C for the reference itself.

[Galaxyrise]

the delay (dead time) after increasing the setpoint = reducing the temperature value
is needed to cool down the LTZ.

Sorry, I managed to misinterpret what yellow meant. I understand now.  I also missed that the graph was in milliseconds!  I was definitely not making measurements quickly enough to see a slope like that. I'm not sure I even can, hehe.

You cannot expect any useful measurements when the regulation loop is not working.
Especially as there are some OPs which have input protection diodes between the + and - input.
(The LT1013 has not, the LTC2057 has around 4-6V zeners, others have only 1-2 diode drops as input limiting).
The LT1112 has only 1 diode drop limiter at the input. (See datasheet page 12 Q9+Q10)

I was not expecting that to influence the current loop, but maybe it is doing something to self heating I wasn't considering.  It does seem like I should do something different to the temperature loop if I want to continue this approach.

The small change in the voltage at pin 4 thus corosponds to a tiny bit of temperature increase.

0.6mV on pin4 should be about 0.27K; would need to be an increase of around a milliwatt, I think?  That seemed like too much when I first considered that possibility.  In fact, it seemed like total LTZ power should have gone down from the values I was measuring, but I could have made some invalid assumptions there (like the current through R3 is the same as the current into Pin8, which Andreas points out may be incorrect with the temperature loop broken.)