Ultra Precision Reference LTZ1000

[janaf]

Thanks TiN but I solved it.

Also, Digikey have a small stock of LZT1000 now, also in single quantities. I haven't seen them on Digikey for years.
I order other parts from them quite frequently.

[janaf]

Hello,

Todays measurement: addition of a resistor in series with the zener cathode. I have called it "R60" to avoid confusion with the R6 used in the LTZ1000 datasheet.

The measurements show, as discussed earlier, that adding a resistor to the zener cathode, decreases temperature sensitivity ie also the sensitivity of the R4/R5 ratio. At the optimal point, "R60", somewhere near 20R, the sensitivity to R4/R5 (temperature) seems totally eliminated.

The bad news is that the sensitivity of the R60 (and R60/R1) will be almost identical to what it was for the original R4/R5 ratio was. So in all, maybe nothing is gained. I can not say if one is better than the other.

Also note that if no other component values are changed, then the output voltage of the circuit increases with "R60". At it's optimal value, 20 ohm, the output voltage from the circuit is up by (EDIT: wrong: one volt, up from 7.1V to around 8.1V) 0.1V from 7.1 to 7.2V (/EDIT)

This also means that all voltages & currents of the circuit will change somewhat, for example the temperature set-point will decrease.

Food for thoughts....

[Andreas]

At it's optimal value, 20 ohm, the output voltage from the circuit is up by one volt, up from 7.1V to around 8.1V.

Hello Jan,

Did you really measure a 1V increase?
On the first view I would expect only 20 Ohms * 5mA = 100mV increase of the voltage.
What happens there?
Is the cirquit oscillating by adding the resistor?
Is the temperature control loop out of regulation?

With best regards

Andreas

[MK]

Looking at the pdf, I see an increase of79,000 microvolts, or about 80mV increase in output level, so that is about right.
It does mean the temperature sensitivity is reduced, but the R60 and R1 need to track, but R1 is needed to be very stble anyway, that does add up to one less ultra stable resistor I think, as anything between 16 and 25 ohms means r4/r5 is more than 5 times less critical, and at 20 ohms shows about 50 times less critical.
I see that as a slight gain in the overall behaviour.

[janaf]

Yes, MK you are right, it's around 0.1V increase (0.079V). I got that an order of magnitude wrong in my head.
You are also right about that it reduces the number of sensitive resistors from five to four. It could also mean further simplifying, omitting R4/R5, an op-amp, diode, feedback etc. But without further testing, I'd keep them. Or ovenize the whole board. The 20R value might need some trimming....

[janaf]

At it's optimal value, 20 ohm, the output voltage from the circuit is up by one volt, up from 7.1V to around 8.1V.

Did you really measure a 1V increase?
On the first view I would expect only 20 Ohms * 5mA = 100mV increase of the voltage.

My misstake. You can see the right values in column C of the PDF, in uV.

The measurements where very stable until "R60" was somewhere above 25 ohm, then output started drifting. I did not investigate why, but it happened repeatedly.

Regards

[macfly]

....
 At it's optimal value, 20 ohm, the output voltage from the circuit is up by one volt, up from 7.1V to around 8.1V. This also means that all voltages & currents of the circuit will change significantly, for example the temperature set-point will decrease. ....

Hi Jan,

I think, there is something wrong. Are you sure about the value of the resistor ?
Perhaps you noticed it before: I have made the same measurement with a 20 Ohms resistor and
as expected, the voltage raised only ~ 100mV.

Regards,

macfly

[janaf]

Yes, you are right, as mentioned a couple of posts up, it's around 0.1V increase (0.079V). I got that an order of magnitude wrong in my head.

The actual values are in column C of the pdf file (add the nominal 7.1xxx, in blue for absolute values).

[Galaxyrise]

The measurements where very stable until "R60" was somewhere above 25 ohm, then output started drifting. I did not investigate why, but it happened repeatedly.

As a guess, raising the output voltage without changing the R4:R5 ratio lowers the temperature setpoint; perhaps you lowered it beyond stable regulation.

[janaf]

Galaxyrise, I don't think so, I was looking at the current to and it was still pretty warm, total current was above 30mA and is usually stable at least down to 25mA in this setup. But I may have overlooked something. In any case, there should be no reason to go above 25R for this resistor (with R1 = 120R)

Any questions, I'd be glad to help.

For measurements and calculation questions, I think it's easiest done column-wise as in the pdf, with the column headers, ie A,B,C.....

[janaf]

Attached is an update for the latest pdf. 

I have added sensitivity of R1 over varying values of "R60". It turns out, as expected, that R1 and R60 cancel each others changes to output, should track each other for best performance.

It also shows that the sensitivity of R1 increases with R60, from the nominal -0.0016 (dV/dR) to around -0.01 at R60=20R, ie the same sensitivity as for R60, as expected.

Sometimes it's good to get confirmations on things that you think you know.

So, bottom line, adding an R60 of 20R, more or less eliminates temperature sensitivity of the circuit and thereby also the demands on R4/R5. But instead you get the R1/R60 pair with high demands on stability and tracking of resistance ratio.

• You get rid of two; R4/R5
• Add one, R60
• Demands on R1 go up from moderate to high

One note is that R4/R5 have one common node, pad 6 of the LTZ1000, while R1/R60 connect to two different adjoining pads, 3 and 4.

[janaf]

Teaser 1 for my next post. 

You may remember from an earlier post that for R2, over a range around the nominal 70K, that the output voltage sensitivity to R4/R5 decreased somewhat with increasing R2. At the same time the sensitivity of output to R2 itself remained constant. I measured this for R2 of 40K to 130K.

I attach that plot again.

Now suppose we increase R2 more, what happens?

Teaser 2: It gets interesting!

[MK]

I think the curve continues it existing tragectory, and that r4/r5 ratio becomes a little less important, look at the temp dependance of VBE in Bob Pease's "what's all this VBE stuff", it goes from 2mv to 2.3 mV sensitivity as Ic reduces, eventually the noise becomes significant.

[janaf]

Yes, it does get interesting!

I kept increasing R2 further and further, and at values in the order of 5.6Mohm, yes, the sensitivity of output to R4/R5 (temperature), passes zero!

At the same time the sensitivity of output to R2 remains constantly near 0.004 (attenuation 1:250)

So by increasing R2 to about 5M6, temperature sensitivity is essentially gone! (like with R60=20R) while the relative sensity of R2 remains the same, no penalty as for R60.

I measured this on 4 different LTZ1000ACH and they all landed in the same region. The spread is such that even if the R2 value is not "spot on", the sensitivity of R4/R5 can be very low without trimming the R2 value. The 5M6 specs do not need to be very high in terms of accuracy, only in stability.

So essentially: Increase R2 to 5M6, the stability / TCR specs remain unchanged, while the two most sensitive resistors, R4/R5 can be replaced with cheapo equivalents.

Then, the bad news: a 5M6 resistor with high stability is hard to get, expensive, whatever. I asked Edwin and yes, he can make them, but they will be big and several times more expensive than the 70K equivalent. The Caddoc USF may be an alternative. Available at 5M. Foils? Don't know if they are made at that high resistances.

Attached; Page 1-2 measured data, page 3 overview wide range plot, page 4 plot near zero crossing for four different LTZ1000ACH

[janaf]

A PS to the previous post: Just checked the sensitivity to R1 and with high R2, R1 actually gets slightly less sensitive, going from 1:700 at R2=70K, goes down to 1:900 for R1 with R2 at 5M6. Not much but at least the right way, no degrading....

[janaf]

I think the curve continues it existing tragectory, and that r4/r5 ratio becomes a little less important, look at the temp dependance of VBE in Bob Pease's "what's all this VBE stuff", it goes from 2mv to 2.3 mV sensitivity as Ic reduces, eventually the noise becomes significant.

Noise does not seem to go up visibly, but I was only measuring with the DMM at 1s aperture. Could it be that while the collector current will go towards zero the base current remains unchanged? Suppose the collector resistor goes to infinity, shut off, what happens to noise then?

[janaf]

Thanks for the input on noise!
 
The current over the resistor is around 1uA (6V/6M). If the transistor optimistically has 1nV/sqrtHz at 1mA, it would have 33nV/sqrtHz at 1uA, or about 100nV to 10Hz. I think the zener (or circuit) has a 1.2uV noise 0.1-10Hz. So even with 6Mohm, and even if the 1nV/sqrtHz was over-optimistic, transistor noise would seem acceptable. But the the 6M resistor noise should also be around 1uV at room temp. 

I think I need to get the noise amp and batteries connected.

[MK]

According to the datasheet the zener has 50nV root Hz at 1Hz, so a 5.6 M resistor for r2 would about double the noise.

I cant see that being viable, but it does show a larger value may be helpful

[Andreas]

Thanks for the input on noise!
 
The current over the resistor is around 1uA (6V/6M). If the transistor optimistically has 1nV/sqrtHz at 1mA, it would have 33nV/sqrtHz at 1uA, or about 100nV to 10Hz. I think the zener (or circuit) has a 1.2uV noise 0.1-10Hz. So even with 6Mohm, and even if the 1nV/sqrtHz was over-optimistic, transistor noise would seem acceptable. But the the 6M resistor noise should also be around 1uV at room temp. 

I think I need to get the noise amp and batteries connected.

You are comparing apples with pears.
nV/sqrtHz is rms voltage (effective AC voltage)
0.1-10Hz is peak to peak voltage
there is at least a factor 6-7 between both.

at low frequencies the (rms) noise voltage of semiconductors
has a 1/f behaviour so you have to carefully look at the measurement frequency.

With best regards

Andreas

[Galaxyrise]

When I played around with lowering the collector current, I remember the difference being pretty obvious on an oscilloscope.

[janaf]

You are right, the datasheet 1.2uV 0.1-10Hz is ptp.

The 1/sqrthz was a crude guestimate, to get a ballpark value of where it may end. Without good measurements or a good spice or other model, noise is out in the unknown...

[janaf]

I did a quick noise measurement 1s aperture on the DMM for 150 seconds. Results; pretty much that I can not measure noise in this setup, mains power, no shielding.
 
I measure near 0.6 uVrms noise with both 68K and 5M6 for R2, while the DMM measures near 0.28uVrms.

The 1.2uVptp 0.1-10Hz should be around 0.0625uVrms (1.2/6/3.2) -1Hz. Right? ie an order of magnitude lower. Hard to measure noise at this level....

Time allowing, I'll add an AC-coupled 0.1Hz-10Hz signal amp, shielding and battery box.

[janaf]

There is a possibility that a combination of an R60 of a smaller value and an R2 of a higher value will get you there without adding a lot of noise [the transistor noise is inversely proportional to the square root of the collector current, and at some point becomes significant over the Zener noise].

Good idea. Just tested it. Using half R60 on a linear scale (10R) and half R2 on a log scale (around 600K) we also get a zero for temperature sensitivity. We also get half the sensitivity of R20 (0.005 instead of 0.01) and unchanged for R2 at 0.004. Maybe a usable compromise.

[Galaxyrise]

That leaves the 7V->10V boost circuit, which is a different topic, and considerably more difficult to get right.  I'm still working on a PWM based circuit, but am trying to work through the variation of Tr, Tf, Tpd, Rsw,

I have a LTC1043 approach that works pretty well in simulation for 7.2V; I'll have to check if there's a good ratio for 7.15.  I've also been considering using more LTC1043s to do the temperature compensation (adding 1/6 of Vbe.)  Time will tell if I can make it work in practice

[Andreas]

I have a LTC1043 approach that works pretty well in simulation for 7.2V; I'll have to check if there's a good ratio for 7.15.

The perfect ratio would be adding 7.15 * 2 / 5 to the already existing 7.15 V giving 10.01V

With best regards

Andreas