Ultra Precision Reference LTZ1000

[Andreas]

I wonder if I could use small signal latching relay, such as TQ2-L2-5VDC that is used on some of the Keithley scanner cards to switch the output between 7V and 10V to see the drift of the 7V->10V step up. With careful thermal planning and routing both lines similarly (placement, junctions) it shouldn't be an issue?

how much insulation - I bought some Spacetherm offcuts so I might use that, but I've read here that overinsulating LTZ1000 is detrimental. In 732B they insulate the whole lot, but I guess it is because LTFLU doesn't have a heater and they heat the whole lot. Spacetherm still might come handy to insulate VHP202Z coupled to small heating resistor.

Hello,

I also use the TQ2-L2 5V relays in my relay multiplexer for the daily measurements.
Since the coil is mostly unpowered there should not be much issue with thermals when using a metal housing.
A contact cleaning routine (like on some instruments on switch on) could be a good idea when you are using it not every day.

I do not have fear with over insulation. Most of the heat goes out by the package over the package pins and the PCB. The main effect of isolation is lesser noise from thermal effects.
If your heater falls out of regulation you will get 50 ppm/K response on the output. (which is hard to ignore).
So if you test your cirquit 5-10 deg C above your max environment temperature you should have enough headroom.

With best regards

Andreas

[VK5RC]

Sounds like enough work for next 3-5 years, that's for sure. LTZ1000-based circuit deceivingly simple at unprepared eye, but quickly morph into great time-vampire project, as you get your feet wet. And you can't escape it afterwards! I fell into this trap 4 years ago, and still no end. Memo for "don't-become-voltnut" members - do not build LTZ circuits 

Maybe you can use just one step-up 7V-10V module, instead of per module. That way you can save some BOM and time, as this circuit is equally difficult, if not more difficult than LTZ REF itself, if goal also not to compromise LTZ's own stability performance.

As for me, I'm tweaking PID for new bigger TEC box, which can fit VK5RC's KX reference. So far no good, need more work. Big chamber slows things a LOT.

TiN Thanks for publishing the work, .  I was hoping for the reference to have a bit more of a 'time-lag'.   It seems to change its internal temperature pretty quickly
Oh well - need to think of version 2!!
Rob

[TiN]

And it ran out of temperature headroom once reached +42.86 °C temperature (PCB) / +41.3 °C (TEC BOX), woohoo. 
As reminder for others, VK5RC's KX reference module have 12.0K/1K temperature setpoint. Based on this I'd recommend keeping ambient below +30C for this setup, to be on a safe side.

I'd say 12K/1K bit too low for travel reference, as for example here in summer ambient can be around +32-36°C, and it's not even considering more warm countries like AU,Africa,etc.

+3°C over that gives figure +128ppm, which translates into ~+42.7 ppm/K of LTZ positive zener tempco without temperature oven running. This just stresses again on fact, that LTZ1000 performance is vitally depends on internal oven stability. 

"Time-lag" is about 300 seconds between my TEC box and KX module temperature sensor. I've reset PID parameters, hopefully will get smoother temperature ramp next days.

DSV-data for those who play with numbers at home. 

[lukier]

Sounds like enough work for next 3-5 years, that's for sure. LTZ1000-based circuit deceivingly simple at unprepared eye, but quickly morph into great time-vampire project, as you get your feet wet. And you can't escape it afterwards! I fell into this trap 4 years ago, and still no end. Memo for "don't-become-voltnut" members - do not build LTZ circuits 

Yes, this is definitely one of these projects And maybe it's a good thing, as voltnuttery is so expensive so it is good to spread it over 3 years. Recently this kind of shopping spree peaked (eBay resistors + Digikey), as I want to build the small reference boards relatively quickly so they can age in some shoebox with a linear PSU for the next year or two while I learn more, test various concepts for the power supply and 7->10 step up and save enough for 3458A

[mimmus78]

I still do not get what is the idea behind a common heater set point.
Do you want to trim the output voltage by setting the temperature (around 50 ppm/K).

The idea is to substitute this 12K/1K signal with a 4 pwm controlled signal in the future
so basically the mother board will be used mainly for checking if everything works.
I accept your advice to add other 3 other 12K/1K positions in case one is not stable
enough.

What i'd like to have is a complete PWM driven temperature set point.

1. it's should be more stable than the resistor divider in the long term
2. you can use it for trimming output of each of the single module
3. maybe you can use it for implementing some minimal temperature compensation

But how do you do the star ground concept + kelvin sensing so that the heater currents do
not influence the output voltage?
Without individual (galvanic isolated) power supplies this will be a demanding task.

High side of each LTZ1000 module is will be buffered, so when connected for averaging it should
not influence each other. My plan is to stick together all four zener kelvin connection on the
lower side. This seems to me exactly how Datron 4910 is made. If possible I'd like to avoid
the 4 (or 5) insulated power supply for this iteration. Does will work? Don't know.

Consider I plan to have different grounding pin for heater, opamps, zener and kelvin
connection on each LTZ1000 module.

[Kleinstein]

The output of the normal LTZ1000 circuit is already buffered. So there is no need to have another buffer before averaging. Even if a LTZ2057 will not add much offset,  it will add a little noise and also possible EMI problems - so it is worth avoiding it if you don't need it.

However there are usually no separate drive and sense lines - so the connectors can get critical (need low R) if low resistors are used for averaging. Connecting the negative sides can be more tricky - one option would be separated supplies or alternatively averaging resistors as well. This would require a buffer for the negative side as well and the negative side would be slightly different from the power supply GND. Depending on the application this might be acceptable or slightly inconvenient.

The positive side buffer might already be more than just a buffer - some of the 7 to 10 V circuits have a high input impedance and might directly connect here.

For the temperature set-point I would be careful with a PWM generated voltage - any failure here could lead to over temperature and there would also be possible noise coming in this way. It might be enough to have the option to measure the actual divider ratio (that is have an output) and allow for some initial calibration measurements with a slightly shifted temperature. A possible drift of the set-point could than be corrected numerical. E.g. you know the LTZs TC with respect to the set temperature (around 50 ppm/K, individually measured) and the measured shift in set temperature (or set voltage).

So the modules could be just normal ref. circuits with just extra connections to have access to the divider and enough sensing lines. No need to use separate LT1006s - the dual LT1013 is just fine. For the individual voltages something like series resistors might be enough protection against something like a short or excess caps. Output supervision might be an option - thus to turn of the heater if the output voltage is much too low.

I don't think it is a good idea to have the temperature sensor close the reference, this area has temperature gradients and extra traces add heat paths. The more suitable place would be the resistors.

[Andreas]

My plan is to stick together all four zener kelvin connection on the
lower side. ....
Does will work? Don't know.

Consider I plan to have different grounding pin for heater, opamps, zener and kelvin
connection on each LTZ1000 module.

Hello,

It´s hard for me to understand how you will do that.
You can either connect the heater+zener currents at the power supply or the kelvin sensing pins.
But not both at the same time without getting ground loops and influencing voltage drops.
Of course you can use sense force OP-Amps for the zener grounds or true differential sensing for each zener.
But this will need additional efforts.

I am also not convinced that a PWM-ratio is easier than a resistor ratio.
It is only different.

Of course you are right that the long term stability should be better.
On the other side you have to
- filter out the PWM-noise with non-perfect filters.
- correct the ratio for T.C. anyway since the rise and fall times of your switches are temperature dependant.
- so carefully select the PWM-frequency between both issues

minor effects: (at least for the LTZ setpoint).
- regard input voltage dependant R,on of the switches that will influence INL linearity.
- use switches with low charge injection
...

with best regards

Andreas

[mimmus78]

It´s hard for me to understand how you will do that.

I think this is a polite way for you tell me that I need to take more attention to this problem I may overlooked 

Connecting the negative sides can be more tricky ... or alternatively averaging resistors as well.

OK let see if I understand you both guys.

PIN 7 of LTZ1000 will be a little bit over the GND and each of the LTZ1000 GND point will be a little different
than the others. If I connect this four pins together I get strange things to happens due to some current flowing
in other paths than "expected" (ground loops).

This means that I need to use an op-amp to cancel out this "offset" and lower everything to a common GND
point (this point than can be used as the LO side of the reference). This 4 current cancellation op-amp must have a
negative rail to push everything down.

Is this correct?

This meas also that special care must be taken to have also the 120 resistor GND side to almost on the same GND level.

I am also not convinced that a PWM-ratio is easier than a resistor ratio. It is only different.

Never told is easy (especially for me). What I was thinking is that the thermal regulation should be a little bit
forgiving so if you can only get it stable to 10ppm (I think) it will be perfectly acceptable. How much is hard to
get it I still have to investigate. Thermal lag may also fix a non perfect pwm filtering if it don't screw the
thermal regulation circuit.

The output of the normal LTZ1000 circuit is already buffered.

Yes but I want to isolate LTZ1000 from capacitative load problems (had a lot of problems in past that I want
to avoid). Than I'd like also to have independent channels for sporadic itercomparison. Don't know if this can
be done with simple resistive averaging.

For the temperature set-point I would be careful with a PWM generated voltage - any failure here could lead to over temperature and there would also be possible noise coming in this way. It might be enough to have the option to measure the actual divider ratio (that is have an output) and allow for some initial calibration measurements with a slightly shifted temperature.

That's nice. Maybe the best is still to use the 13K/1K divider and than add a fine adjustment by PWM.
Measure divider ratio time by time (any 6.5 digits multimeter should be sufficient for this) and digital
compensate for its drift.

Considering that we need just few hundred ppm margin to compensate for divider TC and log term drift
this should be easier to do than a 100% PWM signal. This will also ensure that you will not cook the
reference in case PWM circuit fails (PWM only contribute to a very limited amount on the temperature
set poit). Maybe also a 10/12 bit DAC should be enough stable for the same adjustment purpose.

[Kleinstein]

The outputs for the individual references might want an extra buffer, but for averaging the 4 reference signals inside one does not need a buffer before the resistors to get the average voltage.

If it just to measure the 4 individual references, one could use a resistor or RL combination to isolate the single references from a possible external capacitive load. Just connecting a DMM might not mind an 1-10K resistor in series.

It is only to drive larger loads, like KV divider that one would need a true buffered output, but this would also prefer to have separate drive and sense terminals, not just buffered with 2 terminals, but more like 4 terminal.

Having 4 OPs for current cancellation at the negative side, is just one option, but this also is a lot of effort and it adds the noise and offset of that OP. Depending on the rest of the circuit, there are other options: one would be using 4 resistors to average the negative sides of the refs as well. This is like controlled ground loops. The downside of this is that one would need a kind  of buffer at the low side as well.

To to anything at the GND level one very likely will need a kind of negative supply (could be a virtual GND) anyway. Having a negative supply available would also give a few more Options for the OPs used in the reference circuit (e.g. LT1097 instead of LT1013).

[mimmus78]

Thanks Kleinstein and Andreas for your considerations, I think I made my mind to where to try go.
I'll share my progress in next days.

[lars]

About RC55Y mentioned in an earlier post.

I once tested 10 samples of 1kohm 15ppm/K from Farnell mounted on FR4 boards without power. The first two months the drift was up to 30ppm but after that for a couple of years the drift was about 5ppm/year. The humidity sensitivity were about 25-35ppm for an RH variation of 40% at around 23C. Temperature coefficient varied between about +10 to -15ppm/K. Actually four of the ten samples were at about -15ppm/K. No sample were within +-5ppm/K!

/Lars

[lars]

I did a first try to measure noise on the LT5400-8 as I had a sample mounted on a SO to DIP adapter (FR4 board). As I had no available low noise amplifier I measured with my best DMM at home that is a HP3456A with 0.1uv resolution. I made a bridge out of the LT5400-8 with 1+1k in one arm and 10+10k in the other. As I expected I couldn´t see any difference with 0 or 10VDC applied except the offset DC. At 1PLC I had about 1uVp-p at 10PLC 0.3uVp-p and at 100PLC 0.1uVp-p (last digit).

My setup was probably not the best (see picture). Cheap 1 meter long unshielded banana cables, some silver plated and others even nickel plated.

Lars

[lars]

I have to ask a question related to 1/f noise and resistors for the LTZ1000 standard design that might be important and that I don´t remember to have seen an answer for.
 
From the LTZ datasheet I see that the zener noise at 1 Hz is about 60nV/sqrt-Hz. My question is how much noise (sqrt-Hz at 1Hz) is needed on the respective resistors R1-R5 to get the same amount out as the LTZ?

In another way: how much is the noise attenuated from each resistor to the output. With just a brief check my guess is that the worst one is the 1kohm that seems to be attenuated only about seven times? But as the 1kohm only has 0.5VDC the 1/f noise should not be high even with a thin film SMD like ERA6, PCF0805, RG2012 or RN73 from what I see in the papers from Seifert (in previous link by Kleinstein in post 1919) and Maerki (ETH)??
https://people.phys.ethz.ch/~pmaerki/public/resistor_flicker_noise/20130723a_white_paper_resistor_flicker_noise.pdf

Another question is if 1 Hz noise and DC voltage (eg thermal EMFs) from the resistors have the same attenuations?
Sorry for being a bit lazy and not setting up a simulation but somebody might already have the information or be quicker than me to get a result?

Lars

[Andreas]

Mhm,

my low noise amplifier uses normal metal film resistors.
(1K as input resistor for the high pass).
The noise floor is a factor 5-10 below the noise of a LTZ1000.

And most of the noise comes from the leakage current of the input capacitor.
So I think we do not need to worry about the LTZ cirquit resistors.

with best regards

Andreas

[Kleinstein]

The resistor excess noise only appears if there is a significant (DC) voltage at the resistors. In the low noise amplifier, there will be essentially no DC voltage on the resistors at the sensitive part. So there is no problem at all.

This is different for the LTZ1000 circuit or an amplifier to scale from 7 V to 10 V. However in the LTZ1000 circuit noise from the resistors will like drift of the resistors attenuated. I agree with the first estimate that the 1 K resistor noise would be attenuated least, by about a factor of 7. Also the still low voltage can keep the noise contribution low.
Chances are the 70 K resistors could give more contribution, despite of an attenuation by a factor of about 200.

So resistor excess noise will not be a problem in the reference circuit itself. It is more of a potential problem with a scaling amplifier.

[MisterDiodes]

A couple of notes here RE: Resistor noise in LTZ circuit:

Just because you can't see much excess resistor1/f noise on the Vref output doesn't mean it's not there circulating across the LTZ crystal lattice.  You want any noise in that LTZ substrate as low as possible - even its own Zener will make noise, but as you watch that die over 5~10 years you can see the effects of the die's crystal lattice approaching a truly relaxed state - and even the Zener noise will typically diminish to a much lower value than when it was an infant.  Sometimes not, but usually.  Stress on the die means there are bonds changing position, and that means electron pathways changing position, created and destroyed, etc.  Edwin had a good analogy with water flow:  It's like rocks tumbling and moving around in a small stream:  You can hear the water noise when the stream flows over rocks, but you can't hear the water's noise if the stream flows over a smooth muddy channel.  Same idea applies here, on an electron charge carrier level.

The more current noise you have flowing across the lattice, the less likely it will become stable over time - or it won't become as stable as possible I should say.  That's why we use PWW around an LTZ circuit whenever possible and practical, and that's why the datasheet recommends their use.  ANY other resistor type (unless we invent new physics at some point) will always add excess noise across that LTZ and thus decrease stability - and that's the same reason we don't use an AZ amp as the current driver, with its very high current noise injected into the substrate.

We're just trying to give that LTZ the best possible chance of becoming stabilized with low drift and lowest output noise.  It will take time for everything to stabilize.

The 70k's are the main contributors here.  Even though the absolute value of 70k is not too critical, you really want the lowest noise possible across that LTZ for best long term stability.

RE:7V to 10V booster output:  Yes, you really have to be careful here as any gain error or noise will get boosted up.  And as pointed out: You can't average out or low-pass filter 1/f noise very well - you just get noise energy at lower freq. The goal here is to not generate any noise if you don't have to - in the real world that means generate the least possible noise. 

For use with high rez ADC's and DAC's though you're generally NOT trying to make 10V...you really want a lower voltage as a reference point for more modern chips...usually no more than 5V.  So a lot of times for real applications we're dividing the Vref output DOWN and buffering - which is a little easier. 

[enut11]

Hi All
I now have most of the parts to construct my first LTZ1000 project. All that is missing is the LTZ chip. The main resistors are all Edwin Pettis supplied Type 802 and the PCB is from @mimmus78.

I came across this chip listing on eBay:
http://www.ebay.com.au/itm/401303985615?_trksid=p2055119.m1438.l2649&ssPageName=STRK%3AMEBIDX%3AIT

Is this a good deal at $55US or too risky? If no good, what source of LTZ100ACH would you recommend?
enut11

[julian1]

Heads up, you can buy straight from Linear Tech site, with a debit/credit card.

And contrary to other reports, there is no minimum $250 purchase order (that's for a trade/commercial order).

For LTZ1000CH#PBF I paid $42.85 USD.

There is a USD $54 shipping charge to Aust however. For that reason I ended up getting 2x ltz1000 and 2x lm399.

For me, knowing the provenance of the refs was worth the money, given the time I ended up putting into them.

[TiN]

enut11
That is 100% fake.

Second the comment to buy directly from LT, they are happy to sell you even 1pcs. Or at least from reputable retailer. Digikey has some in stock.
I'll be getting 3-4 chips this week, could chime in 2 for you if you want. Shipping from TW would be 15-20$, I think.

[enut11]

Thanks gents for helping me avoid a big mistake.

TiN, I would like to take you up on your offer. I will PM you.
enut11

[d-smes]

Going back to grounding, isn't Figure I2 from AN-86 the WRONG way to do Kelvin Sensing?   You see, Zener (and other) currents create a voltage drop across the Zener-Force trace resistance (red marks in attached).  This makes current want to flow in the Kelvin sense connection (green arrow) which defeats its purpose.


So, what's the proper way to implement Kelvin sensing?  One way would be to buffer the Kelvin point, but then you'd need a negative supply.  A more practical implementation might be to treat the junction of R5, R1 and Pin 7 as a Kelvin star node and use that as your Kelvin Sense and as a star point for return currents from e.g. resistive divider of a 7V to 10V output converter as shown in second attachment.   Comments?  How do you all do it?

[2N3055]

Output of LZ1000 is taken from pins 3 and 7.. It is combination of zener and transistor that is used...

[Galaxyrise]

I'm not sure why AN-86 appears to suggest creating a ground loop.  The "7V Positive Reference Circuit" in the data sheet is clearer.

[pitagoras]

Is there a reference design PCB for this "normal" positive 7V reference?

[Micke]

On the Eval. board for AD5781, they connect ground to the resistive divider directly to pin 7 on LTZ1000.
I have done like this on my first LTZ1000 design, I have not experienced any drawbacks yet... 
They also have a RC filter between unbuffered 7V and OP-amps with 1.5k and 10µF, quite big capacitance! I thought of doing like this but reduced the cap to 100nF instead, I was afraid that tiniest leakage current of the cap would introduce errors. I was thinking of though if having 10µF use a solid wet tantalum.
Schematic AD5781 Eval board with LTZ1000: https://ez.analog.com/servlet/JiveServlet/download/13219-2-41351/EVAL-AD5791SDZ_LTZ1000A.pdf