KX Reference

[kj7e]

Using one of Pipelie's LNA noise meters, here are some noise measurements comparing the LTZ1000 KX #1 in the Hammond enclosure above, vs the PDVS2 (LM399 based) at 7.14150v.

The LNA has a gain of 10,000, all measurement @ 10mV/div for 1uV/div.

Baseline LNA + scope noise, 280nV p-p;


LTZ1000 over 14 seconds, 1.12uV p-p;


LTZ1000 over 140 seconds, 1.256uV p-p;


So right on the money with the data sheet spec of ~ 1.2uV p-p.

LM399 based PDVS2 over 14 seconds, 6.04uV p-p;


LM399 based PDVS2 over 140 seconds, 5.68uV p-p;

[AG7CK]

A very telling post. Thanks.

[Pipelie]

Thank you! kj7e

Finally , I find a supply for the  Hammond 1590B in local, Thanks for kj7e who provide the part number of the case.

nice work on the black Insulation materials,How did you do that? it looks perfect fit in the case.

use a Coaxial cable to connect the reference and LNA, will get a little better result.

here is some of my test result.

[kj7e]

Thank you! kj7e

Finally , I find a supply for the  Hammond 1590B in local, Thanks for kj7e who provide the part number of the case.

nice work on the black Insulation materials,How did you do that? it looks perfect fit in the case.

use a Coaxial cable to connect the reference and LNA, will get a little better result.

here is some of my test result.

Fresh razor blade, ruler, Dremel tool and a steady hand 

Those where taken with coaxial cables and a dual banana to BNC adapter on the LTZ1000 box.  The LM399 based reference is about where it should be as its much more than a bare reference, there is a 10v buffer, DAC and other circuitry and another output buffer, so lots of things going on there to add a bit of noise.  The LTZ1000A is about where it should be but I have two other KX boards to test.  Ill compare all three soon and see if how they stack up.

[kj7e]

Some additional noise measurements from my latest KX (#3 with LTZ1000A date code 1750).

The test setup;


Baseline LNA/scope noise, 260nV;


14 seconds, 960nV;


140 seconds, 1.3uV;


Found some very interesting things regarding ground loops and my gear with this noise meter, very handy indeed.

[CalMachine]

I've got some measurement results of my 2 KX modules.  Dubbed 'KXBlue' and 'KXGreen' from the label TiN_MaN placed on them.

More tests will be run in the future.  KXBlue had a few ACAL incidents during its ramp, makes it not look so stable.  Here, the TC of the measurement system realllllly pokes through 

[Mr. Scram]

Forgive me my ignorance, but I seem too see a lot of different references and I think most or all are based on the same reference IC. What is the idea behind these variations? Are they attempts to achieve the same thing or do they have differentiated purposes?

[hwj-d]

...
Are they attempts to achieve the same thing ... ?

Yes, the ltz1000 and all around is fun! 

[hwj-d]

... and don't forget this little boy or girl inside all of us, who/she allows to be creative all over the lifetime

[cellularmitosis]

I jumped on the KX train!

I picked up a yard of 100% wool, and cut out a small piece to place between the LTZ and the board, to disrupt air drafts.  Hopefully the wool will be durable enough to be a long-term solution (it is a solid sheet of wool which I poked the leads through before soldering in the LTZ, so it is now permanently part of the build).

I went with hermetic resistors, as this ref will be (eventually) traveling to Taiwan to fetch the TiN Volt.  Hopefully they will minimize any travel-related offsets (e.g. regional humidity differences).

I threaded a 1N5819 schottky into the positive input, for reverse bias protection.

My surface mount soldering is pretty horrific, but it got the job done   C1, 2, 3, and 9 gave me particular trouble -- it felt like I was installing a 1206 part onto an 0805 pad.

Edit: wow, you can get an 8"x8" square of wool for $3 https://www.birdbraindesigns.net/wool-black

[CalMachine]

It's been a little over a year since my 2 KX modules made the journey across the Pacific Ocean to TW. 

With the advent of surpassing the year mark, I'd like to share my current results 

KX "Blue"




KX "Green"

[TiN]

That moment, when you get YT ads showing your PCB design, while watching EEVBlog video...



Not to mention stolen video content from Marco Reps   

[syau]

That moment, when you get YT ads showing your PCB design, while watching EEVBlog video...



Not to mention stolen video content from Marco Reps 

 

[beanflying]

Sux a lot. Have you and Marco tried to copyright strike them or get together with others who seem to have had their designs and videos 'borrowed' for their commercial gain? All us other viewers is give them a dislike or when they leave it available give them a serve in the comments section.

Not that they have a large youtube following but they do have a lot of paid and sponsored noise out there on youtube from content creators.

Code: [Select]

https://www.youtube.com/watch?v=dVEY3BxgHNA

Code: [Select]

https://www.youtube.com/watch?v=1FJXEL3Ff4Q

[grizewald]

People are still making KX references. Well, I am at least. 

As my Test Equipment Addiction has grown, I found myself needing better DC references when I bought my first 7.5 digit multimeter. Having read so much about the LTZ1000 and TiN's KX reference project here on the forum, I decided to build some! I have built two so far and used the project at OSHPark by SvanGool for the PCBs, so I might end up building a third. The BOM is as per the very helpful BOM on xDevs.com and I have used LTZ1000A reference ICs for both builds.

A big thanks to TiN for making this project available to everyone!

Reference number one was a bit of a learning experience as it was my first project with surface mount components. I think it turned out fairly well. I even managed to avoid the traps presented by the various guard traces!



Of course, getting hold of the right resistors for an LTZ1000 reference is a challenge, but between Mouser and my favourite Polish eBay seller, I procured the following resistor set for reference #1.

R5 = Vishay RCK02 12.2955K 0.01% 5ppm
R4 = Vishay S102 1K 0.01% 2ppm   
R3 = Vishay Z201 120R 0.01% 0.2ppm
R6 = Vishay RCK02 68.1K 0.01% 5ppm 
R7 = Vishay RCK02 68.1K 0.01% 5ppm

I think 12.29K actually turned out to be a bit low for R5, so I'll probably be replacing that with a 13K resistor instead.

Having got reference #1 nicely boxed up and discovering how useful it is to put the wires through the plastic frames which hold the end plates of the Hammond case before soldering the wires to the connectors 



I made a second parts order to start building reference #2. This time, I chose a slightly different resistor set with better stability.

R5 = Vishay S102 13K 1% 2ppm
R4 = Vishay Z201 1K 0.01% 0.2ppm   
R3 = Vishay S102 120R 0.01% 2ppm
R6 = Vishay RS92NA 75K 0.1% 2ppm 
R7 = Vishay RS92NA 75K 0.1% 2ppm 

Here is reference #2, all cleaned and waiting for the LTZ1000 and precision resistors to be installed. I skipped the pin headers this time as they don't really provide any benefit once the reference is cased.



Finally, I had reference #1 cased and had also acquired an Agilent 34401A. It seems to agree very nicely with the reference which is impressive for a multimeter which has not been adjusted since its original factory calibration some 20 years ago!

In the background of the shot is reference #2 which now has the resistors and LTZ1000 installed and is being made ready for installation in its case. The precision resistors are quite high off the board because they don't match the spacing of the holes in the PCB and I didn't want to bend the leads too close to the resistors' package. Both references have a foam ring to keep the legs of the LTZ1000 nice and cosy on top of the board and both also have a foam disc underneath the PCB to keep convection disturbances to a minimum there as well.



Both references are now being burned in and have been on constantly for about ten days now. Reference #1 is drifting slowly downwards whereas reference #2 seems to be very stable from the start. Once both references have been powered for at least a month, I'll start logging some long term data, but I noticed something strange in the data which I have already captured.

Below are two plots, one taken on my rather noisy Solartron 7061 and the other (which is currently running) from the 34401A.

(The marked downwards trend, followed by a recovery near the end of the first plot is due to me experimenting with the drift compensation feature on the 7061. Just at the end of the plot, I manually initiated a drift correction. The voltage recovers almost to the original level, but not quite and further testing with reference #1 shows that it's still slowly drifting down, hence why I think I'll swap out R5 for a 13K resistor.)





The common feature here which is puzzling me is the occasional negative going 'pop' from both references. Is this normal during the initial settling in period?

[Dr. Frank]

These downward spikes are typical for the original schematic, which probably picks up any external disturbance. 
These may last several seconds, or several samples, then it's a sign that the oven control being disturbed, if it's only a single point each, then this might be disturbance of the 34401A input only.
Therefore that has nothing to do with settling, this will be a permanent effect.

Check your PSU, if it's possible to shield it, or add a ferrite ring.

Other circuits (from Andreas) greatly suppress these spikes, maybe you go back into the history of this famous "Ultra Precision Reference LTZ1000" thread, where this was demonstrated several times.

This 12.2955K resistor is fine, gives about 55..60°C oven temperature. You should determine this on your own, using your 34401A. Method is described also deeply back in this thread.

Btw: If you encounter longer dips, lasting for many minutes, hours or days, TiN himself called that 'jumpy reference', then this would be caused by the A version itself.
Some samples show these jumps, due to not completely annealed die cast.

Frank   

[Kleinstein]

The measurements with a DMM are only as good as the DMM. So with the 34401 and Solatron 7061 one would see more of an effect of there reference (LM399 and likely 1N829). The large spikes are more like an external problem.

[notfaded1]

The measurements with a DMM are only as good as the DMM. So with the 34401 and Solatron 7061 one would see more of an effect of there reference (LM399 and likely 1N829). The large spikes are more like an external problem.

@Klenstein is so right and I found this out myself.  It started with my 34401A measuring precision voltage reference which eventually led me to the 3458A.  Better voltage reference requires a better DMM and so the tug of war goes!  Then you end up like some others here and need 4 or 5 different 8.5 digit MM which, of course, requires more precision references.  And with all the 3458A's you need a better resistance standard.  That's how it goes: 

B

[grizewald]

Thank you for the replies gentlemen.

I'm fully aware that the meters I own have references far inferior to the LTZ1000, but that doesn't reduce the usefulness of the KX references to me. Quite the opposite in fact as the KX reference is capable of showing up problems on the meter itself and I can still get a good impression of how stable the references are after burn in with my inferior meters. This should let me choose a good time to post them to a friendly cal lab or fellow volt nut for characterisation.

I'm still looking for a proper metrology grade meter and was in the running for a Datron 1281 yesterday, but the bidding quickly passed the limit of my test gear fund in the remaining few minutes of the auction. So for now, I'll have to make do with what I have.

When I first saw these excursions on the 7061, I immediately suspected the meter, but seeing them appear when using the other reference on the 34401A, I began to wonder if it was specific to the KX reference itself.

They are both being run at 12V from a Thurlby PL320 linear lab supply, so the power is quiet and stable.

I looked through the reference #2 log and found the three samples with large negative jumps and indeed just one sample is involved, just as the single line thickness on the plots would imply:

Code: [Select]

2020-03-30 23:46:44,24.90,30.28,1029.25,+7.11377860E+00
2020-03-30 23:46:46,24.90,30.28,1029.25,+7.11371180E+00
2020-03-30 23:46:48,24.89,30.29,1029.26,+7.11377920E+00

as is the big jump of 20ppm:

Code: [Select]

2020-03-31 00:07:38,24.86,30.41,1029.22,+7.11377630E+00
2020-03-31 00:07:40,24.86,30.41,1029.22,+7.11363070E+00
2020-03-31 00:07:42,24.86,30.41,1029.22,+7.11377340E+00

and the third blip at just after 00:30

Code: [Select]

2020-03-31 00:35:35,24.76,30.65,1029.10,+7.11377550E+00
2020-03-31 00:35:37,24.76,30.65,1029.10,+7.11369950E+00
2020-03-31 00:35:39,24.76,30.65,1029.10,+7.11377570E+00

The two spikes on the 7061 log from reference #1 are similar, just one sample blips, one just before midnight and one just after:

Code: [Select]

2020-03-22 23:48:20,25.52,31.14,1039.00,+7.134534
2020-03-22 23:48:36,25.52,31.14,1039.00,+7.134428
2020-03-22 23:48:52,25.52,31.14,1039.00,+7.134538


2020-03-23 00:04:06,25.27,31.40,1038.94,+7.134537
2020-03-23 00:04:22,25.27,31.40,1038.94,+7.134497
2020-03-23 00:04:38,25.27,31.40,1038.94,+7.134535

I also saw a line at just after midnight in the reference #2 log with a problem caused by my logging system where the script seems to have been updating the environmental data at the same time it was being copied to the new line in the log file resulting in missing fields between the time and voltage columns. (The script runs multiple background processes.) Oddly, gnuplot doesn't complain about the empty columns when it processes the file.

Code: [Select]

2020-03-31 00:09:57,24.89,30.51,1029.26,+7.11377890E+00
2020-03-31 00:09:59,,+7.11377990E+00
2020-03-31 00:10:02,24.92,30.51,1029.28,+7.11377880E+00

So I'll need to fix things so that can't happen again!

From this, I'd infer that the reference chips are thankfully not one of TiN's "little jumpers".

The reference that is currently connected to the 34401A is not using the leads in the picture. It is connected with a twisted pair of PTFE insulated silver plated copper wire with low thermal EMF spades on the reference end and Stäubli low mass banana plugs on the meter end. The reference connected to the 7061 would have been connected with the original Solartron measurement cable - also very high quality and shielded.

The only local interference source I can think of is my temperature/humidity/pressure sensor which is sitting on the desk near the reference under test in each case. This sensor reports back to a local MQTT server wirelessly, so there's a 2.4GHz wireless transceiver close by operating for a very short time every minute as the sensor publishes the latest readings.

I can imagine that transmission being capable of causing a disturbance, but with the 34401A returning a new reading every two seconds, I'd expect the disturbances to affect the result much more often than just two readings out of the nearly 41000 that there are currently in the log. I'd also expect the disturbance to have an equal possibility of generating a positive or negative going disturbance and I've never seen a positive one.

To attempt to see if that is the cause anyway, I'll move the sensor a few metres away from the desk and see if I have any ferrites to put on the power leads from the Thurlby supply. The temperature is fairly constant in my apartment, so moving the sensor away from the bench shouldn't affect the relevance of the data to any significant degree. At least it won't get caught by the morning sun as it did today, resulting in the temperature spike at around 08:00 before I shut the blinds on the windows.

edit: The line with the missing columns was not responsible for the largest 20ppm jump. My mistake.

[Kleinstein]

The LM399 reference in the 34401 can by tricky with random jumps from a kind of popcorn noise. So it can appear perfectly stable for minutes and than suddenly just a few µV. A little like TIN little jumper. The only good thing with a well aged 34401 is that there would not be much normal drift with time. Here a new LTZ1000 may actually be inferior. It is still tricky with possible low frequency noise.

About the only way to really judge the reference better than the meter is to use both LTZ1000 refs and use the meter to measure the difference only.

The jumps just a round midnight are somewhat suspicious. One rather strong EMI source are mobile phones. From time to time they send a really strong  (e.g. full power) sync / station search pulse / command (a few seconds, maybe repeated a few times). This may happen just around midnight. 

[grizewald]

The jumps just a round midnight are somewhat suspicious. One rather strong EMI source are mobile phones. From time to time they send a really strong  (e.g. full power) sync / station search pulse / command (a few seconds, maybe repeated a few times). This may happen just around midnight.

That's a very interesting point and I know you've been investigating RFI recently.

I'll make sure to keep my mobile in another room this evening and see if there are any disturbances tonight. I'll be leaving this particular log running for a few more days, so I may have the opportunity to see if there is any correlation. It also just occurred to me that my daughter's bedroom is just behind the bench on the other side of the wall where my bench is. That particular wall is a load bearing wall and is made from concrete and rebar, but I can't imagine the rebar having any effect on such high frequency signals.
 

[grizewald]

I just went back and edited my post from 17:49. It turned out that the -20ppm jump was for real. My mistake.

Reflecting on the discussion so far, I'm finding it very hard to reconcile either popcorn noise in the meters' references or EMI with this problem as I just can't see why it would always result in a negative going spike and never a positive one. Maybe there's something specific about the length of a PCB trace on a specific node in the reference circuit making it act as a perfect antenna for a specific frequency and always manifesting as a negative going spike. 

I'll carry on trying to eliminate possibilities.

[Dr. Frank]

I just went back and edited my post from 17:49. It turned out that the -20ppm jump was for real. My mistake.

Reflecting on the discussion so far, I'm finding it very hard to reconcile either popcorn noise in the meters' references or EMI with this problem as I just can't see why it would always result in a negative going spike and never a positive one. Maybe there's something specific about the length of a PCB trace on a specific node in the reference circuit making it act as a perfect antenna for a specific frequency and always manifesting as a negative going spike. 

I'll carry on trying to eliminate possibilities.

Excuse me,
I think that I have given you the correct hints, as I have seen exactly the same effects (negative dips) either with my 34401A, over 15 years ago (yes, I was probably one of the first amateurs to build two LTZ1000 circuits), I also observed always negative glitches and these were on the order of -1..5ppm, sometimes bigger. Later on, about 10 years ago, with my then freshly bought 3458A, I saw exactly the same glitches, maybe less, beacuse in the meantime I made some improvements on the (linear) power supply. Hopefully, you're not using a cheapo SMPSU ?

The better reference of the 3458A makes things easier, i.e. less noisy and more stable (relative to each other), but the essential characteristics can very well be extracted with a 34401Al, please take Andreas as an example. Thesed glitches you have observed, have absolutely nothing to do with the quality of your 34401A.
Better assemble a 2nd , or a 3rd LTZ1000 reference, to get a really good voltage baseline.
 
My whole story is on the cited LTZ1000 thread, you may better look there.

Causes of these dips might be from switch mode PSUs, either in LED lighting, (noisy) PC PSUs, cellular phones, WLAN, BT, and so on, even from ordinary glitches in your mains supply, from coffee machines, washing, machines, and so on.

I operate all of my 7 LTZ1000 references down in the basement, there is only old school 50Hz incandescent, or fluorescent light, and a lot of concrete with steel armour around.
Anyhow, only additional shielding measures (tuner box instead of aluminium), ferrite rings and of course some blocking capacitors in the right places (Andreas design) mitigate such glitches to zero.

Frank 

[TiN]

grizewald
Very nice write-up, thanks for sharing.

As of jumpy reference. It was one particular reference chip that I talked about. (Out of 30+ various LTZ references I've built total). I don't believe that jumps is something related to LTZ1000ACH version specifically. I mostly use ACH because it is easier to deal with and have much more linear and predictable TC, didn't meet other jumper yet.

That jumpy chip came from eBay A9 3458A board, that I originally bought for my second 3458. After desoldering chip from A9 board and putting onto my KX - it was confirmed that chip itself caused jumps. I tried various things to try mitigate the jumps, but to no avail. Obviously after replacing chip no jumps were present in original design.

Perhaps easy test would be to monitor reference for few days using battery power (such as car/motorcycle 12V battery) inside a shielded can. You can use large cooking can with drilled hole for short shielded cable (or two coax cables) to DMM. This would eliminate possible ground loop currents and power supply noise interference.

Another thing to eliminate the DMM , which can have own jumpy LM399 inside. So for that you can connect both references in opposition and monitor difference with 34401A. Then jumps from DMMs ref could cause much less change, when you just measure few millivolts of difference between REF1 and REF2, and you'd be comparing amplified by 100 input vs internal ppm-level jumps of base 10V range.

[Dr. Frank]

Just a reminder, that somebody else already has given this hint:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg1879310/#msg1879310

Last year, I measured two A version references, for minimizing their T.C.s and I also saw these 0.5ppm shifts.
Maybe I show these measurements elsewhere.
Frank