Author Topic: Ultra Precision Reference LTZ1000 (Read 1343430 times)

Castorp · « **Reply #2900 on:** April 04, 2020, 04:26:53 pm »

Looks just like the LTZ1000 spectra I've measured many times in different ways.

My 89441A does produce 60 Hz internally, and sometimes it does show in the measurements.

notfaded1 · « **Reply #2901 on:** April 16, 2020, 10:25:41 pm »

My precision resistors are coming pretty soon I found out today! I just got these for the binding posts on my references. I'm hoping these will help some.

maginnovision · « **Reply #2902 on:** April 18, 2020, 09:59:39 pm »

Has anybody used Grafana to visualize data while logging? I'm doing this but I'm currently using a single right hand y axis for temps but the Internal temps are pretty far from Ambient. Does anybody know how to add a second Y axis? You can see the issue I have and maybe you can understand why I'd like dual scales to get more detail even if it may not be all that necessary. If you have any tips please share!

EDIT: http://172.119.228.219:3000/dashboard/snapshot/F3ptvJ1tIbHtU7X4roRV3Qlm0h5WboFv?orgId=1 Here is a link to a snapshot, since not everyone is using a 4k monitor.

e61_phil · « **Reply #2903 on:** April 19, 2020, 07:00:26 am »

I used relative ppm instead of volt for the Y scaling in Grafana. Two Y Axis will have different scalings and the reference voltages are hard to compare.

maginnovision · « **Reply #2904 on:** April 19, 2020, 07:06:06 am »

Quote from: e61_phil on April 19, 2020, 07:00:26 am

I used relative ppm instead of volt for the Y scaling in Grafana. Two Y Axis will have different scalings and the reference voltages are hard to compare.

That's on my list. I had some trouble getting the C++ interface for influxdb working right and since it's running I'm just going to let it go and get back to it after the ~7 days it logs(it logs to a CSV file, and a more verbose log as well as influx). Did you do that through math IN grafana or by passing the ppm data?

e61_phil · « **Reply #2905 on:** April 19, 2020, 07:14:53 am »

I stored raw data in influxdb (using Python) and calculated everything in Grafana, but one have to be careful with the queries on big sets of data. I killed my server several times (I have several million entries in influxdb)

maginnovision · « **Reply #2906 on:** April 19, 2020, 09:19:07 am »

Quote from: e61_phil on April 19, 2020, 07:14:53 am

I stored raw data in influxdb (using Python) and calculated everything in Grafana, but one have to be careful with the queries on big sets of data. I killed my server several times (I have several million entries in influxdb)

Would you be able to share how you do that? I'm having trouble calculating the mean and then using that to convert to PPM. The results always end up returning null.

e61_phil · « **Reply #2907 on:** April 19, 2020, 11:25:02 am »

Imho you can’t mix aggregate and non-aggregate data in a query. Therefore, I used a constant in the query. I know it is very ugly, sorry. If you find a good solution to calculate each point together with the mean of all points it would be very nice to know.

SilverSolder · « **Reply #2908 on:** April 19, 2020, 11:31:01 am »

Quote from: e61_phil on April 19, 2020, 11:25:02 am

Imho you can’t mix aggregate and non-aggregate data in a query. Therefore, I used a constant in the query. I know it is very ugly, sorry. If you find a good solution to calculate each point together with the mean of all points it would be very nice to know.

You can, in Oracle at least (Oracle Analytics). Other databases probably have something similar. Look for the OVER clause.

e61_phil · « **Reply #2909 on:** April 19, 2020, 11:36:23 am »

I was taking about influxdb

SilverSolder · « **Reply #2910 on:** April 19, 2020, 01:24:18 pm »

Quote from: e61_phil on April 19, 2020, 11:36:23 am

I was taking about influxdb

Is it possible to export the data and process it further in a different DB, where you don't need the high flow performance?

maginnovision · « **Reply #2911 on:** April 19, 2020, 03:11:20 pm »

Well I attempted both select mean("reading") as "mean"... And it worked but all nulls. I also tried select mean("reading') into "autogen"."mean"... Both cases they create "mean" and it's correctly populated but when using it in math like I said it just returns null for every data point. Maybe I'll try tracking the mean in my program and continuously updating it and then attempting to use the most recent one for the math. In my program I also have it perform 43 readings to stabilize(using a scanner) before logging so perhaps that's another place to get my mean value from, then I'll be able to have 1 constant mean value for every reading point. I'll try some things in a few days when it's finished. If nothing else a hard constant isn't that bad.

TiN · « **Reply #2912 on:** April 20, 2020, 02:25:06 am »

I've swapped reference in experimental 3458A to explore the low-noise theories. It went with few trap pitfalls, but meter is running now and logging first data.
Click on image to see livestream recording (2.5 hours).

Andreas · « **Reply #2913 on:** April 20, 2020, 05:01:13 am »

Hello Illya,

I am missing some kind of air shield for the LTZ1000s to get "real low noise"

with best regards

Andreas

maginnovision · « **Reply #2914 on:** April 20, 2020, 03:13:27 pm »

https://snapshot.raintank.io/dashboard/snapshot/1bEqz2W4r5htRG56dgdJ7j2xl5qnrNEd

I set it up so that I take the first 100 readings and calculate a mean with that to calculate ppm values. That single mean value is logged once per reading. In this case one problem is plainly visible. While those 100 readings were taken the temperature was rising but as logging started the temperature started to drop. I'll have to try a running mean next. It does look better with both readings and temps but I still would like a second temp scale for the ambient temp. This is ok until I can see if that's possible.

nnills · « **Reply #2915 on:** April 21, 2020, 08:32:43 am »

I must say it is a beautiful board you made there @TiN ! One question I have is why you spend your money on a ceramic Rogers material. Is it to reduce mechanical stress due to thermal expansion?

notfaded1 · « **Reply #2916 on:** April 26, 2020, 10:56:13 pm »

Ok I don't want to start major debate but I know many of you here have built multiples of different designs here. I have my LTZ1000's, I now have 4 sets of precision ww resistors for the important ones (this was what took a while). There are two PCB designs that can be ordered I found on github online. Dr. Franks design and cellularmitosis little board. Are there any other options I have? I was thinking of trying a few so two isn't a lot of variations. I'll have to order multiple PCB's anyhow from any FAB but that's not my concern because I'm sure any extras someone will want. Does anyone have any others I can try?

BTW I watched the livestream video of you installing the 4 LTZ board TiN... pretty wild stuff. I'm not going to try that. Nice fix BTW. I'll need a couple more 3458's before I try something like that I think

Thanks,

Bill

SvanGool · « **Reply #2917 on:** April 26, 2020, 11:54:08 pm »

Another one: TIN's singe LTZ1000 board https://oshpark.com/shared_projects/HfKcqjV3
Description here: https://xdevs.com/article/kx-ref

alanambrose · « **Reply #2918 on:** April 29, 2020, 11:16:39 am »

@branadic

I was very interested to see your LNA + AD2 vs. proper-DSA spectrum comparison.

Did you figure out what the shortcomings of the LNA + AD2 route were? It would be great to figure out a light way of getting good spectra instead of having to use a boat-anchor...

TIA, Alan

branadic · « **Reply #2919 on:** April 29, 2020, 11:29:05 am »

Well, the biggest shortcoming is, that you need an LNA, right? There are LNAs for several ranges such as 0.1Hz - 10Hz or 10Hz - 100kHz. What it would need is an LNA for the complete range 100mHz to 100kHz instead. There are several solutions out there such as DC-1MHz or commercial gear like Stanford Research SR560, but they are expensive. Couldn't find a proper DIY solution for the 100mHz - 100kHz range yet.

-branadic-

alanambrose · « **Reply #2920 on:** April 29, 2020, 01:34:33 pm »

I was wondering whether you had any thoughts about the apparent lack of 1/f noise using that set-up - somehow down to the AD2 or the FFT settings?

Alan

Kleinstein · « **Reply #2921 on:** April 29, 2020, 04:12:19 pm »

The LNA from Pipelie uses an AZ OP (or 2) and thus has essentially no 1/f noise, at least with a short. With a actual voltage there would be noise from the coupling cap if an electrolytic cap is used. I am not sure if the noise show for the battery is more from the battery or possibly from the coupling cap.

The downside of the usual AZ OPs is the limited bandwidth, especially if the first stage is used with a high gain. So 0.1 Hz to 100 kHz could become tricky. However 0.1 Hz to some 10 KHz may be realistic for such a setup without an intentional filter for the upper limit.

Similar one may be able to extend the lower limit a little below 0.1 Hz with a larger capacitor - however things are limited there. It is not just the cross over frequency, but also current noise (which gets visible as 1/f noise with a very larger resistance instead of a larger cap) and also dielectric absorption, so that electrolytic caps need really long settling before using it for slow processes.

For measuring down to really low frequencies the way to go would be using 2 reference and measuring the difference with a DC coupled amplifier, at least for the initial stage, possibly all the way with a DC compensation.

branadic · « **Reply #2922 on:** April 29, 2020, 05:13:51 pm »

Just for inspiration one could have a look at the schematic of SR560:

https://doc.xdevs.com/docs/Standford Research Systems/SR560/Stanford_Research_SR560.pdf

and build something similar to it, maybe with reduced functionality and more modern parts available such as MMBF5103 for the JFET, that was tested to have very low 1/f noise:

http://www.angelfire.com/az3/dimitri/images/AX_Dec2018_pp56-58.pdf

-branadic-

alanambrose · « **Reply #2923 on:** May 02, 2020, 06:08:29 pm »

Interesting, following the links I see the NPD5564 pair in the original SR560 is now replaced with a LSK389B pair which is available from Linear (and some other channels) for about $6. Maybe this is drifting off the original subject though...

MiDi · « **Reply #2924 on:** May 03, 2020, 01:00:17 pm »

Quote from: branadic on April 29, 2020, 11:29:05 am

What it would need is an LNA for the complete range 100mHz to 100kHz instead. There are several solutions out there such as DC-1MHz or commercial gear like Stanford Research SR560, but they are expensive. Couldn't find a proper DIY solution for the 100mHz - 100kHz range yet.

I know it gets a bit off topic...

For 0.1Hz to 100kHz Andreas 0.1 to 10Hz LNA could be suitable with some modifications.
With e.g. another LT1037 in second stage and tweaking the lowpass filters to get the cutoff frequency to 100kHz it should do the job and is affordable - I did similar modification a while back for other application.

AN159 10Hz to 100kHz/1MHz (Layout files) might be an option.
Downside is its limitation with low cutoff at 10Hz (1/f corner ~100Hz), so one needs at least second LNA for 0.1 to 10Hz and combine them - not so easy and not so cheap.

If I got it right the SR560 has quite high noise figure at low impedances, it seems better suited for higher source impedances (maybe > 1k $\$\Omega\$$ )


EEVblog Main Site	EEVblog on Youtube	EEVblog on Twitter	EEVblog on Facebook	EEVblog on Odysee

Author Topic: Ultra Precision Reference LTZ1000 (Read 1343430 times)

Share me