Playing with delta sigma converter, poor man's meter.

[0xdeadbeef]

I was once told by someone wiser that Sigma-Delta ADCs are not really designed for constant (DC) voltages and that manufacturers usually provide absolute error numbers for DC applications.
So my (limited) understanding is that Sigma-Delta ADCs are really designed for alternating voltages.

[joeqsmith]

Ended up swapping out the Linear Tech reference for a Thaler 0.6 ppm.  The temperature is just starting to settle.  I'll let it run overnight and we can see if it improves. 

[joeqsmith]

Showing the initial 5000 seconds after a half hour warmup.  It's a bit early to say how much, if any, the new reference improves it.   

[joeqsmith]

Comparing the 34401A for 7hrs.   While the standard deviation is higher for the HP, I just hadn't let it warm up long enough.   If we ignore the first 2.8 hours of data, the old HP does a much better job. 

[joeqsmith]

Repeated a 25 minute sweep using the 1.9Meg Caddock using the new reference.   First comparing it with all the meters I ran this test on, then deselecting some of the flyers and finally just with the HP34401A. 

[joeqsmith]

While I have been using this meter I was never happy with the results.   The last few days I have been thinking about how it could be improved.   The dominant problem has been temperature drift.  To combat this, the box has a fair amount of insulation around the sensitive area with a closed loop heater control.  For all that effort, the results were not very good.   A few ppm hear and there adds up fast. 

I am going to attempt to flatten it out using a BAV199 and resistor, similar to what I showed with the UT61E.   In the attached graph, the initial ramp is with the boards temperature increasing 2.5C with no compensation.     The longer section was with my first attempt at taming it with the BAV199 and a 10K ohm.   It will take some trial and error but the initial results look promising. 

Dave should do a video showing some basic compensation techniques.   I've been a fan of using 4148s for many years as a cheap method.     

***

The office can get fairly warm and to keep the internal temperature regulated, it will need to run above that.   The systems drift is not linear.   Next step is to sweep it over a wide range and plot drift and see is there is an area that looks like it may be easier to tame and be at least 10 degrees over my office on a hot day.    It will take some time to run all these sweeps.  It took several attempts to pick the combo I used on the UT61E.


***

To compensate this circuit, I would normally use my HP34401A but the Brymen is more than stable enough in the 500K mode to detect the drift trends.     

That's about the only thing that is going smooth so far...

[joeqsmith]

After several attempts, its still not very good.   The two yellow boxes are roughly the same temperatures being swept.  The lower is the last attempt.   This is again looking at the Fluke 731B's 1V reference and again, the meter is not being aligned between mods, which is why the vertical scales do not match.   

The temperature control has also been tightened down a bit which will help with the system errors.   

Another option I am thinking about is to just run a temperature sweep and do a fit to this data.  Just forget all this hardware mess and do it in software.  It's a one off meter anyway. 

[splin]

Unless I missed it, this would be much more interesting if you you would tell us which ADC and reference you are using - all I could see was that you have swapped an LT reference for a Thaler 0.6ppm. Is this a murder/mystery thread where we have to work out what you are testing?

Are there any prizes?

[joeqsmith]

These are just random rambles from my hobby.  If you have no interest, then don't follow it.   There are no prizes for the kids.  I have no plans to do any sort of crowd funding or open source design.

The problem is the temperature effects are not linear.   Each stage has its own signature.   The attenuator for example has a different signature for each range.   So attempting to collect all that data and then fit it, makes little sense even for a one off experiment like this. 

After making several more attempts to tame it, I am running out of tricks.  You can see it collecting at 4X speed here after warmup and alignment with the current setup.   
   
https://youtu.be/Kx2uw9XTDqk

[joeqsmith]

Same plots as before showing the new reference.  For the less observant, the part number has been the title for some time.  Green is after several attempts to compensate, compared with prior to compensation (Red) and my HP34401A bench meter (white).   

While it may look like things are getting worse, it's all part of the learning process and understanding the problem.   Two more tests to try.

[chris_leyson]

I think it depends a lot on the architecture of the converter and those designed for audio use are probably not DC accurate. Was comparing a Maxim MAX1416 16-bit dual channel sigma-delta converter against an Agilent 34401A today.
2.5V ref LM4040A, 10:1 frontend attenuator 180k/20k, PGA gain = 1, unipolar unbuffered mode and 60Hz update rate. Given that there is no ground plane, at least 20mV pk-pk digital noise on the 5V supply and it's stuck on top of a noisy Arduino Mega the accuracy and linearity was surprisingly good, +0/-2 LSB  for 0-24.999V FSD or 25,000 counts. Maybe the Arduino fixed point to float is rounding down the last digit.

[joeqsmith]

Welcome  to the club of home made TE.  Matching the performance of the 34401A is proving to be a fun exercise.   Like most of my projects, I tend to bite off more than I can chew.

I don't have any sort of input protection yet because I have only been using the source meter to measure the battery life of some handheld meters.   For this sort of measurement, it's good enough as is.   It seems like it's a good application for GDTs to minimize the error that path will create.   

Next test is running.  Clip showing the first hour, still collecting at 1sps, sped up 16X (like watching paint dry) and still using the Fluke 1V standard.   

https://www.youtube.com/watch?v=xfmNWr-39PI&feature=youtu.be

[joeqsmith]

Ran a couple more tests today.  There's no low hanging fruit to pick.  I think the next step is to try and work on the noise. 

[joeqsmith]

Todays sweep.  Note the placing the beach towel over the unit still appears to have some effect while the control loop compensates for it.  The ADCs and reference are both in inside their own insulated boxes, which are inside a box which is temperature controlled.   

In the first plot, I have switched to double precision but I am still only storing the data as single.  The UT181A software also works in single precision.   None of the meters I have would make use of it.  Using the 181A software to compare the data prior to this last week of testing various ideas, the P-P noise and drift are still about the same.   This is roughly 7 hours of data each. 

A harsh reminder of the time this hobby can suck of you with nothing to show for it.   

[joeqsmith]

During the same test, nothing taken apart, the gain was changed back to 4X where the original data for the VRE3025A was taken.   It's improved ever so slightly.   Wishing now I had stored the data differently as I can't really go back to show how much it's improved.   Splitting hairs.

[joeqsmith]

Comparing the final mods against the HP bench using both the reference and my old linear bench supply which drifts all over.  Should provide a good idea how well the two track.    Had shut everything down last night and I should have let it warm up a bit more.   All in all, the noise and drift look decent. 

[malagas_on_fire]

Hi Joeqsmith


Can we just talk calm  for a while in PM ? You have my word on honesty on previous posts / comments. 

[joeqsmith]

Planning to look at the TI ADS1262 evaluation board to see how it compares. 

[joeqsmith]

Their software it seems is bit lacking but they do provide the source.  As a bonus, it's in Labview which I have some basic grasp of.   They used version 2014 and I am still using 2011.  It wasn't a big deal to save it to the earlier version.   As before, I will run the ADC at the slowest rate which has a side benefit that their motherboard is able to keep up without any problem, unlike the person posting:   

https://e2e.ti.com/support/data-converters/f/73/p/800108/2962005?tisearch=e2e-sitesearch&keymatch=ads1263evm-pdk

The EVM is just sitting on the open table while I sort out the software.  Here are the first 15minutes collected from the Fluke 731B.   Gain of 1, 2.5Hz sample rate.   Large spikes are me sitting in the chair.   I'll let it run for a while and make sure the communications are solid. 

[iMo]

https://xdevs.com/review/ti_ads1262_p1/

[joeqsmith]

https://xdevs.com/review/ti_ads1262_p1/

LabView often is tricky and not very reliable in operation, as even after just ~10 minutes of fiddling with various tool settings I got it to crash with error:

So far, the stripped Labview code I put together seems solid but it will need to run for several hours before I move on.

*****

Ran for a few hours without a problem so I have gone ahead and added the ability to store the data.

[joeqsmith]

I've been using their Labview code (actually, it just calls their lower level code) for a full day now without any problems.   

Normally I would use my cookie can for the shielded box but it won't fit inside my meat packing box which will be required to get the temperature stable.  So I made up a quick enclosure with some PCB material, copper wire mesh and some RF gasket.  The small hole in the top is to gain access to their reset switch.   For this first actual test, I'll just use the towel to keep the air flow down.   

This is the first hour of data collected, played at 16X.  DRATE is 2.5Hz, PGA 2.  Shown with FIR but I have ran a few tests with the SINC4. This is a mistake after too many hours..  Its running SINC4.   

[joeqsmith]

Comparing the TI ADS1262 eval board's current setup against the HP34401A and the my other delta sigma.   The first graph, I ran a fairly long sweep collecting both the HP34401A and with the final changes I made to the source meter.  I was doing something in the lab and bumped the setup.   The data is stored double precision now. 

The second graph, I stripped the DC, or normalized them to zero. I then look at the start and end and attempt to remove the slow drift.  The old HP is doing a much better job with these small errors. 

[joeqsmith]

Graph showing nine hours.  I have no idea why the sudden drop but will continue to let it collect for the day.  The rate of change, considering it is wrapped in the towel, it's hard to believe it's thermal.   

[joeqsmith]

Human error.  Checked it after another four hours and it still hadn't recovered.  As soon as I removed the towel it snapped back.   I need to add some insulation to the inside of the box before a more major SNAFU occurs.   

Also thinking to try the home bundle.  It looks different from the version we tested with the watermark.
https://store.digilentinc.com/labview-home-bundle/