Fluke 8505A : 10V Average 7.5 Digits - Experiences and Opinions wanted

[AG7CK]

I bought this meter defect (USD 50 + three times as much in shipping) out of curiosity and - more importantly - because I wanted a commercially made REFAMP type voltage reference based on SZA263 or LTFLU-1. I have fixed the meter (tantalum syndrome).

It is quite stable (as far as I am able to determine by using the best of my HP 34401A / 34970A / 3456A for comparison). So to me it is absolutely a keeper - a bit odd or almost weird both in user interface and construction / looks, but definitely worth studying and using. I will only use the 10v DC range, 6.5 digits and 7.5 digits average mode.

My first question that I am totally unable to answer myself, is that I suspect the ADC (discrete analog build + microcontroller) of having "missing codes". And I think that the phenomenon even turns up in average mode (which IMO hardly should be possible if the mode is software averaging).

The plot below is 2000 samples over 3.5 hours at constant temperature of a LM399. The samples are 7.5 digit average mode. My Python skills are non-existent but I managed to get descriptive statistics for the data:

count  2000.000000
mean      9.999909
std       0.000002

25%       9.999907
50%       9.999908
75%       9.999910

min       9.999905
max       9.999924 [outlier - my comment]



There is an outlier - a single one with (to me) unknown cause - but all the other 1999 points lie between 9.999905 and 9.999912. That is a span of 7 microvolt or 0.7ppm of 10v.

All values from 9.999905 to 9.999912 are sampled except there are no samples of 9.999908. Should this be possible with SW averaging? Does it point to long time HW "averaging"?

Before I read the manual one more time, I would like to hear if anyone has an opinion on this plot or any other interesting thing about this meter.

Hey - it's a Fluke  [And it was USD 3500+ in 1989. It is amazing how things change ...]

 

[e61_phil]

Hi,

the 34401A has similiar issues when using the 8.5 digits averaged over GPIB. I think the internal calculations run out of precision. The 34401A can only do 10NPLC in hardware everything above is averaging in software. I attached a measurement with 100NPLC setting. One can see there a some missing codes and other codes are more often than they should be. After that I configured the 34401A to use 10NPLC but give 10 samples per trigger. This time I averaged that on my PC (also Python). This time the histogram is much more smoth without gaps.

It would be easier to see the problem if you plot a histogram with 1bin per LSD. It is very easy in Python with the matplotlib.

[AG7CK]

e61_phil

Thanks a lot. I will try the histogram and averaging in Python from 6.5 digit measurements. That will however only show me if all codes are produced on a 10 microvolt level, but it will as you say probably give more information about symmetry. Also, hopefully it should produce the same mean as the 7.5 digit preset mode.

Very easy in Python, you say? OK. Give me a few days.

Thanks again. Good idea and useful information on the 34401A also.

[e61_phil]

Very easy in Python, you say? OK. Give me a few days.

import matplotlib.pyplot as plt
plt.hist(<array of measurements>, <number of bins>, normed=0)
plt.show()

that's it

[3roomlab]

i did it in a long tedious caveman x86 RS232 way
https://www.eevblog.com/forum/beginners/~~-exploring-scpi-thru-rs232-%28problemsolution%29/

[AG7CK]

e61_phil

Thanks for the code. By measuring on a noisier reference, I managed to produce a plot w/o empty bins. I will study the internal and external averaging further when I have the time.


3roomlab

My 8505A has GPIB and no RS232. I use an old National Instruments GPIB-232CT converter in order to avoid VISA  and only use pyserial.

[SilverSolder]

These are good meters - big, heavy, not pretty, but good.  They hold their calibration well.

I think the 'wave' shape in your data is due to room temperature fluctuations during your test.

Also have a look at the 8506A model, which features thermal RMS AC measurement, if you are into that kind of thing.

[JohnPi]

I just bought an 8505A and see the same phenomenon. Here it is compared with an Agilent 34410A measuring a 10 V reference every 10s for ~ 3 days. Temperature is also measured (which causes small deviations in the reference.

Notice the gaps in the 8505A data output codes. It is also much noisier than the 34401A (which isn't spec'ed above 6.5 digits), althugh it is 30 years older than the Agilent.

[dietert1]

The gaps are typical for these instruments (8502, 8505, 8506). They are caused by the ADC circuit that determines 5 bits at a time and saves the remainder for the next cycle. The manual describes how to adjust this. The HP 34401 errors aren't as visible, but we know they are there.

Regards, Dieter

[garrettm]

I'm curious if performing the averaging on the PC would minimize these gaps? These old Fluke meters can supposedly do around 500 samples per second over GPIB with the display off and binary output. (Fastest I got was around 360 samples/sec with a hacked together setup)

The recirculating remainder technique gives 6.5 digit readings at full sampling speed but is inherently more noisy than the multi-slope method. This is why these meters require averaging such large number of samples per reading (2^10 samples IIRC for 7.5 digits). The lack of auto-zero makes them fast, but less stable WRT to temperature changes. This makes the 8505/6 better than the HP 3456A with AZ off, and much, much faster.

Anyways, the gaps are, at least in part, an artifact of how slow these meters are at resolving the LSD. If you use a stable source and incrementally increase the output by 1uV, they will resolve the change without any gaps:

https://www.eevblog.com/forum/testgear/some-old-school-instruments-showing-how-its-done-(hp-3325a-and-fluke-8506a)/msg3448990/#msg3448990

The 8505/6 excel at low bias current input, accuracy at 10 to 19V on the 10V range and the entire 1000 V range: absolutely beats the pants off the HP 3456A in accuracy on the 1000V range and technically could be done at 7.5 digits since the gain is set to 1 like the 10V range.

The Active Filter has some upgrades that can be done to minimize noise and drift as well as improving the composite amplifier on the DC Signal Conditioner. Though I'm not sure if they would be worth the expense to do.