DMM Noise comparison testing project

[barnacle2k]

I am currently gathering data with my Solartron 7081,  NPLC is not an option on this device for 8.5 digits its fixed to 2560 PLC .

From i read this is the current rule set for this experiment?

• shorted input
• full floating point output of the instrument
• choose maximum resolution available 
• measurement duration 24Hrs per variant
• iterate over all available NPLC and DCV range options

[OldNeurons]

Hi everybody,

I own a 34461A since more than 1 month now. This is my first decent bench DMM. Hope than I will not become a volt nut !!!

I am ready to play that game, and already collected some data, but before posting them I would like to submit some observations (see graph below) I made 2 weeks ago.
Initially I wanted to know the warmup time of this DMM, but, when looking at the large drifts, I was a bit disapointed.
I suspected there was a relation between this drifts and the forced cooling system used on this DMM, and decided to log data with the DMM in normal "configuration", and also with the series of holes placed on both sides blocked whith a piece of paper and adhesive tape (but fan still running).
The results are shown on the graph below and pretty obvious.
Furthermore, I noticed that the temperature reported by the DMM is less (yes LESS), when the air entrance are blocked  (-1.5 to -2 degrees Celsius)!!!
I repeated this test 3 times and there is absolutly no doubt about the results.
So, I am really wondering about this cooling design and it's effects.
If I am right, other DMMs like the 34401A do not have any forced colling system.
The 34460A do not have any fan. Is anybody there could do some data logging?
Same with the 34461A. I am curious to see the results of other users doing the same (with and without blocking the air entrance).

Anyway, you are the experts !!! I am looking for your comments.
It would be interesting to know how is measured the internal temperature, where is placed the sensor, etc ...
I am really curious to understand what is happening there.

In addition, a picture of the shunt that I am using.

Jean

[ManateeMafia]

I think I can modify my code but the older meters may or may not play well with the prologix. I will give it a quick test tonight. The SCPI code should be easier.

[Dr. Frank]

Hi everybody,

I own a 34461A since more than 1 month now. This is my first decent bench DMM. Hope than I will not become a volt nut !!!

Hi Jean,

Why not becoming a volt-nuts?
It's fun.. And.. Sorry, you already lost the game.. You are already a volt-nuts, as you bought that DMM, I think.

Btw.: It's Koningsdag, soon.. I' m curious how the Dutch  will celebrate that!

Frank

[blackdog]

Hi Dr. Frank,

"Koningsdag" most peaple wil be drunk around noon, here in Amsterdam.

I celebrate it likely, with several multimeters 

Kind regarts,
Blackdog

[OldNeurons]

Hi everybody,

I own a 34461A since more than 1 month now. This is my first decent bench DMM. Hope than I will not become a volt nut !!!

Hi Jean,

Why not becoming a volt-nuts?
It's fun.. And.. Sorry, you already lost the game.. You are already a volt-nuts, as you bought that DMM, I think.

Btw.: It's Koningsdag, soon.. I' m curious how the Dutch  will celebrate that!

Frank

Hi Frank,

I am afraid you are right, I am contaminated ...!
An yes, it's fun ... but ... expensive and time consuming !

By the way, I am not Dutch, but French. But I spent time in Netherland for my job, years ago, and was there one time during the Koningsdag. Yes, it's something you don't forget!

Jean

[Dr. Frank]

Hi Blackdog,

I already heard some rumors .. we'll be invited by friends to Ouddorp..
dressed in orange, drinking Genever maybe??
Perhaps, we'll find some electronics on the flea markets..

Unfortunately, the next day will be my own birthday.. 

Frank

[lukaq]

If I am right, other DMMs like the 34401A do not have any forced colling system.

Correct, no fan and no holes on case for cooling

[barnacle2k]

i quickly put lukaq's data through a high pass filter that takes care of any drift quite nicely.
It completely smoothed out the weird spikes in that capture.
The histograms look Ok too.
Its an equiripple 600Tap FIR-HP relative corner frequency 0.02. (Propably a bit too high)

Did that to debug the my data processing, the histogram of my 7081 looks abnormal.

(SW used: Matlab)

[macboy]

...
Averaging by math (either by meter functions or computer-aided) should be avoided, as it's just a pandora box of post-processing and not exactly what this study here for. One can take 0.01 NPLC set of data, and give a median from it with zero noise, but that would not show anything about noise pk-pk.

I completely disagree. Oversampling (averaging multiple samples) is a legitimate way of reducing noise and increasing the effective resolution of an ADC, within the limitations of its linearity.

The Solartron meter does 2560 PLC integration? How strange. Then you see that 2560 is 256 x 10 ... I would wager that internally the meter does 256 measurements at 10 PLC each and averages them to provide a single "2560 PLC" reading. I strongly suspect that the HP/Agilent/Keysight 3458A does something similar; multiple sources from data sheets to user manual specifications to technical articles (for example, the April 1989 HP Technical Journal) unambiguously state that integration time is settable from 500 ns to 1 second. Its 1000 PLC setting then must average multiple readings each with a ~1 second integration time. Logically, it makes more sense to assume that multiple shorter readings are taken and averaged than to think that the meters integrate for 20 seconds (1000 PLC @ 50 Hz) or over 50 seconds (2560 PLC @50 Hz).

It is unlikely for anyone to go to the math menu start averaging for taking a single measurement.

It is not necessary to do that. Set the filter to N samples, Repeat mode (not windowed or moving average). Then you get one reading per N samples. Filter can be enabled or disabled with a single button press.

[TiN]

Was not saying that math is useless, key was to reduce number of variables between different meter / participants, so we can have common base for data sets.
We have just range (5 ranges usually) and most popular NPLC (7 values), and it's already 35 data sets for single meter, single member. If every dataset takes 15 minutes to get, excluding setup/change settings - it would already take a day to get everything. Adding math here (which can be calculated differently in different meters or different member applications) - will overcomplicate everything.

And 1000 nplc should be changed to MAX NPLC, as some meters have it as 200 at max, others - 2560, like Solartron example.

[Mickle T.]

ADC of the Solartron 7081 DMM have a single measurement interval equal to 6.25 ms (so-called 1 "GLUG" or 1 quantum of charge). In 8.5-digits mode the floating logic MPU accumulates (not averages!) 8192 of single conversion results. With a 5.24288 MHz clock frequency 51.2 sec interval gives a maximum of 268435456 full scale (bipolar) readings.

[6thimage]

I suspected there was a relation between this drifts and the forced cooling system used on this DMM, and decided to log data with the DMM in normal "configuration", and also with the series of holes placed on both sides blocked whith a piece of paper and adhesive tape (but fan still running).
The results are shown on the graph below and pretty obvious.
Furthermore, I noticed that the temperature reported by the DMM is less (yes LESS), when the air entrance are blocked  (-1.5 to -2 degrees Celsius)!!!
I repeated this test 3 times and there is absolutly no doubt about the results.
So, I am really wondering about this cooling design and it's effects.
If I am right, other DMMs like the 34401A do not have any forced colling system.
The 34460A do not have any fan. Is anybody there could do some data logging?
Same with the 34461A. I am curious to see the results of other users doing the same (with and without blocking the air entrance).

How stable is the temperature of the environment you are in?

I have been told (by someone at Keysight) that the lesser specifications of the 34460A is due to the lack of a fan, with the readings changing depending on whether the tilting bail is covering the side holes. They also said that the fan, in the 34461/65/70A, moves very little air at room temperature and is largely there for use in equipment racks (although this isn't a recommendation to unplug it).

The 34401A does not have a fan, but some of the Keithley's do (e.g. 2001). So a fan is not necessarily a bad thing, but it requires that the airflow is properly designed.

I am intending to perform these noise measurements on my 34461A and 2000, but I currently don't have a good way of shorting the input jacks - has anyone got any cheap recommendations on something I can pick up here in the UK?

[barnacle2k]

Mickle very nice to see you here, you are the undisputed Solartron expert.   
Could you provide a capture of your highly modified 7081?
(LTZ1000 + OP amps's + modified firmware + resistor TC compensation + and god knows what else.)
I used the binary floating point output mode not the DMM style output for maximum resolution. (uses only 48Bit of the 64Bit Double)
Command: FOrmat=Binary, Compressed

The shape of the histogram of my 7081 suggests there is averaging applied. (like you said they accumulate)
There might even be some numerical issues in their implementation, have you had a look at the floating point routines of the floating MCU Mickle?
Or is the floating point conversion done on the earthy MCU?
Because the noise suppression of this averaging doesn't appear to work as well as it should be.

[OldNeurons]

I suspected there was a relation between this drifts and the forced cooling system used on this DMM, and decided to log data with the DMM in normal "configuration", and also with the series of holes placed on both sides blocked whith a piece of paper and adhesive tape (but fan still running).
The results are shown on the graph below and pretty obvious.
Furthermore, I noticed that the temperature reported by the DMM is less (yes LESS), when the air entrance are blocked  (-1.5 to -2 degrees Celsius)!!!
I repeated this test 3 times and there is absolutly no doubt about the results.
So, I am really wondering about this cooling design and it's effects.
If I am right, other DMMs like the 34401A do not have any forced colling system.
The 34460A do not have any fan. Is anybody there could do some data logging?
Same with the 34461A. I am curious to see the results of other users doing the same (with and without blocking the air entrance).

How stable is the temperature of the environment you are in?

I have been told (by someone at Keysight) that the lesser specifications of the 34460A is due to the lack of a fan, with the readings changing depending on whether the tilting bail is covering the side holes. They also said that the fan, in the 34461/65/70A, moves very little air at room temperature and is largely there for use in equipment racks (although this isn't a recommendation to unplug it).

The 34401A does not have a fan, but some of the Keithley's do (e.g. 2001). So a fan is not necessarily a bad thing, but it requires that the airflow is properly designed.

I am intending to perform these noise measurements on my 34461A and 2000, but I currently don't have a good way of shorting the input jacks - has anyone got any cheap recommendations on something I can pick up here in the UK?

Thanks for your feedback.

My room is facing north, and the temperature is pretty stable. May be 1 to 1.5 °C  max. variation over the day when I ran this 9 hours test. For sure, no quick temperature variation that could explain the quick voltage variations. Except my presence, from time to time very close to the DMM ...

For shorting the input jacks, I use a simple copper cable creased like in my previous post. It costs nothing and it's fully efficient.

Looking for your tests results with your 34461A.

Jean

[Mickle T.]

In the Solartron 7081 all of the floating point routines located in the earthed MPU ROM. In "normal" mode (with DIG FILT=off) there is no any filtering of the ADC results, but only (a*x+b) calculations with raw data.
Unfortunately, I can't capture 7081 now, but I have some old data.

[barnacle2k]

In the Solartron 7081 all of the floating point routines located in the earthed MPU ROM. In "normal" mode (with DIG FILT=off) there is no any filtering of the ADC results, but only (a*x+b) calculations with raw data.
Unfortunately, I can't capture 7081 now, but I have some old data.

Thank you Mickle  .
That's exciting news, it mean i could capture the raw Glug data off the serial isolation interface and do my own processing on it.
(but that is offtopic in this thread, i'll let you know if i find anything interesting)

If you do a new capture try using the floating point mode i mentioned above, i cobbled together a small python script to convert the raw captured rs232 bytes to csv floats.
It gives a lot more resolution then the DMM format, which is very advantageous in this specific experiment.

[radioFlash]

Here's a log for a Keysight 34465A on 10V scale with 10 NPLC.

Sample Count   9296
Aperture Enable:   FALSE
Auto Range:   Off
Auto Zero:   On
Input Impedance:   10 M?
Measurement:   DC Voltage
NPLC:   10
Null State:   FALSE
Null Value(Vdc):   0
Range(Vdc):   10
   
   
Measurement:   DC Voltage
Average Reading   3.75852E-08
Min Reading   -1.42748E-06
Max Reading   1.81982E-06
Max-Min   3.247300E-06
Standard Deviation   4.232794E-07

(uploaded as .txt because the forum doesn't like .csv-ugh!)

[TiN]

It does look way too waivy. Please check if there is some AC hum coupled to meter.

[tszaboo]

So 34410A 10NPLC as you asked:

http://pastebin.com/EfHKZJ9t
Model: ,DMM 34410A
Serial Number:,MY47018945
Address: ,USBInstrument1
Date: ,2015-04-16 11:36:11

Sample Count,8772,
Aperture Enable:,False,
Auto Range:,Off,
Auto Zero:,On,
Input Impedance:,High-Z,
Measurement:,DC Voltage,
NPLC:,10,
Null State:,False,
Null Value(Vdc):,0,
Range(Vdc):,10,

[Jf2014]

Here's a log for my Keithley 2001,B06  /A01 ( not calibrated )

CALIBRATION DATE : 15.04.1994
FRONT INPUT SHORT
30 min warm up
Temp 22.8 C - 23.5 C

// SETTING
:SYST:AZER:TYPE SYNC;              // enable autozero sync
:SYST:LSYN:STAT ON;                 // enable line sync
:SENS:FUNC 'VOLT:DC';               // voltage DCV
:SENS:VOLT:DC:NPLC 10;            // NPLC to 10
:SENS:VOLT:DC:RANGE 2;           // range  2V
:SENS:VOLT:DC:DIG 8.5;             // resolution to 8.5 digits
:SENS:VOLT:DC:AVER:STAT OFF; // Filter off

[OldNeurons]

Here is my log for my 34461A (2 months old).

Ranges: 0.1, 1 and 10Volts
Aperture: 10PLC (will come later with 100PLC)
Auto Zero: ON
Input impedance: HighZ (10 Giga Ohms)

Software used : Tera Term and Excel

Edit: Sorry, forgot to name X axis ... Samples ...

[radioFlash]

It does look way too waivy. Please check if there is some AC hum coupled to meter.

I suspect it's thermal effects from my air conditioner as it cycles on and off and the vent unfortunately points towards the meter. I've run the test a couple of times with essentially the same results. I'm using the Fluke 884X-SHORT as the shorting block.

[barnacle2k]

It does look way too waivy. Please check if there is some AC hum coupled to meter.

Use the suggested highpass filtering on the data.
Had good success with a simple IIR filter. (two biquads)
-3db point 0.006 with >60db attenuation for below 0.001.

i can give you the coefficients if you need them, i don't know with what you process the data.
But a filter could be easily scripted in python/c/java/basic/whatever.

[KedasProbe]

To compare the ADC I assume you will need to divide by the used range, also only compare the same number of PLCs, the distribution graphs would then be in ppm. (not nV)