3457a DC noise?

[rigrunner]

My 3457a has always been a bit skittish with the last digit on the low ranges and today i noticed it behaves differently to my other meters when shorting the hi lo terminals with a piece of copper.

Set to the 30mV scale, DC, NPLC = 10, Autozero=on and with a 1.5mm piece of copper wire between hi and lo front terminals i'm seeing the displayed reading jumping around between -00.00040mV and 00.00045mV. It's much worse with Autozero=off.

My other meters (one 6.5 and one 5.5 digit)  don't show the same range of readings, they only vary +/- 2 counts on the last digit when shorted.

Is this normal behaviour for a 3457a or is mine getting unwell?

[3roomlab]

if you can do a short log, you can ascertain for sure by comparing to lukier's reading post#43 https://www.eevblog.com/forum/beginners/o-qn-on-funny-nplc-multislopenoise/  (= abt 120nV NPLC10)

it could also be AC mains noise?
mobile RF interference?
computers? etc

or maybe the PCB is dirty. i got better results after a good "wash". but my DMM is old n used.
it could also be a case of capacitors leaking, so most DIY-er would do a "recapping" with new caps

[lukier]

I short all 4 terminals with copper wire from ethernet cable. 3457A has wonderful math capabilities, use them to collect short statistics on 10 and 100 NPLC.

Also, make sure the meter is warmed up. Just after powering up my offset is something like -4uV and after 30 min it decreases to ~1uV. This drift might appear as large std.dev. when doing statistics.

Might be worth to clean the binding posts and apply some deoxit inside and on the thread. Make sure there are no air currents around the binding posts and therefore the shorting wire.
In general at uV levels and below things get really challenging. Thermocouples everywhere, triboelectric effects, RF noise etc.

[rigrunner]

All the meters have been powered on for 5 hours + and the room temp has stabilised at 20 C.

with NPLC=100, NRDGS=10, MATH STAT:

std dev= 73nV
mean = 112nV

I'll give the posts a clean.

I hadn't considered the electrolytics could be failing   

[lukier]

with NPLC=100, NRDGS=10, MATH STAT:

std dev= 73nV
mean = 112nV

That is actually quite impressive std dev is a tiny bit higher, but man, mean is 112nV! Your meter must be fresh out of calibration or something. On mine I have mean slightly above -1uV.

The only mod I did to my meter is to add insulating foam inside, especially around LM399 as it improves stability, see here: https://xdevs.com/fix/hp3456a/#twk_ref.

[3roomlab]

MOD IT ! ops ... off topic ! 

[rigrunner]

That is actually quite impressive std dev is a tiny bit higher, but man, mean is 112nV! Your meter must be fresh out of calibration or something. On mine I have mean slightly above -1uV.

The only mod I did to my meter is to add insulating foam inside, especially around LM399 as it improves stability, see here: https://xdevs.com/fix/hp3456a/#twk_ref.

Not a cal sticker in sight.  The meter came from a surplus seller in the US so i have no idea of its history either.

I've cleaned the posts and removed and reseated the post <> board connectors.

MOD IT ! ops ... off topic ! 

I'll put some foam around the  reference whilst i have it open. Don't tell 3roomlab...
 

[rigrunner]

I've wrapped copper around all four top posts
After cleaning the posts and the internal connections i see:

mean = 344.00nV (344.00001 e-9)
std dev = 66.69nV (66.699995 e-9)

Noise still seems to be there. The last 2 digits on the 30mv range are still jumping about.

[3roomlab]

i think you should try plotting it, and noting the MAX-MIN (as in p-p noise). it could tell if there is any strange drift. or as kleinstein has ID in my case, "popcorn" noise (see pic, eg @ 17500, entire chunk shifted upwards). so instead of purely numbers, a good look at plots could tell more

[cncjerry]

I ran the test on one of my 3457s for the heck of it.  187e-9 and 228E-9 for mean and std respectively.  The statistics were calculated over 993 samples with NPLC = 100 and default NRDGS.  I am going to do another test with a better shorting strap as I've seen better numbers from this meter.


edit:  I ran it again with wrapped copper strapping.  53E-9 and 42E-9 again for mean and std.  I doubt it would get much better than that with all the noise around here.  I've not run a test with the GPIB attached but my experience is that the noise usually goes up.  I have another 3457 that I keep in my lab where the temp is cold (around 60 degrees) but constant.

Have you tried doing the test in a dark or unlit room away from all TVs, computers, etc? I have a bunch of those ferrite clamps on the AC cord, don't know if they make a difference on the meter but the do on my moving coil cartridge preamp.

[rigrunner]

Whilst i was looking around inside the 3457a i spotted that the line filter was a Schaffner made in Switzerland unit 
It's in the bin now 

Everything back together and powered on for 1 hour gives this plot:

Edit:

Just ran math quickly before i shut everything down. 286e-9 mean 50-e9 std dev from 58 samples NPLC=100

[cncjerry]

Rigrunner, how did you get the plot? Sampling it over GPIB?   

This test has me interested now.  I just ran another test with the lights off in the room, fan off in the adjacent bathroom, etc., trying to get all I can control out of the area.  After 244 samples the mean is 6.398E-9 with STD at 40.237E-9.  I don't remember ever seeing the STD go much below 40E-9.  At one point in this test, it reported 544E-15 for the mean.   That's as close to zero as I've ever seen.   It was reporting xxxE-12 for about the first 100 samples but I think pushing the buttons impact the test. I am going to now add more shielding around it to see if I can get the STD lower.  More to come.

[rigrunner]

Rigrunner, how did you get the plot? Sampling it over GPIB?

Yes. I set the options i wanted with the front panel buttons and then issued a read command over gpib and logged to a file.

I've just quickly tried lights and other bench gear powered off. Lights make a little difference but other bench gear off drops the noise considerably.

I have to give up for this evening. Only a few hours left before i need to be awake again 

[cncjerry]

I'll warm-up my other 3457 tomorrow and run a test.  No matter how many times I run this one, the STD comes in around 40E-9 +/- 2 on this one.  That's only slightly better than yours at 66, definitely not something to worry about.  I'm running NPLC 100 and NRDGS 10.  I tried bumping NRDGS up and the STD didn't change.

[3roomlab]

I'll warm-up my other 3457 tomorrow and run a test.  No matter how many times I run this one, the STD comes in around 40E-9 +/- 2 on this one.  That's only slightly better than yours at 66, definitely not something to worry about.  I'm running NPLC 100 and NRDGS 10.  I tried bumping NRDGS up and the STD didn't change.

could you try NPLC 1 just bare, to see its p-p noise and then the NPLC 100 and NRDGS 10? i am curious how much peak peak noise does it have
(actually i dont quite understand NRDGS, what is it? is it like a NPLC multiplier? an like average to stack multiple NPLC?)

[lukier]

I'll warm-up my other 3457 tomorrow and run a test.  No matter how many times I run this one, the STD comes in around 40E-9 +/- 2 on this one.  That's only slightly better than yours at 66, definitely not something to worry about.  I'm running NPLC 100 and NRDGS 10.  I tried bumping NRDGS up and the STD didn't change.

could you try NPLC 1 just bare, to see its p-p noise and then the NPLC 100 and NRDGS 10? i am curious how much peak peak noise does it have
(actually i dont quite understand NRDGS, what is it? is it like a NPLC multiplier? an like average to stack multiple NPLC?)

No, it is just to limit the number of readings taken after a single shot trigger (normally the meter is running in auto-trigger mode). This way if we know NPLC and the number of samples used in the statistics we are comparing apples to apples I think.

[cncjerry]

I ran it in NPLC 1 with NRDGS 10.  Mean was 11.28E-9 and STD 331.786E-9.  Both still really small numbers but much higher than NPLC 100.

That STD number is .000000331786.

I showed my son the mean at one point was 551E-15 which is .000000000000551

I get a kick out of numbers like this in a home lab.  My GPSDO regularly reports phase errors in the picoseconds.  I realize the length of the antenna isn't quite right, etc, but still our ability to measure at home to microvolts,  frequencies above 20Ghz, signals below -150 or even -160dBm is pretty incredible.

[onlooker]

I ran it in NPLC 1 with NRDGS 10.  ... Both still really small numbers but much higher than NPLC 100.

I think the more fair comparison is between  "NPLC 1 with NRDGS 10" and NPLC 10 assuming the NRDGS samples are averaged in the calculation.

(1 cycle) X (10 sample) ~= (10 cycle) X (1 sample) 

or if use NPLC 100, then compare it to "NPLC 1 with NRDGS 100".

[rigrunner]

could you try NPLC 1 just bare, to see its p-p noise and then the NPLC 100 and NRDGS 10? i am curious how much peak peak noise does it have
(actually i dont quite understand NRDGS, what is it? is it like a NPLC multiplier? an like average to stack multiple NPLC?)

For 3roomlab's curiosity

NPLCC = 1 NRDGS = 10
Upper  430.99996E-09
Lower   -1.1299999E-06
Mean    57.450991E-09
SDev   385.30428E-09

NPLCC = 10 NRDGS = 10
Upper  269.99998E-09
Lower -320.00001E-09
Mean    57.999992E-09
SDev   214.51537E-09

NPLCC = 100 NRDGS = 10
Upper  110.00000E-09
Lower -100.00000E-09
Mean    32.999999E-09
SDev    63.953109E-09

[3roomlab]

wow the 1NPLC p-p is so low. compared to my K2015

using 3600samples = 3.1uV (using 1NPLC log on 3rd Dec) 100mV range
a segment of 100 samples, the p-p is 2.3uV
a segment of 10 samples, the p-p is 1.0uV
o gawd! keithley, why so noisy ! 
NRDGS 10 is meaning 10 samples? i want to see if i can extract same number of samples

@NPLC10 (using 1dec log)
4500 samples = 2uV
100samples = 1uV
10samples = 0.43uV

now im curious @ agilent/HP, they seem to pay alot more attention to make less p-p noise. i wonder if it is true also of 34401a

[rigrunner]

NRDGS 10 is meaning 10 samples? i want to see if i can extract same number of samples

Yes, NRDGS 10 = 10 samples.

Does the noise on your K2015 affect the stability of your low microvolt readings?

[onlooker]

I see NRDGS is set for varies measurements listed here, but I am not so sure whether they are rightly or wrongly used in the noise calculation. Just to continue my last post, here is my understanding:

In essence, NRDGS provides an alternative to the "TRIG" command to trigger the readings (sampling). It is a sort of "burst" mode. But, regardless what NRDGS is set to, how a reading is done is always defined by "NPLC". In other words, if each reading is a data point in the statistics, NRDGS setting is irrelevant.

To compare the noise figures between 2 machines, the (total) integration time per data point (not necessarily per reading/sample) for the calculation should be the same. To a good approximation the total integration time per data_point should be

"total_integration_time_per_data_point = NPLC x samples_per_data_point", assuming the value of a data point is taken as the average over samples per data_point.

Then, It will be interest to know, but, I did not test, as to what will be the actual difference on a 3457a between

data_point= "NPLC=10 and average over 10 samples"

and

data_point= "NPLC=100 and 1 sample"

[rigrunner]

In essence, NRDGS provides an alternative to the "TRIG" command to trigger the readings (sampling). It is a sort of "burst" mode.

An alternative to TRIG? I see it as a companion to it. I've been setting TRIG SGL and NRDGS = nn so that when i clear the math values and trigger the readings i'm always taking nn readings before i read the math registers.

[Le_Bassiste]

wow the 1NPLC p-p is so low. compared to my K2015

using 3600samples = 3.1uV (using 1NPLC log on 3rd Dec) 100mV range
a segment of 100 samples, the p-p is 2.3uV
a segment of 10 samples, the p-p is 1.0uV
o gawd! keithley, why so noisy ! 
NRDGS 10 is meaning 10 samples? i want to see if i can extract same number of samples

@NPLC10 (using 1dec log)
4500 samples = 2uV
100samples = 1uV
10samples = 0.43uV

now im curious @ agilent/HP, they seem to pay alot more attention to make less p-p noise. i wonder if it is true also of 34401a

hmmm... so many samples and the noise is going up... could it be thermal "noise" from air movement in the lab?
i tested a short-circuit (covered with styrofoam for thermal isolation) and easily got approx. 0.5µV p-p for the 0.2V range on my KEI2001 across 4096 readings (NPLC=10, AVG=10, moving, 2 sec between readings), but only after leaving the lab and _not_ coming back before measurements had finished.  plotted the data in excel and applied a trendline mx+b over the data. the trendline's slope was just 2*1e-12, i.e., "flat", so, basically, no long term (thermal) drift that could have screwed the p-p noise voltage. however, seems to me that using SDEV on the data expressed as ppm of full scale range would make more sense.

[onlooker]

An alternative to TRIG? I see it as a companion to it.

I did not mean "alternative to" to be a "replacement of".  Maybe, "a secondary TRIG" is more understandable wording.

I've been setting TRIG SGL and NRDGS = nn so that when i clear the math values and trigger the readings i'm always taking nn readings before i read the math registers.

I think you also set STATS some where before all these. Now, I guess I understand what you were doing: You were taking statistics (mean, sdev) for every nn readings. That is, each reading is one data point for statistics calculation. In this case, the statistics is "independent" of nn, if nn is large enough.   Then, nn=10 samples may be too few for stable statistics.