Author Topic: Handling noise in measurements  (Read 909 times)

0 Members and 1 Guest are viewing this topic.

Offline cncjerry

  • Supporter
  • ****
  • Posts: 488
Handling noise in measurements
« on: January 09, 2017, 08:37:39 AM »
I have a number of meters but the best of which is only a pair of 3457A type.  I also have a pair of 3456a meters which are pretty stable at least compared to those I have but they are only 6.5 digit compared to the 3457A at 7.5 when reading with HIRES.

When I short the leads on the 3457s, after as many as 3K samples, the stdev on either is around 70E-9th.   With everything in the room shut down (lights, TV, etc), I've seen it drop to 40E-9th.  This is with NPLC 100 and the filter off reading and adding the HIRES register.

When I log any of my voltage standards, and they aren't all that great, I see around 20 to 30uV of noise in the plots.  Stdev on these types of tests are usually 6.5E-6.  Watching others with 3458s and high-end voltage standards, I usually note that they are plotting less than 10uV of noise and most times even a lot less on a 10V source.

I hooked both my scope with FFT and my 8568b spectrum analyzer with 85865a preselector and preamp up to the voltmeter while it was reading from a battery operated Geller voltage standard and I note that yes, I see 20 to 30uv of high frequency noise consistent with the local broadcast bands with peaks even higher, up to 70uV at 7Mhz for some reason.  That particular spike has plagued me for years by the way and I can't find the source.

My question is, when people are measuring voltage standards using for instance a 3458a, are they also filtering the input?  Averaging in any way?  It looks like their plots are better by a factor of 10, but if the external noise is the culprit, I would think I'm not the only person with broadcast band noise in the 30uv range, no>?

My voltage standards are generally pretty stable long term.  If I plot a trend line, they are relatively flat since my office is partially built into the side of a sloping hill and the weather is pretty constant in San Francisco.  I might see it go as low as 67 or as high as 69 in my office.

The noise is what is bothering me.  I am using shielded 213 cable with BNC to banana plug adapters.  Even still, I see a lot of noise, again like I stated above, as much as 70uV across a wide band from 100Khz to 103.5Mhz.  I doubt the 3457A is picking up noise much higher than 100khz though it will read HZ to 3M, I believe.

And thinking back to the stdev with the terminals shorted at 70E-9, whatever it is, the noise is coming in on the leads.  I put a shorted 213 cable on the meter and the stdev goes back up to 5E-9 compared to a short right at the meter again at 70E-9.

In the plot below, I measured a Dial-A-Souce KVD type standard over 4000 seconds.  In this case, I had the 3457A filter on at 20th degree.  The trend line in excel is also 20th degree average.  You can see the samples in light blue which are coming from the 3457A after its filter is applied. The more recent samples are at zero in this plot.  After hooking it to the meter, it dropped from around 9.999847V to 9.999835V and then back up to stabilize at 9.999855V.  This unit is new to me since Monday and is reading about 15uV low.  It took about 30 minutes to warm up and then ran for 30 minutes. By the way, from points 191 to 291 I was out of the room and you can see it was pretty flat.

Are others seeing noise like this in their long term plots?
 

Offline CalMachine

  • Regular Contributor
  • *
  • Posts: 90
  • Country: us
    • ENI Labs
Re: Handling noise in measurements
« Reply #1 on: January 09, 2017, 09:57:34 AM »
My question is, when people are measuring voltage standards using for instance a 3458a, are they also filtering the input?  Averaging in any way?  It looks like their plots are better by a factor of 10, but if the external noise is the culprit, I would think I'm not the only person with broadcast band noise in the 30uv range, no>?

When you set NPLC to 100 on a 3458A, it is really setting the NPLC to 10 and then taking 10 measurements and averaging them.
I'm not certain about the 3457, but I would assume they function the same way. 



I am using shielded 213 cable with BNC to banana plug adapters.
 
Those BNC to banana adapters can be pretty bad for precision measurements.  The Pomona ones I've seen are all brass with nickel plating.  You want to use solid core wire preferably bare copper.  If not bare copper, then silver/gold plated copper.  Get some spade/high quality banana connectors and crimp where possible.  Make sure the cable is shielded as well!
« Last Edit: January 09, 2017, 10:04:48 AM by CalMachine »
 

Offline ap

  • Contributor
  • Posts: 45
  • Country: de
Re: Handling noise in measurements
« Reply #2 on: January 10, 2017, 01:40:35 AM »
Actually, what should be used are twisted shielded cables, shielding allone will not do it (respectively, is mediocre only).
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 1248
  • Country: de
Re: Handling noise in measurements
« Reply #3 on: January 10, 2017, 06:57:11 AM »

When I log any of my voltage standards,

 while it was reading from a battery operated Geller voltage standard


Which standards?
Do they have metal housings?
Grounding and guarding concept?
where are the guard pins connected?
do you have a metal desk where all your equipment is located on?

does the reading change when you touch the connection lines, the device housings etc.?

With best regards

Andreas

 

Offline cncjerry

  • Supporter
  • ****
  • Posts: 488
Re: Handling noise in measurements
« Reply #4 on: January 10, 2017, 07:50:49 AM »
Andreas,
The first thing I would like to solve is getting a cable 1 meter long that would give me the same noise level with the distal end shorted as when I short the jacks right at the meter.  I think that would be the first step no matter what I connect.  So I think what I need is a heavy gauge shielded cable, possibly using a twisted pair.  I think I have some shielding braid that I can open and then attach one end to the meter case.  Would it be best to connect the external shield to the guard connection?

I'm not using a metal table but you bring up a good point as I can use one of my heaving metal plate tables instead of a desk.  I'll try to the spade connections, twisted heavy gauge shielded wire and go from there.
 

Offline Conrad Hoffman

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: us
    • The Messy Basement
Re: Handling noise in measurements
« Reply #5 on: January 10, 2017, 08:59:17 AM »
Thermals are a huge issue. Something I've found works well is old twisted pair phone system cable. It's untinned copper and I use it between my Fluke standards and meters with great success. No lugs or anything, just cleaned with Scotchbrite or fine sandpaper. Hopefully you'll get the same results shorting that, or the terminals directly. After the thermals are solved, go after the noise problem with shielding.
 

Offline cncjerry

  • Supporter
  • ****
  • Posts: 488
Re: Handling noise in measurements
« Reply #6 on: January 10, 2017, 07:12:21 PM »
I was able to drop my noise from about 15uV peak to peak to 8uV by using a metal shield around a foam box that I had made to help with temperature.  I then grounded the box to the guard connection.

I had a piece of twisted pair in my hand and then thought the conductors were too small.  I'll give it a try with multiple pairs together.
 

Offline Tony_G

  • Regular Contributor
  • *
  • Posts: 143
  • Country: us
Re: Handling noise in measurements
« Reply #7 on: January 10, 2017, 07:46:51 PM »
Thermals are a huge issue. Something I've found works well is old twisted pair phone system cable. It's untinned copper and I use it between my Fluke standards and meters with great success. No lugs or anything, just cleaned with Scotchbrite or fine sandpaper. Hopefully you'll get the same results shorting that, or the terminals directly. After the thermals are solved, go after the noise problem with shielding.

Wouldn't you get the same benefit with step Ethernet cable? I ask because I just remembered that I have a couple of hundred feet stashed away at home.

TonyG
 

Online VK5RC

  • Supporter
  • ****
  • Posts: 1460
  • Country: au
Re: Handling noise in measurements
« Reply #8 on: January 10, 2017, 08:11:35 PM »
I am interested , why 213? (I presume RG 213)
I found some nasty ground based noise sources from a SMPSU (lead acid battery charger) as well as my LED downlight PSUs. They were also in the order of 20-60uV while attempting to measure 7V DC.
Good hunting!
Whoah! Watch where that landed we might need it later.
 

Offline Conrad Hoffman

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: us
    • The Messy Basement
Re: Handling noise in measurements
« Reply #9 on: January 11, 2017, 03:37:36 AM »
I've never looked at Ethernet cable (analog luddite here) but if it's bare copper it should work. IMO, solid copper, not stranded, is best because you can get it completely clean of oxide on every use. I'm in a rural area, so no strong radio stations, but some things in the lab an house can be an issue. Computers obviously. CFL bulbs. LED bulbs. Any test equipment with multiplexed LED segment displays. Humidifiers with a low setting controlled by a triac. You can get an AM radio and sniff for noise sources. Once you eliminate those and have a shielded twisted pair with untinned copper, one has to consider internal equipment noise sources and stability. I don't mess around with sub-ppm stuff, but even at a couple ppm all those things can matter.
 
The following users thanked this post: Tony_G

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 2187
  • Country: de
Re: Handling noise in measurements
« Reply #10 on: January 11, 2017, 04:03:19 AM »
The HP3457 is know to be relatively noisy - so the older 3456 might actually be better in some cases.

When measuring a considerable voltage, the DMM internal reference is often a major part of the noise. So if you want to test a reference comparable or better than the meter internal one, you need to take the meters own noise into consideration.

One might even consider using two meters in parallel and use the correlation method to separate source noise and meter noise. This way one can measure noise to levels well below the meters own noise - it just takes more time and math.
 

Offline retroware

  • Contributor
  • Posts: 8
  • Country: us
Re: Handling noise in measurements
« Reply #11 on: January 11, 2017, 04:55:43 AM »
The 3457a is particularly noisy in on the 30 volt scale. This is due in part to the fact that they first attenuate it by a factor of 100 and then amplify it by a factor of 10.  The meter is much quieter on the 3v scale were there is no amplification. I'm not sure why HP designed a meter to have such poor performance (noisy, 10Mohm input resistance) on the 30 volt scale.
 

Offline cncjerry

  • Supporter
  • ****
  • Posts: 488
Re: Handling noise in measurements
« Reply #12 on: January 11, 2017, 06:03:06 AM »
The HP3457 is know to be relatively noisy - so the older 3456 might actually be better in some cases.

When measuring a considerable voltage, the DMM internal reference is often a major part of the noise. So if you want to test a reference comparable or better than the meter internal one, you need to take the meters own noise into consideration.

One might even consider using two meters in parallel and use the correlation method to separate source noise and meter noise. This way one can measure noise to levels well below the meters own noise - it just takes more time and math.

I have two x 3456a and 2 x 3457a meters.  What you are suggesting is taking a reading from both and then averaging them or is there an algorithm?  I'm thinking about how I would get them in sync as wouldn't that be important? 

The 3456a is quiet but only 6.5 digits and if I use that meter then it will probably not move at all or very infrequently. 
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 2187
  • Country: de
Re: Handling noise in measurements
« Reply #13 on: January 12, 2017, 04:34:13 AM »
Ideally one would have two meters in sync, and than calculate the cross-correlation function of the two time series you get. With todays PCs this could be done brute force, by the definition - if needed, there are more computational efficient ways that use an FFT. From the Fourier transform of this cross-correlation one gets the power spectrum of the source noise. So one does not get time domain data, but frequency domain.

No seeing much noise on the 3456 might be a problem, as this means quantization is not averaged away.

I also don't understand what was HP's idea behind the voltage ranges of the 3457 - I think the concept is just a bad idea. Maybe they still had 1.18 V ref cells in mind and had difficulties to get 12 / 1.2 V ranges. Having an LCD display (low noise) and the option of an internal scanner where the better points of the 3457.

To look at the noise of a 7 V reference with the 3457, one might have to use an external divider and than use the 3 V range. Still not really convenient.
 

Online guenthert

  • Regular Contributor
  • *
  • Posts: 80
  • Country: us
Re: Handling noise in measurements
« Reply #14 on: January 12, 2017, 06:04:03 AM »
I know of STP, but never got my hands on it;  Fry's has only UTP.  I suppose one could order it online, but I recently got a few meters of low noise (mechanical induced noise is meant here, triboelectric effect I suppose) triax cable.  I'll need to put connectors (4mm plugs I think -- not all my references take spades) on it.  The innermost conductor is tinned solid copper I think, the inner shield (which I would use here as return path) is silver (shiny -- at least shortly after the insulation is removed).  the HP34401 specifies the terminals as 'copper alloy', I couldn't find that specified for the HP3456a (but they are of the same colour).  Not sure, if I'd gain much, if I would use low EMF connectors here, but then, I'm not sure whether EMF or EMI is the bigger problem at my site.  There's a lot of EMI, but it seems to me that both meters happily average that out (e.g. the measured voltage of a 10V reference varies only a few ppm over days, while my oscilloscope finds high frequency noise levels in excess of 100mVpp).
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 2187
  • Country: de
Re: Handling noise in measurements
« Reply #15 on: January 12, 2017, 06:15:58 AM »
When measuring low impedance sources like voltage references, there is no need for triax cables. These are needed with high impedance sources. Even the old mercury ref cells are not that high in impedance that you care about leakage in the pA to low nA range.

To get thermal EMF you need both different materials and thermal gradients. To check how good the materials choice is, one can intentionally increase the temperature gradients. With suitable cables / contacts the sensitivity can be rather low. Surface layers like gold plating or surface oxide has only a rather low effect, as there is essentially no temperature gradient across that thin layer, even if the whole setup has gradients. Usually one can get the thermal EMF smaller than a few µV even without very much care for thermal isolation.
 

Online guenthert

  • Regular Contributor
  • *
  • Posts: 80
  • Country: us
Re: Handling noise in measurements
« Reply #16 on: January 12, 2017, 07:03:38 AM »
When measuring low impedance sources like voltage references, there is no need for triax cables. These are needed with high impedance sources. Even the old mercury ref cells are not that high in impedance that you care about leakage in the pA to low nA range.
I don't care about leak current here.  I realize that I use the triax cable outside its intended application (it's actually leftovers -- originally I ordered it for my KE617).  I don't intend to power the inner shield as guard, but rather use the triax cable as replacement for STP.  The source might be low impedance, but the meter certainly isn't, so as far as I understand I still have an antenna with unshielded cable.

To get thermal EMF you need both different materials and thermal gradients. To check how good the materials choice is, one can intentionally increase the temperature gradients. With suitable cables / contacts the sensitivity can be rather low. Surface layers like gold plating or surface oxide has only a rather low effect, as there is essentially no temperature gradient across that thin layer, even if the whole setup has gradients. Usually one can get the thermal EMF smaller than a few µV even without very much care for thermal isolation.
Yeah, I intended to do such tests and recently found my heat gun again ...
 

Offline Conrad Hoffman

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: us
    • The Messy Basement
Re: Handling noise in measurements
« Reply #17 on: January 12, 2017, 08:19:29 AM »
No, if you connect a low impedance to the meter, the circuit is low impedance and no more susceptible to noise than the lower impedance.
 

Online guenthert

  • Regular Contributor
  • *
  • Posts: 80
  • Country: us
Re: Handling noise in measurements
« Reply #18 on: January 12, 2017, 11:16:09 AM »
Now I readily admit that I don't know the first thing about them radio rays, but if I connect an oscilloscope (1MOhm/13pF input) in parallel (using RG58 coax to 4mm banana) to my 3456a and connect that using two cables to such a low impedance source (e.g. a AD588 "KKMOON" reference) I find some 150mVpp high frequency noise.  If I replace those two cables between the DVM and the reference with another coax cable, the noise goes down to some 90mVpp (if I use a plain 9V block battery the noise is still about 70mVpp).  Hence my curiosity whether a STP (or Triax as "shielded coax") cable would reduce the noise further.

(using two cables with clips to connect to that 9V battery in order to avoid having me hold that thing, the noise is up to some 180mVpp, even though those cables are slightly twisted at one turn every 2" or so)
 

Offline Conrad Hoffman

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: us
    • The Messy Basement
Re: Handling noise in measurements
« Reply #19 on: January 12, 2017, 11:37:22 AM »
Of course it's all relative! We're looking at/for very small signals and drifts, so even a moderately low impedance can pick up noise if the environment has a suitable source of it.  That said, I can't imagine picking up 90-150 mV of noise in a twisted pair to a low impedance device powered by batteries. Do you get the same result with the cable end shorted? How about if you ac couple and connect right across a 9V battery? Something doesn't seem right here, but what's the scope? If you have a very wide bandwidth scope, noise might well be at that level. Low noise work requires limiting the bandwidth, thus my boatanchor Tek plug-in that goes to 10 uV/division also has adjustable filtering down to tens of Hz.
 

Online guenthert

  • Regular Contributor
  • *
  • Posts: 80
  • Country: us
Re: Handling noise in measurements
« Reply #20 on: January 12, 2017, 01:21:45 PM »
[..] That said, I can't imagine picking up 90-150 mV of noise in a twisted pair to a low impedance device powered by batteries. Do you get the same result with the cable end shorted? How about if you ac couple and connect right across a 9V battery?
With the aforementioned slightly twisted cables shorted (those are long clips and now form a 4" by 4" square) it's more than 180mVpp.

Something doesn't seem right here, but what's the scope?
The scope is a Rigol DS1054Z with a bandwidth in excess of 100MHz (my old analog-only CRO "COS5060" finds similar with a bandwidth of 60MHz).  There seems to be some strong components of 25MHz, but I'm not sure, whether that's the original offending signal or just some resonance. 

We're not talking about a remote AM transmitter in a rural setting (someday perhaps), this is heavy duty Silicon Valley electrosmog.  My computer finds easily 6 WiFi networks (a few years ago I lived at a place where I found 70(!) networks), one of them is my own.  Of course their frequency is much too high for me to directly measure.  The largest contributor is my Ethernet-over-Powerline (Homeplug) which I have to disconnect to make sensitive measurements, but haven't today.  I can't control what the neighbor is doing and in summer their a/c is doing terrible things to the power line.  Then there are the omni present cell phones (I can hear in my computer's speaker when my phone is about to receive a call!), wireless phones and a million SMPS.  As much as I'd like to, I can't build a Faraday's cage around "my lab" and power everything from batteries.  I think avoiding noise is a thing of the past and unrealistic these days, other means of dealing with it need to be employed.

If you have a very wide bandwidth scope, noise might well be at that level. Low noise work requires limiting the bandwidth, thus my boatanchor Tek plug-in that goes to 10 uV/division also has adjustable filtering down to tens of Hz.
I do have such a boat anchor too, albeit perhaps not quite as fancy: a Tek AM502 where I can set the upper corner frequency down to 100Hz.  And yes, limiting the bandwidth reduces the noise (which, I believe, is what the op was asking about).
 

Offline Conrad Hoffman

  • Frequent Contributor
  • **
  • Posts: 430
  • Country: us
    • The Messy Basement
Re: Handling noise in measurements
« Reply #21 on: January 12, 2017, 01:44:31 PM »
I don't know if it's legit metrology practice or could cause some other error, but I'd be tempted to make up a multi-pole filter plug, probably including some ferrite, and only with low leakage film caps, to live on the meter or scope input. At least electrical smog doesn't smell bad!
 

Offline Vgkid

  • Super Contributor
  • ***
  • Posts: 1925
  • Country: us
Re: Handling noise in measurements
« Reply #22 on: January 12, 2017, 02:05:55 PM »
Do you have a ups system?
Try running the meter+dut off of one, use shielded cables etc. Though twisted pairs are quite good.
If you own any North Hills Electronics gear, message me.
 

Offline cncjerry

  • Supporter
  • ****
  • Posts: 488
Re: Handling noise in measurements
« Reply #23 on: January 14, 2017, 04:01:51 PM »
I made up a cable from 4 wire twisted with 100% shield.  The stdev as measured on my 3457A dropped from around 6.5E-9 to 5.7e-9, sometimes even lower. Compared to a coax to BNC cable made from RG213 I would say this is progress.  I twisted two pairs on each end.  I was thinking I could make a 4-wire ohm cable out of it as well.

Over the next week I am moving my office to another room.  My workshop has the majority of my equipment but in my office I keep a rack of stuff.  I also have two computers and a known to be noisy TV.  During the move I will be able to shut everything down and switch outlets,etc.  I'm going to try to run some basic tests to see if I can determine how much noise is coming from my A/C mains.  I still think most of the problem is in the air from broadcasts.  For instance, I can hardly use the probe calibration on my TDS784 scope because of all the noise on it from 97.3 and 105.3FM.  I thought the problem was in the calibrator until I hooked my spectrum analyzer up to it and found the FM peaks.  97.3FM, when connecting from my analyzer to my scope calibration pins with a 10:1 Tek probe, is at 2.88mV.  I also have a lot of AM band noise.  I wonder how much of this is getting into my voltage measurements?
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf