How to measure low resistance reasonably accurately?

One good method is to measure twice, one with positive and one with negative current, and subtract. If the currents have equal magnitude, the heating will be the same and the thermoelectric contribution will cancel out.

I'd tend to use an analogue filter and a chopper diff amp running at some reasonable gain ahead of the ADC. Last time I checked, delta-sigmas are only good to 18-19 bits unless you do a lot of averaging.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
ElectroOptical Innovations LLC 
Optics, Electro-optics, Photonics, Analog Electronics 

160 North State Road #203 
Briarcliff Manor NY 10510 

hobbs at electrooptical dot net 
http://electrooptical.net

5% should be easy. With Seebeck coefficients of most metals in the single-digit uV/K range, thermals should be a minor problem, if at all. Just use the same metal for both Kelvin probes, pay some attention to symmetry and isothermy and you should be fine.

Jeroen Belleman

One trick to reduce thermals is to pulse or reverse the current.

Don't go too fast, because some resistors have serious eddy-current transients, especially big shunts and those metal-case wirewound resistors.

And watch for EMI; that's always a hazard in low level measurement.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Great trick, I didn't know of that. Now it is obvious. Thank you, Phil!

OK. I'm going to apply all the reasonable SW techniques, continuous self-calibration, correlated double sampling, etc. Just don't want to screw the analogue part up. :-)

Best regards, Piotr

If you use AC then thermal voltages will cancel, you've then got the problem of induced voltages, but that might be easier to solve as it will be out of phase with the current.

--
  When I tried casting out nines I made a hash of it.

Great trick, I didn't know of that. Now it is obvious. Thank you, Phil!

OK. I'm going to apply all the reasonable SW techniques, continuous self-calibration, correlated double sampling, etc. Just don't want to screw the analogue part up. :-)

Best regards,

That said, the currents do not have to be constant nor precise, just equal. How much for a full four digits, 100 mR range with 10 uR resolution?

Cheers, Harry

Ratiometric measurement is nice. Add a good resistor in the loop and measure its voltage drop, both ways, as well as the drop across the unknown. That way the current source doesn't have to be very good, doesn't actually need to be a current source at all. Most of the delta-sigma ADCs have multiplexers built in, so use the same ADC to measure both voltages.

We do that to measure RTDs against a good Susumu resistor.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

AD7194 claims something like 22+ bits at 10s of Hz and with several differential inputs it could also to measure current with a reference resistor

is AC needed? couldn't measuring at several different currents and extracting slope and offset do?

You already got a lot of great advice. The basic idea is to stay away from pure DC measurements.

I can only add a few tips - use those new low-noise chopper-stabilized op a mps like ADA4522 or OPA189, with good thin-film or bulk metal foil resistor s. As for Sigma-Delta ADCs - the champions are currently in the AD7xxx fami ly - AD7177-2 or the newer AD7768. To get more than 20 meaningful bits you need an external buried zener reference. Or if you go for ratiometric setup - these ADCs can output their internal Vref. They come with tons of other features too.

You can find some neat ideas also in this paper, or the other ones publishe d by the same guys:

Cheers, Nikolai

Good grief. He was only asking for 5% or so.

Jeroen Belleman

The benefits of AC are great, BUT sometimes it's very useful to read the copper resistance of a motor or transformer winding... Thermocouple errors are in the tens of microvolts if you have everything at room temperature and don't hold the probes too tight (or breathe on the subject).. The use of an LM317 for a 100 mA source, and a multimeter that can resolve 100 uV, gives a resistance error in the millivolt range, That covers a lot of the intended territory without AC. It also is available at the cluttered worktable in my basement...

Well, 5% of sub-milliohms is not exactly trivial. Plus, most of this stuff is so cheap now that you can avoid range switching. If you can avoid precision resistors that's even better, as they are an unpleasant bottleneck...

Cheers, Nikolai

d

id

m pure DC measurements.

amps like ADA4522 or OPA189, with good thin-film or bulk metal foil resist ors. As for Sigma-Delta ADCs - the champions are currently in the AD7xxx fa mily - AD7177-2 or the newer AD7768. To get more than 20 meaningful bits yo u need an external buried zener reference. Or if you go for ratiometric set up - these ADCs can output their internal Vref. They come with tons of othe r features too.

hed by the same guys:

Nice... that's a lot of switches on the front end!

George H. (I can't read the paper, but I can look at the figures.. which mostly tells the story.)

But it's fun. I was going to suggest gain and some good DMM. (at least as a check.) George H.

Oh and polarity switch on current of course. GH

Why not use some low frequency AC ?

With AC feed, do you need a chopper to eliminate any preamp drifts ?

g.

Not exactly. You may need to put in DC-blocking link at some point - a capa citor or a transformer - and you will need some kind of rectifier - ideal ly a synchronous rectifier which could be chopper to get out steady signal.

Reversing DC is easier than AC excitation if you are worried by stray induc tances in the reistance you are measuring.

If you wanted to get precise - and 5% accuracy isn't precise - you could us e a tapped transformer in an AC bridge.

Bryan Kibble (of Kibble balance fame) points out that a carefully wound rat io transformer can divide with a precision of about 10ppm.

Ratio transformers are easy to wind, but a total pair to wire up. I've alwa ys wanted to wind one with round-to-flat ribbon cable where you would use i nsulation displacement connectors at either end of the winding to make it e asy to pick the individual tappings, but have so far never got the excuse.

Ratio transformers are almost always much better than anybody really needs, outside of a national standards laboratory.

--
Bill Sloman, Sydney

As long as the DUT and preamp DC errors are less than about 30 % of ADC FSD range, why not do the DC removal in software ?

If you feed the DUT with say 5 or 10 Hz AC to cancel much of both 50 as well as 60 Hz mains interference, doing a synchronous detection shouldn't be too hard. If an ADC and a processor is involved, run FFT and pick the 5 or 10 Hz bin and calculate power spectrum. Actually a DFT might need less processing power, since it could be reduced to a single bin i.e. an I/Q receiver at 5 or 10 Hz.

Which is just a bipolar square wave. If you are worried about stray inductances, sample in the middle of the pulses, far away from the transition areas.