Dear community,
I'm currently working on a setup for testing the accuracy of an impedance sensor. The sensor should be able to measure impedances in the range of 5kOhm-1MOhm. My testing setup consists of resistors and analog switches which are able to change the impedance between the two measurement pins from 5kOhm to 1MOhm with an increment of 5kOhm. The analog switches are controlled by an Arduino Leonardo.
Using a resistor with a fixed value the sensor measures it accurately. The testing setup controlled by the arduino works as well. I can set the resistance using the arduino and I'm able to measure the same value using a multimeter.
But super strange results appear using the sensor to measure the resistance of my testing setup. The only difference between measuring with the multimeter instead of the sensor is that the multimeter uses DC voltage instead of AC voltage. This seems to be a big problem for whatever reason and I have absolutely no idea why this is the case.
I attached the schematic of the circuit and the mesuring results (plot of the relative error between the impedance measured by the sensor and the impedance set by the arduino).
I'm very thankful for all kinds of help. Thank you!

Best regards
Jannis

 

The first obvious question is how does the sensor measure impedance? Can you show the circuitry for the sensor or state its operational characteristics, eg voltage/current, frequency, etc.?

 

Consider that "impedance" includes both resistance and reactance, either capacitive or inductive. So measuring the impedance of anything other than a resistor will depend on the frequency of the applied voltage. And measuring with DC will give different results.
And while the circuit drawing is good, it is not simple to understand how the measurement is actually being done. So at least an explanation of the measurement process will help towards getting an answer.

 

Consider that "impedance" includes both resistance and reactance, either capacitive or inductive. So measuring the impedance of anything other than a resistor will depend on the frequency of the applied voltage. And measuring with DC will give different results.

True MrB2, but if the frequency is fairly low, and therefore strays, parasitics and other transmission line effects are minimal, and the phase around 0deg, I'd expect the DC resistance and the AC impedance to be fairly close, at least at the lower end of the scale.

It may be that thats what the 'relative error' graph shows as its increasing 'impedance' from left to right, but I don't know how to fully interpret that chart. Why does the error drop to zero at 200k and 300k? Is that where we switch in another mux? Is the error an artefact of the mux itself then - maybe voltage related. The circuit is difficult to read, but its obviously the electronic equivalent of a 'decade' box (except maybe not decades as such).

 

We still have no mention as to the method used to measure impedance. And I have not studied the circuit to discover what it is. For measuring resistance, we can use constant current and measure the voltage, or a constant voltage and measure the current, or a bridge and adjust for a null. Those are the three common ways. There is also putting the resistance in an oscillator and measuring the frequency, but that is more complex.
So I am waiting for the TS to explain the measurement scheme used i this system. I hope that they realize that FET switch devices have a fair spread of effective resistances, depending on quite a few things. They are not at all like relay contacts, no matter what the books show.

 

I hope that they realize that FET switch devices have a fair spread of effective resistances, depending on quite a few things. They are not at all like relay contacts, no matter what the books show.

The MAX4737 devices are designed for USB switching with a 3dB bandwidth of 300MHz, 3ohm insertion resistance and 2nS cross-channel skew. An immediate obvious problem is they are single supply with a dynamic range from 0 to +Vcc. AC transmission is therefore not an option unless they are biassed to 2.5v which may be impossible to achieve and meet the design brief.


edit: one solution would be to run the switches off +/-2.5v and the logic off +5v with simple NPN/PNP level converters at the inputs to the switches.

 

The MAX4737 devices are designed for USB switching with a 3dB bandwidth of 300MHz, 3ohm insertion resistance and 2nS cross-channel skew. An immediate obvious problem is they are single supply with a dynamic range from 0 to +Vcc. AC transmission is therefore not an option unless they are biassed to 2.5v which may be impossible to achieve and meet the design brief.

The digital world and digital ICs make no promise to function nicely as analog devices. performance that is excellent in digital operation may not always be satisfactory when handling analog signals.

 

The digital world and digital ICs make no promise to function nicely as analog devices. performance that is excellent in digital operation may not always be satisfactory when handling analog signals.

Too true, however these look ok spec wise; their small signal performance seems OK, its just the offset that appears problematic. There are other pin-compatible quad SPST switches in the range from Maxim that may be a better fit, but a quick scan suggests not.

 

And we still have no mention of what the scheme for measuring is. I mentioned 4 different approaches in post #5, all quite different from each other. It may be that this is a bridge balancing circuit, which is probably the least effective and demands the greatest number of precise parts. It also hhas the lowest resolution of any scheme.

 

Another way around the resistance of the switch itself.



Here the resistance seen between Ref. and "To Device Under Test Terminal" see either 0Ω or 30.1Ω depending upon the component selected to be in the feedback path. The op amp limits the performance of the circuit because of offset and bandwidth to start, but gives the performance a really nice boost, and yes you can use quite a few CMOS transmission gates.

As a friend used to say "If it doesn't work, fix it with feedback."

 

I must be a bit slow today as it is still not clear how the impedance value is determined. So this seems to be another measurement scheme that is different from the four methods that are mentioned. And now, after reading again, I realize that the system is for checking the accuracy of that sensor, not for checking actual impeadance. So I see a scheme to use analog switches to select different resistors as standards. The intrinsic challenge is that the exact resistance of most analog switches depends on the current through them, the supply voltage, and the temperature.
So probably using a mercury wetted reed relay will provide a more repeatable reading. And, to assure that all is correct, a means to verify the voltage across the resistor could be added. This will be valid because resistance measurements depend on passing a current through the resistance being measured. The current may be quite small, but there is a current, and thus a voltage developed that can be measured.

 

@MisterBill2 @DickCappels These switches are max 3.5ohm at +/- 100mA across -40 to +85degC, so < 0.1% of the 5k minimum test impedance required. The problems the TS is seeing are almost certainly the AC bias problem distorting the measurement.

 

Unfortunately I don’t know the working principle of the sensor in detail because this testing circuit above is my task for an internship I’m currently doing. All I know is: There are two electrodes measuring the impedance between them. One electrode has a constant voltage of about 1.5 V the other one has an AC voltage between about 0V and 3V with a frequency in the kHz regime. A uC measures the resulting current and uses a fourier transformation to determine the real and the imaginary part of the complex impedance although only the real part is relevant. In my opinion my circuit behaves much different under the AC conditions than under DC. The plot in my first post showed the applied impedance between the two measuring electrodes using my testing circuit vs. the relative error between the applied impedance and the measured one. These errors are exploding for higher impedances an show some kind of periodicity which is somehow depending on the number of resistors used. The on resistance of the Switches are much too low to make such errors.

 

@jannis_98 Can you put a 'scope on both ends and show us the waveforms:
a) when the value is 'correct'
b) when its 'wrong'

If possible, use another channel to show intermediate points on resistor chain....

On the face of it, if one end is at 1.5v and the other varies 0 - 3v then this should work...

Added thought - do you have a signal generator in the lab?