.. will be updated soon (in construction with a more accurate information)

If the ADC (or the data.acq device) have 100 kohm input impedance, how much accuracy would I lose?

Do I need to buffer this for a better accuracy?

I'm kind of confused to relate the accuracy g and the output voltage here. I would be glad if someone may explain it for a novice.

Setup will be as shown:

..will be updated

There is an interface box for connector input outputs and a dual power supply for the transducers. Number of transducers will be much more but for simplicity I drew two of them. The interface box and the data-acq. board can be very close to each other no prob.

But the transducers (floating wrt the earth) will be around 400 meter away from the box and daq as you see above. A CAT6 twisted shielded cable will be used for each transducer. In the diagram I named the letters for the transducers which corresponds their pins from the datasheet I linked.

The manufacturer states the output impedance as 1 kohm and says load can be 100 kohm. But isn't that a huge problem for accuracy?

I need an accuracy of 0.4 mV.

I'm kind of ignorant about how to interpret the datasheet to the accuracy and impedance of the transducer and ADC. Does this device need calibration or everything is already in datasheet? Is it possible to achieve around 0.4 mV accuracy?

*power ground and signal ground of the transducers (B and F) are internally connected I checked with a multi meter test.

  • \$\begingroup\$ What would your guess be? Something like 1/101? What accuracy are you expecting from the sensor anyway? Any sources of high level interfering signals nearby? \$\endgroup\$ – Brian Drummond Mar 28 '18 at 13:26
  • \$\begingroup\$ Yes I would think that voltage divider way. Is that correct? But the manufacturer recommends 100k load that shocked me since it means huge error. Isnt it? But Im ignorant about these sensors and have a bit hurry for the setup. I edited my answer with all details. The cables will not be on air and they will be indoors on a bridge. But I will use diff ended input daq for common mode issues ect. Please see my edit. \$\endgroup\$ – user1999 Mar 28 '18 at 13:45
  • \$\begingroup\$ 0.4mV on what signal level? 0.4mV on it's own is meaningless... Further, over 400m will be a much bigger issue. You should consider local capture. \$\endgroup\$ – Trevor_G Mar 28 '18 at 13:50
  • \$\begingroup\$ Interest of freq is less than 1Hz. \$\endgroup\$ – user1999 Mar 28 '18 at 13:56
  • 1
    \$\begingroup\$ Basically your problem lacks sufficient specs for Noise ingress, for protection and SNR requirements. There will be other issues. such as optimum shield termination at one end only or both ends from any source of induced EMF. Telephone technology is very robust with CM hybrid transformers , but you are dealling with signals well below this. \$\endgroup\$ – Tony Stewart Sunnyskyguy EE75 Mar 28 '18 at 15:17

You need a resolution of 0.4 mV in order to detect your signal. Accuracy is a different question.

The voltage divider created by the source and load impedances introduces a constant gain error factor that can be calibrated out by applying a correction factor later.

As long as your DAQ has at least 16 bits of resolution, you should be good to go. Since your measurement bandwidth is so low, most sources of noise can be filtered out, and since you're not looking for DC signals either, offsets can be removed as well. All of this (1 Hz low-pass, 0.001 Hz high-pass, gain correction) can be done in the digital domain, after the ADC.

|improve this answer|||||
  • \$\begingroup\$ I think Im confused between the accuracy and resolution here. The ADC I can use is 16 bit +-10V which is enough for that resolution. \$\endgroup\$ – user1999 Mar 28 '18 at 14:30
  • \$\begingroup\$ Can I find that voltage divider factor by applying a known DC through that 400m cable and compare it to what the daq measures? \$\endgroup\$ – user1999 Mar 28 '18 at 14:33
  • 2
    \$\begingroup\$ @newage2000: No, because such a measurement would not include the source impedance of the sensor itself. Furthermore, there are other sources of gain error that you need to account for as well, such as the sensor's own scale factor tolerance of 1%. If you really need to measure the amplitude of the vibration to that level of accuracy (as opposed to merely detecting it), then you need to put the sensor on a calibrated shaker table and work out the required correction factor from that. \$\endgroup\$ – Dave Tweed Mar 28 '18 at 14:52
  • 1
    \$\begingroup\$ @newage2000 as Dave and I have pointed out, you need to do some bench testing. Also, if you decide to add a PCB at the sensor end, going one step further to add a micro and do the detection remotely becomes much more attractive and makes the 400m much less of an issue. \$\endgroup\$ – Trevor_G Mar 28 '18 at 14:54
  • 2
    \$\begingroup\$ Arrgh! Somehow, I'm not making myself clear. It isn't the resistance of the cable that is the issue here -- it is many orders of magnitude less than the source and load impedances. \$\endgroup\$ – Dave Tweed Mar 28 '18 at 15:03

Turning comments into an answer before the comment threads get migrated to chat...

You are confusing accuracy and resolution/sensitivity.

According to the part spec the accuracy over the full range is only 1%, so if you are trying to measure and display 0.022mG of force you are using the wrong sensor. You may be able to calibrate it to closer tolerances, if you also factor in the calibration of the linearity of the device too. However, that is not a trivial matter especially when you factor in temperature variation.

If however you are trying to detect motion down in that range, the device, at least according to the limited information listed, is capable of detecting down to 1uG.

However, you are dealing with very small voltages and as such the signal to noise ratio becomes crucial. Reliably delivering such small analog signals over a quarter mile of cable is no trivial matter.

Your best solution here is to develop small PCBs that will fit on top of the sensor which will receive the analog signal, digitize it locally, then communicate with the central data gathering board via a suitable digital communication protocol. The complete sensor assembly can also be appropriately shielded to further reduce stray noise issues.

The local board can also clean up the power supply before delivering it to the sensor further improving the quality of the sensors output. Indeed over the quoted range it may be prudent to actually deliver a higher DC, or AC, voltage down the line and add regulators to power the sensors.

|improve this answer|||||
  • \$\begingroup\$ The challenges are 1-)For digital signals more than 100meters of CAT is not recommended(?) 2-) I will need 30 of those ADC modules, I dont think I can design them and take that risk for this... But is there a ready ADC module that can convert +-10V at 16 bit resolution? Something like Arduino board but with higher resolution and can accept negative voltages? \$\endgroup\$ – user1999 Mar 28 '18 at 15:56
  • \$\begingroup\$ @newage2000 solving digital transmission >100m is not that hard. Finding an appropriate ADC is beyond the scope of this forum and answer. \$\endgroup\$ – Trevor_G Mar 28 '18 at 16:06
  • \$\begingroup\$ Okay I will stick with analog for now. Just last question. Right now the 400m CAT cable is rolled with 1m diameter, I tested and the noise level is not bad. Im planning it to roll off it outdoors and check out. Maybe rolled cable delude me since it does not represent the real scenario where cable will be straight ? \$\endgroup\$ – user1999 Mar 28 '18 at 16:11
  • \$\begingroup\$ By rolled I mean this: teamluco.co.za/image/cache/data/products/cable/… I onlt tested when the cable is this way in the room \$\endgroup\$ – user1999 Mar 28 '18 at 16:12
  • 1
    \$\begingroup\$ Oh I see so my test doesnt make sense right now. \$\endgroup\$ – user1999 Mar 28 '18 at 16:13

High impedance signals on long cables are likely to carry unintended noise from stray E fields far exceeding your signal.


Buffer the 1k source impedance (1mA ~15V) going into 400m of CAT6 into 100kOhm. The source impedance Z(f) will be near zero but 200 Ohms for RF where there is no feedback gain and no desired signal. I assume the CAT6 cables supply +/-15V DC on 3 wires with 1mA load.

Noise ingress can be caused by high CM noise being converted to DM noise by unbalanced impedance, so that a differential source and receiver is goping to improve this depending on CM/DM impedance ratios * % mismatch * source E field strength and Z(f) of shunt filter.

A magnetic CM SMD choke is adviseable somewhat like those used in telephony to reduce ingress noise and hope to achieve 400uV of noise max.

The purpose was to detect 0.15Hz motion down to 0.022mG.

+/- 7.5V/100mG Full scale. using +/-15V supply.
0.22mG=7.5/100mG = 16.5mVp with Error budget 0.4mV = SNR=41= 32 dB


  • AC line ingress ... are there power lines nearby 3V/m or less?
  • RF AM modulation ingress getting rectified into high impedance ESD protection diodes. Telemetry nearby?

Since the most important informtion is the self-resonant frequencies of the bridge where displacement is double integral of acceleration so the mG level of 0.15Hz has a double amplitude (DA) or 2D @ 1.5Hz which is 10x larger f and 100x larger 2D at 0.15Hz. So the key signal of the bridge vibration is not just A(f) but rather D(f) which is the 2nd integral of A(f). this results in a Displacement amplitude of ....

\$2D=\dfrac{A}{{(\pi f)}^2}\$

My suggestion is to propose a pre-emphasis on the signal just as RIAA did to phonograph recording and FM broadcast does for HF audio in order to improve the spectral SNR.

Except here you want to pre-emphasize the ULF for the desired end result is D, displacement which is the Strain Displacement on the bridge. In otherwords a double integrator with DC restoration.

Re-consider acceleration normally done on bridges is a LPF to block above 20Hz so that phase is preserved at 0.2Hz which is needed 2 decades above the main signal. blocking at 1Hz will result in Group Delay errors computing Displacement at 0.15Hz.

A better way is to use 2 integrators with DC restoring from drift to achieve a BW of 0.002 to 20 Hz ( 4 decades) that is now 40dB better SNR at 0.2Hz after 2nd order integration. Due to group delay in such filters phase error must be considered in the passband by locating breakpoints 2 decades away on either side in order to reduce Phase Error.

Now the telemetry signal with a small PCB is Displacement D(t) instead of G(t) with much higher SNR desired for your application with pre-emphasis at source.

EMC protection is necessary with IEC ingress , but that's another question.

|improve this answer|||||
  • \$\begingroup\$ Interest of freq is less than 1Hz. Still valid what you write? \$\endgroup\$ – user1999 Mar 28 '18 at 13:56
  • \$\begingroup\$ THen it is more an issue of SNR , what is your spec at minimum signal.? \$\endgroup\$ – Tony Stewart Sunnyskyguy EE75 Mar 28 '18 at 13:59
  • \$\begingroup\$ HIs quoted accuracy numbers are nigh impossible. \$\endgroup\$ – Trevor_G Mar 28 '18 at 14:02
  • 1
    \$\begingroup\$ @newage2000 Why should it be painful? Making a bunch of little PCBs that attach to the top of the sensor is not a big deal, especially when they are powered sensors anyway. \$\endgroup\$ – Trevor_G Mar 28 '18 at 14:41
  • 1
    \$\begingroup\$ Yes, @Trevor_G is suggesting doing the ADC there as well, and making the communication with the data gathering point purely digital. This eliminates most of the issues that you're worrying about. \$\endgroup\$ – Dave Tweed Mar 28 '18 at 15:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.