Noisy signal and shielding techniques

[vt92]

Hello all,

I am having a problem with excessive noise on one of the signals in my test system.

The signal source is a third party device (interface box to a sensor) which produces a 0 - 1.0V signal proportional to the signal output.  The signal is connected to a distribution box and then to my NI DAQ device via an SCB68 terminal block.   I am using differential inputs.

As far as I know all of the shielding is correct.  I know that it would be ideal to tie the chassis ground to the signal ground at the source, but it is not feasible. The chassis of the signal source is not grounded, so there shouldn't be a current loop in the shielding.

For a typical signal, I am getting about 20 mv (pk-pk) of higher frequency (>100 Hz) noise and >200 mV of lower frequency noise ~ 20 Hz. 

I have not been able to isolate the noise source or to reduce the noise levels any further.  The lower frequency noise consists mostly of spikes not sinusoidal noise.

Attached is a diagram of the wiring.  I have the shield tied to a 10 megaohm resistor that is connected at the AI GND terminal.  I know this is difference than what is recommend in the documentation, but I have read other resources that recommend this connection.  If I remove the resistor, the noise doubles.

Thanks for your time.

[johnsold]

vt92,

Ah, fun with grounds and shields!

1. Is there a connection to ground inside the signal distribution box as show or did that line just sneak in to the drawing?

2. Do you have the appropriate bias resistors on the inputs?

3. When you say you remove the 10 megohm resistor, do you leave the shield not connected at the DAQ device or is it solidly connected to AI GND?

What is the sampling rate you are using?  Have you measured the signals at the SCB68 with a voltmeter or oscilloscope? If so, do they differ from what you see through the DAQ device?

I suspect that the problem is due to currents flowing through the shield even though the source is not grounded.

Try opening the connection between the shields and ground at various places, particularly between the signal distribution block and the SCB68.  Keep the shields but just ground them at one end.

Lynn

[vt92]

johnsold,

Thanks for the reply.

1. Is there a connection to ground inside the signal distribution box as show or did that line just sneak in to the drawing?  Yes.  I have connected signal (-) and the shield (drain) wire to the chassis at this point.

2. Do you have the appropriate bias resistors on the inputs?  I don't have any bias resistors.  The SCB-68 manual says that you don't need bias resistors for ground referenced inputs.

3. When you say you remove the 10 megohm resistor, do you leave the shield not connected at the DAQ device or is it solidly connected to AI GND?  No, I just removed the shield connection (it wasn't connected at all on that end).

I am using a 2kHz sampling rate.  I haven't drug the scope out yet.  🙂  

Right now, I am trying to narrow down the issue by making a simple cable between the source and the connector block. 

[johnsold · Accepted solution]

1. If the signal is grounded soemwhere, then you no longer have a differential signal. Although any difference in potential between the ground in the signal distribution box and the SCB68 may show up as though it is part of the signal connected to the - input.  Either do not ground it or do not use the differential input configuration.

2. See 1.

3. OK.  Floating one end of the shield often helps.

I am concerned that the ground connection in the signal distribution box may be part of the problem.  What was the reason for making that connection? Why there rather than at one end or the other, or no ground at all?

What is the signal source? What kind of sensor? How many channels are you measuring?

Lynn

[vt92]

Hello,

Thanks again for your response.

I thought it was best to ground the signal ground.  This white paper from NI seems to indicate this:  http://www.ni.com/white-paper/3344/en  (see figure 10 and Table 1)

I tried making a cable as shown below.  The signal noise was very low < 5 mV.  I can't use this as a solution, but it may help me find the problem.

I made the ground connection in the signal box because it was much easier than at the other locations.  I will try moving this connection.

I can't tell you much more about the signal source, it is proprietary.  There is only one channel that is of concern.

[vt92]

One more thing:

This was my original configuration:   Shield grounded at the SCB68 and no connection to signal (-).

 

[johnsold]

OK.  The connection with the grounded signal source and no shield connection at the other end producing low noise seems to indicate that you do have a ground problem. Some interfering signal is getting into your system via the grounds.

Too bad you cannot ground the source.  That is often the best solution.

Try grounding the signal- at the SCB68. I think you will want the shield grounded there also.  It will be important that the shield only be grounded at one point. You may need to try various shield ground points and breaks in the shield.

How well did the original connection work? Did you have differential inputs configured on the DAQ device and the appropriate bias resistors? 

Lynn

[vt92]

The original configuration was awful.  I used a differential configuration, but no bias resistors.  Technically, the source is ground referenced, not floating. 

I am pursing a way to ground the source by modifying the cabling at the source connection.  I don't really want to modify the signal source box in case it get replaced, changed, etc.

[johnsold]

If the signal source is ground referenced, to what ground is that reference? What is the power source for the signal source and how is it grounded?

Lynn

[vt92]

The distribution box has a 12V supply that was tied to the measurement ground.  I earth grounded the 12V supply and removed the connection between 12V (-) and the measurement (-). 

The noise is now < 5mV.

Thanks for your help.