I have been seeing unacceptable noise on my system. On the +/-1V range, the signal will come in with several hundred mV of noise, which I have been able to reduce to 30mV. Through a series of tests and reconfigurations, I have discovered that the noise depends on how the SCB-68 box is grounded, how the computer is grounded, how the box surrounding the voltage dividers is grounded, etc. Nominally, there are no ground loops; however, creating them can actually reduce the noise! I have been attempting to debug this issue(s) for over a week now. So while my summary below is long, it fails to address everything I have attempted; I tried to elaborate upon only the useful information. Since this has stumped me and several others (who are far more experienced than I) in my lab, I’d really appreciate some suggestions.
My system uses two PCI 6229 boards, with 2 SCB-68 boxes used for each board. This noise phenomenon is on both boards. It seems worse when I have all of my signals running into the boards (i.e. more signals, more signal sources, and more signal paths); however, for testing purposes, I have reduced my system to reading the signal from one power supply (two in some cases). All signals are grounded differential analog inputs. I have jumpered together all unused differential analog inputs. The absolute accuracy spec for the +/-1V range is 0.3mV; inputs that are jumpered together, either with a short wire inside of the SCB-68 box or at the end of the cables running into the box, show a level of noise <1mV.
Currently (for debugging purposes) I am looking at the signal measured off of the back of a power supply (Sorensen DLM 150-4). The power supply outputs are isolated from ground. However, the minus side of the power supply is connected to ground by a connection to a vacuum chamber, which is how it will need to be connected during experimentation. The vacuum chamber is tied to laboratory ground and thus shares the ground with the computer. The signal goes into a voltage divider before being sent to the SCB-68. The resistors in the voltage divider are 905 Ohms and 46.5 kOhms; the voltage divider is on a perfboard surrounded by a metal box -- during the experiment there will be several signals that are divided within this same box. The cables are shielded, twisted pairs. The ground on the cable running from the power supply outputs to the voltage divider inputs is attached only to the voltage divider box; the voltage divider box is connected to the SCB-68 box (nominally) for grounding. The ground on the cable running from the voltage divider outputs to the analog inputs, is attached only to the SCB-68 ground.
I found that the noise on my signal would vary when I tied a 1in braided cable from the chamber to the voltage divider box, thereby creating a ground loop between the chamber and computer grounds; it improved by a factor of 2. I found that the noise varied when I connected the braided cable to the SCB-68 as well.
I have tried measuring a low voltage signal directly off of the power supply outputs, instead of using the voltage dividers. On the 3V signal, there is ~20mV of noise at frequencies which could be the power supply switching frequency (20kHz). I think this is reasonable performance for the power supply; but I cannot say whether this noise is from the power supply or the data acquisition system. At the same time, I have been able to reduce the noise on the signal through the voltage divider to ~30mV (by varying grounds and making ground loops); the dominate frequency is still 20kHz.
Since voltage dividers have been used with various NI boards frequently in my lab without issue, I tried using a voltage divider with larger resistors 116 kOhm and 1.46 MOhm. This new voltage divider was not placed in the same voltage divider box; it was placed in a separate box that has its shield attached to the minus output of the power supply (i.e. grounded nominally). I have also used a second Sorensen DLM 150-4 power supply for some cases. Without going into the details of everything I tried (b/c at this point I feel like I’m playing with magic) during this testing, I noticed several curious things. 1) The noise on the signals coming through the voltage dividers is there whether or not the power supply is on. 2) Attaching the grounding cable from the chamber to the shield on one of the voltage dividers, improves the signal on both that channel and the other channel; in other words, it improves the signal on the channel which it is not connected to. This is where using the second power supply was beneficial because I connected the lines going into the different voltage dividers to the outputs of the different supplies; thus, the only commonality between the signals and their shields is that they go into the same SCB-68 box (and hence the same PCI 6229 board, etc.). 3) Even the noise on the signal that runs direct from the power supply outputs into the SCB-68 varied as the grounds on the voltage divider shields were adjusted, including reducing the noise to <10mV.
One last thing to note is that when I ran the computer off an isolation transformer and attached the metal cable from the chamber to the back of the computer, the signals did not improve.
To run my experiments, I have to use several channels and more complex connections, which display substantially more noise. Though I could make-do with a clunky "fix" that simply minimizes noise through awkward grounding to read the voltage off the one power supply, it won't help measuring the signals that I need to measure.
If you've taken the time to read all this, I hope you have a suggestion
or two to help me figure this issue out!
