Reducing Voltage Noise in Differential AI's

[Zoil]

Hello,

This past summer, I built a VI to monitor the data acquision of 20 analog UV dectors with 2 Differential Inputs each.  All seemed well as the program would collect data and record it as I had specified. The program is exceedingly complex, but I can boil down its operation to a nutshell.

Physical System:

I am using LabVIEW 7.0 and a NI PCI6225 card.  THe card is connected by 2 68-pin connectors to two CB-68LPR PCB terminal blocks where the analog input channels are routed to.  All wiring is shielded 18AWG and as much care as possible was taken to isolate power carrying wires from signal carring ones.

Software VI:

The system collects 40 differental analog input voltages at a rate of 1 Hz.  Each of these voltages are converted to proper units (UV or Temperature) and loaded into a 1D array.  The UV values are summed and the temeratures averaged over the 1800 second interval, and are printed to a .csv file every half hour.  There are switches to enable each of the twenty channels, although the first two are delegated to our standardized detectors.  The program worked perfectly with simulated voltages in testing.

Problem:

Now, I am attempting to install the system on the company rooftop, and I realized that there is quite a bit of noise in the AI voltages (as compared with the ancient system that our company has been using since 1975 :P)  The AI voltages appear to have a random walk of as much as 25% of the the value that I get from using a Voltmeter on the physical terminal connections on the CB-68LPR.  From this, then the problem must lie either in my VI software or in my PCI6225. 

Does anyone know of some troubleshooting steps to follow to determine the source of this noise in my Differential analog inputs. 

Thanks,

Zoil

-Attached is my operational VI.  Since I used LabVIEW 7.0, if you are going to suggest specific changes to my VI, please do so with Screencaps as I cannot open recent versions of LabVIEW.

[Zoil]

oh yeah, I forgot to include the SubVI's that I created for my programs execution.  They are attached below.

[tbob]

To minimize inductive noise, the cable from the PCI-6225 to the endpoint device should be shielded twisted pair with the shield connected to AI ground on the PCI-6255 end.  Using unshielded wire could cause stray voltages to be induced into the system.  Not using twisted pair could cause more noise induced into one wire than the other, causing errors in differential readings.  What type of wire are you using?  Are you using the SCH-68-68 cable with a SCB-68 terminal block?  If so, make sure the terminal block cover is closed.

[johnsold]

Zoil,

What is the distance between the sensors and the PCI-6225?  Generally it is best to do the signal conditioning close to the sensor.  Can you put some differential amplifiers or even the PCI-6225 on or near the roof?

Lynn

[F._Schubert]

If your readings are 25% off the value they should be, I'd guess you pick up line frequency because of improper grounding. Try to sample with a rate >100Hz (or >120Hz) and you should see the sine.

To remove this effect, just use RSE instead of differential measurement. To test, just connect one of the inputs to AI GND.

Especially if your UV sensors are simple photodiodes, I'm pretty sure that's it.

Felix

[Zoil]

Thanks for the suggestions everybody. 

Replies:

Tbob- Yes, I am using shielded cables running between the Terminal block and PCI6225.  I am not running shilded cables between the detector inputs and terminal blocks (~3ft) seen below from this picture.  Detectors come in up top and run ~30ft in shielded/braded cable to the rooftop.  Consult the picture below for my wall-mounted setup (only two detectors are plugged into DIN Rails up top:

http://bit.ly/gvj3zG (sorry for the poor quality, I don't have time to find the good Digital camera and upload a good picture)

johnsonold- The Detectors are ~30ft above the top of the white box pictured above.  The cables run ~3ft from the DIN mounted Terminal block to the NI CB-68LPR (green PCB seen in the bottom of the photo)

F.Schubert - Unfortunately, I am stuck sampling at 1Hz with the way I wrote my program.  I can possibly try and rewrite it for greater frequencies (10, 20, 60, 120 etc..) But That would require me to redo all of my subvi calculation.  Also, since my code is rather large, I don't think the PC I'm running it on can handle its iterations that quickly. 

Also, this isn't your everyday UV dtector with a photodiode, it has a phosphor and Peltea Junction to maintain 25degC internal Temp, so a differential input is required for my setup, I believe. 

Update:

I spend the afternoon with my production manager and several voltmeters & an oscilloscope only to discover that the noise signal had been present in our old system as well.  Unfortunately, we are not familiar with the machine code, so we assume that it does some sort of signal averaging.  Is there some way to wire into the DAQmx task a signal averaging feature, such that it samples at say 60Hz but only spits out a value at 1Hz?  Also, since I'm wiring 40 tasks together, is there a way I can average all of them by creating some sort of 60x40 2D array and averaging the rows before I go into my code?

Thanks,

Zoil

[F._Schubert]

I'd just have done this test with MAX and not change the code at all. But since you used an oszi, that isn't needed any more.

At first I'd try if it is possible to eliminate noise. If you like us to help you with this, type of sensors (data sheet), frequency of noise and any schematics would be good.

How to do averaging -> good and bad.

But I think it'd be better to use a filter on the analog wire.

Felix

[terickson]

One more thing to look at.  Whatever your sensors are installed into (fixture, etc.) has a good ground.  I recently troubleshooted a problem with noise on some pressure sensors.  The problem turned out to be that they were installed in a stainless steel pipe that did not have a good connection to ground.  The pipe was isolated by some rubber couplers.  As soon as I grounded the pipe, the noise went away.