Thanks for the additional information. Sounds like you are in an awful noise environment, stepper motors or variable speed AC drives on the NC machines, and welding arcs. All can generate pretty powerful broad spectrum noise.
My order of preference on this kind of problem is:
1. Eliminate the noise.
2. Shield the signal from the noise.
3. Filter the noise.
We can't shut the welder and the NC machines off without adverse economic effects. We'll assume that the equipment is properly installed, grounded and all electrical cabinets closed. So, we can't do much further to reduce the noise at the source.
My next preference would be to shield the signal from the noise. Perhaps you've tried all this, but if not here are some suggestions:
a. If the torque transducer (I'm assume it is strain-bridge based) is far away from its amplifier, try to move the amplifier as close as possible to the transducer. The less distance the signal has to travel unamplifed, the less noise it will pick up.
b. If there is a significant distance between the amplifier output and the A/D board, make sure that the amp has a low output impedence. If the amp is stronger than the coupling of the noise source to the signal cables, it will dominate.
c. Check your ground paths, make sure that there is no way that the signal can share a ground path with anything else.
d. Use shielded cables for the signal, and shielded enclosures for all signal conditioning equipment. Powerful broad-band noise sources can really be difficult to shield, & workmanship becomes a nasty issue. I've had similar applications where it made a big difference how the shields were connected to the enclosure. With the drain wire routed though the enclosure to a single point ground (conventional wisdom) it formed an antenna within the enclosure that re-radiated noise from the drain current to within the cabinet, where it was picked up by signal lines. Grounding the drain wire securely to the enclosure, immediately adjacent to were it entered, tucked flat so as not to form a loop within the enclosure, solved the problem. The SCB 68 might not lend itself easily to this, a NEMA type shielded enclosure around it might help. (If the enclosure has a gasketed door, be sure that the gasket is conductive-- The door won't be effective in shutting out the noise unless it has good electrical contact with the frame all around the periphery.)
Many people claim that the shield should only be grounded at one end, preferably the end nearest the A/D convertor and isolated at the other. I've found this sometimes to be the case, but more often it seems like securely grounding the shield at both ends is better. In rare instances (and I don't understand why) I've had the best luck with shields isolated at Both ends!
e. An often ignored noise path is through the power supplies to the instrument. If you can, run the instruments off a separate circuit, even better off a separate transformer, from the welder and the NC drives. (But don't make a "ground loop" in the process.) I've also found that some kinds of multi-stage (higher priced) surge protectors cut down on the noise coming into the computer and the signal conditioners. If you have separate power supplies for signal conditioning, and have a choice, switching supplies are often better that linear supplies at preventing noise from the power line getting into your signals. Since the switching supply itself generates noise that is probably similar to the noise on the power lines in your shop, its output is built with filters which are very effective at eliminating that kind of noise.
f. Try to avoid having the signal cables travel any significant distance close to potentially noisy power cables. The distance from the signal cable to the power cables should be many times the distance between individual conductors in the cables. If you must have close crossings between these cables, try to make them at about 90 degree angles to minimize coupling.
g. If you can, run the A/D board in Differential mode. This way any common mode noise will get cancelled.
3. Finnally, some suggestions on the filtering issue. My first preference with filtering, and there are reasons why this is a good idea that I won't get into in detail, it to use true analog filtering first, and then, only if needed, add the digital filtering. Digital filtering can't work if the noise spectrum is anywhere near the sample rate of the signal going into the digital filter.
Lets suppose that the output impedance of the transducer amplifier is 50 Ohms, a typical value. If you connect a 5K Ohm resistor directly across the signal terminals on your SCB-68, you will only introduce an error of 1.0% in your signal, and you can calibrate the error out if you need the accuracy. The noise (as big as it may be in voltage) probably isn't coupled very well to the cable to give it much punch. The 5K resistor might swallow up most of the noise. While you are at it, add capacitors across the terminals too. Assuming a 50 Ohm output inpedence from the transducer, a 0.1 uF (high frequency) and 1.0 uF (for the lower stuff) capacitor might absorb a lot of the noise spectrum without getting much of your signal of interest or introducing any significant phase shift. Wire them up neatly with nice short leads between the + and - terminals of your board, or the signal terminal and ground if you're working single ended. If worst comes to worst, they will give you a signal that is clean at the 10 kHz max. sampling rate if you DO end up needing the digital filter.
Offhand, I don't know about the phase shift of the digital filters, if you do end up needing them, we can look into it.
Hope I've been of some help without seeming too didactic about stuff that you might already know as well as I do...
Best Regards, Louis
