Signal EMI protection

Twisted pairs is only good for a twin signals such as in speaker wires. The (simplified) concept is that the same noise will be on both the signal and return and will cancel itself out. Your control signals do not have a return path

Coax cable is designed for a specific frequency range. In that range it is virtually resistance and noise free. (yet more expensive and bulkier than shielded cable)

Best and cheapest is a shielded cable. It will work like an attenae and absorb any external noise. The shield should be grounded on one end only. (this prevents ground loops)

You shouldn't have too much trouble over 3m.

Are you driving into a load resistor at each end or straight into TTL? some sort of matching at each end will help, anything from 75R to a couple of k, depends on the chips, data rate etc. a bit of C to ground may help get rid of some noise,,,it will slug the leading edges of the logic pulses though. There should be plenty of good stuff out there on transmission lines....

Thanks Del,

We continue to see issues even over runs as short as 25 cm.

Both ends are TTL DIO cards. Currently we have both signals lines running inside their own shields. The shields are only connected to ground at one end. The same ground is being run outside the shields.

I've read up on transmission lines and we are trying out these options. Is there something in the implementation of the theory that we might be missing. So far we tried splitting the ground and twisting each signal, running both pairs with a single shield, both pairs with idividual shields, not twisted with single shield and dual shields. We've tried shaping the waveform with caps to remove the hard corners.

All of these have had effects to one degree or another, but none has be totally successful.

Also, any direction on what experiments to attempt would be greatly appreciated.

What data rate are you transmitting at? Is it some sort of recognised format or some scheme you have invented for your own use?

These DIO cards are they your design? If not do you have a circuit diagram for them.

What resistance do you have between input and ground?

What IC are you driving into?

I run bursts of RS232 format data (can't remember the data rate, but it's fairly slow) from our control system to a remote control typically 4m away down unscreened cable, in noisey environments like commercial laundries without too much trouble...mind I don't know how much data is discarded as 'bad' because it doesn't have the right control words at start and finish of the packet.

(This does illustrate that software can help substantially!)

It's only a crude system using 74HC14 schmitt trigger inverters as the input and output device, terminated by about 2.2k if I recall (from input to ground).

The output has a 220R resistor just to protect against it getting shorted to ground.

There are a few ferrite beads & small Cs scattered about to keep out EMI !

I'm sure someone will jump on this and say I'm doing it all wrong !!!

I'll give more detail when I'm back at work tomorrow if it will help.

PS !

Can you play with slowing down the data rate?

If you can't tell the input impedance, switch everything turned off and stick a DVM across the input hopefully it will measure somewhere between 75 and 600 ohms...(ish!) If it measures much higher than say 3k I'd suggest sticking 600ohms across the input and try that.

This is all a bit working in the dark with only a knife and fork....!

Maybe someone who isn't just pretending to be an electronics designer might join in and help? (I'm just the cleaner here! ...English humour...don't panic)

The data rate is pretty slow. Anywhere from 0 to 200 hz. This system is installed on a train and the first signal is generated when a sample is taken. The second signal is used as a reset to let the slave system know that 2000 samples have been taken. It's essentially the real-time synchronization signal between the systems. The samples are send via a network to the slave computer where they are aligned with data that was collected on the slave.

The boards that we are using are off the shelf CTM05 counter/timer boards with a couple of dio pins. Sorry no drawings. I may have to create a schematic by hand if we can't figure this out. The only problem is that we may need to change the board in the future.

I'd realy like to try and keep the solution to the cabling.

The first signal is generated when a sample is taken ... The second signal is used as a reset to let the slave system know that 2000 samples have been taken.

What are the signals?

If they are just a transition from hi to lo or some such then you will have problems (as any noise could trigger it)...if it is a coded word then this is better, and if there is some handshaking even better still.

My comments about resistors still hold good, they can be soldered across the wires at the connector. Measuring the input resistance is probably a good idea.

via a network

words like 'network' make me panic...!

Good luck!

I think that your biggest problem is that the two 0V rails on the two I/O cards are going up and down relative to each other. On a train this might be really difficult to control. You said that you've run a ground lead outside the cable shielding: what about a thick 0V strap directly between the IO cards. The trouble with this is that you're in danger of creating the ground loops mentioned by techno.

How are the two computers powered? In a lab you could easily see this sort of problem if you powered two computers from different benches: plugging them both into the same distribution point would fix it.

If you can force both 0V rails to the same voltage (DC wise) without creating any other problems, and, you still have a problem with the TTL comms. at 200Hz you could afford to slug the signal substantially with an RC network at the receive end. Put say a 1 K resistor in series and a 1 µF capacitor (multilayer ceramic) to 0V as close to the TTL input as possible.

*************

1 µF capacitor (multilayer ceramic) to 0V

***************

Is 1 µF capacitor not too hi to Ground all the data

Not at 200 Hz!

The RC network I have suggested (1000 Ohms, 1 µF) has an RC constant of 1 mSec.

From long experience of commissioning petrol station forecourt installations, (POS, computers, card readers, pumps, tank gauges, compressors, welding shops, etc) it is essential that every piece of peripheral equipment has a good common ground. If you let the signals float then noise will have a much greater effect - even using a 20mA current loop. Twisted pairs are not that good - we could stiill hear the data crosstalking on the audio output. Try linking the computer cases and signal source with a common earth cable.

My longest run of RS232 was about 150metres but we had a booster amplifier at the server and stores terminal, (we still had communication problems until the storeman learned to type properly!).

If you really want a quiet comms link use fibre-optic cable - we used these on some forecourts out to 30m with no problem and there are RS232 adaptors in most distributor catalogues. Fibre optics are also used for secure comms into a sensitive area and nothing gets in or out of the cable except at the terminations.

I agree with the heavy ground, load resistors, twisted shielded conductors and ferrites.

I'm curious, do you think the noise is from the environment or one of the computers. Routing cables away from noisy equipment may help. O, posibly one of the computer power supplies over loaded. I tried out an ISA DIO card a few years ago, and it drew 4 amps with no output. An overloaded switching type power supply might cause some trouble.

Optical isolation would probably help. There are RS232 isolators out there, if you only have one circuit this would work. If you have parallel circuits, you might get a DIO card with built in isolation.

This next bit advice will probably make most of you cringe, and breaks the golden rule most of us have accepted for decades. Ground both ends of the shield. This has made the biggest reduction in noise I've seen. It works!

I've read a couple of places that say to do this. A couple of companies summarize the following article.

http://www.analog.com/UploadedFiles/Application_Notes/41727248AN_347.pdf

The outcome is analog shield at one end, digitial shield at both.

It takes about a week to make a change and observe the results (this problem is not seen in the lab), so I'll update with my progress.

If your System is not frozen, you can use differential signalling with line transmitters receivers.

Additionally, Shielded cables provide good Isolation thus reduces cross-talk. Ground the shielding at the receiver end.

Depends how stringent requirement is?

More over there is always a compromise between Cost & Target.

You may select

Coaxial is out of topic as it is un-balanced Line for very advanced Data-transmission.

The best is Twisted-Individually-Shielded-Pairs In a bulk Shielded cable.

Keep in mind that:

1. SHIELD is a Copper-wire-Braid whose fill-factor is less than 70% at the best.

2. Twisted-pair is itself is Common-Mode Interference rejecting so aviods cross-talk

The economical is Twisted-pairs-Overall-shielded cable.

The Most IMPORTANT things are:

1. Termination on O/P & I/P ends & matching the Terminated-Impedance.

50/75 Ohms are the best & 100/600 Ohms are tolerable.

Terminating with too-hi is problem creater.

2. Cables of the same ZO be selected.

3. Sig-Shields should only be EARTHED on I/P end to aviod Ground-loop currents & Voltage.

Reply to:

Techno #1

*********** Quote 1 **************

Twisted pairs is only good for a twin signals such as in speaker wires.

... Your control signals do not have a return path

********* Quote 1 Ends ***********

Can't understand how a signal is transmitted without return-path? Is it a wire-less transmission?

Or COMMON-Chassiss is being used for return-path [Shield in this case]?

*********** Quote 2 **************

It will work like an attenae and absorb any external noise.

*********** Quote 2 Ends **************

Antenae Picks while Shield Blocks. Copper-wire braid [shield] only blocks Electrostatics

interference, while twists in a pair of wires cancells both Magnetic & Electrostatic.