3
\$\begingroup\$

I am currently designing a device which internally consists of two circuit boards inside a fairly large, grounded metal enclosure. I'm designing one of the boards myself, the other is a bought part which communicates via UART over a pin header connector (ideal for connecting ribbon cable). I'd like to be able to use the 115200 baud speed, but that's not an absolute requirement.

The bought part is available in two versions - one with 3.3V UART and one with proper RS232 voltage levels. Using the 3.3V version would simplify my own board a bit (no need for a MAX(3)232), but I am wondering whether that might have a negative impact on the robustness of the transmission, especially since the device will have to pass an EMC test.

Since I'm a novice electronics designer, I don't have a good intuition for this yet. Do you think the 3.3V level is sufficiently robust over a ~20cm ribbon cable, provided I don't run it right next to the power supply?

Would you recommend putting a ferrite around both ends of the cable? The supply for the secondary board is over the same cable, so I'd expect the sum of currents over that cable to be 0 for all "good" signals, except for high frequency signals, which can run directly to the chassis ground via a 1nF capacitor.

\$\endgroup\$
8
  • 1
    \$\begingroup\$ without a proper (and expensive) EMC test it's hard to tell, but at such low speeds and especially with 3.3V low voltage, I doubt (I am probably just a level above you in terms of electronics design) you will have EMC issues. The issue you could have is voltage drop, and nearby noise of high enough dB to affect your signals. I think with a ferrite collar at both ends you will be fine - digital signals are pretty robust - they are either one, or off, and you would need a pretty hardcore external source to inject enough noise to disrupt it \$\endgroup\$
    – KyranF
    Apr 10, 2014 at 10:37
  • \$\begingroup\$ also, 20cm really isn't very long... \$\endgroup\$
    – KyranF
    Apr 10, 2014 at 10:39
  • \$\begingroup\$ The proper and expensive EMC test will definitely come. I'm trying to increase the chance that we only have to take that test once, though. And yes, I'm not very concerned about emission here but about noise affecting my signal. Voltage drop over the cable should be negligible even for the supply current (<0.01V). \$\endgroup\$
    – Medo42
    Apr 10, 2014 at 10:47
  • \$\begingroup\$ What is the noisiest device in your little box/between the two boards you are using? Just the power supply? Does it have shielded magnetics/transformers? \$\endgroup\$
    – KyranF
    Apr 10, 2014 at 10:54
  • \$\begingroup\$ I haven't really settled on a particular supply yet. There shouldn't be any other strong source of radiated interference in the system though. There is the microcontroller itself at probably 10MHz communicating with some other ICs via SPI, but that all sits directly on top of a large ground plane. Everything else is analog circuitry which is filtered to very low speeds. \$\endgroup\$
    – Medo42
    Apr 10, 2014 at 11:22

2 Answers 2

7
\$\begingroup\$

EMI is NOT about data rate. It is about EDGE rate. The rise time of your signal edge is the determination of the frequency content of your signal.

I know of a company with very poor board layout for high speed signals. But they were using a 40 kHz clock. However, as chip fab is pushing for smaller and smaller feature size, the rise times have increased. With new versions of the "Same old chip they always used" the circuit was failing. The edge rate was giving them frequencies on the board closer to 100 Mhz.

Look at your edge rate with the chips used. If it is possibly a problem, you can do things to limit the edge rate, such as using series termination resistors. Ferrite beads will help kill common mode currents, but it would be far better to not have them.

Don't route any high speed signals on your board close to and parallel with the serial data lines. This is a common error that allows induced high speed signals to extend onto a good antenna for EMI issues.

Once you get above 100 kHz, return current will exist on the closest conductor available. i.e. It will be on the return path that has the highest capacitive coupling and lowest inductance. Higher frequency current flows in fields. This means that current starting down a trace or wire will immediately capacitively couple to whatever return path it finds and induce an opposite direction current. When this is not properly provided, the capacitive coupling will be done in the air with as far a reach as is needed. This is where massive EMI is created.

This is important in both PCB layout and wiring. When possible, give each signal its own current return path. For example, you could have your serial link cable as:

[GND][TX][GND][RX][GND][3.3V] and with a adhesive shield layer if possible.

Depending on what your volumes are or certification needed, it can make sense to get some equipment to do a quick and dirty check of EMI. The $1,500 Rigol DSA815 with some near field probes can give you a great deal of information about problem areas. Also, a large sheet of metal and a LISN can give you a good idea of conductive EMI. These won't give you certification level results, but you can quickly spot possible issues.

\$\endgroup\$
3
  • \$\begingroup\$ I agree, the very sharp switching edges causes the high frequency harmonics - Snubber circuits around switching components for power supplies etc are great to soften that, but for comms you are kind of stuck with series resistors etc. The impedence needs to be -just- enough to stop the spikes emitting from the edge, but not too much to actually distort/curve the signal too much. \$\endgroup\$
    – KyranF
    Apr 10, 2014 at 11:56
  • \$\begingroup\$ Good general EMI advice, but I was asking specifically about robustness of the transmission. My reasoning is that UART (especially at lower speeds) shouldn't be that badly affected by bad signal integrity, because it should sample in the middle of each bit (as determined by the start bit and known clock rate), so by the time the bit is sampled the signals should be in a steady state. Extraneous edges caused by reflections wouldn't hurt it much. I can't ignore emission of course, but I don't plan to run that cable anywhere near the more sensitive parts of my circuit... \$\endgroup\$
    – Medo42
    Apr 10, 2014 at 14:17
  • \$\begingroup\$ I think for less than the cost of ferrite beads, you most likely can go with a shielded cable setup. I think the most likely issue would be induced noise onto the cable that would affect the logic level detection. As others have mentioned, a differential pair protocol with twisted pair line would reduce this considerably. \$\endgroup\$
    – Joe
    Apr 10, 2014 at 15:46
2
\$\begingroup\$

My advice is, be cautious.

By way of example, I've found that using an FTDI USB-to-3.3V async cable, the link is not reliable (no formal testing, but BER around 10^-5 at 115.2kbps) above 38.4kbps. That's over about 1m of shielded cable at TTL levels in free space, not 20cm of ribbon in close proximity to other electronics - so, the comparison is apples and oranges.

RS232 levels will cause their own noise problems, particularly out of an unshielded ribbon cable at high speed. As a rule, if you want to minimise both emissions and errors, go differential; RS485 (or RS422 if you require full duplex comms) should solve all your problems.

\$\endgroup\$
2
  • \$\begingroup\$ I agree, differential will solve emissions and give really good noise immunity. But he doesn't seem to have that option here \$\endgroup\$
    – KyranF
    Apr 10, 2014 at 11:36
  • \$\begingroup\$ There actually is an RS485 / RS422 version of the part, but unfortunately it does not provide hardware flow control which I really want to have available. \$\endgroup\$
    – Medo42
    Apr 10, 2014 at 14:36

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.