I have a situation where I need to connect 10 flat cables (20 way, 0.625mm pitch) to special PCI-E card in my computer. They go from PCI-E card on back to drive bays in front of computer case.

Those flat cables are around 50cm length and come to PCI-E card in one group, from one direction. They look very neat, stacked one to another and taped to stick togeather. I've read that stacking flat cables largely increases EMI, maybe I should consider distancing them somehow?

I have experienced some problems with communication between nodes connected with flat cable and PCI-E card, which was realized to be EMI problem.

First, flat cables didn't have any kind of EMI protection. Then, I installed ferrite cores on both ends of the cables. That helped a bit, but there are still problems.

I have some ideas:

  1. Putting aluminum tape around cables and grounding it on metal case
  2. Separating and twisting flat cables as much as possible (tho that would look ugly and messy so I would like to avoid that)
  3. Distancing them with one from another with aluminum tape and foam tape

Are those good points?
What else can I do to get better EMI protection?

  • \$\begingroup\$ What kind of signals are you carrying? What bandwidth? What interface family (CMOS/LVDS/etc)? And why are you sure it's EMI? \$\endgroup\$ – WhatRoughBeast Aug 25 '15 at 15:51
  • \$\begingroup\$ What is the ratio signal leads vs. ground leads? \$\endgroup\$ – jippie Aug 25 '15 at 17:00
  • \$\begingroup\$ jippie and WhatRoughBeast - I have no idea, your questions are too advanced for my knowledge. I will try to investigate but I can't promise that I can find answers, probably only from manufacturer. \$\endgroup\$ – 10robinho Aug 25 '15 at 18:18

Ideally, put a Faraday cage (a grounded metal box) around everything from the back of the computer to wherever they are going and solidly connect it to the computer earth.

Ribbon cables are not really designed for use in the open.

There are special twisted pair ribbon cables, but that only works if the pinout is right, and it would not generally be helpful if it was not designed in.

As you've found, common mode chokes (the ferrites) help a bit.

The above are just general suggestions for cables carrying logic-level signals. It might be a bit different if the cables were carrying floating signals or had other unusual characteristics.

Differential signalling is good because the return path for the current can be twisted with the other wire so the net field is very low.

  • \$\begingroup\$ Thanks for tips! Maybe I wasn't clear enough, cables never leave computer case, they just go from PCI-E card to drive bays on front. \$\endgroup\$ – 10robinho Aug 25 '15 at 13:48
  • \$\begingroup\$ In that case your signal integrity issues may be related to the type of cable that is being used, so I suggest consulting the manufacturer(s) involved for advice. \$\endgroup\$ – Spehro Pefhany Aug 25 '15 at 13:50
  • \$\begingroup\$ This project is sort of improvisation, distancing nodes from PCI-E card. They shouldn't be distanced by manufacturers idea, but I wanted to bring them to front. Cable they came with is 10 cm, I used much larger, self built cable. \$\endgroup\$ – 10robinho Aug 25 '15 at 13:53
  • 2
    \$\begingroup\$ @10robinho - Sounds like you have also self built yourself some big problems. \$\endgroup\$ – Michael Karas Aug 25 '15 at 14:41
  • \$\begingroup\$ In that case, as @MichaelKaras says, you have a serious problem that may not have a practical solution. \$\endgroup\$ – Spehro Pefhany Aug 25 '15 at 15:28

If you are sure that the problem is crosstalk between the flat cables and not because of an external source, your own suggestion number one would be a good try. Also, if it is a hobby project where you are ignoring specifications of the manufacturer anyway, another more time consuming option would be to just use Ethernet cables; they have 4 twisted pairs each and will eliminate any crosstalk(although I don't know if the impedance of the cable will be a problem in your application).

However, problems because of a bigger length than the manufacturer specified could just as well be due to reflections. Does the system exhibit the same problem with only one flat cable? Then it would rule out crosstalk between the cables.

What tools are you using to measure the signals?

  • \$\begingroup\$ Thanks for your suggestions! I was thinking of Ethernet cables but that would be really huge work to do, and would look really messy. Plus, I'm not sure I would crimp all cables right, I would require hours and hours of work. Also, I would need 25 Ethernet cables, that is A LOT. I've just tired one node only, and it gave problems. That said, looks like crosstalk is not an issue here, but something else, specific to only few flat cables that make problems. What are reflections, why they matter if cable is longer? I don't measure anything, just turn it on and see if it works or doesn't work :) \$\endgroup\$ – 10robinho Aug 25 '15 at 15:37
  • 1
    \$\begingroup\$ Reflections occur when the impedance of the ribbon does not match the impedance of the load. Some of the signal in the ribbon will be reflected back to the source. If the wire is additionally (too) long, the chance is higher that the next signal wave(let's say a square wave representing a bit), will be distorted by the reflection of the current signal wave. I'm not saying this is the reason, though. \$\endgroup\$ – Armannas Aug 25 '15 at 17:17
  • 1
    \$\begingroup\$ On another note, you said that "adding ferrite cores helped a bit", in what way? Did it work occasionally? Can you give more detail? Ferrite cores have different frequency characteristics. You should choose one that fits your situation(frequency). If the PCI-E frequency is 100MHz and the impedance is rated 100 Ohm you should use ferrite cores that are designed for that(see frequency characteristic of katalog.we-online.de/pbs/datasheet/74270093.pdf for example). \$\endgroup\$ – Armannas Aug 25 '15 at 17:33
  • \$\begingroup\$ Thanks for explanation. I added ferrite cores from cables that came with product and put them on cable that I assembled. I have no idea about parameters of this connections. I will contact manufacturer for more information, as I see there is no simple solution here. \$\endgroup\$ – 10robinho Aug 25 '15 at 18:22
  • 2
    \$\begingroup\$ @10robinho have you considered external PCI-E? that might let you move the card closer to the bays and improve the signal enough and or separate the cables to work? \$\endgroup\$ – BenG Aug 25 '15 at 21:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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