2
\$\begingroup\$

I need to design about 1 meter cable for I2C communication. It contains 4 wires, SDA, SCL, ground and power. I plan to use the shielded cable with the shield grounded at one end only.

It looks reasonable to use the twisted pair cable (that has 2 twisted pairs inside.) Which should be twisted with which? I initially assumed that SDA and SCL, the signal wires, should be twisted together (like in USB.)

However, I found on the web random posts that the signal wires should be twisted one with the ground, another with the power instead. How much of this is correct?

\$\endgroup\$
10
  • \$\begingroup\$ Is it not possible to connect either end of the shielding to the GND of each ends? Or just short the GND line from each end wire to the shielding? \$\endgroup\$
    – Carla H.
    Commented Mar 22, 2021 at 10:22
  • \$\begingroup\$ @CarlaH Why would you suggest the shield should be connected on both ends, what would be gained from this? Usually it is intentionally not connected on both ends, but one end only. \$\endgroup\$
    – Justme
    Commented Mar 22, 2021 at 10:29
  • \$\begingroup\$ @Justme TS already mentioned that the shield is connected to GND on one side. And this GND is connected to the other side on the other end of the cable through a wire. If cable shielding is not shorted to GND on the other end, you'd have a meter long antenna for one side. Place this in an industrial setting and see it pick up noises. I mean, how is it usually not intended to be connected on both ends, but one end only? USB cables and coax cables usually implementated with shielding on either ends connected to GND. \$\endgroup\$
    – Carla H.
    Commented Mar 22, 2021 at 10:42
  • \$\begingroup\$ @CarlaH In industrial settings, using shielded cable for buses like RS485, usually it is recommended to connect the shield from one end only. Also same thing applies with USB - the cable shield does connect to connector metal chassis on both ends of the cable - however on the device end the connector shell is not connected to ground but on host end it is, so again grounded on one end only. This is done to make sure no operating currents flow in the shield so it does not become an antenna. But this is offtopic already, the OP did not ask about how to connect the shield. \$\endgroup\$
    – Justme
    Commented Mar 22, 2021 at 10:54
  • \$\begingroup\$ Chassis connection is a different issue though, it is usually an electrical hazard to connect one device/equipment chassis to another's. Would agree that this is offtopic :) \$\endgroup\$
    – Carla H.
    Commented Mar 22, 2021 at 11:04

2 Answers 2

11
\$\begingroup\$

When you twist a pair of wires, it should only ever be signal with its ground, or signal with its complementary signal. If it's well decoupled, twisting a signal with power also works.

Twisting two wires together that carry different signals is asking for enhanced crosstalk, and trouble.

\$\endgroup\$
3
  • 3
    \$\begingroup\$ For I2C in particular, where SDA is sampled on the falling edge of SCL, and the bus is open drain, there is the very real possibility that a 1 can become a 0 because of the coupling. \$\endgroup\$ Commented Mar 22, 2021 at 12:37
  • \$\begingroup\$ How about putting the signal in both wires of the twisted pair? Is that better than signal/GND (which probably means more capacitance) or signal/nothing? \$\endgroup\$
    – NateS
    Commented Apr 10, 2023 at 3:14
  • \$\begingroup\$ Using both cores of a twisted pair to carry the same signal is OK as far as avoiding the crosstalk you would get from putting different signals on the pair. However, you lose the benefit of twisting it with a ground or power, and you use wires inefficiently. If you have two twisted pairs, and two signals, twist each signal with ground, or twist one with ground and the other with power. \$\endgroup\$
    – Neil_UK
    Commented Apr 10, 2023 at 7:31
7
\$\begingroup\$

Quote from I2C specification UM10204:

If the bus lines are twisted-pair, each bus line must be twisted with a VSS return.

Alternatively [...] the SDA line [can be] twisted with a VDD return.

So in short, twisting SDA and SCL is the worst thing you can do.

Twist SCL with GND and SDA with VDD, use decoupling caps between VDD and GND on both ends of the cable, and the cable must also have low capacitive coupling from shield to SDA and from shield to GND.

\$\endgroup\$
3
  • \$\begingroup\$ Could you please provide a little more information in your citation? Where could someone find "UM10204" for example? \$\endgroup\$ Commented Mar 22, 2021 at 13:33
  • \$\begingroup\$ @ElliotAlderson Good point. How should I reference a PDF that is a specific version of a document called the "I2C specification", designated as UM10204 by the manufacturer (NXP) and is extremely easy to find? I can add a link to it, but it's most likely a dynamic and won't work for the next person who clicks the link. \$\endgroup\$
    – Justme
    Commented Mar 22, 2021 at 14:19
  • \$\begingroup\$ At least you could mention that it is an NXP document. I think anytime you quote a document you should tell us who published the document. You and I know that it is NXP (née Philips) but the casual reader might not. You could also add the actual title of the document. \$\endgroup\$ Commented Mar 22, 2021 at 14:46

Your Answer

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

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