I2C over long distance with RJ11 CAT3 cable: how EMI noise will effect differential (but not twisted) pair wires

I have several nodes with Cat3 RJ11 cable in my office. I need to do an I2C communication between several nodes over + 20m and read about the PCA9615 driver. Unfortunately, the Cat3 RJ11 cable has only 4 wires and there is no twisted pair. So do you think I'm going to have a big problem with that? Will EMI noise affect more or less the same the differential (but not twisted) signals?

• how long is a piece of string - or wire? how many errors can you tolerate in your application? noise immunity issues come down to following best practices and what is acceptable for the application. So I think you need to qualify your question quite a bit for anyone to answer. – danmcb Jan 20 at 12:21
• Some two years ago I tried PCA9615 with twisted and untwisted cables, including CAT5 but the results are disappointing. My conclusion is that EMI/noise are usually not the dominant factor. What is important is not to exit the wiring capacitance limit of 400pf. You might like to skim my old posts with some good references: (1) raspberrypi.org/forums/… (2) raspberrypi.org/forums/… /to continue, ... – tlfong01 Jan 20 at 13:10
• – tlfong01 Jan 20 at 13:11
• 20m is really pushing your luck. you'll need multiple tricks to get reliable i2c. (1) do not put SDA and SCL next to each other (2) some (small!) Rseries (3) AC termination (4) rate-limited open drain driver (5) constant current pull-up optimized for cable length (6) reduced bus speed – Pete W Jan 20 at 14:55
• the PCA9615 is made for twisted pair, won't work as well without – Pete W Jan 20 at 15:00

In a twisted pair, any external magnetic field that affects one loop of twist gets canceled out at the next loop of twist. This does not happen on a non-twisted wire pair as it is one single big loop.

• It doesn't happen on coax either (just for the record!). – Andy aka Jan 20 at 12:31
• I think that this is incorrect. In a twisted pair is generally a differential signal, and in this case we want common mode interference to have equal amplitude in each conductor. The twisting is basically just making sure that the two conductors have the same induced current. The twists do not actually cancel, they just keep the two conductors the same average distance from the source. In a coax, the mechanism is shielding, which is a completely different animal. en.wikipedia.org/wiki/Twisted_pair – danmcb Jan 20 at 12:48
• for interests sake, there also exists "star quad" for use with microphones, which takes the whole idea a step further - getting a balanced mic signal to the mixing desk across 30 feet of stage is a demanding app. en.wikipedia.org/wiki/Star_quad_cable – danmcb Jan 20 at 12:52
• @danmcb you need to be clear who you are addressing (me or Justme). If you are addressing my comment then you should put @andyaka in the message. – Andy aka Jan 20 at 13:16
• @andyaka both in fact. – danmcb Jan 20 at 13:17

I need to do an I2C communication between several nodes over + 20m and read about the PCA9615 driver

If you looked at the data sheet for the chip you'll see this diagram: -

The data sheet also says this: -

Construction of the differential transmission line is not device-dependent. PCB traces, open wiring, twisted-pair cables or a combination of these may be used. Twisted-pair cables offer the best performance. A typical twisted-pair transmission line cable has a characteristic impedance of ‘about 100 Ω’ and must be terminated at both ends in 100 Ω to prevent unwanted signal reflections.

And they say this: -

Telecom category 5 (‘CAT 5’) data cable is well suited for this task, but loose wires may also be used, with a reduction in performance.

So, they are not ruling out non-twisted pair when they say this Twisted-pair cables offer the best performance. However, do yourself a favour and cut out whatever error sources might happen.

Will EMI noise affect more or less the same the differential (but not twisted) signals?

It will produce more noise/interference in non-twisted pair cables.

• Thank for your response. "A typical twisted-pair transmission line cable has a characteristic impedance of ‘about 100 Ω’ and must be terminated at both ends in 100 Ω to prevent unwanted signal reflections." If i want to connect more than two nodes, lets say... 10 nodes to that bus, should I put a 100ohm terminator resistor in each node? – BossGandalf Jan 20 at 13:11
• @user273780 No, you only connect the 100 ohm resistors at the distance ends of the cable. – Andy aka Jan 20 at 13:13
• See figure 8 and 9 in the data sheet @user273780 – Andy aka Jan 20 at 13:32
• But in "star configuration" I have several nodes (+20 distance each) connecting in the same point. So if I put 100ohm resistor in that point, where I will put the other terminator resistor? – BossGandalf Jan 20 at 13:39
• Where have you seen this? If it were wired like that you would get problems with data reflections from each slave to the master. The master would also have to drive into a very low impedance as well and that won't work @user273780 – Andy aka Jan 20 at 13:42