How Far Can I2C Go? -- Episode 666

Why 50R cable? If you use typical 75R coax the capacitance per metre would be lower, usually something like 60-70pF/m (go for stranded core if you're going to be flexing the cable a lot - quite a lot of video coax is solid core for fixed installations). Some of the old networking standards that ran on coax used even higher characteristic impedances, with even lower distributed capacitance values (RG62, 93R, 45pF/m, for instance), so if you could find some old lengths, that might work even better.

If you're worried about the resistance per metre - I'm not sure why you need to be? - a larger overall diameter coax will probably have a larger diameter core (to maintain the geometry), so will tend to have lower DC resistance along the core.

The shield has very good current carrying capability (seriously good - in my old Tek portable oscilloscope, they degauss the steel shield that magnetically shields the tube, at start-up, by passing a high current through the shield of the coax that forms the delay line and is wrapped around the tube shield: it goes "thump!" every time you turn it on, which is a bit worrying until you understand the reason for it). There's no reason I can see why you couldn't use the two shields for carrying your power, if you wanted to, though there's the very obvious danger of shorting them. I don't think having an essentially DC potential on the shield affects the screening properties (decouple well at both ends to ensure that, but you'd do that anyway).

Personally, I'd try it with Cat5 cable as well (about 50pF/m) as an experiment - look at the signals at the far end to see if there's much crosstalk beween the clock and the data. With cat5 cable, you could use the spare pairs for power.

Which leads us to: why aren't you looking at the signals with an oscilloscope? That should quickly show you why it succeeds at 10m and fails at 20m.

BTW if you want to find out how robust it is, throw pseudorandom data continuously across the link for 24 hours and see how many errors you get - pressing a button a couple of times isn't much of a test.

thanks for the extensive comment.
So the 50 ohm choice stems from the intended use of I2c on circuit boards and 50Ohms is the standard for impedance controlled traces, so I gave that the best chance of matching with common pullup values etc.. running the power over one of the shields did not work reliably for me(brownout), a big cap on each end might have helped.
Ethernet cable was not reliable for more than a few meters, I suspect the isolation of SDA and SCL to each other made the coax the better choice.
I found my sweet spot at 10 meters 50Ohm coax, with internal pullups and independent PSU.

And on the general why this not that etc. I have very limited time for each video  including sourcing and post production which boils down to usually only one shot to get it done , So i just try until I succeed and wrap it up. Its my intention to get people thinking and trying it themselves and possibly finding better solutions . So I think my goal was achieved by getting you to think hard about the topic and coming up with possible better variants.
So thanks again for chiming in!

thanks for the extensive comment.
So the 50 ohm choice stems from the intended use of I2c on circuit boards and 50Ohms is the standard for impedance controlled traces, so I gave that the best chance of matching with common pullup values etc.. running the power over one of the shields did not work reliably for me(brownout), a big cap on each end might have helped.
Ethernet cable was not reliable for more than a few meters, I suspect the isolation of SDA and SCL to each other made the coax the better choice.
I found my sweet spot at 10 meters 50Ohm coax, with internal pullups and independent PSU.

And on the general why this not that etc. I have very limited time for each video  including sourcing and post production which boils down to usually only one shot to get it done , So i just try until I succeed and wrap it up. Its my intention to get people thinking and trying it themselves and possibly finding better solutions . So I think my goal was achieved by getting you to think hard about the topic and coming up with possible better variants.
So thanks again for chiming in!

Fair enough. I'll stop giving you tasks to do.

And 'well done' on even doing the video, I couldn't manage that in a month of sundays.

My thinking above is all a bit muddled. I was thinking that 100kHz is slow, so the line will look like a lumped capacitor and you needed to minimise the capacitance, but of course the edges are faster and will see 10m as a transmission line, particularly the driven falling edge.

It's messy, because the I2C standard was never intended for this. Philips originally designed it to meander around a circuit board, inside a consumer device like a television, often on a 2-layer board.

Given what you've just said, I think you might be getting it to work not by matching into the cable, but by not matching, so that the end of the cable looks capacitive and slows the edge enough that the reflection that comes back doesn't cause too much disruption. That might explain why more capacitance per metre seems to be giving you a better result. If that is the case, then trying a few hundred pF of capacitance loading the driving end might enable the network cable to work out to the 10m you're getting for the coax.

Another approach might be to accept that the fast falling edge sees the line as a transmission line and use AC termination, so that the termination just exists whilst the signal is changing, but doesn't permanently load things and mess up the signalling levels. For the network cable a 100R resistor in series with however many tens of pF is needed to give a time constant that covers the edge period might do it.

Anyway, a couple of ideas there that anyone reading might experiment with.

But if you're serious about a really cheap, small-scale network, I'd suggest looking at an RS422/RS485 transceiver hanging on a UART port, and sending on a single pair. Writing some simple packet stuff to send and receive under the control of a single master isn't too difficult.

>  I have very limited time for each video  including sourcing and post production which boils down to usually only one shot to get it done 

That seems the faith of every contracted technology video blogger on youtube.

time constraints are part of the game and for me they are a good thing otherwise I would get stuck and procrastinate /never really finish a thing. with the deadline looming I need to find a solution within boundaries. So as long as I can set the limits myself(because I know how long things take me and what hickups to expect, it works great but of course limits project scope.

I learned the hard way to never again let someone set any boundary for any project who is not able to pull it off on their own. If somenone does not have the needed skills / experience they should not make any such Decisions but listen to the ones that do have the skills and let them evaluate.

On this one I set my usual timeframe and it was always out of scope to create a revolutionary bus system, the scope was just to answer asimple questiion with practical application, which it did and got you people to think about it and maybe try your own improved records!