Hi Nick, Thanks for sharing. Good stuff.
I don't have explanations for the things you are seeing. It does make sense to me that running ground + 1-Wire data using the same pair works better than running them using wires from different pairs in the cat5 cable. The pairs are twisted to cancel EMI, although I don't know if that is relevant in this case. I do find it very odd that things work perfectly when the LinkUSB is connected to a laptop but don't work when the LinkUSB is connected to another machine because the 1-Wire network is connected to the LinkUSB, not the computer the LinkUSB is connected to. You could try talking to the LinkUSB directly using minicom (on Linux) or HyperTerminal (Windows). Hit the 'h' key to get help for the basic commands (you can ask the LinkUSB to reset the bus, to find sensors on the bus, and do other things like send commands and read responses. The idea would be to see if the LinkUSB resets or something, but I doubt you'll see anything). When you say "50 to 95% failures reduced to 0 failures"; how are you estimating this, i.e. how are you testing; doing "cat /owfs/28.xxxxxx/temperature"; looking at errors under the statistics directory? Also, what is powering the bus; is it USB or a separate wall wart? Did you try running your tests using the DS18B20 with parasitic power (remember to connect Vdd to ground if you do this)? Please let us know what other things you find since this is good knowledge to have to be able to design reliable networks. Cheers, Eloy Paris.- On 12/15/2011 10:44 AM, Nick wrote: > Please forgive the long post and bear with me on this one. > > My first 1-wire network, snaking around the house with 10 or so sensors and > over a long distance, has been a great success. Almost faultless performance > apart from total data corruption for a short while a couple of times that I > think were from a leaky shower putting water on the outside of part of the > run. I took care over the design and used Clipsal pink cat 5, and it seemed > very robust despite many joins and spurs of several meters in some cases, > many other cables of data and power around, damp walls and perhaps other > impediments. I knew that the networks could be fragile, but it seemed that > in practice they would be pretty reliable based on the experience at home. > > I was wrong, as imagine my surprise when a simple network of about 50 meters > or so in my office building with one 18b20 on the end was almost a total > failure. > > I tested the same setup at home using the same cat 5 and a long run of cable > and had zero errors. Halving the run in the office made no difference, and > while a very short run to a sensor was fine, add on the rest of the run and > it failed again. > > After fiddling a little and getting nowhere, I brought in my laptop, LA and > some other kit to the office to explore more. Curiously, with my Dell laptop > (and same LinkUSB master), the network was now 100% reliable. Plugged into a > different Dell desktop in the office, again a disaster. Default wiring was > orange power, white/orange earth, blue data. Others unconnected but wired in > the RJ45. > > I started experimenting. Here are some of the outcomes that improved things, > and that may go against advice given from Maxim. > > * Attaching just the earth wire from my Saleae LA made the network 100% > reliable (i.e. 50 to 95% failures reduced to 0 failures) > * Using white/blue for earth rather than white/orange, plus a 10K pullup for > the dataline, 95% to 99% reliability. > * Adding a PIC 1-wire slave without or without connecting the data line, > 100% reliable (now attaching LA ground reduced reliability) > * As above but with white/blue for ground gave 100% reliability and no > change if LA ground was attached. > * On different PC, worse problems and the pullup was not so effective. > > I did some other tests, and the final route to near enough 100% reliability > with a single sensor on the end of the run was: > > * Use orange for power > * Blue for data > * Blue/white for earth > * Connect the other earths from the LinkUSB for the length of the run but > not to the slaves > * Add a 10K pullup, though this can be removed with my PIC slaves added. > * It's possible that more slaves will also increase reliability but unknown > at the moment. > > I'm assuming that this boils down to capacitance issues, and that the > changes I made such as choice of what wires to connect for earth etc. simply > affected this, but found it very surprising that the same setup would be > fine with the laptop and yet a failure with other PC's (they are connected > to UPS's in case that's relevant). Effects such as attaching the ground wire > from the LA I found odd too. > > What I took from this is that changing what wires are used for earth in the > cat 5 (or which earth from the master if there's more than one), and how > many earths are used, could be critical and worth doing for anyone with a > similarly misbehaving network. Guidance I'd read said not to connect other > wires to earth, but this was wrong in my case. I would add that in contrast, > the house network uses just orange and white/orange for power, and > green+green/white for data and is fine as I said. > > I would welcome any views about why the experiments made a difference, such > as different PC's, why laptop is fine and not PCs, connecting the LA earth > etc. > > Nick > > ------------------------------------------------------------------------------ 10 Tips for Better Server Consolidation Server virtualization is being driven by many needs. But none more important than the need to reduce IT complexity while improving strategic productivity. Learn More! http://www.accelacomm.com/jaw/sdnl/114/51507609/ _______________________________________________ Owfs-developers mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/owfs-developers
