Said, good measure! Put it into dollars! This helps! Let me have a think about my budget...
On Sun, Jul 13, 2014 at 9:02 AM, Shane Morris <[email protected]> wrote: > Graham, > > Yeah, I suppose I'm not aspiring to the most precise of the measurements, > I want something that'll give me a reasonable accuracy on a budget. > > Inside the RaspberryPi is a free running 1MHz oscillator - if I could > train it with a 1PPS to a good degree of accuracy (say 10ns to 100ns or so) > I'd be a happy man. My synchronisation signal across the network is 10kHz, > so accuracies would need to reflect say twice that for sampling purposes. > The 10kHz is a "best case" - most equipment will respond slower over > network links, and thus not generate as much traffic. Additionally, a "no > change in sensor reading" generates no message over the network, it'll > intelligently trim that to reduce overhead. > > Jason, > > I like DIY. I'm sold on this idea of hacking my RPi to get it to provide > my cluster with NTP signals, and have the 1PPS and 10MHz pop out of a PCI > slot bracket with BNC connectors in it. I'm even going to buy an 8RU > desktop rack to put on top of my 18RU "baby" rack to put the RPis in 1RU > rackmount cases and the 2RU 12VDC power supply shelf in. I'm going to be > running Plan 9 CPU server on the other RPis, and thats where the NTP data > will terminate. I hope I learn something good from this! > > > On Sun, Jul 13, 2014 at 1:53 AM, Graham Haddock <[email protected]> wrote: > >> Shane: >> >> The question I think that is being asked is ... >> What does "precise" mean to you? >> To the nearest order of magnitude, what kind of accuracy are you looking >> for >> on your three signals. This defines the kind of system you will need. >> >> This group normally aspires to the more accurate end of the scale. >> >> If you are doing simple time logging of some process, then you are >> probably at the other end of the possible accuracy scale, and can >> do things much more simply and cheaply. >> >> So ... >> >> 1 PPS: +/- 1 ns? 10 ns? 100 ns? 1 us? 10 us ? >> NTP: +/- 10 ms? 100 ms? 1 second? >> 10 MHz: +/- 10E-6? 10E-9? 10E-12? 10E-14? >> >> --- Graham >> >> == >> >> >> On Sat, Jul 12, 2014 at 3:57 AM, Shane Morris <[email protected]> >> wrote: >> >> > Hal, >> > >> > As much as I'd like to explain the "big picture" in list, it would make >> God >> > awful noise - if you wish to know any details, I encourage you to >> respond >> > to me off list. Given the fact that the robotics is so totally off >> topic, >> > I'm not willing to discuss them here. Thats only out of respect to the >> > topic of the list. The only real stipulations at this design part of the >> > project is 10MHz out, 1PPS out, and NTP out. Please don't think I'm >> being >> > narqy, I'm really not going to pollute the list with off topic chatter. >> I >> > am more than happy to discuss off list, as and when. >> > >> > David, >> > >> > I was planning to use RaspberryPis in some part of the network, and of >> > course, I must be silly, they have ethernet, and can run Real Time Linux >> > (the LinuxCNC distros that have been made for control of CNC machines). >> By >> > the way, the whole network uses heterogeneous CPU types, I'm pretty >> > agnostic to CPU type, as long as it does the job I need it to. The >> actual >> > ethernet interface won't be as deterministic as we'd like, being >> chained to >> > the USB bus, but if one was not to put any other USB devices on, nor >> attach >> > anything that draws power, the USB performance would be good enough for >> > second accuracy NTP frames. This is without any real analysis of any >> spec >> > sheets, although I have this link: >> > >> > http://www.synclab.org/?tag=raspberry%20pi >> > >> > Thats an interesting read in and of itself. An additional link is: >> > >> > http://www.geekroo.com/products/795 >> > >> > Which is a Mini ITX motherboard for RaspberryPi, which can then go >> nicely >> > into a 1RU case. Add LCDs and other bits and bobs as needed (I saw a >> nice >> > little LCD with an ATMega driver taking TTY strings in the ODROID >> Magazine >> > earlier today - it was meant for an ODROID, but it will work with >> anything >> > that'll output VT100 codes). Once in an 1RU case, it looks neat, and >> would >> > work just as well as a $500 NTP ethernet time source second hand off >> eBay, >> > if not much more configurable and hackable. >> > >> > Many thanks for the thoughts! >> > >> > Shane. >> > >> > >> > On Sat, Jul 12, 2014 at 5:50 PM, Hal Murray <[email protected]> >> > wrote: >> > >> > > >> > > [email protected] said: >> > > > I am needing a GPS source of precise time, in three flavours - 10MHz >> > (or >> > > > so), 1PPS, and ethernet NTP. In the beginning, the NTP will be most >> > > > important, and as time goes on, I'll need the 1PPS signal. >> > > ... >> > > > If a static CW12-TIM ethernet clock could be made, I would be >> willing >> > to >> > > try >> > > > my hand at mounting them to mobile robots, again, for synchronised >> > > timing of >> > > > events. >> > > >> > > I'm missing the big picture. Are the robots the end target? What are >> > you >> > > going to do before that? >> > > >> > > Do the robots have a network connection? (maybe only WiFi to a local >> PC >> > > controlling them) >> > > >> > > How accurately do the robots have to be synchronized? >> > > >> > > >> > > -- >> > > These are my opinions. I hate spam. >> > > >> > > >> > > >> > > _______________________________________________ >> > > time-nuts mailing list -- [email protected] >> > > To unsubscribe, go to >> > > https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts >> > > and follow the instructions there. >> > > >> > _______________________________________________ >> > time-nuts mailing list -- [email protected] >> > To unsubscribe, go to >> > https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts >> > and follow the instructions there. >> > >> _______________________________________________ >> time-nuts mailing list -- [email protected] >> To unsubscribe, go to >> https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts >> and follow the instructions there. >> > > _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
