On Sunday 01 January 2017 05:54:05 Bertho Stultiens wrote: > On 12/31/2016 09:18 PM, Gene Heskett wrote: > [snip] > > >> However, if the reference ground is not equal at both ends of the > >> signal line, then you are in trouble. This is why you would use a > >> balanced line (differential), which is not absolutely referenced to > >> ground, but switches on the differential (and you can clamp the > >> lines). > > > > Do you know of a 4 wire to 8 wire and back interface that can > > function at 32 megabaud and doesn't cost 5 grand+? Neither do I. :) > > Laser diodes and detectors that could handle 2x the video speed > > needed to hit an HDTV transmitter would be required, times 4 to do > > it optically. > > > > There are capacitatively coupled chips I have seen the announcements > > for, intended to steal some of the jobs the MOC chips are doing, but > > no clue as to their useable bandwidth. I'll see what google says. > > > > A paper by Silabs > > <https://www.silabs.com/Support%20Documents/TechnicalDocs/CMOS-Digit > >al-Isolators-WP.pdf> says the are using a nominally 10 megahertz > > carrier in a cmos circuit, about 30x too slow for this. > > > > Searching further, TI has a family of them, 3 channels one way, 1 > > the other, would only need one, at 3.49 in 1k lots, claims 100 > > megahertz bandwidth. I'll see how much power it needs, and what the > > small qty price might be. Not available, so I've ordered samples of > > 2 variations, one of which is inverting but I don't see which is > > which. Its an SOIC package, dunno if I could hack a pcb for that. > > > > Perhaps this might be the better method? > > Actually, I do know how to do it cheaper than $5k. The Silicon Labs > isolator chips are rated up to 150 Mb/s, so that is not a real issue.
Nowhere in that paper I just read does it claim 150 MHz, only that the coupleing clock is 10 MHz. > A LVDS driver and receiver can be had for $1..2. Then you need some > support logic and power management (power isolation). All in all, a > three channel system on CAT3 (CAT5 patch-) cabling would be in the > order of $20..30 in components. Add some overhead and we are at about > $50 (disregarding my hours of designing the circuit). This is a $6 chip in our qty's, and needs 4 ea .1 bypass caps on a pcb. An rj45 could be used, but would run up the cost. Solder pads wouldn't bither me a bot, I'm a CET. Since this would normally be a fixed installation, it seems to me a 6 wire ribbon which could steal power from the devices on each side of it would do nicely on a pcb no bigger that an airmail stamp, including an isolated tab for a firm mechanical mount. Put two solder pads on the power connections and you have used all 8 wires in a 9" piece of cat5. And you would have 5 kv of isolation. My problem is in fabbing a footprint for an SOIC chip on dbl sided pcb on my milling machine. I do have the etching mills, I bought a 10 pack for the last project. And once the code is written, more pcb's are just time & worn bits if you dig into the glass, the target is removing the copper without cutting into the glass. Doable as long as the Z post is well lubricated. I think its worth investigating. > (isolator, see f.ex. http://www.mouser.com/ds/2/368/Si864x-51666.pdf) > > A proper digital balanced isolator is no rocket science and does not > need to be too expensive. It just takes time to design, build and > test. > > > However, the real issue is the distance you can carry the signal due > to propagation delay. If you use bidirectional SPI, then you have > clock, data-out (MOSI) and data-in (MISO). > > At 32MHz clock you have a maximum of 15.6 ns of round-trip delay that > can be tolerated on the data-in with respect to the clock > (half-period). That translates into 1.5 meters(*) of cable (velocity > at ~0.6c) and we ignore all the intermediate electronics' propagation > delays. Adding in a buffer/driver (whatever type) at the cable ends > introduces roughly 5..6 ns delay if you are lucky. That would reduce > the effective length to under one meter. > > Not all is lost; If you send much more data than you receive, then you > can still send at high speed where all signals are propagation > balanced. Then, for reading, you reduce the SPI clock frequency > considarably to take the round-trip propagation delay into account. > > > (*) the signal-delay is measured round-trip which is twice the cable's > length. The SPI master sends a clock, traverses the cable, causes a > MISO change, traverses the cable, read by master. Well, in my case, the cable might be 8" long. The power to the pi and the power to the 7i90, originate from the output terminals of a separate 5v supply. But the pi has a ferrite choke on its power lead that the 7i90 lacks. The 5v supply's ground terminal is taken back to the common point bolt with a heavy braid. Now the shield/ground return from the encoder cable, which is not grounded at any other location, shows, depending on how I've hooked it up, anywhere from 400 mv of noise at the x motor power supply switch rate, to at one point over 5 volts p-p. So with the spindle stopped, the encoders velocity outout shows anywhere up to a 3 digit speed output, either direction. This noise, at a 17 kilohertz switching rate, has less than 5ns rise and fall times. One thing I've not tried yet is to ground the shielding on this cable to the encoder, not to the 7i90, but to that common bolt. But the data error is 2 way, I've had the z motor, running along at 5 ipm, suddenly accelerate to 60+ ipm, and run 6" before it unwinds and resumes its commanded move, with this sudden move being in a random direction. The x motors supply came yesterday, so I have the mounting plate out of the box, and am rigging a filter excised from a computer psu mounted on one corner of it. Its input bypass to ground contact is to this plate, and there is a huge ground lug about 3" away that I will use for the common point. The noise at the 50 pin ground connection is easily 10x that on the - power terminal of the 7i90, which tells be that ground at the i/o connectors gets there by a rather circuitous route, so when I have the supplies up and running, if the noise persists, I will move the shield of that cable to the common bolt. Neither the pi, nor the 7i90 actually has a grounded pad that could be used to give a good, big wide braided common point to both. Frankly, its a noise magnet. I have 2 of corcom's brick wall filters on order, rated at 20 amps. I don't know what the rating is of the filters I have is, each is equipted with a 3 amp 5x20 fuse. Based on the amprobe reading on my milling machine , which is under 3 amps with the spindle off, and perhaps 4 amps with it turning 500 revs and a brakeing belt wrapped around the chuck making it do some work, if it blows the 3 amp, a 5 should hold it without blowing. The toroid wire gauge looks to be able to do 6 or 7 amps without excessive heating. Its a test only, to be replaced by the corcom filters when they arrive. I've left room for one of those. The idea Bertho is to see if there is a bulletproof method that will get rid of the noise. Cheers, Gene Heskett -- "There are four boxes to be used in defense of liberty: soap, ballot, jury, and ammo. Please use in that order." -Ed Howdershelt (Author) Genes Web page <http://geneslinuxbox.net:6309/gene> ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, SlashDot.org! http://sdm.link/slashdot _______________________________________________ Emc-users mailing list Emc-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/emc-users
