From 0bf54d8fb8dedf635982b235a4db5de403673cf2 Mon Sep 17 00:00:00 2001 From: Karl Palsson <karlp@tweak.net.au> Date: Mon, 19 Feb 2018 21:07:12 +0000 Subject: hw1: notes on signal integrity --- hw1/signal.integrity.wisdom | 41 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 41 insertions(+) create mode 100644 hw1/signal.integrity.wisdom diff --git a/hw1/signal.integrity.wisdom b/hw1/signal.integrity.wisdom new file mode 100644 index 0000000..47dff01 --- /dev/null +++ b/hw1/signal.integrity.wisdom @@ -0,0 +1,41 @@ +22:59 < karlp> BrainDamage: yeah, proper questions from you of course. building + test harness host board to plug all the nucleo64 boards into + for locm3 testing. Adding a port for plugging a cheap fx2 logic + analyser on, so I can do auto testing with sigrok, +23:00 < karlp> I'm estimating like 20-30mm on the side tap to the logic + analyser port, plus 30-40mm or so between target/host. +23:00 < karlp> how fast can I go on the signals before I need to do anything I + guess? +23:00 < karlp> i2c/uart are probably fine, spi might have an upper limit? +23:06 < BrainDamage> the highest frequency for squarewaves is 2/trise for + signal integrity, assuming say, 30MHz spi that gives us + 30ns period, trise would be say, 1/5th of a semiperiod so + 30/2/5 ~3ns, so that gives us a frequency of approx + ~700MHz for signal integrity, in turn that has a + wavelength of c/700MHz ~430mm, now a common ballpark is + that when your circuit size is comparable to 1/10th of the + wavelength, you start seeing +23:06 < BrainDamage> effects, since you said it's overall 60mm it's more than + 40, so yes, it'll distort the signal a bit leaving it + dangling there +23:07 * karlp loves you +23:08 < karlp> so, is the solution "throw small series resistors right by the + stub off to the dangling trace, so that it dissipates any + reflections on that line? +23:08 < karlp> I guess this also then changes when the LA is plugged in/not in + too. +23:09 < BrainDamage> the LA tends to be a very high impedance, so it doesn't + change a lot when it's in compared to the open circuit case +23:09 < BrainDamage> series termination should work +23:09 < karlp> adds capacitance though, I've foudn that out the hardware with + insufficient i2c pullups, but that's already taken care of. +23:09 < karlp> and yeah, just like 22ohm right on the beginning of the stub, by + the T-junction? +23:12 < BrainDamage> i forgot the exact value for series termination, it was + either 2x or 1/2 of the nominal impedance, if 22 is what + you've seen before then i guess it was 1/2 and yes, that + should fix it +23:14 < BrainDamage> if you feel brave there's faster terminations using + clamping diodes and caps, buuut, i'd say to keep it simple + first + -- cgit