> On Wed. 10 May 2023 at 15:30, <Thomas.Kopp@xxxxxxxxxxxxx> wrote: > > > > > > On 05.05.2023 07:07:03, Thomas.Kopp@xxxxxxxxxxxxx wrote: > > > > > > The datasheet "MCP25xxFD Family Reference Manual" says it needs > an > > > idle > > > > > > bus. > > > > > > > > > > Technically what's needed is an idle condition on the bus as specified > > > > > in the ISO. i.e. 11 consecutive sampled recessive bits after a full > > > > > frame (if one is currently in transmission). > > > > > > > > Right. What happens if another CAN frames comes before there are 11 > > > > consecutive sampled recessive bits? The IFS for CAN is 3 bits? > > > > > > Not quite. Between the Frames is an IFS that's correct but the IFS consists > of > > > an Intermission which is 3 bits long + a bus idle condition of 11 bits. Regular > > > frames have to wait for the IFS to elapse BUT there's an exception for error > > > frames and overload frames. EF may be up to 12 bit, OF are 8 dominant + 8 > > > recessive bits, browsing through the spec I think only 2 OFs can happen > > > consecutively. Adding another 32 bits to the formula should cover this. > > > > Re-reading the spec again I noticed that the part where I wrote > > that there's an "idle condition" after the intermission is > > wrong. What follows the intermission is just "bus idle", > > defined two paragraphs later as "The period of bus idle may be > > of arbitrary length." The 11 recessive bits can be removed from > > the formula again. The longest period (under the assumption > > there aren't multiple/continuous errors on the bus) will be > > Frame + Error Frame (12 bit) + 2 x Overload Frame. > ^^^^^^^^^^^^^^^^^^^^ > > How did you find that a error frame is 12 bits? From section 10.4.4 > "Specification of EF", I can read: > > The EF shall consist of two different fields. The first field > shall be given by the superposition of error flags contributed > from different nodes. The second field shall be the error > delimiter. > > Then: > > Two forms of error flag may be used, the active error flag and > the passive error flag, where > > - the active error flag shall consist of 6 consecutive > dominant bits, and > > - the passive error flag shall consist of 6 consecutive > recessive bits unless it is overwritten by dominant bits > from other nodes. > > Finally: > > The error delimiter shall consist of 8 recessive bits. > > So the error frame should be 14 bits (6 + 8), not 12, right? That was imprecise, you're right - I meant an Error Flag, not Frame and hence the 8 recessive bits were missing. There's an active error flag + passive error flag though which can be 6 bits long each. In section 10.4.4.2 The total length of this sequence may vary between a minimum of 6 bit and a maximum of 12 bit. > For the great total, if you want the absolute worst case, you should > consider that: > > - The error frame may start at any point. For example, you may > observe the first two bits of the intermission and have it > broken by an error frame. It may also appear in the middle of > a data frame. So you may consider cases such as: partial > frame + error frame + intermission + frame + ... > > - The overload frame is required to "destroy the fixed form of > the intermission field". So even if not explicitly specified, > I think that overload frame may start after the second > recessive bits of the intermission, e.g. frame + 2 bits of > intermission + overload frame + 2 bits of intermission + > overload frame + full 3 bits intermission > > - An error frame or an overload frame may trigger another error > frame if a node does not correctly receive one of the bits in > the fixed form delimiter. The only exception is the last bit > of that delimiter (c.f. section 10.11 "Error detection" > paragraph d) "Form error"). In other words, you can have > the two overload frames, then an error frame, then two > overload frames again. > > This is to say that the worst case scenario is just not something > worth consideration. ACK > The assumption that only one error frame occurs is pretty arbitrary. I > would suggest making it simpler and simply ignore error and overload > frames. As long as frame + intermission works well in empirical tests, > I would suggest to stay with that and call it a day. Correct, that's why I wrote: (under the assumption there aren't multiple/continuous errors on the bus) but I agree that's a pretty arbitrary limit. So I'm fine with changing it to one or two full-sized frames + intermission) Best Regards, Thomas
