| ID: | 3897 |
| From: | ge...@igor2.repo.hu |
| Date: | Thu, 2 Apr 2020 17:04:59 +0200 (CEST) |
| Subject: | Re: [pcb-rnd] drc: Antenna detection |
| in-reply-to: |
Hi Vuokko,
On Thu, 2 Apr 2020, Hannu Vuolasaho wrote:
> Hello!
>
> Now that DRC ideas are open, I'll send my ideas.
>
> This feature is in features which doesn't increase productability but may save
> a lot of time.
>
> When there is single connection to routed signal it can cause major ugliness
> and work effort to rest of the routing when going around it. Detecting trace
> which doesn't begin from terminal, polygon or via but which is net should be
> warned.
>
> Most of the time it is misclick or forgotten part of trace. Sometimes it really
> is antenna.
In pcb-rnd terminology this is basically trying to identify a "branch" of
a segment^2 of a network^1 that "leads nowhere", being useless for the
board (not connecting anything to anything else) and potentially being
dangerous (acting as an antenna).
Normally this is something easy to do with human brain but not very easy
to do with algorithms.
My first ideas are these, please think over if it would work correctly for
all positive and negative cases.
(I think polygons don't need to be handled specially, only terminals.)
Method I.
- if a copper object is reachable from exactly one other copper object, it
is surely the end of such an antenna -> violation
- else it may be accessible from zero other objects (stand alone object
in the middle of the desert) -> no problem
- else it is accessible from multiple other _directly connected_ copper
objects; in this case, we need to make an (expensive) map:
1. remove our target object for a moment
2. take all directly connected copper objects and do a search from them,
listing all terminals they reach
3. compare the terminal lists: if we find at least 2 directly connected
objects that yield different lists, we are okay; if all directly connected
objects yield the same terminal list, that means they all connect our
target object to the same things. Alternative wording: our object may have
multiple direct connections, but they all lead to the same terminals, so
it's the same as we had only one connection.
Method 2:
I think in a very indirect way the above really says: if we deleted our
target object, it wouldn't break the current network segment in two or
more new segment. So this other method tries to capture that property more
directly:
1. take a network segment, map all terminals on it and save this as the
master list
2. map all objects of the network segment and iterate over them
3. in each iteration remove one object, while keeping every other object
in place and do the segment mapping from scratch from the first terminal
of the master list
3. if the result of this new mapping is the same as the master map from
step 1, that means removing our current iteration object wouldn't break
the segment, which means (in a sense) the object is useless. It's probably
an antenna.
... unless you have two alternative, parallel paths between 2 terminals!
In that case both paths are violations because removing any object from
either of them will not break the net because the other path is still
there!
Unfortunately I think method I suffers from the same error.
Saying something like "the antenne case has a final object that has only
one copper object connected" is not good either: what if I end my net in a
loop of lines? Then I have an undetected antenna.
So, this problem is harder than it first seems! I will keep pondering, but
I can't promise we can really do it. (Note: whether we can do it or not
doesn't depend on the new drc engine, it depends on whether we can come up
with a formal definition on what we are trying to find in our data.)
Footnotes on our internal code terminology:
^1: a network is a abstract/logical thing, determined by the list of
terminals from the netlist
^2: in the context of the actual geometry of your board (current state) a
network segment is a portion of a network that is already galvincally
connected (with copper objects, not with rats). When you start,
subcircuits placed but no trace drawn, a network has as many segments as
many terminals; when you finish routing, every network is a single
segment. In other words, a segment is what {c f} highlights if you don't
have rats at all.
Best regards,
Igor2
Reply subtree:
3897 Re: [pcb-rnd] drc: Antenna detection from ge...@igor2.repo.hu