That’s a tautological argument, people use curved traces because they think it is important. That only proves people generally follow the herd, it is not proof of any measureable effect.
Guys you walk in circles.
The reason why this feature is not in kicad is not because it is unnecessary in general but because nobody bothered to create a mergeable patch that adds this functionality.
Currently kicad is not particularly well suited for high frequency design. (From reading the forum i get the feeling that there are a lot of missing or half added features. Maybe a this will be better in v5 but i guess some things might still be missing in it.)
Using “free angle mode” routing in OpenGL I made this. My mind cannot handle seeing RF traces with angles even though I’m fully convinced it doesn’t matter. If you want, you can measure up the turning points in your trace using drawing elements for visual help while routing. If you need this exact same trace again you can simply copy it and assign it a new net.
I am guessing that curved traces started with crepe tape or ink resist pen designs.
It would be an interesting experiment with a field solver to see how high frequencies have to be for corners to show
Going in circles? 
All I can say is that we called schematics lumped models. When you get into higher frequencies you need a distributed model. Capacitor coupling do to electric fields causes cross talk to adjacent traces. Adding trace delay tuning and differential pairs is needed so I guessed someone knows the problems. As far as measuring, sometimes you get flaky errors that are hard to measure.
The frequent regurgitation of this topic often requires that you just let people believe what they want to believe as it seems no amount of empirical evidence, or even common sense, is going to convince them otherwise. But it is also quite comical to read the various explanations for why we should avoid right angle corners. This time around they’ve included comparisons to spark gaps and lightning rods, causing plasma and corona, and some new form of capacitor created by electric fields. Not to mention the single electron travelling along a sinusoidal trace. And don’t ask me what lumped and distributed models have to do with any of this.
This topic clearly highlights the vast misconceptions surrounding not only electricity in general but also the propagation of signals through conductors such as on a PCB. The only credible reason given above for avoiding right angle corners is the mention of mechanical stress in flex PCBs. Some people still seem to have this image of electrons racing down a track and trying to negotiate a right angle corner at near the speed of light, some of them fail, hit the edge of the track and bounce back. Imagine their surprise when they learn that, although the signal might propagate at near the speed of light, not only is the velocity of the electrons themselves quite slow (relatively speaking), but they barely move at all, orders of magnitude less than a millimeter (at 500MHz).
That’s not to say that right angle corners never cause any problems but when they do, two 45 degree corners are almost always sufficient. It stands to reason that rounded corners therefore must be even better, and they probably are, but not likely enough to make any difference in most cases.
5 Likes
Which is why some competent simulations with an EM finite mesh field solver would be very interesting
I read on the mailing list that somebody wants to implement such a thing for kicad as his phd project.
Cold winter, so let’s throw a log into the fire… 
I’d love to have at hand one of the books mentioned on the first slides of this presentation
as I suspect they would explain somewhere the gory details of drawings on page 47 (slide 93)…
I won’t try explaining them myself, as I’m just a humble lumberjack
1 Like
You could try explaining the point you are trying to make. 
These images just keep getting reprinted again and again, but hopefully people will notice that the rounded trace is considered as good as the two 45 degree trace, and the one labelled “best” isn’t round at all, but rather a 90 degree corner with a piece missing. I suppose that lets the electrons ricochet around the corner.
Two 45s are considered as good as the curved trace and KiCad does that now.
I would love to get the referenced books, which must have been written long before finite mesh analysis was practical.
The information on vias is limited, so after some searching I found this:
It seems vias start to be visible at about 3 GHz
That link appears to be broken.
If you consider the construction of a via it is not only a much larger impedance discontinuity due to it’s change in shape and size but also due to the corresponding discontinuity in the reference plane(s). The 90 degree corner is far better than a typical via but people don’t seem to pay as much attention to the advice that tells us to avoid vias in RF paths. (See slide 79 on the above rfdesign.pdf).
1 Like
The plain fact that I’m quite ignorant!
And did you see the staircase connections? Amazing!
I guess that electrons accumulate in the corners, like plancton in a squared pool. I just wonder whether they die as it happens to jellyfish… (I discovered that in a backstage tour of a nice acquarium)
Beside the nice talking, a good mention goes to an interview to dr. Howard Johnson
Around time 1:15:00 of the podcast he finally gives a great answer on why 90 degrees tracks aren’t an issue (for most of us).
But the great part is where he explains how that enormous misunderstanding has grown and propagated – finally a funny fact that quenches
my thirst for knowledge in an easy way!
Spoiler: it all began with people working at 50 GHz and 120 mils(!) traces, with multiple stage amplifiers. For them, that added 0.4pF on each corner is an issue. Even working with 10 mils traces is a so ridiculous contribute that can happily be forget.
Now I can sleep well, knowing I won’t dream of electrons bouncing in the corners anymore 
I have a local 455 KB copy of the via paper. PM me with your email if you want to see it
One thing is certain, you should not be building 50 GHz pcbs using cheap FR4 made in China boards
I’m sorry, not to throw more gasoline on the fire, but this paper doesn’t really qualify as “scientific” support of your argument. First of all, after doubting credibility of engineers you might want to use somebody with a little bit more weight than an EMC consultant (who may or may not even be just another engineer), even though he does write papers which are formatted in a very convincing double column formal style and even though it is the first search result that comes up on Google when you do a search for right corner traces emissions
And I’m sorry, but for any paper to claim to be relying on “empirical” “evidence” and including a graph with one of the axes unmarked is downright unscientific.
Based on my experience of tuning RF circuits with a network anylizer I can say that you can change your track impedance by doing minute variation in your layout. The same schematic with the same controlled impedance track would show up at a different spot in the Smith chart just by changing the layout while keeping all the components the same. So I would be careful to take scientific advice on RF layout from people who base their opinion on the the first search result that pops up in Google.
But I should take advice based on some ad hom criticisms and an unpublished anecdote?
Somewhat ironic…
Even “proper science” has a surprising lack of credibility, since up to 70% of results are not easily replicated.
Life in general, and not just engineering, seem to be based on a mix of urban folklore and chinese whispers. Kind of like junk DNA, junk “memes” seem to be propagate themselves, apparently based on attractiveness of the idea rather than any evidence. Most of the time it is largely harmless so never gets corrected.
For example, there is no particular reason to ban use of mobile phones in gas stations. The idea was based on some incident that can’t be verified and was probably caused by static, which actually does cause fires during refueling. But now there are signs warning people of the “danger”.
The incentive for reproducing results is low so it is a self perpetuating problem because it produces a culture where ‘rigor’ is not rewarded.
Absolutely not! I would suggest you leave Google alone, get yourself a network anylizer and test two identical circuits with a controlled impedance trace created using two alternative layouts - one with a right angle traces and one without and document the difference in impedance.
1 Like
I dare to say that in 1960’s (when it all started) east asian productions were not that common…
But as I just checked, Rogers Corp. was estabilished in 1832, then I believe they already had good products for high frequencies.
To me 50 GHz is just unmanageable as 50 G$ or 50 G€ 
Please refrain from personal attacks. (Or is it necessary that the moderators step in and close this?)