gEDA-dev: code sprint ideas

[Peter Clifton]

On Mon, 2007-10-15 at 19:06 +0200, Bernd Jendrissek wrote:
> On 10/15/07, Peter Clifton <pcjc2 at cam.ac.uk> wrote:
> > With square + rounded, the misalignment is not allowed to be as big in
> > both directions at the same time. (Eg.. if the registration is out both
> > in the X and Y directions at once, it might mean mask covers the corner
> > of the pad whilst the X and Y registrations are otherwise in tolerance.
> > (It seems likely to me that the X and Y registrations are independent
> > error terms, not some "distance" offset error in registration.
> 
> IF your X and Y errors are normally distributed, then exactly because
> they are independent variables, you can indeed do a coordinate
> transform into (r, theta) and take advantage of the rotational
> symmetry of the error distribution: your X and Y error limits would
> then be equivalent to a limit on "distance" error.

I wasn't assuming the errors were normally distributed, but I take
you're point. Anyone friendly with a fab house enough to find out how
these registration errors work for real?

> OTOH, if the fab house achieves its tolerances only by grepping out
> the out-of-error-band machine settings, then you'd need a
> square+square pattern to survive their out-of-control process with its
> rectangular error distribution.

Peter