gnucap-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Gnucap-devel] dc command


From: Felix Salfelder
Subject: [Gnucap-devel] dc command
Date: Fri, 29 Mar 2013 12:27:36 +0100
User-agent: Mutt/1.5.20 (2009-06-14)

Hi Al.

the dc command shipped with gnucap .36 has two drawbacks
- non-convergence during sweep is not handled
- when sweeping over more than one value, initial guesses are
  inefficient.

non-convergence could be due to faulty models or to issues with the
solver, however, sometimes it's not possible to quickly improve either
of them. it seems better, to try with a smaller step /automatically/ and
recover -- similar to transient stepping. like for .tran, the trace
option can be used to display (a)ll or also (r)ejected steps.

when sweeping over two variables, the current implementation chooses the
last accepted step as starting point for the next homotopy. it's much
better to take the nearest known operating point. during recursive
sweep, known operating points need to be stacked.
for example, when sweeping through all combinations of xy=00...22,
which .dc does by computing these points line by line, left to right
00 01 02
10 11 12
20 21 22
it starts off at the solution for 02 to compute 10, while 00 would be
closer. note that this will change hysteresis simulation to do something
more natural a well (does it?).

an implementation and some unittests of/for the above is in the
gnucap-uf repo and latest tarball on tool. it quite works in all cases
known to me. however i cant distribute it as plugin, as it needs a
change in keep_ and restore_voltages in u_sim_data.

please consider some of this functionality upstream, or provide the
needed interface (this way or another).

have fun
felix



reply via email to

[Prev in Thread] Current Thread [Next in Thread]