F
Fred Bloggs
Guest
John Larkin wrote:
had to do with times at which the simulation results are stored for
output display and NOT the mathematical time increment size required for
convergence of the integration algorithms which was derived from error
tolerance parameters.
Maybe things have changed, but the original SPICE timestep specificationConsistant isn't the same as accurate.
Timesteps are usually set to a small fraction of the time any circuit
event is expected to take, like 1/100 or 1/1000 or something. You
can't expect to accurately simulate a 5 ms period when time is
quantized to 1 ms. Try 1 us maybe.
My rule is to make the time step as small as possible consistant with
simulation time. When in doubt, change the step time maybe 2:1 in
either direction; if the sim changes visibly, your steps are too
coarse by 10:1 at least.
What's a bummer is to have a circuit with a very high Q or widely
varying time constants; small dt and slow elements make for hour-long
transient runs.
John
had to do with times at which the simulation results are stored for
output display and NOT the mathematical time increment size required for
convergence of the integration algorithms which was derived from error
tolerance parameters.