Re: [Ngspice-devel] gshunt and rshunt

[Paolo N. <p....@ie...>]

On Mon, 2003-07-14 at 10:17, Stefan Jones wrote:

> The ngspice gshunt option seems to be more intelligent. It comes into
> play with Gmin, and when a DC initial solution fails. It just slowly
> increases the diagonal elements until the simulation converges.

Yes, seems more intelligent, find the less disturbing conductance that
will make the system converge. Anyway I am thinking about 2 issues:

- Compatibility: to be compatible with xspice, i think that both options
should be implemented, adding an option like "shuntopt":

shuntopt = 0 (default) -> Alan's scheme (ngspice default)
shuntopt = 1           -> xspice r-shunt scheme

Doing this we can test (tcl/ng)spice against xspice known results.

- Gshunt and rshunt add a conductance to the diagonal elemnts only, to 
  improve matrix condtion. I believe it is interesting to add a minimal 
  conductance between ANY node and ground, because, a parasite 
  resistance between any component and ground is physically present in 
  every circuit. Well, a capacitance in parallel will add even more 
  realism. This can help the simulation to converge and stop signal to 
  rise or fall with irrealistical rates. 

- Sparse library can calculate the condition number of the circuit 
  matrix. I think that this can be printed on the stdout to see how a
  simulation evolves. I will check but, if my memory is not failing, 
  the condition number can be used to estimate the number of correct 
  digits. Why not print this info too ?


Bye,
Paolo  


-- 
--
Paolo Nenzi
GnuPG public key available on keyservers
Key FingerPrint: FCC5 FDA4 4A31 EDE6 982D  AABE 6D48 9D3A 8FC0 A981