LTSpice Slow: help, please.

[Mark Johnson]

It used to be common to save the lineprinter output of SPICE runs, printed on continuous feed "fanfold" paper. People would trace out the different node waveforms of a transient analysis, using colored pencils. Then they'd pull the paper ten feet down the hallway, stand above, and study the circuit performance. You'd save the printouts from ten or twelve different SPICE runs, as a repository of expected circuit behavior, in a big thick stack on the floor of your office.

This was in the days of IBM 370 mainframes, punch card inputs, raised floor computer rooms with access restricted to system operators only, and lineprinter outputs. No video display terminals, no graphics. And yet the world managed to build circuits that worked and exceeded all specifications.

 

[Tournesol]

When your circuit behaves similar to an integrator, it may help to have a jump from the average value to the peak value at t = 0, rather than a sine wave that starts from the zero crossing. Draw a signal that is zero for t < 0 and sin(omega t) for t >= 0 and its time integral and you will see the problem. Draw a signal that is zero for t < p and cos(omega t) for t >= 0 and its time integral and you will see why it may help.

Any practical solution to ask LTSpice to make a fast transient analysis just for DC, replace the OP with the new values, then make a normal transient ?
It is boring: I have wrong values of output voltage, so, I cannot verify is my servo works OK. And, if i want everything working together, it takes too long time.
The problem is the following. If I replace the Servo with a voltage source, It do not contain the AC residual, so the distortions measurements are not accurate enough. And, If set a start time long enough, no way to work, too long at each mod and the OP is wrong.

 

[29285]

Waiting for the integrator to settle may take several 100ms or even more - a simulation over this period may take some minutes. Keep in mind the multitude of sinuoids that have to be calculated. Or the even bigger number if your circuitry is oscillating.

It may help to turn on your signal source at the end of the setup period.

 

[jan.didden]

The problem is the following. If I replace the Servo with a voltage source, It do not contain the AC residual, so the distortions measurements are not accurate enough. And, If set a start time long enough, no way to work, too long at each mod and the OP is wrong.

How about replacing the servo with a DC source plus an AC component?
Or put two sources in series, a DC and AC? You could make the AC part a fraction of Vout for instance.

Jan

 

[MarcelvdG]

Any practical solution to ask LTSpice to make a fast transient analysis just for DC, replace the OP with the new values, then make a normal transient ?
It is boring: I have wrong values of output voltage, so, I cannot verify is my servo works OK. And, if i want everything working together, it takes too long time.
The problem is the following. If I replace the Servo with a voltage source, It do not contain the AC residual, so the distortions measurements are not accurate enough. And, If set a start time long enough, no way to work, too long at each mod and the OP is wrong.

Come on, your simulation time is only a couple of hours. I sometimes have to run simulations that last several days at work.

Anyway, two other tricks you could try besides the step + cosine trick and the use of Hann windows are the use of ideal switches to switch the integrator time constant halfway the simulation and storing the final point of a transient simulation to later read it in as an initial condition for another transient run, although I haven't a clue how that is done in LTSpice.

 

[Tournesol]

How about replacing the servo with a DC source plus an AC component?
Or put two sources in series, a DC and AC? You could make the AC part a fraction of Vout for instance.

Not realistic VS phases and distortions.
Well it works more or less: Real measurements in real world will answer everything ;-).

Thanks to all for your helps.