Interesting idea.. Thanks .. Will try it out..
"It drops to 12 ppm when I replace the capacitor with a 10K resistor. A
pure resistor divider should give better accuracy than that.."
The fact that a pure resistive network, produces these errors suggest
that the errors are due to the spice algorithm used to interpolate for
the zero crossings All interpolation algorithms will converge to the
right answer, when the time step is very small ,approaching zero....
Instead of interpolating, a neighborhood search around the "Measure
points" may produce an effective answer. Normally we ask the question,
"What voltage at a given time?" The question that needs to be asked i "
what time for a given voltage?".
I think multiplying to a lower frequency will ease the time step burden;
but it will not alter the interpolation issue..
And there is the additional multiplication errors..
Remember that I am not even using my circuit yet.. I am simply
measuring LTspice's sine wave generator put through a resistive divider..
cheers
AG
toggle quoted message
Show quoted text
On 9/21/2011 5:43 AM, Kendall Castor-Perry wrote:
Funnily enough I used LTSpice to model a delta-sigma synthesizer a while
back. A 1.6Mpoint FFT was quite good at giving what appeared to be both
precise and accurate answers That's where I discovered that video
acceleration slows down screen plotting, which could take five minutes
to do
a redraw... The simulation stopped working with one of the updates, and I
never revisited it.
But anyway in your particular case, why not use stick a generator of known
frequency on the simulation, and plot the product of that and your own
output, cleaned up with an RC or other simple filter? In other words, just
mix down to a lower IF for display purposes. That'll improve the
'countability' of zero crossings. You'll still need a pretty short
timestep
if you want the accuracy, though. / Kendall
