if I set:
number of noise/freq pairs : 3
Freq1 : 1k
Noise1 : 1e-12
Freq2 : 100k
Noise2 : 1e-18
Freq3 : 1M
Noise3 : 1e-24
I would expect, similarly to what happens with pwl sources in time domain, to get a piecewise linear frequency characteristic.
This is however what I get:
It is linearly interpolating - the problem is that you're plotting on a log-log scale, so that is distorting the linear interpolation.
In essence if you plotted the output noise (squared) (not in dB) versus the linear frequency, you'd see linear interpolation.
That said, in general I'd say that you really want some kind of log log interpolation for noise source - so there's an enhancement request I filed a number of years back asking for precisely this. The CCR number is 6408. I suggest you log a service request and ask for a duplicate CCR to be filed. There's a similar CCR asking for this kind of behaviour with VerilogA noise sources too - CCR 15584.
d'oh! It totally escaped me...In the mean time, I had to come up with some 'log/log plot generator' for my sake , based on Excel.
It is super basic but if anyone is interested and it is allowed, I could attach it here. This is how it looks like:
By all means post it here - I see no problem with that if you're willing to share it.
there it is. Very basic...But I think it is usable to build 'noise profiles' on the fly. It is fixed @10 points per decade but it is easy to change that. One possible improvement would be to allow for kneepoints not on the raster...But that involves some check :)
Hope this is of some use,