Those of you who simulate things like passive LC filters... when you want a simulation that includes the finite Q effect of the inductor, do you tend to use a model that computes the equivalent parallel resistance, Rloss, based on the Q specified on a data sheet at a given frequency (and then use that at*all* frequencies of your simulation, effectively creating Q(freq) )? Or do you simply specify Q and then use that at all frequencies (creating Rloss(freq) )?
For the following, assume you first calculated Rloss at a frequency f0:
The actual Q of an inductor varies with sqrt(freq) until you start approaching self-resonance, so assuming Q is fixed, you underestimate Q above f0 and overestimate below f0. Using just Rloss, you overestimate below f0 and underestimate above f0.