Thank You for your time and informative comments.
I will try to explain my needs and then to summarize my understanding of your comments.
I am developing a very low noise analog front-end with low pass filter for high speed signal acquisition board.
The corner frequency is at few hundred MHz (sorry, I can't be more specific). It is a Bessel (constant group delay) filter. I try to do it with discrete components (L-s and C-s). Those are quite good components with high Q (>10) and low production tolerances (<=2%). My problem is to take parasitics of PCB layout into consideration in selection of discrete L and C values. I plan to calculate the L and C of PCB and reduce values of correspondent discrete components by the same amount. Usually, PCB calculators provide transmission line characteristic inpedance Zo and wave propagation speed Vp. I use them to calculate L=Zo/Vp and C=1/(Zo*Vp) per length's unit. So, correct prediction of PCB absolute equivalent electrical characteristics (Zo and Vp) is quite critical for me. Let say, my design is sensitive to difference in trace inductance of 0.5 nH vs. 0.7 nH and I want to be sure that my prediction is reliable.
Now, to your comments.
1. As I understand, you agree that AppCAD is misleading in prediction of impedance for CPW w/o GNDplane.
Quite surprising for me, but I will take it as it is.
2. You provided results from two more calculators, and they are quite different. Zo of 55.0 ohms vs 63.5 ohms is a big difference for me, as you already understand. You also wrote that "chemandy" calculator is "worse" (or not ?) by definition, because it does not take (or does ?) in account the fringe fields etc. etc.
Which one would you use for my specific need?
Can you or anybody else provide any suggestion, especially at the higher level of the task.
For example, may be, to use significantly lower value of H will be better (i.e. to put a GND plane closer to signal layer). It will increase the Cpcb, which may be compensated by appropriate selection of Cdiscrete with good !!! (0.1+-0.05 pF) resolution, but, simultaneously, it will decrease Lpcb, and it will make selection of Ldiscrete (which are available with not so high resolution, but only with 10+-2 %) much easier.
Any other proposals/advices will be appreciated.
Thank You for your time.