From: Paul
Date: Mon, 14 Oct 2002 08:58:18 +0100
Subject: Re: [TSSP] F A N T C: - It's NEW! - and needs some TSSP input
Bart wrote: > Ran a detail (4) in Fantc: > BTW, with a detail = Time elapsed: 3529.5 secs (900Mhz chugg) Gosh, an hour. Hmm, have to try it on mine when the machine's not too busy. I don't think many users will wait an hour for a result. > fantc tssp measured > Fres(kHz) 236.82 234.9 238 kHz > Lee(mH) 18.14 18.44 > Les(mH) 18.44 18.69 > Ldc(mH) 18.85 18.84 18.67 > Cee(pF) 23.70 24.23 > Ces(pF) 24.56 24.57 > Cdc(pF) 37.31 37.50 > So yes, it converges toward tssp as you mentioned. That's a nice comparison there. fantc uses the same physical model as tssp, so they should converge unless there is a coding error. > Ran fantc at det 2 (236.5kHz, odd it's lower). It will hunt up and down a bit as it converges: both L and C are converging to their final values as detail increases, but they won't necessarily converge smoothly and at the same rate, so from time to time one or other will be 'ahead' and Fres will bob up and down. BTW, that shouldn't give the impression that tssp and fantc compute Fres by working out L and C and applying 1/(2*pi*sqrt(L*C)). They don't. Instead, Fres is worked out more directly, then the V/I profiles are computed from that, then the effective L and C are computed from the V/I profiles. As a final check, you can test it to see that Fres is close to 1/(2*pi*sqrt(Les*Ces)). Since we're not using the Fres formula in the calculations, we're entitled to use it as a cross check. I wouldn't worry too much about whether to use average or nearest wall radius, etc. The difference in most cases is below the meaningful accuracy of fantc, and for that matter tssp. With both tssp and fantc, we get down to a resolution where several effects start to introduce error. Even though tssp is equivalent to a fantc detail setting of around 20-30, it doesn't show any significant improvement over fantc except in one or two special cases where the resonator geometry is very accurately defined, eg isolated smooth electrodes, etc. There seems to be a point, around about detail = 2 or 3, that further detail will not meaningfully improve accuracy, both for tssp and fantc. Seems that fantc conveniently reaches the max meaningful accuracy at a running time that is just tolerable for the user. Some of the factors that limit accuracy are (in no particular order): Material dielectrics; Errors in measured dimensions, esp radius of coils and toploads; Poor definition of ground surface (eg do you use soil level or the tips of the grass blades); Walls and ceilings (ok for those who live in lighthouses, but few of us have cylindrical rooms); Eddy currents and leakage currents; Fres measurement errors; Atmospheric conditions; ok, the list goes on, and altogether they create a baseline of noise for comparison tests. Very nice that fantc digs down to this baseline (circa 1% to 2% error), but does not pretend to go 'beneath'. Fantc seems to hit it just right. Ok, now and again you'll get a coil that matches to 0.1%, but don't get excited because the next coil will be at 3% error. The only way to go to more accuracy is to build pristine resonators, inside precision faraday cages, and even that would only help a bit. For outdoor coils over well defined ground planes, the biggest residual error is that of material dielectrics (coil former mainly), and this I am working on. BTW, if you're left wondering why tssp uses a much higher detail, its partly because I need to make sure that the residual errors are dominated by physical effects, eg material dielectrics, etc, rather than computational approximations, and partly because we take results for the higher modes too, so proportionaly more detail needed for those. -- Paul Nicholson, --
Maintainer Paul Nicholson, paul@abelian.demon.co.uk.