The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl Simulators >> RF Simulators >> difference MIC and FDTD (+ crash) https://designers-guide.org/forum/YaBB.pl?num=1158922730 Message started by Paul Geraedts on Sep 22nd, 2006, 3:58am |
Title: difference MIC and FDTD (+ crash) Post by Paul Geraedts on Sep 22nd, 2006, 3:58am Hi all, Does anyone know where I can find details about the Finite-Difference Time-Domain refinement method as implemented in SpectreRF? I have found a good article on the Multi-Interval Chebyshev refinement method already, but I would like to know when to use which. http://www.ieeexplore.ieee.org/iel5/6899/18566/00855300.pdf?tp=&arnumber=855300&isnumber=18566 As far as I understand, MIC is available in both forced and autonomous PSS and FDTD is only available in forced PSS. I also found a bug in Spectre related to MIC; see the attached logfile. When you reuse *unrefined* PSS data (so the periodic steady-state data itself as written with writePSS without having MIC enabled) while MIC is enabled, a segmentation fault results. It does not matter if you set checkPSS to yes or no. Paul |
Title: Re: difference MIC and FDTD (+ crash) Post by Paul Geraedts on Sep 22nd, 2006, 7:16am As it turns out, a work-around of the segmentation fault is reusing the already *refined* PSS data (instead of the unrefined PSS data). So enabling MIC in both PSS analyses. The periodic steady-state will get even more refined now, so basically it has the same result as decreasing psaratio. Paul |
Title: Re: difference MIC and FDTD (+ crash) Post by byang on Sep 22nd, 2006, 9:42am Hi, Paul, As the original developer of the method (some years ago), sorry for this kind usability issue and the crash. I am not with Cadence now. However, if you can send me your netlist, I can try to relay it to Cadence R&D to take a look. My e-mail address is byang@gemini-da.com. The difference between finite difference method and shooting-Newton method is that the periodicity is enforced exactly in finite difference method. In shooting-Newton method, the solution after a period can differ by some error tolerance. MIC is high-order finite-difference method. If you want very low noise floor and high accuracy, it can help. Regards, Baolin |
Title: Re: difference MIC and FDTD (+ crash) Post by Paul Geraedts on Sep 22nd, 2006, 11:26am Hi Baolin, Thanks for your quick reply! Don't worry about the crash, the work-around does the trick nicely. The netlist that I'm using contains a lot of stuff which I'm not allowed to pass on. Another similar example will probably show the same behaviour. Hopefully somebody from Cadence is reading this as well (Andrew? ;). But thanks for your offer! Yes, MIC seems very useful. I've read somewhere that it can get the noise floor to levels of about -200dB! Great for oscillator design. So am I right that FDTD, as implemented in Spectre, is a refinement method? So that always a (coarse) periodic steady-state has to be calculated first with Shooting-Newton (the combination Harmonic Balance & FDTD is probably a bit silly, right?). So MIC is as exact as FDTD (otherwise, why would you want to use MIC)? I still don't really get when to use which method (FDTD/MIC). Could you give me some general advise? I don't want to bother you too much with it though. Do you know of any good article on FDTD as implemented in Spectre? Thanks again, Paul |
Title: Re: difference MIC and FDTD (+ crash) Post by byang on Sep 24th, 2006, 11:32am Hi, Paul, MIC, as it is implemented in SpectreRF, is one type of FD method. It is just that the integration method is Chebyshev method, which could be high-order, while non-MIC FD method is only up to 2nd-order (Gear or Trap method). Jacob White and Ken Kundert's book on Steady State method has some information on the finite difference method. Baolin |
Title: Re: difference MIC and FDTD (+ crash) Post by Paul Geraedts on Sep 24th, 2006, 2:19pm Hi Baolin, I've never understood that both methods are so similar! I think I'll use MIC exclusively from now on, as it seems to me that MIC is an improvement on all fronts. Or do I overlook situations in which using the 2nd-order method can have advantages? I have seen the book that you mentioned in our library. I'll have a look at it. Baolin, thanks for your posts. You have been a great help! (To get accurate information on these topics is pretty hard for a regular user as myself.) Paul |
The Designer's Guide Community Forum » Powered by YaBB 2.2.2! YaBB © 2000-2008. All Rights Reserved. |