Date: Mon Jul 23 2007 - 05:02:47 PDT
All, I've had a similar discussion last year on another forum with the usual suspects :-) http://www.designers-guide.org/Forum/YaBB.pl?num=1156512831/0 The only safe use of a $freq system function seems to be as part of the printing API ($write, $fwrite, $debug, etc.) as all other uses would essentially enter the problematic field of automatically translating frequency-domain models to time-domain. In that sense a % code might be more appropriate than a system function. Marq To "Geoffrey.Coram" <geoffrey.coram@analog.com> David Miller cc <David.L.Miller@free Verilog-AMS LRM Committee scale.com> <verilog-ams@server.eda.org> Subject Sent by: Re: Proposal for two additional owner-verilog-ams@se environment parameter functions rver.eda.org Classification 13-07-2007 16:36 Hi Geoffrey, Yes I understand for an ac, the circuit is linearized around the op point. I never (at this stage at least) intended for $freq to cause G and C to be recomputed. I have supported $freq for many years and it is mainly a debug function (It is helpful when used in conjunction with $debug()). I also use it in noise analysis, where we can have frequency dependent elements. It has proven helpful to designers when debugging model problems. I can detect models that are dependent on frequency to ensure that they are solved each point (even if that solve doesn't modify the jacobian). Recently a model developer asked me how to load noise freq/pwr pairs from a file. One solution is that you use $fscanf to load the values from the file into an array and then use noise_table(). However this is inefficient as you need to know how many elements are listed in the file so that your Verilog-A array is the correct size. So the other way was to use $table_model() and $freq. This could do the interpolation of the freq/pwr pairs, and then you could use white_noise() to load that pwr value. This worked beautifully, it allows you to have all the benefits of the $table_model interpolation/extrapolation algorithms (not just clamping at the bounds). However it is breaking the definition of white_noise() since now my pwr value is frequency dependent. A couple of ways to solve this: 1. Extend noise_table() and overload the first arg such that if it is a string then it represents a file name which contains the freq/pwr pairs. 2. Add $freq, and a new function 'colored_noise(pwr)' and use $table_model to do the interpolation. I prefer 2. as I am not keen on overloading arguments to be different types. But I was still working through some of the finer points of the colored_noise() function, so wasn't quite ready to propose it. But I thought $freq was fairly simple, so thought I would send it through now for people to critique. Cheers... Dave Geoffrey.Coram wrote: > Arpad - > $abstime works OK in a transient analysis, because the full model is > evaluated > at each timepoint. > > Spice's ac analysis is a different sort of animal: the operating point is > computed, and then *the circuit is linearized about that op-pt*. > Generally, > this means that the model is evaluated during the op-pt and all the > derivatives > are stored off. Then, for each point in the ac analysis, the > derivatives are > used to compute J = G + j w C, where G=conductances and C=capacitances > (derivatives of contributions involving ddt()); G and C are *not* > re-computed > at the new frequency, because Spice assumes that they only depend on the > v-vector (the operating point). > > I agree with David (from Tiburon) about the hazards of $freq expressions > for models that are supposed to work in transient, HB, etc. > > -Geoffrey > > > Muranyi, Arpad wrote: >> Geoffrey, >> >> I don't understand the reason for why the IF statement >> wouldn't get evaluated during the AC sweep. Could >> you please explain that to me? >> >> If we had this in the language, I would expect it to >> work the same as $abstime. Doesn't $abstime=0 in the >> operating point analysis and greater than zero >> thereafter? If I remember correctly, I was able to >> use if($abstime ...) before. Why couldn't $freq do >> the same? >> >> Thanks, >> >> Arpad > -- ===================================== -- David Miller -- Design Technology (Austin) -- Freescale Semiconductor -- Ph : 512 996-7377 Fax: x7755 ===================================== -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean. -- This message has been scanned for viruses and dangerous content by MailScanner, and is believed to be clean.
