Re: gEDA-user: lost newbie

[John Doty <jpd@xxxxxxxxxxxxx>]

On Jan 8, 2006, at 7:45 PM, Dan McMahill wrote:

> Marc wrote:
> > Dan you forgot to mention its cheaper to run a sim before you  
> > actually build anything im not currently in college as everyone  
> > suggests here but i might go do Physics next year im currently  
> > tired of doing college courses i just finished Cisco.
> > Marc :)
>
> For IC design I agree completely and in that area I simulate things  
> almost every day.  For many analog board level designs, I don't  
> agree. There are many circuits for which a hand analysis is  
> completely adequate and a spice simulation is the wrong tool for  
> the job.  If it's a commercial environment, you can quite easily  
> spend more paying for someone to try and come up with a simulation  
> which is even close to reality than having them just build the thing.
>
> What you very quickly run into unless it is something like a filter  
> or a transistor amplifier is that the models either don't exist or  
> aren't good enough to be worth anything.  Take a switching power  
> supply.  You can simulate the power stage (with some modeling  
> effort), but the controller chips are complex and you can spend  
> more time trying to come up with a behavioural model which is even  
> 0th order correct than actually building something.  A lot of the  
> transistor models which are available for discretes aren't very  
> good either.  Sometimes it's due to a large process variation,  
> sometimes because of poor extraction of the model, sometimes  
> because the model used isn't capable of predicting the sort of  
> behaviour of interest.  I came across an example of this recently  
> where someone had some discrete FET models which are not capable of  
> subthreshold modeling but the designer was designing a subthreshold  
> circuit.
>
> I've designed a fairly large number of analog board level circuits  
> and essentially none of the problems I've run into when testing  
> them in the lab would have been predicted by a simulation.  About  
> the only board level simulation I've ever found to be useful is for  
> some filters where I wanted to look at some statistical analysis  
> and in cases where I wanted to model some finite Q's and board  
> parasitics on some RF filters.
> Certainly other cases can be found, but by in large I have not felt  
> that board level analog simulators have been important. I'm sure  
> others will disagree with me on this.

Well, I agree completely. Simulation does not prevent smoke. The  
accuracy of component models is difficult to assess. When you  
simulate, you generally should be seeking the answer to a specific  
question, not a general functional test. What you simulate is  
generally not the circuit you build, but has large portions sketched  
in in idealized ways, and has extra components representing  
parasitics around the piece you're probing.

In many cases, it's much quicker and easier to make a prototype and  
try it. Unless you're a real simulation expert, a real circuit is  
likely to yield more insight than a simulation, too.

John Doty              Noqsi Aerospace, Ltd.
jpd@xxxxxxxxxxxxx