Jim Williams

Okay, sure--that was the right answer all round. But you used the Web like a textbook rather than a crib sheet--i.e. you actually learned something. I use it that way too, and I admit that sometimes when I'm in a horrible time crunch I use cheesy Web calculators for stuff like the spectral absorption of the atmosphere. Buying a book sight unseen is a bit of a gamble, and it takes days to get there, whereas trawling the web for a bit is fast, cheap, and low commitment. Good medicine, as very many others have pointed out.

For a long time I've resisted using expensive commercial packages because I wanted to have my own set of tools--a man with his own tools is a journeyman, a man without is a wage slave. ;)

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics Electro-optics Photonics Analog Electronics

55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

[snip]

BTDT. In fact I had a helluva time switching from pencil schematics and hand-netlisting over to schematic entry ;-)

An engineer at Garmin (Paul Shumaker) came here and personally walked me thru learning PSpice Schematics. We used to meet over lunch, maybe once or twice a year, after the GPS chip was designed, because Paul had a son at ASU... but the son has now graduated and Paul has retired from Garmin... everyone gets old :-( ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
I love to cook with wine.     Sometimes I even put it in the food.

One of my guys, Rob, sometimes designs active filters by hooking a genetic evolution software thingie to an analysis/scoring program, and lets it push poles and zeroes around for a day or so. Seems to work pretty well. He winds up with filters that don't have any standard (Butterworth, Bessel, whatever) form.

I guess you could do the same thing for L and C values in a passive filter. You could even brute-force all combinations of standard values within some sensible ranges, and let it run for days. For a low-order filter, that might work. For a higher-order filter, it might take giga-years.

John

I'll (likely) end up with a 7th order LC plus two damped resonant circuits. The computer would get a major hernia from that one. Sometimes I lean back, squint at all those inductors and caps and come up with an idea which one to tweak and by how much, then the next, and iteratively on and on. The usual challenge is to obtain a fairly flat passband and low sensitivity for component tolerances. Filter software can't really do that. This is the sort of stuff where mathematical methods can frustrate.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

We will never know.

My recollection is Barry Gilbert just has a BSEE. I can't speak for modern electrical engineering degrees, but back in the day, you learned a lot of circuit theory in undergraduate. In grad school, you learned enough to write spice, but would that make you a better circuit designer? I'd say no.

the

et :-)

=A0 =A0 ...Jim Thompson

=A0 =A0| =A0 =A0mens =A0 =A0 |

=A0 | =A0 =A0 et =A0 =A0 =A0|

=A0|

=A0 =A0 =A0 |

Spice is not a design tool. Period end!

:-)

Why not?

John

:-)

It's a design tool in the same way that breadboarding, replacing resistors with trimpots, etc. are design tools.

:-)

If "design" means coming up with the documentation to build electronics that solves some problem, then anything goes. I've often written programs to brute-force search and find the divisors or parts values that I need, and I occasionally breadboard a circuit and fool with it until it works.

In a sense, mathematics itself can be a crutch. If a creature were sufficiently intelligent, everything would be intuitively obvious. When something is not obvious, sometimes we resort to doing a lot of math to get an answer. Just because we can solve a sheet full of algebra and calculus doesn't necessarily mean we actually understand the process any better. I've seen ludicrous mistakes made this way.

(I'm currently having a dispute with a PhD - not one here - about a signal processing issue, in his area of specialization. He's wrong.)

The best situation is to truly see through and intuitively understand the problem and the math. In electronics, sometimes you don't know enough about the parts or the process to apply a rigid math solution, or the math is impossible for mortals to solve. Or simulation/breadboarding is faster or more fun.

I can use Spice to fiddle out the values for a 2nd or maybe 3rd order LC filter. But not for a 7th order. Spice is great for things like tweaking control loops, where the goal isn't clearly defined... you just play with it until the transient response is pleasing.

John

Well, not directly... but John/Rob's approach where you just let an optimizer tweak your values can: You just tell the optimizer that you care about component sensitivity as much (or more... weight things however you want...) as the passband flatness. Optiziming on group delay near the transition region is also a popular pastime.

Small amounts of loss in a filter often seem to work in one's favor as well, making it a bit less sensitive to component variations as well as flattening out ripples in the passband.

I take it your frequency ranges are such that the nifty little 8th- and

10th-order switched cap filters from the likes of Linear Tech, TI, etc. aren't applicable?

---Joel

The NuHertz software is amazing. It will design extreme-performance LC filters of all sorts (including types you've probably never heard of) using standard value, finite-Q parts, and does sensitivity analysis.

I designed a radical filter for us in a few minutes, after I wasted about a week on "classical" methods.

John

If the PC can get through that before the next ice age appears (which doesn't seem too far away when I look at our remaining wood stack ...).

Is there any SW like this available for a test drive?

They would explode in milliseconds because this is high-power stuff :-)

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Ah, thanks, have to check NuHertz then. Although I only have to design this kind of filter once in a blue moon. The usual "design" happens in front of the network analyzer, solder iron to the right, Sierra Nevada Pale Ale on the left :-)

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

I don't think it would break any laws if we let you use our copy now and then, *before* trekking over to Zeitgeist.

The usual "design" happens in

Acck! How can you drink that stuff?

My older daughter loves SNPA and hates mayonnaise. Makes you wonder if that genetics stuff actually works.

John

I don't drink anything since I got my funnel.

{;-)

Jim

We could invite Jeff for a big pint of Russian Ale :-)

They do have a trial version:

Cold from the fridge :-)

Sometimes I also wonder, considering how very different the people can be in your own family. For example, in ours nobody in their right mind would ever want to move much, let alone across an ocean. When we did it we really found nothing to it, except stuff had to be packed a bit more thoroughly and you had to get used to another language (again).

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

It seems to work pretty well, AFAICT... I usually let an optimizer choose any floating-point value it wants for component values initially, and then once it's found something workable, go back and force it to only use E24 (or whatever) values to try to get something manufacturable.

Most all of the RF CAD packages like Microwave Office, ADS, and -- I think -- Ansoft Designer have optimizers like this in them, and there are free trials for all of them... but sadly they're all hideously expensive as well. I know there are some SPICE packages out there with built-in optimizers, but I've never used one and couldn't name one off-hand.

A lot of people use something like Matlab to test out their optimizers (there's lots of code for this freely available), but unfortunately Matlab is pretty slow.

---Joel

the

o on the

t sheet :-)

Let's say you are going to sell a million of whatever you design. Not really a problem in chips. The design has to be solid. You have to know why the elements do what they do. If you don't, something will eventually bite you on the ass. You can be sloppy in software design. Often I think it is a requirement. Software can be patched. Hardware needs to be recalled.

Tweaking filters in spice is probably passable, but you can easily run your pole zeros in an optimizer and force the filter to fit your task. Still it is far better to know what you are doing rather than to hack. With not much effort, you can run the "solver" in excel to optimize filters.

I did a custom chip that was supposed to replace a board design. The board design was such a piece of shit. Fortunately the so-called analog guy at the firm had quit, so I got to fix his crap without much hassle after proving to the customer the stupidity of the design.

Yes, it was a filter issue, though there were logic problems too. They got a better product. We got a smaller chip that was easier to manufacture and sold two million units.

Spice is not a design tool.

the

sheet :-)

You may as well say that math is not a design tool.

What *is* a design tool?

John