Thevenin Theorem

Not true.

John

Seemed utterly boring and pointless in my days...(1970) but it had to be done. None of my logic circuits have two V sources... I guess it's more relevant for analogue stuff? If it doesn't go between 0 and 1, I don't want to know :(

Thevenin reductions are great, and allow complex networks to be reduced a step at a time. I like to, say, design a voltage divider using whatever resistor values make the math easier, compute the Thevenin output Z, and then scale all the values to get a target Z. That's much simpler than a bunch of simultaneous equations.

But it doesn't work for, say, RC networks.

None of my logic circuits have two V sources... I guess it's more

I love you digital guys. All college kids should learn digital logic and Java and stuff like that and ignore electricity, so I can plop a few opamps on a board and charge $5000 for it.

I wish you many, many 1s and 0s.

John

Each to their own , John.... I love my digital world. I just struggle on with the odd analogue bit . Happy Xmas m8.

Funny, when I lay out circuit boards for high-speed logic, I never think ones & zeros, I think about analog behavior of the signal propagating down the path(s) and radiating in to other circuitry.

--
Mark

John Larkin a écrit :

Uhh???

--
Thanks,
Fred.

Exactly. Think about a wired-OR using 200 picosecond logic.

John

It doesn't work for any time-dependant network.

The op is yet another "blogger" who is cutting and pasting textbook stuff and spamming newsgroups in hopes of harvesting google click revenue. We had to tell google to pull us out of that system - they put us in by default - as it was costing us a fortune in useless click charges.

John

For networks containing memory elements, you have to resort to the concept of impedance Z(s) or, if you are considering the sinusoidal steady state, Z(jw) but the concept still holds.

I would point out that the main usefulness of Thevenin's (or Norton's) theorem is not the step-by-step reduction you achieve in a limited class of circuits but the whole concept: Any (reasonably well-behaved) one-port may be substituted by a single voltage source in series with an impedance. This works, for instance, for an antenna: you may substitute the whole universe (including transmitter, the transmission path, interferers, etc) and the antenna structure by a series connection of two elements and the behavior of the true system measured at that port is indistinguishable from the behavior of the model. Not bad!

Or put in 3 resistors and have a 3 bit D/A converter sufficient for many applications :)

Pere

That's cutting it a bit short. Your two 'elements' may in fact be arbitrarily complex, requiring many R, L and C's to model.

Jeroen Belleman

You seem to be saying that, given any linear single-port box, it can be characterized as a linear single-port box. No argument there. But that's not what the OP said. As noted, the OP is trolling for google ad clicks.

Or do delta-sigma in an FPGA and get a 16-bit dac for one R and one C.

John

Thevenin's theorem is only true at one frequency. Hence an antenna can be modeled, but a more complicated network (let's say a combination of R, L, C, transmission lines, and worse yet, *real world* transformers!), where you are interested in its wideband performance, is essentially impossible to determine.

Does there exist an algorithm to take, say, a series of points of impedance vs. frequency, and determine an equivalent circuit? This would be the electrical equivalent of assembling a polynomial (Taylor series) or trig polynomial (Fourier series) from a series of roots. Since electrical circuits transform into polynomials, this shouldn't be too hard, although the process doesn't have to be trivial (finding polynomial roots and residues, seperating additive terms, factoring multiplied terms, etc.).

Tim

--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms

Google "System identification"

VLV

There is software that will take VNA data and deliver a Spice model. There are also companies that will come up with a Spice model for any part you furnish.

John

...and many SPICE tools can also take in frequency-domain data directly. For AC analysis the use is obvious, whereas for transient analysis they often do an inverse FFT and then use convolution at each time step to obtain a response. (PSpice can do this, whereas SI-Metrix unfortunately cannot. Dunno about LTSpice, but knowing Mke Engelhardt I would wager it can.)

Software that generates "native" SPICE models often use a rational polynomial to curve-fit the data (...rational polynomials being easy to simulate in both the AC and time domains). This is actually a bit trickier to do well than it might first appear, since ugly details such as causality rear their heads. When I was in school, one of the more effective algorithms was Bjorn Gustavsen's "Vector Fitting" approach, for which he provides freely-downloadble Matlab code

I've seen several commercial simulators that use his work (Microwave Office is one of them). The math behind it is "high-end undergraduate" stuff -- you need to be pretty familiar with some of the fancier aspects of linear algebra.

Those guys are quite good; we use the RLC library for filter design. The price is also quite reasonable considering all the work that goes into it.

One of the initially overlooked problems with modeling is figuring out just where your reference plane ought to be, and how you deal with mounting pad effects.

---Joel

Sure, but an RC isn't time dependent. (like any linear filter)

Thevenin is just about replacing a network with its apparent output impedance in series with its apparent unloaded output voltage and this can be done with any time invariant linear network.

Of course the voltage source becomes a function of jw, but it's sometimes useful.

--
Thanks,
Fred.

Joel Koltner a écrit :

Do you know of a similar approach in the time domain?

I once had to design a precision high impedance attenuator and it resulted in a multiple time constant compensation network that were hand adjusted. Trimmers weren't allowed and it cost a small fortune to tune it iteratively. Unfortunately the attenuation level didn't allow to operate in the frequency domain and we had to use an accurate 500V step as the test signal. And a sum of exponentials fitting where the time constants are spread on a large time scale doesn't work so well, even when using all the optimizations that naturally come to mind. We had to match two attenuators to better than 100ppm (diff. probe).

--
Thanks,
Fred.

message

But that's not what the spammer said.

John