Least number of steps to solve linear circuit ?

Thread Starter

DarthVolta

Joined Jan 27, 2015
150
If you have a linear circuit, AC or DC , with X# of voltage sources and Y# of current sources, what are the general rules for deciding when it's best to use mesh, node or superposition analysis, and maybe use source transformations and Theiv/Norton ?

I've done them all separate, but given a random circuit where you have to find a single voltage or current, I usually just brute force it with mesh current analysis and a matrix solver, unless there's an obvious quick route.
 

WBahn

Joined Mar 31, 2012
24,853
If you have a linear circuit, AC or DC , with X# of voltage sources and Y# of current sources, what are the general rules for deciding when it's best to use mesh, node or superposition analysis, and maybe use source transformations and Theiv/Norton ?

I've done them all separate, but given a random circuit where you have to find a single voltage or current, I usually just brute force it with mesh current analysis and a matrix solver, unless there's an obvious quick route.
That decision is largely art and experience. Even with the same circuit topology, the specific values and things like whether there are symmetries that you can exploit can have a significant effect.

A couple of general rules of thumb are that if you can place current sources in exterior meshes then that favors mesh analysis while being able to pick a reference node with multiple voltage sources attached favors nodal analysis.

The more experience you have using all of the available tools at your disposal, the more comfortable and confident you will be using a mix of them that is a good set for the particular problem.

This cuts both ways, techniques that you don't spend the time and effort getting comfortable will not be used even when they are the best choice -- For instance, I never use Millman's Theorem and so even if I had a circuit that screamed out for it I wouldn't even know what I was missing. Similarly, I seldom use delta-wye transforms and so even when I spot a circuit where I can tell they would make life much easier it is quicker for ME to use another approach rather than step back and lookup or rederive the transforms.
 

OBW0549

Joined Mar 2, 2015
3,014
If you have a linear circuit, AC or DC , with X# of voltage sources and Y# of current sources, what are the general rules for deciding when it's best to use mesh, node or superposition analysis, and maybe use source transformations and Theiv/Norton?
I'm not aware of any such general rule that can be applied to all circuits. All of the above techniques are usually available, and I find that after gazing at a circuit for a while one of the methods will pop out as a good place to start. As @WBahn said, it's largely a matter of art and experience.

I've done them all separate, but given a random circuit where you have to find a single voltage or current, I usually just brute force it with mesh current analysis and a matrix solver, unless there's an obvious quick route.
Same here; most of the time I end up just writing out the nodal equations and solving them as a simultaneous set. Often, some simplification is possible by use of source transformations or Thevenin/Norton equivalents. Brute force.
 

Audioguru again

Joined Oct 21, 2019
254
Cheat. Use LTSpice.
A simulator program does not read the details of a datasheet like a human engineer. The simulator model has only "typical" spec's for a transistor but a human knows that spec's have minimum and maximum values that must be calculated so that ANY minimum to maximum passing part works perfectly. When you buy some transistors then maybe they all have below typical spec's put still pass the range of spec's. The simulator program does not think about that unless you fiddle with its model.
 
Last edited:

SamR

Joined Mar 19, 2019
1,295
I consider it as a first approximation and not the final exact solution which ultimately depends on the measured device parameters and not the PDF. It is useful.
 

WBahn

Joined Mar 31, 2012
24,853
A simulator program does not read the details of a datasheet like a human engineer. The simulator model has only "typical" spec's for a transistor but a human knows that spec's have minimum and maximum values that must be calculated so that ANY minimum to maximum passing part works perfectly. When you buy some transistors then maybe they all have below typical spec's put still pass the range of spec's. The simulator program does not think about that unless you fiddle with its model.
In general I agree with you. However, you can do better than just typical specs by using a Monte Carlo simulation. If the manufacturer's device model is good, it should contain the information about the variation in the important parameters and then most simulators will randomly pick values for each parameter according to a Guassian distribution out to three-sigma. Of course, you have to be aware that these are about the typical values and so if you have parts that have a systematic bias in them (probably because they came from the same lot), then you may still have issues to deal with.
 

WBahn

Joined Mar 31, 2012
24,853
You can extremely overload a little transistor in LTspice and it does not produce any smoke and does not burn out.
No argument from me. Far too many people use simulators to do all of their thinking for them. By the same token, there's no shortage of people that fail to take advantage simulators to make better designs by running simulations over process parameters. The simulation models also do a much better job of dealing with the nonlinearities of the devices.

When I first started using Spice the results that you got were usually only a reasonable approximation of what you got in actual measurements. So that is what I expected to happen when I started designing ICs (where you have no choice but to simulate your designs). I was amazed at how accurate the models were -- the bias voltages were usually well within 10 mV of what the simulations predicted.
 
Top