Author Topic: Finite element Transient Signal Analysis  (Read 2279 times)

0 Members and 1 Guest are viewing this topic.

Offline aussie_laser_dudeTopic starter

  • Contributor
  • Posts: 46
  • Country: au
Finite element Transient Signal Analysis
« on: April 18, 2020, 11:07:10 am »
Hey guys,
I've got two circuits in an attached image, both appear to be the same circuit, just a resistor and a capacitor in parallel connected to jumper leads.



A naive analysis of the two circuits would suggest that they are identical in performance. My understanding is that there is actually a subtle difference caused by the extra metal track path that could be measured by time domain reflectometry / some advanced transient signal analysis / something along those lines.

I don't have an EE background (but strong in math / physics / science) and would like to do something like a "transient signal analysis" for both circuits where a short square wave pulse is applied across the jumper leads and the charge distribution / electric field / current / voltage etc are calculated at different points along the circuit with respect to time.

To say this another way, let a brief 1V square wave pulse occur during time t=-1000 to t = 0 picoseconds.
   What is the electrical current at position 'X' at time t=0 ps, t=1ps, t=2ps....... t = 1 minute (ps standing for picosecond)?

 My understanding is that circuit a) and b) will have almost identical current with respect to time, but due to speed of light delay / dielectric effect of pcb / time varying charge distributions + electric fields we will have subtle differences. Maybe in some circuits, like kilometer long wires in weird geometries we could easily measure these effects.

I'd appreciate some knowledge about what techniques / software etc are available for these type of calculations. I'm looking for current techniques, and also looking for an understanding of the underlying physics. What principals are in play? eg. momentary charge accumulation in wires, speed of light delay of electric field, mass of electron? velocity distribution of free electrons? i'm guessing on those last two. Obviously there's non-ideal properties of devices like capacitors leaking current. I'm more interested in the physics of electricity dashing around weird pcb track layouts.

Any knowledge would be greatly appreciated :)
John

 

Offline TheUnnamedNewbie

  • Super Contributor
  • ***
  • Posts: 1211
  • Country: 00
  • mmwave RFIC/antenna designer
Re: Finite element Transient Signal Analysis
« Reply #1 on: April 18, 2020, 12:24:55 pm »
Underlying principles is just Maxwells equations.

There are I believe 3 main algorithms that are used in commercial simulators for these kinds of applications:
- Finite-difference Time-domain or FDTD - usually 3D
- Finite Element Method or FEM - usually 3D
- Method of Moments - 2D. Marketing likes to call them 2.5 D (which I can understand) or even 'Planar 3D'. In short, the original MoM algorithm can only work with currents that are planar, but they can be in different planes (in other words, current vectors must always be 0 in the z-axis, but the z-coordinate of the current vector may be non-zero). For much PCB work this is enough, as you can often consider PCB structures as planar. Breaks down once you get to very high frequencies and very thin traces (where your trace thickness becomes a significant fraction of trace width). It has some advantages like only having to consider metals (you need to do some mathematical magic for calculating the substrate one time). There are some 'adavancements' that can also acount for vertical current, and most importantly, that the solvers tend to be way cheaper to buy. Sometimes people also say they are much faster than full 3D simulators, but my experience is that this is often offset by the high-performance adaptive meshers in FEM solvers.


Software (non-exhaustive list, just what I have experience with):

CST Microwave studio (Dassault systems)
EMPro (Keysight)
Advanced Design System/ADS (Keysight) (Note that Keysight has been actively rebranding their software stuff under one family - it is now all called Pathwave I think?)
Momentum Simulator (Keysight, part of ADS)
Lumerical (Lumerical Inc)
HFSS (Ansys)
Microwave Office (Was by AWR under national instruments, but Mentor bought them)
EMX by Integrand Software
Cadence has some solver but their marketing material is just the biggest pile of trash ever that tells you absolutly nill about what their software actually can do.
Sonnet

FDTD simulators: CST Microwave, EMPro in FDTD mode, Lumerical FDTD (Lumerical also has a 2.5 D solver that can do smart stuff for simulating waveguide structures), I think HFSS has an FDTD solver these days too. Comsol probably also has one since comsol can pretty much do everything you want.
FEM simulators: CST Microwave I think has FEM solvers these days, EMPRo in FEM mode, HFSS, ADS FEM solver (still not sure how this is different from EMPro)
MoM simulators: ADS Momentum, EMX, Sonnet, Microwave Office,

Note that all of this software is generally quite expensive. I believe it goes from a few k$/license for some of the 2.5 D solvers, to 250k$ and up for the very advanced 3D solvers like HFSS. As a rule of thumb they are all absolutly horrible to work with and notoriously unstable, but I can't do my job without them. The more you get away from the basic stuff, the more you really needa understand what is going on, because these simulators will only spit out results as good as your input. If you decide to make ports that are a few wavelengths large, you will get nonphysical nonsense out.


About your specific circuit example:

The current will most definitly not be the same. Even if the pathlength were the same, the 'snake-y' trace will have much more significant inductance.
The best part about magic is when it stops being magic and becomes science instead

"There was no road, but the people walked on it, and the road came to be, and the people followed it, for the road took the path of least resistance"
 
The following users thanked this post: aussie_laser_dude

Offline aussie_laser_dudeTopic starter

  • Contributor
  • Posts: 46
  • Country: au
Re: Finite element Transient Signal Analysis
« Reply #2 on: April 18, 2020, 01:12:20 pm »
Quote
Underlying principles is just Maxwells equations.
Nice! I was hoping for this.

Didn't really know what to expect with software, looks like people have actually done a lot of stuff in this area. Thanks for the incredible list!

Quote
As a rule of thumb they are all absolutly horrible to work with and notoriously unstable, but I can't do my job without them.

Does there exist software that will do steps like these?

1. Make a PCB design with Digi-Key component data linked to components in altium/kicad etc (something like my above circuits for example)

2. Create a set of arbitrary input waveforms (a single square wave pulse in my example)

3. Select point on circuit with measurement type. Select simulation time. ('X' mark for current as measurement type and 1 minute simulation time in my example)

4. Hit "run simulation" button, wait a few hours/days while simulation output data is graphed (current vs time graph for position 'X' in my example)

Very interested to know if there's something like this.

Thanks again for the amazing response!
John
« Last Edit: April 18, 2020, 01:35:38 pm by aussie_laser_dude »
 

Offline 2N3055

  • Super Contributor
  • ***
  • Posts: 7466
  • Country: hr
Re: Finite element Transient Signal Analysis
« Reply #3 on: April 18, 2020, 02:38:08 pm »
You should look into transmission line theory. Plenty of literature on the topic..
Take a peek at FEMM and OpenEMS.
"Just hard work is not enough - it must be applied sensibly."
Dr. Richard W. Hamming
 
The following users thanked this post: aussie_laser_dude

Offline TheUnnamedNewbie

  • Super Contributor
  • ***
  • Posts: 1211
  • Country: 00
  • mmwave RFIC/antenna designer
Re: Finite element Transient Signal Analysis
« Reply #4 on: April 18, 2020, 02:39:29 pm »

Does there exist software that will do steps like these?

1. Make a PCB design with Digi-Key component data linked to components in altium/kicad etc (something like my above circuits for example)

2. Create a set of arbitrary input waveforms (a single square wave pulse in my example)

3. Select point on circuit with measurement type. Select simulation time. ('X' mark for current as measurement type and 1 minute simulation time in my example)

4. Hit "run simulation" button, wait a few hours/days while simulation output data is graphed (current vs time graph for position 'X' in my example)


yes and no. Two main reasons: The most powerful solver algorithms are frequency-domain based (FEM and MoM). You can of course use this for time-domain simulation later with FFTs.
Second is that that is simply not how the tool workflow is. To get the full reasoning I would have to write out a basic introduction to EM simulation and RF electronics, which I'm not really in a mood to do.

Usually, the simulation of the PCB behavior is disconnected from the components that are integrated into it. You would first simulate the PCB, and then in software say 'To this passive network model, I connect this here and that here'. You can do something similar to what you mean with ADS, as it can do hierarchical simulation of both schematic (= what components are connected where) and PCB EM, but in the background it just does this separation for you, and only simulates the EM structure if there are any changes on the PCB layout.

As an example, here is how I simulated capacitor layouts. First I made the design in Altium (not shown). I then used the ODB++ format to export the PCB files to ADS, where I imported them into layout. I then did an EM simulation of the layout, adding ports where I would have components (ports are connection nodes). I then place a block representing this PCB network in a schematic view, and attach the capacitors I want to attach and ports (I got the models for these capacitors from the manufacturer). I then simulate the schematic which will show me how the PCB and capacitors behave. Right now I'm in the frequency domain since that is what I'm interested in, but you can also do time-domain simualtions in ADS and other tools.

(Note that this is not meant to be a how-to guide, just trying to give you an idea of workflow).
The best part about magic is when it stops being magic and becomes science instead

"There was no road, but the people walked on it, and the road came to be, and the people followed it, for the road took the path of least resistance"
 
The following users thanked this post: aussie_laser_dude

Offline DaJMasta

  • Super Contributor
  • ***
  • Posts: 2393
  • Country: us
    • medpants.com
Re: Finite element Transient Signal Analysis
« Reply #5 on: April 18, 2020, 05:36:59 pm »
Transmission line theory is a good direction, looking into complex impedance and what factors into it (RF electronics) is going to cover most of the basic principles, and there are tons of simulators of varying complexities, you can probably even get a decent model made in ltspice using lumped element modeling.


Basically, the parasitics of the physical construction of the layout all contribute somewhat to the circuit's performance.  At DC, the resistance of the longer trace is present and would be measurable with a high resolution meter, but would only be minimally important.  Inductors and capacitors are effectively just shorts and opens respectively at DC, but their characteristics present themselves when looking at changing (AC) signals.  The same goes for parasitics - at DC every material has resistance, but at AC every material and circuit element also have inductive and capacitive parasitic elements, and as your frequency of interest increases, these parasitic elements are a larger and larger part of the total system performance.  With the layout you have, a DC constant signal is just the performance of the resistor, a 100Hz AC signal would be dominated by your capacitor, and probably beyond a couple of MHz its performance would be dominated by the added inductance of the extended trace - maybe even into the 100s of kHz region given that the cap is big and the stray trace length is significant.  Above that you will run into resonances based on track length and geometry, your capacitor will start moving towards arbitrarily high impedance because of its own inductive parasitics from its size and construction, and the performance of the circuit as a whole will be quite drastically different than at low speed.

Even with a strong math background, I'd encourage building something like that up feeding it a signal, and measuring.  Measurements probably won't match your early modeling attempts, especially at high frequencies, but it will give you and idea of how drastic these impedance effects can be.
 
The following users thanked this post: aussie_laser_dude

Offline dmendesf

  • Frequent Contributor
  • **
  • Posts: 342
  • Country: br
Re: Finite element Transient Signal Analysis
« Reply #6 on: April 19, 2020, 09:19:05 pm »
If you want to approach this from the physics point then all said now is valid. If however you want to see how it's done from an electronics engineering point of view (by constraining everything to the choices that matter from a "let's make a PCB that works for these devices") then look at IBIS simulations, like https://www.google.com/url?sa=t&source=web&rct=j&url=https://www.st.com/resource/en/application_note/dm00257177-highspeed-si-simulations-using-ibis-and-boardlevel-simulations-using-hyperlynx-si-on-stm32-mcus-and-mpus-stmicroelectronics.pdf
 
The following users thanked this post: aussie_laser_dude

Offline aussie_laser_dudeTopic starter

  • Contributor
  • Posts: 46
  • Country: au
Re: Finite element Transient Signal Analysis
« Reply #7 on: April 19, 2020, 11:49:56 pm »
Quote
If you want to approach this from the physics point then all said now is valid. If however you want to see how it's done from an electronics engineering point of view

Looking to learn all levels and types of approaches from fundamental physics to actual engineering practices.

I'd like to be able to get skilled enough to do these two (difficult) goals:
1. Develop software the can simulate how circuits A and B handle different applied voltages, from DC to MHz all the way to high GHz. Something like a 2.5D FE Maxwell simulator.

2. Gain an intuitive understanding of pcb inductance / capacitance / transmission lines etc that matches simulations

3. Actually make the circuits, measure and confirm everything matches.
« Last Edit: April 19, 2020, 11:52:10 pm by aussie_laser_dude »
 

Offline Mechatrommer

  • Super Contributor
  • ***
  • Posts: 11714
  • Country: my
  • reassessing directives...
Re: Finite element Transient Signal Analysis
« Reply #8 on: April 20, 2020, 01:08:05 am »
sorry to break the party but i guess thats going to be a long way and thick money..

1) try to get the feeling of the mentioned FEM and FDTD.. those space splicing and tensors connection algorithm can be applied to any theory you already know, be it maxwell, newton or molecular energy theory...
2) build rf pcb, buy/rent/borrow rf equipments such as VNA or TDR and measure. sounds simple.
3) apply 1 and 2 and repeat/rectify/recode/relearn many times..

btw, recently tried to characterize few simple pcb sections behavior (3D import) in HFSS, the results are too good to be true... the pcb (longest trace) that i made and verified with real equipments (DSO) is far from it... maybe i did some newbie mistake but i didnt see much options.. its just that too easy to use... export the S12 gain and phase and apply/process to the FFT of ideal square signal, (using external stand alone diy SW) and doing inverse FFT to get the expected output signal based on S12 profile given by HFSS resulting in.. too good to be true result...  :-BROKE
« Last Edit: April 20, 2020, 01:12:30 am by Mechatrommer »
Nature: Evolution and the Illusion of Randomness (Stephen L. Talbott): Its now indisputable that... organisms “expertise” contextualizes its genome, and its nonsense to say that these powers are under the control of the genome being contextualized - Barbara McClintock
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Finite element Transient Signal Analysis
« Reply #9 on: April 20, 2020, 03:23:55 am »
Parts of those things are easy enough to do, but applying them properly and consistently requires all the other things (i.e., "draw the rest of the fucking owl").

For example, trace inductance and capacitance are simply proportional to length.  When it's a simple transmission line structure (a signal trace surrounded by ground), and when what we mean by "inductance" and "capacitance" are the low frequency asymptotic equivalent impedances.

If we model at a similar resolution as the geometry, we should use transmission lines instead of RLC equivalents; but your screenshot shows floating copper, which will couple to the trace on different sides.  So, a whole bunch of transmission lines are going on there already, plus finer details we can't model so easily with just a few transmission lines.

Which is a modelling process, that works the same way that a general impedance might be modeled with a finite number of RLCs, for some degree of fitness.

Which in turn is underlain by mathematical theory of analysis and approximation.

Which means, you can very quickly get into the entirety of an EE curriculum's math offerings (calculus, numerical analysis, statistics?, EM, DSP?) without digging very far into this subject!

Do take confidence, I guess, that your question is not at all insurmountable; answers are not only possible, but mass producible.  Producing them yourself will take a lot of work, but you can get there.

Aside: I recall a thought from Richard Feynman, that he would often contemplate approaches to a simple toy problem in wave mechanics.  Consider a rubber ducky floating in a pool, in which there are many people jostling about, absorbing and creating waves.  The rubber ducky is floating in one place, more or less; given that this is so, can you solve for the entire state of the pool (the waves, positions, directions and amplitudes)?  Can you solve for how many people, and what positions?  Pool size and shape?  (In short, all the boundary conditions.)  How long do you need to observe, to have such-and-such confidence in your answer?  If it is in fact impossible to solve for some of these conditions, why?  What would be the minimum addition to the system to be able to do so?  (For example, the altitude of the rubber ducky might not be sufficient information, but including its angle -- the angle of the waves as it rides over them -- might be.)

Tim
« Last Edit: April 20, 2020, 03:32:22 am by T3sl4co1l »
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline Mechatrommer

  • Super Contributor
  • ***
  • Posts: 11714
  • Country: my
  • reassessing directives...
Re: Finite element Transient Signal Analysis
« Reply #10 on: April 20, 2020, 04:07:45 am »
but your screenshot shows floating copper, which will couple to the trace on different sides.  So, a whole bunch of transmission lines are going on there already, plus finer details we can't model so easily with just a few transmission lines.
if you are refering to my screenshot, the yellow colored are copper traces, the one that touches cyan small box (excitation) is the path needs analysis, the rest are dummy. there is smd resistor and capacitor in between set as lumped RLC elements boundary. the orange color is the gnd with dense vias (surrounded by FR4) and solid ground plane underneath, directly connected to simple model of BNC connector with another excitation box on its output end. both excitations barely touch both signal path and gnd plane. gnd plane is set as reference node for excitations and also set as perfect E boundary. if i remove the perfect E boundary condition, i got different result. if i remove the "gnd plane as reference", i got another different result. there is very little explanation available whats going on with those parameter, i guess i will need a bit of know what and how an EM solver is. but all results are still too good to be true... in real life, there is alot of spectrum lost. the real life pcb only can be considered as 1GHz BW but the HFSS solver indicated they are acceptable beyond 10GHz  :-// :palm: behind this pcb section is nothing high speed, just low speed digital or i2c signals, regulator to the IC that generates signal to the analyzed signal path and many other smd components, i dont think they will matters, if they do, this will be a nightmare to both em modeller and pcb designer. and calculation to solve it will take much more time to complete.
Nature: Evolution and the Illusion of Randomness (Stephen L. Talbott): Its now indisputable that... organisms “expertise” contextualizes its genome, and its nonsense to say that these powers are under the control of the genome being contextualized - Barbara McClintock
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Finite element Transient Signal Analysis
« Reply #11 on: April 20, 2020, 05:42:52 am »
OP's example; but that too.  Note that at frequencies this high (getting close in the 10s of GHz, on up to the 100s GHz that TheUnnamedNewbie has worked with), even via lengths and trace widths matter, so even (1-dimensional) transmission line methods become untenable and you are much better off with full field simulations.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf