Circuit simulation for an online forum

[SpacedCowboy]

Hi folks,

I'm getting towards the end (2-3 months out) of launching a forum sort of like reddit - but without any of the tracking, ads (more tracking), data-harvesting, or selling peoples data; in fact, none of the dark patterns that reddit has these days. One of the differences between blewit.net (because I think they did!) and reddit is that I want to provide really useful editors for different types of 'code' fences - where you type ``` insert stuff here ```. For some examples of these, look in the sandbox - no account or registration needed...

On blewit.net, and the reason I'm posting here, there's a circuit-editor where this text...

Code: [Select]

circuit title="A three-drop I2C bus" grid=12
# A three-drop I2C bus to experiment on. EVERY piece of it is a transmission line,
# the hop from the master to the pull-ups included, and SDA and SCL are the same
# bus span for span, so the two can be compared directly. Simulate it as it
# stands, then try these, one at a time:
#  - the pull-ups R1 (SDA) and R2 (SCL). Each only reaches its own line. Select
#    one and put 1k 2.2k 4.7k 10k 100k in its sweep field. The rise at the far drop,
#    measured 0.3 to 0.7 Vdd as the I2C spec measures it, goes 0.07, 0.16, 0.31 and
#    0.65 us against the 1 us standard mode allows. At 100k it never gets there at
#    all: 100k against the 76pF this bus carries is 7.6us, and a bit at 100kHz only
#    releases SDA for about 5us, so the line ramps and is pulled down again before
#    it arrives. That sawtooth is the bus failing, not the drawing misbehaving.
#    (The measure button reads 10-90% instead, the usual convention for a waveform:
#    2.197 RC against 0.847 RC, so the same edge reads about 2.5x slower there.)
#  - one edge, close up: set the window to 20.4u and the step to 0.2n in the fields
#    beside simulate. The far device rings to -3.59 V as the master pulls SDA down
#  - click the faint 220R at the far device to fit it, and watch BOTH sdaf (the
#    bus) and sdad (the pin behind it). The bus still rings: nothing at a device
#    can stop that. The resistor is a low-pass with the 10pF of the pin, so it
#    helps only while R x Cin is long next to the round trip of the line, about
#    10ns a metre. At 100mm, -3.59 V becomes -0.29 V for no cost in rise time.
#    Stretch that last span to 1000mm and 220R does little (-4.09 to -2.86 V): it
#    wants 2.2k. Scale R with the length, or it looks like it does nothing at all
#  - do NOT sweep that 220R over kilohms. In series with the open-drain pull-down
#    it makes a divider with the pull-up, so the far device stops being able to
#    pull the bus below the 0.99 V a low has to beat: it breaks the bus
#  - the 47p at the device only reaches -3.61 V and slows the rise to 1.28 us, so
#    a resistor is the better cure; fit both and the ring goes at that slower rise
#  - a line is lossless unless the wire is given a trace width (w=), so the ring
#    here is the pessimistic case: real wiring damps some of it away
#  - the far-end pull-up is there too, to compare with the pair at the master
rail VCC 3.3 4,-6
rail VCC 3.3 8,-6
rail VCC 3.3 12,-6
C1 capacitor at 4,-6 rot 90 C=100n
gnd 4,-2
R1 resistor at 8,-6 rot 90 R=4.7k
R2 resistor at 12,-6 rot 90 R=4.7k
#R3 resistor at 42,20 R=4.7k
#rail VCC 3.3 46,20
#wire 38,20 42,20
U1 i2cmaster at 0,2
U2 i2cport at 18,12 addr=0x20
U3 i2cport at 28,12 addr=0x21
U4 i2cport at 38,12 addr=0x22
#Rd4 resistor at 38,4 rot 90 R=220
#Cd4 capacitor at 34,8 C=47p
#gnd 34,8
wire 7,2 8,2 z0=50 len=20mm n=1
wire 8,-2 8,2
wire 7,4 7,20 12,20 z0=50 len=20mm n=1
wire 12,-2 12,20
wire 8,2 18,2 z0=50 len=100mm n=2
wire 18,2 28,2 z0=50 len=100mm n=2
wire 28,2 38,2 z0=50 len=100mm n=2
wire 12,20 18,20 z0=50 len=100mm n=2
wire 18,20 28,20 z0=50 len=100mm n=2
wire 28,20 38,20 z0=50 len=100mm n=2
wire 18,2 18,12
wire 28,2 28,12
wire 38,2 38,4
wire 38,8 38,12
wire 18,20 18,14
wire 28,20 28,14
wire 38,20 38,14
wire 26,12 26,9
label sdam 8,2
label sdaf 38,2
label sdad 38,12
label sclf 38,20
label a0 26,9
probe V(sdam)
probe V(sdaf)
probe V(sdad)
probe V(a0)
sim tran stop=1m step=0.1u

... inside a code-fence will produce this rendered into the forum:

And you'll notice the 'simulate' button, which will run an ngspice-a-like simulation to produce:

where each graph corresponds to a measurement point. To stop you from having to actually type the text-description out, you can also 'edit' the circuit, and you'll see something like:

You can start off with a blank circuit and just draw it, adding probes as you go. There's a whole load of documentation on how it works here which is also interactive so you can play with the parameters of things. Here's what's available:


The editor

• A palette of parts, grouped and searchable. Click to place, drag to move; wires follow the part.
• Wire, label, probe and delete tools. Rotate and flip.
• Select a part and its values are a dialogue, built from the part's own parameter list.
• DRC: a pin connected to nothing, a wire ending in the air, a probe on a net that doesn't exist, two outputs fighting, an open drain with no pull-up, a scope probe with its ground clip off ground. Click a finding and that part is selected, with its values open.
• Comment a part out and it is not fitted: drawn faint, out of the netlist, a two-pin part passing the signal straight through. Click it to fit it again.
  
• A part editor: draw a symbol, set its pins, give it a built-in model or a program.
• A library: save circuits and parts, and use them again.
• Once simulate is pressed, the plots follow every edit as you make changes.

The analyses

• DC operating point: every node voltage, every branch current, and the power each part dissipates. A resistor has a package (0402 to 2512, 1/8 W to 2 W), and one over its rating is flagged.
• DC sweep, for an I-V curve or a transfer characteristic.
• Transient.
• AC, as a Bode plot.
• Noise: output and input-referred density, and the total over the band.
• Monte Carlo: give parts a tolerance, run N times, get a histogram per probe.
• Temperature.
• Parameter stepping over a list of real values (step R1 1k 4.7k 10k), all runs on one graph.
  

The graphs

• Two cursors and a measure table: pk-pk, rms, rise and fall, overshoot, frequency, duty.
• A timebase, log Y, several signals on one graph, graphs set aside.

The parts

• R (with tolerance and package), C, L.
• DC voltage and current sources, sine and pulse sources.
• Diodes: generic, 1N4148, 1N4007, or your own Is and N. Schottky: 1N5817, 1N5819, BAT54. Zener. LEDs by colour.
• BJTs: 2N3904, BC547, 2N3906, BC557. MOSFETs: 2N7000, IRF540, BS250, IRF9540. N and P JFETs.
• An op-amp with gain-bandwidth, slew rate and output rails.
• Switch, voltmeter, ammeter.
• A scope probe, x10 (10 MOhm || 10.8 pF) or x1 (1 MOhm || 100 pF). It loads the node it measures.
• Logic: NOT, AND, NAND, OR, NOR, XOR, D flip-flop.
• Behavioural I2C and SPI masters and peripherals, so a bus can be simulated without modelling a microcontroller.
• You can define your own parts and add pin behaviour, including referencing memory and using logic operations to drive other pins.
• Wires can be transmission lines. Give a wire an impedance and a length, or a trace width over a board stackup, and it delays, reflects and rings.
• Paste a .model card from a datasheet or LTspice's library (D, NPN, PNP, NMOS, PMOS), + lines and all, and a part takes it. What the simulator doesn't model from the card, it lists over the result.
• Subcircuits: a .subckt ... .ends block, placed by a part you make in the part editor.

Some results from the examples

• I2C bench, a master and three ports on 3.3 V, every span a 100 mm line. Pull-up 1k / 2.2k / 4.7k / 10k: rise at the far drop (0.3 to 0.7 Vdd) is 0.07 / 0.16 / 0.31 / 0.65 us. At 100k the bus fails.
• The far device rings to -3.59 V on a falling edge. 220R in series at the device: -0.29 V, same rise time. 47p at the device instead: -3.61 V, and the rise slows to 1.28 us.
• Put a scope probe on that bus. The 0.30 us rise reads 0.34 us with a x10, and 0.70 us with a x1.
• A x10 op-amp stage with 1 MHz GBW, 1 V/us slew and +-12 V rails, given a 1.5 V step: ramps for 12 us, stops at 12 V. As an AC sweep, the corner is at 100 kHz.

How it's built

• Modified nodal analysis, Newton-Raphson with SPICE-style junction limiting, GMIN and source stepping. Trapezoidal integration, with backward Euler for two steps after a digital edge.
• Junction charge, reverse breakdown, series resistance and temperature in the diode model. Noise by the adjoint method: one extra solve per frequency.
• The solver is compiled to WebAssembly and runs in a Web Worker. Nothing is simulated on a server.
• Checked against ngspice-47: divider exact; diode I-V within 0.04%; RC low-pass gain exact to four decimals, phase within 0.0003 degrees; common-emitter 2N3904 bias within 0.09%.

What it isn't

• It's capped at 200 elements. It's for the circuit you're asking a question about, not a whole board. I do have a port of Sparse-matrix arrays as a soon-to-be-used feature which should remove this limitation. Bear in mind that this is a browser though, throwing a full board at it is probably just going to exercise your cpu fans.
• Lines are lossless unless the wire has a trace width.
• The op-amp has no supply pins. The logic and bus parts are behavioural.

What I'd most like to hear: what it gets wrong, which parts and models you'd reach for first, and whether posting a circuit someone can run and change is something you'd use.

[SpacedCowboy]

I guess I misjudged that this might have some appeal. Oh well, live and learn, I guess. It’s there, if anyone wants to use it

[JohnG]

Ok, I'll bite.

First off, it's an impressive piece of work for one person (with AI assistance?).

However, I won't give it further look for the following reasons, all of which are based on my opinions as a decades-long spice/Saber/Spectre user (in descending order of experience), primarily in the analog and power domains with a little RF thrown in.

-It's like  some version of Spice, but much more limited in the simulations it can do.
-Why would I type text of what I want to look at when I can do it directly in a schematic input more easily? I would rather have it output a text and graphic summary of what I am doing, because I often don't know ahead of time, i.e. I am exploring. It's the documentation that's the chore, not the exploration.
-It's not clear where the libraries are or come from, but they look very limited.
-Spice has a track record, your simulator doesn't. Why would I trust it?
-The limitations of solving circuits using spice simulation are about convergence, speed, and making good models of parts that don't exist in the libraries already.
-The limitations of using Spice as a design tool are the usually cumbersome and limited GUIs.
-I don't see a compelling reason to look at it further. I'm happy with Spice-based simulation, just wish it was faster and handled convergence better while maintaining accuracy, or had more sane library management and model-building tools.

I wish that these kinds of AI-assisted efforts could go to building on or improving already existing platforms, rather that recreating things just a little differently. I don't need nor want yet another circuit simulator. For example, if all this effort went into making the KiCAD/ngspice combo much better, I for one would appreciate this more.

These are free opinions, so you can assign whatever value to them you wish.

John

[SpacedCowboy]

Mm. I'm not sure we're approaching this from the same angle. I've seen many a time where people try to describe their circuit in words and ask for advice. Or they draw it on a piece of paper and photo it and upload the photo, or ... there are a legion of ways to do it, but they rarely go off and create something in spice and present the results...

This is, of course, schematic capture, you do just pick up resistors/capacitors/parts and drop them, then wire them. That's the bulk of the text that you're seeing - its the *results* of drawing the circuit and assigning tolerances or transmission-line coefficiencs. The point is that the text is the model, not that you have to type text.

So the idea is that you can draw something, then simulate, and if it still doesn't work, then you post and ask your questions (this is for a web-forum, not someone with decades of experience and their own spice installation). When it's posted, there's a "tinker with this" button that lets people copy and modify the circuit, so you can have a discussion - again, focus is on the forum aspect.

I'm not trying to, nor would I even consider to tell people to not use Spice - that would be nuts. In the same manner as "the best camera you have is the one that you have with you", though, I see some value in being able to interact with, and iterate on, circuits with a reasonable chance (every time I've run the simulator vs spice, it agrees to 0.0001) of the simulator telling you what's happening, in an environment that encourages that.

Perhaps that reframes things, perhaps you are still not convinced, but hopefully the purpose is a bit clearer.

Cheers - and thanks for the comments. In the same way as "you only learn something when you get something wrong", I think there's something to be said for "you only get to improve something when someone challenges you"

[JohnG]

Mm. I'm not sure we're approaching this from the same angle.

Yes, I think you are correct here. I'm putting my opinion there, right or wrong. However, I do this for a living, and am surrounded by the same. We have no problem collaborating and sharing or using the tools. The problems are more about non-convergence or long simulation times, cumbersome analytical tools or poor models. AI helps a lot by scripting things, etc, but it would be nicer if you could just drive the tool properly rather than gluing stuff together with a bunch of scripts. That's why I'd rather see something build on existing tools. I like open-source, that's why I suggest KiCad and ngspice.

Maybe this has a lot more value for others who are trying to learn, or need a circuit and they are the local expert whether the have the right background or not.

This is just one viewpoint. I'm sure there are others, so maybe they will post.

John

[ledtester]

I think having a schematic capture facility would be very helpful for an online electronics forum - especially if it was easy to use from phones and tablets.

Right now schematic have to be created in an external program which usually means using a desktop computer.

[ebastler]

Mm. I'm not sure we're approaching this from the same angle.

It seems to me that the angle you have chosen -- namely to make this editor part of a new forum framework -- is one of the hardest ways to let it gain traction. A forum lives and dies by the critical mass of users it attracts. Even if you provide a brilliant novel way to capture schematics, it will be of very limited use unless there are enough people on the forum who can comment on a posted schematic.

Assuming you have developed a better way (for some users) to describe schematics, wouldn't it make more sense to develop this as a front end for circuit simulation or PCB design programs? You could offer just the editor online, with file export and import in various formats. And with a rendering facility (or image export) which lets users embed the output into forum posts in already existing, popular forums -- eevblog comes to mind...

[Someone]

I wish that these kinds of AI-assisted efforts could go to building on or improving already existing platforms, rather that recreating things just a little differently. I don't need nor want yet another circuit simulator. For example, if all this effort went into making the KiCAD/ngspice combo much better, I for one would appreciate this more.

Or perhaps falstad:
https://www.falstad.com/circuit/
The human readable export is a little ugly compared to the OP's example but the intention is very similar.

[SpacedCowboy]

I think having a schematic capture facility would be very helpful for an online electronics forum - especially if it was easy to use from phones and tablets.

Right now schematic have to be created in an external program which usually means using a desktop computer.

I'm guilty of the same I think - it was developed on a desktop and only tested on a desktop. This is a really good point though, thank you

[SpacedCowboy]

Assuming you have developed a better way (for some users) to describe schematics, wouldn't it make more sense to develop this as a front end for circuit simulation or PCB design programs? You could offer just the editor online, with file export and import in various formats. And with a rendering facility (or image export) which lets users embed the output into forum posts in already existing, popular forums -- eevblog comes to mind...

To be fair, I don't think I've developed an easier way to describe schematics - it's still just {select a part, place it, wire it}. There's optional {configure it} steps to add multiple-variants {eg: 1k,4k7,10k} into a circuit to see what happens during simulation, or to add transmission-line parameters, or .. but the basics of it remain the same.

There are 2 benefits that I can see:

• It adds a transport, which is the text definition - that means you can cut/paste the same text in and get the same circuit, it also means you can "tinker with this" in the editor and you end up with a slightly different definition that's equally transportable. For a thread talking about "why doesn't this do X", being able to point out that the resistor is connected to the gate not the drain, like "this" seems useful to me, and that's driven by being able to easily clone/edit the circuit

• The number of people who are experts in spice is somewhat dwarfed by those who are not. Drawing a circuit, putting an ideal probe at the point you're thinking of and then looking at what happens is a pretty approachable introduction. Being able to put a modelled oscilloscope probe in, instead of the ideal one might go some way to explaining why "it doesn't work" on the bench as well - at least for some.

The big problem with merging into an existing forum is that you need quite a bit of access to how the forum works to do that. The simulator uses a WASM engine, compiled against Kundert's Sparse 1.4, using a c->xc converter which produced ~7100 lines of xc code from the ~14000 lines of C code. That's not a trivial thing to just integrate (even assuming a forum would let someone fiddle with their back-end, which seems unlikely), and that's just the matrix library. The UI setup and simulation engine are on top of that.

But, there's nothing to stop you doing just what you suggest - blewit just asks for a username/password, no details, not even an email. It's really easy to sign up and it's specifically data-minimal on its userbase. It doesn't ask for any user-information because it doesn't want to (and is set up to specifically *not*) track you, or target you with ads (tracking by another name), or use an algorithm to push you anywhere, or any of the dark patterns that currently infest social media.

So, log in, click on 'add a circuit', create one, and screenshot it with the simulation - what you'll miss is the first benefit above (someone else being able to advise by modifying the circuit and showing you) but you wouldn't have had that with the export either.

If you have suggestions for import/export formats, I'm interested to know which ones you'd prefer, and since we currently render as SVG I might add an image-export tool to prevent you from having to snapshot the screen (trivial on the Mac, not sure about Win/Lin).

[abeyer]

So, log in, click on 'add a circuit', create one, and screenshot it with the simulation - what you'll miss is the first benefit above (someone else being able to advise by modifying the circuit and showing you) but you wouldn't have had that with the export either.

But in that case why not just screenshot whatever schematic capture/simulation tool someone already uses?

I get why dealing with existing platforms is hard... but the platform integration seems to be the unique part of what you're doing, so feels like you're going to have to either figure out how to bring it to platforms that have a community already, or stop worrying about the tech completely and work on building a new community from scratch, which is an entirely different endeavor.

[Obin]

One of the things prohibiting me from using something other than KiCad is its speed and the fact I know all the hotkeys. If I now want to draw an schematic to explain something, it will take me less than a minute. And since you can always add images to forum posts, its not a hassle to add screenshots.

Learning a whole new thing just for a forum feels very 'meh'. Especially text based, and to then add a whole GUI, then we just arrive back at KiCad anyhow

[SpacedCowboy]

So, log in, click on 'add a circuit', create one, and screenshot it with the simulation - what you'll miss is the first benefit above (someone else being able to advise by modifying the circuit and showing you) but you wouldn't have had that with the export either.

But in that case why not just screenshot whatever schematic capture/simulation tool someone already uses?

The answer to that question is the first benefit I talked about above - the ability to quickly copy/paste/edit the circuit and show the changes, then (if simulation is the point) running the simulation will hopefully prove the change. You can't do that with a screenshot, no matter where it comes from.

One of the things prohibiting me from using something other than KiCad is its speed and the fact I know all the hotkeys. If I now want to draw an schematic to explain something, it will take me less than a minute. And since you can always add images to forum posts, its not a hassle to add screenshots.

Learning a whole new thing just for a forum feels very 'meh'. Especially text based, and to then add a whole GUI, then we just arrive back at KiCad anyhow

So, it's not text-based, it's a text-description, which means its trivial to store/transfer/copy/edit. The UI is very much a graphical UI.

Anyway, perhaps this will help: there's now a KiCAD import option, just click the import button and it'll open up a file-select box, find your circuit schematic(s) on your local disk and it'll import. If it can match the types to simulation-aware types it knows about, it will; it'll tell you which ones it can't, and it'll offer you the change to enter a spice .model description.

[ftg]

One of the things prohibiting me from using something other than KiCad is its speed and the fact I know all the hotkeys. If I now want to draw an schematic to explain something, it will take me less than a minute. And since you can always add images to forum posts, its not a hassle to add screenshots.

Learning a whole new thing just for a forum feels very 'meh'. Especially text based, and to then add a whole GUI, then we just arrive back at KiCad anyhow

This.
And if I don't have Kicad on the machine, I usually reach out for Digi-Key's Scheme-It for quick schematics or block diagrams.
No login and can export images directly, so no need to even screenshot it.
https://www.digikey.com/en/schemeit/project

[SpacedCowboy]

One of the things prohibiting me from using something other than KiCad is its speed and the fact I know all the hotkeys. If I now want to draw an schematic to explain something, it will take me less than a minute. And since you can always add images to forum posts, its not a hassle to add screenshots.

Learning a whole new thing just for a forum feels very 'meh'. Especially text based, and to then add a whole GUI, then we just arrive back at KiCad anyhow

This.
And if I don't have Kicad on the machine, I usually reach out for Digi-Key's Scheme-It for quick schematics or block diagrams.
No login and can export images directly, so no need to even screenshot it.
https://www.digikey.com/en/schemeit/project

No login means no save, presumably. And an image means no simulation or easy edits for others to help you out.

Did you see my post immediately prior to yours ?

there's now a KiCAD import option, just click the import button and it'll open up a file-select box, find your circuit schematic(s) on your local disk and it'll import. If it can match the types to simulation-aware types it knows about, it will; it'll tell you which ones it can't, and it'll offer you the change to enter a spice .model description

[ftg]

One of the things prohibiting me from using something other than KiCad is its speed and the fact I know all the hotkeys. If I now want to draw an schematic to explain something, it will take me less than a minute. And since you can always add images to forum posts, its not a hassle to add screenshots.

Learning a whole new thing just for a forum feels very 'meh'. Especially text based, and to then add a whole GUI, then we just arrive back at KiCad anyhow

This.
And if I don't have Kicad on the machine, I usually reach out for Digi-Key's Scheme-It for quick schematics or block diagrams.
No login and can export images directly, so no need to even screenshot it.
https://www.digikey.com/en/schemeit/project

No login means no save, presumably. And an image means no simulation or easy edits for others to help you out.

Did you see my post immediately prior to yours ?

there's now a KiCAD import option, just click the import button and it'll open up a file-select box, find your circuit schematic(s) on your local disk and it'll import. If it can match the types to simulation-aware types it knows about, it will; it'll tell you which ones it can't, and it'll offer you the change to enter a spice .model description

I did.
And Scheme-It does allow saving the schematic or block diagram. You just need to log in with your Digi-Key account.
My mention about no login being required was more in reference about the barrier to entry.

[SpacedCowboy]

And Scheme-It does allow saving the schematic or block diagram. You just need to log in with your Digi-Key account.
My mention about no login being required was more in reference about the barrier to entry.

Gotcha, same thing, then

If you're not logged in on blewit, the "sandbox" has a button to press [circuit diagram] on, click that, click [edit] to replace the default I2C bus with your own, and it's the exact same full editor (complete with KiCAD import) as the logged-in one. You just can't save (it'll prompt you to log in) because there's no name to save it under.