2 ms·

I made two diode-related online tools: First tool [1] is for creating spice model from measured voltage and current values. You pass 30 or so measured voltages

by dvh 4y ago

I made two diode-related online tools: First tool [1] is for creating spice model from measured voltage and current values. You pass 30 or so measured voltages and currents it it will generate spice model, for example: .model D1N4007 D(IS=2.24679e-9 N=1.76869 RS=0.647905) Second tool [2] is for finding similar diodes. Let's say you found your mojo diode and you want something similar. You measure 30 or so values from 1uA to 25mA and paste the voltages and currents into the tool, then click on "Similar 3" button and it will display the curves that are most similar to that. The only problem is that there are not that many diodes, so feel free to contribute your curves via github. [1] https://dvhx.github.io/spice-diode-model-js/index.html https://dvhx.github.io/spice-diode-model-js/index.html [2] https://dvhx.github.io/spice-diode-model-js/find.html https://dvhx.github.io/spice-diode-model-js/find.html

exmadscientist 4y ago
Nice visualization! But the optimizer could use some work. It wasn't really able to handle the data I had lying around for an 0402 hyperbright white LED (EAST1005WA0) even though this should have been in its ballpark. (Though this LED is particularly "soft" which might have been troublesome for the basic 3-parameter model?) If you want to improve things the change that will probably get the biggest benefit for the least work is to have the optimizer work in log Is instead of Is directly. (At least, that helped me a lot when I did the super-basic 2-parameter version of this in Excel for somewhat cruder purposes.) If you're up for heavier lifting, because this is only a 3-parameter model, you can probably write out an explicit or approximate explicit solution for 3 input data points, then expand that into a least-squares solution for N data points, then use the actual heavy machinery optimizer as the final correction step. It will run very quickly if you give it a good initial starting parameter set. That parameter set will also be a decent starting position for the more detailed models used by LTspice and friends, should you like to go there.

by dvh 4y ago

exmadscientist 4y ago

Nice visualization! But the optimizer could use some work. It wasn't really able to handle the data I had lying around for an 0402 hyperbright white LED (EAST1005WA0) even though this should have been in its ballpark. (Though this LED is particularly "soft" which might have been troublesome for the basic 3-parameter model?) If you want to improve things the change that will probably get the biggest benefit for the least work is to have the optimizer work in log Is instead of Is directly. (At least, that helped me a lot when I did the super-basic 2-parameter version of this in Excel for somewhat cruder purposes.) If you're up for heavier lifting, because this is only a 3-parameter model, you can probably write out an explicit or approximate explicit solution for 3 input data points, then expand that into a least-squares solution for N data points, then use the actual heavy machinery optimizer as the final correction step. It will run very quickly if you give it a good initial starting parameter set. That parameter set will also be a decent starting position for the more detailed models used by LTspice and friends, should you like to go there.