Maybe a picture would help. I suggest gate shorted to source and sweep the drain voltage plotting Id, I don't see the B parameter doing anything and the curves are still fit as before.
IMO -- unless you are going for the record in low thd, the distortion reduction due to cascoding will be minimal with high gnfb design. Better to spend time on PSR and CMR. Which cascode in a diff amp input will do better for those parameters? Thx-RNM
Nicely called, Scott, no cigar ...
Original results:
Altered test rig as suggested, and got:
From the LTspice help:
Frank
The second graph is not realistic. The sub-saturation curvature transitions abruptly to a linear function. Despite the promising description in Help, it doesn't seem to realize correct behavior.
So it goes.
Exactly ... so, at the moment, "no cigar", not good enough ...
I'll have a bit of a look around, and a play, see what might be possible ...
Frank
We may have some good news, people ... the Lambda parameter - channel length modulation - is crucial, and may have been set completely wrong in the published model. Playing with figures pulled out of the air I now get curves that look exactly like they're supposed to for the Ids vs Vds plot, but in the wrong place. So I'll start fiddling with other parameters, and see if I can get it to fully behave ...
Frank
Frank
Just to show how close one can go to matching the JFET curves, below is a grab from the BF862 datasheet, and LTspice's version thereof with modified parameters. There is still a problem unresolved, because the constant Vgs curves shown have been fudged, LTspice keeps insisting on equal spacing of curves between equal stepped values -- those shown represent other than Vgs of -0.1 through to -0.5. Which means that the curve of drain current as a function of gate source voltage doesn't match the datasheet.
This can be circumvented by adding a behavioural voltage at the gate, but maybe there's some fine tuning elsewhere that can be done ...
Frank
This can be circumvented by adding a behavioural voltage at the gate, but maybe there's some fine tuning elsewhere that can be done ...
Frank
Attachments
This is looking better, but there also still seems to be a kink at the transition. The problem for me was simulating the null of seconds for a simple phase splitter. It always walks down to the kink, where a real device has a sweet spot with a very deep null.
The level 3 model has an ALPHASP parameter that fits the entire curve smoothly with a tanh function which has a continuous derivative.
What did you end up using for lambda and B, and did B start doing something?
You might ask why something as fundamental as lambda is so far off in the extracted model.
The level 3 model has an ALPHASP parameter that fits the entire curve smoothly with a tanh function which has a continuous derivative.
What did you end up using for lambda and B, and did B start doing something?
You might ask why something as fundamental as lambda is so far off in the extracted model.
Last edited:
Been there! Done that! I'm interested in people who have DONE IT and interested in avoiding the pittfalls and interested in the circuit parameters of the SWOPA that I should be looking for; what optimum voltage, what optimum Idss of the BF862. They are probably all over the map out of a sample of 100. Thanks, Ray