Modeling Diode as Varactor in AWR MWO

Dear kkeefe80,

kkeefe80 said:

This doesn't make sense to me, since those curves appear pretty straightforwardly logarithmic and shouldn't take more than a 3rd or 4th order polynomial to approximate accurately. Has anyone else run into this problem with the VRCTR model, and if so how did you remedy it?

First, I must confess that I am not an AWR user, so please have patience with me if my comments are not relevant kkeefe80! However, I have a lot of experience modeling varactor diodes. Hence, when I read your post, it immediately brought some memories to mind, and I finally decided to run a few simulations and send along my thoughts if they are helpful...I hope they provide you some insight!

Using a polynomial to model a the capacitance-voltage characteristic of a biploar or MOS varactor in a numerical simulator is, in my experience, a poor choice. The model is not based on the physics of the device and hence will only be valid over a limited range of data. Further, the slope of the characteristic may not show the same behavior as the slope of the actual device. When used in a closed loop simulation, both of these can cause convergence issues in the simulator or just inaccurate simulation results. Of course, this is only my experience.

To better illustrate these two items, I captured your posted graphic containing the C-V characteristic of the bipolar varactor diode you wish to model. I extracted the data from the plot and did two curve fits: a seventh order polynomial (which I believe you did using the AWR VRCTR model) called Case 1 and a second model based on the expected physics of the diode (Case 2). In both cases, I illustrate the match of the models to the data from your plot, the error of each fit, and the degree to which the derivatives of each model matches that of the data extrarcted from your plot.

I've placed a summary of the effort at URL:

https://www.dropbox.com/s/bd2kcyt3mh369ud/bjt_varactor_c_v_curve_fit_sml_082022v1p0.pdf?dl=0

My view is that the use of a more physical model of the varactor will provide the greatest probability of providing a robust simulator model. In essence, I think it will provide few, if any, convergence issues and a more accurate solution if the voltage range in simulation exceeds the range of the extracted data.

kkeefe80 said:

I'm also considering the PNCAP model, although it's not exactly clear to me how to set its CJ0, VJ, and FC parameters to exhibit the curve that I want.

I did not realize it until I did a AWR search, but it appears the PNCAP model shown in Figure 1 is a physical based model and has the same set of parameters I used in the "physical model" curve fit I performed. Hence, I might suggest you try using the 3 model parameters shown on page 7 in that model. If you have time and patience, perhaps run your simulation using the PNCAP model with these parameters and determine if your simulation results are more to your liking.

I do hope his helps a little kkeefe80.

Shawn

Figure 1 (from https://awrcorp.com/download/faq/english/docs/Elements/pncap.htm)

Shawn,

For starters, your answer goes above and beyond what I'd expect from someone volunteering their free time to help me, so thank you for that.

Unknown said:

Using a polynomial to model a the capacitance-voltage characteristic of a biploar or MOS varactor in a numerical simulator is, in my experience, a poor choice. The model is not based on the physics of the device and hence will only be valid over a limited range of data. Further, the slope of the characteristic may not show the same behavior as the slope of the actual device. When used in a closed loop simulation, both of these can cause convergence issues in the simulator or just inaccurate simulation results. Of course, this is only my experience.

That makes total sense to me; I've had more luck in the past using logarithmic or power functions to model a varactor curve. I'd been unaware of the formula specific to varactor curves that you used in Case 2 of your data collection, though it does appear as if that will be the most effective method going forward. Further, that leads me to wonder why AWR's VRCTR model wouldn't use this formula / set of parameters, but as you pointed out the PNCAP model does, so I guess it's not of any ultimate consequence.

After following similar steps to yours, I was able to get much more sensible results in AWR.

Thanks again,

Kyle

Dear kkeefe80,

Thank you for the update on your progress and thoughts! I am just happy to read you found my comments somewhat useful!

kkeefe80 said:

Further, that leads me to wonder why AWR's VRCTR model wouldn't use this formula / set of parameters,

I agree with your comment Kyle. I was surprised to read that they offered a polynomial curve-fit option as, at least in my experience with MOS and bipolar varactors, their large-signal C-V characteristics are not well approximated by a relatively low order polynomial. Perhaps the motivation is for the use of the model in a small-signal simulation where only perturbations about the DC operating point are considered.

kkeefe80 said:

After following similar steps to yours, I was able to get much more sensible results in AWR.

Excellent! I am glad you are all making forward progress - good luck!

Shawn