empirical for an existing element or theoretical for a design in progress?

Sonnet or AWR has tools for the EM models of a design in progress.

Empirical testing? Have to design a test platform with appropriate balanced paths and calibration for a testing method. Somewhat similar to a test setup for a filter inserted into an RF path.

Transformer's insertion loss can be considered as an independent parameter during initial PA design phase. It only hurts your PAE, but does not affect impedance transformation behavior. One simple way is to set IL=0dB in the first place and go back to run with EM model.

Hi,


after thinking on it more, I don't think MAG is very useful for transformer design because if I understand it correctly, it is not possible to achieve MAG for a lossy two-port passive device. You could argue that it is just a FOM of course...


cheers,
Aaron

Hi rfcooltools.com,


when you say "max available gain", are you referring to the definition of Gmax with a simultaneous conjugate match at both ports? I don't know for sure, but I think this is not possible to achieve Gmax in a passive device when loss is taken into account. Take the simple case where the DUT is a series resistor. I cannot see any way to match this device at both ports using passive components. That's why I think using Gmax as an FOM for a XFMR is a bit arbitrary.


cheers,
Aaron

Hi Aaron,

I think Gmax is usually used for transformer characterization, or as one of passive device merits of figure. When using XFMR for design, T or Pi model can be more useful to gain more circuit insight.

I designed a few CMOS PAs with integrated or off-chip XFMR. I was pretty much following the old-time PA design methodology : Use load-pull technique to do all the followings : 1. Power match to maximize PA output power 2. PAE match to target at power-aided efficiency 3. For linear systems (especially OFDM ones) linear match is also one of the constraints to check AM-AM and AM-PM.

In so doing, I worked on load-pull analysis (after a few some basic checks on P1dB, AM-AM and AM-PM, stability, etc.) for PA core. excluding XFMR, which I consider as part of PA output matching netowork in the configuration I am figuring. The partition is arbitrary. I had included on-chip XFMR as part of PA core, the result was the same, as it should be. But the parasitics  of PA core can play a major role in matching optimization especially when dealing with > 5GHz operating frequency.

So, back to you question : I normally don't worry about transformer characteristic precisely from initial design phase; only the first order model parameters such as turn ratio, magnetically coupling coefficient k, rough L/C parasitic values and so on you will deal with. In the mean time, transformer can be designed to meet your electric and magnetic needs with the best achievable insertion loss (determined mainly by substrate rho, resistive loss, low magnetically coupling coefficient, etc.)

In final stage of design, surely we need to put everything together to see if the result is good.

Hope the above is useful to you.

cheers,
inductioner

Here is the method used by the rhombus industries must be useful for you.
Follow this link "http://www.rhombus-ind.com/app-note/fr-il.pdf"

aaron_do wrote on Aug 2^nd, 2011, 6:56pm:


Hi, for the series resistor Gmax would be just 0dB, but the impedances for load and generator would become infinite (not so special in a Smith chart ). So Gmax is just one FOM, but not the most important one, in your application Gmax may not matter much, because you are limited to a certain range of impedances. Therefore available gain matters more.

Bye Stephan