Spent the past two days working on tuning the RF portion of my circuit. I wasn't sure how it would turn out because it's the first time I tried to make a 2-layer RF board. If you haven't noticed, most of the reference designs are on four layers since the impedance is easier to control that way. It was also good because it gave me time to get to know my network analyzer and me and her are becoming best buds.
When I first started tuning, I was totally in left field. I had used the online transmission line impedance calculators and for a 70-mil microstrip on FR-4 with a 59 mil dielectric, I was supposed to be getting about 63 ohms for characteristic impedance. After tuning the circuit, I calculated the impedance of the receive path to be around 35 + j8 ohms. The imaginary component implies that its an inductive impedance, probably due to the length of the trace.
It was really bizarre because the real part of the impedance was about half what I expected it to be. It was a good lesson for me because I learned that you can't really trust math when it comes to real world performance of RF circuits. For the online microstrip calculators, they assume that you have a trace on the top layer and an infinite ground plane on the bottom layer. However in the case of my board, it was untrue. I actually had a ground plane on both top and bottom layers. I think the top layer ground plane, which was separated from the trace by just 8 mils, affected the impedance pretty drastically. Also, there were contributions from the balun, DC blocking caps, and the differential RF lines coming out of the chip. I suspect that's why my initial tuning values for my matching network were so far off.
It took me about twelve tries to converge on a good impedance match. It actually took me about 8 tries to see the LC resonant dip for the return loss and then an additional 4 tries to center it on 2.4 GHz. The final outcome was that the minimum of the return loss dip was centered at 2.46 GHz (channel 22) with a return loss of -40 dB or about 99.99% power transmitted. My worst return loss value was at 2.405 GHz (channel 11) which was at -16.1 dB or about 97% power transmitted. Still, not too shabby. Word to the wise, those RF inductor and capacitor kits from DigiKey really came in handy.
So this little experience took me pretty deep into the realm of RF design. I can see that RF calculations can only give you a very, very ballpark understanding of how your circuit will behave and that you actually have to measure it to know the real performance. In short, RF design is messy. Luckily, it seems like I survived the ordeal.
Here's some screen captures of the return loss testing with my network analyzer: