# How are the positions of matching network elements determined for RF power amplifiers?

I asked an electrophysicist and he gave me a wishy washy experimental answer. Let me provide an example.

Let's say we want 50 ohm input and 50 ohm output from some particular wideband amplifier. The amplifier is not internally matched. I have full design freedom in terms of PCB traces and component selection. I also have access to network & spectrum analyzers.

I'm afraid of the answer being: "Hook it up to a network analyzer and season to taste" I've completed my undergrad and this type of designing was never mentioned. Assume I don't have the S parameters of the IC.

For reference, assume I'm using this transistor tuned for the 2.4GHz band

• What do you mean by "wide-band"? Do you just mean you're operating at high frequencies, or do you mean that your signal band actually spans close to or above an octave? Jul 25 '13 at 0:32
• The latter as opposed to the former. Frequencies at which discrete components of medium size (0805) can be reasonably used Jul 25 '13 at 17:17

Well, to some extent, the answer is "Hook it up to a Network analyzer and season to taste". Except that for a 20 Watt output transistor, you can't hook it up to a network analyzer unless you feel like letting that magic smoke out of the network analyzer.

In general, the procedure is to select the transistor, design the bias circuits and DC blocks/AC Shorts so that it's biased up at the right bias point, and then get to work. You should start out with the datasheet input and output impedances, and match those to 50Ohms. That particular data sheet doesn't have the S-Parameters at 2.4 GHz, but I bet the FAE could put you in touch with them. (I'm assuming you know how to compute an impedance match)

Once you do that, put that match on the board, and use a signal generator, directional coupler and power meters on the input/output to measure the input and output VSWR, and a Spec Analyzer on the output to measure the output power, and start tweaking it into shape. Part of the problem you'll have is that the S parameters will shift as you drive the amplifier more and more into compression. of course, you want to drive it into compression, because that's where it operates the most efficiently. On the other hand, P1dB is also where the linearity starts to fall off, so it's a fine line to walk with pushing it into compression while keeping the linearity you need.

That's the way I know how to do it, but if you have an Agilent Large signal network analyzer and a Load pull setup, you can do a much better job. Googling around for Power amplifier design tips should find a bunch of good papers, look for things from RF Design Magazine, High Frequency Techniques, and Microwaves and RF Magazine.

caveat: my experience is not so much in rf as in high-speed digital. However high-speed digital is a wide-band problem, because if you're actually switching a voltage on and off at 1 or 10 Gb/s your signal bandwidth is pretty much the whole spectrum from ~0 up to the bitrate, in Hz.

If your signal band is close to or more than an octave, I think you will want to find a design where the positions of your components is not critical.

Techniques like tuned stubs that use a component's position to modify its effective impedance are generally only useful in narrow band systems, meaning where the signal band is substantially less than an octave.

At the very least, these techniques get very complex as the signal bandwidth gets into the octave range, because several tuned elements need to be used together to achieve uniform behavior over the signal band. Which is not to say that such a thing might not be done in certain circumstances, when it's really needed. To do a design like that, if you're not already experienced with similar problems, you'll probably want to spend some quality time with an rf simulator (ADS, for example) before starting to build something. And then use the network analyzer to verify your design once you have built something.

In the wideband systems I'm familiar with, you normally try to simply keep the matching components close enough to the load that the stub leading to the load is electrically short (less than $\lambda/8$, say).