4
\$\begingroup\$

I currently working on a RF design, with frequencies up to about 2GHz.

What is the "rule of thumb" for when to stop using lumped components (Capacitors, Inductors, Resistors), and start using microstrips ect. ?

P.S. I can see that lumped components are available up to about 10GHz, but they are pretty expensive.

Edit: What about decoupling capacitors, on supply lines to pll's ect., should they always be associated with RF Chokes to work probably (at 2GHz) ?

\$\endgroup\$
  • \$\begingroup\$ Can you tell me what kind of design you are doing? This has a lot to do with it, often it is a pick based on what you need and which will be smaller to construct it with. If you have a low enough frequency that you need the space saved by buying lumped components, you will do it, if you just need a cheap product, forget the size, your answer is clear again. If it is smaller for PCB, go for it. \$\endgroup\$ – Kortuk Oct 29 '11 at 9:40
  • \$\begingroup\$ I'm working on a homebrew spectrum analyzer, build mostly using complete blocks. \$\endgroup\$ – JakobJ Oct 29 '11 at 11:00
  • \$\begingroup\$ What frequency range are you planning to support? \$\endgroup\$ – Kortuk Oct 29 '11 at 11:43
  • \$\begingroup\$ Up to 1GHz. Which means I internally will have up to about 2GHz \$\endgroup\$ – JakobJ Oct 29 '11 at 12:59
  • \$\begingroup\$ Range needs a lower bound, what is it? It makes a big difference on your options. \$\endgroup\$ – Kortuk Oct 30 '11 at 7:26
2
\$\begingroup\$

The trouble with discrete parts is that they come with parasitics; e.g. an inductor will have a parasitic resistance, parasitic capacitance, and parasitic capacitance to the board. Generally speaking, as you approach 1GHz, parasitics start affecting your frequency response, even on small SMD components. Even if you buy "10 GHz" components, they'll have parasitics. The hard part of RF design is modeling the parasitics accurately (and knowing where they can be ignored). Also, there's a trade-off between precision and cost.

Microstrip elements are more abstract, so you generally have more control over what's going on. As others have mentioned, though, they are bigger.

Modeling components is time-consuming. Keep in mind, though, for a one-off project, you might find that tweaking the circuit is enough to get it to work. Repeatability over hundreds or thousands of units is a different matter. For this spectrum analyzer project, you don't need to worry about the 10.7 MHz IF blocks, but you will have to be careful with the 1-2GHz stuff. The bandpass filter looks like the most difficult part.

| improve this answer | |
\$\endgroup\$
  • \$\begingroup\$ Okay, thank you. For bandpass filter I've found a good saw-filter (fc=1090MHz), which I'm gonna cascade two times. I just can't see the advantage of sweeping the LO up to about 2GHz, instead of down? (Seems to me the mirror frequencies is not gonna do trouble anyways) \$\endgroup\$ – JakobJ Oct 31 '11 at 8:38
5
\$\begingroup\$

I'm not an expert in RF, but my guess is that the following tradeoffs come into play:

lumped components:

  • you don't have to design them
  • as long as they're physically much smaller than the wavelength, they should work well
  • except for high-power or high-voltage, should be available in 0402 or even 0201
  • if you need to change frequencies, you can do so by changing the BOM + can leave the circuit layout unchanged

microstrip elements:

  • tolerance is controlled by geometry, so can be much better than lumped components (esp. relative tolerances) with higher-Q/sharper rolloff
  • you can do all sorts of things you can't do with lumped components (e.g. resonators, quarter-wavelength transformers, directional couplers, etc)

Wavelength is definitely the key -- 1GHz has a wavelength of about 30cm in air (less in a high-K material like FR4). I've heard various rules of thumb like lambda/2pi, lambda/10, lambda/8, but if your board traces have dimensions of >3cm you probably need to deal w/ transmission line effects for signals > 1GHz.

| improve this answer | |
\$\endgroup\$
  • \$\begingroup\$ lambda/2pi is the one I have gotten from my most reputable sources, but that compared to 8 or 10 is so close it really does not matter. The big thing people forget is if you do not treat it as a tranmission line you still need to take into account its inductance and capacitance as one big term. \$\endgroup\$ – Kortuk Oct 30 '11 at 10:32
1
\$\begingroup\$

There really isn't a hard and fast rule.

The Advantage of lumped components is that they are smaller at these frequencies, so your board area will be smaller. If you go to microstrip elements for filters etc, you should also use a controlled impedance board, otherwise you're in for a bunch of hand tuning.

As far as 50Ohm traces, that's good practice to use those to interconnect circuit blocks, but unless the trace length is greater than 10% of the wavelength, you probably don't need it.

| improve this answer | |
\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.