# Difference between RF and Signal Processing Filters?

1. Jan 15, 2012

### Physineer

I've designed both types of filters for years and can't really explain the difference in clear simple terms. I'm hoping somebody here can help me out with that. First, some working terminology:

RF filter
- a filter with a controlled input and output impedance, usually 50 ohms, to match the characteristic impedance of the transmission lines feeding it

Signal processing filter
- a filter with a high input and low output impedance, can be an active filter, usually used at frequencies below RF (e.g. audio)

A good way to answer the question is in that "intermediate" area, say at 50Mhz. At this frequency both types of filters are practical. Of course cost is always a consideration but the heart of this question is why one would be needed as compared to the other.

There are a lot of subtle things to consider, which I think is the reason I've struggled to come up with clear, concise answer. My replies will only serve to point out some of the subtleties that get overlooked in overly simplistic answers.

2. Jan 15, 2012

### yungman

I don't know your meaning of signal processing filter. When I see signal processing, I think digital filtering already!!! Do you mean lump element( using inductors, capacitors, opamp etc.?) versus RF filters using transmission line elements, couple lines etc.?

That's mainly how I see it in filter designs. At low frequency, you use lump elements because the λ is so long make using transmission line impractical. At high frequency, any active electronics introduce addition roll offs and varying impedance with frequency. Also parasitic start to take over when using lump elements.

It is easier to design filters that don't need impedance matching at low frequency or have a lot more option in input and output impedance ratio as you see in lump element filter table. In RF, impedance is everything.

I don't think is that practical using transmission line or couple line filter at 50MHz.

3. Jan 15, 2012

### Physineer

Hi yungman, thanks for the reply. Rather than replying point by point, I think I'll highlight the one comment that, if followed through on, has the best chance of getting to the bottom of it:

Can you summarize, as concisely as possible, why "impedance is everything" in RF in your view? This principle alone, which I agree with, if properly stated, could serve to explain the difference.

(Several of your other questions and comments get into the subtleties I was referring to initially, which tends to produce a lengthy explanation. Again, I'm looking for the simplest possible explanation.)

4. Jan 15, 2012

### yungman

The first and foremost, it is hard to get RF power, gain of each stage is quite low compare to low frequency circuits. You have no problem gain a gain of 100 out of a low frequency stage. Where as for RF amp, gain is expensive and you have to maximize the power transfer......ie matching impedance.

Second thing which is as if not more important: at RF stages, because of parasitic, you have feed forward and reverse, meaning what happen at the output can feed back to the input and affect stability. That's the reason they use S parameters that has S12 and S21 to represent the output to input and input to output respectively. This make matching very important as any mismatch in impedance will cause reflection and bounce back to the former stage. In low frequency where λ is so much larger than the physical size of the components, you don't really worry about this kind of stuff. You put an opamp in between two stages, you pretty much isolate the two. But stability of RF circuits depends heavily on the matching of the impedance.

To get a little ahead, we not only draw the S11 S22 on the Smith Chart, we also draw the input and output stability circle at various frequencies that show the region of the chart that will cause instability and should avoid it. This is all about input and output impedance.

For low freq circuits, you don't worry about any of these. You take a signal, you make you input input impedance high so you don't load the signal down. You buffer the signal and go through filter and wala, you got a nick filter!!!

5. Jan 15, 2012

### Physineer

Thanks again yungman!

I asked about filtering, where the gain is typically less than one (even in the passband, due to insertion loss), specifically to avoid this consideration. I think it will help keep the explanation simpler. So, even when G<1, impedance matching is still important. Why (or under what conditions)?

All true. But why at lower frequencies (e.g. 20Mhz) is a large mismatch, which still causes reflections, acceptable? For example, a 20Mhz driver (source) might have a real impedance of less than 1ohm. A 20Mhz filter could have an input impedance of >1Megohm. The transmission line (circuit trace) between them is typically in the 10-100's of ohms range. Those are large impedance mismatches.

It depends on the trace lengths that connect the input and output of a filter to the other stages, that determines whether signal reflections, of a given frequency, will be an issue. What specifically is the criteria that determines whether or not this will be the case? Why?

Let's get back to what I asked about initially, the "in between" frequencies (e.g. 50Mhz). Everything you wrote above is just at true for an "RF" circuit at 50Mhz as a low-frequency circuit (e.g. audio). What I mean is that op-amps are available that operate into the 100's of Mhz, so why not use the same approach for a 50Mhz RF filter as used in low-frequency filtering if it's so superior?

6. Jan 15, 2012

### yungman

50MHz is not low frequency by any stretch but RF really don't come in mind at this frequency, it is easy to get opamp to do amplification. At that frequency, you better worry about termination if structure is larger than 1/10λ. By RF today, we talk over 1GHz. If your RF means 100MHz or under, things get a lot easier and don't even worry about transmission line elements, you can forget about what I said about S parameters.

Last edited: Jan 15, 2012
7. Jan 16, 2012

### Physineer

Hi yungman, thanks again. We're starting to whiddle it down.

Your primary criteria seems to have come down to whether or not the structure is larger than 1/10λ. So let's drill down on that.

If it's less, than reflections aren't a problem - why not?
If it's more, than reflections are a problem - why?

In other words, let's delve into what happens when reflections occur and to what extent performance is impacted based on the magnitude of the impedance mismatch and the length of the structure.

8. Jan 16, 2012

### yungman

I should re-phrase, it is not that there is no problem, but it is not as critical if the structure is less than 1/10λ. First let me show you a simple example of the impedance change at one end of a transmission line when the other end is terminated with a fixed impedance.

$$Z_{in}(l)=z_0\;\frac{Z_L+jZ_0\tan \beta l}{Z_0+jZ_L\tan \beta l}\;\hbox { where } \;l \;\hbox { is the length of the line.}$$

As you can see, when the length of the tx line get longer compare to the wave length, the input impedance start to change. If you structure is long, the dimension of the connection become a transmission line and start change the value. You don't get the same as you expect.

As for the termination of the transmission line, any miss match will cause a standing wave along the line. The amplitude will change along the line. The more miss match, that more drastic it will change. You need to look up standing wave pattern as this is getting into the heart tx line theory. You'll find this information in engineering EM books.