# Parallel transmission lines / TL switches

1. Oct 1, 2011

### msbe

If i wanted to make a 1:N coaxial RF switch (say ~100MHz, nothing too high), could you simply route N traces in a star pattern on a PCB and implement a mechanical switch to either connect to your measurement point, or terminate in the characteristic impedance (say 50 Ohm).

I'm getting thrown off a bit because I haven't come across much work on parallel transmission lines. The theory would be that if it's terminated in 50ohm it appears as an infinite length transmission line, i.e. has zero reflection. That would seem to imply that the signal would be split in power N ways. N-1 of them terminate in 50 ohms and have no reflection. The other one would go to the load and reflect back based on whatever the impedance seen is.

It seems like this is the design most RF switches use (absorptive switches that either connect to the output or terminate internally in 50ohm or 75ohm depending on the IC), albeit they use FET's typically for the actual switching.

Is this right? Or does connecting multiple transmission lines in parallel actually change the characteristic impedance seen going into the selected channel? I could see this since the power of the input signal is nessicarily lost due to the 50ohm terminations, but if the switches were open it would set up standing waves from the reflection.

Maybe the switch would simply present as an insertion loss by being terminated? So a 1:2 switch would reduce output power by 3dB since half of the power would go to a 50ohm termination?

Last edited: Oct 1, 2011
2. Oct 2, 2011

### yungman

Are you designing for single frequency or wide band? for single frequency, there are a lot of way to deal with this, for wide band, it gets a little tricky.

For wide band, you need to have all trace to individual switch in 50Ω transmission line. At the star point, you'll see 50/N ohm. So to avoid reflection and max power transfer, the input has to be in impedance equal 50/N ohm. Or else you are going to have standing wave between the source and the star junction.

You might be better off using N single pole double throat ( from input route to either A or B) in series. Say the first switch, A is the output1, B connect to the second switch. A of the second switch is output2 and B connect to the third switch where A is the output3 and B to the forth switch...................

With this, you only deal with one single path at a time by choosing which output to turn on and no termination on any of the output needed. You need to find RF switches of fet switch for 50 ohm.

Last edited: Oct 2, 2011
3. Oct 2, 2011

### msbe

Single frequency.

Ok thanks for that, yeah thinking about it now seems obvious the impedence will drop just like resistors in parallel.

If your input is 50ohm and say N=4, i wonder if you could match 3 channels with 200ohm, so input impedence would be 200/4= 50ohm.

Series idea is a thought, i'll have to think more about it, thanks.

4. Oct 2, 2011

### yungman

Single freq would be easier, use 1/4 wave line to transform from 50 ohm input to 50/N at the star junction. For transforming from 50 to 50/N using quarter wave line, the impedance of the line is $Z_0=\sqrt{50\times(\frac {50} N)}$ With this, you have 50 in and all 50 out. But if you working at 100MHz, the quarter wave is going to be long.

But I think the series switches works better because you don't have to worry about the unselected output termination.

Last edited: Oct 2, 2011
5. Oct 2, 2011

### msbe

Yeah that's the problem haha, VHF is in the region of pain where 1/4wave is pretty damn long, and implementing on a PCB requires some interesting trace routing

6. Oct 2, 2011

### yungman

It is the amount of terminations in the star scheme that bugs me more than anything else. As for the size and the large width of the quarter wave trace, you can use high diel constant to show the velocity and thinner dielectric to to narrow down the trace. It is still doable. It is the amount of the termination that the unselected switch that make it a lot more complicate.

You sure you cannot find any multi pole solid state RF switch in one package? But if you layout carefully using small footprint solid state RF switch, the total length of the path is still much shorter than the wave length. 100MHz is not the most critical in the world.

At 100MHz, you sure you cannot use the normal fet multiplexers IC? I had seen very low on channel resistance switches 10+ years ago, they might have much better and lower on resistance now. Don't try to design as if you are dealing with 1GHz signal.

7. Oct 2, 2011

### yungman

For VHF, there are plenty of cheaper op-amp around that you can buffer the input with opamp using 50 input termination, then drive into the multiplexer and then buffer each of the output with an opamp. In the middle between the input opamp and output opamps, make the distance very short and don't even worry about termination. Make sure you have something like a 10K resistor from input of each ouput opamps to ground so the input does not float when it is not selected. Might be cheaper this way.

8. Oct 2, 2011

### skeptic2

One of the problems with a 1:N switch is the lack of isolation between the outputs. As long as you're splitting the input between N outputs this isn't a major problem, although VSWR on one output would get reflected to the other outputs as well. You would have more serious problems if you tried to combine the N outputs into 1 input.

However, if the number of outputs is an integral power of 2, there is a better solution. Look up Wilkinson Splitters. A 50 ohm input is split between two 70.7 ohm TLs, each 1/4 wavelength long to transform 50 ohms to 100 ohms. The load ends of the TLs are connected with a 100 ohm resistor. The 100 ohm impedance of the TLs in parallel with a 100 ohm resistor gives 50 ohms. In addition the short path through the resistor combined with the half wavelength path through the sum of the two 1/4 wavelength lines gives you isolation between the two outputs. You can stack Wilkinson Splitters on each output to attain any integral power of 2.

The problem is the physical size of the solution for VHF but there are some ways around that too. Instead of doing everything on a PC board you could use 70 ohm coax and just put the output junctions on a PC board.

Another solution I discovered and haven't seen published anywhere is that instead of using a transmission line to transform 50 ohms to 100 ohms (or any other impedance) use a pi or T network in which each of the 3 elements of the pi or T has the same impedance as the transmission line that would have be used, i.e. 70 ohms. You have to use the same network, either pi or T on each leg. Also when using a pi or T network instead of a transmission line, the bandwidth is not as great. This makes it possible to use Wilkinson Splitters at VHF frequencies.

Edit: See http://www.microwaves101.com/encyclopedia/wilkinson_nway.cfm for N-way Wilkinson Splitters.

Last edited: Oct 2, 2011
9. Oct 2, 2011

### skeptic2

I should have mentioned in the above post that the pi or T matching networks must have the middle component different from the two end components. A pi network might have inductors for the legs and a capacitor for the bridge or vice versa.

10. Oct 2, 2011

### yungman

Oh yeh, I forgot using lump element impedance transformation. My bad!!! At 100MHz, it is absolutely doable.

In my suggestion using opamp to buffer and short traces, I don't think you need to worry about termination inside the circuit.

11. Oct 2, 2011

### skeptic2

It's not strictly impedance transformation and that's why you can't do it with L matching networks. The pi or T preserves the 1/4 wave phase shift making it possible to maintain isolation between outputs.

12. Oct 2, 2011

### yungman

As I said, for VHF, you don't really look at it as RF and microwave and quarter wave stuff. There are so many easier way of doing this. Look into RF multiplexer and use opamp to buffer input and output and you don't need to worry impedance matching inside the circuit. It's under 200MHz!!!

With careful layout and choice of components, I don't see any problem of getting over -30dBm channel isolation. Even at over 2GHz, it is not hard for me to get -25dBm+ isolation.