Build 50 ohm power splitter (question for RF guru)

[KJDS]

I've just built one, it isn't pretty but it's working reasonably. just used two 50ohm resistors, so good input return loss, which is ok going into identical matches. There's not much room to get a soldering iron in to put a third resistor in.

Amplitude balance to 6GHz is excellent, phase balance was better than 4 degrees to 6GHz.

[KJDS]

Phase and amplitude comparison between the two channels.

[eurofox]

Do take a good look at the fingers inside the female connectors on the splitter ... you never know how well they've been treated (as I said before, it would be better if all connectors on equipment was male).

I will follow you advice and put a permanent "sex changer" 

[Alexei.Polkhanov]

Weinschel - nice splitter. Looks like they are "Aeroflex" now. It does not seem to have any side panels or screws that you can take out to pick inside and see how it is built.

I would be dying of curiosity   

[KJDS]

Weinschel - nice splitter. Looks like they are "Aeroflex" now. It does not seem to have any side panels or screws that you can take out to pick inside and see how it is built.

I would be dying of curiosity   

I'll guess that it's a thin film circuit with the resistors being part of the process.

[G0HZU]

Glad the OP has found something suitable

However, I do feel a bit frustrated by the lack of spec/requirements data posted up. When the alternative was a precision splitter from HP costing $1400 I assumed you were looking for something very precise with a high spec. All my comments were aimed towards producing a 'precision' RF splitter up to many GHz.

Last year I did make what I would class as a precision splitter when I was evaluating my old (made in 1965?) HP RF voltmeter as it measures gain and phase to 1GHz. It was just a temporary lashup splitter but my target spec was:

VSWR  better than 1.05:1 on all ports
Phase tracking <1 degree
Amplitude tracking << 0.1dB

The aim was to get this performance across the whole range of the instrument and I designed it using my own (measured) models of the chip resistors I used. I also did a full EM simulation of the PCB.

I used 100R resistors in parallel to make up the 50R resistors for a delta splitter because the parasitics were far better when compared to a single 50R chip resistor.

I can try and dig it out and remeasure it and post up the results if anyone is interested but this was very much a 'precision' splitter that was designed on a simulator and the measured results matched the simulator pretty much spot on.

The secret is to get accurate models of the chip resistors and I measured the 100R resistors right out to 3GHz to get a decent model to put into the simulator. I also select on tested each of the 100R resistors on a decent DMM to get as close to 100R as possible for each. I chose the best 6 resistors from a strip of maybe 30.

But I think it will be a lot tougher trying to make (let alone model) something decent that runs to 10GHz. I've never tried it.

[eurofox]

Glad the OP has found something suitable

However, I do feel a bit frustrated by the lack of spec/requirements data posted up. When the alternative was a precision splitter from HP costing $1400 I assumed you were looking for something very precise with a high spec. All my comments were aimed towards producing a 'precision' RF splitter up to many GHz.

Last year I did make what I would class as a precision splitter when I was evaluating my old (made in 1965?) HP RF voltmeter as it measures gain and phase to 1GHz. It was just a temporary lashup splitter but my target spec was:

VSWR  better than 1.05:1 on all ports
Phase tracking <1 degree
Amplitude tracking << 0.1dB

The aim was to get this performance across the whole range of the instrument and I designed it using my own (measured) models of the chip resistors I used. I also did a full EM simulation of the PCB.

I used 100R resistors in parallel to make up the 50R resistors for a delta splitter because the parasitics were far better when compared to a single 50R chip resistor.

I can try and dig it out and remeasure it and post up the results if anyone is interested but this was very much a 'precision' splitter that was designed on a simulator and the measured results matched the simulator pretty much spot on.

The secret is to get accurate models of the chip resistors and I measured the 100R resistors right out to 3GHz to get a decent model to put into the simulator. I also select on tested each of the 100R resistors on a decent DMM to get as close to 100R as possible for each. I chose the best 6 resistors from a strip of maybe 30.

But I think it will be a lot tougher trying to make (let alone model) something decent that runs to 10GHz. I've never tried it.

I really appreciate your post and from other members here, it is very instructive.

My first try was with a Wilkinson splitter but got too much loss of signal.
Next I try with a T splitter with normal 1% resistors (16,5 ohm), I have a quite big collection but no 16,5 and assume it was not available and did no check and just use 2 resistors in // and understood it was not perfect. Since one member told that it is a standard value I check it and immediately order it.
I was also “hunting on ebay” for an existing splitter, all I could find that was affordable was outside EU …. Big delay, high shipping cost, customs clearance, duty to pay ? and finally was lucky to find one in UK that is virtually unused according the ebay poster.

eurofox

[G0HZU]

I think you will get good results with the Weinschel splitter

I've found my old evaluation jig for my HP8405A VVM and here's the splitter section from it. I needed a precise 6dB split when checking the performance of my HP8405A after I had repaired and had aligned it as per the service manual.
The idea was to make something that was very, very precise in amplitude and phase tracking out to 1GHz and also I wanted the VSWR to be very low.

It wouldn't be good enough for what you need but it does show what can be achieved with a few SMD resistors costing less than 1 penny each.

Note: You can't really see it but the PCB artwork under the SMD resistors has been optimised using Genesys+Sonnet EM to tune out any remaining parasitics from the stacked 100R resistors. I've got the VNA warming up and I'll post up a few measurements.

I've written 3GHz on it so I'll test it right out to the limit of my old VNA (3GHz) and post up the results. I remember that it was remarkably good up to the wanted 1GHz bandwidth so prepare to be impressed

[G0HZU]

OK here's the first few plots and it is actually much better than I remember!

I remember that I did spend a bit of time adding/removing solder to optimise the performance but the results are definitely 'precision' class. I'm prepared to admit there may be an element of luck in the tolerances of the milling as this PCB was milled on my T-Tech router. Therefore, I seriously doubt I could replicate these results on a second unit but I'd hope to get very close.

The first image is the return loss on the sum port. It is better than 35dB at 3GHz although there will be a degree of measurement uncertainty here on my old VNA. However, it measures about 50dB return loss for a precision SMA 18GHz load at 3GHz and this is not the same load I did the calibration with.

The second plot is the amplitude tracking. Again the results are staggeringly good even to 3GHz. I aimed for <<0.1dB at 1GHz and met this easily. The two traces are virtually on top of each other on a 0.1dB/div scale.

[eurofox]

I just wonder if you put the resistors in the middle leaving space between resistors if you cannot improve the parasitic effect because now since they are so close to each other there should be a capacitance effect and again I'm not an RF expert.

[G0HZU]

Yes, ultimately, this splitter will lose performance at higher frequencies due to the layout. I think the performance will probably degrade quite quickly above about 5GHz because I have used 'big' SMD 100R resistors and stacked them on top of each other to get 50R.

But it only had to work to 1GHz to test the old HP8405A 

See below for the phase tracking.

The result is pretty good although it depends on the torque setting for each SMA connector. I've tried to keep it the same on both tests.
Port 1 was measured for phase and normalised and the trace below shows the phase error between port 1 and port 2.

You can see I met my target of <1degree error at 1GHz.

See also the sum port VSWR plot that shows about 1.03:1 at 3GHz.

[G0HZU]

I checked on the Minicircuits website and they do a 4.2GHz splitter with SMA connectors (ZFRSC-42+) and it performs about the same as my homemade splitter.

They cost about $70 from Minicircuits.

http://217.34.103.131/pages/s-params/ZFRSC-42+_GRAPHS.pdf

I had to do a bit of tweaking with tiny bits of solder to optimise mine but you can see that it's possible to get this level of performance from a low cost splitter

[rfbroadband]

you can push these designs up to many GHz (> 10GHz). Here are a few things to consider

1) SMD resistors: From an RF point of view a resistor is not a resistor. It is a resistor with a series inductance, parallel capacitance and two C to gnd from each side. If you would calculate the transfer function of a resistor you would find that a resistance value ROpt exists where the resistor is really (almost) just a resistor. If R < ROpt the R has inductive behavior and for R>ROpt is has capacitive behavior. Example DC resistance R is 200Ohms but at 5GHz |Z| = 100 Ohms etc. ROpt depends on the geometry of the SMD resistor. 0402 resistors often have ROpt at around 100 Ohm. This is why two 100 Ohms 0402 resistors in parallel are a much better 50 termination that just one 50 Ohm resistor. I have built a 20GHz precision termination using that method that had > 20 dB return loss up to 20 GHz.

2) Obviously choose an RF PCB material that supports the desired frequency range...

3) choose the correct connector type that supports your desired freq. range (SMA vs SMB etc)

4) be very careful on how you design your SMA to PCB connector footprint for your PCB. It is very easy to introduce resonances that basically destroy your transfer characteristic. It gets even more complex if you decide to use a multi layer board (more than just two layers)

5) if you really want to push the frequency range (> 10GHz) you will have to optimize and compensate for steps in width of the transmission line relative to the with of your SMD components

6) don't know how much power you want to dissipate..0402 is nice for broadband RF applications there are limits on how much power they can dissipate.

hope that helps

[eurofox]

you can push these designs up to many GHz (> 10GHz). Here are a few things to consider

1) SMD resistors: From an RF point of view a resistor is not a resistor. It is a resistor with a series inductance, parallel capacitance and two C to gnd from each side. If you would calculate the transfer function of a resistor you would find that a resistance value ROpt exists where the resistor is really (almost) just a resistor. If R < ROpt the R has inductive behavior and for R>ROpt is has capacitive behavior. Example DC resistance R is 200Ohms but at 5GHz |Z| = 100 Ohms etc. ROpt depends on the geometry of the SMD resistor. 0402 resistors often have ROpt at around 100 Ohm. This is why two 100 Ohms 0402 resistors in parallel are a much better 50 termination that just one 50 Ohm resistor. I have built a 20GHz precision termination using that method that had > 20 dB return loss up to 20 GHz.

2) Obviously choose an RF PCB material that supports the desired frequency range...

3) choose the correct connector type that supports your desired freq. range (SMA vs SMB etc)

4) be very careful on how you design your SMA to PCB connector footprint for your PCB. It is very easy to introduce resonances that basically destroy your transfer characteristic. It gets even more complex if you decide to use a multi layer board (more than just two layers)

5) if you really want to push the frequency range (> 10GHz) you will have to optimize and compensate for steps in width of the transmission line relative to the with of your SMD components

6) don't know how much power you want to dissipate..0402 is nice for broadband RF applications there are limits on how much power they can dissipate.

hope that helps

It is very instructive and confirm ideas I got.

What about not using PCB at all and connect the SMA terminal to copper wire the SMD resistor in the air in a Y shape?
In theory we should avoid at least the parasitic influence of the PCB.

[G0HZU]

I think you can do away with the PCB for low cost and low performance homebrew splitters up to about 1GHz. However, in terms of a precision homebrew splitter for several GHz I think you have to use a PCB. Especially if you want it to be rugged and reliable.

In my case I exploited the PCB in order to fine tune out the remaining parasitics from the piggy backed 100R resistors.
I have a jig here that lets me measure SMD components very accurately up to about 3GHz so I measured two 100R resistors in piggy format and then used the model from this to help me choose how to 'end' the microstrip on my board such that the return loss up to 1GHz (at each port) was really low. As I said earlier, there is a subtle shape at the end of each microstrip to fine tune the performance. I used Genesys and Sonnet to help me optimise this shape at the end of each of the three microstrip sections.

I used some old GIL 2032 PCB material 0.03" thick because it gave the optimal microstrip width to suit the delta pattern of 0603 resistors. I also used decent SMA PCB end launchers soldered top and bottom to give a really good transition to the PCB.

So although the circuit board I posted up looks very simple, a lot of modelling was done behind the scenes to give me a good head start.

If I wanted to make something (decent) for higher frequencies I have some 0.02" Rogers 4003C material here that would suit 0402 resistors and I'd use some similar looking (but better) Suhner end launchers. However, I'd have to resort to using resistor models from Modelithics because I have no way here to model a pair of 0402 resistors out to 10GHz+.

So I'd be working fairly blind and I would need to use one of the 20GHz VNAs at work. I've made resistive splitters for use up to about 12GHz before but these were just on a synthesiser PCB and the performance wasn't critical in any way as I just wanted to crudely split an LO signal. So they were very much fit and forget.

[G0HZU]

I suppose if I was going down the path of the OP I'd be trying to avoid having lots of SMA transitions in the overall 'pulser' system so I'd be looking to build it all on a single PCB. i.e. the pulser and the splitter and I'd use GCPWG to give me more flexible control over the dimensions of the transmission line at various points of the design.