Build 50 ohm power splitter (question for RF guru)

[eurofox]

I'm not an RF dude and ask opinion of people with knowledge in this area.

I'm working on a ultra fast pulser, I build my own power splitter (because it is extremely expansive) with 16,5 ohm resistors in T connection, they are connected to SMA terminals and there is more than 1 cm space around the 1% resistors. It is build in a quite massive aluminium box.

I just wonder what kind of bandwidth I can reach this way? I know that some RF foam is available and wonder if it is useful in my case.

I order as well to improve the design 33 ohm 1% SMD resistors (1W), in parallel I will get the requested 16,5 ohm.

Thanks in advance for your input. 

eurofox

[T3sl4co1l]

If the layout is correct, chip resistors are good into the 5-10GHz+ range.  And whatever fractional nanoseconds that corresponds to.  Mind that low values will tend to look inductive (physically, they look more like transmission line than terminating resistor), and high values, capacitive (same reason, opposite direction).

Obviously, something as high as a 1M resistor ain't gonna be 1M impedance at 10GHz, not when the footprint pads themselves have lower capacitive reactance between them through the substrate alone.

Tim

[acbern]

the problem is not so much what chip resistor you use but how the layout is, how tightly controlled your microstrip impedance is and so on.
to judge you would have to say much more about how you did it. the other question is what you need in terms of reflection, amplitude/phase match (i.e for your application, what does 'work' mean?). 10ghz can be reached, sure, even with FR4 if done right and with some constraints, but in the end you will have to measure this with a vna.

[eurofox]

My pulse generator is in the area of 30ps rising edge, amplitude from 200 mV to 800 mV.

The resistors got about 1 cm from SMA terminals and between each other in Y shape, from top and bottom of the aluminium container 1 cm as well.

The resistor are based on 2 different 1% 1/4W in parallel to get the 16,5 ohm.

[ejeffrey]

How much power do you need?  Adding resistors in parallel will increase parasitic capacitance.

[eurofox]

How much power do you need?  Adding resistors in parallel will increase parasitic capacitance.

1/4W is ok, I don't think that resistors of 16,5 ohms are available, you have to build them with 2 resistors in parallel.

Maybe on special order but then I can better buy an Agilent splitter, is only 1400 US$ 

[KJDS]

Will this do? I've got a few lying about, yours for £20 plus postage.

http://www.e-meca.com/rf-power-divider/divider_specs.php?divID=3&specsID=22

Alternatively, I'd go for using 16.8R resistors, or a triangle of 51R resistors. Use chip resistors otherwise the parasitics will kill you. Have the parts touch each other at the junctions.

[G0HZU]

There's several ways to try and make a successful design. However, you do need access to a decent VNA if you are looking for high performance.

You can either adopt an iterative approach armed only with the VNA and experiment with star and delta configurations of various SMD resistor packages to see which is best.

Or you could try and get decent models of the resistors you are using (up to many GHz) and try and simulate the design. This can be very useful in terms of learning 'faster' than the VNA approach but ultimately you will still end up with the VNA and the 'best' combination of resistors and teasing/snipping it with a scalpel and maybe even adding some tiny slivers of copper tape to get best balancing of the various parasitics in the resistors and the layout. Then when you get a good result, make a neat (final) version.

I would start off with the delta configuration using a pair of small 100R SMD resistors stacked in order to get 50R for each arm. The resulting parasitics from this may well mean it performs better than one with a single small 50R SMD resistor per arm but it all depends on what spec you are trying to meet for phase balance/loss/match for the three ports.

[Marco]

Will this do?

Those are quadrature combiner/dividers, phase shift together with the limited frequency range is unlikely to give you two nice pulses as outputs.

Eurofox, why didn't you make a delta splitter? It uses 3 Z0 resistors.

http://www.microwaves101.com/encyclopedia/Resistive_splitters.cfm

(The Owen splitter they describe is also interesting, more loss but also more isolation.)

PS. there's some 4GHz splitters on ebay for 30 bucks. Of course they have female connectors, which might be trashed. (Why aren't male connectors standard on equipment? Seems more reasonable to me to put the easily broken gender on the cable.)

[Alexei.Polkhanov]

If you translate 30ps rise time into bandwidth using simple rule of thumb formula:

BW (in GHz) = 0.35 / TR (nanoseconds) or in your case BW(GHz) = 0.35/0.03 = 11.6 GHz

So you need power splitter for 12 GHz. Example for 7.8 GHz shown on picture below. Geometry will be different for 12 GHz.  Also note that white material is NOT FR4 - you need low loss high frequency substrate such as Rogers "Duroid" or Alumina. Search for "duroid" on Ebay - you'll probably find some.

In order to get correct geometry for your power divider you can search for free tools online or use commercial tools such as Agilent ADS (Advanced Design System) or Microwave Office etc. - they are very expensive (>$25K however). If you use right data for substrate and keep the dimensions withing +- 5% of simulated then you will probably get something close to desired result even if you have no access to VNA.

I would try to find power divider for ~12GHz first.
If you work with this kind of frequencies you kind of start to appreciate why some of that stuff cost so much.

[KJDS]

Will this do?

Those are quadrature combiner/dividers, phase shift together with the limited frequency range is unlikely to give you two nice pulses as outputs.

A quadrature splitter has the outputs 90 degrees apart in phase, which these don't.  Whilst they aren't perfect for the job, I suspect that unless the eorofox has access to a 6GHz VNA. or preferably a 12GHz one,  then they will work better than anything he can make.

[ejeffrey]

1/4W is ok, I don't think that resistors of 16,5 ohms are available, you have to build them with 2 resistors in parallel.

16.5 Ohm is an E96 value and available from all the major manufacturers.  Or you can use a delta configuration with 49.9 ohm.  If you can't get those easily, the nearest available from the E24 series (51 ohm or 16 ohm) are pretty close, especially the 51 ohm.  You may find that the reflection from the slight impedance mismatch is less objectionable than the extra reactance of additional components.

[ejeffrey]

A quadrature splitter has the outputs 90 degrees apart in phase, which these don't.  Whilst they aren't perfect for the job, I suspect that unless the eorofox has access to a 6GHz VNA. or preferably a 12GHz one,  then they will work better than anything he can make.

What you linked to looks like a wilkinson power divider.  As you say, the outputs are in phase, not in quadrature like a hybrid.  Still, a 2 GHz wilkinson divider is not a great choice for a pulse splitter, it will definitely screw up your rise time.   A simple resistive power divider is actually not that hard to get right the first time, and you should easily be able to beat that device even without a VNA -- at least in terms of preserving the sharp rise-time of your pulse.

[Alexei.Polkhanov]

You guys have no faith in simulation tools. VNA is very useful, but it should not be just trial and error approach.

[eurofox]

If you translate 30ps rise time into bandwidth using simple rule of thumb formula:

BW (in GHz) = 0.35 / TR (nanoseconds) or in your case BW(GHz) = 0.35/0.03 = 11.6 GHz

So you need power splitter for 12 GHz. Example for 7.8 GHz shown on picture below. Geometry will be different for 12 GHz.  Also note that white material is NOT FR4 - you need low loss high frequency substrate such as Rogers "Duroid" or Alumina. Search for "duroid" on Ebay - you'll probably find some.

In order to get correct geometry for your power divider you can search for free tools online or use commercial tools such as Agilent ADS (Advanced Design System) or Microwave Office etc. - they are very expensive (>$25K however). If you use right data for substrate and keep the dimensions withing +- 5% of simulated then you will probably get something close to desired result even if you have no access to VNA.

I would try to find power divider for ~12GHz first.
If you work with this kind of frequencies you kind of start to appreciate why some of that stuff cost so much.

This calculation I did as well, my RF instruments are limited, counter up to 20Ghz, generator up to 3,2Ghz, SA Rigol + TG 1,5Ghz, USB SA up to 3Ghz.
Sampling scope up to 11Ghz.

[eurofox]

1/4W is ok, I don't think that resistors of 16,5 ohms are available, you have to build them with 2 resistors in parallel.

16.5 Ohm is an E96 value and available from all the major manufacturers.  Or you can use a delta configuration with 49.9 ohm.  If you can't get those easily, the nearest available from the E24 series (51 ohm or 16 ohm) are pretty close, especially the 51 ohm.  You may find that the reflection from the slight impedance mismatch is less objectionable than the extra reactance of additional components.

You are right, I checked in my stock and I was supposed to have almost all current values and didn't find it but it is available and just order them.

[G0HZU]

You guys have no faith in simulation tools. VNA is very useful, but it should not be just trial and error approach.

I do RF simulations pretty much every day at work using Genesys, AWR and Sonnet but in this case I think the simulation accuracy will only be as good as the models for the chip resistors.

Modelithics do some decent models up to many GHz for various PCB substrates/thicknesses but I still think the 'final' trimming of this (homebrew) circuit can only be done with a decent VNA and a scalpel etc.

Note: The 7.8GHz splitter PCB you posted an image of isn't a broadband design.

[KJDS]

You guys have no faith in simulation tools. VNA is very useful, but it should not be just trial and error approach.

I have a lot of faith that I'd get one working quicker sat at a 20GHz VNA with a time domain option than anyone would with CST Studoi, MWO, ADS or any other simulator. If I were doing the design to put into volume production then I'd start by spending a lot of time characterizing the resistors, then even more time with CST studio to put together a full 3d EM model, then building and testing, but for a one off, build something and tweak it with a scalpel. It's too simple to spend much time playing with maths up front.

[Alexei.Polkhanov]

Yes, this is true. Unfortunately only few manufacturers like TDK (http://product.tdk.com/en/technicalsupport/tvcl/) have models for caps and resistors that can be used at these frequencies. I am not even sure how accurate these models are.

There is no doubt that with VNA things are much easier and faster, but if you don't have access to VNA that covers these frequencies what do you do?

[Richard Head]

If you don't have access to a VNA then you will never know if your design is correct. No amount of simulation will replace a proper real worl test.

[Marco]

Couldn't he read in the data from the sampling scope and take the FFT from the pulse source and the split source and divide them to get the transfer function?

[acbern]

Couldn't he read in the data from the sampling scope and take the FFT from the pulse source and the split source and divide them to get the transfer function?

not really, unless the sampling ferquency would be much much higher to get a meaningfull response.

if this is for a serious measurement and not just somehow a splitter doing something (at 30ps tr it does not spund that way), the only way to get a reliable result (without a wideband VNA and lots of work and costs to build an RF PCB, buy appropriate RF resistors and so on) is to buy something. this is not an issue of simulation, the parasitics, the tolerance, and the unknowns of the parts models, will be the main problem. as this is for pulse measurement, a flat phase ressponse is needed too over the whole frequency range. in order not to distort the pulse, the 11.6ghz (0.35 factor) will not do it either.   
most cost efficient are used picoseocnd pulse lab splitters on ebay. they are also built for exactly that application.

[Marco]

He could measure if it were flat to 11 GHz at least, which would still be a worthy effort for a DIY splitter (and not that much short of the connector limited bandwidth).

Weinschel splitters seem to be the better deal compared to the HP and Picosecond pulse labs splitters ... some sold for 70 bucks not that long ago and there's one on there for 150 bucks buy me now at the moment.

[eurofox]

He could measure if it were flat to 11 GHz at least, which would still be a worthy effort for a DIY splitter (and not that much short of the connector limited bandwidth).

Weinschel splitters seem to be the better deal compared to the HP and Picosecond pulse labs splitters ... some sold for 70 bucks not that long ago and there's one on there for 150 bucks buy me now at the moment.

I just score this one 

http://www.benl.ebay.be/itm/Weinschel-93459-Model-1515-Power-Divider-DC-to-18-GHz-SMA-fbb22-8-/400732429250?pt=UK_BOI_Electrical_Test_Measurement_Equipment_ET&hash=item5d4d839bc2&_uhb=1

Based on the discussion my idea was not so bad to use resistors in "air" isolation, no PCB.

Like I said in the post I'm not an RF dude, I manage very well and have instruments till 3-4 Ghz.

Just bought a 1Ghz probe for my MSOX3104 

I really appreciate all answers, it was interesting. Above 4-5 Ghz we enter in a complete new world that look more to a microscopic plumbing than electronics.

Thanks 

eurofox

[Marco]

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).