4 way wilkinson power divider

[anum_pirkani]

I am designing a 4 way wilkinson power divider but i am not getting the desired results.
I have to implement it wideband i.e 5GHz-15GHz.

After division into 4 parts, the S parameters have variations from the desired value and there is a difference between S21, S31, S41, S51.

Please help me out.

Each wilkinson network has N=3 and i have used 3 such two way wilkinson dividers.

I have attached the design and its output.
The design has been made in ADS.

 

[Darktrax]

The splitter you show in your picture does not look like a Wilkinson. It appears to be a split, and re-split, with multiple quarter-wave matching sections, possibly not all at the same electrical length.

The defining feature of a Wilkinson is the resistor between (in phase) outputs, and in the case of multi-stage Wilkinsons, resistors in each square branch.There is also a version of a Wilkinson where two resistors are used separated by two quarter-wave line sections, making the whole structure look like a variant of the "rat-race" splitter, and not requiring the outputs layout to come close so as to reach the resistor.

Here is a link to a comprehensive collection of Wilkinsons design information..
--> **broken link removed**
**broken link removed**
To try for wider bandwidth, there is another class of Wilkinsons.
**broken link removed** Then, to get the value of the Wilkinson resistor technique without having to bring outputs together, we have the beautiful variation called "Gysel Splitter".
**broken link removed**

There are lots of examples, many using ADS. Keep in mind, these work as splitters. To use as combiners, the signals have to be in phase.

 

[moreho]

Why...
1- no resistors
2- Too Many sections
3- unequally spaced ports
4- 120% Bandwidth From Wilkinson! Its very Hard To make Wilkinson wideband, ...

 

[anum_pirkani]

thanks a lot for your detailed reply.

I am actually trying to make a wideband wilkinson in frequency range of 5GHz-15GHz with 10GHz being the center frequency. I went through a research paper according to which, if the ratio of f2/f1 is 3, three branches of microstrip line are to be used each having different impedences.
Z1= 1.1497*Zo
Z2=1.4142*Zo
Z3=1.7396*Zo

Similarly, a resistor is to be placed after every branch.

So, i have made a two way wilkinson at the beginning according to this numerical data and then just copied and pasted this 2way wilkinson at each of its output to get a 4 way wilkinson.

Without resistors, i should get S21, S31, S41, S51 approx equal and close to -6.01 but i am getting variations in it.
When i place resistors in the design, i do get correct isolation, S11, S22, S23 and so on, but the problem i am facing in are transmission coefficients i.e S21, s31, s41, s51.
I am making this design in layout window of ADS.

- - - Updated - - -

P.S: when i make this design using ideal microstrip lines, i do get correct results

 

[Darktrax]

Hi anum_pirkani

All quarter-wave resonant combiners have some kind of limited bandwidth.
The very nature of any splitter that uses trick quarter-wave resonant line lengths in its structure is going to be limited in bandwidth in some way. Where you see attempts to mix or add a several of such matching sections, sized to favour other frequencies to cover a wider band, it usually also involves more insertion loss, multiple peaks in the return loss and pass characteristic.
There are some great designs, and also, some Wilkinson-like arrangements have bands wide enough to be very useful, but the bandwidths in quarter-wave resonant networks will always be limited by the fact they are, inherently, quarter-wave.

Do invoke the search on this forum for the previous thread on Wilkinsons.
It was deeply discussed, and for me, it taught me several key points.

1. The addition of a resistor between the outputs brings together ports that are 180 degrees out of phase, loaded in a pseudo-terminated way, akin to a termination across a balanced line, and so they will cancel. This means that for signals at the split ports that are in phase, there is isolation. If re-combining signals that were previously split, and amplified, this is perfect. Also good for combining signals from multiple antennas, the coherent signals coming from a single source can be combined without some being re-transmitted, and yet be matched. This property decides at the outset whether the Wilkinson is OK for the application. If it has to combine separate unrelated signals, it will not have the isolation property, and you may have to use something else.

2. The multi-stage, multi-terminated Wilkinsons can offer stretched bandwidths, with some trade-off, but it brings about a layout issue. The need to "bring together" the branches to meet the resistor, and the fact you cannot have "zero length" resistors, forces characteristic multiple semi-circles or squares. This is where the closely related "Gysel" variant that splits the resistor into two unbalanced terminations (of a sort), gets around the problem.

3. The Wilkinsons, via Gysel, morph into the "Rat-Race" class of combiner-splitters. Just like anything that depends on quarter-wave features, they have limited bandwidth, which is often good enough if the frequency is high enough. Just like Wilkinsons, coherent in-phase signals are necessary if used as combiner. If as a splitter only, there is no problem.

4. Regular quarter-wave splitters. These are very well known. Essentially, the input has to be matched to some impedance at the junction which is the result of two branches coming together in parallel. Then, for each branch, they need their own quarter-wave transformer sections to make their ports also match to the impedance seen at the junction. The junction impedance cannot be the same as that of the ports. The impedance of a quarter-wave transformer line between them is usually the geometric mean of the impedances at the ends. The splitter in your picture looks like one of these.

5. Wider band things
Various forms of coupled line. The Lange Coupler comes to mind here. Usually, there is some attempt by various tricks to equalise the odd and even mode impedances of the coupled section. Coaxial sections, with magnetics added to inhibit currents on the outside of the shield, with balun tricks can yield a splitter-combiner with more than an octave of bandwidth, and impressive performance. See the pictures of the topside and underside of a older technology ANZAC H-8-4. It works from about 900MHz to 2GHz. Not easy to implement in stripline or microstrip, but great for when a Wilkinson just will not do.

P.S. If your layout version fails, then maybe the substrate definition is different to the circuit input version.

 

[anum_pirkani]

i get it...!!!
your detailed explanation on splitters had been very useful for me. You are right, the impedences at all output ports can never be exactly matched. No mater what i do, i am getting a variation in the output ports.
After tuning and changing lengths of various branches, i have been able to reduce the gap between ports to around 2db which is acceptable in my case.
thanks a lot for your kind help.