Wide-spread constructions of a "Coaxial splitter" or a "PCB bridge" do not provide any outputs matching nor crosstalk isolation. They can be used for two "non-interfering" antennas with short cables.

So, i decided to design (model in HFSS) and build a "right" power splitter based on Wilkinson splitter. Its features are: good matching of all 3 ports, good output channels separation (up to tens of dB), low insertion loss (ideal -3dB from input to both outputs, -3dB from any output to input).

I choose to use the most stable, predictable and cheap dielectric: air ;) . So, the construction is basically a microstrip of thin (0.1 ... 0.5mm) copper or brass over a conducting plate. The advantages of such an approach are low sensitivity to the thickness of the sheet, and the ability to achieve almost any given impedance of a line by varying trace width.

This construction is the result of many attempts of modelling in HFSS package:

The input signal from a connector enters a 50-ohm line, which then splits to two 71-ohm traces 0.25 lambda in length. Both these traces meet through a 100-ohm resistor, and pass to output connectors througs 50-ohm lines. The "extra" 50-ohm lines are needed to correctly lead the wave from connectors to the splitter itself, and to make enough space between output ports to fit connectors.

Here are the simulation results:

Return losses on input port (blue line) and output ports (red line),  crosstalk between outputs (green line).



SWR for input port (blue line) and output ports (red and green lines). No more than 1.06 in working band.



Transfer ratios: In to Out (blue line) and Out to In (red line). About -3.4 dB, which is close to ideal.



So, let's DO it!

Base plate is made of 2-mm thick 2-sided FR-4 sized about 95mm х 60mm. It has 3 holes for connectors, and 3 small holes around every of them for mounting screws.



The edges of connectors' cases are cleaned and tinned. After the installation of the connectors they are soldered to the inner copper layer of the base plate.



The conductor is made of 0.1mm...0.5mm-thick copper or brass sheet. I printed the drawing of the traces, cut it from paper and glued the template to the sheet. Then with a razor and a metal ruler I cut the shape. I made a 1-mm "notch" in the middle of each side where a connector will be soldered.
Now we can clean the microstrip, cut the center pins of connectors to be 2 mm above the plate, and solder! Use a 2-mm thick "spacer" between the trace and the plate to keep distance. After that, apply a polyethylene spacer to the "meeting point" of legs to prevent bending.

The 100-ohm resistor between the legs is made of two 200-ohm or even three 300-ohm SMD parts in parallel. Try to "spread" them over the length of edges.


Now, clean all the flux, and assemble the case! Its side walls are 15 mm and made of brass, as well as the lid.



Ready for testing! Here are the Netstumbler results (for DLink DI-524):
Signal/noise ratio is 44 dB through the "NogiFroggy" and 47 dB while direct antenna connection.

This splitter is now "in operation" with DLink DI-524, sector antenna similar to DLink ANT24-1200, and a 16-dB panel antenna.

All the drawings (in PDF) and models are on my page.

Enjoy!