Go carefully when feeding an end fire array. Unlike a vertical collinear array, the coupling between ports is substantial. Even with a perfect well isolated splitter, you won't get the current distribution and hence patterns that you expect.
What you need to do is simulate the array in NEC or something, and experiment with the voltage magnitudes and phases on the elements, until you have the current distribution or radiation pattern that you want. Then only can you start to design the feed network - three different transformers and delay lines to achieve the feed voltages you want.
Some end fire arrays are designed for feeding simplicity. The LPDA has a simple transmission line running between elements, they each take some current and the rest flows on down the line. The yagi generally has a zero feed point voltage on all but one of the elements, which is of course easy to feed. There are others, like a yagi with two or three driven elements, the W8JK and the HB9CV with two, but the common theme is that the feedline design is an integral part of the antenna design.
I tried a quick simulation of four cases:
- Simple $\lambda/2$ dipole.
- Result: Well matched at 445 MHz. Gain=2.1 dBi.
- Three dipoles $\lambda/4$ apart. Fed with $-90^\circ, 0^\circ, +90^\circ$ current sources.
- Gain = not calculated for current sources, sorry. F/B = 7 dB.
- Currents on the three are equal, and phases are as requested.
- Three dipoles $\lambda/4$ apart. Fed with $-90^\circ, 0^\circ, +90^\circ$ voltage sources.
- Gain = 8 dBi, F/B = 6 dB.
- Impedance at ports: Insane. One port 1.5:1, one 3:1, last port has power flowing back from it, SWR is negative, hard to describe.
- Currents on the back two are equal, almost no current on the front dipole.
- Three dipoles $\lambda/4$ apart. Fed with transmission lines: first each dipole has a $\lambda/4$, 86.6$\Omega$ line, which would transform 50$\Omega$ into 150$\Omega$. Then each has a 150$\Omega$ line back to a common source, one zero length, one $\lambda/4$, one $\lambda/2$. This would be the naive phasing to get the end fire effect.
- Gain = 5.6 dBi, F/B = 5 dB.
- Impedance at port: SWR of 1.5:1.
- Currents quite mismatched, middle dipole has about 60% of the current of the side two. Phases are about right for end-fire.
The next step, which I can't easily automate, would be to play with the transmission line lengths and impedances to achieve equal current, correct phasing. As a guess, it might pay to split half the power to the middle dipole, and 1/4 to each side one, keeping the -90, 0, 90 phase relationship.
In summary, N-port equal power equal phase splitters are for collinear arrays and perhaps broadside arrays, but for end-fire things get a lot more complicated.