Power Splitter, Power Divider, baluns, va3diw

Power splitter combines or divides power. It works in bi-directional way. You can split, you can divide. You can use it for combining antenna signals, or use it as a two-port output device. I recommend to start with the article from [1], [6] w6pql, and [7] about power splitters, dividers, and baluns.

Test it at low power, than at full power. Two channel scope is good. You would be suprized how many mistakes are in handbooks. These devices have many physical forms dictated by frequency, bandwidth, impedance, and used power. Everything works on paper and simulation program. It is a different story running the device at full power. These circuits work also on 2.4 and 5GHz. The theory behind is solid.

The HF wire splitter can be wound on heavy high permeability 43 or 73 ferrite material, few turns. Core 3.5 by 3.5 mm hollow cylinder. Wound as bifilar. Measure the characteristic impedance. Z = sqrt(L/C). Start wire and the tailing wire shall be connected together. The broadband isolation of the first antenna to the second antenna is not extraordinary. Mechanical design significantly influence mutual isolation between the two ports, symmetry, and overall performance. Digging into details, it becomes not simple as expected.
Try multiple designs and ferrite materials. The HF material "43" and "73" seems to work best for HF. Try materials with u_r=850 to u_r=2000. Fair-Rite, NY product. Two hole ferrites "73-2402" or larger 73-202 from the Houston, Texas store. They are nice. You can make the purchase on phone. It is a customer oriented service with nice chat and instant feedback. As it should be.

Then there is the power splitter with three resistors. Not worth of considering.

The picture depicts two-core wire power splitter with tap published for years by ARRL. It has low port isolation, significant loss, and impedance mismatch. The picture was many times published without any performance notes. For twisted-wire version throw away the T1 core on the left. Use only the two sections on the right side T2. It looks like Wilkinson.
Running the splitter at high power can cause ferrite core saturation. Think about the wire impedances as a transmission line based device [4]. It is useful to have a L-C meter and oscilloscope on hand to check reactances. A spectrum analyzer will reveal the IMD₃ distortion in transistors and ferrites! Where you stand. The IMD₃ numbers are useful to know.
Motorola power splitters work with flat transmission line on the core. Hand made transmission line with strips of copper on Teflon tape gives better performance and bandwidth. The coaxial cable version is power bingo, requiring larger cores. Looking at the spectrum analyzer to manage 5 MHz bandwidth coupling hundreds of miliwatts, the C1 absence made no measurable difference. Experiment with C1 for wider bandwidth to compensate for leakage inductances.

Practical Example
The point of interest was to split two power signals and provide adequate isolation between the two split ports. Minimize the IMD3. More inductance per turn helps the transformer symmetry and lowers magnetic core flux. Saturation starts slowly. Each ferrite material has different properties. Motorola recommends to keep the magnetic flux under 100 Gauss = 0.01 Tesla for linear mode. The structure can be resistively balanced with sharp minimized crosstalk between the two split ports. That's the 100 ohm resitor.

Crosstalk & signal leakage, and nonlinearity in each of the branches, can hit -52dBc to -55dBc for IMD3 products distortion. That means the device is working right. It can work better. Make sure all ports have righ load impedance. A signal generator has rarely 50 ohm output. Connect resistive trimmer 220 ohm to the split output. Second trimmer 1k ohm between the two split ports. Small attenuator between the signal source and the device can fix impedances. The isolation between the two ports for 10% bandwith can reach >40dB at 10MHz.
T2 inductor reactance shall be more than 18uH/6t with 43 ferrite material @10MHz. jX = 1.13k ohm. The rule of thumb and rule of 20's = 20*50 ohm = 1000 jX. The balance and isolation of Ant1 and Ant2 ports is quite sharp when tweaking the resistors.

Signal Harmonics and signal shape distortion will limit the symmetry and balance floor.
Use sine wave.
While working on-site in office, I had only $10 cheap B&K signal source made with OPAmps.. The performance sucks. Thinking why this multi-billion corporation does not have few bucks for a regular signal generator. Dreaming about my own signal generator sitting on the bench.
Use 10dB 50 ohm attenuator for the signal input. The input/output impedances are never identical and never pure real. The input impedance shall be 50 ohm. Spectrum Analyzer has 50 ohm attenuator on the input. Watch the output on spectrum analyzer.

This corporate Agilent EMI Spectrum Analyzer with LCD screen is weird.
The replacable CMOS battery unit in the rear carries also clock generator and configuration memory in one pack. If the replaceable battery unit is flat, the spectrum analyzer sometimes loads calibration numbers. Sometimes it won't load any calibration numbers during boot. Forget replacing the CMOS battery.
You need the actual calibration numbers stored in this unit. Check the present day/year/time on the screen. If you see weird date and time, then it needs battery unit replacement and calibration. It means without working battery unit, you can measure numbers large as a barn.
If you buy used similar spectrum analyzer, after few years it becomes useless.
Thinking what's going on in Agilent Santa Rosa, CA. Was it designed by engineers, or filibusters? I never saw similar anomalies with old HP CRT scren spectrum analyzer.

Where the IMD3 comes from?
It comes partially from the ferrite transformer. For curiosity I compared ultralinear transistors for CATV distribution network, 2N5109, and 2N2222 at 10MHz. Two Norton type of amplifiers with ferrite power splitter. There is not significant difference between 2N5109 and 2N2222. The difference was about 3 to 3.5 dB in IMD₃.
To cover the generic limits:
The circuit needs larger cores with form factor related to delivered power, better isolation between the two power amps to be split, collector output nonlinearity shall be isolated. Signal amplitude on the collector can be a limiting factor. Lowering transistor gain might help.
The performance in common emitter configuration is weak for all above mentioned transistors. If you see HF high IP3 LNA with common emitter super-duper article, the maggot has no clue what he is talking about. It is about circuit architecture.

The IMD₃ numbers were confirmed on pocket-size spectrum analyzer of my own design.
Every 1dB drop in input amplitude resulted in 3dB drop of IMD₃ products. For power level of one hundred milliwatts, the IMD₃ is -80dBc or hidden in the noise floor.

The configuration with micro-coaxial cable or transmission line described in Motorola appnotes is another way, how to make power splitters, dividers, and baluns. It can split 0.5 Watt HF signal. The core loss is hardly measurable with 43 material. A band-pass output filter cleans split output signal harmonics and adds IMD₃ measurement consistency with repeatability.

Resources
[1] Pozar: Microwave engineering
[2] Fair-Rite ferrite materials, NJ
[3] muRata ferrite materials
[4] Motorola Appnotes
[5] va3diw balun notes
[6] w6pql notes
[7] w8ji notes