The design is of a 50:75Ω autotransformer using a single Fair-rite 5943003801 (FT240-43) core. Note that this data is NOT for an Amidon #43 core.
The winding is a total of 6t tapped at 5t, wound as two close wound layers. The prototype used a single conductor stripped from LAN cable. The winding configuration is an attempt to minimise leakage inductance for improved broadband performance.
This is unconventional, but then the conventional designs on Youtube etc do not usually have credible evidence of low InsertionVSWR or other performance.
Above is the prototype with load (2 x 150Ω resistors measuring 74.8Ω at DC). Note the very short connections (see Length matters .)
Above is a sweep with the output terminal shorted to the common terminal. Total leakage inductance is taken as 45nH. This is an important design parameter.
Above is a SimNEC model of the transformer and comparison with measurement of the uncompensated prototype.
The value of parameter Ll in D1 is half the measured total leakage inductance. Parameter cse was adjusted to calibrate the model uncompensated response (cyan) to the measured uncompensated response (green.)
The value of Ccomp was adjusted to obtain best compensated VSWR response.
Modelled core loss at 100W continuous is <5W. The FT240 sized core is capable of about 10W dissipation in free air at room temperature, so the core has headroom to accommodate a worst case VSWR=2 mismatch.
For a working transformer, I would use PTFE insulated silver plated copper wire, wound in the same geometry. Thicker conductor decreases leakage reactance, thicker insulation and loose windings increases leakage reactance.
The transformer would handle 1kW sensibly processed SSB telephony again at room temperature (see Average power of SSB telephony .)