As Thomas has noted, it's actually not difficult at all, following the design invented by Joe Dunsmore at HP, and implemented since in almost every broad-band VNA. I've seen a paper with a version that worked well from 100KHz to past 11GHz.
I've published the excerpt of Dunsmore's original publication here
This coupler works at high frequencies as a TLT (Transmission Line Transformer) built using coax or hardline, and at lower frequencies as a common mode choke. Dunsmore suggests using low-permeability high-frequency ferrites first, followed by high-permeability low-frequency ferrites. So there are three different effects at play here which all pan out to very broadband performance. The low frequency performance depends on the amount and permeability of the ferrite used. It fails at low frequency because the inductance is too low so the 50R impedance does not match the opposite leg of the bridge.
Dunsmore suggests an assymetrical bridge (same ratio, different impedances on two sides) to reduce through loss to 1.6dB, but Paul McMahon VK3DIP realised that a 6dB equal-arm bridge can use ferrite also in the corresponding arm, and that improves low frequency performance by a factor of 10. The VK3DIP design was copied by a German, which was copied in the Ukraine and then cloned (badly) in China, so you can buy the following product on EBay and Aliexpress. Although it is non-functional as shipped, it is easy to repair it to function as VK3DIP's intent, and with good semi-rigid cable trimmed properly and better resistors, it functions quite well from 100Hz through 3GHz. Still better than 20dB directivity at 2.4GHz:
transverters-store was the Ukraine company, now defunct. The fault with the Chinese clone is that one pair of 100R resistors are not joined at both ends to form a single 50R. I have test data on the stepwise improvements we made to one of these, and the final result. The 3GHz limit is due to poor trace impedance management and the large resistors being used. Just going to 0402 precision 50R resistors on a proper impedance-controlled PCB should get this up to 8 or 10GHz.
