I'm pretty sure thermal noise is half phase noise and half amplitude noise. In other words, -177 dBm/Hz from phase noise and -177 dBm/Hz from amplitude noise gets you -174 dBm/Hz thermal noise. So technically it's a low phase noise splitter just because it has low loss.

Absolute lower limit of phase noise for a passive, room temp device is -177 dBc/Hz, and if that splitter is 0.6 dB excess loss, I'd think that would put the phase noise at -176.4 dBc/Hz, but this lists -175 dBc/Hz.

That said, I'm sure piezo effect from substrate dielectric or capacitors will cause phase noise, so maybe they use low-glass substrates or certain caps.

This is interesting. Maybe there is more to it:

https://www.researchgate.net/publication/321324476_Phase_noise_measurements_with_a_cryogenic_power-splitter_to_minimize_the_cross-spectral_collapse_effect/link/5b1be357aca272021cf46626/download

So the above paper describes how an isolation resistor in a power splitter introduces a differential thermal noise that messes up (underestimates) the DUT phase noise in a cross-correlation phase noise analyzer. They fix it by cooling the splitter to 4K but 77K may also work.

Maybe the Holzworth splitter is a purely reactive type (though not with 38 dB isolation), specifically needed for their analyzers, though the paper mentions issues with reactive types too, though I don’t understand what they mean.

It’s also neat to see the splitter isolation decrease near 4K, which makes sense as the resistor is changing.

Note I’m using the term power splitter, but maybe a power divider would behave differently.

Thanks for the reply! I only checked back now ... the project has been on hold due to other issues.

As I understand it, the thermal noise floor in a 50 ohm system is at an absolute level of -174dBm/Hz.

However, the Holzworth spec says -175dBc/Hz (max -170dBc/Hz), i.e. it's relative to the carrier (and specified at a particular offset frequency from the carrier). I don't think thermal noise would be at a level that's relative to the carrier, would it?

Anyway, thanks for the link to the article, I'll have a look at it.