The phase response might be tricking the VNA into aliasing your measurement.

If it's a high sided LO, that mean the output frequency is being swept backwards too. So the GD (derivative of phase wrt. freq) still comes out positive.

R&S has a two-tone GD measurement technique. I have never used it, but they say it can even be used to measure satellite links. In this they talk about using a reference mixer fed from the same LO as your mixer, bringing it back up to RF, and I assume that would take care of the frequency inversion as well.

https://cdn.rohde-schwarz.com/pws/dl_downloads/dl_application/application_notes/1ef98/1EF98_1e.pdf

OK, I think I get it. The reason I asked is because of an "Excel mixer" I am playing with. Here is a page on it, it is still a mess

https://www.microwaves101.com/encyclopedias/excel-s-parameter-mixer

when I "mix" and amplifier's S-parameters to another frequency using high-side LO and look at group delay it is negative. But that is not a physical thing, just a figment of a slightly bogus way to simulate a receiver, right?

Thanks

You can achieve negative group delay in a medium with anomalous dispersion (like seawater). This just means the signal arrives prior to the carrier (group velocity > phase velocity; what makes it anomalous). Media with high insertion loss can also yield negative group delay such as just outside a bandpass filters passband.

It turns out that when you "mix" S-parameters with high-side LO, you need to take the conjugate of S21 and S12, and then group delay comes out as expected. I did not make this up, it is detailed in a paper by Dr. Dylan Williams...I'll post some updates on this shortly....

Thanks to Frauke for pointing this out. Here is the reference:

D. F. Williams, F. Ndagijimana, K. A. Remley, J. A. Dunsmore and S. Hubert, "Scattering-parameter models and representations for microwave mixers," in IEEE Transactions on Microwave Theory and Techniques, vol. 53, no. 1, pp. 314-321, Jan. 2005,