Thanks for doing this!
(and now for the picky questions…)
From your spectrum plots, I’m surprised at the amount of out-of-band energy with the interdigital/cavity filters.
My bet – it’s signal leakage around the filter, such as through the plastic SDR shell. When I’m testing filters, LNAs, other sensitive RF stuff on my bench, I put the mess-to-be-tested in an enclosure made of fine mesh copper cloth. Makes a big difference, particularly with SDRs that have an internal AGC action.
You could try wrapping the RF part of the signal chain (SDR, filters, etc) in aluminum foil (copper would be better) connected to the outer braid of the antenna feed, and the USB shield.
Another approach is to put the whole lot (filters, SDR, Raspberry Pi) in a metal enclosure (metal box or a copper mesh enclosure) with a SMA connector on one end, and Ethernet and power on the other. The rule of thumb is no openings electrically larger than 1/4 wavelength of the highest frequency you’re dealing with. That’s 2.5cm for 3 GHz, pretty big. I like to go as small as I can, with copper mesh that has about 2mm spacing.
cheers
bob k6rtm