But I think direct conversion from 1 GHz is a bad idea, because it will be very hard to get two exactly the same mixers and low phase error between LO IQ lines due to very high frequency, even 1 mm wire difference or a small solder drop will lead to a huge phase unbalance. Such receiver will have bad performance and will be very hard to design PCB and tune it.
The need to match the mixers and get the correct phase relationship between the LOs is of course a big part of why this is often done on-chip in the GHz range. But I see no reason to consider it intractable for a board level design.
We're not talking about putting something together with bits of wire. A PCB layout for the sorts of small surface mount components being discussed here will have tolerances much tighter than 1mm. And even a 1mm length on a typical PCB microstrip is only about 2 degrees of phase shift at 1GHz. If the LO signal paths are laid out to be identical except for a specific
difference in length between them, it should be straightforward to get the phase error within a fraction of a degree. Solder joints etc should have little effect, as they will be very similar between the two paths.
Of course, the constraint of this approach is that the phase relationship would only be correct at a specific LO frequency. So this is only viable if the receiver is able to use a fixed LO, with any further tuning being handled in baseband processing.
There will also be differences in conversion gain between the two mixers, which will affect I/Q balance, but that can be corrected for at the baseband stage.
The better approach is to do down conversion to some IF, for example 10.7 MHz and then feed to quadrature mixers.
I don't think it's necessarily a better approach.
If you downconvert a signal to a 10.7MHz IF, then that conversion will be equally sensitive to interference at an image frequency that is 21.4MHz away from the signal of interest (either above or below it depending which side you put the LO). At HF or low VHF, that image frequency will be well out-of-band, and far enough away from the desired signal that it can be easily rejected by practical filter designs at the RF stage. At 1GHz though, an image 21.4MHz away is much closer to the signal in relative terms, and much harder to reject adequately. For some standard frequencies, there are SAW filters available that may do the job, but otherwise it may be impractical to achieve the necessary rejection.
In this regard a direct conversion architecture is a major advantage, since you get built-in image rejection without difficult filtering requirements at the RF front end.