It seems most phasing receiver designs I've seen use 90 degrees shift in the LO and 90 degrees shift in the AF chain. Some use 90 degrees in the RF signal path and 90 degrees in AF. An interesting variant, TinySDR, apparently uses 45 degrees in RF, 90 degrees in LO, Polyakov mixers, and 90 degrees in AF (via software). (EDIT: I'm not sure if I interpreted the TinySDR phase shift scheme correctly - corrections would be welcome.)
One unusual combination that I haven't seen implemented is 90 degrees in the RF signal path and 90 degrees in the LO, with no phase shifting in the AF chain. This combination is described here but with no implementation: http://www.qsl.net/g3cwi/dc.html
Taking the example of TinySDR, if we replace the RF RC phase shift network (which, I think, may provide a 45 degree shift) with a transformer and amplitude-balancing pot as described earlier in this thread, then we could get wideband 90 degrees phase shift in the RF path, while needing only one adjustment as we tune up and down the band. Since the LO also produces 90-degree phase-shifted I/Q outputs, we mix (using normal mixers, not frequency-doubling Polyakov mixers) the 0 degree RF signal with the I and the 90 degree with the Q, add the AF outputs, and it seems we're ready to go for LSB reception. Swap the I/Q for USB. No AF phase shifting needed!
Seems like this could simplify the overall design. Thoughts?
EDIT: this won't work (and explains why I couldn't find any implementations of the idea). The reason is that we're essentially multiplying sin*sin and cos*cos in this scheme, both of which yield cosines. Therefore the AF signals generated in this way cannot be summed to eliminate one sideband.