But you seem to be asking for a simple way to achieve it, and to do so for any and all VFO frequencies.
Im sure it probably IS possible to build an image canceling DC receiver using polyakov type mixers but I dont beleive it would be simple or "non-critical"!
90 Degree phase shift is much easier.
I think you're right about 90 degrees being easier - at least, it's easier to find information about. Quick summary of what I've found so far:
To create 45 degree phase shift we can:
1) use one series RC network at the VFO, between signal and ground, with capacitor at top. In-phase signal taken from top of network, "quadrature" (octiture?) signal taken from RC junction. R is set equal to the capacitive reactance. Covering different frequencies requires changing both the phase network (R and/or C) as well as an amplitude-balancing adjustment.
2) use two series RC networks, one at each branch of the split RF signal path. Both RC networks are adjusted for a total phase difference between the two signal paths of 45 degrees. Covering different frequencies requires tuning (possibly) both RC networks, and an amplitude-balancing adjustment. (EDIT: I don't think this option is needed in practice. Polyakov mixers double the LO frequency and LO phase shift, and hence need 45 degree phase shift at the LO. But Polyakov mixers don't double the RF, so we should use a 90 degree phase shift in the RF path with Polyakov mixers.)
3) use a 90 degree phase shift arrangement for the VFO (see below) plus a 45-degree RC phase shift for the RF signal. (I saw a design, TinySDR, that appears to do this; I hope I did not misinterpret the intention and function of the phase shift networks.)
4) other ways, perhaps using logic gates? I don't have any information on this at present.
To create 90 degrees phase shift we can:
1) use one RC series network at the VFO, connected across a transformer secondary winding, and with the RC junction grounded. Signals at opposite ends of the secondary winding will by definition be 90 degrees out of phase since the R and C are in series. Covering different frequencies requires only one adjustment: an amplitude-balancing adjustment of either R or C so that R is equal to (?) the capacitive reactance.
2) use 2 series RC networks in the RF signal paths, adjusted for a total of 90 degrees phase shift. As with the 45 degree case, covering different frequencies requires potentially changing both networks, and an amplitude correcton.
3) use flip flops and a f*4 VFO to generate I/Q signals over the entire VFO range with no phase adjustments required. Output will be square wave.
4) use a DDS to directly generate the I/Q signals.
5) other ways not yet researched...
I thought Polyakov mixers, given their good reputation, performance, simplicity, and interesting properties like only requiring a f/2 VFO, might simplify an all-band DC receiver design. But it seems that if we want a phasing
DC receiver, the Polyakov mixer may actually complicate things due to the necessary 45 degree phase shift.
Corrections, additions, suggestions are welcome!