That's a neat bit of software!

 

But I've noticed in all the all-pass design procedures, there are choices of optimising, Monte Carloe-ing, adjusting and tuning the RC  part of the networks.  But none seem to take into account the effect of mismatch of the nominally matched pair of resistors around each opamp.   Or of residual amplitude mismatching.

 

But it does seem that if you are prepared to spend the time getting all the Rs and Cs within a micro percent of the correct values, then apply the same effort to amplitude and phase matching the channels, a quite phenomenal performance wrt. sideband suppression is theoretically possible.

 

But I guess in any practical scenario, most constructors will tweak for an acceptable performance and leave it alone.   For upconverting at LF, we're only interested in narrow band signals - the entire WSPR segment is only 200kHz wide - and a simple 2 stage (one opamp in each of I and Q) is adequate.  In that case a pair of preset or variable resistors can be tweaked quite quickly.   40 - 50dB supression seems almost guaranteed over a 100Hz using a predictable mixers like fast CMOS switches, with a comparable level of carrier suppression.  I've built two now, one with FST3125 and a balanced output transformer,  and one with a dual quad switch, FST3257.  The latter gives somewhat better performance, but is slightly more complex.

 

Andy
www.g4jnt.com