Hardware and vintage methods:
There is the phasing method of using two gilbert cells each with L.O. and signal respectively phase shifted by 90 degrees.
Interested in the history , I found a 1946 paper by R.B. Dome "Wideband phase shift networks" ( will come up on search)
where he described audio filters that can give a +45 and -45 degree phase shift of suitable accuracy, directly on the baseband audio.
His filters do not hold 45 degrees over the baseband, but he claimed the pair of them track with about 45 degrees difference over audio range.
Assuming I understood his filters, the model in qucs was unable to achieve phase shift accuracy to sufficiently suppress the unwanted sideband.
However using that idea to phase shift L.O and incoming signal for a 14 MHZ down converter works very accurately, just with two pairs of RC networks.
(tracking with fraction of a degree difference over the 20 metre band.)
I have artworked a circuit to try but not sure if I will build it as it is also easy to filter the wanted downconverter signal .
Incidentally I am using a Nooelec AD9850 40MHz DDS Function Generator Module to injection lock a vintage tube receiver.
Trying various methods, the best way so far is by using the 6BE6 converter tube and feeding the DDS into the control grid with the oscillator otherwise a conventional Hartley on the oscillator grid , which stays on level both when locked to DDS or with DDS off.
It is working well from 7 MHz to 18 MHz, but does have spurs (usually out of the way) and I will try a better filter and shielded box.
Attached is screenshot of GUI written in bash and gtkdialog to control the AD9850 in serial mode.
https://app.box.com/s/v24an4ea1iiib9jvcpyeh49j9wx0518ihttps://app.box.com/s/w328smgvhyn0ya4eb3bb0yvbsvx06fuahttps://app.box.com/s/jjyuodfiqw90o39pabxoviwwn7gpdcn3https://app.box.com/s/tf7936ph90rf4vljgbo427sevbtn6i85