A regenerative receiver design was published in 2001 by N1BYT called the Wheatstone Bridge Regenerative (WBR) receiver. The design was interesting because it achieves good reverse antenna isolation (little oscillator energy can escape to the antenna) while still allowing antenna energy to enter and be amplified by the regenerative detector. The WBR achieves the reverse antenna isolation passively; no additional isolating amplifier stage is necessary.
If you're not familiar with the WBR circuit, AA1TJ writes more about it here: http://aa1tj.blogspot.jp/2011/08/ancients-have-stolen-our-inventions.html. Please see that article to understand the basic idea.
I was intrigued by this idea of passive reverse isolation but couldn't completely grasp how the circuit worked. Until now. A 1928 patent clearly describes the WBR operation: https://www.google.com/patents/US1667513. I have seen no references in online discussions to this patent; I found it myself after a lengthy period of investigation, while attempting to understand the operation of and research the background of the WBR receiver.
The key point from the 1928 patent's circuit description is that, unlike a normal regenerative receiver, the antenna signal does not directly excite the oscillator's tank at all! Instead it flows through the inductor's center tap and simultaneously goes through both arms of the bridge, which leaves both ends of the capacitor, and both non-grounded ends of the inductor, at the same voltage -- thus not exciting the tank. By going through the bridge, the signal reaches the input of the amplifying active device. Only after amplification and regenerative feedback does the signal then first appear at the tank.
Another interesting point is that the antenna energy is not filtered by the oscillator tank before it reaches the active device. This means that the antenna signals (in N1BYT's design, where the antenna load Z1 is an untuned inductance) encounter the non-linear active device without the benefit of filtering and thus AM blanketing or IMD are more likely to occur than in a receiver where the antenna signals are first filtered by the oscillator's tank. The 1928 patent uses an additional tuned circuit in place of N1BYT's untuned Z1 for additional input signal filtering.
So to summarise: Forward coupling from the antenna into the detector is high, because the antenna signal flows symmetrically from the tap point through both arms of the bridge (without exciting the tank) whereupon the signal reaches the amplifier input, is amplified, and is regeneratively fed into the tank for the first time. Reverse coupling from the oscillator to the antenna is very low -- almost zero -- because the inductor voltages at opposite sides of the center tap cancel, leaving no oscillator voltage at the center tap to excite the antenna.