This was my original prototype design to see how well a simple superhet could be made to function, and is still the "guinea pig" for testing new ideas.
In order to derive the maximum performance, i decided to go for a fairly high IF frequency to eliminate images. The usual 1700kC is low enough to achieve reasonable gain, but selectivity suffers. Both these issues are nicely avoided by using IF regeneration. Its also out of the range of interferring AM broadcast stations.
The best and latest in valve mixer technology was in the 60s when Philips created the ECH81 triode/heptode. This valve had little frequency pulling between mixer and oscillator, and is fairly tolerant of oscillator voltage level into grid 3. By using a variation of the Vackar type oscillator, thus keeping the Oscillator level constant over the tuning range, the effective gain is also kept constant over a wide tuning range. The triode oscillator section is fairly low gM, and has to be driven hard to get into the higher SW bands.. The ECH84 came later, and although designed primarily for synch seperation and gated AGC in TV service, appears to exhibit slightly better performance as a mixer in this application.
Because of the very high impedance of the dynamic load on the mixer anode, a result of the "Q" multiplication effect of the regenerative IF, the conversion gain is increased dramatically, therefore, an RF attenuator is required to limit overloading in the IF. The high conversion gain also allows a relatively lossy passive bottom coupled dual tuned bandpass filter on the input, again improving mixer performance by narrowing the RF bandwidth, thereby reducing out of band noise. This is quite noticeable when comparing this to my simpler RX having just the one RF tuned circuit.
Using the Pentode screen voltage to vary the IF gain, results in very smooth and controllable regeneration with little frequency shifting. I did try a second IF bandpass filter between mixer and IF in an attempt to improve the IF shape factor. This worked to a point, but something I had not envisaged was the "double humping" when adjusting the regeneration. As the "Q" increased, the effective coupling between the two coils increased, causing extreme interaction and effectively splitting the signal into two several kCs apart at the critical threshold.
I would like a 3:1 transformer in the detector anode to gain a free voltage stepup to increase the audio gain, have yet to find anything suitable.
This radio I used 5 pin DIN sockets , and PVC plug in coils, an easy way to change bands and makes superhet tracking design slightly easier. Air cored coils and air spaced tuning capacitors are the ultimate for stable and "drift free" oscillators. The best tuning capacitors are made from brass, these have less thermal expansion than aluminium vanes, good luck finding these. Polyvaricons are abysmal, only slightly better than varicap diodes.
Having no tuning slugs improve the thermal effects on the coils, but also makes adjustment very time consuming, needing to add or remove turns to get it just right, but its worth it to maintain the high "Q". This set of coils covers 5 to 15megs, tuning is touchy. More coils and less tuning range would be the next goal with this one.
Some may find this approach interesting, I know back then i had lots of fun building and experimenting with it.
I suggest you build up a regen using a valve , this is the easy way to visualise whats happening I think. Transistors are low impedance devices and in my experience dont match well with high impedance, high Q coils. Even if the transistor base is tapped down the coil, it may impedance match ok, but the coil is still loaded with the equivalent of a semi shorted turn or two. Valves operate with voltage change, transistors require current changes, and transistors are always forward conducting, always loading the previous circuit. , valves are normally not..
I have built transistorised regens , they have never worked as well as valved equivalents in a direct comparison, The best transistor one I have working uses a dual gate mosfet, but is compromised by the relatively low voltage headroom, limiting the dynamic range. DG Mosfets are getting quite rare now, and harder to find.
In all my valved superhets, I have not got your problem of regen changes with front end tuning, there are no IF gain changes . The oscillator and RF circuits all tune together. What I do have though is a bloody annoying change of regen setting from say threshold at 4.5 megs, then if I go to 12 megs, I need to change the setting very slightly.... I have tried unsuccessfuly to fix this as it would neatly get rid of the regen control. The cause is oscillator leakage at the mixer anode, the regen IF is more effected as the oscillator frequency gets closer to the IF frequency. Thats another reason I tried a double tuned BPF between mixer and IF, to try and remove as much of the osc signal as possible.. This only is a nuisance with megahertz frequency changes, ... tuning over say a 1 meg bandwidth requires no change in the regen setting at all.
So, I live with that small annoyance , after all, it is a minimalistic radio.
Other very good reasons to try valves is the ease to get them working, and what can be achieved with just a few stages. There is always the "cool" factor as well. And a biggy is the dynamic range, purely down to the higher voltages used.