The "noisy regen" discussed elsewhere is simply a cross-coupled oscillator connected to an LC tank, with the audio taken off of the emitter.
Normally, for the two transistors, the base of one transistor is connected to the collector of the other, giving two base/collector wire pairs. One base/collector wire pair is normally connected to the LC tank (the "hot" wire pair); the other base/wire collector pair is tied to Vcc (the "grounded" wire pair).
In examining various circuits and ideas about the noisy regen by @Sean O'Connor , I realized that it is possible to disconnect the base and the collector of the "grounded" wire pair, in order to achieve some more useful results. One result seems to be that we can keep the base of the "grounded" wire pair still going to Vcc as normal, but for the collector of that wire pair, we insert an AF load resistor going to Vcc. We then ground the collector at RF with a 10 nF bypass capacitor, allowing the collector to be at RF ground as is required for oscillator operation.
By doing this, we now have an amplified collector output that can be used to extract the audio. This should give higher-level AF output than extracting the audio from the emitter.
An LTspice simulation confirms this. Here, the regen is just below critical threshold (adjusted for about 600 Hz -3 dB bandwidth, measured with an AC analysis and a Bode plot), resonant at about 1.887 MHz. A 1 uV peak-peak modulated signal is fed into the tank. AF output is taken at two locations: AF1 is taken from the collector load resistor; AF2 is taken from the emitter, as is normally done. Each AF output is then additionally filtered by a 4.7k/10nF RC network, and the output voltage is graphed.
As you can see from the simulation results, the red V(AF1) trace is larger than the turquoise V(AF2) trace, indicating that the AF1 output -- taken from the collector of the modified "noisy regen" -- is larger than the AF2 output, taken from the emitter.
These voltage outputs represent the unloaded AF output of the regen. But as long as the following AF amplifier has a high input impedance, then the results should still be similar even when loaded by the following AF amplifier -- the collector AF output should still be higher than the emitter AF output.
And we still have preserved the useful ability of the cross-coupled oscillator to reliably oscillate with any any LC tank. So this modified "noisy regen" would seem to be an improvement over the standard "noisy regen". I hope to try it in hardware soon.
To make a more complete analysis, it is also necessary to test the performance of the noisy regen in its normal configuration, without the collector resistor. It is possible that the mere presence of the collector resistor is altering the biasing conditions and hence changing the amount of AF output we get at the emitter. So, to test the noisy regen in its normal configuration, I removed the collector's AF load resistor and connected the collector directly to Vcc, as is normally done.
The AC analysis and Bode plot indicated that, without changing the regeneration control (the variable resistor in the emitter), the system was still under the oscillation threshold, but gain had increased slightly and hence the bandwidth had narrowed slightly (about 80 Hz narrower bandwidth, for a -3 dB bandwidth of 560 Hz without the collector resistor, compared with a -3 dB bandwidth of about 640 Hz with the collector resistor), and the resonant frequency shifted up about 300 Hz. This very small change in operating conditions indicated to me that the emitter's AF output would still be about the same as before.
A transient analysis confirmed that the AF output at the emitter did not noticeably change, indicating that the presence or absence of the collector resistor does not significantly affect the amount of AF present at the emitter. Therefore, adding a collector resistor and taking the AF off of the that resistor should yield higher output than trying to take the AF off of the emitters, as is normally done in the noisy regen.
It's worth noting that the amount of AF we are getting out, even from the collector, is still dismally small. In the best case (taking the AF off of the collector resistor), we have a microvolt-level input RF signal that gives about microvolt-level AF output. Reading some old articles about optimised single-tube regenerative detectors, it seems that microvolt-level RF input signals could give millivolt-level AF output, and when optimised, the detector gain (ratio of AF output voltage to RF input voltage) could be 7000 or more. So our little noisy regen circuit above, powered from 0.65 volts, is clearly lacking in the gain department compared to old tube detectors. The regenerative gain of the RF signal is obviously there -- the microvolt-level input RF signal gets regenerated up to a few hundred microvolts RF across the LC tank -- so the cause of the low AF output is the inefficiency of the envelope detection process.
The low detector gain doesn't really matter so much as long as we have a high-input-impedance AF amplifier that provides enough gain for these low-level AF signals. Better detector gain might be obtained with higher supply voltages and more careful biasing to maximize the nonlinearity and hence the AF output voltage.