I'm continuing the previous discussion (started in the thread about a tube-based regenerative superhet) here into a new post.
There are no final schematics yet, since the circuit is still a work in progress. This thread will hopefully describe that progress.
In my linked post above, I was discussing with @dayleedwards about the general problem in a regenerative superhet (with a fixed IF) that the LO tuning can change the regeneration level slightly.
I mostly fixed this by using an amplitude-stable LO, the cross-coupled BJT oscillator. I connect this LO to an unbalanced, single-BJT mixer (through a link winding from the oscillator tank onto the BJT emitter), and take the mixer output from the collector, which leads to the regenerated LC tank. In this configuration, the LO tuning almost does not affect the regeneration level. But, I noticed something funny.
I previously wrote:
With this new, mostly-amplitude-stable LO, I can indeed set the regeneration so that it is just below threshold over the entire tuning range, and leave it there when I tune the LO over its entire range. There is a slight decrease in regeneration as the set is tuned higher in frequency, but this is not enough to significantly affect the sensitivity of the set, as tested with signals from a small ferrite rod antenna.
I thought that the decrease in regeneration, when tuning the LO to higher frequencies, was due to the oscillator waveform, when tuned higher, placing an increased load on the regenerated tank and reducing the regeneration.
But just now I noticed that the decrease in regeneration was slow and gradual. To combat a separate problem of unwanted AF oscillation due to supply voltage ripple, I recently have been modifying a high-C RC decoupling network in the power line of the receiver. The high-C RC network serves to temporarily stabilise the supply voltage if the AF power amplifier suddenly draws a lot of current on AF peaks, by preventing sudden changes in supply voltage and only allowing the supply voltage to change very slowly.
Therefore, a slow -- and not instantaneous -- change in the regeneration level is indicative of a slowly changing supply voltage. And I observed that the regeneration level changed slowly, not instantaneously, as I tuned the LO. Even when snapping the LO from maximum to minimum very quickly, the regeneration level changed not instantly, but slowly over a few seconds. So I now think that what is actually happening is that at higher frequencies, the oscillator is pulling more current, which causes the supply voltage to sink (after a second or two as the RC network stabilises), which finally causes the regeneration level to drop lower. Note that my power supply is a single 1.2 volt cell, so I don't have the luxury of regulating a higher voltage down to a lower stable voltage.
With previous oscillators (Colpitts, Hartley, or Vackar), I also had the problem of LO tuning affecting the regeneration level of the fixed-frequency regenerative IF stage. These oscillators will change their output amplitude with frequency. I had assumed this varying output amplitude was causing the change in regeneration level. But now I'm not sure anymore.
To summarise, there seem to be two factors at play here:
Sagging supply voltage at higher LO frequencies causing a decrease in regeneration.
Differing LO output levels depending on tuning, which leak through the unbalanced mixer into the intermediate-frequency LC tank, thus placing a differing load on this regenerated LC tank and changing the regeneration level.
For testing, factor 1 can be eliminated by powering the LO off of a separate battery. Then, tests can be conducted again with a Colpitts, Hartley, or Vackar oscillator (which will change the output amplitude with frequency). If these oscillators no longer change the regeneration level with LO tuning, then we can conclude that factor 2 is not significant.
Continued in part 2.
Single-diode voltage regulation actually works in this 1.2V receiver!
I am now powering all stages directly from the 1.2 volt battery, except for the following three stages: the cross-coupled LO, the cross-coupled regenerative IF stage, and the last IF amp attached to the regenerative stage's tank. Those three stages are running off of 725 mV -- Vcc goes through 50 ohms through a forward-biased 1N4148 to ground, and the "regulated" 725 mV supply voltage is taken off of the top of the diode (with a large-uF filter capacitor also at the top of the diode). We're using about 24 mA of current with this simple "voltage regulator" arrangement (1.2 volts through 50 ohms through the diode to ground).
It was enjoyable to confirm that the good old cross-coupled oscillator -- for both the LO and the regenerative stage -- can run off of this small 725 mV. The one IF amp connected to the regenerative stage's tank is also, necessarily, powered off of 725 mV, which very likely leads to poor gain in this IF stage.
However, the poor gain of the last, 725 mV-powered IF stage is more than offset by the massive gain increase in the rest of the receiver. Thanks to the single-diode voltage regulation applied to the LO and the regenerative stage, these stages are now sufficiently stabilised (with regard to supply voltage) that no motorboating occurs and all of the other stages can now be powered directly off of the 1.2V battery, with no RC decoupling network needed and hence no voltage drop across the RC network. This full 1.2V power for the 3 IF amps, for the detector stage, and for all the AF stages leads to a quite noticeable boost in volume -- the no-signal mixer noise is now uncomfortably loud in the headphones (I will add a manual volume control potentiometer later).
As for frequency stability: Reception of strong CW reception still causes some chirp, but it seems less than before. SSB signals are more intelligible than before, but there still is some warbling present.
So single-diode regulation has improved the receiver gain and has improved the frequency stability, though the oscillation frequency of the LO and regenerative stage still can be influenced slightly by the dynamic AGC action. Further improving the frequency stability would require buffer stages, but I don't want to increase the receiver complexity any further, so I'm going to accept the current frequency stability as is.
Here are some thoughts and experimental results about voltage stability.
The two most voltage-sensitive stages -- the cross-coupled oscillator for the LO, and the cross-coupled oscillator for the regenerative IF amplifier -- can probably run on quite low supply voltages, such as 0.7 volts.
By using a forward-biased silicon diode in series with a resistor, and connecting the free end of the resistor to Vcc, the voltage at the top of the diode will be, more or less, 0.7 volts, due to the voltage drop across the diode. The available voltage and current will likely be enough to power the cross-coupled oscillator. The voltage available on top of the diode will not be a perfectly stable 0.7 volts, of course -- there will be some ripple if the Vcc supply voltage has ripple. Some LTspice simulations (which I can't seem to locate now) indicated that with a 1.2 volt supply voltage, a millivolt-level voltage ripple on the Vcc line would be reduced by about a factor of 10 at the top of the diode, thus providing a crude form of voltage stabilisation, even at the low supply voltage of 1.2 volts.
Separately from the above considerations about diode-based voltage regulation, I did some AGC tests in hardware, increasing the standing AGC bias to a very high quiescent level -- thus greatly attenuating all signals, and making the AGC level fairly stable since no signal was now strong enough to "activate" the AGC any further. This then yielded clear SSB and CW reception. So it's definitely the dynamic AGC action that is causing the frequency instability when listening to SSB and CW.
Even if the supply voltage could be stabilized with a diode, it's still possible that AGC action might undesirably change the LO frequency and the oscillating regenerative detector's frequency. It's conceivable that when AGC changes the gain of the IF strip, then the input impedance and output impedance of the IF strip might also change, and these impedance changes might then affect the frequency of the LO and the oscillating regenerative stage. Avoiding this might then require buffer stages with stable impedances, thus increasing the receiver complexity.
For SSB or CW reception, probably the simplest practical method is simply to disable AGC when listening to SSB or CW. But some manual gain control would still be needed to avoid overloading the regenerative stage. Therefore, disabling AGC while still allowing manual gain might be most easily accomplished by adding a switch in the AGC signal line, which when closed allows normal AGC operation, and when opened disconnects the AF signal from feeding back into the AGC transistor. In this open-switch condition, then the AGC transistor's bias (and the IF gain) is determined solely by the variable resistor on its base, which can then be used as a crude form of manual IF gain control (crude because its action will be exponential; a small increase in base current will cause a large increase in transistor current and hence a large decrease in IF gain).
Since it's fairly easy to do, I think I will try using the one-diode voltage regulator to power three stages: the LO, the regenerative stage, and the last IF amp attached to the regenerative stage's tank. I can foresee four possibilities: (1) the 0.7 volts supplied by the one-diode regulator is insufficient to power the LO or the regenerative stage; (2) the set may start to motorboat again due to the different supply voltages in different parts of the receiver; (3) it may work and SSB/CW reception may be possible without distortion; or (4) SSB/CW reception may still be distorted due to insufficient voltage regulation, due to impedance changes caused by AGC action, or due to a combination of both.
If that fails, then I'll just go with the switchable AGC option.
Part 11: Voltage stability optimisation
This is a small update from part 10 concerning voltage stability.
Since this receiver is powered off of only 1.2 volts, there is no good way to regulate the voltage, since voltage regulation chips or diodes would require a much higher voltage in order to be able to regulate it down to a lower voltage.
Powering everything directly off of Vcc leads to motorboating due to internal battery resistance and small Vcc voltage ripples causing unwanted AF feedback and oscillation.
Therefore, part of the receiver must be powered downstream of an RC decoupling network (to isolate and filter out Vcc voltage ripples and prevent them from reaching another part of the circuit where they can cause a feedback loop), and part of the receiver must be powered directly from Vcc.
The final power supply scheme I settled on is shown below. The most voltage-sensitive parts of the receiver are the LO and the regenerative Q multiplier. Even small voltage ripples on these stages will cause noticeable shifts in the LO frequency, the oscillation frequency of the regenerated IF, and the amplitude of the regenerated IF signal. Because these stages are so sensitive to supply voltage variations, these stages should be powered from the most stable voltage available, which is the full Vcc voltage across the battery. In addition, the last IF amp shares the LC tank with the Q multiplier, so it too should be powered directly from Vcc. This scheme is the best we can do to ensure voltage stability, and hence frequency and amplitude stability, of the LO and Q multiplier stages. This scheme isn't perfect, because there is still some internal battery resistance that may cause the actual voltage across the battery to vary with the current draw, but this cannot be avoided.
All of the other parts of the circuit are then powered through a 100 ohm/1000 uF decoupling network.
Results:
No motorboating or unwanted AF oscillation was observed.
Distortion: Now, because the AF power amp is powered through an RC network, very strong signals can cause the final power amp to generate distorted output. This was previously determined to be caused by the presence of the RC network. The distortion sets in at a lower signal level than when the power amp was powered directly from Vcc. But, the good news is that the AGC can compensate for this. The AGC bias just needs to be set a little higher (setting the AGC transistor a little more into conduction), resulting in less IF gain. This then keeps the IF signal strength below the level where it would cause AF distortion in the final power amp.
Reduced AGC dynamic range: Ideally, we want the IF gain to be set so high that mixer noise (which corresponds to the weakest signal that can be received) to be amplified to a loud and comfortable listening level. However, due to the distortion described in point 2 above, we can no longer run the IF gain at this high of a level, and need to reduce the quiescent IF gain to avoid AF distortion on strong signals. Unfortunately, by reducing the quiescent IF gain, this also reduces the gain on very weak signals like the mixer noise. This means that weak signals (and mixer noise) are only weakly audible. Previously, when the AF power amp was powered directly from Vcc, very weak signals and very strong signals could be regulated by the AGC to be about the same strength. But now, since very weak signals are reduced in strength to allow sufficient AGC action on very strong signals, the end result is a reduction in AGC dynamic range.
Frequency stability: Setting the Q multiplier into a moderately-strong oscillating condition, strong CW signals still exhibit a noticeable of "chirp" or LO frequency change when the CW signal changes from off to on. I couldn't get many SSB signals in my reception test this evening, but the few SSB signals I heard seemed to be somewhat intelligible, but not very clear. There still seems to be some warbling in the speech, which again likely indicates some fluttering of the LO frequency caused by varying current draw as the AGC is dynamically responding to the signal strength and modulating the IF gain. Anyway, I think there is nothing more that can be done to combat this phenomenon, as long as the receiver is limited to being powered off of a single 1.2-volt cell. It's as good as it's going to get.
Excessive AF gain: As a test, powering the 3-transistor direct-coupled AF amplifier directly from Vcc (and not from the filtered Vcc_2) led to higher AF gain, but the AF gain was actually too much for comfortable listening and led to noisy audio.
Excessive IF gain: As a test, powering the IF stages directly from Vcc led to noticeably higher IF gain and a noticeably higher noise level, because the no-signal mixer noise was now being amplified to a very high AF level, which reduces the available dynamic range for signals above the mixer noise. The excessive IF gain could be more-or-less compensated by adjusting the AGC bias even higher.
Optimum gain: The excessive gain described in point 5 and point 6 above indicates that the current receiver (as shown in the above schematic) seems to have reached an almost optimum amount of gain -- enough for loud headphone listening, but not so much that there is excessive noise in the no-signal condition, which might overload the Q multiplier or overload the AF amplifiers. However, this optimum amount of gain may partially depend on the Q of the inductors in the IF LC tanks. If one uses inductors with a different Q than my current inductors (which have unknown Q), then this might change the total IF gain and lead to insufficient of excessive IF gain. For the IF tank inductors I am currently using shielded, red OSC coils, intended for the LO of AM BCB receivers, that I rewound. At some point, I hope to try hand-wound toroidal inductors for my IF tanks, although this will make aligning the IF tanks more difficult as toroidal inductors aren't really adjustable (except for adjusting the turn spacing).
Now, I think the core parts of the receiver are working about as well as can be expected. Next steps will be minor improvements:
Adding an LED to show signal strength (the AGC current).
Adding a manual volume control, which may need another AF buffer stage.
Implementing LO-RF tracking.
Adding a bandspread capacitor. The bandspread capacitor, like the main tuning capacitor, will tune both the LO and the RF sections simultaneously, and must be a dual-gang capacitor. I'll probably aim for around 500 kHz of bandspread at the low end of the tuning range (around 5 MHz), which will then probably cover 2-3 MHz at the high end of the tuning range (around 20-ish MHz).
Maybe adding a BFO. Because the Q multiplier can already be set into the oscillating mode for CW/SSB reception, the only reason to add a BFO would be for single-signal selectivity, running the Q-multiplier below threshold as a narrow filter and tuning the BFO to one or the other side of the filter peak to emphasize one or the other sideband. I'm not sure if it will be worth the trouble to add a BFO in this receiver, since it's not really optimised for CW/SSB listening due to the lack of voltage regulation and some small LO drift.
Some other major improvement ideas, which I may or may not tackle, would be:
Increase the AGC dynamic range by controlling two, not just one, IF stages via AGC.
Investigate whether different LO topologies (instead of the cross-coupled oscillator) might have improved frequency stability when faced with ripples on the Vcc voltage. The cross-coupled oscillator's frequency is somewhat sensitive to voltage, whereas traditional LC oscillators may be less sensitive. An LTspice simulation might be a quick way to check this.
Superhets with common-base IF amplifiers (like mine) seem to be unusual. But I found a circuit with an IF strip similar to mine, using common-base amplifiers. On p. 17-25 of the following document, a "pocket superhet" is described: https://worldradiohistory.com/BOOKSHELF-ARH/Author-Groups/Edwin-Bradley/Transistor-Circuits-for-the-Constructor-No-1-Edwin-Bradley.pdf . The text states that:
I have to say that this mirrors my experience also. I have 4 tuned IF stages. When I tried common-emitter stages, the whole IF strip oscillated. It seems that neutralization, with some empirically-determined negative feedback capacitor(s), is needed to inhibit oscillation. Also the collector output generally needs to be connected to an inductive tap on the LC tank (not the top of the LC tank) to avoid oscillation; this requirement for a tapped coil makes self-wound IF tanks more troublesome.
Part 10: AGC
AGC is working. Latest schematic is below.
AGC circuit description
AGC voltage is taken off the collector of the last AF power amp via a 100 nF capacitor. This feeds backwards into Q141, which is biased non-linearly as a detector. Increasing average AF voltage causes an increase in average current flow through Q141, lowering the base bias on the first IF amp Q132. VR18 sets the bias for the AGC detector Q141.
The AGC bias setting is somewhat tricky:
Set AGC detector bias to zero.
Set the regenerative stage somewhat below the oscillation threshold.
Disconnect and reconnect the mixer from the IF amp and confirm the mixer noise becomes inaudible and then audible again.
Slowly increase AGC detector bias until noise goes down, indicating a reduction in IF gain. Then decrease AGC detector bias slightly until the noise comes up again.
Tricky part: as the AGC begins to take effect, it will affect the supply voltage seen downstream of the R192/C195 decoupling network on the battery. This in turn will affect the regeneration level. Therefore, the regeneration level must again be adjusted to be somewhat below the oscillation threshold, and the entire procedure should again be repeated from step 3.
To test the AGC, hold a noise-generating device (like a modern smart phone 😀) near the RF tank and mixer. The noise level audible in the headphones should barely change. If the noise level jumps up, it indicates that the AGC is not working properly.
Where to derive the AGC voltage, and its interaction with manual AF volume control
AGC voltage is currently derived from the output (collector) of the last AF power amp.
An attempt was made to take the voltage from the input (base) of the last AF power amp, or equivalently from the output of the 3-transistor direct-coupled AF voltage amplifier Q128/Q129/Q130. This did not work and AGC action was not noticeable.
So the AGC voltage needs to be derived from the output of the last AF power amp. This then requires that the manual AF volume control for the receiver be placed at this same point in the audio chain -- right at the headphones. It is not possible to place the manual AF volume control anywhere before the last AF power amp, because in that case the manual AF volume control would then undesirably affect the AGC voltage.
Undesirable voltage fluctuation on the regenerative stage
The Q-multiplier Q139/Q140 is powered downstream of the Vcc decoupling network R192/C195.
Unfortunately, as the AGC is operating to vary the gain of the IF stage, the voltage seen downstream of the Vcc decoupling network will fluctuate due to the varying current draw.
This causes fluctuation of the voltage available to the regenerative stage, affecting the regeneration level and possibly also slightly shifting the frequency of the regenerated tank.
It is not possible to power the Q-multiplier Q139/Q149 (and the attached IF amp Q135) directly from Vcc. Attempting to do so causes AF motorboating of about 5 Hz.
Garbled SSB reception due to AGC affecting regeneration level and frequency of the regenerative stage
SSB signals around 7 MHz were tuned in.
Case 1. Pushing the regenerative stage over the oscillation threshold yielded warbling, barely-intelligible speech, consistent with a rapidly varying signal strength and frequency, likely caused by the AGC action. When tuning in to strong CW signals, "chirp" could be heard as the AF signal strength caused the AGC to reduce the IF gain, affecting the regenerative stage's frequency and causing the CW signal's received pitch to change.
Case 2. Running the regenerative stage below threshold, and tuning a separately-powered BFO (a GDO) to 7 MHz -- the signal frequency -- yielded good and clear SSB reception with almost unnoticeable levels of warbling or distortion. This indicates that the LO frequency is sufficiently stable even in the face of varying current draw caused by AGC.
Case 3. Tuning the BFO to 3 MHz (the IF) again yielded garbled, warbling, barely-intelligible speech. Speech quality was very bad, but slightly better than Case 1. Presumably, the AGC action is causing rapid fluctuation of the regeneration level (due to voltage fluctuation of Vcc_2) and possibly rapid frequency shifts of the regenerated tank's frequency.
Though SSB reception is possible in Case 2, it is not desirable to run the BFO at the signal frequency. It would be more desirable to run the BFO at the IF frequency. But the regenerated IF amplitude (and possibly frequency) are unstable due to Vcc_2 being unstable. And we cannot connect the regenerative stage directly to Vcc because this causes motorboating.
I need to think about this some more. A separate battery to power the regenerative stage would probably work. But having been successful thus far in using only one 1.2V battery, I would prefer to continue using only one 1.2V battery to power the set.
A compromise solution might be to connect the AF power amp to a separate RC decoupling network -- which I previously determined introduced AF distortion at high AF levels. But now with AGC working, the AGC might be able to keep the signal levels below the point where bothersome AF distortion begins. Then, having isolated the power-hungry AF power amp with its own RC network, I might then be able to connect the regenerative stage directly to Vcc without the emergence of unwanted AF oscillation.
Other problems: 1 Hz AF oscillation
When a very strong signal is tuned in -- like the local BFO signal -- an AF oscillation of about 1 Hz can occur. Probably, the RC decoupling network on the battery needs to be increased to combat this.
To stabilise the LO voltage when adjusting IF gain, I'm now powering the LO (Q54/Q55 in the below schematic) directly from Vcc instead of the filtered Vcc_2. This has mostly eliminated the problem of the LO changing frequency when the IF gain is adjusted. I still think there is a several-hundred Hz shift (tested with the LO at about 7 MHz) that may occur when IF gain is altered from zero to maximum. This is a lower frequency shift than before (when the LO was powered from the Vcc_2), but still is potentially problematic for SSB signals. So, SSB reception may not be possible, or may be distorted, when AGC is used.
The latest schematic is below.