This is a recovered file. The images in this post may be out of order, and there may be duplicates.
vladn: solid state frequency-compensated regen?Post by qrp-gaijin » Sat Jun 30, 2012
Switching to this thread (since the subject is not Clapp specific). This is a continuation of this thread:viewtopic.php?t=4472 and the original thread on tube branch related to hybrid feedback regens:viewtopic.php?t=3714 I have put up two circuits illustrating a minimalistic jfet regen using the hybrid feedback. The second one adds a Vackar divider and is more suitable for SW use. Please note - I have not verified these as drawn although they are relatively close to what I used for testing (also see comments below on the gate voltage regen control):qrp-gaijin wrote:How would one calculate the appropriate capacitive divider values and ratio? I start at 1:3 ratio and adjust experimentally. The main purpose of the Vackar divider is balancing disturbances introduced by the jfet gate and drain. The secondary purpose is an additional mean of the loop gain control.qrp-gaijin wrote:In terms of device operating point and frequency shift with regeneration is there any difference to raising the gate potential as opposed to lowering the source potential? There are three subtle differences: - oscillation amplitude is better controlled with positive bias at the gate/grid (I mentioned that before somewhere in the tube section); - the RC time constant shaping AF bandwidth is fixed; - (possible) simulation shows that there is some compensatory effect of simultaneously increasing Ids and decreasing Vgd reducing frequency shift, not sure it is simulation artifact or real effect, testing is needed.qrp-gaijin wrote:I don't quite folllow how "adding a series control element will increase drain signal amplitude." Could you elaborate? jfet has a relatively high output impedance that drives a relatively low impedance feedback network, hence the signal amplitude at the drain is small. If you increase the load impedance by inserting a series element in the drain circuit the amplitude at the drain will increase.qrp-gaijin wrote:Now if only there were some way to get a handle on those higher order terms (you seem to have little trouble with them, but K3NHI did).Look, we went over this several times in other threads . Let me summarize: - I have tested the method on two very distinct circuits and got over a magnitude improvement compared to conventional feedback by removing the linear term alone. I documented this to the best of my abilities - check the graphs on pg23 and pg25 of my main report, these are actual measurements; - the remaining higher order terms are hard to predict because they depend on complex physical phenomena in the resonant circuit that determine it's Q(f); - no I do not know how to remove the remaining terms; - I do not find them objectionable in practical use, take a look at videos that I made; - there is always room for improvement in any design...
Here's a brief work log of my attempts and results so far with the hybrid feedback Armstrong-Vackar topology. I don't have much to report yet; the capacitive gate divider and the means of regeneration control seem a bit tricky. The initial schematic was as follows: I started with an Armstrong topology. Q2 is a 2SK192A. Tank inductor Tr2.1 was about 40 turns on a red T50-2 toroid. Tickler winding Tr2.2 was 6 turns. C21 is a fixed "throttle capacitor". Construction was done on a solderless breadboard. Initially I started with source resistor R2=3k with the gate directly grounded through the coil Tr2.1, eliminating C3 and C4. This was a known good Armstrong regenerative detector configuration and I could confirm oscillation around 7385 kHz. Further tests, unless noted, took place at this frequency. Next I inserted the gate capacitive divider C3 and C4 and grounded the gate directly through a 1M resistor. A gate divider of 27pF/27pF caused oscillation to stop. 1000pF/27pF allowed weak oscillation. Removing C4 with C3 at 1000pF allowed stronger oscillation. Further tests were done with C3 at 1000pF and without C4. Next I inserted the VR2, R10, C20, and R18 components to allow variable gate bias for regeneration control. With source resistor R2=3k, Q2 would continue oscillations regardless of the setting of VR2. With source resistor R2=20k, oscillation was weaker and could be stopped and started by adjusting VR2. Oscillation would stop below about 7000 kHz. The range of regeneration control was very limited; full regeneration resulted in a very weak oscillation. Next I decreased gate-voltage-limiting resistor R10 to 20k, and also decreased source resistor R2 to 10k. The regeneration control was fairly well behaved, allowing smooth transition from no oscillation to full oscillation. Q2 would oscillate down to about 5615 kHz with regeneration voltage at the gate at maximum, measured at 2.689V. To try to get the set oscillating even lower, next I reduced gate-voltage-limiting-resistor R10 to 10k. Q2 would oscillate down to 5295kHz with gate voltage at 4.032V. R10 was next reduced further to 3.3k; Q2 oscillated down to 5120kHz at gate voltage 5.450V. Source resistor R2 was lowered to its original value of 3.3k. Q2 would oscillate down to 3910kHz at gate voltage 4.697V. Gate-voltage-limiting resistor R10 was reduced to 1k. Q2 oscillated down to 3850kHz at gate voltage 5.494V. Maximum oscillation frequency in this configuration was 7615kHz. At higher frequencies, due to the reduced source resistor, gate bias could no longer stop Q2's oscillation. I also tried replacing the source resistor with a 10k pot, but the regeneration control seemed more abrupt and unreliable at lower frequencies. My current conclusions: - Gate bias for regeneration control seems well behaved, but requires tweaking the source resistance as well depending on frequency. - There is little frequency shift with gate bias adjustment at oscillation threshold; frequency shift increases as gate bias is further increased and oscillation amplitude increases. - The capacitive divider at the gate caused oscillation to stop, and only a very large tapping ratio allowed oscillation to continue, over a very narrow range of frequencies. Seeing as how gate bias adjustment for regeneration control requires also tweaking source bias for reliable regeneration over only a modest range of frequencies (3.8 MHz - 7.3 MHz), I'm not sure if this gate bias regeneration scheme is usable for a wide-range regen. Since the entire point of this project is to make a regen whose regeneration level needs no adjustment over a wide frequency range, the gate bias adjustment + periodic source bias tweaking method doesn't seem completely suitable. Not sure how to proceed next. I could control supply voltage for regeneration, or continue investigating the source-bias-only regeneration. All of the above represents only incremental modifications to an Armstrong oscillator. I haven't even tried inserting the Vackar feedback capacitor yet. There are many things that can cause the set to stop oscillating, and it seems to require a rather tedious and careful step-by-step alteration of the topology from Armstrong to Armstrong-Vackar in order to discover what set of parameters will allow oscillation to continue. EDIT: Some better progress. The following Vackar-Armstrong hybrid feedback circuit oscillates from 3320 to 7860 kHz. I eliminated the capacitive divider at the gate, inserted the Vackar feedback capacitor, and also rerouted the tickler to be shunt fed instead of series fed. Making C29 variable seems to hold promise as a means of regeneration control. A polyvaricon was able to pull the set out of and into oscillation, but a very low capacitance was needed to pull the set out of oscillation. When I tried with a varactor, the minimum capacitance was too great and the set continued oscillating. Anyway: the important thing is, I have an oscillating hybrid feedback topology. Next step will be to build it properly on a copper clad board, confirm the throttle capacitor regeneration works, then attempt to minimize the tilt. I still remain open to the possibility that the varying Q of the varactor may make things more difficult than they need to be, but I still want to try using varactor tuning. Few notes related to your *last* diagram (if possible try to keep reference designators for similar components the same across variants): - the value of R21 should be >1meg, otherwise you are loosing Q; - either C27 or C29 can be removed; - gate voltage range from 0 to Vcc/2 is a good starting point for regen control; - tweak C28 for linear term minimization, this will affect total feedback as well so you have to compensate by tweaking the source resistor; - for frequencies much above 5MHz use type6 or type7 core; My suggestion is not to focus on throttle capacitor regen control at least for now, once the simple version works satisfactory with the linear term removed we can discuss options for a more sophisticated regen control. Last edited by vladn on Sun Nov 25, 2012 6:16 pm, edited 1 time in total.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Sun Nov 25, 2012 6:16 pm Some more random ideas and questions. I realize the questions might not have clear answers at this point. Consider them research issues . 1. Regeneration control. Maybe a BJT circuit with base bias control would work. I don't have much experience with BJT regens, but base bias seems to work okay (certainly better and over a wider range than my gate bias experiments). 2. Tickler turns. Why is it necessary to reduce the tickler turns below that required for a normal Armstrong regen? Can the tickler turn ratio affect the tilt and/or the nonlinear terms? 3. Tank reactance. Is there a recommended value for Armstrong-Vackar designs? Does tank reactance affect the tilt? Last edited by qrp-gaijin on Sun Nov 25, 2012 6:35 pm, edited 1 time in total.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Sun Nov 25, 2012 6:31 pm vladn wrote:Few notes related to your *last* diagram (if possible try to keep reference designators for similar components the same across variants): Sure, sorry about that. I was in a hurry to note down the current circuit constants and behavior so I did a copy-paste, which renumbered everything. vladn wrote:- the value of R21 should be >1meg, otherwise you are loosing Q; - gate voltage range from 0 to Vcc/2 is a good starting point for regen control; OK. I note that my orignal and somewhat unsuccessful experiments with gate voltage regen control were done with a straight Armstrong topology, not the Armstrong-Vackar. But when I changed to Armstrong-Vackar, the set seemed to oscillate more reliably over a wider range of frequencies. So I will try again with the gate voltage control, and will increase R21 to improve the Q.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Sun Nov 25, 2012 9:30 pm qrp-gaijin wrote:Maybe a BJT circuit with base bias control would work.Yes. qrp-gaijin wrote:Why is it necessary to reduce the tickler turns below that required for a normal Armstrong regen?Because the two feedback paths are additive. qrp-gaijin wrote:Can the tickler turn ratio affect the tilt and/or the nonlinear terms?Yes, and so does the value of the feedback capacitor C28. Increasing tickler turns and decreasing C28 increases the total feedback. Changing only one of the two values changes both the tilt and total feedback. Changing tickler turns and C28 in the same direction changes the tilt. qrp-gaijin wrote:Is there a recommended value for Armstrong-Vackar designs?The goal is to get the highest possible unloaded tank Q. This will ensure minimal tank coupling to the amplifier and therefore minimum frequency detuning, phase noise etc. Amidon site has some Q(f) graphs for various materials for different # of turns, this gives a good starting point. Also note that type 6 and 7 materials have a noticeably better thermal stability than type 2.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Mon Nov 26, 2012 12:22 am vladn wrote:Increasing tickler turns and decreasing C28 increases the total feedback. Changing only one of the two values changes both the tilt and total feedback. Changing tickler turns and C28 in the same direction changes the tilt. Very interesting. However, adjusting the number tickler turns is one of those things that is easier said than done. It's one reason that I started my experiments with the Seiler-Vackar - I was hoping to be able to adjust all parameters just by swapping out some caps. But as noted in an earlier post, I couldn't get it oscillating properly over a wide range with the 1 uH coil I tried and my 500 pF varactor. Maybe I'll give it a try again with a larger-reactance coil and no Vackar capacitive divider (short circuiting your C3 in the Seiler/Vackar figure on p. 16). Have you successfully prototyped in hardware the Seiler-Vackar variant? In light of the above observations on feedback and tilt adjustment in the Armstrong-Vackar, and looking at the Seiler-Vackar figure on p. 16 of the hybrid feedback paper, is it correct to say that: increasing C5 and decreasing C2 will increase the total feedback; changing only one will change both tilt and total feedback; changing C5 and C2 in the same direction will change the tilt? Sounds nice in theory and easier than fiddling with tickler turns - if it can be made to work in practice! One idea I have is a grounded-base Colpitts oscillator (see http://www.ke3ij.com/bigloop.htm ), which is a reliable starter in my experience. The interesting thing is that the amplifier output (collector) is tuned, not the input. However if I'm understandig your paper correctly, the hybrid feedback network can be reversed, so I can treat the collector as the "input" and the emitter as the "output" I should be able to use the Seiler-Vackar feedback network as-is (it will actually be reversed in terms of input and output, but that's OK, right?).
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Mon Nov 26, 2012 2:22 am qrp-gaijin wrote:However, adjusting the number tickler turns is one of those things that is easier said than done. You can adjust for tilt using C28 only and then tweak the total gain with either the source resistor or the input divider or both. Before any tilt tweaks increase the value of R21 though.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Mon Nov 26, 2012 6:46 am I can't seem to get the gate bias regeneration control working well. Experiments continue. In the mean time, I was looking for BJT regen circuits and I found this rather interesting one. It almost looks like a hybrid feedback circuit. What do you think?http://www.theradioboard.com/rb/viewtop ... 5753#25753
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Mon Nov 26, 2012 3:26 pm qrp-gaijin wrote:I can't seem to get the gate bias regeneration control working well. Experiments continue.It would help if you describe the problem.qrp-gaijin wrote:In the mean time, I was looking for BJT regen circuits and I found this rather interesting one. It almost looks like a hybrid feedback circuit. What do you think?It is a pure class 2 (Vackar-like) feedback. Also the tank is heavily damped via R3.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Mon Nov 26, 2012 10:24 pm vladn wrote:qrp-gaijin wrote:I can't seem to get the gate bias regeneration control working well. Experiments continue.It would help if you describe the problem. Basically I can't bring the set out of oscillation at the high end of the tuning range (7 MHz) with "reasonable" (3.3k-20k) values of source resistance and zero volts up to Vcc/2 on the gate. If I raise the source resistance up to 40k, I can bring the set out of oscillation for a very narrow and limited range of frequencies at the high end of the tuning, but as I tune lower that range the set cannot be brought into oscillation anymore. Also, at lower frequencies, as I slowly tune the set lower and nudge the regeneration control back and forth, the regeneration control drastically shifts the tuning, maybe due to the unreasonable operating point brought about by the high source bias. For instance, with source bias at 3.3k the set will tune (always oscillating regardless of gate bias) from about 3.8 to 7.3 MHz over the full range of varactor capacitance, but with 40k source bias and fiddling with the gate bias, for about 75 percent of the range of varactor capacitance, the set only tunes down to about 6.5 MHz. Then the remaining 25 percent of the varactor capacitance (at the high end of varactor capacitance) doesn't oscillate at all. So the unreasonably high source bias plus the gate bias is restricting the tuning range somehow, is pulling the frequency greatly with gate bias adjustment as frequency decreases, and allows regeneration control only for a very limited range of frequencies at the high end of the tuning range. My Vackar feedback cap at the bottom of the tank was at 100nF to make the set behave in Armstrong-only feedback mode. (Decreasing it didn't help with the regeneration control.) I've basically tried everything except reducing the number of tickler turns, which I avoided because it requires unsoldering the coil and cutting wire. But I think now I need to reduce tickler turns to reduce the amount of tickler feedback. In your experience, does gate bias regeneration control work well over a wide (several MHz) range? NOTE: I monitored oscillation strength and frequency on a nearby receiver. With 3.3k source bias I am sure I was listening to the fundamental frequency. With 40k source bias I didn't verify this, so I may in fact have been listening to the second harmonic. That could make sense: the very high bias maybe (?) is introducing a large amount of inter-junction capacitance which would both lower the oscillation frequency and the range, lowering the maximum oscillation frequency so severely (down to 3.5 MHz) that I was listening to the second harmonic instead of the fundamental at 7 MHz. Just a theory.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Tue Nov 27, 2012 4:39 am Could you post your latest schematic with component values ? It would be *very* helpful to replace the varactor with a plain air variable capacitor for now. The varactor Q varies with frequency as well as the oscillation amplitude making things really complicated. Moreover, as the oscillation amplitude rises varactor will become a rectifier overriding the "tuning" voltage and shift the frequency drastically! This is why I strongly advised again using wide tuning varactor in the initial experiments. (At the very least try a standard dual varactor circuit, it has a better dynamic range. But even then I have doubts it will work.) Regeneration control via gate bias is nearly identical to the source resistor control (although adjustment range is smaller). Other differences are subtle.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Tue Nov 27, 2012 12:39 pm vladn wrote:Could you post your latest schematic with component values ?vladn wrote:It would be *very* helpful to replace the varactor with a plain air variable capacitor for now. OK, I will try that next. I hope I can eventually get the varactor version working though as it could make for a small and precise tuning mechanism if two ten-turn pots are used. Things left to try for this circuit, in order: 1. Replace varactor with air variable cap. 2. Reduce tickler turns. 3. Rebuild circuit properly on copper ground plane (not on solderless breadboard). If I can't get this circuit working, I may try a grounded-base BJT Armstrong or Hartley configuration, and use base bias to control regeneration. Example of a potential grounded-base Hartley regen: viewtopic.php?t=2959 , though N1TEV uses some "tricks" with that regen that might complicate transformation to a hybrid feedback topology (e.g. attempting to smooth the regeneration control by simultaneously decreasing emitter bias resistance and increasing the shunting of the tickler, or running the circuit at a very low current). Anyway, that's the backup plan - I still have some hope for the JFET circuit.
millwoodPostby millwood » Tue Nov 27, 2012 10:22 pm R22 is 47k?
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Tue Nov 27, 2012 11:29 pm millwood wrote:R22 is 47k? I tried values from 3.3k (reasonable) to 47k (unreasonable IMHO - the JFET must be practically at pinch off). Only with source bias around 40-47k was I able to use gate bias voltage to pull the set out of oscillation. Lower values of source resistance caused the set to continuously oscillate regardless of gate voltage. So, it looks like either too much tickler feedback, and/or the varactor is messing things up.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Wed Nov 28, 2012 2:20 am qrp-gaijin wrote:So, it looks like either too much tickler feedback, and/or the varactor is messing things up.Both. The varactor causes self-reinforcing frequency shift/snap once the oscillation amplitude grows. And you have too much feedback. The latter can be fixed by adding a divider at the gate rather than reducing the tickler coil (but it can make the varactor problem even worse as the tank amplitude can increase further with the divider). Personally I would go in this sequence: step 1 (conventional armstrong feedback) - replace varactor with an air variable capacitor; - add divider (15pf/15pf), if you have a quality trim cap use it in the divider; - ground plane; - trim the divider until you can follow the oscillation threshold over the entire tuning range with a source resistor around 10-20k; step 2 (hybrid feedback) - remove (short) C27; - reduce C28 to around 1nF, this will increase feedback, trim gain back down using the gate divider; - check tilt, adjust C28, trim gain, repeat step 2a (optional if you are building RX rather than VFO) - add RF buffer and AF stage (RF buffer may require some tilt trimming) - test the thing for reception; step 3 (varactor study) - experiment with varactor, definitely use a dual varactor circuit (but it may still fall short of expectations); - if it fails use capacitor+varactor (fine tuning) arrangement, that will definitely work; step 4 (advanced regen control study) - we can try a more advanced regen control, I have it partially tested but it needs some work;
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Wed Nov 28, 2012 4:46 am Okay, good , now I know exactly what to do next. I also have a better appreciation for just how delicate of a balancing act this is. It's quite a bit more work to set up than a plain regenerative receiver, but it's remarkable that it can be made to work at all, with the whole system perched right on the brink of instability (for under-threshold reception) even as the tank is being tuned and the changes in the complementary feedback paths are perfectly cancelling out.
vladn wrote:my unverified guess is that (ii) dominates the regen control effect. The higher is the drain RF load impedance (low throttle capacitance) the larger is the amplitude of the amplified RF signal at the drain. That signal is inverted with respect to the gate signal and fed back to the gate via parasitic Cgd of the JFET. This introduces negative feedback (as opposed to the positive feedback from the source to the tank). Varying the amount of the negative feedback by changing the drain load you adjust regeneration. Again this is only a guess. Consider the following N1TEV design, which uses a fairly common Armstrong tickler and throttle capacitor arrangement with a JFET:http://www.electronics-tutorials.com/re ... ceiver.htm First, I always understood that the throttle capacitor controls the amount of current flowing through the tickler. The RFC prevents current flowing up to Vcc, so instead it flows through the adjustable throttle. Would you agree that this description is accurate for a JFET Armstrong throttle capacitor and tickler? The drain signal is inverted, so the tickler is wound, adjacent to the main coil, with its hot end (to the drain) and cold end (to the throttle capaitor) reversed with respect to the main coil, to effect in-phase feedback. Then, according to your argument above, it would seem that the reducing throttle capacitance not only reduces the current flow through the tickler (thus reducing tickler-based feedback), but it also increases the drain load impedance, which then feeds the impeded inverted signal back to the gate via parasitic capacitance, further reducing feedback beyond that aleady effected by tickler current reduction. So there are two feedback control mechanisms with the tickler and throttle cap: tickler current control and drain impedance control. Using non-inverted tickler feedback from the source, but keeping the drain throttle in place, thus prevents direct control of feedback through the tickler and leaves only drain impedance control to control feedback indirectly (through parasitics). Interesting. I have seen at least one Hartley that uses a throttle capacitor: N1TEV's 2010 design in CQ magazine. The explanation is the usual "RFC backs up the RF signal preventing it from traveling to Vcc and instead forcing it through throttle capacitor". It sounds simple enough, but perhaps there is indeed more than meets the eye. (The issue of "where do the excess electrons accumulated on the top plate of the throttle capacitor go, when the gate signal drops into a valley and constricts the JFET channel" is still bugging me; given enough time, the electrons will bleed off through the RFC, but what about when they're not given enough time as the gate signal is wiggling up and down at RF? I guess some low amount of average current leaks through the RFC, bleeding off enough electrons from the top plate such that the top plate's charge does not grow without bound.)vladn wrote:I do not quite like using device parasitics for any control (as it may not be repeatable from device to device), this is personal and subjective, indeed it may work well, it just goes against my engineering/aesthetic intuition Believe me, I want to get the gate bias regeneration control working, but it's not cooperating. So I turned to a method that I have more experience with, the throttle capacitor. I am still hammering away at getting gate bias regeneration working; now, with my almost-tilt-balanced prototype, it should be easier than with a non-tilt-balanced setup (where required gate bias would vary greatly with frequency). And speaking of design repeatability, I'm a little concerned that hybrid regen designs might not be easily repeatable due to the large amount of tweaking that needs to be done. Lack of design repeatability would be regrettable, as it would discourage casual experimentation with the very elegant hybrid feedback idea.
vladn wrote:4mH is way too much, even a quality multi-section 4mH choke will have the self-resonanse frequency way below SW. At your frequencies 4mH choke will have very low and capacitive ! impedance. Really? Most regens I see use 2.5 mH chokes, and one N1TEV design I built says that the set will work all the way down to long wave with no component changes. I'll do some reading on the matter, but what is the ideal range for impedance and self-resonant frequency for chokes used in regens in the drain line?vladn wrote:qrp-gaijin wrote:L1b was about 8 turns at first, reduced to 4 turns later.Apparently that's too much. If you check my 6AS6 experiments I used 1-2 turns for L1b (~50t for L1a). In your JFET case 2 turns is a good starting value. But the interesting thing is that the throttle cap with the same number of tickler turns was able to achieve proper tilt compensation. Again, this is from memory, but I would classify my experiences as follows: 1. Inverting Armstrong-Vackar, 4 mH in drain line, gate bias regeneration control. Result: C2 reduction can properly decrease the oscillation threshold (relative to high-range frequencies at ~6 MHz) at mid-range frequencies of ~4 MHz, but cannot decrease the oscillation threshold at low frequencies at ~2.5 MHz. Oscillation requires more feedback at 2.5 MHz regardless of C2 setting. So there is significant non-linearity here: C2 can compensate tilt at 4 MHz, but not at all at 2.5 MHz. 2. Non-inverting Armstrong-Vackar and gate bias regeneration control (with Vackar capacitive divider at gate). Tried both without 4 mH choke, and with 4 mH choke bypassed by 0.1 uF. Result: same as case 1. 3. Non-inverting Armstrong-Vackar, 4 mH choke in drain, 365 pF throttle capacitor for regeneration control, gate grounded through 1M resistor (with Vackar capacitive divider). Result: C2 reduction can properly reduce and reverse the tilt such that 2.5 MHz requires less feedback (less throttle capacitance) than 6 MHz. C2 enlargement can properly reverse the tilt such that 2.5 MHz requires more feedback than 6 MHz. It seems that the gate bias regeneration control method is somehow causing more energy loss at low frequencies requiring more feedback no matter how much class-2 feedback is introduced. Yes, it could be argued that I just have too much class-1 style feedback. I will test that and reduce the turns. But: the non-inverting Armstrong-Vackar with throttle cap (and with choke) worked fine with the existing tickler feedback, whereas the exact same circuit and tickler, rewired for gate bias regeneration control, didn't work properly. So it seems that something is fundamentally different between throttle cap regen control and gate bias regen control in this circuit. Again, in this non-inverting Armstrong-Vackar configuration, the C2 vapacitor has a dual role of Vackar network capacitor *and* source bypass capacitor. I suppose that the value of the C2 bypass capacitor may have some frequency-dependent effect on the required feedback level when changing the gate bias - an effect that perhaps is not present when the gate bias is fixed (at ground, relying only on source bias) and when using the throttle cap to control regeneration. This is just a guess. In the inverting Armstrong-Vackar, the source bypass cap is separate from the Vackar network cap C2. However, very similar poor tilt adjustment behavior (using gate bias) exists.qrp-gaijin wrote:Solderless without proper ground plane is not a good testing platform at HF and higher frequencies. I've seen very strange behaviour from RF circuits that work reliably and repeatably on with a proper ground. It's entirely possible my gate bias woes are due to this. My intuition seems to say there's something more, though, since the same circuit works with the throttle cap but doesn't work with gate bias. And one more data point: base bias in a BJT almost seems workable. I could reduce but not reverse the tilt using base bias (non-inverting, tuned collector, grounded-base Armstrong-Vackar). I didn't see the low-frequency non-linearity I saw with gate bias control. The tilt was reducing, across the band, but I couldn't eliminate it (too much class 1 tilt), and further C2 reduction stopped oscillation or led to parasitic oscillations (C2 is also the emitter bypass cap in this design). This would seem to suggest that reducing tickler turns (and class 1 feedback) would make the base bias regen control topology work. Again, I'll try that (reducing tickler turns even further), and maybe that will fix the gate bias problems too. The BJT base bias regeneration control worked well immediately without any capacitive divider gain adjustment (as was needed to get gate bias regeneration control working), so that may be the least-trouble, easiest topology to get working. It also has the advantage that no throttle cap or RFC are required. Anyway, more experiments will follow.
Since we've got a good theoretical discussion going about both the throttle capacitor and about hybrid feedback topologies, I thought I'd mention for exploration another conceivable idea for a tilt-balanced regen. (I think I mentioned it briefly before in another thread.) It's probably much more trouble to realize in practice than the existing hybrid feedback approach (combining class 1 and class 2 feedback types), but who knows, maybe it might be interesting to analyze and/or try to implement this alternative idea. The idea: take a dual gang variable capacitor. Use one gang to control tuning and another gang to control regeneration as a throttle capacitor. Higher frequencies will automatically receive less total feedback due to decreased throttle capacitance. Therefore this scheme should be appropriate to counteract the tilt for class 1 oscillators (Hartley, Colpitts, Armstrong). To make the throttle capacitance level track exactly the regen frequency, it would be necessary to use use padders and trimmers in parallel with the throttle capacitor gang, as well as to control overall loop gain. Such a scheme presents a number of interesting questions. First of all, for a class 1 oscillator, what is the shape of the function Cth(f) that specifies the required throttle capacitance Cth to reach oscillation threshold for a given oscillator frequency f? This will depend (as being discussed in the other thread on throttle capacitor function) on device parasitic capacitance and the oscillator feedback configuration (number of tickler turns, etc.). Second, is it possible to achieve the required shape of the Cth(f) function when the physical Cth is ganged with the tuning capacitor Ctu? I can hazily imagine how one would could use a trimmer in parallel with the throttle capacitor gang, and a padder in series with the throttle capacitor gang, to achieve a desired capacitance swing for the throttle capacitance. Then the padder and trimmer might be optimized along these lines: Given: Ctu(f) = 1/(L*(6.28*f)^2) And for a capacitor with equally sized gangs, the throttle capacitance is equal to the tuning capacitance: Cth_unadjusted(f) = Ctu(f) Then, adding the padder and trimmer, the adjusted throttle capacitance is: Cth_adjusted(f) = 1 / ( 1/(Ctu(f) + Ctrimmer) + 1/(Cpadder) ) Then, find Ctrimmer and Cpadder to minimize the error between Cth_adjusted(f) and the ideal Cth(f) function. Since the ideal Cth(f) function's shape will depend on the oscillator feedback arrangement (tickler turns, gate divider, bias, etc.), these parameters all could also be optimized to minimize the error. It might also be possible or necessary to alter the relationship between Ctu(f) and Cth_adjusted(f) by adding a trimmer/padder to the capacitor gang used that is used to control oscillator frequency. This particular scheme seems impractical enough so that I probably won't be trying to figure it out anytime soon, but maybe someone is interested in pursuing and/or formalizing the matter. It seems like there should be enough information in this thread and the throttle cap thread to take a stab at the problem. Note that Dave S.'s page at http://makearadio.com/others/regenrx4.php makes a mention of this kind of "ganging" scheme for automatic regeneration control, but the idea isn't formalized. (Search that page for the phrase "automatic control of regeneration".) On a different note, it is interesting to observe that the above page refers not to a "throttle condenser" but instead to a "variable bypass condenser", so the concept of the "throttling" role of the capacitor had perhaps not yet taken hold.
vladn wrote:Could you re-draw the simplified circuit with reference designators ? Second jfet is the grounded gate feedback jfet. Aha, that changes everything. OK. The following diagram (D1) is the original simplified circuit (mostly based on KR1S's diagram with a sprinkling of the Ten Tec diagram). I eliminated the RF preamp stage. I also moved the grounded-gate JFET (T2) to the lower-right corner of the diagram to emphasize the fact that it is part of the source follower output/feedback loop, not part of the input. Diagram D1: The next diagram (D2) represents my misinterpretation of your post, where I thought the "second JFET" was the input JFET (T1). Here you see what I mean that there would be three feedback paths (through T2 to tank, through tickler magnetic coupling to tank, and through tickler to Vackar capacitor C9). Diagram D2: The next diagram (D3) represents what I think is the correct interpretation of your post: eliminating the original direct-from-T2-to-tank feedback, and instead routing it through an Armstrong tickler and the Vackar cap. I agree that this looks like it would work. EDIT: Now I think this will not work. The T2 output is non-inverting, but using C9 as the Vackar cap requires an inverted output. See updated diagram D3a. Diagram D3 (EDIT: maybe incorrect due to non-inverting T2 output; see corrected diagram D3a below): Diagram D3a: updated diagram D3 to use non-inverting Vackar-style feedback. I think this is correct, and D3 is incorrect. C9 is just an RF grounding cap; C10 is the (non-inverting) Vackar feedback cap. Probably, the RFC is not needed here. The next diagram (D4) represents my proposed idea to augment the existing direct-from-T2-to-tank feedback path with an additional Vackar feedback path from T2's source into C10. Note that since the T2 output is non-inverting (with respect to the tank), we need to feed the Vackar signal into C10, not C9 (C9 is just an RF grounding cap, and C10 is the Vackar cap). I am not sure if (a) the original feedback path is class 1 (if not, then adding class 2 Vackar feedback is inappropriate), and (b) if the signal magnitude of the added Vackar feedback path will be sufficient (it doesn't go through the grounded gate amp, but is taken directly off the preceding source follower). Diagram D4 (as in D3a, probably, the RFC is not needed here): Assuming the basic idea of diagram D4 is plausible, then maybe we could add another grounded-gate amp T3 for the Vackar feedback, and try to balance the feedback paths with a potentiometer VR3, as shown in diagram D4a below. You mentioned such an idea (balancing two separate amplifier paths) at the end of the hybrid feedback paper. Diagram D4a: possible dual-feedback, dual-amplifier topology. This is starting to move beyond the realm of "simple" regenerative receiver topologies... (EDIT: added DC blocking cap between VR2 and VR3, moved RFC to T3 drain) In diagrams D3 and D4 (and variants) it does seem, as I mentioned before, that the frequency response of T2 will affect the tilt. Would you agree? Also, what is the class of the unaltered feedback mechanism in the original diagram D1?vladn wrote:BTW you can simulate it and see what happens at the higher frequencies. Just break the feedback loop at the second jfet drain and drive the V-A path with a current source. Hm, OK. I have mostly figured out how to simulate the idealized VCCS/VCVS hybrid feedback models, but I haven't tried running simulations with actual JFET models yet. Maybe soon. EDIT: added possible dual-amplifier balanced topology as diagram D4a.
rp-gaijin wrote:No. For the AC sweep simulation, R2 is disabled (open circuit), I1 current source is enabled, I2 current pulse is disabled, and open-loop output is taken at point "out" on the left side of R2. For transient analysis to check if oscillation is possible, R2 is enabled, I1 is disabled, and I2 is enabled.Are you using the current source to drive the base in the AC sweep ? If so then you introduce synthetic downward tilt in the frequency response - the device Miller capacitance and the infinite impedance source forms an integrator.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Sat Jan 12, 2013 5:05 am vladn wrote:Are you using the current source to drive the base in the AC sweep ? If so then you introduce synthetic downward tilt in the frequency response - the device Miller capacitance and the infinite impedance source forms an integrator. Yes, I am using the current source to drive the base in the AC analysis (with an open loop). Is there a better way? I assume your previous comments, about needing to run the AC sweep with the amplifier oscillator below oscillation threshold, do not apply when running with an open loop, correct? Last edited by qrp-gaijin on Sat Jan 12, 2013 5:12 am, edited 1 time in total.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Sat Jan 12, 2013 5:10 am qrp-gaijin wrote:Yes, I am using the current source to drive the base in the AC analysis. Is there a better way?Use a voltage source instead. Or a current source shunted to ground by a realistic impedance of the feedback path. In all my simulations I used the voltage source assuming that I use a transconductance amp. Bipolar is a current device but driven from a low impedance node here operates "close to" the transconductance mode (voltage driven current source).
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Sat Jan 12, 2013 6:25 am vladn wrote:Use a voltage source instead. OK. The results were: with the common-base, no change: the circuit remained tilt balanced with exactly the same circuit parameters. (As I said, the common-base is a reliably-performing circuit.) For the common emitter, first I realized that the circuit as posted has a problem when simulated in open-loop mode: the base bias is cut off when the loop is opened. I fixed this by connecting the base directly to RFC2 to Vcc, and connecting Lfb to the base in a shunt-feed instead of series-feed arrangement. The result was that the tilt could not be balanced at all: too much class-1 feedback was always present. I'll continue to experiment in the simulator, but the tentative conclusion is again that the common-base seems like the most well-behaved arrangement. Edit: After some experimenting in the simulator I realized that when opening the loop it is important to make sure that all components that affect the active device biasing or bypassing should be kept in place when opening the loop or else the active device will not act realistically when the loop is opened. So I opened the loop at the collector output where it enters the H-V network input port, drove the H-V input port with a current source, and measured output at the collector. This ensures that the parts of the network that affect base bias and bypassing remain intact. This yielded tilt balance with reasonable values of, as I recall, Lfb=3e-4 * Lt and Vackar feedback capacitor C5=12n. I think the plausible simulator results are enough to warrant a physical prototype.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Re: Toroids and excessive class-1 feedbackPostby qrp-gaijin » Sat Jan 12, 2013 3:27 pm vladn wrote:qrp-gaijin wrote:sometimes a feedback coil inductance of 0.00001 uH is needed Are you sure the loss model is realistic ? I have not observed such extremes neither in simulation nor in real circuits. You can also increase the Vackar divider ratio, the threshold condition will be achieved at higher Lf and lower Cf values. It still seems that at higher HF, e.g. covering 10 MHz-30 MHz with a 1.5 uH coil (Rs=10, Rp=200k), it may not be possible to achieve tilt balance with a toroidal inductor since the minimum one-turn feedback coil inductance is 0.05 uH and the coupling is a fixed, high value, both of which seem to introduce too much class-1 feedback. Regarding the Vackar divider ratio being able to change the required Lf and Cf at threshold, do you have an example circuit where this works? I'm not seeing it in my simulations.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USARe: Toroids and excessive class-1 feedbackPostby vladn » Sat Jan 12, 2013 3:41 pm qrp-gaijin wrote:It still seems that at higher HF, e.g. covering 10 MHz-30 MHz with a 1.5 uH coil (Rs=10, Rp=200k), it may not be possible to achieve tilt balance with a toroidal inductor since the minimum one-turn feedback coil inductance is 0.05 uH and the coupling is a fixed, high value, both of which seem to introduce too much class-1 feedback. Regarding the Vackar divider ratio being able to change the required Lf and Cf at threshold, do you have an example circuit where this works? I'm not seeing it in my simulations.Make C8 (on your latest diagrams) 10x larger. Then to get the circuit back to the oscillation threshold you will have to reduce C5 and increase Lfb (simultaneously, to maintain zero tilt). In my 6as6 RX I've used 1t feedback coil (50t main coil) and then adjusted Cfb to about 24nF to get the zero tilt. I used an 8pin DIP socket (high quality - round pins) with one row grounded and the other row connected to the Lfb, then put few caps in parallel in the socket. I also have a C0G trimmer in the Vackar divider. Using both a socket and a trimmer makes tilt and threshold adjustments quite easy and solderless.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Re: Toroids and excessive class-1 feedbackPostby qrp-gaijin » Sat Jan 12, 2013 4:01 pm vladn wrote:Make C8 (on your latest diagrams) 10x larger. Then to get the circuit back to the oscillation threshold you will have to reduce C5 and increase Lfb (simultaneously, to maintain zero tilt). But C5 can only be reduced down to a certain minimum value before it starts to introduce tilt non-linearities, right? I'm still not able to achieve tilt balance in my simulated common-emitter 10-30 MHz design with Lf >= 0.05, Vackar ratio increased 10x, and decreasing C5. I'll continue to see if I can get it balanced, but there may be some circuit limits that are being run up against as we move into the upper HF region. Either that or my common-emitter open loop model is still messed up (again, I'm currently opening the loop where the collector drives the H-V network, instead driving the H-V network with a current source, and measuring collector voltage). I may also try to balance the common-base in the same frequency range. I think the common-base has no Miller capacitance so maybe it's easier to adjust and/or analyze.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Sat Jan 12, 2013 4:12 pm You need to re-estimate Rp and Rs for that frequency range. Take a Q of 200, calculate both for this Q and 20MHz, then take a half of Rs and double the Rp. It is a crude model, a much more accurate one can be derived from Q(f) plots from the Amidon site. This requires an optimization software to be written and the curves to be digitized (I have the methodology but it requires work).
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Sun Jan 13, 2013 1:20 am vladn wrote:You need to re-estimate Rp and Rs for that frequency range. Take a Q of 200, calculate both for this Q and 20MHz, then take a half of Rs and double the Rp. It is a crude model, a much more accurate one can be derived from Q(f) plots from the Amidon site. This requires an optimization software to be written and the curves to be digitized (I have the methodology but it requires work). OK. I think I am starting to see the basic problem, which you again state above but which I had forgotten: Q (and physically non-observable parameters Rs and Rp) are not constant, but instead are a non-trivial function of frequency. Nevertheless: assuming use of a toroid, even if we were to use the physically-observable Q(f) plots from the Amidon site, and used some tilt optimization method that uses Q(f), is it not still possible that, for some frequency range, it may be impossible to remove the linear term of the tilt? This would be because of the physically-dictated restrictions on the to-be-optimized parameters k, Lf, and Cf. The coupling efficient k cannot be reduced at all for a toroid, so its value cannot be optimized. Similarly, lower bounds exist on the to-be-optimized parameters Lf, which cannot be reduced to less than one turn, and Vackar feedback capacitor Cf, which practically speaking probably cannot reliably be reduced below 1 pF (and lowering Cf less than max(C_tuning) will introduce tilt non-linearities, but I suppose that doesn't matter if the non-linear optimizer is taking this effect into account). Perhaps I'm just worrying too much and misinterpreting simulator results, and in practice everything will work fine, but is it correct to say that to your knowledge, no one has yet built a hybrid feedback regen for upper HF?
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Sun Jan 13, 2013 4:33 am qrp-gaijin wrote:Perhaps I'm just worrying too much and misinterpreting simulator results, and in practice everything will work fine, but is it correct to say that to your knowledge, no one has yet built a hybrid feedback regen for upper HF?As far as I can tell the method is crude (hard to predict exact Lfb,Cfb without measuring Q(f)) but robust (stable to parameter variations, I was so far always able to converge to zero tilt by adjusting Lfb,Cfb and the divider ratio in hardware). It worked for MW and mid-HF. I do not see a reason for it not to work at the upper HF but I have not tested it. You should try to choose the core material / size / wire diameter / #turns such that the Q peak is inside the frequency range of interest (for many reasons, not just tilt compensation). I do however recommend getting a feel for the simultaneous variation of the divider ratio and the Lfb,Cfb pair such that BOTH the tilt and the network gain are nearly constant in the simulator before proceeding to the hardware.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Sun Jan 13, 2013 8:17 am I had some time to work on the hardware prototype today. The BJT common-base Hartley-Vackar showed slight class 1 tilt even with no Hartley feedback inductance (cold end of inductor directly grounded). However it was very nearly tilt-balanced with 1.6nF as the Vackar feedback capacitor. Vackar divider was 4pF/47pF. Regeneration control was abrupt (snapping into oscillation). I then reworked the circuit to be the common-emitter variant. This refuses to oscillate at all. Vackar divider is 4pF/4pF. Vackar feedback cap is 1.6nF. Hartley feedback coil is two turns. The QUCS circuit simulator says it should oscillate, but it doesn't. One lingering possibility is that the fixed caps I am using may not be NP0 (though they are marked with a black dot). This would introduce unknown losses. Next time I buy parts I'll pick up some known-good NP0 or silver mica caps and will try again. Edit: Common-emitter is now oscillating. Problem was a poorly placed emitter bypass cap (it should be mounted directly on the emitter regen control pot). This was discovered by poking around the circuit randomly, always a good thing to do when an oscillator won't start. I happened to touch the wire leading from the emitter to the regen control pot, and suddenly it oscillated. By touching it I was grounding it; otherwise, it was probably picking up stray RF from the tuning capacitor nearby. Regeneration control is smooth. Hand capacitance seems about the same as with the common-base, meaning that the hand capacitance problems with the common-base were due mainly to shielding and physical chassis instability, not to the floating variable cap. Physical placement of the two RFCs in common-emitter seems noncritical (I was worried that two RFCs in close proximity might cause unwanted feedback and instability). As for tilt, there seems to be some tilt nonlinearity; mid-band requires less feedback than low/high band (currently tuning 3.5-10 MHz). I'll continue to try to tweak the tilt in this topology.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Some tilt dataPostby qrp-gaijin » Sun Jan 13, 2013 10:06 am Here's some measured tilt data from my common-emitter Hartley-Vackar prototype. Vackar divider is 4pF/4pF. Vackar feedback cap is 1640pF. Hartley feedback coil is 2 turns. Main inductor is an additional 59 turns (T50-6 yellow core). Tuning capacitor is 150 pF max. I measured the amount of emitter resistance required for a barely-detectable onset of oscillation on my nearby station receiver with an antenna of about 10 cm wire. 3500 kHz 42.0k 4500 kHz 44.6k 5500 kHz 44.5k 6500 kHz 44.7k 7495 kHz 42.4k 8510 kHz 35.5k 9495 kHz 37.4k 10500 kHz 33.6k There is some non-linearity evident; required feedback dips mid-band and rises at the band edges. A simple analysis, attempting to cancel only the linear tilt, would call for more Hartley/less Vackar feedback. Next I'll try to increase the Vackar feedback capacitance. I hope I'm not boring you all with this project. The tilt optimization certainly presents an interesting challenge to get the most out of a simple (and in my case, low-voltage) regen circuit.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Mon Jan 14, 2013 6:28 am Antenna coupling changes the frequency dependence, sometimes drastically. You may want to add an RF buffer to get a better feel for the threshold dependence. The numbers look reasonable (you should be able to reduce the remaining tilt). Yes the higher order terms are there but I think you should also get the measurement for the Vackar-only and Hartley-only configurations in the hardware and compare.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Mon Jan 14, 2013 6:57 am vladn wrote:Antenna coupling changes the tilt. You may want to add RF buffer to get a better feel for the threshold dependence. Sorry I didn't make that clearer - the regen has no antenna connected to it yet. The 10cm antenna was connected to the monitoring receiver (FT-817ND) to pick up the onset of oscillation from the nearby regen. I have no spectrum analyzer or scope, so such crude measurement methods must suffice. Yes, I will add an RF amp. The circuit will be be the same as the RF amp in this circuit: http://www.electronics-tutorials.com/re ... ceiver.htm , with the emitter resistor adjusted for 2.5 mA of current when running off of 1.4V. Either I will use inductive coupling of the amp signal into the tank inductor, or I will just connect the amp collector to the oscillator collector below the choke (similar to your earlier proposed idea of connecting a grounded-gate amp directly to the drain RFC).vladn wrote:The numbers look reasonable (you should be able to reduce the remaining tilt). Yes the higher order terms are there but I think you should also get the measurement for the Vackar-only and Hartley-only configurations in the hardware and compare. I plan to do that. One thing that still puzzles me though is the behavior of the earlier common-base circuit I tried. Even with no Hartley feedback coil, there was some slight class-1 feedback, even with a small Vackar feedback cap (820 pF with a tuning cap of 150 pF). I can only assume that stray couplings caused by my messy wiring, and/or parasitic capacitances within the BJT, were causing the class-1 feedback even with Lfb=0. You mentioned earlier that "follower" topologies can introduce non-linear tilt when driving capacitive loads. I don't consider a common-base topology a "follower" but perhaps something similar with happening with the common-base to introduce parasitic class-1 feedback that could not be balanced.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Biasing common-emitter BJTPostby qrp-gaijin » Mon Jan 14, 2013 10:33 am Got a question about regen biasing. My current common-emitter BJT prototype is built as follows. This was based on the following RA3AAE circuit: That RA3AAE circuit was discussed here as being detuning-resistant:viewtopic.php?p=28151#28151vladn wrote:The detector operates with Vcb=0 for DC which minimizes detuning. You can see that in my variant, with the Vackar capacitive divider and hybrid feedback, I need to add two RFCs. This seems somewhat odd to me, although it does preserve Vcb=0, it does oscillate, and it does seem resistant to detuning as regeneration is adjusted around threshold. I was considering a way to remove RFC2. For instance, we could use feedback bias as follows: In the simulator, this does oscillate. However, Vcb is no longer 0 for DC, which was the original design feature. What would be the side effects if I were to go with the feedback bias variant? I suppose detuning with regeneration adjustment would increase. Anything else?
qrp-gaijin wrote:One thing that still puzzles me though is the behavior of the earlier common-base circuit I tried. Even with no Hartley feedback coil, there was some slight class-1 feedback, even with a small Vackar feedback cap (820 pF with a tuning cap of 150 pF).A common base arrangement has a very low active input impedance at the emitter side and requires more energy recirculation (hence tighter amplifier coupling) which is why I did not study it. I don't know the answer to your question off the top of my head. I may need to run the simulator and see what happens there. On your common emitter circuit. Something I do not quite like is the inclusion of the RFC into the tank circuit with significant coupling (1:1 divider ratio or 1:4 impedance conversion). Standard RFCs usually have neither good temperature stability, nor high Q. I would either increase the divider ratio or do the opposite - connect the collector to the main coil without the divider and pull the coil middle tap to Vcc via a single RFC (the middle tap has the lowest impedance for RF). The latter will increase the effects of the collector loading but the RFC effects will be virtually eliminated.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Mon Jan 14, 2013 10:19 pm vladn wrote:I would either increase the divider ratio or do the opposite - connect the collector to the main coil without the divider A rather annoying tradeoff: remove RFC effects but suffer increased collector loading, or decrease collector loading but suffer RFC effects. It seems that connecting the collector to an additional lower tap on the coil would be the high-Q way to solve the problem: collector loading reduced due to the low tap (replacing the capacitive Vackar divider), and no collector RFC effects (collector, and base, go to Vcc through the RFC at the Hartley tap).
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Tue Jan 15, 2013 12:47 am qrp-gaijin wrote:A rather annoying tradeoff: remove RFC effects but suffer increased collector loading, or decrease collector loading but suffer RFC effects.Intuitively I think RFC is worse, but I am not sure.qrp-gaijin wrote:It seems that connecting the collector to an additional lower tap on the coil would be the high-Q way to solve the problem: collector loading reduced due to the low tap (replacing the capacitive Vackar divider), and no collector RFC effects (collector, and base, go to Vcc through the RFC at the Hartley tap).Yes, it may be a good solution, but I need to check the simulation - the inductive divider is not grounded. (edit)No go unfortunately, the divider must be between the tuning cap and the ground. It kind of works for mid-tap 1:1 ratio (with higher non-linear terms) but not for high division ratios.(/edit)
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Tue Jan 15, 2013 4:06 am vladn wrote:qrp-gaijin wrote:connecting the collector to an additional lower tap on the coilNo go unfortunately How about a low-turn-count separate coupling coil from the collector to Vcc (replacing the collector RFC), magnetically coupled to the tank? Turns count would be reduced to just sustain oscillation. One complicating factor would be that the new collector coil would be magnetically coupled to both the tank coil and the base's Hartley feedback coil. No idea what that would do to the tilt. Edit: I tried 2 simulations, one with no divider (collector goes directly to top of tank; Hartley tap goes through RFC to Vcc) and one with a new low-inductance link coupling for the collector. Both had the same problem: too much class-1 feedback unless the feedback and/or collector link inductance were reduced to the order of 1e-5 * Ltank (Vackar cap=1.6nF). Tuning range was 3.5-10 MHz, Ltank=15 uH. I'm going to focus on why the no-divider (collector directly to tank) method seems to have an excess of class-1 feedback (more so than with the 4pF/4pF Vackar divider), and if there's any way to increase Lfb to more reasonable values. Maybe extra Rs and/or Rp tank loading is one way (but I cringe at the thought of intentional tank damping). Another alternative might be to forget inductive class-1 feedback and instead try a Colpitts-Vackar.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Wed Jan 16, 2013 4:51 pm qrp-gaijin wrote:Edit: I tried 2 simulations, one with no divider (collector goes directly to top of tank; Hartley tap goes through RFC to Vcc) and one with a new low-inductance link coupling for the collector. Both had the same problem: too much class-1 feedback unless the feedback and/or collector link inductance were reduced to the order of 1e-5 * Ltank (Vackar cap=1.6nF). Tuning range was 3.5-10 MHz, Ltank=15 uH. I'm going to focus on why the no-divider (collector directly to tank) method seems to have an excess of class-1 feedback (more so than with the 4pF/4pF Vackar divider), and if there's any way to increase Lfb to more reasonable values. Maybe extra Rs and/or Rp tank loading is one way (but I cringe at the thought of intentional tank damping). Another alternative might be to forget inductive class-1 feedback and instead try a Colpitts-Vackar.Is that the exact same circuit that you posted last time ? (BTW is there a way to refer to a specific post number on this board ?) I'd like to see/check the results but I need to know precisely what you are doing and observing. If you can post (or PM) a link to your latest simulation this would help avoid bloating the thread with minute details.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Thu Jan 17, 2013 12:12 am vladn wrote:qrp-gaijin wrote:Edit: I tried 2 simulations, one with no divider (collector goes directly to top of tank; Hartley tap goes through RFC to Vcc) and one with a new low-inductance link coupling for the collector. Both had the same problem: too much class-1 feedback unless the feedback and/or collector link inductance were reduced to the order of 1e-5 * Ltank (Vackar cap=1.6nF).I'd like to see/check the results but I need to know precisely what you are doing and observing. Let's start simply. To begin with, here is the circuit where I removed the input (Robert/Vackar) divider. Normally, the collector would go directly to the top of the tank inductor. For open loop analysis I break the loop at the H-V network input, drive the network with a current source, connect the collector to Vcc through a 1-ohm load resistor, and measure voltage at the collector. Parameters: Lt=15uH, Rs=10, Rp=200k, Ct=1*/(((6.28*f)^2)*(Lt)), f swept from 3 to 10 MHz. Q1 is a 2N3904. For Lfb=1e-4 * Lt: For Lfb=1e-5 * Lt: For Lfb=1e-6 * Lt: The only way to balance the tilt is with a tiny feedback inductance of Lfb = 1e-5 * Lt. If Lfb is larger (e.g. 1e-4*Lt), reducing Vackar cap C5 cannot compensate. I'm getting some other unexpected behavior with variants of this circuit (e..g. class 1 feedback even with the feedback coil removed), but let's address that later. First: is such a small Lfb expected? If not, what is causing the excess class-1 feedback? Parasitics? After we get this sorted out, next we might examine the circuit variant with the extra link winding on the collector for impedance transformation.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Thu Jan 17, 2013 2:20 am This is your problem: vladn wrote:You need to re-estimate Rp and Rs for that frequency range. Take a Q of 200, calculate both for this Q and 20MHz, then take a half of Rs and double the Rp. It is a crude model, a much more accurate one can be derived from Q(f) plots from the Amidon site. This requires an optimization software to be written and the curves to be digitized (I have the methodology but it requires work). Let's say the middle of your frequency range is approx 6MHz, then assuming Q=200 Rs(equiv)=2*pi*f*L/Q=2.8ohm Rp(equiv)=2*pi*f*L*Q=113k Splitting the losses in half we take the half for Rs (1.4ohm) and double the Rp (220k). You have used much higher value of Rs hence the difference in the simulation results. With Rs=1.4ohm I got flat response with Lf=0.5e-3*Lt which is doable. Again I want to emphasize that the model is crude because we do not know the exact Q(f) and use arbitrary Rp+Rs mix instead. A real circuit (albeit tube) covering this frequency range and using a toroidal type 7 iron core required much higher ratio of class 1 to class 2 feedback compared to the simulation, so you should be OK.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Thu Jan 17, 2013 4:27 am vladn wrote:This is your problem: vladn wrote:You need to re-estimate Rp and RsWith Rs=1.4ohm I got flat response with Lf=0.5e-3*Lt which is doable. Right. I had carelessly assumed the values I was using for Rs and Rp were OK for low-HF, but I see now the values are quite critical (though physically unobservable). This is what I get with Lf=5e-4*Lt, Rs=1.4, Rp=220k. Is it close to your result? There seems to be more non-linearity than with my previous results. I suppose the previous greater Rs and greater damping smoothed out the non-linearities. Hopefully later today I will get a chance to re-simulate the collector link coupling with the more realistic loss values. As I recall, other than the low Lf needed, the collector link method seemed to return reasonable tilt behavior, so it might be a way to implement an inductive divider without using the lossy RFC at the hot end of the tank.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Thu Jan 17, 2013 5:00 am I think you have an aliasing problem. Can you add more points to the sweep command (not tuning steps but finer resolution in the sweep itself)? And use the log scale on the vertical axis if possible.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Thu Jan 17, 2013 10:38 am vladn wrote:I think you have an aliasing problem. Good call. That fixes it, and is something I should have been and from now on will be more careful about when dealing with simulation results. Next: reducing the collector tank loading. We said before that a capacitive divider isn't ideal (as it requires a lossy RFC at the tank's high-Z point), and an additional inductive tap divider on the main coil won't work (since the resulting inductive divider is ungrounded). So, the next question is, will an additional RF-grounded link coil at the collector work? As far as I can tell, the answer is yes. The circuit (loop has been broken between collector and link coil): Lt=15uH, Rs=1.4, Rp=220k, Lfb=1e-3*Lt, Lin=1e-3*Lt (Lin=new input link into H-V network). The tilt: Any problems with this approach? And while we're on this circuit... is it possible to add a "bandwidth-compensated" bandspread control here as discussed in the other thread? Edit: looking at the circuit again... has this now become a Reinartz oscillator?
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Fri Jan 18, 2013 1:33 am qrp-gaijin wrote:Any problems with this approach?I think it is OK conceptually. Requires even more complex inductor though... qrp-gaijin wrote:And while we're on this circuit... is it possible to add a "bandwidth-compensated" bandspread control here as discussed in the other thread?Yes, just add a large variable cap in parallel to the Q1 base for fine tuning. In general it should "sit" on the feedback node of the H-V network, opposite to the tuning cap.qrp-gaijin wrote:Edit: looking at the circuit again... has this now become a Reinartz oscillator? I am getting lost myself with naming conventions. Is Reinartz = Meissner ?
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Fri Jan 18, 2013 2:43 am vladn wrote:qrp-gaijin wrote:Any problems with this approach?I think it is OK conceptually. Requires even more complex inductor though... Well, it seems to be the only remaining way to reduce collector loading. If it improves oscillator stability and noise, I think an extra link coil isn't too much added complexity. And the link can be relatively easily adjusted just like a Roberts/Vackar divider - remove turns until oscillation is just barely maintained.vladn wrote:qrp-gaijin wrote:"bandwidth-compensated" bandspreadYes, just add a large variable cap in parallel to the Q1 base for fine tuning. This would be a variable cap from the base to ground?vladn wrote:Is Reinartz = Meissner ? That's something I've been wondering about for a while as well. In the context of a BJT, I'm referring to the non-inverting common-base oscillator with inductive coupling links from both emitter and collector into the tank. I've seen this arrangement referred to as both a Reinartz and a Meissner. In the latest proposed circuit, we have a tank that is coupled to the base and collector (not emitter and collector as in the normal Reinartz) and is thus an inverting configuration. In addition we've raised the cold end of the tank (and the feedback coil) above ground with the Vackar cap and have fed additional inverted feedback energy into that junction. So it seems it could be argued that this is an inverting Reinartz/Meissner configuration to which Vackar feedback has also been added.
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Sat Jan 19, 2013 5:28 pm I can confirm that the latest circuit I posted (with the collector link coupling) works in hardware. A single link turn could sustain oscillation but required very high regeneration levels (i.e. very low emitter resistance). I increased it to three turns (being careful to keep the winding sense the same as the main tank coil for proper phasing) and with a Vackar cap of 1640 pF it is very close to tilt-balannced. It seemed that the number of collector link turns affected the tilt. The almost-balanced configuration with three link turns was not balanced with one link turn. It seemed that there was excessive class-1 feedback with only one link turn. However, I'm not seeing that effect in the simulator. Maybe the extremely low emitter resistance required when using only one link turn is disturbing the tilt. Also, when using only one link turn, frequency varied quite a bit with regeneration adjustment near threshold. With three link turns, frequency shift is reduced. The collector link winding does seem, as expected, to help stability compared with connecting the collector directly to the tank top. Monitoring the regen's oscillation on a nearby receiver, there was noticeable warble if the link winding was not used. With the link winding, the impedance transformation should be about 1:413. Next I will work on the RF amp and some way of extracting audio from the regenerative stage.
vladnPosts:853Joined: Sun Nov 02, 2008 7:15 amLocation: NJ, USAPostby vladn » Sun Jan 20, 2013 9:06 pm It looks like it is possible to use the "feedback" tap of an H-V network as a bidirectional node in a "two-point" type oscillators like negative resistance oscillators or two transistor non-inverting feedback style oscillators. The optimum values for C_fb and L_fb are different in this arrangement. One interesting property of such an arrangement is that the next (non-linear) term has the opposite sign - the frequency response curve has a sight dip in the middle of the frequency range (as opposed to a slight increase). There could be a way (at least in theory) to use this somehow to compensate the next term (but it is not an easy problem to solve).
qrp-gaijinPosts:2822Joined: Sun Feb 28, 2010 2:12 pmContact:Postby qrp-gaijin » Mon Jan 21, 2013 4:10 am vladn wrote:One interesting property of such an arrangement is that the next (non-linear) term has the opposite sign - the frequency response curve has a sight dip in the middle of the frequency range (as opposed to a slight increase). There could be a way (at least in theory) to use this somehow to compensate the next term (but it is not an easy problem to solve). When you say "use this ... to compensate the next term" do you mean "having observed both dip and hump shapes for the second order term, design a way to control the dip/hump parameter and achieve zero for both first and second derivatives"? In one of the many BJT variants I simulated, I believe I was able to control the dip/hump (second order term sign) by altering the emitter bypass capacitance. (Not sure about if this was the parameter I altered, but I do remember that altering some parameter allowed me to change the dip into a hump.) I'll have to check and re-simulate that to make sure it wasn't an aliasing artifact. In general, I was getting very odd behavior with low values (0 pF-20 pF) of emitter bypass capacitance and no Hartley feedback inductance. Maybe parasitic Colpitts-style feedback was taking place (I was specifically trying to investigate combining inverting Vackar feedback with non-inverting Colpitts feedback).
After many tries over the 3 years since this thread started, I have finally achieved an important goal: I have successfully built a hybrid-feedback, tilt-balanced, low-voltage, varactor-tuned, BJT regenerative receiver with a practically fixed regeneration level.
Back to my varactor-tuned hybrid-feedback circuit - here it is:
circuit-2.png (76.6 KiB) Viewed 1331 times
Q1 is an RF amp to allow coupling in the RF signal while presenting the detector with a constant load; Q2 is the regenerative detector.
The key to success was the use of a high-Q 1SV74 varactor. In a previous attempt at a varactor-tuned, hybrid-feedback regenerative receiver I had used a low-Q 1SV149 varactor, designed for AM BCB use and having a Q of 200 at 1 MHz, which approximately equals a Q of 20 at 10 MHz. On the other hand, the new 1SV74 varactor I am using has a Q of 50 at 50 MHz, which approximately equals a Q of 250 at 10 MHz.
The disadvantage of the higher-Q varactor is a more limited capacitance swing, and hence a more limited tuning range. On the other hand, one advantage is that a limited tuning range is easier to tilt-balance than a wide tuning range.
The important thing is that the varactor Q must be higher than the coil Q, or at the very least, the varactor's equivalent series resistance should be almost constant over the entire tuning range. If these conditions are met, then the predominant losses in the tank are the Rs (serial losses) and Rp (parallel losses) of the coil, and as described in vladn's paper, the hybrid-feedback approach can passively equalise the threshold regeneration level for any arbitrary mix of Rs and Rp. The worst thing is a low-Q varactor with ESR that varies greatly as the varactor is tuned - and unfortunately the 1SV149 varactor is just such a varactor when used at HF.
The circuit currently tunes from 7 MHz to 10 MHz.
Here are some notes on the construction:
1. Determining the threshold regeneration level: It's easiest to connect the regen to an AF amplifier and to listen for the threshold noise as you adjust the regeneration control. Once you have found the threshold level, you can then tune the regen higher or lower in frequency and observe if you need more or less regeneration to reach threshold. One problem with this approach is if you have too much or too little feedback in the circuit (i.e. the regen is never able to oscillate, or the regen is always oscillating), then even adjusting the regeneration control you will never cross the oscillation threshold and thus you will hear nothing from the AF amplifier.
2. Determining the proper amount of feedback: I am repeating a principle stated by vladn earlier in this thread, but the principle bears repeating: Increasing L2 increases the total feedback and alters the tilt towards class-1 behavior. Decreasing C12 increases the total feedback and alters the tilt towards class-2 behavior. To increase the total amount of feedback without altering the tilt, you must increase L2 and decrease C12 simultaneously. Conversely, to decrease the total amount of feedback without altering the tilt, you must decrease L2 and increase C12 simultaneously.
3. Construction order
3a. The first thing I did was to build the circuit with L2 being a short circuit and C12 (the Vackar feedback capacitor) set to 1000 pF. Then, I could verify that the oscillation was controllable below, through, and above threshold. Furthermore, regeneration behaved as a class-2 oscillator, i.e. required regeneration to reach threshold increased monotonically with increasing frequency.
3b. Then, I set C12 to 100 nF (essentially an AC short circuit with 0 ohms reactance) and increased L2 to become 5 turns on the cold end of L1. Again, I could verify that oscillation was controllable below, through, and above threshold. Furthermore, regeneration behaved as a class-1 oscillator, i.e. required regeneration to reach threshold decreased monotonically with increasing frequency.
3c. I reduced C12 back to 1000 pF to re-introduce Vackar feedback into the circuit. This gave too much feedback. I could hear no output from the AF amplifier (no threshold noise) as I adjusted the regeneration control over its entire 100k range, and based on the principle in #2 I knew that the total circuit feedback was too much and that the circuit was always oscillating. I also verified on a spotter receiver that the regen was always oscillating no matter how I adjusted the regeneration control.
3d. I reduced L2 to 3 turns and increased C12 to 2000 pF. I could verify slight class-2 behavior (required regeneration at the low end of the tuning range was slightly lower than at the high end of the tuning range), indicating a slight excess of Vackar-style feedback.
3e. I incrementally added capacitors in parallel to C12, finally arriving at 2320 pF for almost perfectly level tilt.
vladn wrote: General guidelines for feedback branch balancing: C2 should be several times bigger than maximum value of Ct. L2 should have slightly fewer turns than in conventional Armstrong regen (perhaps 80% or so). Once you get the amount of total feedback right (operating in the middle of regen pot adjustment) - tweak C2 to minimize tilt. Decrease it if the regen level pot needs to be advanced more at the bottom frequency vs top frequency of the tuning range. Increase C2 if the opposite is true.
4. There seems to be a tiny tendency for oscillation in the middle of the tuning range indicating a non-linear excess of feedback. There also seems to be a tiny bit of excess Vackar feedback. I might try adding another 100 pF in parallel with C12 to see if it helps.
5. Connecting an active antenna to the RF amp Q1, I could verify that I could set the detector below threshold and tune over the entire tuning range without the set breaking into oscillation, and I could hear several SWBC stations in this state. However, I could not hear the band noise. This indicates the detector is somewhat insensitive, which might be fixed by a separate detector stage.
6. Again using the active antenna and setting the set slightly into oscillation, I could verify that I could tune over the entire tuning range with good sensitivity and with easy phase-locking onto AM SWBC carriers for synchronous reception, thanks to the weak oscillation level maintained over the entire range by the tilt-balanced, equalised regeneration level.
6a. Adding a separate, explicit, aactive limiter stage (as in Kovalenko's automatic regeneration scheme) would help maintain the oscillation level at an even lower amplitude for better synchrnous reception. LTspice simulations indicate Kovalenko's scheme should work even when running off of 1.2 volts, as my receiver is. The combination of explicit gain limiting and passive tilt-leveling (with hybrid feedback) would then start to echo vladn's 6as6 regen design.
7. One thing I was worried about was that the heavy coupling of the collector into the tank might upset the equalised regeneration level, as it allows the BJT parasitic capacitances to influence the tank. However, this seems not to be significant enough to worry about. A smaller link coupling, magnetically coupling the collector to the main tank coil, could be used to reduce tank loading by the collector.
8. Currently the set tunes 7-10 MHz. At higher frequencies (approaching 30 MHz) parasitic BJT capacitances of #7 may upset the equalised regeneration level. Also, the 1SV74 varactor Q will be lower at higher frequencies (e.g. the 1SV74 varactor Q will be only approximately 83 at 30 MHz). In a future experiment, I may try some different VHF varactors (FV1043) with even higher Q that should have Q>300 over the entire HF spectrum (specified Q=100 at 100 MHz, i.e. Q=333 at 30 MHz, Q=3333 at 3 MHz).
9. All potentiometers are high-quality 10-turn potentiometers.
10. C12 must consist of high-quality, RF-grade capacitor(s) such as NP0 or C0G types.
11. I'm not convinced the RF amp coupling to the detector is optimal. Setting the RF attenuator to maximum oddly seems to desensitize the receiver instead of increasing sensitivity (when using the active loop antenna). This may have to do with the fact that Q1 (RF amp) is directly coupled to the RFC of Q2 (the regenerative detector). It's probably better to have the Q1 load be another secondary winding on L1.
12. I'm allowing the varactor reverse bias to fall to 0 volts, which is generally not recommended. Varactor Q is lowest in this condition, so it's possible that limiting the minimum bias to 1 volt may reduce the variability in varactor ESR and hence improve the equalisation of the regeneration level.
13. I would guess the circuit should be reproducible, with the same circuit constants, using a variable capacitor instead of a varactor.
Here's what the chaotic breadboard looks like.
board.jpg (179.46 KiB) Viewed 1340 times
The coil is would as shown below. The top 3 turns on the coil are L2; the rest of the turns are L1.
coil.jpg (151.54 KiB) Viewed 1340 times
If there is any interest I could post a video, but the current construction is not physically stable so the reception frequency is subject to warbling and microphonic effects.
Big thanks again to vladn for your advice throughout this thread. From the start, I was determined to use varactor tuning and to limit myself to a 1.2-volt power supply. Finally, I have achieved this goal to my satisfaction.
I plan to rebuild this receiver into a more stable and usable form. Probably for the next iteration I will focus on bandswitched coils, which will require tilt balancing for each separate coil. For future iterations I might try to mimic vladn's superlative 6as6 design and add a dynamic limiter, scale-invariant gain compression, bandwidth control, RF AGC, etc. - all running of course off of 1.2 volts.
I came up with a new way of applying hybrid feedback (Colpitts-Vackar) to my simple BJT regen, and applied the hybrid feedback to a circuit with a ferrite rod and a wide-tuning 1SV149 varactor. First, I present an unsuccessful circuit: fer-armstrong-vackar.png (83.01 KiB) Viewed 1044 times This is the same circuit as the tilt-balanced Armstrong-Vackar I mentioned in my previous post. However, in the new circuit of this post I am using a shortwave ferrite rod antenna for the tank inductor instead of a T50-6 iron core. The idea was to eliminate the need for an external antenna. You can see the ferrite rod in a previous receiver here: viewtopic.php?p=50381#p50381. In this experiment I also returned to using AM BCB 1SV149 varactor, which is noticeably lossy at shortwave frequencies. The frequency-dependent losses in the varactor combined with the frequency-dependent losses in the ferrite rod antenna make for a receiver that is likely difficult to reproduce. Nevertheless, I wanted to try the combination of the 1SV149 varactor (which has a wide tuning range) plus the ferrite rod antenna, and see to what extent it was possible to use hybrid feedback to level out the regeneration level. In the above circuit I used the same approach as before of using Vackar-style feedback (via feedback capacitor C12, holding the base above RF ground) plus Armstrong-style feedback (via the L2 tickler). The good news was that with L2 being a short-circuit (no Armstrong feedback), I could verify Vackar-style tilt (required regeneration increases with frequency) over most of the tuning range. That indicates that the frequency-dependent varactor losses and the ferrite rod losses are somehow balancing each other, such that the inherent Vackar-style tilt can emerge without being distorted by the varactor and ferrite rod losses. The bad news was that adding the L2 tickler into the circuit always overwhelmed the circuit with too much Armstrong-style feedback, resulting in Armstrong-style tilt (required regeneration decreases with frequency). Even with L2 reduced to a single turn very loosely coupled to the cold end of the ferrite rod, that single loosely-coupled turn was enough to cause the tilt to shift from Vackar-style to Armstrong-style. One approach to fix this would be to increase the Vackar-style feedback by decreasing C12. Unfortunately that isn't really possible because C12 should not be reduced too small, as it should always be much lower impedance than the tuning capacitance (provided by D1, which is a maximum of about 500 pF). The other approach to fix this is to reduce the Armstrong-style feedback. Unfortunately it's not possible to reduce the feedback below a single loosely-coupled turn. So instead, I opted for Colpitts-style feedback (which has the same tilt behavior as Armstrong-style feedback) as shown below. fer-colpitts-vackar.png (81.58 KiB) Viewed 1044 times Here, I removed the tickler L2 and added a new Colpitts-style feedback path through C1, with a tiny value of 4 pF, just to add a tiny bit of Colpitts feedback from the collector back into the emitter. Note that if C1 is not present (i.e. C1 reactance is infinity), the circuit is just a Vackar-style regen. On the other hand if C1 is present (i.e. reactance less than infinite) and the Vackar capacitor C12 has near-zero reactance (e.g. a high value like 100 nF), then the circuit becomes a common-base Colpitts regen, with the Colpitts divider formed by C1/C11, the tuning capacitance D1 being in parallel with both capacitors of the divider, and the cold end of L1 being grounded through the near-zero-reactance C12. With both feedback paths C1 and C12 in the circuit, the circuit becomes a hybrid feedback regen using both an inverting feedback path (collector-base) for the Vackar feedback, and a non-inverting feedback path (collector-emitter) for the Colpitts feedback. After some theoretical investigation I had recently discovered this topology, but this is the first time I tried it in hardware. The results are good and this method does work to balance the tilt. The required regeneration adjustment tuning over the entire tuning range (~4-12 MHz) is very small. There is a slight excess of Colpitts feedback, so C1 should be reduced (reducing Colpitts feedback) or C12 should be reduced (increasing Vackar feedback). There is also some noticeable nonlinear behavior in the tilt (especially when the varactor reverse bias approaches zero volts at the extreme low end of the band) indicating that a perfectly flat regeneration level across the entire tuning range is likely not possible when using the lossy varactor and the lossy ferrite rod. Even if not perfectly flat, the mostly-flattened regeneration level does make for a much smoother operating experience with reduced need for regeneration adjustment, plus the advantage that the ferrite rod removes the need for an external antenna. This combination of ferrite rod antenna plus reduced regeneration adustment will likely make a handy portable receiver. I aim to rebuild my prototype into a usable receiver soon.