Perfect balance at low frequency. I believe what 6A3 was referring to is the difference in parasitics between the cathode and anode, so the impedances aren’t necessarily equal across frequency even if the resistors (and the loads on the other side of the coupling caps!) were hand matched to .01%. It’s a subtle point, that with 30 dB of feedback, goes away. With no (or limited) GNFB, one is at the mercy of what tubes “do” by themselves. Some people like it that way.
Both output impedances are grossly unequal. Output signal swing is heavily limited. No CMRR/hum suppression at all. Cant't say that I like the Concertina. It's just a cheap solution.
Best regards!
Best regards!
The open loop gain rolls off with the frequency increase. You want 1st-order rolls off. Adding 2nd pole is the formula for an oscillator.
20dB NFB means, at low frequency or DC, it has 20dB NFB. Not at high frequency.
30dB NFB at DC is not an issue, either. As long as you keep the open loop gain below 1 (unity) at 100KHz (with a good output transformer). In general, it would be stable.
PS: A good circuitry should be stable even without any load connected to the output transformer. The phase margin or gain margin should not differ much with the load. Sadly, I see the majority tube amps is only marginal stable without the load.
20dB NFB means, at low frequency or DC, it has 20dB NFB. Not at high frequency.
30dB NFB at DC is not an issue, either. As long as you keep the open loop gain below 1 (unity) at 100KHz (with a good output transformer). In general, it would be stable.
PS: A good circuitry should be stable even without any load connected to the output transformer. The phase margin or gain margin should not differ much with the load. Sadly, I see the majority tube amps is only marginal stable without the load.
Last edited:
The driving-point impedances are grossly unequal, but if the loads are the same, the effective impedance in the cathode is the same as in the anode. Therefore the gains are the same. You must drive equal loads to get this. That includes the reactive part, which you may not be able to control. The limited output swing is a big deal, but where do you typically see it used? Driving EL84’s or 7591’s in pentode, where 12 volts peak gives full output. Or in a Williamson, where there’s a gain stage in between.
I am not sure that the unequal impedances of the concertina outputs is a real issue.
The current in Rp is always equal to Rk.
The only exceptions are:
Any grid current (So . . . use real good tubes, And keep the grid signal swing away from the grid-to-cathode contact potential). Problem solved.
Xc of cathode to filament, can be a problem at high frequencies. Pick a low capacitance cathode to filament tube, and do not use very high resistance Rk & Rp. (The ratio of Rk to Xc at high frequencies needs to be taken care of). Problem solved.
DC leakage from cathode to filament, from tubes that did not have elevated filaments, that would have prevented degradation of the filament to cathode insulation material. Elevate the filament. Problem solved.
Any Un-equal load impedance of the next stage.
Did I cover all the things that can prevent good performance of the concertina?
Proper design makes concertinas work very well.
But that still does not make me like concertinas.
$0.03
adjusted for inflation
The current in Rp is always equal to Rk.
The only exceptions are:
Any grid current (So . . . use real good tubes, And keep the grid signal swing away from the grid-to-cathode contact potential). Problem solved.
Xc of cathode to filament, can be a problem at high frequencies. Pick a low capacitance cathode to filament tube, and do not use very high resistance Rk & Rp. (The ratio of Rk to Xc at high frequencies needs to be taken care of). Problem solved.
DC leakage from cathode to filament, from tubes that did not have elevated filaments, that would have prevented degradation of the filament to cathode insulation material. Elevate the filament. Problem solved.
Any Un-equal load impedance of the next stage.
Did I cover all the things that can prevent good performance of the concertina?
Proper design makes concertinas work very well.
But that still does not make me like concertinas.
$0.03
adjusted for inflation
Impedances not an issue. To get the medium mu cathodyne to run correct, aim for the integer divisor of approx 0.33 that is if pentode anode is approx 90V, near starvation diet, then triode grid section cathode volts becomes 85V then anode should be around 260V and so on. The trick is to use a stabilized stepped B+ for both (low noise) ; run the pentode section from a lower B+ voltage. If triode section runs at 360V then I use approx 220V B+ for the pentode section with 150K anode resistor. I aim triode section with equal 18K+18K resistors with 5mA through current.
The joker is going to select the pentode screen resistor and that interferes with the anode volts too! Here we come across the 1/4 or 1/5 current rule, so quite commonly implying a screen resistor of approx 910K is used, and that coincides with lowest distortion. That resistor value for many versions of NOS Mullard ECF80 and Ei; Philips Valvo E80CF; Westinghouse 7643 fits well. That inplies the actual screen voltage may be around 25-35V measured with a 10M DVM. Don´t be tempted to fit a low voltage cap as switch on B+ could be up in the hundreds.
There is another trick up ones sleeve, the g2 to cathode capacitor which I use as a subsonic rolloff. A 220nF will provide a roll off -3dB at 10 Hz; with output transformers going down close to that frequency, then stability is evermore important.
Using the 0.33 integer I can get 25+25V RMS onset symmetrical clipping at 0.5% distortion. This is an excellent building block for a balanced CCS diff driver and even direct drive into output stage so long one stays in class A. Apply 20dB gnfb and astonishingly low distortion figures can easily be obtained.
Yes, some have remarked the cathodyne triodes section PSRR is poor but decoupling caps are cheap, and that applies for the heaters as well. Use ´em.
I use the cathodyne for my rather noisy 250W +250W parallel output EH KT90 amp; distortion at 240W @ 1Khz is a mere 0.15% and 5Khz 0.3%.
Big and powerful. That´s how I like´em
Bench Baron.
The joker is going to select the pentode screen resistor and that interferes with the anode volts too! Here we come across the 1/4 or 1/5 current rule, so quite commonly implying a screen resistor of approx 910K is used, and that coincides with lowest distortion. That resistor value for many versions of NOS Mullard ECF80 and Ei; Philips Valvo E80CF; Westinghouse 7643 fits well. That inplies the actual screen voltage may be around 25-35V measured with a 10M DVM. Don´t be tempted to fit a low voltage cap as switch on B+ could be up in the hundreds.
There is another trick up ones sleeve, the g2 to cathode capacitor which I use as a subsonic rolloff. A 220nF will provide a roll off -3dB at 10 Hz; with output transformers going down close to that frequency, then stability is evermore important.
Using the 0.33 integer I can get 25+25V RMS onset symmetrical clipping at 0.5% distortion. This is an excellent building block for a balanced CCS diff driver and even direct drive into output stage so long one stays in class A. Apply 20dB gnfb and astonishingly low distortion figures can easily be obtained.
Yes, some have remarked the cathodyne triodes section PSRR is poor but decoupling caps are cheap, and that applies for the heaters as well. Use ´em.
I use the cathodyne for my rather noisy 250W +250W parallel output EH KT90 amp; distortion at 240W @ 1Khz is a mere 0.15% and 5Khz 0.3%.
Big and powerful. That´s how I like´em
Bench Baron.
With a gain stage in between the concertina and output bank it sounds like a Williamson-ish design that only needs single digit volts out of the concertina to begin with. Mine only needs 3V out of it - and the front end is on its own mosfet 300V power supply. Very quiet, unless I get an input tube with h-k leakage, in which case I simply plug in another. On the the other hand I’ve got a little 14 wpc zero global feedback amp that uses a concertina to develop over 75V p-p into a pair of low mu triodes. Off a 260V supply. I didn’t know if it would have it in it, but apparently it does. It’s got enough oomph to budge the 37.5 V cathode resistor voltage and push up into class AB. Pleasantly surprised, but I’ll take the extra dB and a half.
I use a Zobel network on the output of my amps, at least my recent headphone amp designs. This reduces the tendency of the amp to cross 0dB gain, then swoop back up and re-cross 0 dB at a higher frequency. This is the very definition of conditional stability.
I should drag my "Kingfisher" 7591A push-pull amp back in and give it a run on the gain-phase analyzer at work to see if an output Zobel network would help with overall stability. That amp was plagued with intermittent oscillation, and I finally got it to shut up by some cut-and try measures. It sounded quite nice afterwards, but I've never been quite convinced that it has adequate gain and phase margin. I used the power and output iron from a Fisher 500B receiver to build that amplifier.
I should drag my "Kingfisher" 7591A push-pull amp back in and give it a run on the gain-phase analyzer at work to see if an output Zobel network would help with overall stability. That amp was plagued with intermittent oscillation, and I finally got it to shut up by some cut-and try measures. It sounded quite nice afterwards, but I've never been quite convinced that it has adequate gain and phase margin. I used the power and output iron from a Fisher 500B receiver to build that amplifier.
Anyway, if one opts for a balanced driver section with a CCS in the common cathode line, as benchbaron said, what's the use of an additional Concertina when the balanced driver would do the job equally well with the 2nd triode's grid grounded?
Best regards!
Best regards!
Well, I wouldn't dare to say that it doesn't work, see e.g. the famous K&H Telewatt amplifiers, featuring EL36's, EL500's or EL5000's as finals.
Best regards!
If a CCS is used, not just a small cathode resistor. A big cathode resistor to a -200V supply works pretty good too. Decent enough CMRR for use in my my bass guitar amp. The other Mullards I’ve done use an MJE340 to a -15V rail.
Zobels also allow one to do “stupid” things like clip into an open circuit, without the output transformer or tubes to suffer the fate of Alderaan.
Using ex TV video tubes as drivers, even strapped as triodes as I do resulting in 25dB gain is a gift, coupled with low internal Z... .. Surplus TV tubes are notorious for wild tolerances between makes, rafted with inconsistencies, but video distortion back in 1960 wasn´t an issue with 405/625 line TV or used as an 807 driver in my G8xxx radio rig. A CCS used in the common cathode (Williamson config) corrects any misbalance between these wild tubes, even though the Cathodyne outputs exhibit perfect balance.
Here´s the trickster.
Measuring each output of the cathodyne with a standard x10 scope probe into a voltmeter or equiv with 10M input impdance, newcomers may be surprised to find out the outputs aren´t quite equal. Don´t be fooled by the probe loading effects ! A high quality Tek probe P6109 or sim will provide the correct result.
However, the 12HG7/12BY7 in full pentode mode unscreened with 40dB gain used for audio stages really isn´t practical for audio interstage as microphony can be bad. These tubes were simply not designed for such. However HK amp designs used them, but requiring copious amounts of feedback to tame. The 12HG7 and lesser extent the 12BY7 are RF wild birds, these tubes willl oscillate nicely with a CCS in the anode; so for audio work I always use a anode resistor as a damper and driven harder than other types, otherwise these tubes also respond well to small amounts of degenerative feedback in the cathode. The drawbacks to this method is well known.
Carbon resistors to many may seem crummy and outdated but I use them on these tubes and other, in the right places.
Great Stuff.
Bench Baron
Last edited:
I didn´t mention that any knowledgable wanting to use ex TV tubes in power amp circuits would avoid the often awful mismatch hassle, and request near matched tubes from a vendor;... then as others rightly mentioned the CCS in the Williamson driver could be dispensed with. With my rambled stock of NOS TV tubes, it took me many hours burning the night oil selecting tubes which were close-enough matches.
However, my dummy test rig is set up just with front end and cathodyne driver using simulated loads for pairs under test, the duff and horrid spec ones can be quickly weeded out.
Home checking for characteristic tube match with pentodes connected as triodes and balance, apart from measuring anode voltages, a scope probe on the joint cathode (CCS connection ) will see 2x f at quite low amplitude perhaps not quite equal despite the cathodyne outputs being equal. The neg CCS in this instance does a grand job. A 1Khz as an input signal will appear on the CCS as 2Khz or act as doubler at quite low amplitude. The symmetrical amplitude or equality of the waveform is the one to take note, and swap tubes, label accordingly. Later, as part of the power amp mainframe I modified by breaking the joint cathode by fitting degeneration resistors in each cathode leg which adds local feedback. It´s all in the Radiotron Handbook, nothing new under the sun.
I am not particulary tidy worker, but this pie looking test rig modified so many times has been with me for nearly 55 years. The black colvern pot, silvered mica caps, and tubes are all from that exciting era.
Keep at it !
Bench Baron
However, my dummy test rig is set up just with front end and cathodyne driver using simulated loads for pairs under test, the duff and horrid spec ones can be quickly weeded out.
Home checking for characteristic tube match with pentodes connected as triodes and balance, apart from measuring anode voltages, a scope probe on the joint cathode (CCS connection ) will see 2x f at quite low amplitude perhaps not quite equal despite the cathodyne outputs being equal. The neg CCS in this instance does a grand job. A 1Khz as an input signal will appear on the CCS as 2Khz or act as doubler at quite low amplitude. The symmetrical amplitude or equality of the waveform is the one to take note, and swap tubes, label accordingly. Later, as part of the power amp mainframe I modified by breaking the joint cathode by fitting degeneration resistors in each cathode leg which adds local feedback. It´s all in the Radiotron Handbook, nothing new under the sun.
I am not particulary tidy worker, but this pie looking test rig modified so many times has been with me for nearly 55 years. The black colvern pot, silvered mica caps, and tubes are all from that exciting era.
Keep at it !
Bench Baron
The output stage distortion contribution in a Williamson is significantly more than from the driver stage, but that said, a 6SN7 typically has noticeably lower harmonic distortion than a 12AU7. It could be worthwhile doing apples to apples HD comparison of the driver stage when trying alternative tube types, or even circuit configuration changes. The driver stage in the Williamson is designed not to reach grid-conduction limits, so as to maintain the loading balance of that stage on the cathodyne out to many hundreds of kHz (assuming a reasonable layout with only minor parasitic capacitance coupling to chassis of wiring and coupling caps).
Agreed, 12AU7 thd noticeably worse than other comparables and yet it revels in Philips amp circuits. That tube for me is a dead end.
Morgan Jones gave it also a thumbs down. Some years ago with my standard jig I did a comparison of various B9A minature pentode output tubes configured as triodes with Pa 6-12W ratings specifically as balanced drivers in mind, exactly over the distortion issue. 6CL6,6CB6,6CH6 12H/G and loads more ex video stuff and so on and found the distortion didn´t vary a great deal but their gain with a given anode load was the dominating factor. (Replacing the resistive anode load with a CCS resulted in a perfect 4-15MHz oscillator). Obviously, casually mentioning this without figures, currents and so on isn´t scientific, but instant plug in results from the 12BY/HG group put me in the desired direction. It is time consuming changing connections as some have center tapped heaters.
Readers peruse at http://www.pmillett.com/pentodes.htm
In consideration for my double 250W parallel P_P power amp, which posed the real achilles heel for driver stages driving parallel output stages at 600V + running fixed bias without running into signifigant distortion with a minimum 6dB voltage headroom, above output stage clipping (nasty requirment), is that grid resistor /tube often stated as max 50Kohms/2 tubes, is a punitive load not yet considering reflected output stage Miller cap.
On my jig I simulated output stage loading with each driver output shunted via DC blocking cap then by a 22K and 500pF cap to ground; that loading is punitive to a driver stage. It implied running the drivers at least 20mA anode current each side, quite a hammer compared to the conventional designs using 5-8mA one comes across. Having done listening tests on several amps I am no fan of "muddly sounding" cathode followers. As a retired switchmode/ magnetics designer I have become accustomed to the standard Totem pole drive concept of class A driver that actively pulls down in both directions, to properly discharge Miller.
There is alot going on here in the beyond, but without doubt the driver stages are often under-represented in a power amp designs; which requires good test equipment and time.
"Experiment and enjoy it, but more important know what one is doing".
Bench Baron
Morgan Jones gave it also a thumbs down. Some years ago with my standard jig I did a comparison of various B9A minature pentode output tubes configured as triodes with Pa 6-12W ratings specifically as balanced drivers in mind, exactly over the distortion issue. 6CL6,6CB6,6CH6 12H/G and loads more ex video stuff and so on and found the distortion didn´t vary a great deal but their gain with a given anode load was the dominating factor. (Replacing the resistive anode load with a CCS resulted in a perfect 4-15MHz oscillator). Obviously, casually mentioning this without figures, currents and so on isn´t scientific, but instant plug in results from the 12BY/HG group put me in the desired direction. It is time consuming changing connections as some have center tapped heaters.
Readers peruse at http://www.pmillett.com/pentodes.htm
In consideration for my double 250W parallel P_P power amp, which posed the real achilles heel for driver stages driving parallel output stages at 600V + running fixed bias without running into signifigant distortion with a minimum 6dB voltage headroom, above output stage clipping (nasty requirment), is that grid resistor /tube often stated as max 50Kohms/2 tubes, is a punitive load not yet considering reflected output stage Miller cap.
On my jig I simulated output stage loading with each driver output shunted via DC blocking cap then by a 22K and 500pF cap to ground; that loading is punitive to a driver stage. It implied running the drivers at least 20mA anode current each side, quite a hammer compared to the conventional designs using 5-8mA one comes across. Having done listening tests on several amps I am no fan of "muddly sounding" cathode followers. As a retired switchmode/ magnetics designer I have become accustomed to the standard Totem pole drive concept of class A driver that actively pulls down in both directions, to properly discharge Miller.
There is alot going on here in the beyond, but without doubt the driver stages are often under-represented in a power amp designs; which requires good test equipment and time.
"Experiment and enjoy it, but more important know what one is doing".
Bench Baron
I realise this post was from august LOL, but I wanted to comment upon this simple truth, in order to push the subversive philosophy that Thorsten Loesch expoused some time ago, which I consider very wise (and much easier for single ended amps!), and I am currently listening to on my Sweet peach modded amplifier, with thunderous bass and crisp clean treble:
Don't put the OPT inside the loop at all.
The transformer does require careful driving - the aim of the amplifier here is to ensure the primary waveform is correct, and lets the secondary do what it wants. Well, I say 'does', but perhaps it doesn't.
I consider this a good idea because:
- The bass rolls off in a transformer so it doesn't get much feedback anyway.
- The treble rolls off in a transformer so it doesn't get much feedback anyway.
- The midrange probably doesn't need much.
The advantages are:
- Any tube to tube feebcak loop is now way faster.
- The speaker cone and secondary are not fighting via a laggy loop, one simply follows the other.
I did wonder if the treble would be gone, and the bass would be gone.. instead I found that both are there, and sound magnificent.
Of course one needs to check the tube feedback is tidy, we don't want any UHF signals in there, but eliminating the OPT from the loop is all good in my experience.
I think today most people include the transformer in the loop as it's easier, a nice neat low voltage to feedback. But I can recommend resisting this luxury, and chopping it out of the loop 😀
I'm not a fan of using descriptions like 'faster' and 'laggy'. They relate imho to the risetime of a step disturbance, and amps with global negative feedback can have quite fast risetimes. The difference in risetime between with/without GNFB could well be way up above 100kHz (<10us). I doubt that has an influence on speaker cone response, or distortion in the audio range. For sure, that implies an output transformer suitable for GNFB, but even an OPT with significant parasitics may be able to be tamed with the right approach.
Checking that tube feedback is tidy is much easier nowadays, especially for diyers, as many more people have better scopes and other gain-phase tools, and access to a collective experience of others that have gone down a particular amp and output transformer path. If there is no collective experience about a particular OPT using GNFB, then yes many diyers would be advised to not use GNFB, but then there is still risk of applying plate feedback if it hadn't been used before or is not a clone effort where it was used.
Checking that tube feedback is tidy is much easier nowadays, especially for diyers, as many more people have better scopes and other gain-phase tools, and access to a collective experience of others that have gone down a particular amp and output transformer path. If there is no collective experience about a particular OPT using GNFB, then yes many diyers would be advised to not use GNFB, but then there is still risk of applying plate feedback if it hadn't been used before or is not a clone effort where it was used.