grimberg said:
That whole thing got started because Morgan Jones suggested that an LM317 wasn't a good enuff current source
dave
That was my exact finding - when I compared a bypassed LM317 in the cathode to a decent plate CCS (cascoded 10M45S or a Gary Pimm CCS) it just didn't stack up.
I imagine the group buy CCS would also perform very well in this application.
regards
pm
mach1,
Would you consider sharing your circuit and experience using the 12B4 in a linestage? If not here maybe in a separate thread called "12B4 linestage variations" or something similar.
🙂
Would you consider sharing your circuit and experience using the 12B4 in a linestage? If not here maybe in a separate thread called "12B4 linestage variations" or something similar.
🙂
I have the original diyAudio CCS boards and wired all of them for HV PNP CCS for plate load.
I'm going to try CCS plate load and choke plate load on my 12B4 😉
I'm going to try CCS plate load and choke plate load on my 12B4 😉
As the guy who innocently started this bypassed-LM317-CCS-in-the-12B4A-cathode discussion, please allow me to clarify, pontificate and spew:
1.) As I explained in earlier posts (not well enough apparently), the role of the bypassed LM317-based CCS in the cathode is VERY modest – its ONLY purpose is to keep the operating point (Ip, and therefore Vpk) of the high-octane 12B4A stable and predictable. As a small side benefit, the RC time constant in the cathode circuit is somewhat lengthened (lower frequency roll-off, or a smaller cap requirement for the same roll-off). But this is a minor benefit. For less than a buck, the LM317 cheerfully performs this simple task quite effectively, and it might even be cheaper than the handsome 1W cathode resistor it replaces.
2.) When a CCS is bypassed in the cathode, as it is here, it is no longer a CCS to audio signals. The audio signal never even sees this CCS, and this CCS never sees audio signals, at least above a few Hertz. The CCS is COMPLETELY swamped by the large bypass cap, which shorts the cathode firmly to ground for all audio AC signals, if we’ve picked the cap well. This does make the choice of the bypass cap the Achilles tendon of this design. It has to be large (>100uF), and therefore might best be a composite of electrolytics and smaller parallel bypasses. Here you can season to taste. Try combinations of your favorite Black gates, big fat poly caps, precious Russian Teflon caps, or those ancient PIOs waiting for a gig in your parts bin. I had to leave something to keep the tweakers busy.
3.) A CCS used as a plate load is a far, FAR, (did I say FAR?) more critical and demanding role for any CCS than is a bypassed cathode CCS, and it must perform an entirely different function. It can’t hide behind a bypass cap. Its extremely high dynamic impedance at all audio frequencies (we only wish!) makes sure than the only load that the gm of the triode sees is the rp of the tube itself, looked at in the Norton equivalent fashion. When the planets align thus, the gain equals rp*gm = mu (well, also throw in a minus sign for those who watch polarities). And mu is generally the most linear of the three interlocked triode parameters. When I hear some folks debating using a CCS as a plate load or as a bypassed cathode bias device, I have to scratch a body part. That’s like debating whether cotton is best used in pillow cases or gym socks. Apples and oranges. Cheese straighteners and door knobs.
4.) The LM317 CCS would suck grievously as a plate load or in almost any role where it would be exposed to audio voltages (if it could even withstand the voltage swings). Its internal opamp-style feedback circuits are just not fast enough or linear enough to be worthy. I would never suggest using one as a plate load, nor even as an un-bypassed CCS in the cathode of a CF. In this modest BYPASSED cathode assignment however, we’ve outlawed any audio voltages across the LM317 above a handful of Hertz. Not allowed, Sign posted.
5.) For a plate-load CCS, I’ll go out on a limb and say that any SS effort short of an all-out, damn-the-torpedoes Gary Pimm-style CCS is probably not going to be worth the effort. I said “probably”. Please prove me wrong. Unadorned MJE350s and IRFXXs just ain’t going to cut it here; sorry. You need not only very high resistance, but constant resistance as a function of voltage swing, AND you need not only very low shunt capacitance, but constant capacitance as a function of voltage swing. The latter is a real heart-breaking challenge for silicon-based creatures. I haven’t tried the 10M45, but I have my doubts, knowing my sonic preferences and my own ears’ idiosyncrasies, despite my admiration for some of its adherents. It’s just another opamp trying to play audio in a tube circuit. Any trace of solid-state’s variable capacitance in the plate-load CCS will take the sound down a notch or two, deflating some of the triode glory, as those of you who’ve read my previous rants about that topic may recall me saying. Without an all-out, full-blown, turbo-charged CCS, I’ll stick instead very gladly with a nice cheap wirewound resistor load on the plate - so long as we keep RL>>rp, a necessary requirement for linear operation of almost any triode. Here, we’ve been talking about RL values that are 6 to 8 times rp, not too shabby. (A comparably-loaded 12AX7 would need a plate load resistor of over 500K, a huge B+, and it couldn’t drive anything more intimidating than the grid of a CF, just to give a comparison of relative plate loading). A simple wirewound plate load can sound dramatically more natural than most run-of-the-mill semiconductor CCSs, unless you like a bit of artificial sheen or zip to the top end. If I ever have my doubts about what top end should sound like, attendance at any acoustic music event, perhaps a string quartet or an un-amplified acoustic guitar with vocal will immediately remind me that natural high-end is silky smooth and free of the electronic textures we’re so used to hearing. Well-conceived triode designs can almost get there. This year’s latest Krell doesn’t cut if for me, but, as always, YMMV. But I digress…
6.) I never suggested this design as the end-all of line stages. It was an easy derivation of a CD output stage I had happily used for years. It can certainly be improved upon, but I will say this: I think it performs extremely well for what it is, a very simple and relatively cheap design. In fact, I will put it up against almost any other line stage topology, at least matched dollar-for-dollar, including that Akimbo or Akidding (what is it?). Sometimes simplicity wins - a mantra I’ve too rarely listened to myself, being an engineer always wanting to make design “enhancements”. The advantage of this design is the very high standing plate current, the low plate resistance (no CF is needed), a low value of mu that’s about perfect for line-stage use without feedback, and, in case nobody noticed, very low Miller capacitance up front. That latter factor is important when a volume control does the driving. Oh, and low mu means that the Vgk has a large negative value – this means that incipient grid current variation is safely out of the picture, something that cannot be said with many higher mu tubes, even though they may never stray into positive grid territory. They can get too close for comfort though. This last forgotten factor matters when the source resistance is relatively high – say, again, like when a volume control is doing the driving. Yes, there is something magical about the sound of a 12B4A used this simple way. I wholeheartedly endorse the shunt gas tube regulator PS design to complement it.
7.) Can it be improved? Of course! You betcha! But not without taking a giant leap up the ladder of complexity and cost. The choke-loading idea is great, but you’ll need a beefy choke which can handle 25mA+ of standing current, has maybe 30H or more of inductance, and has low self-capacitance. Suddenly you’ve got two very big, very heavy and very expensive pieces of iron and copper on your hands. But it could be done, and it might even sound even better. Or you could try the uber-CCS-as-plate-load mentioned above. Granted, it’s harder to make a uber-CCS which cranks out 30mA than it is to make one that performs well at 2mA. But, still, another worthy project, just well beyond the realm of the simple and cheap. And, of course, you can play with power supplies for months.
Geesh, I do ramble when it’s way past my bedtime. Sorry, and good night…
1.) As I explained in earlier posts (not well enough apparently), the role of the bypassed LM317-based CCS in the cathode is VERY modest – its ONLY purpose is to keep the operating point (Ip, and therefore Vpk) of the high-octane 12B4A stable and predictable. As a small side benefit, the RC time constant in the cathode circuit is somewhat lengthened (lower frequency roll-off, or a smaller cap requirement for the same roll-off). But this is a minor benefit. For less than a buck, the LM317 cheerfully performs this simple task quite effectively, and it might even be cheaper than the handsome 1W cathode resistor it replaces.
2.) When a CCS is bypassed in the cathode, as it is here, it is no longer a CCS to audio signals. The audio signal never even sees this CCS, and this CCS never sees audio signals, at least above a few Hertz. The CCS is COMPLETELY swamped by the large bypass cap, which shorts the cathode firmly to ground for all audio AC signals, if we’ve picked the cap well. This does make the choice of the bypass cap the Achilles tendon of this design. It has to be large (>100uF), and therefore might best be a composite of electrolytics and smaller parallel bypasses. Here you can season to taste. Try combinations of your favorite Black gates, big fat poly caps, precious Russian Teflon caps, or those ancient PIOs waiting for a gig in your parts bin. I had to leave something to keep the tweakers busy.
3.) A CCS used as a plate load is a far, FAR, (did I say FAR?) more critical and demanding role for any CCS than is a bypassed cathode CCS, and it must perform an entirely different function. It can’t hide behind a bypass cap. Its extremely high dynamic impedance at all audio frequencies (we only wish!) makes sure than the only load that the gm of the triode sees is the rp of the tube itself, looked at in the Norton equivalent fashion. When the planets align thus, the gain equals rp*gm = mu (well, also throw in a minus sign for those who watch polarities). And mu is generally the most linear of the three interlocked triode parameters. When I hear some folks debating using a CCS as a plate load or as a bypassed cathode bias device, I have to scratch a body part. That’s like debating whether cotton is best used in pillow cases or gym socks. Apples and oranges. Cheese straighteners and door knobs.
4.) The LM317 CCS would suck grievously as a plate load or in almost any role where it would be exposed to audio voltages (if it could even withstand the voltage swings). Its internal opamp-style feedback circuits are just not fast enough or linear enough to be worthy. I would never suggest using one as a plate load, nor even as an un-bypassed CCS in the cathode of a CF. In this modest BYPASSED cathode assignment however, we’ve outlawed any audio voltages across the LM317 above a handful of Hertz. Not allowed, Sign posted.
5.) For a plate-load CCS, I’ll go out on a limb and say that any SS effort short of an all-out, damn-the-torpedoes Gary Pimm-style CCS is probably not going to be worth the effort. I said “probably”. Please prove me wrong. Unadorned MJE350s and IRFXXs just ain’t going to cut it here; sorry. You need not only very high resistance, but constant resistance as a function of voltage swing, AND you need not only very low shunt capacitance, but constant capacitance as a function of voltage swing. The latter is a real heart-breaking challenge for silicon-based creatures. I haven’t tried the 10M45, but I have my doubts, knowing my sonic preferences and my own ears’ idiosyncrasies, despite my admiration for some of its adherents. It’s just another opamp trying to play audio in a tube circuit. Any trace of solid-state’s variable capacitance in the plate-load CCS will take the sound down a notch or two, deflating some of the triode glory, as those of you who’ve read my previous rants about that topic may recall me saying. Without an all-out, full-blown, turbo-charged CCS, I’ll stick instead very gladly with a nice cheap wirewound resistor load on the plate - so long as we keep RL>>rp, a necessary requirement for linear operation of almost any triode. Here, we’ve been talking about RL values that are 6 to 8 times rp, not too shabby. (A comparably-loaded 12AX7 would need a plate load resistor of over 500K, a huge B+, and it couldn’t drive anything more intimidating than the grid of a CF, just to give a comparison of relative plate loading). A simple wirewound plate load can sound dramatically more natural than most run-of-the-mill semiconductor CCSs, unless you like a bit of artificial sheen or zip to the top end. If I ever have my doubts about what top end should sound like, attendance at any acoustic music event, perhaps a string quartet or an un-amplified acoustic guitar with vocal will immediately remind me that natural high-end is silky smooth and free of the electronic textures we’re so used to hearing. Well-conceived triode designs can almost get there. This year’s latest Krell doesn’t cut if for me, but, as always, YMMV. But I digress…
6.) I never suggested this design as the end-all of line stages. It was an easy derivation of a CD output stage I had happily used for years. It can certainly be improved upon, but I will say this: I think it performs extremely well for what it is, a very simple and relatively cheap design. In fact, I will put it up against almost any other line stage topology, at least matched dollar-for-dollar, including that Akimbo or Akidding (what is it?). Sometimes simplicity wins - a mantra I’ve too rarely listened to myself, being an engineer always wanting to make design “enhancements”. The advantage of this design is the very high standing plate current, the low plate resistance (no CF is needed), a low value of mu that’s about perfect for line-stage use without feedback, and, in case nobody noticed, very low Miller capacitance up front. That latter factor is important when a volume control does the driving. Oh, and low mu means that the Vgk has a large negative value – this means that incipient grid current variation is safely out of the picture, something that cannot be said with many higher mu tubes, even though they may never stray into positive grid territory. They can get too close for comfort though. This last forgotten factor matters when the source resistance is relatively high – say, again, like when a volume control is doing the driving. Yes, there is something magical about the sound of a 12B4A used this simple way. I wholeheartedly endorse the shunt gas tube regulator PS design to complement it.
7.) Can it be improved? Of course! You betcha! But not without taking a giant leap up the ladder of complexity and cost. The choke-loading idea is great, but you’ll need a beefy choke which can handle 25mA+ of standing current, has maybe 30H or more of inductance, and has low self-capacitance. Suddenly you’ve got two very big, very heavy and very expensive pieces of iron and copper on your hands. But it could be done, and it might even sound even better. Or you could try the uber-CCS-as-plate-load mentioned above. Granted, it’s harder to make a uber-CCS which cranks out 30mA than it is to make one that performs well at 2mA. But, still, another worthy project, just well beyond the realm of the simple and cheap. And, of course, you can play with power supplies for months.
Geesh, I do ramble when it’s way past my bedtime. Sorry, and good night…
May I saw at that limb a bit? Once one cascodes that unadorned MJE350 with a faster, more fragile PNP, good things happen, both with respect to source impedance and capacitance. Likewise, an unadorned DN2540 is a mediocre CCS, but in a simple self-biased cascode, they work superbly as plate loads. Gary Pimm has some nice graphs showing this. His turbo'd CCS look nice, but really, once you've got the dynamic impedance above 10M across the band, does it really help to get it up to 500M?
I only make this point because the wirewound plate loads are impractical if you want to run the tube with generous amounts of current.
I only make this point because the wirewound plate loads are impractical if you want to run the tube with generous amounts of current.
SY said:
Hi SY,
How was ETF? And Amsterdam?
Sure...saw away. Cascoding certainly is the very first step in adorning a CCS transistor. And of course the lower capacitances of a faster transistor will help enormously. Keep going…
No, I do not think that once you get into the mega-ohms that further increases in resistance will help. Heck, a theoretical "CCS" that could muster just a few tens of kilo-ohms, if those kilo-ohms were dead constant over signal swing, would probably be adequate. There are diminishing returns when you go way above Rccs>>rp. BUT, it is the shunt capacitance of those plain Jane CCSs, and even many simply-adorned cascodes, that is the problem. This small shunt capacitance can change with signal swing, often by a lot percentage-wise. A wirewound doesn't tend to do that. Nor do Gary Pimm's fascinating CCS designs. To his credit he has actually measured shunt C in several CCS variants (see link), as you mentioned. His better designs get down to a tiny fraction of a picofarad. Even I, Mr. Tube Bigot, would agree that if the absolute value of this evil little capacitance is very low, then we can safely ignore significant percentage variations in its value over signal swing. Gary achieves an amazing 0.023pF in his Rev. 5. At that low absolute starting value, I don’t care if it swings wildly to 0.02pF on one signal peak, and to 0.03pF on the opposite peak, because that’s still too low in absolute terms to worry about. A lot of times however, even here in beloved DIYAudio-land, I see cascode CCSs where the lower transistor’s base/gate (speaking of the transistor connected directly to the triode’s plate) is biased via a low-impedance path to ground, maybe via a bypassed resistor divider or diode/zener/LED bias network. Sure the low frequency impedance of the collector/drain that the plate sees is high enough, but the shunt capacitance may not be, since Cdg/Ccb is shunted to ground, or through a low impedance path to ground. In other words, the shunt C doesn’t get the benefit of cascoding. And then you have new-found textures and sheen on music.
Impractical? We’ve been discussing variations on a generously-biased (25mA to 30mA) 12B4A design for more than a year using wirewound plate loads in the 6.2 to 8.2K range (well above the 1k rp of a cooking 12B4A, but not into CCS territory). Sure, these get a bit toasty, but a 10 watt WW part is dirt cheap, and might I say, easy on the ears.
Gary Pimm's CCS measurements
ETF was quite nice. Tim di Paravicini was indeed larger than life.
You've motivated me to go back to the bench and try to get an indirect measure of the capacitance in an MJE350 cascode. My stage measurements in the past have given no indication that this is a measurable issue in-band, but hey, I'm curious.
These days, I'm tending more toward the cascoded DN2540 path- a couple of FETs, a couple of resistors, and the capacitance is wicked low. Ridiculously high dynamic impedance. Two terminals, no batteries.
You've motivated me to go back to the bench and try to get an indirect measure of the capacitance in an MJE350 cascode. My stage measurements in the past have given no indication that this is a measurable issue in-band, but hey, I'm curious.
These days, I'm tending more toward the cascoded DN2540 path- a couple of FETs, a couple of resistors, and the capacitance is wicked low. Ridiculously high dynamic impedance. Two terminals, no batteries.
You went to Amsterdam and all you talk about is Tim di Paravicini? You need to travel with me next time 😉
Pimm's measurement technique looks pretty solid for assessing the absolute value of the shunt C of a CCS. But measuring the effect of its variability within an audio circuit suggests discriminating between two forms of IMD: the conventional AM type, and the PM type, the latter being my main concern. My contention is that even very low levels of PID are obnoxious. A spectrum analyzer doesn't tell you which is which.
Cordell developed such a test set years ago, although I'm not sure it's sensitive enough for us tube-heads. I've been considering improvements. If only I had more time, dammit.
The relatively simple doubled-DN2540 CCS does look promising.
Pimm's measurement technique looks pretty solid for assessing the absolute value of the shunt C of a CCS. But measuring the effect of its variability within an audio circuit suggests discriminating between two forms of IMD: the conventional AM type, and the PM type, the latter being my main concern. My contention is that even very low levels of PID are obnoxious. A spectrum analyzer doesn't tell you which is which.
Cordell developed such a test set years ago, although I'm not sure it's sensitive enough for us tube-heads. I've been considering improvements. If only I had more time, dammit.
The relatively simple doubled-DN2540 CCS does look promising.
Attachments
Brian Beck said:
Wouldn't THAT be the ideal gift for the one who has everything?! Gotta get me one of those. I have lots of cheese that needs straightening. Source?
Sheldon
anybody tried how 12B4 sounds with foolish load in range of 1K2 to 2K7?
I'm talking about this at least 6 pages......and nobody..........
nota bene-I don't have sole piece of 12B4
I'm talking about this at least 6 pages......and nobody..........
nota bene-I don't have sole piece of 12B4
formerly choky, i'll take you up on that... B+ 200V 2K2 load.
i'm not going to get fireworks, am i? 😀
i'm not going to get fireworks, am i? 😀
hi all
come on lads is someone gowning too build this 12B4a pre amp as its meant to be a start valve/tube with no CCS and bring it back to a good preamp
come on lads is someone gowning too build this 12B4a pre amp as its meant to be a start valve/tube with no CCS and bring it back to a good preamp
