Yes >> http://ayumi.cava.jp/audio/ULKF/img221.png
&
yes >> http://ayumi.cava.jp/audio/ULKF/img225.png
10% KNF (cathode negative feedback) used in the above examples.
&
yes >> http://ayumi.cava.jp/audio/ULKF/img225.png
10% KNF (cathode negative feedback) used in the above examples.
If you run into problems with a transformer winder not wanting to wind a transformer with CFB taps, you can always break it out into a parafeed design. The attached diagram is just an example for a single ended circuit. You can also do a push-pull parafeed design too. All you have to specifiy is the correct impedance or turns ratio, primary inductance, and standing DC current for the "plate choke" to obtain your desired percentage of CFB. A 6.6K to 150 ohm single ended transformer is equal to a plate choke with a 15% CFB winding.
Attachments
Thanks for the link. Unfortunately, this article discusses output stages with tertrodes and penthodes with CFB, not pure low-mu triodes with CFB.
Just a quick post, since it has been a while - haven't had enough time to spend in the "lab" recently, but the variac has arrived and awaits testing. The major obstacle at the moment is getting an entire room full of stuff cleaned up and organized enough to be productive. Having moved to a new home not that long ago, everything is in boxes, in the way, and no longer organized in the old familiar ... uh... "geography". 
It's time consuming to have to dig through 6 or 8 boxes to find that part you know is there somewhere...
Anyway, in the last little while I have at least had time to do a bunch more reading... Soaked up O. H. Schade's "Beam Power Tubes" - a good one, wish I'd read it long ago. Also chunks of RDH4 and an old transformer design text, as well as vacuuming up some vintage transformer info from around the web. I now have a much better appreciation of the limitations of transformer design. All of this has influenced my thinking somewhat...
Transformer design becomes significantly compromised as more complications are added. Given that there are few, if any, good tools to correct poor output transformer performance, I now think a better approach is to keep the transformer design focused on achieving wide bandwidth. This means jettisoning unnecessary features, and instead adapt the output stage circuitry to achieve the desired goals.
For example, a zener string with capacitor bypass can be used to lower the g2 voltage, rendering separate screen grid windings unnecessary. The caveat is that B+ must then be tightly regulated to keep g2 bias stable. UL taps are also best brought out from the ends of a winding layer - this is easier to build and gives lower leakage inductance than taps brought out from mid-layer. Although this restricts the available choice of tapping ratios, it may not be a serious handicap, as there remain other options for tuning the distortion performance by manipulating g2 voltage, g1 bias, and load impedance.
I'm undecided at this point whether CFB holds any significant advantage over "Schade" feedback or not. Certainly, the latter is easily adjusted without changing the transformer, but it does result in a lot more resistance in the g1 circuit - something I wish to avoid. It is tempting to give up on CFB at this point and just go back to standard off-the-shelf transformers... Sure would be a lot easier. But for now, I am still going to pursue a custom output transformer with CFB windings.
Hopefully I will make some progress on the UL measurement rig soon.
It's time consuming to have to dig through 6 or 8 boxes to find that part you know is there somewhere...Anyway, in the last little while I have at least had time to do a bunch more reading... Soaked up O. H. Schade's "Beam Power Tubes" - a good one, wish I'd read it long ago. Also chunks of RDH4 and an old transformer design text, as well as vacuuming up some vintage transformer info from around the web. I now have a much better appreciation of the limitations of transformer design. All of this has influenced my thinking somewhat...
Transformer design becomes significantly compromised as more complications are added. Given that there are few, if any, good tools to correct poor output transformer performance, I now think a better approach is to keep the transformer design focused on achieving wide bandwidth. This means jettisoning unnecessary features, and instead adapt the output stage circuitry to achieve the desired goals.
For example, a zener string with capacitor bypass can be used to lower the g2 voltage, rendering separate screen grid windings unnecessary. The caveat is that B+ must then be tightly regulated to keep g2 bias stable. UL taps are also best brought out from the ends of a winding layer - this is easier to build and gives lower leakage inductance than taps brought out from mid-layer. Although this restricts the available choice of tapping ratios, it may not be a serious handicap, as there remain other options for tuning the distortion performance by manipulating g2 voltage, g1 bias, and load impedance.
I'm undecided at this point whether CFB holds any significant advantage over "Schade" feedback or not. Certainly, the latter is easily adjusted without changing the transformer, but it does result in a lot more resistance in the g1 circuit - something I wish to avoid. It is tempting to give up on CFB at this point and just go back to standard off-the-shelf transformers... Sure would be a lot easier. But for now, I am still going to pursue a custom output transformer with CFB windings.
Hopefully I will make some progress on the UL measurement rig soon.
I have been selling the SSE PC board for over 8 years. It uses EL34's, 6L6GC's, 6550's or KT88's. They can be wired in triode, UL or pentode via jumpers or a front panel switch. Cathode feedback (CFB) using the OPT secondary is also jumper or switch selectable. GNFB is not a "simple" option via jumper or switch, but it can be implemented. There is not enough reserve gain in a two stage amp for a lot of GNFB.
About half of the builders implement the switches. Most report the same results I see.
An EL34 in triode mode without any feedback provides the best emulation of a DHT amp, and works great on full range speakers and the typical "DHT" music choices. A KT88 really likes UL with CFB. Here 14 or 15 watts is available with enough damping to deal with things like 15 inch speakers and Pink Floyd or Depeche Mode!
As mentioned before the common thing for a ham radio operator to build in the 1960's was a transmitter using an 807 or other 6L6 based variant. A plate choke is needed to prevent spurious oscillations. They were typically made by winding about 10 turns of #26 (or so) enameled wire over (and in parallel with) the body of a 2 watt carbon composition resistor of about 100 ohms. It was usually mounted right at the top cap if an 807 or 1625 tube was used. Space wound bare wire was common in military surplus and commercial transmitters. Some were even silver plated!
A simple but of course nowhere near perfect way to test selectable UL can be built up by using two OPT's. The most choice of "ratios" can be obtained if both OPT's have multiple speaker impedance taps. Wire one OPT up to the tubes in the usual manner. Wire the second OPT's plate leads up to the output tube's screen grids and connect its CT to the screen grid supply voltage. This can be a different power supply for a different screen voltage, but NEVER turn it on before applying plate voltage. The screen grid will MELT!
With the secondaries wired in parallel the tube sees triode mode (neglecting the OPT imperfections). By adjusting the way the secondary taps are connected the "ratio" between the plate and screen grid "taps" can be changed. Crowhurst (?) advocated this method for a low cost way of emulating the MacIntosh twin coupled circuit. His paper might have even preceeded Mac.
I have tried it for multi tapped UL using cheap OPT's and it does work. It's not perfect but a well stocked junk box can create this for free. The OPT's need not be the same, or even the same impedances. Use the best or biggest one for the plate circuit. Try reversing the polarity of one or both transformers if strange in band suck outs or resonances occur. A damping resistor between the two screen grids might help too.
I built the mosfet circuit I posted about 5 years ago and did see some spurious oscillations during transitions, but it was a sky wired layout hanging in the air above an SSE board. A clean layout should resolve the spurious stuff. The comutation effects should not occur if you are using a sweep tube since the screen voltage will always be considerably lower than the plate's B+ voltage and should never go above it. The plates B+ supply can be used to feed the mosfet and no diodes are needed. My experiments all used a sacrificial Chinese 6L6GC which despite many trips into the red zone, is still alive!!
Ditto.....the SW guys around here get a good laugh at my code. I "invented" a language I called C-- when writing code for test boards. I simply left out all the hard stuff in C that I didn't understand, like pointers. Now 10 years later someone invented the Arduino. Their compiler understands all my old code! There is now an Arduino compatible board using a 80 MHz 32 bit PIC chip that eats my old code.
I have found that the models available for LT spice don't do grid current well, and I have yet to get realistic results attempting to model a screen driven amp. This casts a doubt on UL results.
There was a "discovery" published by a couple of college boys in Electronic Design News a few years ago about a "boosted triode mode" using a 6L6GC that provided extra power output by operating the screen grid 100 or so volts higher than the plate. All of this was "discovered" in a simulation and nobody bothered to actually build the circuit. I knew what would happen, but built it anyway and the 6L6's screen grid lit up like a Christmas Tree! EDN did not answer or publish my email with a glowing picture.
LT spice will not complain about things like dissipation or voltage ratings. It will happilly let me successfully simulate a 500 watt amp with a pair of 6L6GC's.
I learned that a succesful simulation does not guarantee a working circuit. A simulation that doesn't work is often a good predictor of failure providing you aren't trying something where the models are known to be inaccurate, like heavy A2 or screen drive.
Note: the models used in the TubeCad simulators seem to do OK with A2 or AB2. The power output, distortion, and even harmonic distribution numbers matched up pretty good with real world amps I built.
SE amp cad told me I could get 40 watts from an 845 in A2, which nobody believed. I built the amp and get 41.
The PP calculator predicted some ridiculous power and distortion numbers running KT88's in triode AB2 (don't remember the #'s now). I built the amp several years ago and got the numbers the calculator predicted.
Whoa, big post - thanks Tubelab... Was wondering what you might have to say on the subject.
Years ago, I built an EL34 amp with 6.6k output transformer / 40% UL taps (push-pull, mind you)... I implemented the switch to select between UL and triode, and quickly settled on UL. Triode mode was too dark sounding and lacked transparency - possibly due to a little HF rolloff resulting from Miller effect, and was audibly distorted compared to UL. It was also soft in the bass, despite a stiff transistor regulated power supply and oversized OPT. After a few years I removed the switch because it simply wasn't getting used. I also put in a 3PDT switch to select between GNFB networks and fussed with that one, changing the network choices once or twice. Ultimately picked something like 6 or 8dB of feedback as the best based purely on subjective testing. Eventually removed that switch too. However I've always wondered how the amp would sound with only local feedback at the output stage.
Now that i have these nice 6BG6-GA / 7027A or whatever they really are, I think its time to find out. The old EL34 amp, rapidly approaching its 20th anniversary, is probably due for an overhaul. So I have the additional motivation to have a replacement ready when I take it out of service. I'm really enjoying a return to designing with tubes, after working with solid state almost exclusively for the last 10 years.
The two-transformer trick is an interesting one - I may try it, though I don't think i have suitable iron at the moment. Your warning about power sequencing is well taken - I understood as much, but it bears repeating so thanks for calling special attention to it.
C-- ... Heheh, I like that. I don't write C or assembler very often any more. Modern scripting languages are so much easier and more efficient to write with, and on modern processors, no longer suffer serious performance disadvantage vs. C for most uses. Python and unix shell are my tools of choice these days.
I really like the TCJ PP calculator too. Nice to hear that it correlates well with reality. Unfortunately, the version I have only does triode mode, and I believe thats still the case. I'd love to see it expanded to pentode and UL, but I can imagine this would be a huge undertaking.
Turning to the matter at hand, I must be going off the deep end, because I'm starting to seriously consider what it might take to wind my own OPTs...
Years ago, I built an EL34 amp with 6.6k output transformer / 40% UL taps (push-pull, mind you)... I implemented the switch to select between UL and triode, and quickly settled on UL. Triode mode was too dark sounding and lacked transparency - possibly due to a little HF rolloff resulting from Miller effect, and was audibly distorted compared to UL. It was also soft in the bass, despite a stiff transistor regulated power supply and oversized OPT. After a few years I removed the switch because it simply wasn't getting used. I also put in a 3PDT switch to select between GNFB networks and fussed with that one, changing the network choices once or twice. Ultimately picked something like 6 or 8dB of feedback as the best based purely on subjective testing. Eventually removed that switch too. However I've always wondered how the amp would sound with only local feedback at the output stage.
Now that i have these nice 6BG6-GA / 7027A or whatever they really are, I think its time to find out. The old EL34 amp, rapidly approaching its 20th anniversary, is probably due for an overhaul. So I have the additional motivation to have a replacement ready when I take it out of service. I'm really enjoying a return to designing with tubes, after working with solid state almost exclusively for the last 10 years.
The two-transformer trick is an interesting one - I may try it, though I don't think i have suitable iron at the moment. Your warning about power sequencing is well taken - I understood as much, but it bears repeating so thanks for calling special attention to it.
C-- ... Heheh, I like that.
I don't write C or assembler very often any more. Modern scripting languages are so much easier and more efficient to write with, and on modern processors, no longer suffer serious performance disadvantage vs. C for most uses. Python and unix shell are my tools of choice these days.I really like the TCJ PP calculator too. Nice to hear that it correlates well with reality. Unfortunately, the version I have only does triode mode, and I believe thats still the case. I'd love to see it expanded to pentode and UL, but I can imagine this would be a huge undertaking.
Turning to the matter at hand, I must be going off the deep end, because I'm starting to seriously consider what it might take to wind my own OPTs...
Turning to the matter at hand, I must be going off the deep end, because I'm starting to seriously consider what it might take to wind my own OPTs...
Nah ... Quite comfortable at this deep end, HifiZen. I have since adopted it myself after also having had some problem with transformer winders. Nobody wants to layer-wind any more. I in turn don't like the random winding style, mainly because it takes up more space. Also, I cannot calculate transformer parameters accurately without having layer winding.
Quite some fun too, only you must not be concerned about the "time-is-money" routine. I built myself a winder (sorry for being momentary OT) using wood panels(!), a car wiper motor, increasing the speed on the wiper-side x2 and a series BJT speed controller using a sewing machine foot control! Counter etc., and it works quite comfortably thank you.
Just a point with all this technical analysis. One must not forget that feeding the screen is not like feeding a control grid. The screen draws current and thus have its own 'internal resistance' analogous to rp. That's where some difference creeps in between a pentode (EL34) and a beam tetrode (KT88, 6L6). The latter can create its own eventual distortion pattern because of the large variation in G2-current in a beam tube with output. Do this in class-AB to boot and the design becomes rather a labour of experimentation. Problems can be easily solved though by using the correct R.C shunt compensation between G2-anode or over the two cathodes in the case of cathode feedback.
Lastly (although I think it has been mentioned somewhere) - Schade feedback is an 'external' thing; the tube 'does not know about it' and still keeps pentode characterisitcs. UL is an internal thing; it changes the tube characteristics themselves to 'best-of-triode + best-of-pentode' as the graphs show. Where given, it is interesting to compare the distortion vs output characteristics vs bias vs load impedance for various output tubes (Mullard EL34 shows this for several conditions.)
Feels like forever since I posted on this topic - it has been over two months now. Anyway, I wanted to poke my nose in to let everyone know that I haven't given up.
In the interval, I've been away from home almost 6 weeks, so that has really slowed things down. My investigation into transformers and transformer design has continued, taking me on a little detour through the realm of coil winding, and off into the world of CNC... fascinating and fun stuff. I decided to build a small CNC router, and am quite excited about using it on future projects!
A good deal of what I learn with the router would apply to CNC coil winding. However I'm still not sure whether I want to start winding my own transformers. The commercial options are very limited and expensive for what I want to do, so if I really intend to follow this path, it seems almost certain that I will need to wind my own. But, the subject of transformer design and winding is deep enough that it's somewhat at cross purposes to my ultimate goal of just building a good amp. Early in the year, I was hoping to have something new to show at BAF '13, but that's already unrealistic, even if I don't wind any iron. Regardless, I'd like to make some headway and have an output stage design by the end of the year.
I guess that's all for now. I'll have some more time to ruminate on this transformer winding question while I finish up the CNC router and get it running.
The adventure continues....
In the interval, I've been away from home almost 6 weeks, so that has really slowed things down. My investigation into transformers and transformer design has continued, taking me on a little detour through the realm of coil winding, and off into the world of CNC... fascinating and fun stuff. I decided to build a small CNC router, and am quite excited about using it on future projects!
A good deal of what I learn with the router would apply to CNC coil winding. However I'm still not sure whether I want to start winding my own transformers. The commercial options are very limited and expensive for what I want to do, so if I really intend to follow this path, it seems almost certain that I will need to wind my own. But, the subject of transformer design and winding is deep enough that it's somewhat at cross purposes to my ultimate goal of just building a good amp. Early in the year, I was hoping to have something new to show at BAF '13, but that's already unrealistic, even if I don't wind any iron. Regardless, I'd like to make some headway and have an output stage design by the end of the year.
I guess that's all for now. I'll have some more time to ruminate on this transformer winding question while I finish up the CNC router and get it running.
The adventure continues....
OK, time for another follow-up post on this topic, since a lot more time has passed since my last post, and I've done plenty more thinking and reading in the meantime, but alas, precious little doin'...
As a practical matter, I've had precious little time for hobbies in the past months, CNC included. Since I don't foresee a reversal of this trend any time soon, I've decided to drop the idea of winding my own OT's in the interest of actually getting an amplifier built some time this decade. I've also, regrettably, decided I will have to significantly scale back my investigation of UL operating points for the same reason, although I will still be doing some work to optimize an output stage employing a modified UL topology which makes use of an off-the-shelf OT. Hopefully, what comes of that will still be useful to other DIYers who are also restricted to off-the-shelf transformers.
My recent thinking on the subject of UL and CFB is as follows. I'll start with my thoughts on UL...
1) For a conventional UL setup, the literature appears to confirm that for a broad family of tube types (say, KT-66 and 6L6-oids), there isn't going to be a big difference in the distortion profile for the commonly encountered tapping ratios of 33%, 40%, or 43%. These ratios should all give roughly comparable performance. Below about 25%, the full benefit of UL is clearly not realized, so this region is of little interest, while tappings from 50% to 100% appear to be of little or no practical value - one might as well go straight to full-triode if that is the desired set of characteristics.
2) Adding separate screen windings to provide UL screen drive from a lower bias voltage than the plate supply tends to compromise the OT's performance. Since it is critical to achieving the full benefit of UL operation that the screen taps have accurate voltage ratio and phase relative to the full primary voltage at all frequencies, a compromise in the OT quality is especially undesirable. A conventional tapped primary OT has a clear advantage in this respect, and the same rationale also argues against the addition of a separate winding for CFB (more on this later).
3) There appears to be significant freedom to tailor an output stage distortion profile through a combination of B+ voltage, quiescent current, and plate-to-plate load impedance choices. This freedom expands if one considers the use of a series voltage-dropping circuit to reduce the screen DC potential from the primary DC voltage by a fixed amount. A string of LEDs, for example, might do nicely, or perhaps a more elaborate solid-state shunt regulator type circuit, or even just a simple dropping resistor with bypass cap - there is no shortage of options. This further obviates the need for separate screen grid windings.
Conclusion: conventional off-the-shelf OTs actually already offer a good range of options with which to achieve most if not all of the benefit UL has to offer. There is still a bit of a guessing game as to which exact combination of primary Z and tapping ratio will extract the best performance, but within reason they should all behave well.
Now on to the matter of CFB...
From the perspective of electrode potentials inside the tube, there is actually no difference between CFB (furnished by subtracting part of the OT primary winding from between the UL taps and placing the same number of turns in series with the cathodes), and the use of plate-to-grid1 feedback (as illustrated in Fig.33(c) of O.H.Schade's "Beam Power Tubes").
There is also no difference in the drive impedance seen at the secondary - the feedback action is equal in both cases, and reduces the effective rp by the same amount. Actually, there is a small difference, which is that the CFB winding will bring some OT core distortion components into the feedback loop, whereas plate-to-g1 feedback would not.
However, if desired the so-called "E-Linear" circuit which takes FB from the UL tap to grid 1 would include similar OT core distortions in the feedback loop.
External to the tube however, I count three differences worthy of consideration -
First, plate-to-g1 feedback introduces additional resistance in the g1 drive circuit, which need not be present in the case of CFB. This resistance is unwanted if attempting to push into class AB2, or achieve extreme bandwidth, but could be dealt with by inserting a cathode follower (or FET source follower). In the absence of such a follower, class AB2 is likely not intended, in which case the series resistance may be helpful in mitigating grid current induced "blocking distortion".
Second, there are large voltage swings occurring at the output tube cathodes in the case of CFB, placing a heavy strain on the heater insulation. This reliability risk is avoided with grid1 FB.
Third, "Schade" FB is very easy and inexpensive to modify. With CFB, you are probably stuck with only one choice of feedback ratio. While this is not a technical difference, it could be a very significant practical one for the DIYer.
Considering all of this, I am now leaning firmly towards plain UL with feedback brought to grid 1 from either the plates or UL taps. Very conventional and boring! I suppose the standard circuits really aren't all that bad after all ... who knew? But, this has been a very fascinating and enlightening study exercise.
I suppose I ought to cover a side topic, which although not directly related to the output stage topology, has influenced my thinking and preferences.
Regardless of whether "Schade" feedback or CFB is employed, the voltage swing demanded from the driver stage is huge. The difficulty of swinging very large voltages with low distortion runs counter to the goal of building a complete amplifier circuit with desirable distortion character. Some of the preferred driver circuits, such as the Mu-Follower or simple cathode followers, also place large AC voltage swings on heater-cathode insulation, again raising the same reliability concerns as CFB.
Pentodes are an interesting driver option, and seem to be in vogue for bringing feedback from output tube plates or UL taps to the driver plates. However, I am aesthetically drawn more towards utilizing the gain of a such pentode driver to achieve the short-loop output stage feedback, but not it's full output swing (current or voltage). Pentodes can be very linear when swinging smaller voltages. So if feedback is brought from the output tube plates to the driver grids, then the driver pentodes can not only swing small, very linear output voltages, but additionally benefit from significant local feedback around themselves. The driver tubes thus only swing the output voltage needed to drive the output tubes normally, plus the small additional error voltages required to linearize the output stage. I feel this take greater advantage of a pentode driver than the fashionable "plate-to-plate" feedback style.
The Lafayette KT-550 and HK Citation II designs both make for interesting study, and illustrate nicely the driver+output design concept I think I am converging on. I have a different front end in mind, but that's for a different thread.
As a final note, I have acquired a pair of Dyna-clone A-431 transformers for prototyping purposes (~5k impedance when using the 8Ω secondary). Their reputation suggests they should deliver excellent UL performance, but I wonder how other modern transformers would compare - Hashimoto HW-60-5 or Lundahl LL1620 (anyone know the UL tap ratios on either of those?).
Alright, enough for now. I'll be back at some point in the future to post updates on how the UL operating point measurements go.
Keep on keepin on...
As a practical matter, I've had precious little time for hobbies in the past months, CNC included. Since I don't foresee a reversal of this trend any time soon, I've decided to drop the idea of winding my own OT's in the interest of actually getting an amplifier built some time this decade. I've also, regrettably, decided I will have to significantly scale back my investigation of UL operating points for the same reason, although I will still be doing some work to optimize an output stage employing a modified UL topology which makes use of an off-the-shelf OT. Hopefully, what comes of that will still be useful to other DIYers who are also restricted to off-the-shelf transformers.
My recent thinking on the subject of UL and CFB is as follows. I'll start with my thoughts on UL...
1) For a conventional UL setup, the literature appears to confirm that for a broad family of tube types (say, KT-66 and 6L6-oids), there isn't going to be a big difference in the distortion profile for the commonly encountered tapping ratios of 33%, 40%, or 43%. These ratios should all give roughly comparable performance. Below about 25%, the full benefit of UL is clearly not realized, so this region is of little interest, while tappings from 50% to 100% appear to be of little or no practical value - one might as well go straight to full-triode if that is the desired set of characteristics.
2) Adding separate screen windings to provide UL screen drive from a lower bias voltage than the plate supply tends to compromise the OT's performance. Since it is critical to achieving the full benefit of UL operation that the screen taps have accurate voltage ratio and phase relative to the full primary voltage at all frequencies, a compromise in the OT quality is especially undesirable. A conventional tapped primary OT has a clear advantage in this respect, and the same rationale also argues against the addition of a separate winding for CFB (more on this later).
3) There appears to be significant freedom to tailor an output stage distortion profile through a combination of B+ voltage, quiescent current, and plate-to-plate load impedance choices. This freedom expands if one considers the use of a series voltage-dropping circuit to reduce the screen DC potential from the primary DC voltage by a fixed amount. A string of LEDs, for example, might do nicely, or perhaps a more elaborate solid-state shunt regulator type circuit, or even just a simple dropping resistor with bypass cap - there is no shortage of options. This further obviates the need for separate screen grid windings.
Conclusion: conventional off-the-shelf OTs actually already offer a good range of options with which to achieve most if not all of the benefit UL has to offer. There is still a bit of a guessing game as to which exact combination of primary Z and tapping ratio will extract the best performance, but within reason they should all behave well.
Now on to the matter of CFB...
From the perspective of electrode potentials inside the tube, there is actually no difference between CFB (furnished by subtracting part of the OT primary winding from between the UL taps and placing the same number of turns in series with the cathodes), and the use of plate-to-grid1 feedback (as illustrated in Fig.33(c) of O.H.Schade's "Beam Power Tubes").
There is also no difference in the drive impedance seen at the secondary - the feedback action is equal in both cases, and reduces the effective rp by the same amount. Actually, there is a small difference, which is that the CFB winding will bring some OT core distortion components into the feedback loop, whereas plate-to-g1 feedback would not.
However, if desired the so-called "E-Linear" circuit which takes FB from the UL tap to grid 1 would include similar OT core distortions in the feedback loop.
External to the tube however, I count three differences worthy of consideration -
First, plate-to-g1 feedback introduces additional resistance in the g1 drive circuit, which need not be present in the case of CFB. This resistance is unwanted if attempting to push into class AB2, or achieve extreme bandwidth, but could be dealt with by inserting a cathode follower (or FET source follower). In the absence of such a follower, class AB2 is likely not intended, in which case the series resistance may be helpful in mitigating grid current induced "blocking distortion".
Second, there are large voltage swings occurring at the output tube cathodes in the case of CFB, placing a heavy strain on the heater insulation. This reliability risk is avoided with grid1 FB.
Third, "Schade" FB is very easy and inexpensive to modify. With CFB, you are probably stuck with only one choice of feedback ratio. While this is not a technical difference, it could be a very significant practical one for the DIYer.
Considering all of this, I am now leaning firmly towards plain UL with feedback brought to grid 1 from either the plates or UL taps. Very conventional and boring! I suppose the standard circuits really aren't all that bad after all ... who knew?
But, this has been a very fascinating and enlightening study exercise.I suppose I ought to cover a side topic, which although not directly related to the output stage topology, has influenced my thinking and preferences.
Regardless of whether "Schade" feedback or CFB is employed, the voltage swing demanded from the driver stage is huge. The difficulty of swinging very large voltages with low distortion runs counter to the goal of building a complete amplifier circuit with desirable distortion character. Some of the preferred driver circuits, such as the Mu-Follower or simple cathode followers, also place large AC voltage swings on heater-cathode insulation, again raising the same reliability concerns as CFB.
Pentodes are an interesting driver option, and seem to be in vogue for bringing feedback from output tube plates or UL taps to the driver plates. However, I am aesthetically drawn more towards utilizing the gain of a such pentode driver to achieve the short-loop output stage feedback, but not it's full output swing (current or voltage). Pentodes can be very linear when swinging smaller voltages. So if feedback is brought from the output tube plates to the driver grids, then the driver pentodes can not only swing small, very linear output voltages, but additionally benefit from significant local feedback around themselves. The driver tubes thus only swing the output voltage needed to drive the output tubes normally, plus the small additional error voltages required to linearize the output stage. I feel this take greater advantage of a pentode driver than the fashionable "plate-to-plate" feedback style.
The Lafayette KT-550 and HK Citation II designs both make for interesting study, and illustrate nicely the driver+output design concept I think I am converging on. I have a different front end in mind, but that's for a different thread.
As a final note, I have acquired a pair of Dyna-clone A-431 transformers for prototyping purposes (~5k impedance when using the 8Ω secondary). Their reputation suggests they should deliver excellent UL performance, but I wonder how other modern transformers would compare - Hashimoto HW-60-5 or Lundahl LL1620 (anyone know the UL tap ratios on either of those?).
Alright, enough for now. I'll be back at some point in the future to post updates on how the UL operating point measurements go.
Keep on keepin on...
I agree with your whole post. The quoted bit is quite true if one wants to add the extra effect of the cathode feedback (only phase reversal will require cross-coupled NFB if I understand correctly). Yes, it makes one think. (Not to over-complicate: Using triode drivers will still cancel their 2nd harmonic products, and pentodes can be finicky for best performance distortion-wise, still.)
I would add that I have never done over-experimentation with UL. As the tap proceeds, characteristics mostly do not render a definite optimum point as per published data - it depends on a trade-off between characteristics. In this respect you might also look up the very analytical articles by Langford-Smith (of RDH-4 fame) and Chesterton, in the old Australian 'Radiotronics' magazine. I do not have a direct reference, but it was a series of three articles in the May, June and July 1955 editions (they are on internet). (Caviat: Just be careful: In there the writers use impedance rations, not turns-ratios for the taps. Confusing; the specific impedance ratios are never used in practice.)
Another illutrative graph of how the parameters vary with an KT88 can be found in various articles - again apology for not having that now. I think googling 'ultra-linear' will yield results. I will try and post next time.
Thus, as said, your points taken. Might make me rethink future efforts .... (Though not of such consequence when one winds own transformers - or should one include the lot in a smaller cathode winding of the driver? Maybe the voice coil volatge will be sufficient? Now that's a point ....)
FOLLOW:
The KT88 graphs can be found at ultra-linear. It is along article, starting with the Hafler-Keroes contribution. About 66% down you will find a heading: Ultra-linear circuit characteristics That will describe KT88 operation, followed by the intended graphs, on a horizontal axis of screen taps 0% - 100%.
These differ from other tubes, but the tendencies are the same. (Notice that here the screen initially actually contributes to the output as one might expect, although Langford-Smith did not find that. Nevertheless.)
The KT88 graphs can be found at ultra-linear. It is along article, starting with the Hafler-Keroes contribution. About 66% down you will find a heading: Ultra-linear circuit characteristics That will describe KT88 operation, followed by the intended graphs, on a horizontal axis of screen taps 0% - 100%.
These differ from other tubes, but the tendencies are the same. (Notice that here the screen initially actually contributes to the output as one might expect, although Langford-Smith did not find that. Nevertheless.)
Many Radiotronics Magazine Scans are here:
electron Tube Data sheets - AWV, Amalgamated Wireless Valve Company
Slow loading - be patient.
The 3 Ultralinear Articles are in May, June, July 1955 (1955-05, 1955-06, 1955-07).
Ultralinear + Cathode Feedback was named "Super Triode" by Menno (despite Japanese guys having previously used that name for something completely different), a summary can be found here:
http://www.mennovanderveen.nl/nl/download/download_3.pdf
Also look at the VDV-2100-CFB/H Toroidal Output Transformers which are designed for combined UL and CF. I have a pair of these sitting on my shelf. Intending to throw a quad of KT88 at them.
Also worth having a look at Rob's work (RMS Acoustics)
Motional feedback intro
He found a pair iof KT88 was fine to drive the VDV2100-CFB/H.
Note that that tranny has 10% Cathode Feedback Windings and 30% Ultralinear, When you combine the 2 you get and effective 40% Ultralinear (Cathode to Screen volts ratio).
Cheers,
Ian
electron Tube Data sheets - AWV, Amalgamated Wireless Valve Company
Slow loading - be patient.
The 3 Ultralinear Articles are in May, June, July 1955 (1955-05, 1955-06, 1955-07).
Ultralinear + Cathode Feedback was named "Super Triode" by Menno (despite Japanese guys having previously used that name for something completely different), a summary can be found here:
http://www.mennovanderveen.nl/nl/download/download_3.pdf
Also look at the VDV-2100-CFB/H Toroidal Output Transformers which are designed for combined UL and CF. I have a pair of these sitting on my shelf. Intending to throw a quad of KT88 at them.
Also worth having a look at Rob's work (RMS Acoustics)
Motional feedback intro
He found a pair iof KT88 was fine to drive the VDV2100-CFB/H.
Note that that tranny has 10% Cathode Feedback Windings and 30% Ultralinear, When you combine the 2 you get and effective 40% Ultralinear (Cathode to Screen volts ratio).
Cheers,
Ian
I haven't read through the whole thread, but I see a lot of agonizing over UL, "Schade" and OT limitations.
First some general comments. On the cheaper Edcor and Hammond UL OTs that I have measured, the central B+ to UL winding section(s) is generally well coupled to the secondary for UL purposes. I often see (measuring leakage inductances of the individual primary sections to the secondary, correcting for turns) one of the outer plate to UL sections poorly coupled. A poorly coupled section is best driven by a high impedance pentode plate since the leakage L is automatically compensated for by a current source.
The UL taps are a good place to take voltage feedback from, due to their close coupling to the secondary. +1 for UL mode, but the screens have a non-linear impedance due to the large variation in screen current. -1 for UL.
The usual plate to grid1 "Schade" resistive approach makes for a difficult to drive grid for the driver stage. Going back to the driver grids instead requires cross-coupled neg. feedbacks. This means the feedback(s) has to go thru the OT in class B mode, phase issues. In any case, the assymmetric leakage L's at the plate winding taps make for a poor feedback source, they should come off the UL taps instead.
If you take the feedback resistors back to the driver cathodes, then no more cross-coupling or hard to drive grids. (well the driver grids require more signal, but this is low level already)
The RCA handbook amplifier uses this (plus some Schade). Then taking the resistive feedbacks from the UL taps makes for more accurate coupling to the secondary, and helps attenuate the signal some as well.
On the downside, feedbacks to the driver stage cathodes seem to get poor "sound" reviews. Probably because they work too well to remove the distortion. With pentode drivers, the local loop gain can be phenomenal (no signal thru the OT, ie, no cross coupling in feedbacks, minimal phase issue). Maybe why RCA lightened up on it with partial "Schade", or to soften the saturation/clipping.
So far a standard UL OT will work fine. If you are going to wind your own OT, then I would add a winding with the same number of turns as the UL to UL taps, insulated for ground voltage, and put on as an afterthought (ie, kept out of the way of the closely coupled windings.) Then put caps between its ends and the UL taps. Voila.... closely coupled now. Can use it for a separate UL winding or a CFB winding OR:
Going further yet, The UL sections are the best coupled sections, but we normally use only one of them at a time in class aB. Why not use both at once all the time, a 2X better OT then, since the leakage L is halved. So we go Circlotron/Mac like by connecting the plates to the usual plate taps, and the cathodes to the extra winding ends (opposite UL tap end). Center point of the extra winding is grounded. (no need for two floating B+ like the Circlotron this way) Each tube effectively uses both UL sections and one plate section now. The plate section is operated by a high Z plate, so leakage does not matter there.
For 40% UL taps, this effectively puts 1.4 X turns on each tube from the usual, so the primary Z is doubled. So the entire OT can have its turns dropped by 30% (0.707) to get back to the same primary Z. This all works out to the same number of primary turns as the usual OT (including the extra winding here now). You now have a 1/2 Primary Z OT xformer to wind (as far as the tightly coupled windings are concerned) so expect another 2X bandwidth performance. Total is 4X better now, for no extra total primary turns. Ie, Mac performance with an easy to wind OT. Note that the tubes now have one UL to B+ section (in the added winding actually) acting as CFB, giving 28% CFB (40/(100+40). You can still do the above driver scheme too, or others.
ALL OTs should be wound this way by default. So flexible, can do most anything well.
First some general comments. On the cheaper Edcor and Hammond UL OTs that I have measured, the central B+ to UL winding section(s) is generally well coupled to the secondary for UL purposes. I often see (measuring leakage inductances of the individual primary sections to the secondary, correcting for turns) one of the outer plate to UL sections poorly coupled. A poorly coupled section is best driven by a high impedance pentode plate since the leakage L is automatically compensated for by a current source.
The UL taps are a good place to take voltage feedback from, due to their close coupling to the secondary. +1 for UL mode, but the screens have a non-linear impedance due to the large variation in screen current. -1 for UL.
The usual plate to grid1 "Schade" resistive approach makes for a difficult to drive grid for the driver stage. Going back to the driver grids instead requires cross-coupled neg. feedbacks. This means the feedback(s) has to go thru the OT in class B mode, phase issues. In any case, the assymmetric leakage L's at the plate winding taps make for a poor feedback source, they should come off the UL taps instead.
If you take the feedback resistors back to the driver cathodes, then no more cross-coupling or hard to drive grids. (well the driver grids require more signal, but this is low level already)
The RCA handbook amplifier uses this (plus some Schade). Then taking the resistive feedbacks from the UL taps makes for more accurate coupling to the secondary, and helps attenuate the signal some as well.
On the downside, feedbacks to the driver stage cathodes seem to get poor "sound" reviews. Probably because they work too well to remove the distortion. With pentode drivers, the local loop gain can be phenomenal (no signal thru the OT, ie, no cross coupling in feedbacks, minimal phase issue). Maybe why RCA lightened up on it with partial "Schade", or to soften the saturation/clipping.
So far a standard UL OT will work fine. If you are going to wind your own OT, then I would add a winding with the same number of turns as the UL to UL taps, insulated for ground voltage, and put on as an afterthought (ie, kept out of the way of the closely coupled windings.) Then put caps between its ends and the UL taps. Voila.... closely coupled now. Can use it for a separate UL winding or a CFB winding OR:
Going further yet, The UL sections are the best coupled sections, but we normally use only one of them at a time in class aB. Why not use both at once all the time, a 2X better OT then, since the leakage L is halved. So we go Circlotron/Mac like by connecting the plates to the usual plate taps, and the cathodes to the extra winding ends (opposite UL tap end). Center point of the extra winding is grounded. (no need for two floating B+ like the Circlotron this way) Each tube effectively uses both UL sections and one plate section now. The plate section is operated by a high Z plate, so leakage does not matter there.
For 40% UL taps, this effectively puts 1.4 X turns on each tube from the usual, so the primary Z is doubled. So the entire OT can have its turns dropped by 30% (0.707) to get back to the same primary Z. This all works out to the same number of primary turns as the usual OT (including the extra winding here now). You now have a 1/2 Primary Z OT xformer to wind (as far as the tightly coupled windings are concerned) so expect another 2X bandwidth performance. Total is 4X better now, for no extra total primary turns. Ie, Mac performance with an easy to wind OT. Note that the tubes now have one UL to B+ section (in the added winding actually) acting as CFB, giving 28% CFB (40/(100+40). You can still do the above driver scheme too, or others.
ALL OTs should be wound this way by default. So flexible, can do most anything well.
Last edited:
Well, I don't know about agonizing... personally, this is more of a pleasant cogitation. Having designed around the commonly-accepted wisdom in the past without delving too much into the details, I am now making an endeavour to dive deeper and really think about the whys and wherefores. In some respects, this is more of an intellectual journey than a practical one, though the end goal is of course to build a better amplifier.
Some thought provoking responses here...
Johan -
your comment about the fussiness of pentodes is well taken. I have not really worked with them much before, but reading Schade's paper sparked an interest, and I'm quite appetized by what I have read elsewhere, too! The RDH4 treatment of pentodes is very good, and has sold me on the concept of giving them a try as driver and/or input tube.
That KT88 graph is indeed a good "summary" plot. There's a similar graph (I forget where, probably one of the Langford-Smith / Chesterman articles) which also shows the IMD curve - quite interesting, and reinforces that the ~40% turns range is about optimal, if minimizing IMD is the goal (and it is certainly one of my goals for this next amp design).
Ian -
Thanks, three good references I have not seen before. The AWV Radiotronics archive looks like a treasure trove of information. The RMS Acoustics page looks pretty good too - I'll have to spend some time going through these. I see Rob has a nice looking auto-bias circuit for his amp... That's another side-topic I've been exploring lately, but also a discussion for another thread.
smoking-amp -
Lots of food for thought here...
My conceptual notion of screen current is that it's AC components are essentially the worst of the distortion components (most of the high order junk) subtracted from the cathode current as it transits the tube. Re-uniting these electrons with their long-lost plate brethren should thus mostly cancel the high order distortions, leaving the more benign low-order distortion currents inherent in the g1-k interaction. Frank Blohbaum puts this to good use with his (also partition-noise-cancelling) BestPentode circuit idea. I've yet to really test this hypothesis myself, however Frank's data in Linear Audio vol.0 is compelling, and it seems to mesh with my read of Schade. If correct, then although the screen currents are indeed very distorted, the primary winding portion between B+ and UL tap should be carrying the (presumed linear) summed plate + screen currents, while the portion of the primary winding from UL tap to plate should be carrying the most distorted current in the OT. If you buy that argument, then UL scores a +2! (?) ... and it argues even more strongly for taking FB from the UL tap and not the plate.
More good points you raise about asymmetric leakage L's involved with FB from the plates, and about the cross-connected FB being more exposed to the class-B discontinuities. Hadn't thought too much about that so far. Taking FB to the driver cathodes is certainly a thought, though at first blush seems difficult with high-gm types (low-impedance FB network). In any case, my instinct is to apply the bulk of the driver tube's gain into it's own FB loop, keeping the amount of FB in the output tube loop relatively modest. One could certainly play with that ratio and search for a nice balance.
Having settled on a conventional OT for the current project, I'll have to return to the custom OT concepts one day when I am ready to invest the time. But I like your thinking on the OT compromises and solutions. I would perhaps pick a ~17% CFB ratio, or ~20% UL tap with equal tertiary, and keeping the same total # of turns. That would give a more conventional ~40% UL ratio seen internally by the tube, and reign in the grid drive voltage swing needed. The caps to improve tertiary coupling are a nice touch, similar to the old trick of bypassing a 1:1 IT with caps to improve BW and phase at the high frequencies ... Fisher 50A and the like. An elegant solution, filed for future reference.
Regards,
Some thought provoking responses here...
Johan -
your comment about the fussiness of pentodes is well taken. I have not really worked with them much before, but reading Schade's paper sparked an interest, and I'm quite appetized by what I have read elsewhere, too! The RDH4 treatment of pentodes is very good, and has sold me on the concept of giving them a try as driver and/or input tube.
That KT88 graph is indeed a good "summary" plot. There's a similar graph (I forget where, probably one of the Langford-Smith / Chesterman articles) which also shows the IMD curve - quite interesting, and reinforces that the ~40% turns range is about optimal, if minimizing IMD is the goal (and it is certainly one of my goals for this next amp design).
Ian -
Thanks, three good references I have not seen before. The AWV Radiotronics archive looks like a treasure trove of information. The RMS Acoustics page looks pretty good too - I'll have to spend some time going through these. I see Rob has a nice looking auto-bias circuit for his amp... That's another side-topic I've been exploring lately, but also a discussion for another thread.
smoking-amp -
Lots of food for thought here...
My conceptual notion of screen current is that it's AC components are essentially the worst of the distortion components (most of the high order junk) subtracted from the cathode current as it transits the tube. Re-uniting these electrons with their long-lost plate brethren should thus mostly cancel the high order distortions, leaving the more benign low-order distortion currents inherent in the g1-k interaction. Frank Blohbaum puts this to good use with his (also partition-noise-cancelling) BestPentode circuit idea. I've yet to really test this hypothesis myself, however Frank's data in Linear Audio vol.0 is compelling, and it seems to mesh with my read of Schade. If correct, then although the screen currents are indeed very distorted, the primary winding portion between B+ and UL tap should be carrying the (presumed linear) summed plate + screen currents, while the portion of the primary winding from UL tap to plate should be carrying the most distorted current in the OT. If you buy that argument, then UL scores a +2! (?)
... and it argues even more strongly for taking FB from the UL tap and not the plate.More good points you raise about asymmetric leakage L's involved with FB from the plates, and about the cross-connected FB being more exposed to the class-B discontinuities. Hadn't thought too much about that so far. Taking FB to the driver cathodes is certainly a thought, though at first blush seems difficult with high-gm types (low-impedance FB network). In any case, my instinct is to apply the bulk of the driver tube's gain into it's own FB loop, keeping the amount of FB in the output tube loop relatively modest. One could certainly play with that ratio and search for a nice balance.
Having settled on a conventional OT for the current project, I'll have to return to the custom OT concepts one day when I am ready to invest the time. But I like your thinking on the OT compromises and solutions. I would perhaps pick a ~17% CFB ratio, or ~20% UL tap with equal tertiary, and keeping the same total # of turns. That would give a more conventional ~40% UL ratio seen internally by the tube, and reign in the grid drive voltage swing needed. The caps to improve tertiary coupling are a nice touch, similar to the old trick of bypassing a 1:1 IT with caps to improve BW and phase at the high frequencies ... Fisher 50A and the like. An elegant solution, filed for future reference.
Regards,
Last edited:
The local feedback network for driver cathodes is not really a problem with high gm drivers because you only need a little feedback signal then. The feedback attenuator automatically ends up with a low value cathode R in the divider, and when taken from the UL taps, could be even lower impedance. Take a look at these schematics (which could use some high gm driver updating, like 6JC6 or 6BN11/6J11 maybe instead of the 6CB5 or 6AU6, that 12AU7 can go too):
http://www.pmillett.com/file_downloads/RCA_HiFi.pdf (pages 14,15) or the RCA RC-30 tube handbook, pages 696,697 or with the cathode feedbacks only: Electronics 35
Combining the screen current back with the plate current in UL should be quite helpful for reducing the distortion. One might even consider a Mosfet follower/cascode assist to put it back in at the plate nodes instead. (cap from drain to plate, gyrator from drain to B++) Several versions of that around.
http://www.pmillett.com/file_downloads/RCA_HiFi.pdf (pages 14,15) or the RCA RC-30 tube handbook, pages 696,697 or with the cathode feedbacks only: Electronics 35
Combining the screen current back with the plate current in UL should be quite helpful for reducing the distortion. One might even consider a Mosfet follower/cascode assist to put it back in at the plate nodes instead. (cap from drain to plate, gyrator from drain to B++) Several versions of that around.
Last edited:
Guys,
For your possible interest. Schematic at post #602 "Schade" style feedback from the anodes vs. schematic post #604 identical except feedback from the UL taps.
http://www.diyaudio.com/forums/tubes-valves/72536-el84-amp-baby-huey-61.html
There was something intially atractive to the sound when taking feedback from the UL taps but I eventually abandonned it and went back to anode connections after extended listening and swapping back and forth a few times. Yves (dupv) felt the same (BH post #828) and felt that the "focus and sound stage" were not as good with feedback from the UL tap.
Don (smoking-amp) - thanks for the RCA stuff. I also have a print out of your little scribble from BH post #753 in my "development folder" to think about.
There was some good discussion in there (posts 600 thru' 750) with good input from Don and Revintage and some less useful stuff from Alex on his way to getting "sin-binned" AGAIN.
Have been thinking about 6CL6 for drivers for the quad of KT88 into VDV-2100-CFB/H with UL + Cathode Feedback. Have plenty of 6AU6 too. I don't need really grunty drivers as I will use 1 off mosfet source follower(current source loaded) direct coupled to each output tube to buffer the driver and take firm control of the output tube grid, Robs bias servo (modified with current source load to -ve supply) to apply individual bias to the source follower gates.
Cheers,
Ian
For your possible interest. Schematic at post #602 "Schade" style feedback from the anodes vs. schematic post #604 identical except feedback from the UL taps.
http://www.diyaudio.com/forums/tubes-valves/72536-el84-amp-baby-huey-61.html
There was something intially atractive to the sound when taking feedback from the UL taps but I eventually abandonned it and went back to anode connections after extended listening and swapping back and forth a few times. Yves (dupv) felt the same (BH post #828) and felt that the "focus and sound stage" were not as good with feedback from the UL tap.
Don (smoking-amp) - thanks for the RCA stuff. I also have a print out of your little scribble from BH post #753 in my "development folder" to think about.
There was some good discussion in there (posts 600 thru' 750) with good input from Don and Revintage and some less useful stuff from Alex on his way to getting "sin-binned" AGAIN.
Have been thinking about 6CL6 for drivers for the quad of KT88 into VDV-2100-CFB/H with UL + Cathode Feedback. Have plenty of 6AU6 too. I don't need really grunty drivers as I will use 1 off mosfet source follower(current source loaded) direct coupled to each output tube to buffer the driver and take firm control of the output tube grid, Robs bias servo (modified with current source load to -ve supply) to apply individual bias to the source follower gates.
Cheers,
Ian
re: Ian
Hey, thanks for reminding me about that idea (753), I had forgotten all about it!
EL84 Amp - Baby Huey - Page 76 - diyAudio
Pentode does the grunt work, and the CCS'd triode lays down the law. An application of a composite amplifier stage, with a Vamp and an Iamp in parallel. Should be usable in lots of places.
I wonder why the difference in sound between UL and plate taps. I'm always suspicious that some distortion is "adding" something extra to the sound. Would be interesting to compare FFTs between the two.
Hey, thanks for reminding me about that idea (753), I had forgotten all about it!
EL84 Amp - Baby Huey - Page 76 - diyAudio
Pentode does the grunt work, and the CCS'd triode lays down the law. An application of a composite amplifier stage, with a Vamp and an Iamp in parallel. Should be usable in lots of places.
I wonder why the difference in sound between UL and plate taps. I'm always suspicious that some distortion is "adding" something extra to the sound. Would be interesting to compare FFTs between the two.
Last edited:
Hi gents!
Sorry to dig up an old thread, but.
Does someone by chance has the content of Ayumi's Lab ULCFB articles with the will to share it? The link seems to be broken.
404 Not Found
Thanks!
Sorry to dig up an old thread, but.
Does someone by chance has the content of Ayumi's Lab ULCFB articles with the will to share it? The link seems to be broken.
404 Not Found
Thanks!
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.