A CFP is sensitive to the drive impedance from the ~VAS because a high impedance means that the ~"poles" of both devices are too close together. An active pull-off vs a resistor on the OP may work but gets complicated, a tricky cost/benefit decision. And we have to watch for clipping stability.
This is what I was talking about with an NPN only output stage, I snipped out the PNPs
added a diode. Didn't try to run the simulation but I believe that it could be made to work
as flawed as it might be. Use whatever front end you like, you could leave it with gain or not:
You could make the top half a normal triple EF, then set the lower half to unity gain.
While it is tempting to remove Q15 and 16, take a look at the clipping/sticking behavior with
and without before doing it.
added a diode. Didn't try to run the simulation but I believe that it could be made to work
as flawed as it might be. Use whatever front end you like, you could leave it with gain or not:
You could make the top half a normal triple EF, then set the lower half to unity gain.
While it is tempting to remove Q15 and 16, take a look at the clipping/sticking behavior with
and without before doing it.
Last edited:
It still requires Q11. That type (or anything like it) wasn’t around way back when. Best you could do was an MJE350 or one of those TO-66 versions.
I don’t know how much of a stabilizing effect running a CFP in parallel with an EF has, which was what the original circuit tried to do. And then put the whole thing in a local 4 stage CFA. In any case I bet it‘s complicated and at the mercy of how good the models are. I dont think I’d try to build anything like it.
I don’t know how much of a stabilizing effect running a CFP in parallel with an EF has, which was what the original circuit tried to do. And then put the whole thing in a local 4 stage CFA. In any case I bet it‘s complicated and at the mercy of how good the models are. I dont think I’d try to build anything like it.
You are correct about the CFP being sensitive to the VAS source impedance. VAS circuits with Miller compensation tend to have a low output impedance at the higher frequencies where it matters. However, sometimes that impedance can peak at some higher frequencies. For that reason I often put a Zobel series R-C network on the output of the VAS shunting to ground. Sometimes 100 pF and 100 ohms is enough of a shunt. Simulating the output impedance of the VAS circuit can help with this issue.
Cheers,
Bob
I agree with your concerns. Bryston amplifiers have a pretty good reputation, but that particular Bryston appears not to be a very good design.
Cheers,
Bob
I'm talking about a modern day quasi, you need drivers no matter what so Q11 should not be a problem.
It sounds like you've built all sorts of versions, so it is just another alternative. I was going to say check
out the Crest that Bob mentioned but it looks like from your response you are familiar with them.
I think that a problem with CFPs is that stability is difficult enough but if you let the bigger devices saturate
then it becomes worse, so keep both the driver and outputs out of saturation.
The basic Bryston that used a unity gain triple like that (with MJL3281 and 1302 off each driver) was pretty good, reliable, and stable. The circuit above is about a bridge and a half too far.
Modern-day Quasi’s don’t need to do anything “stupid” - one can drive the upper bank with MJE15032 and lower with 15033 and not look back. Turn it on and it works. As long as the rail voltage is constant. With todays sustained beta outputs triples aren’t “necessary”, although it doesn’t hurt when driving four pairs of speakers with an amplifier.
The Crest 8000 series was the original H, G or whatever class. Straightforward implementation, lots of MJL3281/1302. I have a rack full of the more recent version - the CA series. They used complementary mosfet banks to switch in the upper rails in a hard switched arrangement. And switched over to cheaper Toshiba outputs. I’ve done both types of class G,H but for the big stuff its hard switched. Fitting 20 active outputs and three pairs of commutators on a heat sink is enough of a battle - I don’t need to go doubling or tripling the package count. I use the QSC type commutators, but on more conventional EF3 based amplifier circuits, which can share a common power supply.
Modern-day Quasi’s don’t need to do anything “stupid” - one can drive the upper bank with MJE15032 and lower with 15033 and not look back. Turn it on and it works. As long as the rail voltage is constant. With todays sustained beta outputs triples aren’t “necessary”, although it doesn’t hurt when driving four pairs of speakers with an amplifier.
The Crest 8000 series was the original H, G or whatever class. Straightforward implementation, lots of MJL3281/1302. I have a rack full of the more recent version - the CA series. They used complementary mosfet banks to switch in the upper rails in a hard switched arrangement. And switched over to cheaper Toshiba outputs. I’ve done both types of class G,H but for the big stuff its hard switched. Fitting 20 active outputs and three pairs of commutators on a heat sink is enough of a battle - I don’t need to go doubling or tripling the package count. I use the QSC type commutators, but on more conventional EF3 based amplifier circuits, which can share a common power supply.
You asked for a quasi that doesn't oscillate for your stock of 2N3773s so I gave you one, but nevermind.
These are good points, but I am not sure there is any benefit to a modern-day quasi.
The 3-EF Locathi output stage is what I usually use. It costs an extra Vbe of headroom compared to a double, but the slightly smaller headroom tends to keep the output transistors a little further away from saturation against the rail with maximum signal swing. The other thing that is a bit more challenging with the triple EF is that there is an additional Vbe to contend with regarding thermal tracking. For this, I put the predrivers on a common heat spreader along with and additional Vbe multiplier transistor. This all results in a twin Vbe multiplier bias spreader.
As an alternative for a triple, one can make the pre-driver/driver a CFP. Yet another approach is to make the pre-driver/driver a diamond buffer, but this requires the addition of a pair of current sources to supply current to the drivers.
Cheers,
Bob
So-called sustained beta power transistors are nice, but doubles still do not cut it for supplying the needed amount of total output stage current gain if you are serious about high current output capability with sufficiently low loading of the VAS for low distortion.
Cheers,
Bob
Again, I'm not suggesting anyone build one, or that it is good, was just responding that he
wanted a stable one to use up his stash of 2N3773s.
For replicating old-world performance triples are often unnecessary, but for low Z loads and high supply voktsge they still are. A PL700 rebuild as a double is definitely a stretch, but a 400 isn’t - if all you’re trying to do is replicate original capabilities with modern parts (and ok with putting in flatpacks). It seemed a little silly to use a triple on the last little amp I built that used a 44V single supply, but didn’t seem too silly on the one that ran off 68V since it’s mission could make it drive 2 ohms or even less. If you want your PL to drive 4 ohms at war volume, you use a triple… with modern parts.
For state of the art performance you don’t use doubles of any kind, or QC’s of any kind. That separate thermal tracking is something I’ve been doing for a long time, but not necessarily for the same reasons. Any time an amp that uses separate heat sinks for the NPN and PNP banks is run at war volume they need to be tracked separately. Ask me how I know this…. My solution was a string of four diode-connected BD139’s - one mounted to an NPN output, another to the bank’s driver - and the same on the PNP side. Leach style, but with the tracking all split up. The multiplier itself gets mounted with the predrivers, which tend to run isothermal under normal use, and only heat up at all at war volume. But no more runaway, even on supplies up to +/-140 volts. In the application, I can also live with slightly overcompensated bias - that comes down some at war volume.
Yeah, I talk a lot about using big amps at war volume, which is not the typical end use for most people who frequent this site. And if you go over to the PA section, you will find that most users who are thinking they can save any money on a commercial system are instructed not to try to build anything themselves. But it’s a personal challenge of mine to build stuff with this capability (or even more than what one could afford at retail) that goes back to where this all started for me. The only figure of merit that mattered was the decibels per hertz-dollar that one could generate on the dance floor of a rave. Back then, the Phase Linears or something similar built out of surplus parts bubbled to the top of the list, as did all home brew speakers. Modern stuff improved, and I had more money to tackle my PA needs with a credit card, but I saw equipment evolving away from what I deemed “necessary”, in terms of decibels per hertz-dollar. Speakers got inefficient, amps wouldn’t tolerate the blasting you could do with an old BGW or CS800. So I’m off building horn loaders you need a forklift to move, and amps as big as the old Crests that can do full power 15 seconds at a time, 1/3 on, 2/3 off, bridged into 4 ohms - indefinitely. Don’t ask an $8000 14000 watt 2U amp to do that - it will get pissy and shut down on you.
For more sane uses at home, I’ve found that just using best practices, one can build an amp that will more than do the job and have no audible faults can be built out of spare parts from the dB/$Hz quest. I also like to play with optimizing vintage topologies, getting better sound than you ever could in the 70’s, out of very modest stuff. The dB/$Hz quest often sits on the shelf for years….
Yeah, it would be nice to put all those 2N3773s to work in a unit that puts out 700 watts per channel. Not the end of the world if I can’t - but the search will at least continue in the background.
For state of the art performance you don’t use doubles of any kind, or QC’s of any kind. That separate thermal tracking is something I’ve been doing for a long time, but not necessarily for the same reasons. Any time an amp that uses separate heat sinks for the NPN and PNP banks is run at war volume they need to be tracked separately. Ask me how I know this…. My solution was a string of four diode-connected BD139’s - one mounted to an NPN output, another to the bank’s driver - and the same on the PNP side. Leach style, but with the tracking all split up. The multiplier itself gets mounted with the predrivers, which tend to run isothermal under normal use, and only heat up at all at war volume. But no more runaway, even on supplies up to +/-140 volts. In the application, I can also live with slightly overcompensated bias - that comes down some at war volume.
Yeah, I talk a lot about using big amps at war volume, which is not the typical end use for most people who frequent this site. And if you go over to the PA section, you will find that most users who are thinking they can save any money on a commercial system are instructed not to try to build anything themselves. But it’s a personal challenge of mine to build stuff with this capability (or even more than what one could afford at retail) that goes back to where this all started for me. The only figure of merit that mattered was the decibels per hertz-dollar that one could generate on the dance floor of a rave. Back then, the Phase Linears or something similar built out of surplus parts bubbled to the top of the list, as did all home brew speakers. Modern stuff improved, and I had more money to tackle my PA needs with a credit card, but I saw equipment evolving away from what I deemed “necessary”, in terms of decibels per hertz-dollar. Speakers got inefficient, amps wouldn’t tolerate the blasting you could do with an old BGW or CS800. So I’m off building horn loaders you need a forklift to move, and amps as big as the old Crests that can do full power 15 seconds at a time, 1/3 on, 2/3 off, bridged into 4 ohms - indefinitely. Don’t ask an $8000 14000 watt 2U amp to do that - it will get pissy and shut down on you.
For more sane uses at home, I’ve found that just using best practices, one can build an amp that will more than do the job and have no audible faults can be built out of spare parts from the dB/$Hz quest. I also like to play with optimizing vintage topologies, getting better sound than you ever could in the 70’s, out of very modest stuff. The dB/$Hz quest often sits on the shelf for years….
Yeah, it would be nice to put all those 2N3773s to work in a unit that puts out 700 watts per channel. Not the end of the world if I can’t - but the search will at least continue in the background.
Here's a Class H by our own astx, BJT and FET which was something I was thinking about for the PL700:
I believe that he still considers it a prototype: Food for thought.
https://www.diyaudio.com/community/...mp-200w8r-400w4r.235194/page-156#post-7428118
I believe that he still considers it a prototype: Food for thought.
https://www.diyaudio.com/community/...mp-200w8r-400w4r.235194/page-156#post-7428118
It’s a flatpack based double EF, with mosfet uppers. His voltage makes it more of a 400 than a 700, but you could go to 100V rails. And you’d probably want EF3 that high. There are a lot of different circuits that could be used to reliably get PL700 output levels. Here’s one in progress, interrupted by the recent home build along with everything else. Flatpack based, class H (hard switched), designed to shoehorn into some chassis (es, I have 6 of them) that I snatched up for $15 each. +/-96V with dual (again, cheap surplus) toroids and capable of 360W/8R, and over 500 into 4, and won’t burst into flames at 2 ohms. Full suite of protection circuitry, all on one board. Prototype built years ago ran beautifully spread out all over the bench. Every bit as capable as a PL700, but unfortunately won’t look or feel anything like one. Four of them in a rack will make some serious noise (and these are for top cabs, bigger things envisioned for kick bins and subs).
The trick is gettning something into the PL700 box and still have it be as reliable and trouble free.
The trick is gettning something into the PL700 box and still have it be as reliable and trouble free.
Attachments
Even the Phase Linear 400 uses triples. Although operation at high current is a good reason to use triples, using triples is not just for that reason. The nonlinear load impedance that the VAS sees driving a double causes distortion. This can occur even at lower current swings. Consider a double with driver and output current gains of 50 driving a 4-ohm load. The net current gain of 2500 only increases the nominal load impedance seen by the VAS to 10,000 ohms, and that 10,000 ohms can be nonlinear, especially when going through the crossover region from the impedance seen through the N outputs versus that of the P outputs. The potentially very significant net beta mismatch between the top and bottom will cause distortion. With a triple, that mismatch matters little. The cost of the extra pair of predriver transistors is very low.
Having said that, the PL400 is not a very good design. It has an undegenerated input stage and a one-transistor VAS with a bootstrapped load. Ditto for the PL700.
Cheers,
Bob
The old PL400 isn’t hard to out-do with a modern double. I’m not saying I’d do it, but it’s possible. You’re right - it’s not that good of a design. My first PL-ish homebrew used EF3 with 80 MHz C2565/A1095 drivers and D424/B554 outputs but it used lower rails (58V) and a current source loaded VAS. And the op amp stage from the S2. Sounded really nice, and could be tortured bridged at 4 ohms. I built a stripped down version for home use with EF2, same voltage, only 400 VA transformer, half as many power transistors. But I boosted the VAS from 6 mA to 25. Still sounded better than my stock 400. It just ran a 10” 8 ohm 3-way at dorm-room volume. There were another 4 iterations over the years, ultimately becoming more like the BGW than the PL.
Maybe I am missing something or there are different semantics at work here. This does not look like a class H design. There is only one rail.
The MOSFETs hold the Vce of the inner BJTs at an approximate constant voltage on the order of 32 V. This is not a bad design, but I just don't think it would be called class H. Having said that, it is good that the MOSFETs share the total dissipation with the BJTs. Indeed, by holding Vce of the BJTs roughly constant, the SOA of the BJTs is preserved, while the more SOA-robust MOSFETs bear the brunt of the SOA burden. The MOSFETs act like source followers feeding the collectors of the BJTs, and the MOSFETs are fed with a level-shifted version of the output signal.
Notice also that this design does not save power dissipation as does a class H design. Total output stage power dissipation of this design is essentially the same as that of a class AB amplifier. I could be mistaken, but this amplifier seems to share some of the concept of the Nelson Pass Stasis approach.
Cheers,
Bob
I have a working PL400 that I am refurbishing/redesigning. It will get a completely new IPS/VAS, but I am tempted to keep the Quasi output stage because TO-3 complementary power transistors seem really hard to find, and adapting the physical design to use TO-264 plastic devices seems difficult. The PL400 was biased almost into full class B, with only a few mA of idle bias. In the working version, upping the bias to about 15 mV across the 0.22-ohm emitter resistors significantly reduced the distortion, as expected. I think I can revise the quasi to act more like a PNP than it currently does. We'll see. The design also needs better bias tracking, so I'll use a twin Vbe spreader in the revised design. I'll also add a loudspeaker protection circuit which, God knows, every PL should have had! I'll also add an output L-R network.
Any thoughts on how to fit it with plastic output transistors will be appreciated. I have a PL700 that I will use vertical power MOSFET output devices on, so I will have to confront that issue on that design. I have thought of one way to do it, but it is not pretty and gives up some thermal conductivity.
Cheers,
Bob
MJ15024/5 or 21193/4? Not that hard to find these days unless they’re out of stock at the moment. Adapting to flatpacks IS the problem with updating the design - how the hell do you mount them? Flatpacks are the only option for sustained beta high fT outputs. But all 15024/5 driver and output does work pretty well. Unfortunately, those 80 MHz drivers wouldn’t be possible. One could use four outputs and shoehorn MJE15032/3 drivers in there somewhere, perhaps on an internal heat sink or on some blank space between outputs. And use that separate thermal tracking idea.
The original 66546/PL909 combo didn’t like to be biased up to anything close to optimum. Oscillations appeared when output hits around 4Vp-p. Back it off to zero or crank it till it’s hotter than the blazes of hell and damnation and it went away. The one I rebuilt with all 2SD424’s and the others that went full comp were stable at any bias.
The original 66546/PL909 combo didn’t like to be biased up to anything close to optimum. Oscillations appeared when output hits around 4Vp-p. Back it off to zero or crank it till it’s hotter than the blazes of hell and damnation and it went away. The one I rebuilt with all 2SD424’s and the others that went full comp were stable at any bias.
I'll look into the availability of those power transistors in the TO-3 package for the PL400 upgrade after I see what I can get out of the existing output stage combo. Now that my new book is in print I have some time to putter around with various projects that pose technical challenges.
I did not see any instability when I increased the bias in the existing PL400 output stage, but I am not surprised to hear if there were oscillations in some cases, since the high-frequency rail bypassing is not great and there is no output coil. Of course, the Quasi design can be improved to be more stable.
For the PL700 vertical MOSFET design I am considering, I will put a 6-3/4 X 1-1/2 X 1/4 inch aluminum heat spreader mounted behind each of the four heat sinks on the inside of the chassis to mount the plastic devices. The heat spreader will be bolted to the inside of the steel chassis with a heat sink compound interface. The finned heat sinks will also be mounted to the steel chassis with heat sink compound on the other side. Heat will thus have to pass through two such mountings to the heat sinks. I am hopeful that the large area of the interfaces will provide an adequate thermal conduction path.
With the heat sinks at a maximum human-safe temperature of 55C, I am hopeful that the maximum internal heat spreader temperature will not exceed 70C, under worst-case conditions, allowing for a 15C thermal loss through the Aluminum-Steel-Aluminum interface. These are just guesses off the top of my head. Each aluminum heat spreader will form an output stage module with 2 or 3 N-channel/P-channel pairs. I'll see how it goes. I'll first do thermal tests with one of the modules mounted in the unit with temperature monitoring on the heat spreader and on the heat sink with a known amount of power dissipation on the heat spreader created by one or more power resistors mounted on it.
Cheers,
Bob
I did not see any instability when I increased the bias in the existing PL400 output stage, but I am not surprised to hear if there were oscillations in some cases, since the high-frequency rail bypassing is not great and there is no output coil. Of course, the Quasi design can be improved to be more stable.
For the PL700 vertical MOSFET design I am considering, I will put a 6-3/4 X 1-1/2 X 1/4 inch aluminum heat spreader mounted behind each of the four heat sinks on the inside of the chassis to mount the plastic devices. The heat spreader will be bolted to the inside of the steel chassis with a heat sink compound interface. The finned heat sinks will also be mounted to the steel chassis with heat sink compound on the other side. Heat will thus have to pass through two such mountings to the heat sinks. I am hopeful that the large area of the interfaces will provide an adequate thermal conduction path.
With the heat sinks at a maximum human-safe temperature of 55C, I am hopeful that the maximum internal heat spreader temperature will not exceed 70C, under worst-case conditions, allowing for a 15C thermal loss through the Aluminum-Steel-Aluminum interface. These are just guesses off the top of my head. Each aluminum heat spreader will form an output stage module with 2 or 3 N-channel/P-channel pairs. I'll see how it goes. I'll first do thermal tests with one of the modules mounted in the unit with temperature monitoring on the heat spreader and on the heat sink with a known amount of power dissipation on the heat spreader created by one or more power resistors mounted on it.
Cheers,
Bob