BC said:
It's not off topic. To the extent that I stuck a preamp in the thread, then it's reasonable to talk about it.
If you're wanting a bipolar buffer somewhat similar in flavor to this, look on Walt Jung's website. The caveat being that there are four (maybe more) transistors per buffer rather than two. If I recall correctly, John is listed in the credits as having contributed to that design.
I prefer doing it this way because it's simpler and more elegant. JFETs also offer a much higher input impedance than bipolars. It's pretty much a win-win situation from my point of view.
EUVL said:
I gave up on simulation back around the time I posted the Aleph-X. These days I regard simulation as a parallel universe; one having entertainment value but not much intersection with the reality we live in. I prefer to solder real parts and know, rather than simulate and think I know.
I have no earthly idea where you might look to get models for parts. If Toshiba (or OnSemi, or Fairchild, for that matter) don't have them then I don't have any other suggestions. It seems that people frequently criticize the accuracy of published models, anyway.
There's a silver lining to my time-constraint cloud in that by the time I actually turn on my soldering iron I've pretty much thought things through and made most of the big decisions. All that remains is to tidy up a few details. This circuit only went through two iterations to get it to this point, and the only real problem with the first one was parts availability for the Siliconix JFETs. The circuit worked properly the first time I turned it on. The only thing I needed to do was cut back on the gain and I already had a whole bag-full of ideas I wanted to try--the trick being to do so without using negative feedback. While I was at it, I went ahead and made the Ver. 2 circuit fully balanced just to save time later on, because I already knew I was going to build a balanced amp for the midranges after assembling the single-ended version you see here.
If you look closely at the pictures, you'll see two wires sticking up from the left-hand side of the front end board with a jumper between them. That's where I tied off the unused drivers. The only thing I need to do to get a fully balanced 100W amp is pull those wires and tag in a bias circuit and output bank. Presto! Instant 100W version. I've already got the parts--just needed heatsinking. I was on the verge of going water-cooled for the 100W when MikeW sent me some huge heatsinks. I hope to get some time this weekend to etch circuit boards and get the output situation squared away.
For what it's worth, the balanced/bridged version should have slightly less distortion due to cancellation of second harmonic. That may be outweighed by other factors, so I'm not counting on a big reduction. It's not a primary concern for me anyway, though I understand that others feel otherwise.
Grey
Somehow I get more and more the feeling that mosfets have really some very nice properties that make them particularly suitable for simple amps.
Especially their relation between bandwidth and parasitic capacities seem to be in a very sweet spot, so that there's often no frequency compensation needed. That's also really nice about the GR-25, a nice performer without the necessity for FC-tuning.
Ever seen a bipolar amp without frequency compensation?
(I thought I would do something to get the thread back to the front, it really deserves more attention IMHO).
All the best, Hannes
Especially their relation between bandwidth and parasitic capacities seem to be in a very sweet spot, so that there's often no frequency compensation needed. That's also really nice about the GR-25, a nice performer without the necessity for FC-tuning.
Ever seen a bipolar amp without frequency compensation?
(I thought I would do something to get the thread back to the front, it really deserves more attention IMHO).
All the best, Hannes
Sure, ie the renowned Hiraga 20W and 30W models. Those are current output (common emitter), current feedback class-A opamps (thus very much GFB) in the end and don't need explicit compensation by design.h_a said:
The output stage of the GR-25, as a traditional source follower design, is prone to overshoot with capacitive loading like any other source/emitter follower. With perfect layout, bypassing and care with the gate stoppers one can normally make sure that this overshoot doesn't break out into sustained oscillation.
- Klaus
h_a said:
Each kind of part has its own weaknesses. MOSFETs, particularly vertical MOSFETs, have relatively high capacitance and, worse still, that capacitance is non-linear. To a large degree you can ameliorate the problem by driving it with lots of current coming from a low impedance. In the GR-25, I picked up a 2 x 4" and 'whacked the problem upside the head' by using 100mA bias in the driver stage and choosing 500 Ohm resistors to ground reference the front end and lower the Zout.
There's nothing keeping you from using lateral MOSFETs in the output. They're not as robust in terms of current (vertical MOSFETs excel at this), so you'll need more of them in parallel, but their capacitance is more linear. The downside being that the addition of more lateral MOSFETs adds to the cumulative Gate capacitance the driver stage sees, so the reduction in capacitance per device is offset by the addition of more devices. It will be a balancing act, but there's absolutely no reason why it shouldn't work.
Bipolars aren't as bad in terms of capacitance, but they consume lots of base current. The heavy drive stage bias should accommodate reasonable bipolar drive requirements easily, but as with lateral MOSFETs you'll probably need more devices to share the load. Bipolars also have trouble interfacing with the real world without a Zobel network at the output. Charles Hansen says he has methods of handling this sort of thing without resorting to stuff at the output, but he regards the information as proprietary. If it works as I suspect, it will be very much circuit dependent, so it would not be a drop-in solution. More of a cut-and-try sort of thing that would depend on PCB layout and a hundred other variables that don't resolve easily at the planning stage.
KSTR said:
If you have a preferred signal and "load from hell" that you would like me to try, I'd be happy to torture the amp for a while. I've got a channel on my bench at the moment (I'm playing with bias circuit variations), so it won't be a big problem.
Grey
Hhm, I would think you have already tried the usual square wave test into varying capacitve loads (in 1-3-10 steps, from like 1nF to 3uF or so. This stepping is important, as the problem zone with complex conjugate poles in the transfer function is only within a specific range of cap loads and also depends much on the driver properties), in parallel with a ohmic load. With feedback designs in particular (not yours, then) it is useful to have the small signal squarewave with its quick slopes riding on a LF sine wave so that the output stage also has varying large signal (and we know that oscillation often happens close to the rails when the capacitances increase rapidly). Another problem can arise with paralleled MOSFETs, in switching applications they often show parasitic oscillation in the transitions, closer to gigahetz than megahertz frequencies-- quite certainly not a problem here, with big enough gate stoppers.GRollins said:
SOA / thermal stability wise an "industry test" I sometimes use is feeding the DUT with a pink noise signal, while the output is loaded with a current source also fed by a (different) pink noise signal. The current source can be replaced by a bigger amp (wider voltage swing) and a resistor. This puts true 4-quadrant stress on the output stage. One can also use sines of different frequencies. This is surely a heavier load condition than will ever appear with real speakers.
I don't expect much to happen with your output stage, though, just because the absence of feedback. And if there is oscillation it's a parasitic effect (inductive drain leads especially, and inductive drive). Textbooks (and SPICE, for that matter) tell that a loaded follower will overshoot but not self-oscillate if the rest of circuit is perfect -- which it never is....
- Klaus
PMA,
Feel free to post your results. You know I'm not a big proponent of simulation, but I'm sure others will find it interesting.
The circuit is so obvious that I can't believe no one has posted a list of other threads discussing the same thing. Once you have the concept of a self-biasing JFET complimentary follower in hand, surely it would be natural to put one after a difficult source such as a "passive" preamp or in front of a difficult load. To put it in both places is the next step.
Being a follower, it has local degeneration and correspondingly low distortion and Zout. The exact values will change somewhat depending on what resistor values you choose, but it should still perform well, regardless.
Using 2SK170/2SJ74s will give even better performance at the expense of some voltage swing. The obvious solution to that is to cascode the followers. Depending on how you arrange the cascodes, you can achieve some pretty remarkable things. Given that there's no (voltage) amplification required of the circuit as a line stage or crossover, I used the 2SK246/2SJ103 because they allowed for plenty of voltage swing on their own and left the cascode versions on the back burner.
If only Lovoltech had made a P-ch power JFET we could really have something to talk about...
Grey
Feel free to post your results. You know I'm not a big proponent of simulation, but I'm sure others will find it interesting.
The circuit is so obvious that I can't believe no one has posted a list of other threads discussing the same thing. Once you have the concept of a self-biasing JFET complimentary follower in hand, surely it would be natural to put one after a difficult source such as a "passive" preamp or in front of a difficult load. To put it in both places is the next step.
Being a follower, it has local degeneration and correspondingly low distortion and Zout. The exact values will change somewhat depending on what resistor values you choose, but it should still perform well, regardless.
Using 2SK170/2SJ74s will give even better performance at the expense of some voltage swing. The obvious solution to that is to cascode the followers. Depending on how you arrange the cascodes, you can achieve some pretty remarkable things. Given that there's no (voltage) amplification required of the circuit as a line stage or crossover, I used the 2SK246/2SJ103 because they allowed for plenty of voltage swing on their own and left the cascode versions on the back burner.
If only Lovoltech had made a P-ch power JFET we could really have something to talk about...
Grey
With some trivial modifications, the front end of the GR-25 will make an excellent transconductance output preamp. If you prefer a more conventional follower output, strap on a Line Drive as an output and you'll have a low Zout preamp. Yes, the folded cascode/driver is a little overdone for that application, so replace the MOSFETs with 2SK246/2SJ103 JFETs and you've got a lot better preamp than most, with or without feedback.
Grey
Grey
Just as a footnote, I'm actually using 10 Ohm pots for DC offset adjustment, but that's because I happened to get lucky matching N-ch and P-ch parts against each other. I used 50 Ohms in the schematic to leave a wider adjustment window for anyone who might want to build the circuit.
I'm tempted to build one with the 2SK170/2SJ74, but I doubt I'd be able to hear any difference compared to the 2SK246/2SJ103 because there's no gain to boost the noise, and the local degeneration probably swamps any other differences. Then again, I might just get a wild hair some rainy afternoon and do it anyway.
The idea of turning the volume knob further clockwise than you're accustomed to with a preamp that has gain can be unnerving at first. To put things in perspective, put a test signal of, say, .5V into your normal preamp, then hook a meter to the output and turn the volume knob until it says you're getting .5V out. By definition, that's 0dB gain. On my Premier Three, that happens at about the ten o'clock position. On a circuit like this, it's when the knob is cranked fully CW. As long as you have sufficient gain to drive your amplifier to a reasonable listening volume, there's absolutely no reason to add more distortion and noise than necessary. If it should turn out that more gain is needed for just one input, then put a gain stage only on that one input, ahead of the volume knob. That way your other inputs won't have to go through unnecessary circuitry and whatever noise the gain circuitry adds will be turned down by the volume pot when you're listening at low volumes. In a regular preamp the pot is ahead of the line amplifier and that noise and distortion is there all the time, on every signal, whether you want it or not. This way, you can have your cake and eat it too.
PMA,
Thanks for posting the analysis.
Grey
I'm tempted to build one with the 2SK170/2SJ74, but I doubt I'd be able to hear any difference compared to the 2SK246/2SJ103 because there's no gain to boost the noise, and the local degeneration probably swamps any other differences. Then again, I might just get a wild hair some rainy afternoon and do it anyway.
The idea of turning the volume knob further clockwise than you're accustomed to with a preamp that has gain can be unnerving at first. To put things in perspective, put a test signal of, say, .5V into your normal preamp, then hook a meter to the output and turn the volume knob until it says you're getting .5V out. By definition, that's 0dB gain. On my Premier Three, that happens at about the ten o'clock position. On a circuit like this, it's when the knob is cranked fully CW. As long as you have sufficient gain to drive your amplifier to a reasonable listening volume, there's absolutely no reason to add more distortion and noise than necessary. If it should turn out that more gain is needed for just one input, then put a gain stage only on that one input, ahead of the volume knob. That way your other inputs won't have to go through unnecessary circuitry and whatever noise the gain circuitry adds will be turned down by the volume pot when you're listening at low volumes. In a regular preamp the pot is ahead of the line amplifier and that noise and distortion is there all the time, on every signal, whether you want it or not. This way, you can have your cake and eat it too.
PMA,
Thanks for posting the analysis.
Grey
Having come to solid state from fiddling with tubes, I still find it vaguely unsettling that some solid state circuits have input impedances at 10k and below. To me, 75-100k is about the minimum that "feels" right. It doesn't help that I know solid state is quite content with such things--I just have this persistent gut feeling that something's awry.
FOOTNOTE: Anyone who is the opposite of me and feels that anything above 47k is an affront to the natural laws of the universe is welcome to put lower value resistors at the inputs of either the GR-25 or the Line Drive. They're quite insensitive to that sort of thing. END FOOTNOTE
I'm not sure that both front and back buffers are required in all cases. If someone is using a good quality 100k pot, then as far as the source is concerned, it's a nice, tidy 100k resistive load. That's not always the case, of course. If you've got a volume control that presents a varying 5k load to the source, then I'm all for buffering the silly thing. At least conceptually, I thought it would be better if I went ahead and showed fore and aft buffers and told people they could do without, than to show one and tell them to add another if they felt they needed it.
When I have some spare time--probably on the second Tuesday of next week--I'll try some 1M resistors at the input. As long as it doesn't interact with source or cable inductance and/or capacitance in some ugly way, I may decide to leave it.
Given the simplicity of the circuit, I'll probably run up several sets of them and sit down with a half-dozen volume controls to give them a listen. Given that the actual pots or circuits (relays, optical, what have you) will be buffered (another good time to use both buffers), I'll know that any differences I hear stem from the controls themselves. If I don't hear any difference, I'm free to use the cheapest one.
The first thing on my to-do list is to see if I can borrow a cosmetic second Shallco from John--just one that's sitting around gathering dust in a box, you understand...
Grey
FOOTNOTE: Anyone who is the opposite of me and feels that anything above 47k is an affront to the natural laws of the universe is welcome to put lower value resistors at the inputs of either the GR-25 or the Line Drive. They're quite insensitive to that sort of thing. END FOOTNOTE
I'm not sure that both front and back buffers are required in all cases. If someone is using a good quality 100k pot, then as far as the source is concerned, it's a nice, tidy 100k resistive load. That's not always the case, of course. If you've got a volume control that presents a varying 5k load to the source, then I'm all for buffering the silly thing. At least conceptually, I thought it would be better if I went ahead and showed fore and aft buffers and told people they could do without, than to show one and tell them to add another if they felt they needed it.
When I have some spare time--probably on the second Tuesday of next week--I'll try some 1M resistors at the input. As long as it doesn't interact with source or cable inductance and/or capacitance in some ugly way, I may decide to leave it.
Given the simplicity of the circuit, I'll probably run up several sets of them and sit down with a half-dozen volume controls to give them a listen. Given that the actual pots or circuits (relays, optical, what have you) will be buffered (another good time to use both buffers), I'll know that any differences I hear stem from the controls themselves. If I don't hear any difference, I'm free to use the cheapest one.
The first thing on my to-do list is to see if I can borrow a cosmetic second Shallco from John--just one that's sitting around gathering dust in a box, you understand...
Grey
Grey-
That is a beautiful circuit! Being someone who appreciates no feedback designs -- it find it particularly impressive.
With any luck, I may get to build it.
If you do get boards made to sell, I would be interested. (If not, I have learned its not that hard to etch your own.)
Thanks for posting it.
JJ
That is a beautiful circuit! Being someone who appreciates no feedback designs -- it find it particularly impressive.
With any luck, I may get to build it.
If you do get boards made to sell, I would be interested. (If not, I have learned its not that hard to etch your own.)
Thanks for posting it.
JJ
Offering boards is something that would depend on how much interest people show. I made my own and you're certainly welcome to do so.
I haven't bothered to post regulator circuits because it was all quite specific to my circumstances. I always regulate low level circuits, which for present purposes includes the front end of this amplifier, but this time I decided to regulate the output rails down to 25V from a raw voltage of around 30V or thereabouts. I did so mostly to limit heat dissipation because I knew I was going to be biasing the output pretty hard. Naturally, it's gratifying to see an output rail with no ripple and only a few millivolts of residual noise. This happens to help the distortion specs to the extent that rail nasties are not reflected in the output. I will probably regulate the 100W version's rails as well because I'm planning on using the same type of transformers, hence same rail voltage, and the same bias. Whether I regulate output rails for any further versions of the circuit will depend on what transformers/rails I might come up with.
If you use less bias or have a transformer that happens to give you exactly the rail voltage you want, feel free to dispense with the output regulation, but don't be surprised if the distortion rises a bit as a result. You could adopt a middle path and employ a capacitor multiplier. That would give you substantially quieter rails at the expense of not locking in the voltage. If you go for an actual regulator you could use something simple like a MOSFET with a Zener-referenced Gate. I used a somewhat more complex circuit, but stopped short of protection circuitry and sundry other bells and whistles. I did put in a slow-start function because I've got a lot of capacitance to charge. I just pulled a number out of thin air and went for about a two second ramp-up. It takes the sting out of turn-on.
Naturally, you could use some combination of resistors or inductors with your cap bank if you want something really simple.
Grey
I haven't bothered to post regulator circuits because it was all quite specific to my circumstances. I always regulate low level circuits, which for present purposes includes the front end of this amplifier, but this time I decided to regulate the output rails down to 25V from a raw voltage of around 30V or thereabouts. I did so mostly to limit heat dissipation because I knew I was going to be biasing the output pretty hard. Naturally, it's gratifying to see an output rail with no ripple and only a few millivolts of residual noise. This happens to help the distortion specs to the extent that rail nasties are not reflected in the output. I will probably regulate the 100W version's rails as well because I'm planning on using the same type of transformers, hence same rail voltage, and the same bias. Whether I regulate output rails for any further versions of the circuit will depend on what transformers/rails I might come up with.
If you use less bias or have a transformer that happens to give you exactly the rail voltage you want, feel free to dispense with the output regulation, but don't be surprised if the distortion rises a bit as a result. You could adopt a middle path and employ a capacitor multiplier. That would give you substantially quieter rails at the expense of not locking in the voltage. If you go for an actual regulator you could use something simple like a MOSFET with a Zener-referenced Gate. I used a somewhat more complex circuit, but stopped short of protection circuitry and sundry other bells and whistles. I did put in a slow-start function because I've got a lot of capacitance to charge. I just pulled a number out of thin air and went for about a two second ramp-up. It takes the sting out of turn-on.
Naturally, you could use some combination of resistors or inductors with your cap bank if you want something really simple.
Grey
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.