Re: CLASS AB???????
Incidentally, ALL current-feedback amps. run in class-AB at their feedback summing node....
http://www.anasoft.co.uk/EE/currentfeedbackmyth/currentfeedbackmyth.html
Fred Dieckmann said:
Incidentally, ALL current-feedback amps. run in class-AB at their feedback summing node....
http://www.anasoft.co.uk/EE/currentfeedbackmyth/currentfeedbackmyth.html
class A vs. class AB
Not to be too nitpicky, but strictly speaking, to say that some active devices run in class AB is to say one of the devices switches off somewhere in the cycle, which isn't true of current feedkback amps as you (or the article you linked) say. Just because the current may change greatly relative to the quiescent point does not mean it is class AB. I will admit that with the latest in auto-biasing, non-switching and error correcting circuit techniques, the definition of class-A and even class-AB has been blurred a bit, but calling something class AB that doesn't actually stop conducting at some point in the cycle does not follow traditional nomenclature [where class-A is conduction over the whole cycle, class-B is conduction over half the cycle, and class-AB (not really a class to be honest) is somewhere in between] and is greatly misleading. Call it non-constant current class-A or irregular class-A or SOMETHING, just not class-AB.
Not to be too nitpicky, but strictly speaking, to say that some active devices run in class AB is to say one of the devices switches off somewhere in the cycle, which isn't true of current feedkback amps as you (or the article you linked) say. Just because the current may change greatly relative to the quiescent point does not mean it is class AB. I will admit that with the latest in auto-biasing, non-switching and error correcting circuit techniques, the definition of class-A and even class-AB has been blurred a bit, but calling something class AB that doesn't actually stop conducting at some point in the cycle does not follow traditional nomenclature [where class-A is conduction over the whole cycle, class-B is conduction over half the cycle, and class-AB (not really a class to be honest) is somewhere in between] and is greatly misleading. Call it non-constant current class-A or irregular class-A or SOMETHING, just not class-AB.
Unstable VAS current
Regarding the unstable VAS current problem of complementary diff. amp inputs with current mirrors, I believe the problem is even more insidous than suggested so far. For perfectly matched transistors in the diff. amps. the VAS current is undefined as mentioned. But for real world transistors with unmatched offsets, several cases result. If both diff. amps. have opposite offsets, they can both request high current from the VAS or both can request zero current from the VAS, depends on which way the offsets are. If both offsets are in the same direction, then the feedback can adjust the output offset to compensate, but the magnitude of the two diff. amp. offsets are unlikely to be identical, leading back to the first case with large or zero VAS current requested. This is all because of the very large gains available here. So the VAS stage always operates in meltdown mode or class C (no gain until input signal exceeds residual offset).
I notice that Slone provides current limiting in the VAS and a resistor to set a minimum current, bandaids, not elegant. Possible class C operation is not fixed.
The diff. amp. degeneration resistors need a trim pot in one side of each diff. amp. to null out the offsets. Then a resistor from upper VAS base to lower VAS base can be used to set bias current in the VAS. Still, with such high gain, this may not be stable with temperature.
The current source in the VAS collector idea won't work (I don't think) because it will either just have zero volts across it or near rail to rail voltage across it. The opto-isolator or other type servos may fix the VAS bias stably but won't fix the unbalanced currents in the diff. amps. so still need the trimmers.
On a somewhat related note, I often hear it mentioned (for a single diff. amp. design with mirror) that the mirror maintains balance of current in the diff. amp. But I don't see any provision ever provided for supplying the bias current to the VAS base. So in reality the diff. amp. will unbalance sufficiently to provide this. Am I missing something here or is this considered so small as to be negligable.
Finally, on a somewhat off topic question, what analog design book mentions the Baxandall super pair? I've looked thru a bunch of books with no luck. Thanks. Don
Regarding the unstable VAS current problem of complementary diff. amp inputs with current mirrors, I believe the problem is even more insidous than suggested so far. For perfectly matched transistors in the diff. amps. the VAS current is undefined as mentioned. But for real world transistors with unmatched offsets, several cases result. If both diff. amps. have opposite offsets, they can both request high current from the VAS or both can request zero current from the VAS, depends on which way the offsets are. If both offsets are in the same direction, then the feedback can adjust the output offset to compensate, but the magnitude of the two diff. amp. offsets are unlikely to be identical, leading back to the first case with large or zero VAS current requested. This is all because of the very large gains available here. So the VAS stage always operates in meltdown mode or class C (no gain until input signal exceeds residual offset).
I notice that Slone provides current limiting in the VAS and a resistor to set a minimum current, bandaids, not elegant. Possible class C operation is not fixed.
The diff. amp. degeneration resistors need a trim pot in one side of each diff. amp. to null out the offsets. Then a resistor from upper VAS base to lower VAS base can be used to set bias current in the VAS. Still, with such high gain, this may not be stable with temperature.
The current source in the VAS collector idea won't work (I don't think) because it will either just have zero volts across it or near rail to rail voltage across it. The opto-isolator or other type servos may fix the VAS bias stably but won't fix the unbalanced currents in the diff. amps. so still need the trimmers.
On a somewhat related note, I often hear it mentioned (for a single diff. amp. design with mirror) that the mirror maintains balance of current in the diff. amp. But I don't see any provision ever provided for supplying the bias current to the VAS base. So in reality the diff. amp. will unbalance sufficiently to provide this. Am I missing something here or is this considered so small as to be negligable.
Finally, on a somewhat off topic question, what analog design book mentions the Baxandall super pair? I've looked thru a bunch of books with no luck. Thanks. Don
Re: Unstable VAS current
I have a PDF file of Hawksford's AES article where he introduces the Hawksford cascode and mentions the Baxandall super pair (not by that name though). I don't know of any book that discusses it. I can email you the paper if you'd like. I found the Hawksford cascode to have lower distortion than the super pair when used with MJE340/350 drivers. There was also some discussion of configurations that look like a mix of the Hawksford cascode and the super pair (near the end of the "Amplifier Topology Subjective Effects" thread), but the only way I could prevent the thing from showing oscillations in LTSpice was using Gear integration instead of trapezoid. This leads me to believe that they aren't stable, though I don't have conclusive proof of this.
smoking-amp said:
I have a PDF file of Hawksford's AES article where he introduces the Hawksford cascode and mentions the Baxandall super pair (not by that name though). I don't know of any book that discusses it. I can email you the paper if you'd like. I found the Hawksford cascode to have lower distortion than the super pair when used with MJE340/350 drivers. There was also some discussion of configurations that look like a mix of the Hawksford cascode and the super pair (near the end of the "Amplifier Topology Subjective Effects" thread), but the only way I could prevent the thing from showing oscillations in LTSpice was using Gear integration instead of trapezoid. This leads me to believe that they aren't stable, though I don't have conclusive proof of this.
pdf and linsley hood
Hi Mikek, how does the Linsley Hood design work? I looked thru a bunch of amp schematics by Hood, but don't see anything applicable, must be in a Wireless World article. I probably have it somewhere. Only thing I can think of is an Hitachi amplifier that had two VAS 's and a bridge output, but would just be the same as two whole amplifiers then. Or maybe an opto from the VAS back to a split in the common feedback connection of the two diff amps so as to cause a differential offset between them.
Hi Andy, I have a xerox copy of the article, but a pdf file would be nice to have too. My filing system consists of making MORE copies of important papers so I have a better chance of finding them!
Hi Mikek, how does the Linsley Hood design work? I looked thru a bunch of amp schematics by Hood, but don't see anything applicable, must be in a Wireless World article. I probably have it somewhere. Only thing I can think of is an Hitachi amplifier that had two VAS 's and a bridge output, but would just be the same as two whole amplifiers then. Or maybe an opto from the VAS back to a split in the common feedback connection of the two diff amps so as to cause a differential offset between them.
Hi Andy, I have a xerox copy of the article, but a pdf file would be nice to have too. My filing system consists of making MORE copies of important papers so I have a better chance of finding them!
jam said:
AAhh...from that japanese site....tried this...but found currents grossly unbalanced in diff. stages....hood's much better...and simpler...
Note that Hood's design does not balance currents in the diff. stage as effectively as the generic Self design...but it's an improvement on many so-called dual complementary diff. stage designs...
Linsley Hood design
Ah, interesting! Thankyou Mikek. R8 and R9 provide the VAS bias feedback in a convenient manner. TR1 allows adjusting out the offset difference in diff amps. (Q11 typo I think, NPN type) Should be able to tweak this into a more conventional shape to get low distortion. R8 and R9 only need to feedback DC info, so can put caps from Q9,Q12 ends of R8,R9 to respective rails. Now can put another 100 Ohm resistor in each emitter circuit (Q9,Q12) to act as conventional degeneration. Q9,Q12 can be upgraded to fancier configurations, maybe Baxandall super pair or Hawksford cascoded VAS designs. Lots of possibilities.
Ah, interesting! Thankyou Mikek. R8 and R9 provide the VAS bias feedback in a convenient manner. TR1 allows adjusting out the offset difference in diff amps. (Q11 typo I think, NPN type) Should be able to tweak this into a more conventional shape to get low distortion. R8 and R9 only need to feedback DC info, so can put caps from Q9,Q12 ends of R8,R9 to respective rails. Now can put another 100 Ohm resistor in each emitter circuit (Q9,Q12) to act as conventional degeneration. Q9,Q12 can be upgraded to fancier configurations, maybe Baxandall super pair or Hawksford cascoded VAS designs. Lots of possibilities.
RE: try this for size
Hi Jam, this schematic is interesting too. The VAS bias current is fed back through an interesting network here. Sort of like using an opto-isolator to sample the VAS current except with no opto! I think I would put a capacitor across the X and Y points to slow down the bias feedback, no need for speed in the bias feedback in either this design or the Linsley Hood design. This design needs some trimpot(s) in the input diff amps. sources to null out the offsets. In either design, I would replace the FETs with some precision matched bipolars like Analog Devices MAT series. FETs are too mismatched. (Would need to add tail current sources and degeneration resistors with these bipolar inputs to get DC input levels to work out.) Also the current gain in this design is awfully low due to the common base VAS stages.
The type of VAS bias feedback network used here could be put to use to make a design that is automatic in nulling the diff. amp offsets by connecting the common mode correction current to one side of the input diff. amp degeneration resistors (ie. at the emitters of the global feedback connected diff. transistors, degen. resistors need to be added) instead of to the current mirrors. This would correct the offsets AND control the VAS bias automatically. Could use an opto isolator instead also. Needs some reworking to get DC operating levels to work out.
Hi Jam, this schematic is interesting too. The VAS bias current is fed back through an interesting network here. Sort of like using an opto-isolator to sample the VAS current except with no opto! I think I would put a capacitor across the X and Y points to slow down the bias feedback, no need for speed in the bias feedback in either this design or the Linsley Hood design. This design needs some trimpot(s) in the input diff amps. sources to null out the offsets. In either design, I would replace the FETs with some precision matched bipolars like Analog Devices MAT series. FETs are too mismatched. (Would need to add tail current sources and degeneration resistors with these bipolar inputs to get DC input levels to work out.) Also the current gain in this design is awfully low due to the common base VAS stages.
The type of VAS bias feedback network used here could be put to use to make a design that is automatic in nulling the diff. amp offsets by connecting the common mode correction current to one side of the input diff. amp degeneration resistors (ie. at the emitters of the global feedback connected diff. transistors, degen. resistors need to be added) instead of to the current mirrors. This would correct the offsets AND control the VAS bias automatically. Could use an opto isolator instead also. Needs some reworking to get DC operating levels to work out.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.