PART II
The complementary differential is a marvel, but it’s not quite enough to build a complete amplifier on its own. The circuit has four (potential) signals coming out, but they’re limited as to current delivery and too far away from ground to make much use of them as they are. Some way to increase the available current and to bring the disparate signals back together is needed.
Fear not, there is a way. Actually, there are several ways, but the way I chose to use for the GR-25 is the one that John used in the JC-3. After all, what good’s using the topology if you’re not going to use all of it? (Don’t argue…we’ll get to variations sooner or later…[Probably later, the way my life runs.])
The circuit John used is called…it’s called a…er…well, honestly, I don’t know what to call it. I’ve tried to get John to give the thing a name of some sort, but he’s not offered one to date. Like the JFETs in the differential, it does double duty. It’s both Common Source and Common Gate at the same time. Incidentally, note that I’ve used MOSFETs in these positions. The nomenclature for a MOSFET is the same as for a JFET, so we’ll still be talking in terms of Gates, Sources, and Drains.
For the time being let’s look at Q9 and ignore everything else. Q9’s Gate derives its signal from Q6 and R2. That’s not particularly outlandish taken on its own. All things being equal (which they aren’t, but we’ll come back and fill in the blanks shortly), Q9 would take that signal, amplify it, and toss it out into the world via its Drain. Q9’s Drain sees Q11’s Drain as a load, and since the impedance at a MOSFET’s Drain is very, very high indeed, the potential gain is absolutely enormous. 60dB or more, all at once. Whew! But before we deal with that, we need to address the signal at Q9’s Source. That signal comes from the current flowing through R9 from Q5. This is a curious state of affairs…driving a transistor through two of its orifices at once! (I hear that’s illegal in the entire state of Mississippi and in certain counties in Tennessee. South Carolina doesn’t admit that it can be done at all.) Nonetheless, that’s what’s happening here. The interesting part is that since the differential offers signals of both polarities it’s possible to drive Q9 both ways in an additive sense. Remember that when we’re using an FET in Common Source mode, the input signal needs to be out of phase with the intended output signal. When we’re using it in Common Gate mode, the input needs to be in phase. By simply attaching the Gate and Source to opposite sides of the differential, Q9 can be driven both ways at the same time. Determining the legality of this action in your state is your responsibility.
Given that most MOSFETs—including the ones I used in the GR-25—are enhancement mode devices, that means Q9’s Gate needs to be more negative (or less positive, if you prefer) than its Source for it to bias properly. That’s easy enough to arrange, just make the sum of R2 and R10 large enough that the voltage drop across them forward biases Q9’s Gate relative to its Source, which sees only the voltage at R9. As a practical matter, the MOSFETs in the GR-25 want to see about 2-3V between their Gates and Sources before they’ll turn on. This somewhat simplified schematic shows two fixed resistors. In the full schematic I used two resistors and a pot because even matched MOSFETs tend to have some variation in Vgs (the voltage between the Gate and Source) and some means of fine tuning is needed in order to get everything to balance properly.
Whether the term folded cascode applies to this part of the circuit is very much open to debate. A normal folded cascode would get its signal only through its Source. Its Gate would be connected to a voltage divider or some other fixed voltage reference and would not see an active signal at all. On the other hand, a normal Common Source connection would not have a signal introduced to its Source. So what exactly is this thing, anyway? As I said, John declines to give it a name and to my eye the Source connection, even though it’s proportionately dwarfed by the signal at the Gate, is there nonetheless. Setting aside the relative amounts of signal and amplification, visually the thing looks more like a folded cascode to me. Clearly, that’s a subjective judgement, not a technical one, and other people will come to other conclusions. For the time being, it’s enough to say that it works and be on about our business.
This is the first place I will be modifying the GR-25. The first change will be to insert bipolar transistors in place of these MOSFETs, while leaving the topology untouched. There are numerous other possible variations, though. Perhaps the most obvious would be to go to a straight folded cascode. Another would be to use those same devices in Common Source mode. From there current mirrors are only a hop, skip, and a jump, and could easily spawn a half-dozen sub-variations. And there are other possibilities as well. Choosing between them comes down to a question of hours in a day. How many variations can you look at before you run out of time? For Versions 1.x and 2.x I stayed fairly close to John’s original topology, just to have a baseline to compare everything else to.
As the schematic stands right now Q9, Q10, Q11, and Q12 are where the majority of the voltage gain takes place. The N-ch and P-ch devices stand facing each other and see each others’ Drains as loads. Although not strictly true, it’s not too far off the mark to say that each N/P pair serve as active loads for one another. A conventional active load is a current source, undriven, whereas both ends of this portion of the circuit are driven, but given the enormous load they present to each other, the potential gain is quite large.
For those who feel that negative feedback is the ultimate goal, this sort of hookup is bread and butter. But for someone attempting to build a non-feedback circuit, that much gain represents a burden. We need some way to deflate that gain. Since the impedance is the majority of the problem, the simple and obvious way to bring the gain down is to lower the load seen by the MOSFETs. R17 and R18 lower the impedance at those two nodes by several orders of magnitude. Gain drops accordingly. They also serve to ground reference the signal on both sides and bring some DC stability to the amp.
R13, R14, R15, and R16 are Gate stoppers. They don’t do much in the audio sense; they’re there to keep the MOSFETs’ Gates from turning into oscillators. The capacitance present at a MOSFET’s Gate has an unfortunate tendency to interact with the rest of the circuit to produce oscillations up in the megahertz range. Very unpleasant. Let’s skip that part of the evening’s entertainment, shall we? The value of the stopper resistors isn’t necessarily all that critical, but if you decide to play around with them, it’s a good idea to have an oscilloscope handy. If you drop the value too far you can get back into oscillation trouble. One side effect of the stopper resistor value is that it helps determine bandwidth. If you are having RF problems with a MOSFET amplifier, raise the value of the stopper resistors. Bandwidth will drop and you may be able to cure the RF problems right there, without having to do anything else.
And with that, I’ve about covered the “turning the corner” part of the circuit, regardless of what John wants to call it. (Plus I’m running out of time…again…)
Grey
The complementary differential is a marvel, but it’s not quite enough to build a complete amplifier on its own. The circuit has four (potential) signals coming out, but they’re limited as to current delivery and too far away from ground to make much use of them as they are. Some way to increase the available current and to bring the disparate signals back together is needed.
Fear not, there is a way. Actually, there are several ways, but the way I chose to use for the GR-25 is the one that John used in the JC-3. After all, what good’s using the topology if you’re not going to use all of it? (Don’t argue…we’ll get to variations sooner or later…[Probably later, the way my life runs.])
The circuit John used is called…it’s called a…er…well, honestly, I don’t know what to call it. I’ve tried to get John to give the thing a name of some sort, but he’s not offered one to date. Like the JFETs in the differential, it does double duty. It’s both Common Source and Common Gate at the same time. Incidentally, note that I’ve used MOSFETs in these positions. The nomenclature for a MOSFET is the same as for a JFET, so we’ll still be talking in terms of Gates, Sources, and Drains.
For the time being let’s look at Q9 and ignore everything else. Q9’s Gate derives its signal from Q6 and R2. That’s not particularly outlandish taken on its own. All things being equal (which they aren’t, but we’ll come back and fill in the blanks shortly), Q9 would take that signal, amplify it, and toss it out into the world via its Drain. Q9’s Drain sees Q11’s Drain as a load, and since the impedance at a MOSFET’s Drain is very, very high indeed, the potential gain is absolutely enormous. 60dB or more, all at once. Whew! But before we deal with that, we need to address the signal at Q9’s Source. That signal comes from the current flowing through R9 from Q5. This is a curious state of affairs…driving a transistor through two of its orifices at once! (I hear that’s illegal in the entire state of Mississippi and in certain counties in Tennessee. South Carolina doesn’t admit that it can be done at all.) Nonetheless, that’s what’s happening here. The interesting part is that since the differential offers signals of both polarities it’s possible to drive Q9 both ways in an additive sense. Remember that when we’re using an FET in Common Source mode, the input signal needs to be out of phase with the intended output signal. When we’re using it in Common Gate mode, the input needs to be in phase. By simply attaching the Gate and Source to opposite sides of the differential, Q9 can be driven both ways at the same time. Determining the legality of this action in your state is your responsibility.
Given that most MOSFETs—including the ones I used in the GR-25—are enhancement mode devices, that means Q9’s Gate needs to be more negative (or less positive, if you prefer) than its Source for it to bias properly. That’s easy enough to arrange, just make the sum of R2 and R10 large enough that the voltage drop across them forward biases Q9’s Gate relative to its Source, which sees only the voltage at R9. As a practical matter, the MOSFETs in the GR-25 want to see about 2-3V between their Gates and Sources before they’ll turn on. This somewhat simplified schematic shows two fixed resistors. In the full schematic I used two resistors and a pot because even matched MOSFETs tend to have some variation in Vgs (the voltage between the Gate and Source) and some means of fine tuning is needed in order to get everything to balance properly.
Whether the term folded cascode applies to this part of the circuit is very much open to debate. A normal folded cascode would get its signal only through its Source. Its Gate would be connected to a voltage divider or some other fixed voltage reference and would not see an active signal at all. On the other hand, a normal Common Source connection would not have a signal introduced to its Source. So what exactly is this thing, anyway? As I said, John declines to give it a name and to my eye the Source connection, even though it’s proportionately dwarfed by the signal at the Gate, is there nonetheless. Setting aside the relative amounts of signal and amplification, visually the thing looks more like a folded cascode to me. Clearly, that’s a subjective judgement, not a technical one, and other people will come to other conclusions. For the time being, it’s enough to say that it works and be on about our business.
This is the first place I will be modifying the GR-25. The first change will be to insert bipolar transistors in place of these MOSFETs, while leaving the topology untouched. There are numerous other possible variations, though. Perhaps the most obvious would be to go to a straight folded cascode. Another would be to use those same devices in Common Source mode. From there current mirrors are only a hop, skip, and a jump, and could easily spawn a half-dozen sub-variations. And there are other possibilities as well. Choosing between them comes down to a question of hours in a day. How many variations can you look at before you run out of time? For Versions 1.x and 2.x I stayed fairly close to John’s original topology, just to have a baseline to compare everything else to.
As the schematic stands right now Q9, Q10, Q11, and Q12 are where the majority of the voltage gain takes place. The N-ch and P-ch devices stand facing each other and see each others’ Drains as loads. Although not strictly true, it’s not too far off the mark to say that each N/P pair serve as active loads for one another. A conventional active load is a current source, undriven, whereas both ends of this portion of the circuit are driven, but given the enormous load they present to each other, the potential gain is quite large.
For those who feel that negative feedback is the ultimate goal, this sort of hookup is bread and butter. But for someone attempting to build a non-feedback circuit, that much gain represents a burden. We need some way to deflate that gain. Since the impedance is the majority of the problem, the simple and obvious way to bring the gain down is to lower the load seen by the MOSFETs. R17 and R18 lower the impedance at those two nodes by several orders of magnitude. Gain drops accordingly. They also serve to ground reference the signal on both sides and bring some DC stability to the amp.
R13, R14, R15, and R16 are Gate stoppers. They don’t do much in the audio sense; they’re there to keep the MOSFETs’ Gates from turning into oscillators. The capacitance present at a MOSFET’s Gate has an unfortunate tendency to interact with the rest of the circuit to produce oscillations up in the megahertz range. Very unpleasant. Let’s skip that part of the evening’s entertainment, shall we? The value of the stopper resistors isn’t necessarily all that critical, but if you decide to play around with them, it’s a good idea to have an oscilloscope handy. If you drop the value too far you can get back into oscillation trouble. One side effect of the stopper resistor value is that it helps determine bandwidth. If you are having RF problems with a MOSFET amplifier, raise the value of the stopper resistors. Bandwidth will drop and you may be able to cure the RF problems right there, without having to do anything else.
And with that, I’ve about covered the “turning the corner” part of the circuit, regardless of what John wants to call it. (Plus I’m running out of time…again…)
Grey
So...when are the PCB's gonna be ready??......just kidding! I've been following this thread from the beginning, learning a lot, but still clueless. We need to genetically enhance you, so you can work 48 hours straight and need only 6 hours rest, so you'll stop running out of time. I hope to eventually build this amp. Thanks Grey.
-john
-john
GRollins said:
My first thought was to agree with you, at least to the extent that it partakes somewhat of the folded cascode nature. That would follow from what I've read about similar but unbalanced circuits (omitting the Q9/11 output side). The explanation I recall handwaved about how the f.c. addition might provide a lower gain but wider bandwidth path (I think this must have assumed a shunt capacitor across your R1/9). And that might be true for the single-sided version.
But if I trust the calculations I just did, for the balanced arrangement (and with no passive rolloff, though I don't think that matters since it would be applied symmetrically, hence rolling off the f.c. injected signal as well), the small signal gain is EXACTLY the same as if the input stage had a single R (equal to R1 + R9) and the 2nd stage had a completely separate source resistor with the value of R9. (describing only the upper half, and as in this last bit sometimes only the upper right quarter). There is obviously a DC biasing difference, since the 2nd stage's currrent increases the DC gate bias, though with the values present in the full amp I don't think that's terrifically significant, but it might make the need for N/P matching of the 2nd stage a bit more important, as that bit of positive feedback will only increase whatever DC mismatch they have.
Assuming my modelling and math don't have any errors - possible, but I did this between interruptions this morning - the cascode portion doesn't seem to have any effect on the audio signal... well, also assuming perfectly matched components. Along with checking these results, I also want to try it without assuming all that symmetry. Offhand I'm certain only that this will produce rather more long-winded equations...
(What is it with this thread and people e-mailing instead of posting?)
Imagine the following in a rich, fruity, on-air, radio DJ tone of voice...
And top 'o the morning to you...time to rise and shine! Here at G-R-E-Y radio, we play the hits, only the hits, and nothing but the hits, 24-7. Except when we're reading letters from our loyal listeners. And speaking of which, we'll now go to our on-air mail segment, where we reach into the mail bag and drag out a randomly chosen letter to read over the air.
Okay...I'm reaching into the bag...I've got a letter...I'm opening it...and it says:
Dear Grey,
Last night was wonderful! I've never felt anything like that. When you started licking my...
Whoa!
Sorry, wrong letter, wrong bag...
So...I reach into my other mail bag...the one with the audio letters in it...and...let's see...this one's from "Frantic in Fresno." It says:
Dear Grey,
How about those series resistors at the input? How do I know whether I need them and how do I calculate them?
Well, Frantic, it's like this. You remember the Gate stoppers for the MOSFETs? The ones to keep the capacitance from turning into an ad hoc oscillator? Small town version of the same thing. A JFET doesn't have as much capacitance--not by a long shot--but it can still kick up a fuss under the wrong circumstances. So we put in a Gate stopper. Now, as to calculating same...again, it's like the MOSFETs. You've got a lot of latitude. If you live a charmed life--like us DJs, what with groupies and all--you may be able to get by without one. Kinda like ridin' bareback, if you know what I mean...ahem...but if you're having trouble with the ladies, er, oscillations, that is, you'd be better off with some protection. Ya dig? Try something like 1k. If that doesn't knock it out, work your way up to 10k. Just look at it on a scope and see if anything weird or kinky is going on. If it's weird, keep it to yourself. If it's kinky, give me a call--maybe we can work something out.
Just remember, you're creating a voltage divider at the input. The amp's overall gain will drop a bit depending on the relative values of the series resistor and the one to ground. You can also make a case for using lower value resistors all the way 'round, due to the noise that arises from high resistor values. But since there's not that much gain between the input of an amp and the speaker--not compared to the rest of the system, anyway--a little noise will remain a little noise 'cause it only gets amplified about 20x or so.
And now we'll get back to the music. Remember, the music comes first, here at G-R-E-Y radio. And while Mr. Plant is crooning Whole Lotta Love I'm going to get back to that first letter. I kinda liked the direction that was headed...
Grey
P.S.:
www.ceitron.com
www.mcmelectronics.com
www.bdent.com
www.matchaknob.com
etc. More will come to me (just the way the ladies do) if I stop and think about it. Or maybe others will post sources, too.
For that matter, PART III is coming, too. But not yet. Maybe I shoulda played Sixty Minute Man...
Imagine the following in a rich, fruity, on-air, radio DJ tone of voice...
And top 'o the morning to you...time to rise and shine! Here at G-R-E-Y radio, we play the hits, only the hits, and nothing but the hits, 24-7. Except when we're reading letters from our loyal listeners. And speaking of which, we'll now go to our on-air mail segment, where we reach into the mail bag and drag out a randomly chosen letter to read over the air.
Okay...I'm reaching into the bag...I've got a letter...I'm opening it...and it says:
Dear Grey,
Last night was wonderful! I've never felt anything like that. When you started licking my...
Whoa!
Sorry, wrong letter, wrong bag...
So...I reach into my other mail bag...the one with the audio letters in it...and...let's see...this one's from "Frantic in Fresno." It says:
Dear Grey,
How about those series resistors at the input? How do I know whether I need them and how do I calculate them?
Well, Frantic, it's like this. You remember the Gate stoppers for the MOSFETs? The ones to keep the capacitance from turning into an ad hoc oscillator? Small town version of the same thing. A JFET doesn't have as much capacitance--not by a long shot--but it can still kick up a fuss under the wrong circumstances. So we put in a Gate stopper. Now, as to calculating same...again, it's like the MOSFETs. You've got a lot of latitude. If you live a charmed life--like us DJs, what with groupies and all--you may be able to get by without one. Kinda like ridin' bareback, if you know what I mean...ahem...but if you're having trouble with the ladies, er, oscillations, that is, you'd be better off with some protection. Ya dig? Try something like 1k. If that doesn't knock it out, work your way up to 10k. Just look at it on a scope and see if anything weird or kinky is going on. If it's weird, keep it to yourself. If it's kinky, give me a call--maybe we can work something out.
Just remember, you're creating a voltage divider at the input. The amp's overall gain will drop a bit depending on the relative values of the series resistor and the one to ground. You can also make a case for using lower value resistors all the way 'round, due to the noise that arises from high resistor values. But since there's not that much gain between the input of an amp and the speaker--not compared to the rest of the system, anyway--a little noise will remain a little noise 'cause it only gets amplified about 20x or so.
And now we'll get back to the music. Remember, the music comes first, here at G-R-E-Y radio. And while Mr. Plant is crooning Whole Lotta Love I'm going to get back to that first letter. I kinda liked the direction that was headed...
Grey
P.S.:
www.ceitron.com
www.mcmelectronics.com
www.bdent.com
www.matchaknob.com
etc. More will come to me (just the way the ladies do) if I stop and think about it. Or maybe others will post sources, too.
For that matter, PART III is coming, too. But not yet. Maybe I shoulda played Sixty Minute Man...
maney,
Yep...I'll be the first to admit that--with a mere 10 Ohms under its tail--the thing doesn't get much signal as a folded cascode. As I believe I noted above, to me it's a visual thing. I look at the schematic and I see a folded cascode, regardless of the relative proportions of the signals at Gate and Source. I have no quarrel with those who want to call it a Common Source connection. It's more technically correct.
But I still see that link at the Source...
Grey
Yep...I'll be the first to admit that--with a mere 10 Ohms under its tail--the thing doesn't get much signal as a folded cascode. As I believe I noted above, to me it's a visual thing. I look at the schematic and I see a folded cascode, regardless of the relative proportions of the signals at Gate and Source. I have no quarrel with those who want to call it a Common Source connection. It's more technically correct.
But I still see that link at the Source...
Grey
GRollins said:
A simple common-source stage (R1 is the resistor from gate to common; R2 from source to common, and plain i is the output of the preceeding stage) has output:
io = 1 / (1 + gmR2) * gmR1i
The first term is the effect of the source degeneration, while the second is obviously just the input voltage times the transconductance. When the complementary input current is fed into the source, it doesn't change a great deal:
io = 1 / (1 + gmR2) * gm(R1+R2)i
So the ratio of the common-source and folded cascode gains is the same as the ratio of the resistors. No surprise, huh?
I'm skipping the single-ended bootstrapped circuit (R1 returned to the source rather than the common) because its gain is obviously the same as an undegenerated common source amp, at least at "low" frequencies, and offers no new insight. On to the circuit you posted as foldcasc.pdf. I'm modeling one half of the complementary circuit, and have interesting results only for the simplifying case where the resistors are perfectly matched. To confuse thing s just a little, R1 in the equation is the value of eg., R1 and R2 in Grey's circuit, and R2 is the value of eg., R9 and R10. I solve for the difference of the signal currents in Q10 and Q9.
i1 - i2 = 1 / (1 + (gm1 + gm2)R2) * (gm1 + gm2)(R1 + R2)i
Just about what you'd expect from uncoupled common source except for the doubling of the effect of the source degeneration resistors. Hmmm, I take it back about no other interesting results - it strikes me that the equations before merging for the simplified case fail to show any assymetry due to gm mismatches - the differential terms depend only on the sum of the device transconductances (assuming eg., R9 = R10), which is unexpected.
-- Martin "I don't think I'll be adding the FET's square law to the model, though" Maney
Good Evening ladies and gents,
In that last post (the one with all the math) , what is Gm???
sorry for the ignorance, but in my line of work it means General Motors. as for Mr. Rollins anything you think can be multiplied by a factor of 10, all you see here is the tip of the iceberg...
The Truth Is Out There...
Regards, Elwood
PS, I have Pictures
In that last post (the one with all the math) , what is Gm???
sorry for the ignorance, but in my line of work it means General Motors. as for Mr. Rollins anything you think can be multiplied by a factor of 10, all you see here is the tip of the iceberg...
The Truth Is Out There...
Regards, Elwood
PS, I have Pictures
GRollins said:
Nah, that would be Gm * Rout, and I doubt anyone has thought about transistors that way since maybe the early days of sand.
Mu is significant with tubes because it often limits the gain one can get from a tube with not too extreme circuits.eyoung: gm is the transconductance: ratio of change in output current to change in input voltage. Not usually mentioned when talking about bipolar transistors because its value varies exponentially with collector current. You might want to look over Borberly's articles titled JFETs: The New Frontier at the articles page on Borberly 's site.
That's why I said 'kinda' and 'sorta.' Elwood thinks in tube-ese, where the term current doesn't have the same meaning it does in solid state-ese. The idea of an apples/oranges comparison like transconductance isn't as intuitive as voltage/voltage.
There's a time for nit-picking. This ain't it.
Grey
There's a time for nit-picking. This ain't it.
Grey
GRollins said:
Obviously, i don't know Elwood or his background, but I learned about transconductance originally in the context of tubes, so this distinction makes no sense to me at all. Frankly I remember more discussion and use of gm and rp than of mu - the latter was a sort of hazy general classification of small triodes that was seldom if ever used in design (at least that I recall - thirty year old memories may have aged less than perfectly).
What, didn't everyone once cherish a much-banged-up copy of the Radiotron Designer's Handbook?
I've known Elwood for nearly forty years. He and I have been up to mischief that wouldn't bear repeating in a public forum (even if the statute of limitations has run out [which in some cases it may not have]). In fact, he's the only person out of the nearly 100k members here that I've met face to face; our relationship far predates this site, or the Web, or even personal computers, for that matter. I know how he thinks (and he knows how I think, for that matter).
(He's also the only person on the planet who might actually be in possession of pictures of me in...er...compromising positions.)
I'm a firm believer in the Suzuki method. Teach them to play music. Let them enjoy themselves. Leave the theory for later...if ever. If you whack people upside the head with theory right off the bat, you'll chase them off. It's the philosophy I've tried to live by for years and I did my damnedest to build it into the DNA of this place. Whether I succeeded or not only time will tell, but it seems to be working so far.
Patience is the key, though I'll be the first to admit that mine has lapsed on occasion. When the time is right, people will ask deeper questions. Analogies work better for certain types of questions than pages of math and "therefore it is obvious that..." conclusions.
Or, to put it another way, jackrabbit starts at stoplights are dramatic, but rarely serve much purpose.
Grey
P.S.: So Elwood is at a stoplight in his Firebird. A Corvette pulls up next to him and guns his engine. Elwood blips his throttle in response. The light turns green and the race is on. With great difficulty, both cars haul down in time to keep from running the next stoplight, which is red. Elwood and the 'Vette driver are running their engines at redline. The light turns green, clutches and tires scramble for traction, and they both scream down the next block. Next light is red. They pull up, side by side, engines roaring, full throttle. The light turns green, the Corvette blasts across the intersection at ninety percent of the speed of light...straight into the waiting arms of a cop.
Elwood drove away at a sedate pace more appropriate for a blue-haired granny, grinning.
You see, he knew the speed trap was there.
Cunning, Elwood is.
Don't mess with the 'Wood.
(He's also the only person on the planet who might actually be in possession of pictures of me in...er...compromising positions.)
I'm a firm believer in the Suzuki method. Teach them to play music. Let them enjoy themselves. Leave the theory for later...if ever. If you whack people upside the head with theory right off the bat, you'll chase them off. It's the philosophy I've tried to live by for years and I did my damnedest to build it into the DNA of this place. Whether I succeeded or not only time will tell, but it seems to be working so far.
Patience is the key, though I'll be the first to admit that mine has lapsed on occasion. When the time is right, people will ask deeper questions. Analogies work better for certain types of questions than pages of math and "therefore it is obvious that..." conclusions.
Or, to put it another way, jackrabbit starts at stoplights are dramatic, but rarely serve much purpose.
Grey
P.S.: So Elwood is at a stoplight in his Firebird. A Corvette pulls up next to him and guns his engine. Elwood blips his throttle in response. The light turns green and the race is on. With great difficulty, both cars haul down in time to keep from running the next stoplight, which is red. Elwood and the 'Vette driver are running their engines at redline. The light turns green, clutches and tires scramble for traction, and they both scream down the next block. Next light is red. They pull up, side by side, engines roaring, full throttle. The light turns green, the Corvette blasts across the intersection at ninety percent of the speed of light...straight into the waiting arms of a cop.
Elwood drove away at a sedate pace more appropriate for a blue-haired granny, grinning.
You see, he knew the speed trap was there.
Cunning, Elwood is.
Don't mess with the 'Wood.
PART III
When we last left it, the circuit was finally beginning to resemble the GR-25 front end. In this installment, I’m going to cascode the cascoded cascodes (always wanted to say that…) and tidy up a detail or two. No, I’m not going to go so far as to step through the addition of the degeneration resistors; there will be a bit left to your imagination. The full schematic is available in the first post and you can compare it with this one at your leisure. If the differences between the full schematic and this one still bother you, ask, and I’ll fill in the blanks. For the time being, my intention is to leave this schematic a little bit on the spare side so it’s not quite so visually confusing.
The output cascodes (Q13, Q14, Q15, Q16) serve pretty much the same function that the front end cascodes did. In the Version 1 circuit, they took over part of the rail voltage from the paralleled JFETs that I was using in the “turning the corner” position. Those JFETs (J310/J271) worked fairly well, but I wanted more current and the Siliconix parts I was using were devilishly hard to get. I dumped the JFET idea and put MOSFETs in their place. That did a marvellous job of taking care of the current problem and also solved the voltage limitations that I’d faced using the JFETs. The tradeoff was that I had accumulated a lot of Gate capacitance that I needed to deal with. I grumped and grumbled and cussed and fumed about it, but eventually faced the fact that if I wanted to get that capacitance under control, I’d need to keep the output cascoded.
Any gain device has parasitic capacitance between any two of its pins. This is true for all devices, whether tube or solid state. In the case of a MOSFET, there is capacitance between the Gate and the Drain, and also between the Source and the Drain. Those two capacitances have to be charged and discharged every single time the Drain moves from its idle condition. The result is that bandwidth is lowered and distortion is introduced. A cascode works by drastically limiting the voltage changes that the Drain of the gain device sees. The Source of the cascode is a low impedance connection that requires very little of the gain device, voltage-wise. Less voltage swing equals less charging and discharging of the capacitances at the Drain, and life is better. Given that the two devices are in series, the same current that flows in the gain device will flow in the cascode, so the signal hardly knows the cascode is there (…at least in theory).
So the cascodes stayed in, but I changed from the IRF610/IRF9610 that I had been using to the same 2SK2013/2SJ313 parts that I used in the folded position. (Yes, I know some people—John being one—aren’t happy when I call it a folded cascode.) A simple voltage divider sets the reference voltage at the Gates of the cascodes. The voltage divider could just as easily go from rail to ground, but I chose to run it from rail to rail so as to keep the currents from flowing in the ground plane on the circuit board. The full circuit splits R20 and R23 into two resistors in series to dissipate heat a little better. I use RN60 resistors, which are 1/4W in MILSPEC terms, but 1/2W in commercial terms. The split values are just barely perceptably warm to the touch. Since they’re carrying DC only, I’m not worried about AC effects. In a perfectionist sense they’re going to generate a slightly greater amount of thermal noise, but this is an amplifier, not a phono stage, so I’m not bothered by it. If you want to use a larger wattage resistor there, feel free to do so. For that matter, you could put capacitors there to short the noise at the Gate to ground and stabilize the DC reference even further. There are always little things you can find to fiddle with if you use enough magnification on your microscope while looking for imperfections.
The load resistors (R17, R18) are split now, to give two outputs to facilitate attaching the bias circuit and output stage. Imagine this as being analogous to butterflying a thick steak. The resistance is left attached at the ground end and the single output becomes two at the other ends.
I’ve also put in resistors at the inputs; probably should have from the beginning. I left out the series resistors as they’re not really necessary to understanding the circuit and I was trying to keep the conceptual schematic simple. The series resistors are one of those things that fall into the category of “when in doubt, put them in.” If you have to ask me if you need them, put them in just to be on the safe side.
The degeneration resistors under the Sources of Q3, Q4, Q13, and Q14 in the full schematic are there to linearize the MOSFETs and drop the gain a bit more. The final gain of the amp is a delicate balancing act. Nearly every resistor in the circuit changes the gain in some way. Some up. Some down. I’d like to say that this circuit can be freely modified…but…well, it can, but you’d better have an oscilloscope and some meters handy. You’re not going to want to change things at random—particularly not five or six things all at the same time. Modifying this circuit is something you’ll want to approach incrementally.
Grey
When we last left it, the circuit was finally beginning to resemble the GR-25 front end. In this installment, I’m going to cascode the cascoded cascodes (always wanted to say that…) and tidy up a detail or two. No, I’m not going to go so far as to step through the addition of the degeneration resistors; there will be a bit left to your imagination. The full schematic is available in the first post and you can compare it with this one at your leisure. If the differences between the full schematic and this one still bother you, ask, and I’ll fill in the blanks. For the time being, my intention is to leave this schematic a little bit on the spare side so it’s not quite so visually confusing.
The output cascodes (Q13, Q14, Q15, Q16) serve pretty much the same function that the front end cascodes did. In the Version 1 circuit, they took over part of the rail voltage from the paralleled JFETs that I was using in the “turning the corner” position. Those JFETs (J310/J271) worked fairly well, but I wanted more current and the Siliconix parts I was using were devilishly hard to get. I dumped the JFET idea and put MOSFETs in their place. That did a marvellous job of taking care of the current problem and also solved the voltage limitations that I’d faced using the JFETs. The tradeoff was that I had accumulated a lot of Gate capacitance that I needed to deal with. I grumped and grumbled and cussed and fumed about it, but eventually faced the fact that if I wanted to get that capacitance under control, I’d need to keep the output cascoded.
Any gain device has parasitic capacitance between any two of its pins. This is true for all devices, whether tube or solid state. In the case of a MOSFET, there is capacitance between the Gate and the Drain, and also between the Source and the Drain. Those two capacitances have to be charged and discharged every single time the Drain moves from its idle condition. The result is that bandwidth is lowered and distortion is introduced. A cascode works by drastically limiting the voltage changes that the Drain of the gain device sees. The Source of the cascode is a low impedance connection that requires very little of the gain device, voltage-wise. Less voltage swing equals less charging and discharging of the capacitances at the Drain, and life is better. Given that the two devices are in series, the same current that flows in the gain device will flow in the cascode, so the signal hardly knows the cascode is there (…at least in theory).
So the cascodes stayed in, but I changed from the IRF610/IRF9610 that I had been using to the same 2SK2013/2SJ313 parts that I used in the folded position. (Yes, I know some people—John being one—aren’t happy when I call it a folded cascode.) A simple voltage divider sets the reference voltage at the Gates of the cascodes. The voltage divider could just as easily go from rail to ground, but I chose to run it from rail to rail so as to keep the currents from flowing in the ground plane on the circuit board. The full circuit splits R20 and R23 into two resistors in series to dissipate heat a little better. I use RN60 resistors, which are 1/4W in MILSPEC terms, but 1/2W in commercial terms. The split values are just barely perceptably warm to the touch. Since they’re carrying DC only, I’m not worried about AC effects. In a perfectionist sense they’re going to generate a slightly greater amount of thermal noise, but this is an amplifier, not a phono stage, so I’m not bothered by it. If you want to use a larger wattage resistor there, feel free to do so. For that matter, you could put capacitors there to short the noise at the Gate to ground and stabilize the DC reference even further. There are always little things you can find to fiddle with if you use enough magnification on your microscope while looking for imperfections.
The load resistors (R17, R18) are split now, to give two outputs to facilitate attaching the bias circuit and output stage. Imagine this as being analogous to butterflying a thick steak. The resistance is left attached at the ground end and the single output becomes two at the other ends.
I’ve also put in resistors at the inputs; probably should have from the beginning. I left out the series resistors as they’re not really necessary to understanding the circuit and I was trying to keep the conceptual schematic simple. The series resistors are one of those things that fall into the category of “when in doubt, put them in.” If you have to ask me if you need them, put them in just to be on the safe side.
The degeneration resistors under the Sources of Q3, Q4, Q13, and Q14 in the full schematic are there to linearize the MOSFETs and drop the gain a bit more. The final gain of the amp is a delicate balancing act. Nearly every resistor in the circuit changes the gain in some way. Some up. Some down. I’d like to say that this circuit can be freely modified…but…well, it can, but you’d better have an oscilloscope and some meters handy. You’re not going to want to change things at random—particularly not five or six things all at the same time. Modifying this circuit is something you’ll want to approach incrementally.
Grey
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.