Once upon a time, there were such things as tubes. They were, for all intents and purposes, N-channel devices. There were no P-channel tubes available, so a great deal of time, effort, and ingenuity went into developing circuits that used only N-ch devices.
One of the many clever circuits that came out of the tube era was the Circlotron. Since the Circlotron topology has put in only sporadic appearances since transistors gained ascendancy, it is not as familiar as other topologies. Essentially, it is a push-pull circuit comprised of two single-ended amplifiers operated nose to tail. That doesn’t sound too unusual until you notice the power supplies. Plural. There are two of them. In a Circlotron, you have one amplifier circuit, followed in series by a power supply, then by the other amplifier circuit, then another power supply, which in turn leads back to the first amplifier circuit, all in a circle. In the old days the conceptual schematics were frequently drawn in a literal circle to emphasize the series nature of things. Given that modern electronics CAD programs aren't easily convinced to draw circular circuits, nowadays a Circlotron diagram usually ends up looking something like this:
One of the many clever circuits that came out of the tube era was the Circlotron. Since the Circlotron topology has put in only sporadic appearances since transistors gained ascendancy, it is not as familiar as other topologies. Essentially, it is a push-pull circuit comprised of two single-ended amplifiers operated nose to tail. That doesn’t sound too unusual until you notice the power supplies. Plural. There are two of them. In a Circlotron, you have one amplifier circuit, followed in series by a power supply, then by the other amplifier circuit, then another power supply, which in turn leads back to the first amplifier circuit, all in a circle. In the old days the conceptual schematics were frequently drawn in a literal circle to emphasize the series nature of things. Given that modern electronics CAD programs aren't easily convinced to draw circular circuits, nowadays a Circlotron diagram usually ends up looking something like this:
Attachments
Many tube topologies were adopted by solid state practitioners once transistors became widely available. Some were not. The Circlotron has not been totally forgotten, but there have been few commercial examples. Jim Bongiorno updated the circuit to use solid state components in the Eighties and received a patent for his efforts. The next commercial example came from Atma-Sphere, but Ralph Karsten used tubes once more. His innovation was to direct-couple the output stage, thus getting rid of the pesky output transformer. He also received a patent. (Atma-Sphere, incidentally, is still in business. They can be found on the web at www.atma-sphere.com.)
A separate historical line joins at this point; that of the power JFET. Small signal JFETs are common devices. They generally have modest gain and are—by our standards—rather limited in the amount of voltage that they can withstand. They do provide a really high input impedance, however, and the transfer curve is frequently useful for audio work. They are also quiet. Those are powerful arguments for using JFETs in audio equipment, but at some point it becomes necessary to use either MOSFETs or bipolar transistors if either high voltages or currents are needed.
Back during the Seventies, there were power JFETs for a short period of time, but they followed the Dodo bird and the dinosaur and became extinct. During their short lifespan, they garnered a lot of attention and people have been on the lookout ever since for similar devices.
Not so long ago a company by the name of Lovoltech began manufacturing power JFETs again. They designed their parts for computer power supplies. It never occurred to them that people in the audio branch of electronics might be interested in such a thing as a power JFET. In fact, they were rather flustered to find that people wanted to use their parts for linear circuits at all. From my conversations with them, I gather that they think we're a little odd, but that doesn't mean that they won't sell us parts. This is one of those situations where money talks.
The primary problem with the Lovoltech power JFETs is that they have very limited power dissipation capabilities. The TO-251 package is absurdly small—roughly the size of your little fingernail. They are also limited in the amount of voltage they can take. But they can take a reasonable amount of current and that's enough to be getting on with.
The trick to taking the voltage (and with it, the heat) load off of the Lovoltech JFET is to use a circuit called a cascode. A cascode is simply one gain device sitting on top of another one's shoulders. The cascode device need not be the same as the one it is sitting on.
Nelson Pass has published several variants on the cascoded power JFET. All of his circuits to date have been single-ended. Visit www.passdiy.com and have a look at the Zen versions 8 and 9. Note that there are a number of variants on the Zen #9, so it may take you a while to read through it all. I highly recommend that you read both papers.
Judging from the number of power JFETs in the field at the moment, I would say that there are quite a few Zen 8s and 9s being built.
But then there's me. I persist in being a dinosaur (not the extinct variety mentioned above…at least not yet). I have speakers that need more power than the Zen amplifiers can comfortably produce. Granted, they can be scaled up, but some perverse part of me wanted to go in a different direction. There aren't a lot of solid state topologies that will allow the use of two N-ch devices. I considered building a quasi-complimentary amp for a while, then thought about a bridged design such as the Aleph-X. But in the end, I decided to go a different route: I decided to resurrect the Circlotron.
On the plus side, the Circlotron offers a like-device topology. You can use it with all N-ch output devices. It also has the curious property of using the N-ch devices in opposition. As one conducts more, the other conducts less. This lowers distortion because the transfer curves of the devices bow in opposite directions and essentially cancel out.
On the negative side of the equation, the Circlotron requires a little more complicated power supply than most amplifiers. Those two power supplies mentioned earlier aren't the usual plus and minus with ground in the middle that you're used to seeing, they are two identical power supplies, completely separate from one another. Don't panic though, many power transformers these days have two separate secondaries that can be used independently. That sort of transformer will work admirably for this circuit.
The cascode device will need a voltage at the Gate in order to bias the power JFET. There are several ways to accomplish this, but I chose to use a separate power supply. Given that the voltage and current required are very modest, this need not be an expensive proposition.
If you build the circuit the way I've built mine, this means four separate power supplies per channel, two large ones and two small ones.
As I've posted elsewhere, I got fascinated with the process of biasing the JFET. There are any number of other aspects that I could have gotten interested in, but that's the one that grabbed my imagination. Why? I don't know. It just happened that way. The way I've chosen to use in this circuit is clearly a variation on the scheme that Nelson uses in the Zen circuits. Varying the voltage at the Gate of the cascode device (in this case, a MOSFET) raises or lowers the Source, which in turn raises or lowers the voltage seen by the Drain of the power JFET. The JFET's Gate is held at ground potential. Given the number of people who have read Nelson's papers for the Zen amps, this should be a readily understandable variation.
Note that this biasing system is not recommended for class AB or class B operation. This is a class A biasing circuit. There are other biasing circuits that will work better for AB and B. I'll try to get some of those posted later. The reason I went with this one first is that it requires no adjustment. None. You choose the bias by the parts you put into the circuit. After that, everything takes care of itself. I went with this circuit because there are still people who don't understand how to bias the Aleph-X. I figured that they would appreciate a circuit that would not require a lot of fiddling.
Each half of the amplifier circuit is much like Nelson's Zen circuit. That's to be expected, given that one cascode looks pretty much like another. The difference between my circuit and Nelson's is the way I set the voltage at the cascode's Gate. Simply put, I use a small NPN transistor to watch the Source resistor for the JFET. The NPN sets the voltage at the cascode's Gate such that it sees its Vbe across the JFET's Source resistor. You set the bias by choosing the value of the Source resistor. It's that simple.
After that, it's a question of putting two of these circuits together in a Circlotron arrangement.
Yes, you can take half of the circuit and run it single-ended, if you want. Be my guest.
As far as I know, no one has ever produced a commercial version of a cascoded Circlotron. It's a new circuit. With that in mind, I decided to call this circuit the New-tron. Hopefully, that name will have neither positive nor negative connotations.
For the moment, I am presenting this circuit as a “hot follower.” Built as per the schematic, each cascode stack runs at about Iq=1A. That’s a little more than you might expect to see if you divide the Vbe of the NPN by the value of the Source resistor, but the NPN’s base current drives the actual value up a bit.
At the moment I am using 25V for V1 and V2. Yes, you can run it at higher or lower voltages. To a first approximation, your Vout measured peak to peak will equal one rail. Vrms will be that divided by 2.828. Set your current by adjusting R1, R8, R13, and R19 or add more cascode stacks slaved off of the inside ones. Or both.
Bias A and Bias B are set to 15V. Nothing fancy there, just one of those little Tamura transformers with rectification and a TO-92 case 7815. Would the circuit benefit from more elaborate regulation? Probably. However, my intent was to get the circuit going and that was the first part that came to hand.
Yes, the circuit will tolerate the usual parts changes. I used the IRF644 because I still have some I’m trying to use up. Feel free to use IRFP044s or IRFP240s or whatever. Keep in mind that changing the ZTX450s to another part will alter the bias by slight amounts. In fact, I spent an entire week just playing with different ways to fine tune the bias just within this one circuit. I had to slap myself… “Grey, stop playing. The intent here is to post a circuit that doesn’t require fiddling.” Oh. Right. So I took out all the pots and went back to the circuit as you see it here.
How do you drive it? Given that it doesn’t have a front end, you’ll need a balanced signal that will swing all the voltage that you intend to see at the output. I will be posting an IRF610 differential circuit in the next day or two that will do for starters. Those of you who have a spare Aleph preamp lying about can use that as long as your speakers are fairly efficient. I have other front ends that I will post later. I also have other versions of this output stage and other output stages that are different.
It’s all a question of time, folks. I had hoped to get this thread going back in January. It didn’t happen. Now it’s May and I thought that it would be best to do it on the installment plan rather than try to get everything into one monolithic post.
As always, questions and comments are welcome, but my time is increasingly fragmented and I may not get back to this thread as often as I’d like.
Grey
A separate historical line joins at this point; that of the power JFET. Small signal JFETs are common devices. They generally have modest gain and are—by our standards—rather limited in the amount of voltage that they can withstand. They do provide a really high input impedance, however, and the transfer curve is frequently useful for audio work. They are also quiet. Those are powerful arguments for using JFETs in audio equipment, but at some point it becomes necessary to use either MOSFETs or bipolar transistors if either high voltages or currents are needed.
Back during the Seventies, there were power JFETs for a short period of time, but they followed the Dodo bird and the dinosaur and became extinct. During their short lifespan, they garnered a lot of attention and people have been on the lookout ever since for similar devices.
Not so long ago a company by the name of Lovoltech began manufacturing power JFETs again. They designed their parts for computer power supplies. It never occurred to them that people in the audio branch of electronics might be interested in such a thing as a power JFET. In fact, they were rather flustered to find that people wanted to use their parts for linear circuits at all. From my conversations with them, I gather that they think we're a little odd, but that doesn't mean that they won't sell us parts. This is one of those situations where money talks.
The primary problem with the Lovoltech power JFETs is that they have very limited power dissipation capabilities. The TO-251 package is absurdly small—roughly the size of your little fingernail. They are also limited in the amount of voltage they can take. But they can take a reasonable amount of current and that's enough to be getting on with.
The trick to taking the voltage (and with it, the heat) load off of the Lovoltech JFET is to use a circuit called a cascode. A cascode is simply one gain device sitting on top of another one's shoulders. The cascode device need not be the same as the one it is sitting on.
Nelson Pass has published several variants on the cascoded power JFET. All of his circuits to date have been single-ended. Visit www.passdiy.com and have a look at the Zen versions 8 and 9. Note that there are a number of variants on the Zen #9, so it may take you a while to read through it all. I highly recommend that you read both papers.
Judging from the number of power JFETs in the field at the moment, I would say that there are quite a few Zen 8s and 9s being built.
But then there's me. I persist in being a dinosaur (not the extinct variety mentioned above…at least not yet). I have speakers that need more power than the Zen amplifiers can comfortably produce. Granted, they can be scaled up, but some perverse part of me wanted to go in a different direction. There aren't a lot of solid state topologies that will allow the use of two N-ch devices. I considered building a quasi-complimentary amp for a while, then thought about a bridged design such as the Aleph-X. But in the end, I decided to go a different route: I decided to resurrect the Circlotron.
On the plus side, the Circlotron offers a like-device topology. You can use it with all N-ch output devices. It also has the curious property of using the N-ch devices in opposition. As one conducts more, the other conducts less. This lowers distortion because the transfer curves of the devices bow in opposite directions and essentially cancel out.
On the negative side of the equation, the Circlotron requires a little more complicated power supply than most amplifiers. Those two power supplies mentioned earlier aren't the usual plus and minus with ground in the middle that you're used to seeing, they are two identical power supplies, completely separate from one another. Don't panic though, many power transformers these days have two separate secondaries that can be used independently. That sort of transformer will work admirably for this circuit.
The cascode device will need a voltage at the Gate in order to bias the power JFET. There are several ways to accomplish this, but I chose to use a separate power supply. Given that the voltage and current required are very modest, this need not be an expensive proposition.
If you build the circuit the way I've built mine, this means four separate power supplies per channel, two large ones and two small ones.
As I've posted elsewhere, I got fascinated with the process of biasing the JFET. There are any number of other aspects that I could have gotten interested in, but that's the one that grabbed my imagination. Why? I don't know. It just happened that way. The way I've chosen to use in this circuit is clearly a variation on the scheme that Nelson uses in the Zen circuits. Varying the voltage at the Gate of the cascode device (in this case, a MOSFET) raises or lowers the Source, which in turn raises or lowers the voltage seen by the Drain of the power JFET. The JFET's Gate is held at ground potential. Given the number of people who have read Nelson's papers for the Zen amps, this should be a readily understandable variation.
Note that this biasing system is not recommended for class AB or class B operation. This is a class A biasing circuit. There are other biasing circuits that will work better for AB and B. I'll try to get some of those posted later. The reason I went with this one first is that it requires no adjustment. None. You choose the bias by the parts you put into the circuit. After that, everything takes care of itself. I went with this circuit because there are still people who don't understand how to bias the Aleph-X. I figured that they would appreciate a circuit that would not require a lot of fiddling.
Each half of the amplifier circuit is much like Nelson's Zen circuit. That's to be expected, given that one cascode looks pretty much like another. The difference between my circuit and Nelson's is the way I set the voltage at the cascode's Gate. Simply put, I use a small NPN transistor to watch the Source resistor for the JFET. The NPN sets the voltage at the cascode's Gate such that it sees its Vbe across the JFET's Source resistor. You set the bias by choosing the value of the Source resistor. It's that simple.
After that, it's a question of putting two of these circuits together in a Circlotron arrangement.
Yes, you can take half of the circuit and run it single-ended, if you want. Be my guest.
As far as I know, no one has ever produced a commercial version of a cascoded Circlotron. It's a new circuit. With that in mind, I decided to call this circuit the New-tron. Hopefully, that name will have neither positive nor negative connotations.
For the moment, I am presenting this circuit as a “hot follower.” Built as per the schematic, each cascode stack runs at about Iq=1A. That’s a little more than you might expect to see if you divide the Vbe of the NPN by the value of the Source resistor, but the NPN’s base current drives the actual value up a bit.
At the moment I am using 25V for V1 and V2. Yes, you can run it at higher or lower voltages. To a first approximation, your Vout measured peak to peak will equal one rail. Vrms will be that divided by 2.828. Set your current by adjusting R1, R8, R13, and R19 or add more cascode stacks slaved off of the inside ones. Or both.
Bias A and Bias B are set to 15V. Nothing fancy there, just one of those little Tamura transformers with rectification and a TO-92 case 7815. Would the circuit benefit from more elaborate regulation? Probably. However, my intent was to get the circuit going and that was the first part that came to hand.
Yes, the circuit will tolerate the usual parts changes. I used the IRF644 because I still have some I’m trying to use up. Feel free to use IRFP044s or IRFP240s or whatever. Keep in mind that changing the ZTX450s to another part will alter the bias by slight amounts. In fact, I spent an entire week just playing with different ways to fine tune the bias just within this one circuit. I had to slap myself… “Grey, stop playing. The intent here is to post a circuit that doesn’t require fiddling.” Oh. Right. So I took out all the pots and went back to the circuit as you see it here.
How do you drive it? Given that it doesn’t have a front end, you’ll need a balanced signal that will swing all the voltage that you intend to see at the output. I will be posting an IRF610 differential circuit in the next day or two that will do for starters. Those of you who have a spare Aleph preamp lying about can use that as long as your speakers are fairly efficient. I have other front ends that I will post later. I also have other versions of this output stage and other output stages that are different.
It’s all a question of time, folks. I had hoped to get this thread going back in January. It didn’t happen. Now it’s May and I thought that it would be best to do it on the installment plan rather than try to get everything into one monolithic post.
As always, questions and comments are welcome, but my time is increasingly fragmented and I may not get back to this thread as often as I’d like.
Grey
Attachments
Anyone who needs parts may check the Lovoltech thread in the Marketplace/Group Buys forum.
I suppose it goes without saying that this circuit will take a lot of heatsinking if you intend to generate more than a few watts. Power supplies will need to be robust, also.
I don't have it drawn in the schematic, but you are also invited to add Nelson's "cascode modulation" to the circuit. You'll need separate modulation for each side.
Grey
I suppose it goes without saying that this circuit will take a lot of heatsinking if you intend to generate more than a few watts. Power supplies will need to be robust, also.
I don't have it drawn in the schematic, but you are also invited to add Nelson's "cascode modulation" to the circuit. You'll need separate modulation for each side.
Grey
Very interesting Grey.
One quick question: Isn't there a form of cascode modulation already present in your circuit? I've only had a chance to do a quick once over of your circuit, but when a signal is presented to Q4&5, they will modulate changing the voltage of the source resistors, and thus the voltages at the base of the ZTX450's. Won't this then somewhat modulate the bipolars, causing the current across the bias resistors to modulate, thereby modulating the voltage at the gates of the cascode MOSFET's?
As I said, this commentary is based only on a quick once over look, so I may be blowing smoke. Please correct me if I am wrong.
Cheers, Terry
One quick question: Isn't there a form of cascode modulation already present in your circuit? I've only had a chance to do a quick once over of your circuit, but when a signal is presented to Q4&5, they will modulate changing the voltage of the source resistors, and thus the voltages at the base of the ZTX450's. Won't this then somewhat modulate the bipolars, causing the current across the bias resistors to modulate, thereby modulating the voltage at the gates of the cascode MOSFET's?
As I said, this commentary is based only on a quick once over look, so I may be blowing smoke. Please correct me if I am wrong.
Cheers, Terry
Errata:
In Post #2, I said that the peak output would be approximately one rail. Make that both rails added together.
No surprise: I'm waaay into sleep deprivation.
As an example, I've got the critter on the bench set for 25V rails on the output. It'll swing 50Vp-p or a little over 17Vrms. That's around 40W into 8 Ohms if you've got the current.
(I've already caught three typos in this post and am probably missing others. If I say something goofy, just nod and say that I warned you. Those who feel that everything I say is goofy will not be surprised no matter what I do.)
gl,
I am the cobbler who had no shoes. I never listen to things; I spend what little time I have available at the bench.
No pictures, sorry. I still have an old Cannon AE-1--the kind of camera that runs on (gasp!) film. However, I have no scanner or other way to convert from "real" pictures to digital. I have promised myself that as soon as I run out of film I will begin looking at digital cameras.
(Three more typos.)
Besides, no one wants to look at this thing...it's ugly.
metalman,
At first glance, you'd think so, but as always the devil is in the details. One, the bipolar would flip the polarity of the modulation signal, thus increasing distortion, not decreasing it. Two, it's a moot point because the RC filter at the base of the NPN takes out nearly all the signal, leaving only variations less than 1Hz.
Incidentally, the values of the RC filter can be fiddled around according to whatever you have in your junkbox. All that matters is that you get the cutoff frequency really, really low. I chose to use a relatively high value resistor expressly so that the base current would bump up the bias a bit. If you use a larger capacitor and smaller resistor, you'll get closer to the .65V/Rsource=Iq formula that you would expect.
lumanauw,
I've got more versions of this amp that you can shake a stick at, but I'll need to get a front end or two posted first. At this point, all I really want to do is fall on a pillow and snore, but my daughter needs feeding and tending to, so that's not going to happen.
(Another typo...and two others whilst typing this comment...man, I am toast.)
Grey
P.S.: Oh hell, just when I thought it was safe to go back in the water...I just thought of another way to bias the circuit. Blast and bother!
In Post #2, I said that the peak output would be approximately one rail. Make that both rails added together.
No surprise: I'm waaay into sleep deprivation.
As an example, I've got the critter on the bench set for 25V rails on the output. It'll swing 50Vp-p or a little over 17Vrms. That's around 40W into 8 Ohms if you've got the current.
(I've already caught three typos in this post and am probably missing others. If I say something goofy, just nod and say that I warned you. Those who feel that everything I say is goofy will not be surprised no matter what I do.)
gl,
I am the cobbler who had no shoes. I never listen to things; I spend what little time I have available at the bench.
No pictures, sorry. I still have an old Cannon AE-1--the kind of camera that runs on (gasp!) film. However, I have no scanner or other way to convert from "real" pictures to digital. I have promised myself that as soon as I run out of film I will begin looking at digital cameras.
(Three more typos.)
Besides, no one wants to look at this thing...it's ugly.
metalman,
At first glance, you'd think so, but as always the devil is in the details. One, the bipolar would flip the polarity of the modulation signal, thus increasing distortion, not decreasing it. Two, it's a moot point because the RC filter at the base of the NPN takes out nearly all the signal, leaving only variations less than 1Hz.
Incidentally, the values of the RC filter can be fiddled around according to whatever you have in your junkbox. All that matters is that you get the cutoff frequency really, really low. I chose to use a relatively high value resistor expressly so that the base current would bump up the bias a bit. If you use a larger capacitor and smaller resistor, you'll get closer to the .65V/Rsource=Iq formula that you would expect.
lumanauw,
I've got more versions of this amp that you can shake a stick at, but I'll need to get a front end or two posted first. At this point, all I really want to do is fall on a pillow and snore, but my daughter needs feeding and tending to, so that's not going to happen.
(Another typo...and two others whilst typing this comment...man, I am toast.)
Grey
P.S.: Oh hell, just when I thought it was safe to go back in the water...I just thought of another way to bias the circuit. Blast and bother!
I stole a few minutes this afternoon and scratched out the schematic for the simple front end I've used off and on during my fiddlings. This is not a be-all, end-all front end, just a decent one. In its favor: It's simple and relatively cheap.
In a pinch, it might make a fair-to-middlin' medium gain line stage.
The rail voltages are specified as +40V and -10V. A couple of thoughts:
--Use matched devices for Q2 and Q4.
--It would actually be better if the positive rail were closer to 45V, 40V was just a convenient voltage for me at the time. I didn't need that last few volts of swing on the output; you'll get more symmetrical clipping if you use a slightly higher rail.
--The negative rail is specified as -10V. Feel free to make it -40 V or -45V if it makes you happy, but be sure to heatsink Q3 accordingly. The device can take it, as long as you make arrangements for the heat.
--As always, the rails are important, but since both the positive and negative rails sum to DC, it will take some of the worry out of the power supply end of things. That doesn't mean you can be a slacker, just that you don't have to lie awake at night worrying about five decimal places of accuracy for the rail.
--It draws roughly 30mA of current, total. Call it 15mA per side. Q2 and Q4 will dissipate about .6-.7W each. I'm comfortable running TO-220 devices up to about half a watt without a heatsink. This is slightly over that, and although I'm sure things would have been just fine without, I went ahead and put heatsinks on them. Nelson would, no doubt, run them naked--he ain't happy until he smells burning flesh. I prefer my steak medium-rare rather than well-done.
--Q3--assuming a 10V rail--is okay without a heatsink. If you use a higher rail, use a heatsink.
Just hook this circuit directly to the output stage posted earlier. Note that the output stage is a follower, so it does not invert phase. With that in mind, the negative feedback comes back to the same side of the front end. It does not cross over to the other side like it did in the Aleph-X. If you do swap it over, you'll have positive feedback. This is only recommended if you like listening to the old Chicago track "Free Form Guitar" or live Jimi Hendrix recordings (e.g. "The Star Spangled Banner"). You will not actually need the recording--the circuit will provide all the sound effects you can stand.
Don't say you weren't warned.
Grey
P.S.: I've had complaints about the resolution of the schematics. I feel your pain, as the expression goes. DIY doesn't allow anything over 1000 pixels and I'm at wit's end trying to get a readable schematic in under their limits.
My schematic program exports BMP files. I am resizing and converting them to GIF in another program. There may be better ways of doing this, but &%$)!@# I've got better things to do with my time than sit around trying to convert files all day long. As it is, I was frantically pounding the keyboard this afternoon, running late getting to work, all because of all this file size/format/conversion nonsense.
Grumble.
In a pinch, it might make a fair-to-middlin' medium gain line stage.
The rail voltages are specified as +40V and -10V. A couple of thoughts:
--Use matched devices for Q2 and Q4.
--It would actually be better if the positive rail were closer to 45V, 40V was just a convenient voltage for me at the time. I didn't need that last few volts of swing on the output; you'll get more symmetrical clipping if you use a slightly higher rail.
--The negative rail is specified as -10V. Feel free to make it -40 V or -45V if it makes you happy, but be sure to heatsink Q3 accordingly. The device can take it, as long as you make arrangements for the heat.
--As always, the rails are important, but since both the positive and negative rails sum to DC, it will take some of the worry out of the power supply end of things. That doesn't mean you can be a slacker, just that you don't have to lie awake at night worrying about five decimal places of accuracy for the rail.
--It draws roughly 30mA of current, total. Call it 15mA per side. Q2 and Q4 will dissipate about .6-.7W each. I'm comfortable running TO-220 devices up to about half a watt without a heatsink. This is slightly over that, and although I'm sure things would have been just fine without, I went ahead and put heatsinks on them. Nelson would, no doubt, run them naked--he ain't happy until he smells burning flesh. I prefer my steak medium-rare rather than well-done.
--Q3--assuming a 10V rail--is okay without a heatsink. If you use a higher rail, use a heatsink.
Just hook this circuit directly to the output stage posted earlier. Note that the output stage is a follower, so it does not invert phase. With that in mind, the negative feedback comes back to the same side of the front end. It does not cross over to the other side like it did in the Aleph-X. If you do swap it over, you'll have positive feedback. This is only recommended if you like listening to the old Chicago track "Free Form Guitar" or live Jimi Hendrix recordings (e.g. "The Star Spangled Banner"). You will not actually need the recording--the circuit will provide all the sound effects you can stand.
Don't say you weren't warned.
Grey
P.S.: I've had complaints about the resolution of the schematics. I feel your pain, as the expression goes. DIY doesn't allow anything over 1000 pixels and I'm at wit's end trying to get a readable schematic in under their limits.
My schematic program exports BMP files. I am resizing and converting them to GIF in another program. There may be better ways of doing this, but &%$)!@# I've got better things to do with my time than sit around trying to convert files all day long. As it is, I was frantically pounding the keyboard this afternoon, running late getting to work, all because of all this file size/format/conversion nonsense.
Grumble.
Attachments
balanced cascode modulation
Nice circuit, Grey - thanks for sharing!
If you'll indulge me, I'd like to propose adding a balanced form of Nelson Pass' cascode modulation technique. I believe that circlotrons benefit from some type of cross-coupling that would better enable each half of the circuit to sense the full voltage across the output.
The attached schematic shows how you could cross-couple two cascode bias networks using a simple resistor network. The 15k (nominal) resistor does the cross-coupling, and would be adjusted in situ for best performance. In the attached circuit, constant current sources feed current to the bias network from the output stage power supplies, avoiding the need for a separate bias supply at the cost of a few more parts. I think you could similarly cross-couple other preferred bias arrangements, including the one you show above.
I see the effects of balanced cascode modulation as similar to what happens in triode circlotron output stages such as the Atma-Sphere. Because triodes "see" the voltage at both their anodes and their cathodes, and because each circlotron anode follows the opposing cathode, each half of the stage is better able to independently control both output terminals. A benefit of the circuit shown here is that the anode (drain) signals reflect the output rather than the B+ supplies, so they are relatively free of power supply noise even with unregulated supplies.
Lacking a power FET model, I verified this circuit's basic operation in simulation by scaling the currents down by a factor of 100, in essence simulating a line-stage version of the design. Some component values in the schematic were then re-scaled for a power amplifier. Offered in good faith, but build at your own risk!
Nice circuit, Grey - thanks for sharing!
If you'll indulge me, I'd like to propose adding a balanced form of Nelson Pass' cascode modulation technique. I believe that circlotrons benefit from some type of cross-coupling that would better enable each half of the circuit to sense the full voltage across the output.
The attached schematic shows how you could cross-couple two cascode bias networks using a simple resistor network. The 15k (nominal) resistor does the cross-coupling, and would be adjusted in situ for best performance. In the attached circuit, constant current sources feed current to the bias network from the output stage power supplies, avoiding the need for a separate bias supply at the cost of a few more parts. I think you could similarly cross-couple other preferred bias arrangements, including the one you show above.
I see the effects of balanced cascode modulation as similar to what happens in triode circlotron output stages such as the Atma-Sphere. Because triodes "see" the voltage at both their anodes and their cathodes, and because each circlotron anode follows the opposing cathode, each half of the stage is better able to independently control both output terminals. A benefit of the circuit shown here is that the anode (drain) signals reflect the output rather than the B+ supplies, so they are relatively free of power supply noise even with unregulated supplies.
Lacking a power FET model, I verified this circuit's basic operation in simulation by scaling the currents down by a factor of 100, in essence simulating a line-stage version of the design. Some component values in the schematic were then re-scaled for a power amplifier. Offered in good faith, but build at your own risk!
Attachments
All,
I realized that I had said that the biasing circuit was the only real difference between Nelson's basic building block and mine. There's actually another difference--I take the signal from the Source of the JFET, whereas he takes the signal from the Drain of the MOSFET.
jh6you,
Let's take a closer look at the biasing circuit in Isotope One.
--Assume the amp to be at idle, so there's no signal present. The NPN sets its Vbe across the Lovoltech's Source resistor. This results in a current through the load resistor (R5) that holds the Gate of the cascode device at the Lovoltech's Vds + the MOSFET's Vgs. So far, so good. In fact, it's a current source as long as there's no signal at the Lovoltech's Gate.
--Assume that the circuit is in use, but still within class A operating parameters, meaning that there's signal all the time; the stack never turns off. The RC network at the base of the NPN filters out the signal, keeping C1 charged via a small amount of DC drawn from the main bias current. Again, the cascode device's Gate is held at the proper voltage to bias the Lovoltech.
--Assume that the Lovoltech has been driven into shutoff during part of the amplification cycle. No conduction at all. The voltage across the JFET's Source resistor collapses and C1 begins to discharge. The '450 tries to adjust the bias by pushing the cascode's Gate higher, but it's not conducting. When the signal comes back, C1 begins to recover, and this brings down the cascode's Gate, but for an instant the Lovoltech is biased too hard.
In other words, you begin to modulate the cascode, but the wrong way. Distortion increases.
Granted, I'm overstating the case. The NPN's base current won't draw down C1's charge that much during a mere half cycle of music, but if it's more than a passing peak, you're likely to see some degradation of performance.
Yes, it's possible to use either a JFET or a MOSFET instead of a bipolar. But at that point, you've lost the simplicity and predictability of the NPN's Vbe and life gets much more complicated.
Why bother? I've got other biasing schemes that will do a better job. The entire point of Isotope One is to build an amp that doesn't need pots and fiddling. The more complicated mess-with-it circuits will come later.
Incidentally, I've only got one RC drawn. If you want, you can use an RCRC (or more) and increase the cutoff slope.
Another note: This is why I regulated Bias A and Bias B to 15V--so that if the NPN did, in fact, slap the bias rail, then it wouldn't blow the MOSFET. If you go for some variation of this circuit that actively sets the bias but runs from a rail higher than about 20V or so, you might want to throw in a Zener to protect the MOSFET.
Joe,
People are certainly welcome to post other front ends and variations on the output stage.
Could you elaborate on how you see the circuit working? I'm sure it's just because I'm sort on sleep, but I don't see the cascode modulation part of it.
Using a current source to drive a resistor in order to set the bias definitely works. I've got one of those in the queue, but if anyone else wants to post one, go ahead.
Note that either the CCS or the resistor will need to be variable in order to set the bias, unless you intend to add more circuitry to automate it (I may or may not do that later--I've got a couple of other things I want to do first).
Grey
I realized that I had said that the biasing circuit was the only real difference between Nelson's basic building block and mine. There's actually another difference--I take the signal from the Source of the JFET, whereas he takes the signal from the Drain of the MOSFET.
jh6you,
Let's take a closer look at the biasing circuit in Isotope One.
--Assume the amp to be at idle, so there's no signal present. The NPN sets its Vbe across the Lovoltech's Source resistor. This results in a current through the load resistor (R5) that holds the Gate of the cascode device at the Lovoltech's Vds + the MOSFET's Vgs. So far, so good. In fact, it's a current source as long as there's no signal at the Lovoltech's Gate.
--Assume that the circuit is in use, but still within class A operating parameters, meaning that there's signal all the time; the stack never turns off. The RC network at the base of the NPN filters out the signal, keeping C1 charged via a small amount of DC drawn from the main bias current. Again, the cascode device's Gate is held at the proper voltage to bias the Lovoltech.
--Assume that the Lovoltech has been driven into shutoff during part of the amplification cycle. No conduction at all. The voltage across the JFET's Source resistor collapses and C1 begins to discharge. The '450 tries to adjust the bias by pushing the cascode's Gate higher, but it's not conducting. When the signal comes back, C1 begins to recover, and this brings down the cascode's Gate, but for an instant the Lovoltech is biased too hard.
In other words, you begin to modulate the cascode, but the wrong way. Distortion increases.
Granted, I'm overstating the case. The NPN's base current won't draw down C1's charge that much during a mere half cycle of music, but if it's more than a passing peak, you're likely to see some degradation of performance.
Yes, it's possible to use either a JFET or a MOSFET instead of a bipolar. But at that point, you've lost the simplicity and predictability of the NPN's Vbe and life gets much more complicated.
Why bother? I've got other biasing schemes that will do a better job. The entire point of Isotope One is to build an amp that doesn't need pots and fiddling. The more complicated mess-with-it circuits will come later.
Incidentally, I've only got one RC drawn. If you want, you can use an RCRC (or more) and increase the cutoff slope.
Another note: This is why I regulated Bias A and Bias B to 15V--so that if the NPN did, in fact, slap the bias rail, then it wouldn't blow the MOSFET. If you go for some variation of this circuit that actively sets the bias but runs from a rail higher than about 20V or so, you might want to throw in a Zener to protect the MOSFET.
Joe,
People are certainly welcome to post other front ends and variations on the output stage.
Could you elaborate on how you see the circuit working? I'm sure it's just because I'm sort on sleep, but I don't see the cascode modulation part of it.
Using a current source to drive a resistor in order to set the bias definitely works. I've got one of those in the queue, but if anyone else wants to post one, go ahead.
Note that either the CCS or the resistor will need to be variable in order to set the bias, unless you intend to add more circuitry to automate it (I may or may not do that later--I've got a couple of other things I want to do first).
Grey
I'm missing something here.
We would seem to lose the 'triode-ness' of the J-fet by using it as a follower (no voltage gain), and then we are using a hex-fet for the voltage gain.
What about using a J-fet for the voltage gain?
http://www.supertex.com/pdf/datasheets/DN2535.pdf
Run it with about 40mA or so, smack in the triode region.
And a 400W big honker J-fet for the follower?
http://www.ixys.com/99192.pdf
Lot easier to mount a TO-247 than the TO-252 (Lovoltech) package.
We would seem to lose the 'triode-ness' of the J-fet by using it as a follower (no voltage gain), and then we are using a hex-fet for the voltage gain.
What about using a J-fet for the voltage gain?
http://www.supertex.com/pdf/datasheets/DN2535.pdf
Run it with about 40mA or so, smack in the triode region.
And a 400W big honker J-fet for the follower?
http://www.ixys.com/99192.pdf
Lot easier to mount a TO-247 than the TO-252 (Lovoltech) package.
GRollins said:
Basically, it's a voltage divider. This may be easier to see if you get rid of the two top 620 ohm resistors, which have a purpose but are not essential here for basic operation. Then increase the remaining 620 ohm bias resistors to 1.2k or whatever value gives the correct bias voltage for the type of MOSFET you are using. Leave the cross-coupling resistor connected across the tops of the two bias resistors.
In actual operation, the swinging output signal produces a voltage imbalance across the cross-coupling resistor. This causes current to be subtracted from the bias resistor on the more positive side of the circuit and added to bias resistor on the more negative side. The result is an attenuated and inverted image of the output signal superimposed on the cascode bias at the gate of each MOSFET.
For a given output signal level, the magnitude of this modulation is a function of current, which in turn is a function of cross-coupling resistance. Higher resistance lead to smaller currents and hence smaller modulations, and vice versa. You would adjust this resistance experimentally for lowest distortion, then perhaps replace it with an equivalent fixed value.
FWIW, if I were prototyping this, I'd use fixed resistors for the bias and make the current sources adjustable over a small range, say 5-7mA. The LM317 CCS (see LM317 datasheet) has worked well for me where there's little or no signal voltage across it, which is the case here.
I hope this is clear - let me know.
djk,
You mentioned JFETs, but the links you posted are both for MOSFETs. If you know of another source for power JFETs, please let us know.
I'm assuming that you mean for the DN2535 to be used in the front end. I don't see that it should be a problem to substitute it directly into the front end circuit in place of the IRF610s. If you try it, let us know how it works out.
I hope to get another front end posted fairly soon. I've got one resistor value I want to take another look at, then I'll have to go through all this file format foolishness again, which takes more time than I can easily justify. After that, I'll stick to back end topologies for a little bit. With a half dozen schematics in hand, people can mix and match according to their wants and needs.
Yes, the IXYS MOSFET could be used instead of the IRF644. I've got handfuls of '644s lying about, so I tend to use them just because they're handy. Once they're gone, I'll get something else.
Now, if your intent is to use the IXYS device to replace both the IRF644 and the LU1014D, that will work too. No one says that you have to cascode the ouput stages in a Circlotron, and as far as I know, no one has ever done so before, at least on a commercial basis. If you want to build a more conventional MOSFET Circlotron, that would be a good starting point. You'd have to scrap the biasing circuit I posted, but it won't be that hard to come up with another one.
There's no question that the TO-247 case is easier to mount than the TO-251, but unfortunately the TO-251 case is what Lovoltech chose to use. Given that Lovoltech is the only company making power JFETs at the moment, we're kinda stuck.
The output of a Circlotron doesn't have any voltage gain. It's a follower. The gain it provides is in the current realm. That's why the front end has to be able to swing all the voltage you will need at the output. In fact, followers lose a small amount of voltage swing, so the front end has to do a little more to make up for that loss.
In our favor is the fact that this topology effectively doubles the voltage swing, which gives us a 6dB or so head start. In a 'normal' single-ended amp, only one side of the front end differential's gain is used. The other side's gain is thrown away...literally. Frequently there's not even a load resistor on the 'back' side. In a design such as this, both sides of the front end differential are used, which gives us a "free" 6dB more gain, all for the price of a single resistor. The Circlotron output topology cheerfully accepts this balanced signal and we're set to go.
Joe,
Again, it may just be that I'm short on sleep (still/again)...but it looks as though you're using negative feedback at the cascode's Gate, whereas Nelson is using positive feedback. If true, then I would anticipate that distortion would actually rise a bit.
Perhaps Nelson will weigh in here and give his opinion.
Like you, I used fixed resistances and a variable CCS to bias the cascode stack. Works like a charm.
Grey
You mentioned JFETs, but the links you posted are both for MOSFETs. If you know of another source for power JFETs, please let us know.
I'm assuming that you mean for the DN2535 to be used in the front end. I don't see that it should be a problem to substitute it directly into the front end circuit in place of the IRF610s. If you try it, let us know how it works out.
I hope to get another front end posted fairly soon. I've got one resistor value I want to take another look at, then I'll have to go through all this file format foolishness again, which takes more time than I can easily justify. After that, I'll stick to back end topologies for a little bit. With a half dozen schematics in hand, people can mix and match according to their wants and needs.
Yes, the IXYS MOSFET could be used instead of the IRF644. I've got handfuls of '644s lying about, so I tend to use them just because they're handy. Once they're gone, I'll get something else.
Now, if your intent is to use the IXYS device to replace both the IRF644 and the LU1014D, that will work too. No one says that you have to cascode the ouput stages in a Circlotron, and as far as I know, no one has ever done so before, at least on a commercial basis. If you want to build a more conventional MOSFET Circlotron, that would be a good starting point. You'd have to scrap the biasing circuit I posted, but it won't be that hard to come up with another one.
There's no question that the TO-247 case is easier to mount than the TO-251, but unfortunately the TO-251 case is what Lovoltech chose to use. Given that Lovoltech is the only company making power JFETs at the moment, we're kinda stuck.
The output of a Circlotron doesn't have any voltage gain. It's a follower. The gain it provides is in the current realm. That's why the front end has to be able to swing all the voltage you will need at the output. In fact, followers lose a small amount of voltage swing, so the front end has to do a little more to make up for that loss.
In our favor is the fact that this topology effectively doubles the voltage swing, which gives us a 6dB or so head start. In a 'normal' single-ended amp, only one side of the front end differential's gain is used. The other side's gain is thrown away...literally. Frequently there's not even a load resistor on the 'back' side. In a design such as this, both sides of the front end differential are used, which gives us a "free" 6dB more gain, all for the price of a single resistor. The Circlotron output topology cheerfully accepts this balanced signal and we're set to go.
Joe,
Again, it may just be that I'm short on sleep (still/again)...but it looks as though you're using negative feedback at the cascode's Gate, whereas Nelson is using positive feedback. If true, then I would anticipate that distortion would actually rise a bit.
Perhaps Nelson will weigh in here and give his opinion.
Like you, I used fixed resistances and a variable CCS to bias the cascode stack. Works like a charm.
Grey
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.