Ah,
Metropolis..... A wonderful movie! I loved Giorgio Moroder's soundtrack too.
Notice the knife switch in the background in the shape of a crucifix - now there's an image of modern life for you!
Graham, this is the beginning of a very fine SE amplifier, with low order distortion only, primarily H2. It will sound sensational, and if you visit me again soon I'll play one for you. You thought the AKSA was good; wait until you hear this one! Here's a few points you might like to ponder.
1. If you use the bipolar/mosfet hybrid in CFP your Zout will drop to below 50 milliohms. This means - you guessed it - gate drive problems are all but eliminated, and you need no global feedback! You can then even consider a mouse power tube to perform the voltage amplification - a lofty choice, I can assure you!
2. A bipolar transistor CCS in the +ve rail will give you superior performance, in my view, since the drain/collector impedance will not be affected by the protection zener you have from gate to drain. Further, the supply voltage can be lowered considerably, obviating two separate power supplies (although the voltage amp mosfet circuit will need to be thoroughly decoupled from the output device supply, of course.)
3. A CFP output stage requires no Zobel for stability, since there is no GFB and tolerance to reactive loads is nothing short of monumental. This amp will drive an ESL63 speaker very well.
4. Because the tempco of the mosfet is self-limiting AND its transconductance is limited, you can get away with very small DC resistance in the drain of a bipolar/mosfet CFP hybrid for current control. This means you can use the DCR of the choke to good advantage; as little as 300 milliohms is sufficient. In my experience an IC servo control here is quite unnecessary; I use a simple LED voltage reference. While this will give you about a volt of output offset for 3.3A of standing current, it will open up the possibility of a bipolar power supply, with the cold end of the choke feeding a negative 1V supply, thus eliminating the output cap. In practice, this means judiciously center-tapping the present single AC supply.
I think your design direction is very impressive!!
Cheers,
Hugh R. Dean
aksaonline.com
Metropolis..... A wonderful movie! I loved Giorgio Moroder's soundtrack too.
Notice the knife switch in the background in the shape of a crucifix - now there's an image of modern life for you!
Graham, this is the beginning of a very fine SE amplifier, with low order distortion only, primarily H2. It will sound sensational, and if you visit me again soon I'll play one for you. You thought the AKSA was good; wait until you hear this one! Here's a few points you might like to ponder.
1. If you use the bipolar/mosfet hybrid in CFP your Zout will drop to below 50 milliohms. This means - you guessed it - gate drive problems are all but eliminated, and you need no global feedback! You can then even consider a mouse power tube to perform the voltage amplification - a lofty choice, I can assure you!
2. A bipolar transistor CCS in the +ve rail will give you superior performance, in my view, since the drain/collector impedance will not be affected by the protection zener you have from gate to drain. Further, the supply voltage can be lowered considerably, obviating two separate power supplies (although the voltage amp mosfet circuit will need to be thoroughly decoupled from the output device supply, of course.)
3. A CFP output stage requires no Zobel for stability, since there is no GFB and tolerance to reactive loads is nothing short of monumental. This amp will drive an ESL63 speaker very well.
4. Because the tempco of the mosfet is self-limiting AND its transconductance is limited, you can get away with very small DC resistance in the drain of a bipolar/mosfet CFP hybrid for current control. This means you can use the DCR of the choke to good advantage; as little as 300 milliohms is sufficient. In my experience an IC servo control here is quite unnecessary; I use a simple LED voltage reference. While this will give you about a volt of output offset for 3.3A of standing current, it will open up the possibility of a bipolar power supply, with the cold end of the choke feeding a negative 1V supply, thus eliminating the output cap. In practice, this means judiciously center-tapping the present single AC supply.
I think your design direction is very impressive!!
Cheers,
Hugh R. Dean
aksaonline.com
Great photo - thanks for enlarging upon it (in both senses).
"...if I short the 100k in series with the signal, the distortion improves quite a bit. What is happenening here? "
The thing I noticed immediately about this circuit is that it looks like it derives from a valve circuit. The simple single-ended design, the big inductor, the use of 100k+ bias resistors and so on. And this is fine as long as compensation is made for the differences between FETs and valves. To my mind FETs are far more closely related to BJT than valves and come with a similar but different plethora of non-linearites.
All other things being equal (grounding, psu, etc in good shape) then I'd hazzard a guess that the distortion effect you observe is due to variation of *both* Cgs and Cgd, especially Cgd if you have very high voltage gain in the driver. Both capacitances vary quite a lot with voltage/charge. This is the ugly face of the FET. You may notice in BJT V gain stages that an auxilliary cap (say 20-50pF) is added from C to B and this is to swamp the BJTs variable capacitance. In general, BJTs have significantly lower parasitic capactiances than similarly rated FETs.
My advice would be to find out the nominal and likely variation of capacitance values for your FET and then check the circuit still works as you want it to.
"...if I short the 100k in series with the signal, the distortion improves quite a bit. What is happenening here? "
The thing I noticed immediately about this circuit is that it looks like it derives from a valve circuit. The simple single-ended design, the big inductor, the use of 100k+ bias resistors and so on. And this is fine as long as compensation is made for the differences between FETs and valves. To my mind FETs are far more closely related to BJT than valves and come with a similar but different plethora of non-linearites.
All other things being equal (grounding, psu, etc in good shape) then I'd hazzard a guess that the distortion effect you observe is due to variation of *both* Cgs and Cgd, especially Cgd if you have very high voltage gain in the driver. Both capacitances vary quite a lot with voltage/charge. This is the ugly face of the FET. You may notice in BJT V gain stages that an auxilliary cap (say 20-50pF) is added from C to B and this is to swamp the BJTs variable capacitance. In general, BJTs have significantly lower parasitic capactiances than similarly rated FETs.
My advice would be to find out the nominal and likely variation of capacitance values for your FET and then check the circuit still works as you want it to.
Traderbam,
I built one with dual IRF mosfets top and bottom, using a current source as shown, and measured it into an 8R resistive load! Actually, it was 38 milliohms with a total stage current of 3A. I'd imagine with a single device it would be around 50 milliohms.
This is 1 impressive circuit......
Cheers,
Hugh
aksaonline.com
I built one with dual IRF mosfets top and bottom, using a current source as shown, and measured it into an 8R resistive load! Actually, it was 38 milliohms with a total stage current of 3A. I'd imagine with a single device it would be around 50 milliohms.
This is 1 impressive circuit......
Cheers,
Hugh
aksaonline.com
resistance is futile
If you have made this circuit and measured it then that must be right. But my estimation for dc output resistance disagrees. It goes like this:
Let's say the CFP has a BJT with beta of 100, the GS resistor is 500-ohms and the Gm of the FET is 10. Educated guesses. Then this makes the beta of the CFB = 100*500*10 = 500k.
Now the driver stage has a collector loading of 100k. So the CFB is effectively fed from a 100k source. The output resistance is approximately the input resistance divided by the output stage beta = 100k/500k = 0.2 ohms. (This is slightly too low because the gm of the BJT hasn't been taken into account).
In the existing circuit (with no CFP) the output resistance is determined only by the Gm of the output FET, which, using my earlier estimate, would be 1/10 = 0.1 ohms.
So adding the CFP atleast doubles the dc output resistance.
Where have I gone wrong?
BAM
If you have made this circuit and measured it then that must be right. But my estimation for dc output resistance disagrees. It goes like this:
Let's say the CFP has a BJT with beta of 100, the GS resistor is 500-ohms and the Gm of the FET is 10. Educated guesses. Then this makes the beta of the CFB = 100*500*10 = 500k.
Now the driver stage has a collector loading of 100k. So the CFB is effectively fed from a 100k source. The output resistance is approximately the input resistance divided by the output stage beta = 100k/500k = 0.2 ohms. (This is slightly too low because the gm of the BJT hasn't been taken into account).
In the existing circuit (with no CFP) the output resistance is determined only by the Gm of the output FET, which, using my earlier estimate, would be 1/10 = 0.1 ohms.
So adding the CFP atleast doubles the dc output resistance.
Where have I gone wrong?
BAM
Off topic? Nonsense!
Metropolis is alway on-topic with me! I am the proud owner of a properly purchased Giorgio Moroder video, a CD soundtrack of same, also a video of a version with the crookest piano soundtrack you ever heard, a paperback screenplay, and text version of Frau von Harbou's original novel. I wish I could live in that movie sometimes...
GP.
moses said:
Metropolis is alway on-topic with me!
I am the proud owner of a properly purchased Giorgio Moroder video, a CD soundtrack of same, also a video of a version with the crookest piano soundtrack you ever heard, a paperback screenplay, and text version of Frau von Harbou's original novel. I wish I could live in that movie sometimes...GP.
Attachments
3-way CFP?
Hi Hugh.
Sorry I hadn't replied to your posts before. Yeah, I am interested in trying a CFP but as I want to keep an N-channel fet I might have to do a 3 stage. I'll just stick a lot of emitter degeneration in the second stage to try and keep the gain down to a reasonable level. Might even revisit a few ideas from the CDA amp. e.g. the error signal multiplication transformer cct, and/or the error feedforward cct using an opamp. They both worked more or less ok at the time but were not quite suitable for what I was doing then. BTW, who performs the "Post Modern Blues" track on that CD?
Hi traderbam
What I am also going to do is try a unity gain opamp buffer ahead of the nfb resistors and lower these to about 3k3 and 33k. I hope the much lower cct impedance will then make any fet capacitance variation unnoticeable. How much trouble wil I get into off you guys for using an opamp?
GP.
Hi Hugh.
Sorry I hadn't replied to your posts before. Yeah, I am interested in trying a CFP but as I want to keep an N-channel fet I might have to do a 3 stage. I'll just stick a lot of emitter degeneration in the second stage to try and keep the gain down to a reasonable level. Might even revisit a few ideas from the CDA amp. e.g. the error signal multiplication transformer cct, and/or the error feedforward cct using an opamp. They both worked more or less ok at the time but were not quite suitable for what I was doing then. BTW, who performs the "Post Modern Blues" track on that CD?
Hi traderbam
What I am also going to do is try a unity gain opamp buffer ahead of the nfb resistors and lower these to about 3k3 and 33k. I hope the much lower cct impedance will then make any fet capacitance variation unnoticeable. How much trouble wil I get into off you guys for using an opamp?
GP.
Aksa,
I'm trying to say that the CFP circuit as drawn doesn't really have an output impedance of its own...rather it has a high beta. Like a very high gain BJT. This beta is 500k (according to my educated guesses). So the output Z of the amp using the CFP depends on the Z that the input of the CFP sees. In Circlotron's circuit this is 100k-ohms. So the output Z is 100k/500k=200mohms. This is twice as high as the simple FET follower (according to my educated guesses).
In other words the driver stage output Z is critical to calculating the output stage output Z. In this case the output stage does not define the output Z on its own.
I recognize that the CFP has the benefit of reducing the influence of the FETs Cgs and Cgd and to linearize the gm.
I would be careful about omitting the Zobel network - the CFP is a feedback system in itself with a minimum 90deg OL phase lag because of the FET. If a pure capacitive load is driven this will add another 90deg. The Zobel will help to mitigate the phase shift. Especially beneficial when loop feedback is used.
I'm trying to say that the CFP circuit as drawn doesn't really have an output impedance of its own...rather it has a high beta. Like a very high gain BJT. This beta is 500k (according to my educated guesses). So the output Z of the amp using the CFP depends on the Z that the input of the CFP sees. In Circlotron's circuit this is 100k-ohms. So the output Z is 100k/500k=200mohms. This is twice as high as the simple FET follower (according to my educated guesses).
In other words the driver stage output Z is critical to calculating the output stage output Z. In this case the output stage does not define the output Z on its own.
I recognize that the CFP has the benefit of reducing the influence of the FETs Cgs and Cgd and to linearize the gm.
I would be careful about omitting the Zobel network - the CFP is a feedback system in itself with a minimum 90deg OL phase lag because of the FET. If a pure capacitive load is driven this will add another 90deg. The Zobel will help to mitigate the phase shift. Especially beneficial when loop feedback is used.
Hi Bam,
OK, I can't argue with your theory; seems sound to me.
However, I have built it, worked it over, tweaked it, and listened to it - for about five years now, and it's very good, works brilliantly.
To answer your points:
>I'm trying to say that the CFP circuit as drawn doesn't really have an output impedance of its own...rather it has a high beta.
If you select a stage current, a drive circuit, and a load, then surely it must have a Zout. At 3A, two paralleled mosfets top and bottom, 50V supply, and 8R load, I measured better than 40 milliohms.
The circuit block as shown has a beta, or voltage gain, of one. The driver has high gain, of course, determined by the combination of it's collector load and the transconductance of the mosfet. While it makes sense to figure the Zout from the source impedance of the driver and the current magnification, I measured the Zout by reading off the output voltage unloaded versus loaded, and using the familiar (1 - loaded/unloaded) x 8R.
I seem to remember a 180K bias resistor on the base of the CFP driver, supplying roughly half Vcc, and a source impedance (from a tube, in fact) of around 15K.
>Like a very high gain BJT. This beta is 500k (according to my educated guesses). So the output Z of the amp using the CFP depends on the Z that the input of the CFP sees. In Circlotron's circuit this is 100k-ohms. So the output Z is 100k/500k=200mohms. This is twice as high as the simple FET follower (according to my educated guesses).
In my circuit, this same calculation is 15.2K/500K, which is 30.4 milliohms, close to the 38 milliohm measured. Pretty close.
>In other words the driver stage output Z is critical to calculating the output stage output Z. In this case the output stage does not define the output Z on its own.
Ah, I see now! Yes, of course, I was relating a portion of my amp design, not giving it all away!!! Moreover, I have not revealed how I interface the tube to the CFP; there is a further trick here which I cannot talk about.
>I recognize that the CFP has the benefit of reducing the influence of the FETs Cgs and Cgd and to linearize the gm.
I would be careful about omitting the Zobel network - the CFP is a feedback system in itself with a minimum 90deg OL phase lag because of the FET. If a pure capacitive load is driven this will add another 90deg. The Zobel will help to mitigate the phase shift. Especially beneficial when loop feedback is used.
Yes, Bam, perhaps, but I found that a Zobel slightly degraded the sound and yet the amp is quite stable into an ESL63, which it has been happily driving now for about three years!
Thank you for your continuing interest.
Cheers,
Hugh
www.aksaonline.com
OK, I can't argue with your theory; seems sound to me.
However, I have built it, worked it over, tweaked it, and listened to it - for about five years now, and it's very good, works brilliantly.
To answer your points:
>I'm trying to say that the CFP circuit as drawn doesn't really have an output impedance of its own...rather it has a high beta.
If you select a stage current, a drive circuit, and a load, then surely it must have a Zout. At 3A, two paralleled mosfets top and bottom, 50V supply, and 8R load, I measured better than 40 milliohms.
The circuit block as shown has a beta, or voltage gain, of one. The driver has high gain, of course, determined by the combination of it's collector load and the transconductance of the mosfet. While it makes sense to figure the Zout from the source impedance of the driver and the current magnification, I measured the Zout by reading off the output voltage unloaded versus loaded, and using the familiar (1 - loaded/unloaded) x 8R.
I seem to remember a 180K bias resistor on the base of the CFP driver, supplying roughly half Vcc, and a source impedance (from a tube, in fact) of around 15K.
>Like a very high gain BJT. This beta is 500k (according to my educated guesses). So the output Z of the amp using the CFP depends on the Z that the input of the CFP sees. In Circlotron's circuit this is 100k-ohms. So the output Z is 100k/500k=200mohms. This is twice as high as the simple FET follower (according to my educated guesses).
In my circuit, this same calculation is 15.2K/500K, which is 30.4 milliohms, close to the 38 milliohm measured. Pretty close.
>In other words the driver stage output Z is critical to calculating the output stage output Z. In this case the output stage does not define the output Z on its own.
Ah, I see now! Yes, of course, I was relating a portion of my amp design, not giving it all away!!! Moreover, I have not revealed how I interface the tube to the CFP; there is a further trick here which I cannot talk about.
>I recognize that the CFP has the benefit of reducing the influence of the FETs Cgs and Cgd and to linearize the gm.
I would be careful about omitting the Zobel network - the CFP is a feedback system in itself with a minimum 90deg OL phase lag because of the FET. If a pure capacitive load is driven this will add another 90deg. The Zobel will help to mitigate the phase shift. Especially beneficial when loop feedback is used.
Yes, Bam, perhaps, but I found that a Zobel slightly degraded the sound and yet the amp is quite stable into an ESL63, which it has been happily driving now for about three years!
Thank you for your continuing interest.
Cheers,
Hugh
www.aksaonline.com
Circlotron shows some backbone; laughs at peer pressure.
Well I put the opamp buffer in as I said I would and it fixed up the wobbly virtual earth node that was indeed getting pushed around by external things, probably by the varying fet capacitance. I replaced the 100k & 1M with a 3k3 and 33k as I said. Used half a Texas LTC2272 because I had a few. Will try and post a cct about 12 hours from now. With the source resistor bypassed and bootstrapping applied that little non-audio fet has a gain of almost exactly 3000 Close the loop from the speaker output and it drops to almost exactly 10. No oscillation problems but when I turn it off, as the rails are going down it makes a single blurt noise. Is it trying to tell me something?
Now that I'm satisfied that the cct likes to be driven with a low Z I've got no problem using an emitter follower buffer instead of an opamp. It has to swing +/- 2.7 volts into 3k3. Will anyone out there encourage my laziness by suggesting a cct?
GP.
PS. thanks heaps for the interest, guys / blokes / dudes.
PPS the opamp sounded just fine, but I was having imagination attacks because everyone tells me they are just no good.
GGP.
Well I put the opamp buffer in as I said I would and it fixed up the wobbly virtual earth node that was indeed getting pushed around by external things, probably by the varying fet capacitance. I replaced the 100k & 1M with a 3k3 and 33k as I said. Used half a Texas LTC2272 because I had a few. Will try and post a cct about 12 hours from now. With the source resistor bypassed and bootstrapping applied that little non-audio fet has a gain of almost exactly 3000
Close the loop from the speaker output and it drops to almost exactly 10. No oscillation problems but when I turn it off, as the rails are going down it makes a single blurt noise. Is it trying to tell me something?Now that I'm satisfied that the cct likes to be driven with a low Z I've got no problem using an emitter follower buffer instead of an opamp. It has to swing +/- 2.7 volts into 3k3. Will anyone out there encourage my laziness by suggesting a cct?
GP.
PS. thanks heaps for the interest, guys / blokes / dudes.
PPS the opamp sounded just fine, but I was having imagination attacks because everyone tells me they are just no good.
GGP.
Re: Circlotron shows some backbone; laughs at peer pressure.
Tell you what, close your eyes (very important) and just listen. What do you hear?
When I do it myself, I don't hear my overkill (according to some, the overkill). I hear purely the signals coming from my Ortofon FF15 + TT JVC direct drive, 70's, which are surprisingly good for being one the cheapest pickup available (25 USD). I'm talking about my phono amp.
It's important not to see the gear.
Circlotron said:
Tell you what, close your eyes (very important) and just listen. What do you hear?
When I do it myself, I don't hear my overkill (according to some, the overkill). I hear purely the signals coming from my Ortofon FF15 + TT JVC direct drive, 70's, which are surprisingly good for being one the cheapest pickup available (25 USD). I'm talking about my phono amp.
It's important not to see the gear.
This is it now.
I really have to get some better speakers for testing. What I am using are a pair of Plessey C100 10" things in a single 100 litre sealed box, with a pair of ~40mm cone tweeters from some old Philips tv just hanging in mid air on their wires.
The ones I have inside the house you just don't want to know about ^2.
GP.
I really have to get some better speakers for testing. What I am using are a pair of Plessey C100 10" things in a single 100 litre sealed box, with a pair of ~40mm cone tweeters from some old Philips tv just hanging in mid air on their wires.
The ones I have inside the house you just don't want to know about ^2.GP.
Attachments
Mother of all constant current circuits.
During the last week on and off I had been messing around with various constant current circuits, both mina and other peoples. The purpose is to use it as a drain / collector load instead of a resistor because it can act like a super high value resistor in that when the signal voltage across it changes the current through it hardly changes at all. Therefore, the transistor or fet attached to it gives heaps of gain and very good linearity. Anyway, here is something I threw together late this afternoon, and wow does it go! It has a slope resistance of greater than 400 megohms. Here's how I measured it:
I got a zero to 60vdc supply and attached it to the cct and put a 100R resistor in series, and put a decent meter across the resistor. Cranked up the voltage and it starts to work at a little over 10v. The voltage drop across the resistor (i.e. the current) stayed constant from 20v to 60v. What do I mean by constant?? I cant remember the exact figure but it was about 51 mA so say there was 5.12345 volts across the 100R resistor. At 20v the reading dithered from 5.12345 to 5.12346. At 60v the reading *still* dithered from 5.12345 to 5.12346 so you see it changed LESS THAN 10 microvolts across 100 ohms, which means less than 0.1 microamps even though the applied voltage changed by 40 volts. R=V/I = 40/0.0000001= 400,000,000 ohms mimimum! I'm satisfied with that.
Looking at the cct a little further, if you ran a resistor from the fet drain to supply rail, and replaced the LM317 "output" resistor with a collector-emitter connected across it you would have an outrageous cascode stage would you not? The output would be pure current and soooo linear. What's more, if you had a current source instead of the top load resistor, the stage gain would be nothing short of ridiculous raised to the power of a googolplex. If nfb was applied to bring the gain down to a sane level, the distortion should be improved by a similar amount. Overkill to the max.
GP.
During the last week on and off I had been messing around with various constant current circuits, both mina and other peoples. The purpose is to use it as a drain / collector load instead of a resistor because it can act like a super high value resistor in that when the signal voltage across it changes the current through it hardly changes at all. Therefore, the transistor or fet attached to it gives heaps of gain and very good linearity. Anyway, here is something I threw together late this afternoon, and wow does it go!
It has a slope resistance of greater than 400 megohms. Here's how I measured it:I got a zero to 60vdc supply and attached it to the cct and put a 100R resistor in series, and put a decent meter across the resistor. Cranked up the voltage and it starts to work at a little over 10v. The voltage drop across the resistor (i.e. the current) stayed constant from 20v to 60v. What do I mean by constant?? I cant remember the exact figure but it was about 51 mA so say there was 5.12345 volts across the 100R resistor. At 20v the reading dithered from 5.12345 to 5.12346. At 60v the reading *still* dithered from 5.12345 to 5.12346 so you see it changed LESS THAN 10 microvolts across 100 ohms, which means less than 0.1 microamps even though the applied voltage changed by 40 volts. R=V/I = 40/0.0000001= 400,000,000 ohms mimimum!
I'm satisfied with that.Looking at the cct a little further, if you ran a resistor from the fet drain to supply rail, and replaced the LM317 "output" resistor with a collector-emitter connected across it you would have an outrageous cascode stage would you not? The output would be pure current and soooo linear. What's more, if you had a current source instead of the top load resistor, the stage gain would be nothing short of ridiculous raised to the power of a googolplex. If nfb was applied to bring the gain down to a sane level, the distortion should be improved by a similar amount. Overkill to the max.
GP.
Attachments
CFP with N channel MOSFETS??
Ok, I have a MOSFET follower amp.. with a +12 -12 split supply. lower mosfet configured as a CCS. upper mosfet gain of 1.
I have always had trouble with gate capacitances and driving them adequately.
how does one use the CFP stage with N channel mosfets?
Thanks!!
Richard
Ok, I have a MOSFET follower amp.. with a +12 -12 split supply. lower mosfet configured as a CCS. upper mosfet gain of 1.
I have always had trouble with gate capacitances and driving them adequately.
how does one use the CFP stage with N channel mosfets?
Thanks!!
Richard
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.