Compromise stability in doing so ?, I really dont think its a good idea or worth the effort.
Who says I would be compromizing stability? The bar there is pretty clear there I guess, >=60deg phase margin, >= 12dB gain margin.
In my opinion 60 is too low, shoot for min 70 and ideally around 75 to 80. At last someone in another thread has posted info why this is what one should design for. As an example, my own CFA which has half the complexity of your VFA design comes were close in matching it in THD if I were to opt for same phase margin.
Since you like complexity you should try the patented circuit I showed in your thread, it would make a very high performance CFA. In the vas include a hawksford cascode to increase loop gain. Ontaba said he could get 80 db loop gain at 80 khz, this is the way to get there but good luck getting it stable with slow outputstage for audio amp.
60 is quite conservative - many control systems are sucessful with 30 degree margin - but you do want to allow for loading and operating point variation
audio amp designers seem to ignore 2-degree of freedom control theory - design for feedback's advantages of desensitivity, distortion reduction with stability limits - then correct forward gain peaking with responsible filtering
in audio amp feedback/compensation most seem to strictly equate step response with "quality"
audio amp designers seem to ignore 2-degree of freedom control theory - design for feedback's advantages of desensitivity, distortion reduction with stability limits - then correct forward gain peaking with responsible filtering
in audio amp feedback/compensation most seem to strictly equate step response with "quality"
And there is an actual issue that plagues both CFA and VFA topologies: The dreaded slow output stage
I don't think there are extremely high Ft lateral FETs around. Perhaps better chance with the verticals, but that comes with its complex temp compensation scheme.
I can show a little stripped version, but there is nothing special about it, its typical application, just refined. If you look at Cyrus 3 schematic and add beta enhanced vas, thats it. You shoud be able to better 100 db THD20 alone with it just using shunt comp. I get 6 ppm THD20 on 100W using cordell models. I add feedforward, but that Im not at liberty to show. Actually its better to call it distortion reduction by cancelation and is more in line with ideas such as R Marsh has shown in the forum. If you look at member mcd99uk s thread youll see a very similar amp although he adds error correction. Edmond uses CMCL and DTMC but I havent experimented wit this yet and maybe there is more to be gained.
If anyone reads my thread you'll see the evolution of the design. It's hit a problem which looks like a grounding issue at present possibly due to the high current through the feedback resistors. But I could be wrong.
I started with error correction but then adopted Edmond's DTMC which allowed the design to equal the error corrected THD but without the error correction which I changed to a plain DC servo.
I started with error correction but then adopted Edmond's DTMC which allowed the design to equal the error corrected THD but without the error correction which I changed to a plain DC servo.
Well its very linear too, its just not workable with todays output devices, works good with small signal, AD use it some of their opamps.
Is this one you mean, current input, current output? or CMA?
Attachments
Hi Guys
Wrenchclone, your schematic in post-119 needs buffers between the VAS and mosfet gates. Lots of capacitance there to slew and slow things down. I suspect that is the slew-rate bottle neck in your circuit.
Bob, I've never built an amp that needed a servo to set the DC offset. To me, servos seem like a "bandaid" on either a poor design or to accommodate a poor decision. Why do you need DC response? You can't hear zero hertz and your speakers will die from it. The servo craze peaked back in the 1980s with all those DC-to-greenlight amps from Japan - great specs, but "DC"?
I've built a few amps using an input section like your mosfet amp, except with BJTs instead of jfets right at the input. No servo. You state "servo not shown" in a later form of that schematic in your book.
I like gain that rolls off DC and have no problems using a high-value cap to achieve this. I've built CFAs with AC coupled feedback and had no DC offset issues nor need for adjustments even without matched devices. When everything is matched, there is no current through the shunt element of the feedback loop anyway.
A point I make regularly is that every opamp output and DC amplifier output should have a resistive tether to ground. Adding one simple inexpensive part helps the circuit establish low DC offset much more readily. App notes tend to be stripped of nearly everything required to make them work in the real world, and then the legacy of poor design grows...
Andrew's nx and sx amps work pretty well and follow the basic simplicity RNM desires while attaining pretty good performance.
The LC circuit may not be the best reference for learning about CFA. To get really good THD even with matched parts, requires a two stage circuit, such as Andrew's and Wrenchclone's. As an aside, the number of stages in CFA may be couned differently than for VFA - not sure. An ambiguity occurs in Walter Kester's "High Speed Operational Amplifiers" treatise, written for Analog Devices, where both the "basic CFA" and "2-stage CFA" look the same except the latter shows the three compensation caps.
Most claims of "good sound" from builders merely reflect the satisfaction of building anything that works first, and does not sound horrible right out of the gate. The first amp I ever built only had a response to 37kHz, but it was a learning experience mostly in how to make such measurements with a scope. That thing never passed being a hair ball of tacked together parts, but it worked. The only amp I built that blew up was a 100W app from Motorola. The latter actually made sound before it blew up, which was not bad, so satisfying in its novelty; never heard the hair ball.
When I mentioned the use of CFPs previously, it was for the input conversion circuit as a possible means to reduce the impedance error at the feedback node.
An alternate input circuit - which may be cast aside by y'all without comment - is one I built around 1990. It uses comp-diffs driving comp EFs in a unity buffer arrangement. I used 5-pin SIP diff and CM packages, so wiring was dead simple. The goal was just to have a very wide bandwidth buffer. This would be most akin to Doug Self's (is his name okay on this thread? hehe) discrete opamp buffer, shown in the Small Signal Audio Design book, achieving 4ppm THD up to 10kHz (fig.3-17e). His uses a single diff, CM, single EF and current sources, with usual Cdom.
With whatever input stage becomes "acceptable", I think it is worth looking into bootstrapping it so that Early effects are minimised. It depends on whether such a method saves parts or has better effects than cascoding everything in sight. The above buffer circuit example has the advantage of <2mV offset between input and output, so as a CFA input stage, the NFB node would be pretty low-z as another loop monitors it.
I am happy to escape miller compensation and CFA easily does that. Even if required for high-frequency stability, it does not come into play until well above the hifi band.
Have fun
Kevin O'Connor
Wrenchclone, your schematic in post-119 needs buffers between the VAS and mosfet gates. Lots of capacitance there to slew and slow things down. I suspect that is the slew-rate bottle neck in your circuit.
Bob, I've never built an amp that needed a servo to set the DC offset. To me, servos seem like a "bandaid" on either a poor design or to accommodate a poor decision. Why do you need DC response? You can't hear zero hertz and your speakers will die from it. The servo craze peaked back in the 1980s with all those DC-to-greenlight amps from Japan - great specs, but "DC"?
I've built a few amps using an input section like your mosfet amp, except with BJTs instead of jfets right at the input. No servo. You state "servo not shown" in a later form of that schematic in your book.
I like gain that rolls off DC and have no problems using a high-value cap to achieve this. I've built CFAs with AC coupled feedback and had no DC offset issues nor need for adjustments even without matched devices. When everything is matched, there is no current through the shunt element of the feedback loop anyway.
A point I make regularly is that every opamp output and DC amplifier output should have a resistive tether to ground. Adding one simple inexpensive part helps the circuit establish low DC offset much more readily. App notes tend to be stripped of nearly everything required to make them work in the real world, and then the legacy of poor design grows...
Andrew's nx and sx amps work pretty well and follow the basic simplicity RNM desires while attaining pretty good performance.
The LC circuit may not be the best reference for learning about CFA. To get really good THD even with matched parts, requires a two stage circuit, such as Andrew's and Wrenchclone's. As an aside, the number of stages in CFA may be couned differently than for VFA - not sure. An ambiguity occurs in Walter Kester's "High Speed Operational Amplifiers" treatise, written for Analog Devices, where both the "basic CFA" and "2-stage CFA" look the same except the latter shows the three compensation caps.
Most claims of "good sound" from builders merely reflect the satisfaction of building anything that works first, and does not sound horrible right out of the gate. The first amp I ever built only had a response to 37kHz, but it was a learning experience mostly in how to make such measurements with a scope. That thing never passed being a hair ball of tacked together parts, but it worked. The only amp I built that blew up was a 100W app from Motorola. The latter actually made sound before it blew up, which was not bad, so satisfying in its novelty; never heard the hair ball.
When I mentioned the use of CFPs previously, it was for the input conversion circuit as a possible means to reduce the impedance error at the feedback node.
An alternate input circuit - which may be cast aside by y'all without comment - is one I built around 1990. It uses comp-diffs driving comp EFs in a unity buffer arrangement. I used 5-pin SIP diff and CM packages, so wiring was dead simple. The goal was just to have a very wide bandwidth buffer. This would be most akin to Doug Self's (is his name okay on this thread? hehe) discrete opamp buffer, shown in the Small Signal Audio Design book, achieving 4ppm THD up to 10kHz (fig.3-17e). His uses a single diff, CM, single EF and current sources, with usual Cdom.
With whatever input stage becomes "acceptable", I think it is worth looking into bootstrapping it so that Early effects are minimised. It depends on whether such a method saves parts or has better effects than cascoding everything in sight. The above buffer circuit example has the advantage of <2mV offset between input and output, so as a CFA input stage, the NFB node would be pretty low-z as another loop monitors it.
I am happy to escape miller compensation and CFA easily does that. Even if required for high-frequency stability, it does not come into play until well above the hifi band.
Have fun
Kevin O'Connor
Last edited:
Hi Guys
Jfets have no direct relationship between gate current and channel current - at least not a reliable one. Since CFA is all about "current steering", it is much more straightforward to use BJTs. BJTs also have a reliable Vbe voltage and emitter resistance. Matching jfets is much more difficult than matching BJTs. Worse is finding compliments let alone matching them to n-channels.
One point that may help with performance is simply to operate the devices at high-ish currents. This gets most BJTs into their high GBW range, although small signal devices may already have enough gain and bandwidth even at low currents. Higher currents also keep the small signal stages in class-A, as Andrew did in his very nice sx and nx amps.
To keep internal device capacitances low from a design perspective, each device should have a "significant" voltage across it. For many devices, 10V is enough; for others much more is needed. So, this might require a hit in overall efficiency of the output stage if one wishes to achieve high slew rates if even for simple specsmanship.
A benefit to increased supply rails is increased energy storage in the main filters. This will certainly help transient response.
Have fun
Kevin O'Connor
Jfets have no direct relationship between gate current and channel current - at least not a reliable one. Since CFA is all about "current steering", it is much more straightforward to use BJTs. BJTs also have a reliable Vbe voltage and emitter resistance. Matching jfets is much more difficult than matching BJTs. Worse is finding compliments let alone matching them to n-channels.
One point that may help with performance is simply to operate the devices at high-ish currents. This gets most BJTs into their high GBW range, although small signal devices may already have enough gain and bandwidth even at low currents. Higher currents also keep the small signal stages in class-A, as Andrew did in his very nice sx and nx amps.
To keep internal device capacitances low from a design perspective, each device should have a "significant" voltage across it. For many devices, 10V is enough; for others much more is needed. So, this might require a hit in overall efficiency of the output stage if one wishes to achieve high slew rates if even for simple specsmanship.
A benefit to increased supply rails is increased energy storage in the main filters. This will certainly help transient response.
Have fun
Kevin O'Connor
Last edited:
It is 80dB @20k, not @80k. And 80dB is not the limit.
It reaching 90dB with bjt input, but I like the sound with jfet.
The output is an unique paralleled irf 510, not the slow one, because this single hexfet also could do 10W+ @60MHz in classE amplifier.
My own experiments with this circuit :
Distorsions peu connues des étages d'entrée des amplis
An externally hosted image should be here but it was not working when we last tested it.
Other posts show the impressive harmonic distorsion numbers and the benefit obtained by a 3 transistors Wilson current mirror.
An externally hosted image should be here but it was not working when we last tested it.
Last edited: