"A prime reason was that the "CFA" properties made it more convenient and easier for the customer to stay close to optimal compensation as the gain was altered to suit different circuits. "
It may be a bit more nuanced than that (just saying). In a VFA you have to trade gain for BW. In the CFA, provided certain caveats are met, you don't have to do that because it does not have to be dominant pole compensated (fewer gain stages and fewer poles)
However, if you sim it and comp for high PM at ULGF the gain BW advantage diminishes. At 90 degrees PM a CFA may have an octave more CL ULGF than a comparable VFA, but it it's no longer gain BW independent. If you are prepared to go to 30 degrees PM, then they are clearly independent. So in a CFA you have to sacrifice PM to get gain BW independence.
The upshot is that for practical power amps, you probably won't get this feature because you need decent PM's. For an IC driving a well defined load, CFA is perfect solution . . .
It may be a bit more nuanced than that (just saying). In a VFA you have to trade gain for BW. In the CFA, provided certain caveats are met, you don't have to do that because it does not have to be dominant pole compensated (fewer gain stages and fewer poles)
However, if you sim it and comp for high PM at ULGF the gain BW advantage diminishes. At 90 degrees PM a CFA may have an octave more CL ULGF than a comparable VFA, but it it's no longer gain BW independent. If you are prepared to go to 30 degrees PM, then they are clearly independent. So in a CFA you have to sacrifice PM to get gain BW independence.
The upshot is that for practical power amps, you probably won't get this feature because you need decent PM's. For an IC driving a well defined load, CFA is perfect solution . . .
what "high frequency"?
THD above "conventional audio frequency" ~ 20 kHz wouldn't appear to explain much of anything "audibile"
as long as the ultrasonic distortion components are low enough not to generate IMD in the air or ear - not so hard even if we accept "only" -80 dB THD 20k
while multiple high frequency tone's difference IMD products folding down to our sensitive hearing regions may be audible, not masked by fundamentals
high feedback excess loop gain can be >60 dB to over 20 kHz with 2-pole compensations and still use ~2 Mhz loop gain intercept with >60 degree phase margin
and the IMD products are reduced by the feedback at the difference frequency, ie easily > 60 dB below 20 kHz - not limited by the falling loop gain beyond 20 kHz that may not squash their ultrasonic THD or sum IMD to as low
THD above "conventional audio frequency" ~ 20 kHz wouldn't appear to explain much of anything "audibile"
as long as the ultrasonic distortion components are low enough not to generate IMD in the air or ear - not so hard even if we accept "only" -80 dB THD 20k
while multiple high frequency tone's difference IMD products folding down to our sensitive hearing regions may be audible, not masked by fundamentals
high feedback excess loop gain can be >60 dB to over 20 kHz with 2-pole compensations and still use ~2 Mhz loop gain intercept with >60 degree phase margin
and the IMD products are reduced by the feedback at the difference frequency, ie easily > 60 dB below 20 kHz - not limited by the falling loop gain beyond 20 kHz that may not squash their ultrasonic THD or sum IMD to as low
Last edited:
The important properties for the servo would be fast recovery, no phase inversion nasties, unity gain stable, hi z IP. The very low cut off limits it's contribution to amplifier sound - just my input on this.
One trick you can use to even further minimize the contribution from the servo is to dial the initial offset out with a pot and then connect the servo it you've got it right, the servo error output will be zero ( or very close). Then the servo just corrects for drift and perhaps initial offset, and you get maximum dynamic range for the servo.
One trick you can use to even further minimize the contribution from the servo is to dial the initial offset out with a pot and then connect the servo it you've got it right, the servo error output will be zero ( or very close). Then the servo just corrects for drift and perhaps initial offset, and you get maximum dynamic range for the servo.
Hi Richard,
Thanks for putting up that JFET distortion plot. Your example of offset trim for zero and minimal THD point not being the same is correct and well-understood by most. For BJTs and JFETs, lowest THD is usually at or close to the point where the idle current in each device is the same. Indeed, it is actually more appropriate to say the point at which transconductance of the two devices is the same. In many amplifiers, the sameness of operating current is enforced by the current mirror load.
However, your example does not support your assertion that I questioned (you asserted that the servo correction should not be injected along with signal). It also does not support your second assertion that a DC servo should not be used to correct basic offset. We always try to design for fundamentally small offsets and drift, but there will often be enough left over to warrant correction.
Part of the issue is the DC gain of the amplifier, which will usually be 20-30 when no electrolytic is used. A monolithic Dual JFET pair with 10mV offset will result in an output offest of 200-300mV, which is too much. Of course, one can spend a lot more and get by with an expensive Dual JFET with only 1mV offset.
Where in the topology the servo correction is most appropriately applied can sometimes depend on topology or where the source of the offest is located. However, more often than not, and in particular for conventional VFAs, the correct point is in fact at the feedback node. For BJT input pairs, the origin of the offset is usually Vbe mismatch and input bias current mismatch (or input bias current flowing through different DC path resistances on the pos and neg inputs). In this case it is best to inject the servo correction signal at the input to the diff pair and NOT at the current mirror (for example). Correcting offset at an otherwise balanced current mirror does exactly the wrong thing.
Bear in mind that when a DC servo offset correction is applied at the input to the diff pair (along with the feedback signal), the current mirror load STILL enforces the operating current equality in the diff pair.
There is nothing wrong with injecting the DC servo correction along with the feedback signal, and in fact in most VFA amplifiers that is the correct place. I have not thought carefully enough about the CFA case, but your generalization was not limited to CFAs.
Cheers,
Bob
All quite true and understood. My generalization did not apply to a specific circuitry or topology... such as, if a circuit does use current-mirrors or not... 'your results my vary'.
Since you have covered many of the issues.....
To expand -- In practice, IF you do not filter the desired DC servo voltage sufficiently and many do not IMO and if there are any power turn-on/off funny stuff with the servo going into the input. I guess you could have a relay delay on the output... but that isnt a great solution. I havent seen any SIM of turn on and turn off while monitoring the output, for asymmetrical or symmetrical topologies, so I cant say if this or that circuit has a problem. Thus, generalization.
Thanks for answering the issues for us re. dc servo for specific topology/circuitry. I am sure many will benefit.
-RNM
Last edited:
Define "serious CFA circuit" and you may have a big surprise. Or perhaps anything that doesn't match your views can't be "serious"?
Good points Bob - I dont think there is enough consideration of the issues around imbalance issues - be they due to device mismatches or deliberate 're-balancing'. Seems some spice analysis to establish the size of the problem would be a good thing. One for a rainy day . . . I'm too busy for the next few months.
I am sure you understand how this little CFA can be compensated.
For me, time to hit the bed.
Attachments
That is exactly how prof. leach did it. Read about WHY he did it that way.
For a CFA , it is "the lesser of two evils" ...
OS
This is not an accurate representation of what I wrote or what I think Scott meant.
You misunderstand my comment if you think I mean "CFA" are "much easier to compensate"
My point was that once the IC team had done the work to optimise the "CFA" then it would work well over a moderate variation in gain without much customer effort.
To stay close to optimal in a VFA as the gain is altered requires more customisation.
I don't see any reason why the creator of the amplifier has an easier job in either choice, VFA or "CFA"
Best wishes
David
Last edited:
Actually "nuanced" was exactly the word that came to mind for Scott's post but I didn't use because to write "Well informed, perceptive and nuanced" looks like flattery.
And I didn't repeat his point that the benefit was typically for lower gain applications, because he had already made that clear.
So I think we have a similar view here.
Best wishes
David
DZ --- All you have to do is read the literature and you will find the same info as SW said here by those authors. This is not new info. Though factory optimized CFA are easy to apply by the customer, it is also easy to optimize. We have not -- those who designed and built CFA power amps done a good job of figuring out why we like the easy nature of the CFA in implementation.... especially using discrete parts on a large area of circuitry and wiring.
When I see SIM'ed circuits using small pfd values for comp, I know they wont work in a wired up circuit with all the strays involved.... so the actual end result is usually lesser and a bunch of trial and error and tweeking to make it stable. Maybe there is less of that with CFA and its low Z circuitry etc. It just comes from 35 years of playing with these things and those here who have done so as well. At least it is so when pushing the envelope to the max of performance and using the devices which give the lowest thd/IM etc...... when not pushing to the edge, maybe it is a different story. Then you dont see much nuances anyway.
Thx-RNMarsh
When I see SIM'ed circuits using small pfd values for comp, I know they wont work in a wired up circuit with all the strays involved.... so the actual end result is usually lesser and a bunch of trial and error and tweeking to make it stable. Maybe there is less of that with CFA and its low Z circuitry etc. It just comes from 35 years of playing with these things and those here who have done so as well. At least it is so when pushing the envelope to the max of performance and using the devices which give the lowest thd/IM etc...... when not pushing to the edge, maybe it is a different story. Then you dont see much nuances anyway.
Thx-RNMarsh
Last edited:
with limits set by low speed power devices and/or packaging/mounting/wiring parasitics the easiest place to push is higher loop gain with higher order loop gain shaping when the critical range of frequency is so low as is audio compared to practical loop gain intercepts
if you want to effectively push up speed you need to use MOSFET in RF packaging ops
you say you understand the role of fast processes in monolithic CFA sucess, but you sure don't seem willing to address the converse - the practical fact of relatively slower output stages which is the audio power amp reality - and how that limits "the CFA advantage" in audio power amps
if you want to effectively push up speed you need to use MOSFET in RF packaging ops
you say you understand the role of fast processes in monolithic CFA sucess, but you sure don't seem willing to address the converse - the practical fact of relatively slower output stages which is the audio power amp reality - and how that limits "the CFA advantage" in audio power amps
Last edited:
Well, I am willing to try that as well. I found microwave devices to be very linear. Especially those designed for class A single-ended operation. But can osc easily if not very careful....strip-line package with smd can be a challenge. But I am up for a new challenge. BTW - I use mostly MOSFET OPS.
A couple years ago I decided I needed to be able to measure what I could design. I set my goal to be able to read harmonics to -160dB. I was told it was impossible, basically. After spending over $70K on this quest, I learn they were correct. BUT, I did get to -150dB (re 1v). This is real measurment and not SIM. And, in the process I learned a lot about distortion analyzers and oscillators from some very helpful guys. Now I can accuratly read harmonics well below my personal design threshold for audio of -100dB. Note the word - accurately. But, I had to try all sorts of things before ending up with ShibaSoku AD725D and Audio Precision 2722A and many others by HP and Panasonic Industrial and a few sound card like products. I can say with great certainty that below -100dB harmonic levels there are a lot of nuances and you find them in the design, topology, layout, parts, grounding, power supplies etc etc. And, it is from my own experience that the CFa and the more simplified compl-push-pull topologies are easiest to get very, very low THD at all freqs and especially HF in actual build. I have been hoping to learn why that is. Still not sure... except that Bonsai comes the closest in describing the reason IMO.
Thx-RNMarsh
A couple years ago I decided I needed to be able to measure what I could design. I set my goal to be able to read harmonics to -160dB. I was told it was impossible, basically. After spending over $70K on this quest, I learn they were correct. BUT, I did get to -150dB (re 1v). This is real measurment and not SIM. And, in the process I learned a lot about distortion analyzers and oscillators from some very helpful guys. Now I can accuratly read harmonics well below my personal design threshold for audio of -100dB. Note the word - accurately. But, I had to try all sorts of things before ending up with ShibaSoku AD725D and Audio Precision 2722A and many others by HP and Panasonic Industrial and a few sound card like products. I can say with great certainty that below -100dB harmonic levels there are a lot of nuances and you find them in the design, topology, layout, parts, grounding, power supplies etc etc. And, it is from my own experience that the CFa and the more simplified compl-push-pull topologies are easiest to get very, very low THD at all freqs and especially HF in actual build. I have been hoping to learn why that is. Still not sure... except that Bonsai comes the closest in describing the reason IMO.
Thx-RNMarsh
Last edited:
I already told the following story.
In an analog big mixing desk we use summing amps in the bus. Usually OPAs in inverting configuration. The problem i always faced was the sound quality deteriorate with the number of tracks plugged in the same bus.
Using Current feedback OPAs (OP260 allowing a usable feedback impedance value), the problem was solved for me and the sound quality remain consistent, whatever the number of tracks.
We organized in the 80th, with several sound engineers working as "free lance" in my studio, a blind test session to compare a pair of modified mixing bus with original ones. The result was obvious, in favor to the CFAs.
Please notice that, in inverting configuration, there is no active device in the feedback path, and the error signal is passively subtracted. It is the configuration that present the best performance for LTP based OPAs. Notice too that the feedback impedance was the same in the two compared bus, apart the little cap in parallel with the feedback impedance, necessary for stability with much of the voltage feedback OPAs for stability, removed with the CFA.
Now, i use those OP260 in my personal preamp and everywhere i can in all the analog stage of my hifi system. I had the occasion to try a lot of various OPAs in my preamp. The OP260 are still there. And i find exactly the same qualities than in my power amplifier: "Clarity", "Ease", "little details", "separation between instruments", "fluidity (or natural) of the trebles", "absence of fatigue".
I'm sorry, we did not made distortion measurements at this this time: Sound engineers are those kind of stupid people who trust their ears more than numbers (may-be because they are paid for this ?).
I'm so sorry for those people witch do not believe in their earing ability, as they are obliged most of the time to listen to their musical productions.
Naaah.
The mistake we made earlier in this thread was to say you have to close the loop at the same frequency as VFA because of the OPS phase shift, which is in this 1-3 MHz range where in a VFA you get a decent PM. As you pointed out earlier, the additional gain stages in a VFA cause the additional phase shift. You dont have those in a CFA so you can close a CFA loop at a higher frequency, and get the benefits.
(I just checked the sx-Amp sim and the loop can be closed at ~4 MHz and still have 80 degrees of PM. In the practical amp, Its running set it at 2.5 MHz with 80 degrees PM and its perfectly stable.)
Last edited: