Thank you Edmond,
If we could just stay on subject and leave the comparisons alone for now we might get somewhere. It seems more than half of all the posts are either comparisons or objections to the topic. Those people should start there own thread, why there is no such thing as a CFA amplifier and let the rest of us get down to the subject.
To me this is the equivalent to saying that since there is no real lead in a pencil we should never call it a lead pencil and just discount that is the name we use in common language. Enough of the argument about semantics, it is getting real tired as an argument.
If we could just stay on subject and leave the comparisons alone for now we might get somewhere. It seems more than half of all the posts are either comparisons or objections to the topic. Those people should start there own thread, why there is no such thing as a CFA amplifier and let the rest of us get down to the subject.
To me this is the equivalent to saying that since there is no real lead in a pencil we should never call it a lead pencil and just discount that is the name we use in common language. Enough of the argument about semantics, it is getting real tired as an argument.
Last edited:
On the contrary, there is such a beast as a current feedback amplifier: it possesses shunt applied-series derived (current) negative feedback, has, ideally, zero and infinite input and output impedance respectively and delivers a current output for a current input.
The topology far too many folk erroneously refer to as "CFA" possesses series derived-shunt applied (voltage) negative feedback and has a high input impedance and low output impedance. It is, therefore, a voltage feedback amplifier. Elementary my dear Watson. 😎
Agreed, why don't you start and set an example that others may follow. I did a search and haven't find much of your contribution, other than complaining about the lack of focus.
If you don't feel qualified to contribute, why don't you formulate some specific questions (not like "how do I build the best CFA?") and then perhaps somebody will pick up and answer.
Michael,
I thought that the original premise was exactly that, "possesses shunt applied-series derived (current) negative feedback, has, ideally, zero and infinite input and output impedance respectively and delivers a current output for a current input" Can we just agree on a basic premise such as this and move forward? That is all we need to do and stop the sideshow about what we are talking about. Of course I imagine just as in a vfa there will be alternative ways to achieve the end results, that is what we should be looking at and why each implementation does what it does and what works best in each case.
I thought that the original premise was exactly that, "possesses shunt applied-series derived (current) negative feedback, has, ideally, zero and infinite input and output impedance respectively and delivers a current output for a current input" Can we just agree on a basic premise such as this and move forward? That is all we need to do and stop the sideshow about what we are talking about. Of course I imagine just as in a vfa there will be alternative ways to achieve the end results, that is what we should be looking at and why each implementation does what it does and what works best in each case.
Waly,
I am most surely not qualified to put up any schematics that I would not expect those in the know to already have seen and studied. I am not asking for the best cfa there is as I don't expect there is a single way to do this. I just want to learn by seeing and understanding circuit designs that are discussed and analyzed in a rational manner. That is why I keep quite and try to just watch, but it is a few who seem to dominate the conversation with what doesn't seem to be of much value to the discussion. I have seen some of your comments that are very detailed and those I appreciate, your actual analysis of particular modes of operation.
I am most surely not qualified to put up any schematics that I would not expect those in the know to already have seen and studied. I am not asking for the best cfa there is as I don't expect there is a single way to do this. I just want to learn by seeing and understanding circuit designs that are discussed and analyzed in a rational manner. That is why I keep quite and try to just watch, but it is a few who seem to dominate the conversation with what doesn't seem to be of much value to the discussion. I have seen some of your comments that are very detailed and those I appreciate, your actual analysis of particular modes of operation.
A question then to start the discussion and perhaps others will ask similar questions. Can we define a series of transistors that meet the requirements for low impedance in the input stage, what we should look for in a device in the first stage and which devices are currently available, not just the most sought after limited availability devices, something we can easily purchase from a Digi-key or Mouser type of supplier.
Ok, for a diamond input stage. I'll start with a couple of suggestions... BC5x0c or BC3x7-40 and use cascoding. Cascode transistors - something with large (ish) Vce capabilities like 2N5401/5551.
The BC5x0c and BC3x7-40 have good low Vce characteristics and high gain.
Almost any small signal transistor will do in the input stage (diamond buffer), from BC540/550 to 2N5551/5401. The diamond buffer, being an emitter follower, has very relaxed requirements. Collector current is in the mA range, Vce worst case equals the +/- power supplies and Beta should be reasonable high.
The low impedance at the inverting node (I suppose this is what you mean) has almost nothing to do with the transistor type. It's mostly a property of the topology. That "low impedance" is never a requirement.
For cascoding, you can't beat the Fairchild KSC3503/KSA1381 pair.
P.S. IE10 drives me nuts. Each time I'm typing in "cascode" it stubbornly autocorrects to "cascade". Have to go back each time and correct the word.
Last edited:
What's your opinion regarding transistor selection when taking into account the small Vce across the input transistors of a diamond input? There is an effect that's been dubbed (on here) as "quasi saturation". The BC5x0 is supposed to be a good choice when taking this into account.
Transistormarkj made some measurements showing this effect. See link.
http://www.diyaudio.com/forums/solid-state/236002-amp-design-attempt-number-2-simpler-19.html#post3581526
Last edited:
Waly,
Yes Microsoft has a way of ******* most of us off at some point or another. Can't you just add the correct spelling to the dictionary that IE10 is using so that it stops trying to auto-correct a word it does not know?
Yes Microsoft has a way of ******* most of us off at some point or another. Can't you just add the correct spelling to the dictionary that IE10 is using so that it stops trying to auto-correct a word it does not know?
When a device with lightly doped collector regions (read: more common (but not only) in high voltage devices) operates at high injection levels (read: high collector currents), the internal base-collector junction is forward biased, while the external base-collector junction is reversed biased; as such, under these circumstances, the DC current gain and the unity gain frequency fT falls sharply.
Indeed, the curve traces for the 2N3904 clearly show the quasi saturation effect. However, I am not sure this necessary matters in the diamond buffer input stage. If you want a great example of quasi saturation, take a look at the MJL4281/MJL1302 power pair. They are used as emitter followers in output stages and I never heard of anybody complaining about.
What those traces are also showing is the high Early voltage and also the sustained beta (with Ic) for the 2N3904, and the very poor Early voltage and very poor Beta-Ic for the BC550 over 20-25 mA (that's why these devices need to be cascoded when used in gain stages, to take advantage of the high beta).
The input diamond buffer doesn't need to be biased beyond 1-5mA so any of these devices will do good.
P.S. It is very hard to make recommendations without seeing a schematic, or at least a set of requirements. E.g. the Vce requirement (and perhaps the quasi saturation impact) in the diamond buffer depends on where the transistors collectors are connected.
Last edited:
When the input diamond buffer in a current feedback amp ( if that exists ) is not in the global feedback loop i see trouble.
I do not think that a conventional diamond buffer is the best solution for a CFB amp.
I do not think that a conventional diamond buffer is the best solution for a CFB amp.
Waly,
Here is a HEAVILY stripped down version of the diamond input from my amp I'm currently at the prototype stage with.
Purely, presented as point of discussion. The diamond is biased at approximately 1mA.
PS. I read you are a student. Can I ask what level you are studying at? I have studied electronics at degree level but you are way more knowledgeable.
Paul
Attachments
In this case, quasi saturation may matter. Spice supports the Level II Gummel Poon model (includes quasi sat) but I still have to see such device models...
Honours Bachelor + Master's EE, finished my dissertation, now in a 2 year post graduate program abroad. Nothing about audio.
mcd99uk, are you seeing any benefits in performance by bootstrapping your input buffers to the complementary level shifter emitter as opposed to using a simple Zener reg and tying the collectors of the buffers to these? This would avoid the quasi sat problem that seems to be a concern.
Zener noise is easy to deal with, although it requires an additional low value resistor and a cap.
Zener noise is easy to deal with, although it requires an additional low value resistor and a cap.
Any links to these which us unwashed masses can unnerstan please Scott.
My naive interest is always getting the same performance with greater simplicity.
For those of you that are interested, there's a very nice blog by Prof Sergio Franco here
Demystifying pole-zero doublets | EDN
Some of you may recall he wrote some articles in the late 80's demystifying CFA in EDN.
Demystifying pole-zero doublets | EDN
Some of you may recall he wrote some articles in the late 80's demystifying CFA in EDN.
Thanks for this obvious-with-hindsight, insight Joachim. Perhaps this sets the ultimate performance limit for Diamond IPS CFAs.
Waly hints at this in his #518 post on Diamond IPS and I shall now put your statement alongside his seminal explanation.
My $0.02 is that there is an element of Marshy's Distortion Cancellation in this .. both for the Baxandall pairs which form each leg .. as well as the 'symmetrical' nature of the input Emitter followers.
Anyone comment on whether the Emitter followers outside the FB loop set the ultimate THD level in Diamond IPS CFAs?
Which begs the question whether the simple VSSA IPS like my #500 has any ultimate limit to reducing THD with more Loop gain. If so, trying for 1 ppm THD would only complicate stuff things to no benefit.
Last edited:
A first steb in answering this question will be to try the input buffer with resistive loading vs CCS a la mcd99uk. I suspect CCS will offer some small benefit.
Of course, a more radical approach would be to use something like AFEC, which would deal with this nicely. However, I think in the spirit of exploring things in some detail and really getting to the non- linearities an dealing with them, this approach is something I would shelve in this discussion for now.
Joachim, do you have any proposals or ideas for the input stage?
Of course, a more radical approach would be to use something like AFEC, which would deal with this nicely. However, I think in the spirit of exploring things in some detail and really getting to the non- linearities an dealing with them, this approach is something I would shelve in this discussion for now.
Joachim, do you have any proposals or ideas for the input stage?