would a folded cascode stage benefit from a Boxall connection?
say a pnp folded cascode, referenced through a npn emmitterfollower (and a pull down resistor to ground) of which the collector is connected to the emitter of the folded cascode. thus forming a Boxall pair. i am going to try this on my amp one time.
say a pnp folded cascode, referenced through a npn emmitterfollower (and a pull down resistor to ground) of which the collector is connected to the emitter of the folded cascode. thus forming a Boxall pair. i am going to try this on my amp one time.
Last edited:
It was not my intetion to have low feedback, I just could not get high feedback level without distortion degradation.
Boxall is great for anything that wants a very high output impedance, particularly low capacitance, until as I say the transistors run out of gas. What Boxall needs to work well is to be fed at the main Q emitter by an impedance a good deal higher than the emitter impedance. Now, note that we often approximate this as 1/gm, but when you have a very high impedance at the collector and a big voltage swing, that approximation begins to break down. Controlling peaking will require some compensation somewhere, it seems usually just a little lumped C at the output. Distortion from variable output C with voltage of the transistor is still greatly reduced, although things are slowed down a bit. And of course whatever buffers this output node needs to have itself a small and well-behaved input capacitance.
Aldridge-like stages, i.e. bootstrapped cascodes (popularized by Hawksford) have often a higher voltage burden but also recycle the control electrode current and hence also benefit from that current passing back through the "lower" transistor with as close to unity gain as possible. Walt Jung I believe has a comprehensive and comparative study on his website of some of the alternatives, from the perspective of current sources.
Another thing that is nice about these recycling schemes: the changes in base current of the main device versus temperature, and the noise in that base current, are as well recycled. So you wind up with at least the shot noise in the auxiliary transistor in Boxall for example, but there is a very small contribution from the main transistor. With a smaller geometry auxiliary transistor with high beta you can do very well. However, when used in current mirrors and mirror-like structures with current gain, the ballasting emitter resistors need to be large before their contribution gets small.
Achieving low series (voltage) noise is pretty easy in CFAs. It's the parallel ("current") noise at the feedback input that's often the limitation, and as mentioned it is often that arising from the mirrors, unless they have large and many-volts-across resistors; those then tend to slow things down a lot, although some bypassing with C can recover this and move noise out-of-band. EDIT: Dadod has a good compromise here with the 1k loads, thus referring about 5.8pA/sq rt Hz to the input. Good bandwidth is still achieved, and trying for a lot more is probably running into limitations elsewhere, like the output stage.
Last edited:
Yes.
Simplifying the design is part of the art in designing .... while giving up nothing that is required in the process. That results in the optimum design. But, it ain't as easy as it ends up looking.
Thx-RNMarsh
One can approach noise in CFAs the tried-and-true but brute-force way of massive parallelism, using bigger geometries or paralleled parts, more quiescent current. But it's a slow climb to higher performance, and looking at the overall signal chain and the S/N ratio delivered to the power amp's input to begin with, it may be mostly academic.
drew a quick crude schematic so function is clear. would the current transfer ratio be closer to -1 than a normal folded cascode? (i do not know the specific terms but think i make sense 😛 )
🙂😎
[shows an example of an inherently linear circuit.... something good to learn how to do in any case - adds to one's pallet of design choices/flexibility]
-RM
Do you mean before the peaking is tamed? Not sure how to characterize it other than having a negative component that is corrected by the external C. For that I think I showed 270 femtofarads for the test load of the open-loop N mirror of 100k. This was to get flat response -3dBr at about 8.5MHz.
With the P version and in push-pull with the N (some drafting to be done to clean it up) I found the best flat response down a bit and needing a little more C. The cancellation of 2nd is partial and could probably be fine-tuned, but I see about a 10dB further reduction in distortion for the same 8V p-p swing at 10kHz.
These are medium-bandwidth limited-power and voltage parts, and a better design for the power amp application would probably use high-voltage devices like you show (2SA1381, 2SC3503). I'll try to incorporate those although I don't know how good my models are (they exist as "ECG" parts in the library).
Boxall by itself really wants a lot of Z in the emitter to have good distortion performance into small load R. Although the transfer ratio may be a little better there may be little improvement, even degradation of distortion. Where it shines is in the capacitance reduction and with large load R. There is also going to be less thermal distortion due to alpha change with self-heating, but this is pretty small to begin with. The simple common-base stage is pretty decent!drew a quick crude schematic so function is clear. would the current transfer ratio be closer to -1 than a normal folded cascode? (i do not know the specific terms but think i make sense 😛 )
However with an I source load for the folded cascode, and a high-Z signal drive, things get better in a hurry. Now the Vbe change of both stages has no effect, and the remaining effect of alpha modulation with output voltage and current swing dominates the simple common-base stage. But the Boxall pumps most of this back into the main emitter, reduced only by the alpha of the auxiliary transistor, for nearly perfect behavior except for the peaking at high frequencies.
i see, thanks 🙂 in the actual circuit there is a ccs load for the folded cascode (instead of this 1K i drew) and a high output impedance stage driving the emitter.
drew a quick crude schematic so function is clear. would the current transfer ratio be closer to -1 than a normal folded cascode? (i do not know the specific terms but think i make sense 😛 )
I do not see the diagram, am I the only one?
I had an additional insight about this variant [EDIT: see post 940] which is a pleasant surprise. Typically the auxiliary transistor for base current feedback in Boxall is run at a much lower current than the main one. As a result the [EDIT: Boxall portion of the] circuit's overall current noise is mostly due to shot and excess noise in the aux transistor's base current, and this is pretty small. But what is nice here is that we are also recycling most of the aux transistor base current, this time with two alphas worth of loss and some loss to the emitter ballasting R. But it means we can essentially ignore its noise contribution as well. If one compares this to the configuration where the aux base is tied directly to the input of the mirror, that one results in about 6dB (for a unity gain magnitude mirror) more output current noise than the base current noise alone. If the return is done according to the patented circuit, the noise simply appears at the output with unity gain.
So what? one might ask. Well, having no impact from aux Q noise allows us to increase its quiescent emitter current and increase the bandwidth of the stage, without a noise penalty. Of course we need to be mindful of the noise in whatever is supplying that emitter current (unless it too is recycled, but that is messy), and it does subtract from the main Boxall transistor current. But I see for the same example, 100k load, with the output load current adjusted to compensate, a bandwidth increase with the smaller required response compensation cap to 20MHz from the initial 8.7MHz. Distortion is also better by about 6dB, still mostly 2nd.
The noise will still be dominated by ballasting resistor thermal noise when those are 500 ohms, and there is a small contribution from base current noise of Q1, Q3, and a little from Q2. There will be a breakeven point when ballasting Rs get larger and base current noise is more important; for nominal beta = 400 devices the noise in the limit of large R is about 3pA/sq rt Hz, neglecting excess noise. There is also e sub n modulation of current, from Q1 and Q3, but if these are low rbb' parts it can be fairly small.
Last edited:
Could you tell if I introduced your enhacement CM from #940, the schematic is here http://www.diyaudio.com/forums/solid-state/240712-cfa-topology-audio-amplifiers-95.html#post3623952. Sorry by mistake I directed that question to Manso first.
Yes I saw that (and was afraid that I had missed something of Manso's 😱). Not to worry!
Nothing that would change the picture , good results but nothing
exceptionnal , moreover if we look at the inflated component count
to yield what compare pale to a VFA using half the transistors ,
not only on THD but also on IMD.
At thoses 50W/8R a NDFL based VFA will typicaly do sub tenth ppm
at 1khz , below 1ppm and 5 ppm at 10khz and 20khz respectively ,
IMD 400 + 7khz being inexistant while 19 + 20 khz IMD residuals
are at -120dB or so , all numbers that are possible with 17 transistors ,
far from the 28 i saw recently here , hence , i m still skeptical
about the , well , current CFA frenzy...😉