Which node(s) to worry about in CFB amplifiers

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Clarification: I'm talking here about amps resembling the topology of integrated "current feedback" opamps, like the EL2020, to name the veteran.

These are still voltage controlled voltage sources (Vout = A * Vin), in contrast to transcoductance amplifiers (Iout = Vin / Z), which were also named "current feedback" originally.

For a discrete implementation, see for example Per-Anders' headphone amp QRV06 http://home5.swipnet.se/~w-50719/hifi/qrv06/downloads_qrv06.html which was discussed here at http://www.diyaudio.com/forums/showthread.php?s=&threadid=23241&perpage=15&highlight=&pagenumber=1

In speculating about the pro and cons of this topology, most attention was spent to the "current feedback", the low impedance feedback node. In the case of integrated amplifiers, where it is most important which nodes are exposed, this design gave (disputed) benefits of seperately selecting gain and bandwidth. But in a discrete implementation, where all variables can be adjusted, this doesn't make a big difference.

Having the low impedance feedback node does make the feedback path one small signal follower delay faster, which isn't that much a difference.

IMHO the interesting node is the link between input stage and VAS:

In QRV06 and many other CFB designs, the collectors of the input diamond drive the inputs of current mirrors, a low impedance node (about 100 Ohm in the case of QRV06). Consequently, when looked at in open loop, the gain of the input stage is low, about 22dB. In other designs this gain is still lower.

In LTP input stages, all done in "old skool" manner, the collector load is a high impedance node, at least for low frequncies. The collector has to to drive a) a (possibly cascoded) current mirror output b) a (possibly beta enhanced) VAS transistor base and c) Cdom.

So, possibly, the perceived differences between CFB and other topologies boil down to a boring pole-zero discussion of the amplifier in open loop.

And, for the experimental mind, both concepts can be mixed, and the "missing combinations", input diamond with high-Z load and input LTP with low-Z load, be tried.

(Not that I assume, they are untried. There simply are no new ideas in analog audio I assume, and somebody having a better overview will most likely even be able to name commercial products with these topologies).

Regards,
Peter Jacobi
 
Single ended current feedback topologies have been around at least as long as complementary transistors

I am exploring your question in the single ended “current feedback” amp circuit below

The underlying topology of Q1-3 with emitter feedback R9/10 was a mainstay of 1960’s solid state amplification – I’ve been playing with this sim aiming towards a “retro” solid state headphone amp for 300 ohm headphones - i tell someone i can design solid state circuits with distortion products below the noise floor and then they ask "but are the odd harmonics higher than the even?"


My concept for D1 is that it permits a higher value of R1 which, in parallel with a suitably high Q2_Rpi, hopefully increases loop gain faster than the degeneration provided by D1’s small signal impedance reduces it on the other side of Q2 - but I’m well on the way to establishing that it really doesn’t work that way…

For an interesting indication of which is the “sensitive node”, try changing out Q2, a high Hfe device (such as the 2n5087 which was my original intention) and explain the decrease in distortion – I see -110 dB 2nd with the lower gain 2n3906 model vs -105dB with the 2n5087 @ 1 mWrms output, this indicates my theory is wrong (or that its foolish to put much faith in the models) and perhaps I should scrap the input Z boost at Q2_b and use a cascode or super-pair for Q2

Q3,4,7 form a push-pull class A complementary output stage that allows >300 mW into 300 Ohms within the power dissipation limits of fast small signal Qs ( Zetex e-line 1 W “to-92”, classic to-5/to-39 cans or just parallel a couple of 600 mW to-92 for Q3,4)


I’m certain the state of the art in predicting the audibility of varied distortion mechanisms won’t allow for the degeneration of such into a “…boring pole-zero discussion of the amplifier in open loop”

In fact @ > 3hrs old this thread is already overdue for a obligatory “current feedback sucks in audio amplifiers” from jocko
 

Attachments

  • cf_head0.png
    cf_head0.png
    43.5 KB · Views: 479
Peter, I suggest that you fool around with my simulation files and see for yourself what is important. Judging from your thoughts I imagine that you are not up to date the this topology (no offence). You must throw away what you know about voltage feedback and start think different (Apple!).

You have 3 things:

The feedback resistor sets the speed of the amp

Totally forbidden to have a capacitor across the feedback resistor

The current mirror stage is a high impedance node, must be followed by a buffer.

I have find a lot of good documents. Check at AD, TI and NS. Lot's of good and understandible material.

I have uploaded my simulation files somewhere here. If you can't find send me a message.
 
Hi Per-Anders,

Thank you for you fast answer, but I'm in doubt what you are wanting to tell me.

peranders said:
Peter, I suggest that you fool around with my simulation files and see for yourself what is important. Judging from your thoughts I imagine that you are not up to date the this topology (no offence). You must throw away what you know about voltage feedback and start think different (Apple!).
[...]
I have uploaded my simulation files somewhere here. If you can't find send me a message.

I still have your sim files from the last discussion (use CFP boosters or not). I've used them to study the amp in open loop.

peranders said:
You have 3 things:
The feedback resistor sets the speed of the amp

Totally forbidden to have a capacitor across the feedback resistor

The current mirror stage is a high impedance node, must be followed by a buffer.

I'm aware of points 1 and 2.

Re point 3: Of course the current mirror has high Z output - but it has a low Z input. So the input stages is working into low Z node, which superficially (or really) means throwing away possible open loop gain.

I was just wondering whether the the input stage working into high or low Z node may be the most significant difference between CFB and LTP amps. As this would be difference between one and two voltage gain stages and possibly different Bode plots of the open loop amp.

peranders said:
I have find a lot of good documents. Check at AD, TI and NS. Lot's of good and understandible material.

I've been through lots of this stuff. But beside all the valuable technical info, you must add a grain of salt, as you shouldn't believe the 'CFB is the magical solution' hype, sometimes seen (in older docs when this point was more heatedly discussed). In some paper from 'the other side', it was soberingly remarked that CFB only brings back one parameter to tune (speed via the In- source Z), which was taken from him before (speed via external comp C).

Regards,
Peter Jacobi
 
Hi jcx, All,

jcx said:
Single ended current feedback topologies have been around at least as long as complementary transistors

I am exploring your question in the single ended “current feedback” amp circuit below.

Thanks for joining the exploration party.

I'm aware that we are most likely bound to operate the VAS single ended, when the input stage has significant voltage gain.

A related problem gave raise to this long thread:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=16796

jcx said:
The underlying topology of Q1-3 with emitter feedback R9/10 was a mainstay of 1960’s solid state amplification [...]
Yep. I always think "El Cheapo" when it comes up:
http://sound.westhost.com/project12.htm

Also going to the beginnings of my Elektor collection greeted me with this heritage. For example I've entered the 1975 Equin in the simulator (does well as long as you stay with slow output BJTs):

PDF http://www.linearaudio.de/scratch/equin-1.pdf
LTSpice http://www.linearaudio.de/scratch/equin-1.asc

jcx said:
My concept for D1 is that it permits a higher value of R1 which, in parallel with a suitably high Q2_Rpi, hopefully increases loop gain faster than the degeneration provided by D1’s small signal impedance reduces it on the other side of Q2 - but I’m well on the way to establishing that it really doesn’t work that way…

The effect of the additional diode drop would be inclusive, when the usual cookbook recipes (Self et. al.) are applied, i.e. replacing Q2 with a "beta enhanced" stage = Darlington.

jcx said:
I see -110 dB 2nd with the lower gain 2n3906 model vs -105dB with the 2n5087 @ 1 mWrms output, this indicates my theory is wrong (or that its foolish to put much faith in the models) [...]

Replace the LTSpice 2N3906 default model with the Fairchild one, for a test.

Anyway, 5k is more than 100, but by a recent poll, 71 of 100 LTP input stages are working into current mirrors, a much higher impedance node. I'll post schematic for this in a minute.

Regards,
Peter Jacobi
 
pjacobi said:
Not that I assume, they are untried. There simply are no new ideas in analog audio I assume

In fact the input topology of the Quad 303 can be considered as current feedback in some sense. Here are 2 app notes about the ins and outs of CFB:

http://www-s.ti.com/sc/psheets/sboa081/sboa081.pdf
http://www-s.ti.com/sc/psheets/slva051/slva051.pdf

Note that the value of the feedback resistor defines part the open loop gain with CFB. The lower the resistance the higher the open loop gain. So imagine what happens at high frequencies when you put a capacitor across the feed back resistor...

Cheers ;)
 
So, as promised, the schematics of the other two combinations:

A) CFB input stage working into a high Z node
(as discussed, this should better be asymmetrical)
PDF: http://www.linearaudio.de/scratch/cfb-asym.pdf
LTSpice: http://www.linearaudio.de/scratch/cfb-asym.asc

Q1 and Q4 make the CFB input stage (half diamond), the collector of Q4 is working against the high Z node consisting of a current source, VAS transistor's base and Cdom - just as in a standard design with LPT input stage. VAS transistor Q6 works against current source Q9.
Remaining Qs define the bias currents. Output stage is highly stylized as a power follower with 100k input imdance and 0 output impedance.
Fancy cascoding can be added at several places, especially poor Q4, providing significant amount of the total voltage gain, would profit.

B) LPT input stage working into low Z node
(this can be done asymmetrically or symmetrically, to highlight the similarity with EL2020 [or QRV06] CFB designs, this example is symmetrical)
PDF: http://www.linearaudio.de/scratch/lpt-sym.pdf
LTSpice: http://www.linearaudio.de/scratch/lpt-sym.asc

OK, these are standard LTPs Q5/Q6 and Q7/Q8. But here the collectors of Q5 and Q7 go into low Z nodes, the inputs of the current mirrors Q1/Q2 and Q9/Q10. Output stage is highly stylized as above.
As the voltage gain is done alone by the current mirror outputs, without cascoding there wouldn't be much. So, and for added variety with JFETs, they are cascoded.

Summary:
Your standard CFB layout (as in EL2020 or QRV06) has (approximately) only one voltage gain stage, as the input stage is working into a low Z node.
An archetypical "standard" audio amplifier with LTP inputs (as seen in the Self, Slone and numerous other books) has two voltage gain stages, as in the input stage is working into a high Z node.
But you can also have it the other way around, as shown above.

Regards,
Peter Jacobi
 
pjacobi said:
Summary:
Your standard CFB layout (as in EL2020 or QRV06) has (approximately) only one voltage gain stage, as the input stage is working into a low Z node.
An archetypical "standard" audio amplifier with LTP inputs (as seen in the Self, Slone and numerous other books) has two voltage gain stages, as in the input stage is working into a high Z node.
But you can also have it the other way around, as shown above.

Regards,
Peter Jacobi

Hi Peter,

Sorry, but I am missing the point what you actually want to tell us.

Regards
 
Folded Cascode

Pjotr said:

When seeing the schematic for the THS4001 in this paper, I realized that I've forgotten another important topology: folded cascode.

This is another possibility to let a LTP input work into a low Z node and consequently resulting in a single voltage gain stage.

See these schematics:

Symmetrical version:
PDF: http://www.linearaudio.de/scratch/folded-cascode-sym.pdf
LTSpice: http://www.linearaudio.de/scratch/folded-cascode-sym.asc

Asymmetrical version:
PDF: http://www.linearaudio.de/scratch/folded-cascode-asym.pdf
LTSpice: http://www.linearaudio.de/scratch/folded-cascode-asym.asc

Some observations:
- It's not straightforward to mix this with a diamond input stage, as you must add 180° somewhere for the right polarity.
- Especially the symmetric version behaves in simulation very similiar to a diamond+current mirror topology. It would be most interesting to compare measured and perceived performance
- Further cascoding with JFETs will give those low THD numbers in simulation, which are often laughed about (but not near -300dB as Bernhard reqularely achieves...)
- Curiously, in the asym version K2 and K3 are of same (low) height wheras in the sym version K3 is 10dB lower. Go figure!

Regards,
Peter Jacobi
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.