Not from what I have read - they are matched very well now. How well is "True" matching I wouldn't venture to guess.
I agree none rely only on matching for their performance. There many methods of cancelling distortion... few of which have been discussed here... besides matching and diff pair cancl and nfb. Did you see my C matching demo awhile back in the beginning -- a jFET follower... that is the type of thing being done today -- mostly to extend bandwidth for commercial apps. But it reduces distortion as well and can be applied to audio as my quick little demo indicated the dramatic reduction using this means.
Thx-RNMarsh
Last edited:
Though the world seems to like my little idea as so many versions of it have been developed.... I don't think of it as a CFA. But... who am I to fight popular perceptions. I still think of it as a compl Push-Pull circuit with nfb applied to input pair emitter (low Z port). I guess that's too much of a mouth full.
So should this hybrid be called a CFA? What do you guys think, now?
Thx-RNMarsh
So should this hybrid be called a CFA? What do you guys think, now?
Thx-RNMarsh
Last edited:
Member
Joined 2009
Paid Member
I guess the gap is in the understanding of "matching". For example, it is impossible to match PNP/NPN devices for all the Is, Cbc and Ft. Reason is, Is depends of the carrier mobility (in a 3:1 ratio, electron to holes, requiring about 3x the NPN device area for the same Is). 3x device area means much larger Cbc, also the Cbc dependence of Vcb is different for p-n vs, n-p base collector junctions. It also means a lower Ft for the PNP. There's always a compromise, but no sane designer will rely on PNP/NPN matching for reaching certain high frequency linearity target. DC and LF matching is less than half the device "true" matching story.
most theory papers don't make a distinction for symmetry in a amp topology classification scheme – they are usually shown as abstract dependent sources that the engineer is supposed to recognize can be realized either single ended or symmetric
below is from:
http://www.diyaudio.com/forums/soli...s-self-wants-your-opinions-9.html#post3411700
a few feedback classification schemes have been used over the years - circuit theory PhD have to publish something
"Current Feedback Op Amp" is a "trivial name" - popularized, understood but not strictly fitting a proper classification scheme - like the VAS/TIS "controversy" Trivial name - Wikipedia, the free encyclopedia
http://web.archive.org/web/20070128220917/http://people.ee.ethz.ch/~hps/publications/2000cas.pdf
for compensation there is also lots of literature - since audio power amps mostly use op amp toplogies - and some similar limitations like slower outputs
but I don't know of any universal classification/naming scheme - most options are amp gain stage topology dependent - you can find such awkward things as "Multipath Reverse Nested Miller Commpensation"
Cherry's Nested Feedback Loops do show a fairly "regular" structure for adding gain stages - but still is tied to amp stage topolgy/circuit implementation details
below is from:
http://www.diyaudio.com/forums/soli...s-self-wants-your-opinions-9.html#post3411700
a few feedback classification schemes have been used over the years - circuit theory PhD have to publish something
"Current Feedback Op Amp" is a "trivial name" - popularized, understood but not strictly fitting a proper classification scheme - like the VAS/TIS "controversy" Trivial name - Wikipedia, the free encyclopedia
http://web.archive.org/web/20070128220917/http://people.ee.ethz.ch/~hps/publications/2000cas.pdf
from: http://www.diyaudio.com/forums/soli...feedback-not-suitable-audio-6.html#post947696 - thread closed, original reference dead - I give a archive.org link
for compensation there is also lots of literature - since audio power amps mostly use op amp toplogies - and some similar limitations like slower outputs
but I don't know of any universal classification/naming scheme - most options are amp gain stage topology dependent - you can find such awkward things as "Multipath Reverse Nested Miller Commpensation"
Cherry's Nested Feedback Loops do show a fairly "regular" structure for adding gain stages - but still is tied to amp stage topolgy/circuit implementation details
Attachments
Member
Joined 2009
Paid Member
Increased OLG, by itself, is generally easy to get. However, the limit on how much you design for is stability. The symmetrical amplifier, which for purposes here I'll equate with a push-pull VAS, are, I believe, largely the symmetrical behavior of the VAS and the fact that twice as much VAS output current is available for a given amount of VAS bias current.
Cheers,
Bob
Member
Joined 2009
Paid Member
Yes. BUT.
I don't have much concern for that level of matching to get very low distortion in audio (low frequency).... the circuits/topology I used didn't and I just gave an example(s). These are not SIM... they are measured results, too.
And very simple and easy to keep stable without interaction with HD (phase shift and phase tweeking).... and, That seems to me a BIG plus.
-THx-RNMarsh
Last edited:
Question
Question for everybody:
Where the hell is this "simplicity" requirement, for a WFA (Whatever Feedback Amplifier) coming from?
Does somebody believe that the "quality" of an audio system is inverse proportional to the number of solder joints? Parts cost pennies, anyway, who cares e.g. about 10 more transistors? Not to mention that the low power part is perhaps 5% of the cost in any decent audio power amplifier.
Or is this some sort of game, like "who can build the simples amplifier that nobody can subjectively distinguish from a 50,000 quid boutique unit". If so, the game was won and it's over since before I was born.
Or everybody is targeting the level of understanding of a low electronics skills DIYer, with a 10 quid slush fund, and who should be able to build one of these things with a gas oven heated soldering iron? I understand that a DIY friendly design should not require a spectrum analyzer, but from here to promote designs that are deeply compromised (look at the VSSA PSRR!), only for the sake of "simplicity" it's, to me, a long way.
Question for everybody:
Where the hell is this "simplicity" requirement, for a WFA (Whatever Feedback Amplifier) coming from?
Does somebody believe that the "quality" of an audio system is inverse proportional to the number of solder joints? Parts cost pennies, anyway, who cares e.g. about 10 more transistors? Not to mention that the low power part is perhaps 5% of the cost in any decent audio power amplifier.
Or is this some sort of game, like "who can build the simples amplifier that nobody can subjectively distinguish from a 50,000 quid boutique unit". If so, the game was won and it's over since before I was born.
Or everybody is targeting the level of understanding of a low electronics skills DIYer, with a 10 quid slush fund, and who should be able to build one of these things with a gas oven heated soldering iron? I understand that a DIY friendly design should not require a spectrum analyzer, but from here to promote designs that are deeply compromised (look at the VSSA PSRR!), only for the sake of "simplicity" it's, to me, a long way.
Night thought
Everybody should agree that feedback relies on an algebraic sum of two signals (it's a subtractive process in negative feedback)
These signals have to be of the same nature, either both voltages or both currents (a current cannot be subtracted from a voltage...).
To sum two currents, each has its own path, they are in parallel and then join at a common point.
Concerning feedback, I see neither two paths of parallel currents nor a point where they join in the circuits currently named here Current Feedback Amplifiers.
Everybody should agree that feedback relies on an algebraic sum of two signals (it's a subtractive process in negative feedback)
These signals have to be of the same nature, either both voltages or both currents (a current cannot be subtracted from a voltage...).
To sum two currents, each has its own path, they are in parallel and then join at a common point.
Concerning feedback, I see neither two paths of parallel currents nor a point where they join in the circuits currently named here Current Feedback Amplifiers.
Subtle things about the 'CFA' topology -
A couple variations I tried back in the 1970's was looking at a single transistor gain stage THD. I noticed while measuring distortion what affect varying the supply voltage would have on the THD. I found there was an unexpected THD null at a certain supply voltage. And, different load R would shift the null and require a different dc supply voltage to obtain a THD null. I could have used the transistor output C and gain multiplipication of Co and put a similar value as a variable C between the input and emitter (I think it was a bipolar?) and null the THD with the C trimmer...... Depending on the Rload resistor value etc the null would be broad or narrow....
If that is the case, you can have ps voltage of two different values - no problem as long as you have plenty of margin so you don't clip the signal. [Then don't need to worry as much with matching
]
A differential input stage with same type devices would require added C cross-coupled at the output to cancel the C's. I mentioned this near the beginning. Called Neutralizing technique. It also works well. To do it in Compound-Complimentary Push-Pull (or CFA), the cancellation comes naturally. Just need to get close the same C's (matching?) and some tweaking of added C or matching devices closer. Then throw in little amount of NFB to clean up the remainder and provide other bennies. These are all alternative to high nfb to clean up everything... not that its bad - just and alternaive and it is simple, cheap and uses few parts.
The techniques of distortion cancellation rather than reduction, is not new but with this symmetrical topology it is easier in my sense of the word. Maybe not any lower than other methods but its simple (OK sophisticated, if you prefer) Subtle, for sure. But you can see why at higher and higher freqs - well beyond audio - it is very hard to reduce distortion thru just nfb alone.
Thx-RNMarsh
A couple variations I tried back in the 1970's was looking at a single transistor gain stage THD. I noticed while measuring distortion what affect varying the supply voltage would have on the THD. I found there was an unexpected THD null at a certain supply voltage. And, different load R would shift the null and require a different dc supply voltage to obtain a THD null. I could have used the transistor output C and gain multiplipication of Co and put a similar value as a variable C between the input and emitter (I think it was a bipolar?) and null the THD with the C trimmer...... Depending on the Rload resistor value etc the null would be broad or narrow....
If that is the case, you can have ps voltage of two different values - no problem as long as you have plenty of margin so you don't clip the signal. [Then don't need to worry as much with matching
] A differential input stage with same type devices would require added C cross-coupled at the output to cancel the C's. I mentioned this near the beginning. Called Neutralizing technique. It also works well. To do it in Compound-Complimentary Push-Pull (or CFA), the cancellation comes naturally. Just need to get close the same C's (matching?) and some tweaking of added C or matching devices closer. Then throw in little amount of NFB to clean up the remainder and provide other bennies. These are all alternative to high nfb to clean up everything... not that its bad - just and alternaive and it is simple, cheap and uses few parts.
The techniques of distortion cancellation rather than reduction, is not new but with this symmetrical topology it is easier in my sense of the word. Maybe not any lower than other methods but its simple (OK sophisticated, if you prefer)
Subtle, for sure. But you can see why at higher and higher freqs - well beyond audio - it is very hard to reduce distortion thru just nfb alone. Thx-RNMarsh
Last edited:
Well many here are on a budget - they have families or what ever and this hobby comes from any time and money left over. If cost is not a concern then we would all have nice shiney new Audio-Precision instrumenets and the highest of hi quality in everything. Of course it matters. And, some build because it is the only way they can enjoy better sound at the price they can afford.
In industry it also matters as size means cost increases... larger realestate from larger pcb, larger box, etc. You can reduce the parts count and make the remaining parts smaller, of course.
But there is something else. A sort of challenge-- how sophisticated can we get... simplifying can not only save money but get results that are just as amazing as a circuit with many times the number of parts/transistors. There is a sense of elegance in doing that. As well as the challenge. How good can you make it with the fewest number of parts? Its my challenge to others and to my self. It turns out this topology does it for me in all these ways. is it better? that's never been the correct question for me.
Thx-RNMarsh
Last edited:
So, for me we have come full circuit.... in the beginning I said we would eventually get to the cancellation techniques... my favorite area. We are there. And, this topology has endured for a varity of reasons. Some not understood or appreciated. I hope you guys dig in deeper to the distortion cancelling techniques and find ways to apply them to audio circuits. Look towards the many patents in the area - esp the MOSFET topologies/designs/circuits... many found there.
Thx-RNMarsh and Cheers !!
xooxoxoxoxo
Thx-RNMarsh and Cheers !!
xooxoxoxoxo
To posit that its a requirement would be an error. Rather 'everything should be as simple as possible, but no simpler'. I build amplifiers and evaluate them by their subjective qualities and I tend to find a general pattern emerging, that simpler means better SQ. Until it gets too simple that is.
Indeed true, but missing the point as far as I can see.
If anyone's promoting compromised designs then by all means point out where compromises were made and suggest improvements. Engineering's not about compromising, its rather about optimizing, a totally different process.
Alas, most of the evils in simple CFAs are due to their symmetrical nature.Bigun said:
Requiring VFAs to be symmetrical would degrade one advantage they have over CFAs. With moderate complexity, VFAs have better Loop Gain and are easier to design for 1 ppzillion THD20k. Making them symmetrical would require FAR more complexity to achieve the same results and Loop Gain. Bob's reply to Bigun is pertinent.
I think Self deals with the evils of 'symmetrical' in his new book and I would love to be able to see his detailed pontificating on this.
I'm going to distinguish 'symmetrical' from Cordell's 'push-pull VAS' #1934 which is also one of my favourite topologies though not for his reasons.
kgrlee said:
Thanks for this Bob and I can only commend your very fair comments.Bob Cordell said:
Guilty mi'lord. If I may answer both Bob & Waly as I'm the main perp demanding Simplicity ..Waly said:
Firstly, I have a small but reliable amount of Blind Listening Test data that seems to suggest 'simple sounds better'. I certainly don't claim it always applies but it also suggests 'complex DUN sound better'.
Alas this was with VFAs, not CFAs.BTW, it was this series of tests which led to the choice of the 'push-pull VAS' topology mentioned above .. but often Engineering decisions are made for very mundane reasons like ease of repair etc.
Secondly, there are VERY large hidden costs for complexity. As only one example, the need for and ease of rework on a PCB (and also its reliabilty) is exponentially related to the number of solder joints .. arguments as to the exact relation to the QA gurus please.
I'm very old fashioned in this and have had young engineers on the carpet cos their 'new' design had more parts than what it replaced.
In these SMT days and microscopic bits, the DIYer faces even more problems from complex circuits .. simply just checking joints & correct parts.
Thirdly, once you get below 100ppm THD20k, layout, decoupling & earthing has a far greater effect on performance than your circuit and use of Golden Pinnae parts. My small brain has difficulty doing optimum layouts even with my simplistic circuits .. let alone more complex ones.
Lastly, if I can get less than 1ppm THD20k 50W, good stability, good PSR bla bla with a very simple circuit, I need to see a lot more good stuff to justify more complexity. Many complex circuits do worse than my naive, simple & simplistic circuits is all respects.
Bob brings up the subject of tweaks. IMHO, a stable tweak is OK for DIY and possibly even in Production. The offset tweak in VSSA is of this sort. To get a good handle on these issues, you really need a small scale production batch. This is my biggest practical worry on CFAs at present.
But if Bob uses his beloved FET i/ps, the easiest solution to offset is a DC servo and this also solves the problem for CFAs.
Alas, the OPS bias current tweak is often a long drawn out process. Both Self & Bob deal at length with Thermal stability and I've done some work myself in da old days but I don't think the final solution is out yet. I'd like to play with some simple 'servoed' solutions but the requirement is NOT simple.
Foul Bob! I'm going to call 'religious dogma' on this and your evil assertions about Class B IPS. But who am I to complain after singing Palestrina's longest setting of my Credo?Bob Cordell said:
Waly, Esperado's and other's simple enhancements to VSSA improve its PSR a lot. The resultant circuit is still 'simpler' than the equivalent VFA.
Member
Joined 2009
Paid Member
For me the attraction to 'simple' circuits is not based on reliability, pcb re-work or expense. I find the cost of my projects is mostly my time, my patience and the loss of half of my basement. I like simple circuits because they look nice when they are drawn out. I like that they are easier to understand and explore in terms of the operation, pcb layout is usually much easier too.
From what I understand, you can only sum currents at a simple node - two electron flows become one. Then in order to amplify the result you need to convert the current to a voltage referenced to a 'signal ground' because all active amplifying devices (e.g. transistors, triodes, FETs) operate based on an input which is a differential control voltage (e.g. base-emitter, grid-cathode, gate-source).
If instead of operating with currents you want to subtract two voltages then you can apply them directly to the control inputs of an active device.
All the error amplifiers I see in feedback amplifiers must ultimately operate in the voltage domain because that is the physics of their operation.
[p.s. the way I see it: bipolar devices are not current operated devices, but they have low input impedances compared with other active devices and must be driven from a source that can provide enough current to maintain the control voltage being applied]
From what I understand, you can only sum currents at a simple node - two electron flows become one. Then in order to amplify the result you need to convert the current to a voltage referenced to a 'signal ground' because all active amplifying devices (e.g. transistors, triodes, FETs) operate based on an input which is a differential control voltage (e.g. base-emitter, grid-cathode, gate-source).
If instead of operating with currents you want to subtract two voltages then you can apply them directly to the control inputs of an active device.
All the error amplifiers I see in feedback amplifiers must ultimately operate in the voltage domain because that is the physics of their operation.
[p.s. the way I see it: bipolar devices are not current operated devices, but they have low input impedances compared with other active devices and must be driven from a source that can provide enough current to maintain the control voltage being applied]
Last edited:
I plainly agree.
I do not see any algebraic sum of currents but I clearly see an algebraic sum of voltages in the feedback process of CFAs.
In simple inputs à la JLH class A, the output current of the input stage is mainly determined by the transconductance of the input transistor which depends on the differential voltage between its base and its emitter (Vbe).
Waly
For me personally it does not matter what type of negative feedback . It is better when she's not at all like a vacuum tube amplifiers .
Then the following requirements on the speed of the loop NFB , the bandwidth .
Even better when the output stage has a negative output impedance at minus 0.25 ... 0.3 ohms. In this case, the best control of the diffuser at the amplifier .
The output stage in my opinion it is best to do in class AB with a corrector Hauksforda , it allows for some tuning to get a negative output impedance .
Operation of the amplifier in Class A, reminds me of the struggle between two sumo wrestlers are simultaneously awkward draw a thick felt-tip pen waveform audio . A driver can be performed by any simple scheme , such as a symmetrical folded cascode with the output buffer . If the tube amplifiers provide good sound is distorted more than 2 percent, with such a structure can be obtained distortion is not higher than 0.05 %, and the sound quality unattainable any other structure.
[FONT=Arial, sans-serif]http://www.rowen.ch/en/poweramp.php[/FONT][FONT=Arial, sans-serif] [/FONT]
best regards
Petr
For me personally it does not matter what type of negative feedback . It is better when she's not at all like a vacuum tube amplifiers .
Then the following requirements on the speed of the loop NFB , the bandwidth .
Even better when the output stage has a negative output impedance at minus 0.25 ... 0.3 ohms. In this case, the best control of the diffuser at the amplifier .
The output stage in my opinion it is best to do in class AB with a corrector Hauksforda , it allows for some tuning to get a negative output impedance .
Operation of the amplifier in Class A, reminds me of the struggle between two sumo wrestlers are simultaneously awkward draw a thick felt-tip pen waveform audio . A driver can be performed by any simple scheme , such as a symmetrical folded cascode with the output buffer . If the tube amplifiers provide good sound is distorted more than 2 percent, with such a structure can be obtained distortion is not higher than 0.05 %, and the sound quality unattainable any other structure.
[FONT=Arial, sans-serif]http://www.rowen.ch/en/poweramp.php[/FONT][FONT=Arial, sans-serif] [/FONT]
best regards
Petr