john curl said:
John,
While I won't argue with many years of design and production experience - which I lack - I cannot but request some more specificity as far as possible without intruding on proprietary grounds.
If the superiority of discrete designs can be qualified by measurements, and the difference in results also is above perceptual thresholds for the best listeners, then I have no argument.
If the superioriy cannot be evaluated objectively by existing measurement protocols, but can be reliably established by reputed listeners, I may not be convinced but leaving space to future developments that could fill the void.
If the superiority boils down - and I respect this - to marketing strategies, than again I will not be convinced but have no qualms regarding to each one getting what he wants better and is willing to pay for.
I am not particularly happy with IC's, but have found them nice building blocks for designs that, objectively measured, and listened by moderately discriminating people, have been quite satisfactory.
Anyway, thanks for you response.
Rodolfo
JPV, don't take Dr. Leach's comments too literally The optimum Re drops 15-25mV of voltage at idle. This was proven by a VP of HP, back in 1971 or so, in the HP Journal, also using a computer simulation. I agree with Bob on an inverting servo with a second inverter. It is less problematic than a non-inverting servo. I know from experience.
As far as IC's are concerned, I like a high slew rate of 100V/us or more, and fully balanced driver stages. IC's have a problem with this.
As far as IC's are concerned, I like a high slew rate of 100V/us or more, and fully balanced driver stages. IC's have a problem with this.
ash_dac said:
It doesn't translate directly to GBW, since the gain depends both on input and output voltage. However, we may find the theoretical maximum amplitude bandwidth product at the amplifier output for sine waves by observing that
d/dt A sin(w t) = w A cos(w t)
That is a sine wave with amplitude A and ang. freq w needs a slew rate of at least w A. Hence, a slew rate of 100 V/us = 100 MV/s gives
w A = 2 pi f A = 10^8 => f A = 16 MVHz
So we could for instance output at most 1 V peak at 16 MHz, or 100 V peak at 160 kHz etc. To get GBW, you have to factor in also the gain of the amplifier.
Note that this is the theoretical maximum based on the slew rate figure alone.
JPV said:Concerning gm doubling
Mr Cordell, in case of complementary feedback pair it is perhaps right but in emitter follower pair like in the T topology, I do not understand gm doubling.
If you read the paper from Leach (http://users.ece.gatech.edu/~mleach/papers/classab.pdf) on gm doubling it seems undeniable with spice simulation, analytical demonstration and an explanation why the misconception .
He uses two matched transistors which is certainly not true in a real life but the argument remains.
If this is true, then the T circuit is really great not only because it loads very lightly like any darlington but also because the drivers are allways in class A ( emitters not connected to the load but between themselves through a resistor) makes them a very good low impedance to suck the charges out of the output stage. This is very good to reduce the remaining crossover spikes.
M. Curl, I understand that this topology is 40 years old but is it not the same for all of them
I am in the process of learning and I would love to understand what are the subtle distortion mecanismes that still exist in the output stage and how to solve them ( except layout of course)
When I say IC I mean front end like the LM4702. Experienced designers can certainly design discrete high voltage gain stages with good desensitivity and stability which is given for free by the LM4702 in the 100W+ ( I am not selling for National !!).
Mr Curl, what do you dislike in a specific IC like this one driving an output stage in this range of power
If all this is true, ( for the diy) making a medium power amplifier boils down to choosing between BGT and MOSfet for efficient reliability and protection mecanism. With use of servo and bridging, he can make a very high power with a BGT/MOSfet choice for ease of paralleling.
Mr Cordell, why do you prefer inverting servo and where can I read on the use of servo to remove offset.
Many Thanks to all of you
Jean-Pierre Vanderreydt
I chose the example of the CFP output stage to demonstrate gm doubling because it provides a more extreme example due to the very low output impedance of each individual CFP. However, the T circuit also demonstrates gm doubling if it is over-biased, or biased into Class AAB. It's just that in source follower output stages, the individual output impedance of each half of the source follower is a little higher and a function of current (gm of the output transistor), so proper choice of the bias current can result in a smoother crossover transition.
Neverthe less, even in a T circuit, if you bias the output stage at, say, 200 mA and use 0.33 ohm emitter resistors, you'll see gm doubling. Just note that the re of each output transistor at idle (1/gm) is only about 0.13 ohm, reasonably small compared to the emitter resistor. At idle, net output impedance is about (0.13+0.33)/2 =0.46 ohms. At high currents, where re is very small, output impedance is on the order of 0.33 ohms. Here, in this admittedly over-simplified case, gm has not doubled, but it has increased by a factor of 1.4.
I prefer the inverting servo to the non-inverting servo for at least two reasons. First, to make a non-inverting integrator, you have to use two good integrator caps. That alone is not worth the cost of a dual op amp over a single op amp. In some cases the inverting integrator makes it easier to use a smaller integrating capacitor as well. Secondly, the use of the inverting integrator followed by an inverter gives more design options and flexibility, and little or no worries about common mode effects. I usually put in a second pole to further reduce the effects of the servo in the audio band, and I also use the output of the inverting integrator as part of my dc output detect circuit.
Bob
We can't speak of output impedance of emitter follower when biasing current don't flow through it. We may speak only of the whole amp with all whistles and bells, and does not matter, are output transistors in CC or CE, the difference is only in the feedback path, if we speak of the amps with the same power gains. So, we can't say that emitter resistors are in series with output resistance, they are simply part of it that don't add to it proportionally. Didtortions are results of non-linearities of transistors, i.e. exponential laws, parasitics resistances and capacitances, etc... And emitter resistors if to be honest add to output resistance more than their value because they create a negative feedback by current (both by output and across both transistors) linearizing transfer function of each follower, but a global feedback by voltage decreases the output resistance significantly.
M. Cordell,
I understand your arguments but the development of Leach shows on the other hand that for low resistors in the emitter, there is no gm doubling. I need to mull over still.
It is important because no gm doubling ( in normal case of biasing), then no gm distortion.
What is left : non linear Vin Iout and Hfe. Is this not well handeled by the CC topology?. All this should let the overall feedback make a very efficient job if good front end.
MOS or BJT ( fast ones) should they not be very close from a distortion point of view?
Thank you for the servo info. Can you rely on this to protect your speakers?
Mr Curl, I suppose you are refering to a paper from Oliver?
Do you know how to get a copy?
JPV
I understand your arguments but the development of Leach shows on the other hand that for low resistors in the emitter, there is no gm doubling. I need to mull over still.
It is important because no gm doubling ( in normal case of biasing), then no gm distortion.
What is left : non linear Vin Iout and Hfe. Is this not well handeled by the CC topology?. All this should let the overall feedback make a very efficient job if good front end.
MOS or BJT ( fast ones) should they not be very close from a distortion point of view?
Thank you for the servo info. Can you rely on this to protect your speakers?
Mr Curl, I suppose you are refering to a paper from Oliver?
Do you know how to get a copy?
JPV
JPV, if you follow my advice and bias your emitter resistors to have 15-25mV across them, then so called Gm or Beta doubling will not be too much of a problem. However, you will find that many people tend to use more than 25mV across the emitter resistors at idle, and therfore suffer a real problem at low levels. They use larger value emitter resistors, because they think that it is easier to generate a stable bias, or because they don't know any better.
It is a real problem, because when the output stage is operating at class A, then both sides of the output stage are contributing to the output, and the open loop output impedance is 1/2 that of when only one device is contributing.
It was first solved by B. Oliver at HP back in about 1970, but it is not too important that you read the paper itself.
It is a real problem, because when the output stage is operating at class A, then both sides of the output stage are contributing to the output, and the open loop output impedance is 1/2 that of when only one device is contributing.
It was first solved by B. Oliver at HP back in about 1970, but it is not too important that you read the paper itself.
