mikeks said:
I haven't been following the desensitizing discussion as closely as I should, but the role of Rs is intriguing. Let's assume that the whole EC output stage is fed from a low-impedance source, as might be the case with a VAS that has local Miller compensation feedback or some other form of local feedback. Indeed, for purposes here, assume it is fed by a voltage source. The impedance at the emitter of the emitter follower at the front end of the EC stage will then be just 1/gm of that transistor. It will be a pretty low value, but will fluctuate a bit with amount of error correction due to the changing collector current of the EC differencing transistor. This is where Rs is connected. Current flowing in Rs will act against the gm of the transistor to develop a voltage.
What is also interesting is that, if the EC is doing its job perfectly, there will be essentially no signal across Rs, since the output node will be perfectly tracking the input. I guess maybe the desensitizing issue comes into play when, indeed, the EC is not doing a perfect job, thus causing a signal to appear across Rs and thus causing a signal current to flow that is a reflection of the residual error.
It is a little less clear, in simple english, how the resulting current flow effects a further error reduction, thus desensitizing the circuit. Take the top half where we are driving the output positive. Let's say there is an error in that the output is not quite going as positive as it should (less than perfectly unity gain). That will cause a slight increase in the voltage across RS and thus in the current flowing in Rs. This in turn will cause a very slight drop in the voltage at the emitter of the input emitter follower. This would seem to be in the direction of decreasing the apparent voltage of the input signal, causing the output to perhaps drop a bit further. This would seem to be in the wrong direction of that desired. Did I get this right? Is this what the "flagrant violation" is?
Note also that DC current will flow in Rs, causing a further current flow in the emitter follower and consequent increase in its gm. Also, if other, perhaps relatively fixed, currents are drawn from this emitter follower (as in my circuit), its gm will be increased and the relative effect for a given value of Rs would seem to decrease.
Mike, am I on the right track here with this verbal explanation?
Cheers,
Bob
Hi, Anatech,
Let's see the upper half only. The input signal is entering 2 transistor, 210a and 240a. The actual output diamond buffer is 240a.
210a and 202a is the "correcting loop". It senses input (base of 210a) and output (emitor of 202a). If there is any difference sensed, it modulates the CCS value that goes to base of 250a via current mirror of 220a+230a.
Let's see the upper half only. The input signal is entering 2 transistor, 210a and 240a. The actual output diamond buffer is 240a.
210a and 202a is the "correcting loop". It senses input (base of 210a) and output (emitor of 202a). If there is any difference sensed, it modulates the CCS value that goes to base of 250a via current mirror of 220a+230a.
Hi lumanauw,
What I see is a diamond transistor created by 210a and 202a. The current through 202a is mirrored by 220a and 230a and possibly affects the input signal through 240a, also driving 250a and therefore 250b and coupled to 240b.
Round and round it goes. If you connected the emitters of 250a and 250b to the output signal output, it looks like a pair of diamond buffers with the second buffer (240, 250 pairs) modulated by the current drawn from the first one so that the second one is driven by current and voltage.
So what I'm saying is that 210a and 202a form the diamond trnasistor that modulates the other stuff.
I'm still lost on how it works for error correction. I feel like the answer is right there but I can't see it. Thanks Daniel.
-Chris
What I see is a diamond transistor created by 210a and 202a. The current through 202a is mirrored by 220a and 230a and possibly affects the input signal through 240a, also driving 250a and therefore 250b and coupled to 240b.
Round and round it goes. If you connected the emitters of 250a and 250b to the output signal output, it looks like a pair of diamond buffers with the second buffer (240, 250 pairs) modulated by the current drawn from the first one so that the second one is driven by current and voltage.
So what I'm saying is that 210a and 202a form the diamond trnasistor that modulates the other stuff.
I'm still lost on how it works for error correction. I feel like the answer is right there but I can't see it. Thanks Daniel.
-Chris
I'd like to wish all of you a very happy new year and a most prosperous and enjoyable 2007. I have greatly enjoyed our interactions on this thread over the last couple of months, and have learned a lot from your combined knowledge, experience, wisdom and inquisitiveness.
Happy new year,
Bob
Happy new year,
Bob
mikeks said:Experimentation reveals that this arrangement appears surprisingly insensitive to modest (~10%) imbalances in the error amplifiers.
This probably because the complementary error amps. tend to cancel such errors; this also appears to be attribute of Hawksford's scheme.
In fact the observed desensitivity is due to the use of two EC loops in parallel, which gives a maximum reduction in sensitivity of 50% over that pertaining to a single EC loop.
Thus Hawksford's equation here necessarily becomes:
0.5/(1+K4)
Bob Cordell said:
Hi Bob , let's make some noise...
When we talk about error correction , we are talking about trying to achieve a output similar to the input in every way (except gain).
My favorite method to access the overall linearity , is a adaptation of the David Hafler null test , but everybody here don't agree to much with the idea.
What is your opinion about the usefulness of this test , to measure overall linearity?
Here is a link for my test set up.
Post #8 at
http://www.diyaudio.com/forums/showthread.php?s=&threadid=13415
Kind regards.
Tube_Dude said:
Hi Bob , let's make some noise...
When we talk about error correction , we are talking about trying to achieve a output similar to the input in every way (except gain).
My favorite method to access the overall linearity , is a adaptation of the David Hafler null test , but everybody here don't agree to much with the idea.
What is your opinion about the usefulness of this test , to measure overall linearity?
Here is a link for my test set up.
Post #8 at
http://www.diyaudio.com/forums/showthread.php?s=&threadid=13415
Kind regards.
It's a good technique. I actually used the electronic version of the null technique to measure the very low distortion of my MOSFET power amplifier I did back in the early Eighties. The technique is referred to briefly in my JAES amplifier paper on my web site at www.cordellaudio.com.
I still use the technique to this day, having implemented it formally in a box I call the Distortion Magnifier. It is also described briefly on my web site. I demonstrated it at the RMAF workshops. It can easily achieve a 60 dB null on a power amplifier. It incorporates both amplitude and phase matching. In the box, I can add back in controlled amounts of the signal after the null, so that the box provides a controlled amount of distortion magnification by 20 dB or 40 dB. Thus thus gives my distortion instrumentation either 20 or 40 dB of increased dynamic range, either for spectral analysis or THD analysis, or both together. Using this, I can see down to -140 dB THD components at 1 kHz fundamental. I can see to below -120 dB for either THD20 or 19+20 kHz twin tone.
Bob
Bob Cordell said:
Hi Bob
Thanks for sharing!
The first time that I see your circuit schematic with error correction , was back in 1985 in a Siliconix Incorporated Mospower applications book . In that time we don't have Net and all the knowledge came from books and publications like Audio Amateur and others...
In your design what really impress me , was the performance of the full power 500 Khz square wave into a 8 Ohms load...
A very good definition, Distortion Magnifier!
Maybe we can add "All Types of Distortion Magnifier"