powerbecker said:
I think I did, somewhere along the previous 1000 posts or so
Here is something. I forgot exactly what it was, but IIRC: top to bottom: power amp 50W in 8 ohms, output stage only but now unloaded, Vas while driving loaded output stage.
This is with separate error correction inthe output stage and the Vas stage, NO global feedback.
Jan Didden
janneman said:
I think I did, somewhere along the previous 1000 posts or so
Here is something. I forgot exactly what it was, but IIRC: top to bottom: power amp 50W in 8 ohms, output stage only but now unloaded, Vas while driving loaded output stage.
This is with separate error correction inthe output stage and the Vas stage, NO global feedback.
Jan Didden
Thank you Jan,
it seems that the (or yours) EC work worse for the VAS then for the powerstage. Really no wonder because the VAS is slow.
My present thinking is, that the discussed kind of EC suffers (as NFB) also under limited bandwidth.
So there is possibly no real/big improvement against the use of common NFB !??
Heinz!
powerbecker said:
Yes you are right, in the sense that I think the common thing in the distortion of my amp is in the way the summers are implemented. They are the same in the Vas and the power stage so both show similar distortion characteristics. I hope that when I improve the summer(s) the whole thing improves, but I have no real ideas so far.
You are also correct that it looks not particularly better or worse then 'normal' global nfb amps. It was not my intend to improve on that, but trying to get to the same quality level without global nfb, and my results show it is possible. Of course I can try to further improve by applying extra loop gain and global nfb, but for the time being I like to see if I can improve the amp in its current topology and then publish it.
Jan Didden
Brian's Cordell 1984 o/p stage model v1
Here is version 1. I've used the actual passive parts values that Bob gave me from his 1984 board. Most of the small power transistors are right. For drivers I have chosen ZTX651 and ZTX751 as approximations to MPSU07 and MSPU57. The output FETs are Andy C's, substitutes for the original IRF132 and IRF9130. Output bias is about 150mA. I've included Zetex's model statements in text files.
I have included a voltage source V9 so that the output device feedback loop gain can be measured as -v(OUT)/v(FB). R18 is the bias adjust resistor (a pot in the real circuit). V8 can be used to measure the closed loop response as v(OUT)/v(IN). The output is loaded with an 8-ohm resistor R18. I have a cap placeholder C8 for checking stability. The output current source I1 can be used to measure output Z as v(OUT)/i(I1). V6 and V8 provide the +/-11V input offset that Bob describes in his paper.
Here is version 1. I've used the actual passive parts values that Bob gave me from his 1984 board. Most of the small power transistors are right. For drivers I have chosen ZTX651 and ZTX751 as approximations to MPSU07 and MSPU57. The output FETs are Andy C's, substitutes for the original IRF132 and IRF9130. Output bias is about 150mA. I've included Zetex's model statements in text files.
I have included a voltage source V9 so that the output device feedback loop gain can be measured as -v(OUT)/v(FB). R18 is the bias adjust resistor (a pot in the real circuit). V8 can be used to measure the closed loop response as v(OUT)/v(IN). The output is loaded with an 8-ohm resistor R18. I have a cap placeholder C8 for checking stability. The output current source I1 can be used to measure output Z as v(OUT)/i(I1). V6 and V8 provide the +/-11V input offset that Bob describes in his paper.
loop gain response
Here's the loop gain of Cordell v1, with v(IN)=0 and an 8-ohm load.
The LF gain is about 31dB, with a knee at about 75kHz. The unity gain frequency is 3.3MHz with 88 deg phase lag. The system is very stable.
This result is consistent with the actual measurements Bob made in terms of the reduction of THD when the feedback was applied (see post #743).
Bob, did you measure the loop gain of the actual circuit?
Here's the loop gain of Cordell v1, with v(IN)=0 and an 8-ohm load.
The LF gain is about 31dB, with a knee at about 75kHz. The unity gain frequency is 3.3MHz with 88 deg phase lag. The system is very stable.
This result is consistent with the actual measurements Bob made in terms of the reduction of THD when the feedback was applied (see post #743).
Bob, did you measure the loop gain of the actual circuit?
Attachments
Here's the evil capacitor load test. This is the feedback loop gain measured with C8=2uF. This may be of general interest to show what capacitive loads can do to feedback circuits.
The knee has been reduced to 64kHz and unity gain frequency is now 750kHz with 172 deg phase lag. This is very unstable and is the reason why an output inductor is essential in the real circuit.
The knee has been reduced to 64kHz and unity gain frequency is now 750kHz with 172 deg phase lag. This is very unstable and is the reason why an output inductor is essential in the real circuit.
Attachments
Now the overall transfer function into 8 ohms load, V(out)/V(IN) and the input Z, V(IN)/I(IN)
The -3dB point is about 15MHz.
The input Z is very high, being over 1 Mohm at 20kHz and is capacitive. The input Z drops to 3k at 10MHz and sees a second pole above this frequency.
The -3dB point is about 15MHz.
The input Z is very high, being over 1 Mohm at 20kHz and is capacitive. The input Z drops to 3k at 10MHz and sees a second pole above this frequency.
Attachments
So then the question is, what comprises the other? I threw around a couple of suggestions off the top of my head. Mr Cordell says PIM is not an issue (though some people at the diyhifi forums disagree with his assessment; I don't have the knowledge to judge for myself...); thermal distortion is another one I mentioned, but D. Self says it only is a problem with chip amps and not discrete designs... masking effects of the distortions of problem sources and speakers perhaps? But I think that can't be the only thing. What else is there? Is it possible to have types of nonlinear distortion that would not show up in THD measurements?
I know janneman will freak out again that I'm going off topic, but I think it's relevant.
Nixie said:
You're asking the right questions.
While I say that THD-20 is very important, and that a full IM spectral analysis is more important, these alone appear not to tell the whole story. Again, just because they don't tell the whole story, don't throw them out. These are static distortion tests on the lab test bench, much like most of the other proposed tests. I generally like to have very low THD-20, as long as it is not done at the expense of other performance, because it is harder for other distortions, like crossover and high-order IM, to sqeek through in a well designed amp where THD-20 is very low. It is also true that this approach may unfairly penalize more benign types of distortion like second and third order.
But if THD-20 and related traditional distortions told the whole story, why would things like capacitors and cables make a difference? Part of the answer lies in amplifier misbehavior that is not caught on the test bench, either because it is not stimulated there or because the loading is too benign. Power supply related shortcomings will often not show up on the test bench. EMI susceptibility may often not show up on the test bench. Parasitic oscillation bursts may often not show up on the test bench. Thermal distortions, to the extent that they are an issue, may often not show up on the test bench. Clipping behavior and protection behavior into real speaker loads will not show up on the test bench. The list goes on...
Bob
Continuing from here and here respectively, the equivalent circuits below are used to demonstrate that loop-transmission in the major-loop described by summers S1 and S2 (here op. amps. U1, U2), is very small, being much less than unity if an ideal output stage (E1) with exactly unity gain is used.
If non-linearity is introduced courtesy of a diode in series with the ideal output stage, then the magnitude of loop-gain increase to virtually unity, proving that loop-transmission is proportional to extracted error.
This is consistent with this analysis.
If non-linearity is introduced courtesy of a diode in series with the ideal output stage, then the magnitude of loop-gain increase to virtually unity, proving that loop-transmission is proportional to extracted error.
This is consistent with this analysis.
