OS wrote ""Oh where, oh where... to put them diamond collectors"
I tie them to the Zener regs. That way all 4 diamond buffer trannies see the same Vce and Ic and you can get really good thermal stability. I'm getting 1-2 mV. No servo needed and direct coupled which might explain the very good ( ok, I'm going to get subjective here - apologies!) bass sound.
I tie them to the Zener regs. That way all 4 diamond buffer trannies see the same Vce and Ic and you can get really good thermal stability. I'm getting 1-2 mV. No servo needed and direct coupled which might explain the very good ( ok, I'm going to get subjective here - apologies!) bass sound.
Mcd99uk wrote "The question is the feedback network sets the ULGF (My personal limit is 1Mhz) but can you set the compensation network (VAS shunt and 2 pole miller) to override this, is this a problem?"
I've done a lot more thinking about this, and concluded that if you force the ULGF to be the same as a VFA, you are probably creating a suboptimal design. I've done quite a few sims the last week or so snd will update my CFA vs VFA article accordingly.
Bottom line: in the OL condition, the CFA OP pole falls below the UG intercept. In a VFA it's falls above the UG intercept. To solve this problem in a VFA, you use dominant pole comp with a c 1-2 MHZ ULGF, or a variation (MIC etc). All of these comp schemes result in pole splitting that pushes the LF pole down in frequency, and the HF pole below the UG intercept, thus stabilizing the VFA.
In a CFA, you don't need dominate pole comp for the reason I noted above about the OPS pole. You can therefore close the loop at a higher ULGF - there is no requirement for pole splitting to ensure stability. The result from my 1st cut sims is that at 20 kHz, you can get an additional 12 dB of loop gain compared to an MC VFA.
Now, if you go trying to raise the OL gain of a CFA in the mistaken notion that more feedback is better, you end up with the OPS pole above the UG intercept ( in the open lop condition). In that case, you have to apply a heavy hand to stabilize the CFA, and you lose the benefits of wide loop gain BW, and closed loop BW.
This clearly explains why some CFA's can have their ULGF closed at 3-5 MHz ( MOSFET or 30 MHz OP trannies) and still remain stable. It also clearly explains why CFA's are easier to comp, and more tolerant in general of capacitive loads and have equal or lower distortion than VFA even though their OL gains are higher. So, in a CFA, less is more.
I've done a lot more thinking about this, and concluded that if you force the ULGF to be the same as a VFA, you are probably creating a suboptimal design. I've done quite a few sims the last week or so snd will update my CFA vs VFA article accordingly.
Bottom line: in the OL condition, the CFA OP pole falls below the UG intercept. In a VFA it's falls above the UG intercept. To solve this problem in a VFA, you use dominant pole comp with a c 1-2 MHZ ULGF, or a variation (MIC etc). All of these comp schemes result in pole splitting that pushes the LF pole down in frequency, and the HF pole below the UG intercept, thus stabilizing the VFA.
In a CFA, you don't need dominate pole comp for the reason I noted above about the OPS pole. You can therefore close the loop at a higher ULGF - there is no requirement for pole splitting to ensure stability. The result from my 1st cut sims is that at 20 kHz, you can get an additional 12 dB of loop gain compared to an MC VFA.
Now, if you go trying to raise the OL gain of a CFA in the mistaken notion that more feedback is better, you end up with the OPS pole above the UG intercept ( in the open lop condition). In that case, you have to apply a heavy hand to stabilize the CFA, and you lose the benefits of wide loop gain BW, and closed loop BW.
This clearly explains why some CFA's can have their ULGF closed at 3-5 MHz ( MOSFET or 30 MHz OP trannies) and still remain stable. It also clearly explains why CFA's are easier to comp, and more tolerant in general of capacitive loads and have equal or lower distortion than VFA even though their OL gains are higher. So, in a CFA, less is more.
Last edited:
Kgrlee - Didn't look at the overload conditions. I was just having a play with the sims at the time with no real purpose. Waly's post came to mind and had a quick experiment. When I have the PC up and running again I will try to answer your question.
Bonsai, thank you for your response. Already at single digit THD @ 20KHz in sims so not after more feedback. Was more worried about slew rate and using optimum resistor values in the diamond buffer as per Edmond's suggestion earlier. By following these recommendations, my basic understanding is that the ULGF will increase in my case as I have a shunt feedback resistor of 39R.
Still have the latest experiment to build. Fortunately have a paper copy of the schematic.
Will have a read of your VFA vs CFA paper later on.
Many Thanks
Paul
Bonsai, thank you for your response. Already at single digit THD @ 20KHz in sims so not after more feedback. Was more worried about slew rate and using optimum resistor values in the diamond buffer as per Edmond's suggestion earlier. By following these recommendations, my basic understanding is that the ULGF will increase in my case as I have a shunt feedback resistor of 39R.
Still have the latest experiment to build. Fortunately have a paper copy of the schematic.
Will have a read of your VFA vs CFA paper later on.
Many Thanks
Paul
Please post as soon as you do because the claims don't make sense to me.
Maybe the lack of plots to illustrate your point has lead me to misunderstand, but the comments seems to contradict feedback theory.
Kent Lundberg's paper on the "Control-centric View" of compensation is the best explanation I have seen.
Have you read it? It would make a useful common point of reference if you have.
Best wishes
David
David, did you get any chance to check on .asc file from my 200 W CFA thread?
BR Damir
No more debate, that's a much fairer summary of Waly's attitude😉
On it now.
Best wishes
David
Last edited:
We don't disagree, and I have said pretty much the same thing in my book. THD-20 is a good test, not because we can hear harmonics of it, but because it is a symptom of HF nonlinearity. I have generally stressed that 19k + 20k CCIF two-tone IM is the better test.
Note that, in the old days, it required an expensive spectrum analyzer to do 19+20k CCIF correctly, capturing the odd-order products that were spaced only 1kHz away from the signal components and each other. Sometimes people just looked at the even-order components, like 1kHz, with less expensive methods, but this was not very revealing. Back then, I developed a 3-tone test (MIM) that also sent odd-order products down to the lower frequencies where they could easily be measured. With modern PC-based spectrum analysis at hand, there is little excuse for not using 19kHz+20kHz testing.
Cheers,
Bob
Yes. And if HD is not such a problem, (at reasonable amount), because harmonics belong at any musical instrument, IM is much more damageable, as it add 'non harmonic' content, we can assimilate with modulated noises, blurring the signal.
To make a photographic comparison, HD is a little like color errors, while IM can be seen as noise or grain on a film that ruin the sharpness.
I remember Dolbys, modulating tape's hisses, was sometimes more annoying that the hisses itself.
Last edited:
In what way does it contradict feedback theory? I see no contradictions and no feedback rules are broken.
Hi -
This is all true about IM's affect. I wonder actually how important IM will be with an amplifier that produces less than -100dB THD ? It is very likely the IM will be extreamly low as well. If the IM is also very low or under -100dB also, then it is no more or less useful than THD.
I think a multi-tone tests - with as many tones as is humanly possible - is the best test. Because it is more like music signals and the total affect of the unwanted tones generated is more audible and more like the affect we would hear on music signals/sound. [3 tones is better but not enough.]
Thx-RNMarsh
Never thought about this, but, if we try F1+F2+F2, is the IM the same than the one produced by🙁F1+F2) +🙁Fi+F3)+🙁F2+F3) all alone ?
It can be DOUBLE the normal F1+F2, etc It is called a TRIPLE TONE. (or TRIPLE BEAT)
Last edited:
Bob, I know I've asked this before but I'd like to do so again in the hope that you might do the sums or get some keen mathematician to do them for your next edition.
It's to get the relation between each type of intermod product and the corresponding HD product. This is exact for any 'wideband' device .. 'wideband' being defined as flat to 200kHz for an audio device.
I did this circa 1980 (up to 7th order) when I kud kunt, reed en rite but it is now well beyond the ken of my single senile brain cell.
_______________________
Two listening tests have bearing on this subject:
http://sound.westhost.com/articles/intermodulation.htm Rod Elliot's very simple listening test suggests the pundits who claim, "2nd harm. good, 3rd & odd order harms. bad" are pontificating through the wrong orifice.
Peter Fryer's Intermodulation Distortion Listening Tests actually simulated only 1st order Intermod, ie the f1-f2 stuff that Rod says ISN'T generated by 3rd and odd order non-linearities. The important point about this series of tests, for us, is that the sound of this type of intermod is what most people would ascribe to amplifiers.
I was involved with Peter's Listening Tests.
A possible explanation for our listening test results might be that ...
.. its the lower frequency intermod products that are most evil and the sidebands around 19 & 20kHz are innocuous. 😱
More DBLTs indicated.
Hi OS and Paul,
Let me explain the CCS noise cancellation in more detail and hope it will answer your questions. But first of all, I address only two aspects of the diamond IPS, thus not the cascode (and recycling of its base current), TIS or OPS. So if you don't get an improvement of a whole amp, it is most likely that distortion or PSRR of these subsequent stages swamps or masks the improvement of the IPS. Also bear in mind that I have used perfectly matched PNP-NPN pairs. Using arbitrary models of real trannies can easily ruin the performance.
Below you see a slightly modified version of the 2nd IPS form post 3135. This one is the most easiest to analyze, as all four emitter resistors are equal, also the collectors of the first trannies are tied to the FB input and the two FB resistors replaced by equivalent parallel impedance (just for ease of calculation, later on, you can split them again).
The effect of noise or hum is simulated by injecting an error signal - current In - by means of VCCS G1*, parallel to the top CCS. Now let's follow this signal. It not only creates a voltage (Vn) at the base of Q4, but also propagates via Q1 to the feedback node, where it creates a voltage Vfb.
The conditions for cancellation of CCS noise are:
I4 = I5 (the AC component, of course).
This only is possible if I6 = IQ1 (=~In).
With a standing current of 1mA, the emitter resistors are determined by the condition for lowest distortion, i.e. RE = 0.5/gm = 13 Ohms (that is, at 27 degrees Celsius!!!) . So now we only have to calculate R6 (which equals Rfa || Rfb, the feedback resistors in a final design).
Vn = In (RE + 1/gm) = In.RE + In.2.RE = 3.In.RE (1)
I4 = (Vn - Vfb) / (RE + 1/gm) = (Vn - Vfb) / (3.RE) (2)
I5 = (Vn ) / (RE + 1/gm) = (Vn ) / (3.RE) (3)
From (2) and (3) it follows that Vn - Vfb = Vfb (4)
or Vn = 2.Vfb = 2.In.R6 (5)
From (1) and (5) it follows that 2.R6 = 3.RE
or R6 = 1.5 RE = 19.5 Ohms.
With this value and using my models, I7 (the output current) should be ~32dB lower than In.
Quite easy, isn't it? But if the collectors of Q1 and Q2 are tied to the emitters of Q3 respectively Q4, the calculation gets far more complicated. In that case I prefer to leave it to my simulator to figure that out (I'm too old for such old fashioned paperwork).
If you desire to implement these ideas, I urge you to first play with the circuit below and use complementary models. Later on, you can add a real TIS, OPS etc. and see what happens then.
Good luck.
Cheers, E.
PS: Next project: How to recycle the base current of the cascode trannie (a la Hawksford or Baxandall).
* Do you see this Mikeks-II? Just one noise source and no "two uncorrelated noise sources" blah blah blah 😛
Attachments
I do and I think it your analysis in #3135 is not correct.
a) Noise # hum.
b) The way you are modeling the noise in #3135 doesn't seem to be correct (a VCCS with zero volts input?).
Reasoning for the above is that hum adds linearly as voltages (or currents), while noise always adds linearly as power. You are doing an AC (linear) analysis, so adding the VCCS may model only the hum.
You can do an AC noise analysis in spice but then, for an apple to apple comparison, you have to calculate the input referred noise by defining the output, the noise source, and you also need to know the stage gain. It is unclear how you did this in #3135, also no (comparative and absolute) numbers for the noise performance are given.
P.S. I see you now "fixed" the noise source, by using a VCCS controlled by the input voltage in #3667. Not sure why you insist in using a VCCS for noise modeling (and I don't think this new way is anywhere correct). You can simply define, for noise analysis, the source at the input, the output at the output, and then calculate the input referred noise by knowing the AC gain.
Further, in #3135 you claim that Cob effect cancellation requires constant Vce. That is incorrect, it requires constant Vcb. Though I would agree that your cascode common emitter is theoretically better in cancelling the Cob. In your case, the Vcb variation is defined by the variation of a practical ~1.3V voltage source (like two forward biased diodes) with the current ( say +/-50mV), while mine keeps Vcb at the reference voltage (say 15V) +/- the input signal (say +/1V). That would make about 4% for yours vs. 7% mine. I'm saying theoretical because in my opinion, both variations are way to small to be of any practical importance.
Regarding the common base stage of the cascode, I could be missing something, but IMO a high Early voltage device is mandatory. Otherwise, the common base device would be subject to the same Early effect that the cascaded common emitter device is. What I'm saying is that e.g. cascoding two identical devices doesn't make much sense.
Finally, I think what you are proposing here is essentially an intellectual game. You mentioned yourself that this works only with perfectly matched transistors (modeled here as ideal CCCS), and temperature compensation is not yet considered. This, and the fact that I would not expect anyway to have the IPS dominate the distortions in the global picture, IMO makes this setup of little practical interest.
P.P.S. I thought I'm on your ignore list, but I see you are following me pretty close?!
Last edited:
Out of curiousity and to get an idea about the distortion, I've had a go at trying to change my wild fifth-order idea into something that at least works in LTSpice (but not in reality, as there are still plenty of loose ends).
20 kHz, 20 V peak into 8 ohm:
2nd harmonic: -133.9 dB with respect to the fundamental
24th and largest harmonic: -125.767 dB with respect to the fundamental
2 kHz, 20 V peak into 8 ohm:
2nd and largest harmonic: -153.647 dB with respect to the fundamental
Recovery from overload: it recovers, with some ringing
See the attachments. It is somewhat unpractical in its present form, for example because it is an inverting amplifier with 100 ohm input resistance, because it is excessively sensitive to mismatch and because the driver transistors aren't big enough, but I just wanted to check the idea.
The LT Spice and FFT settings were: no compression, reltol=1e-5, 2048 points per sine, 32768 total, first four sine waves not included in the FFT, rectangular window, no binomial smoothing.
20 kHz, 20 V peak into 8 ohm:
2nd harmonic: -133.9 dB with respect to the fundamental
24th and largest harmonic: -125.767 dB with respect to the fundamental
2 kHz, 20 V peak into 8 ohm:
2nd and largest harmonic: -153.647 dB with respect to the fundamental
Recovery from overload: it recovers, with some ringing
See the attachments. It is somewhat unpractical in its present form, for example because it is an inverting amplifier with 100 ohm input resistance, because it is excessively sensitive to mismatch and because the driver transistors aren't big enough, but I just wanted to check the idea.
The LT Spice and FFT settings were: no compression, reltol=1e-5, 2048 points per sine, 32768 total, first four sine waves not included in the FFT, rectangular window, no binomial smoothing.
Attachments
Its great to have accurate explanations of how a circuit works or to correct and add to the reason/explanations given.
There are a lot of models and ways to explain. Just as there are many backgrounds and levels of knowledge here. Deep device physics knowledge isn't necessary.... just a nice option. Engineering simplified formulas and descriptions that lead to circuits that get the desired results are needed.
Now, most of the math and models are included in SIM software..... we are left with old and new circuit topologies and the understanding needed for what it takes to improve given specs.
What was the question?
Thx-RNMarsh
There are a lot of models and ways to explain. Just as there are many backgrounds and levels of knowledge here. Deep device physics knowledge isn't necessary.... just a nice option. Engineering simplified formulas and descriptions that lead to circuits that get the desired results are needed.
Now, most of the math and models are included in SIM software..... we are left with old and new circuit topologies and the understanding needed for what it takes to improve given specs.
What was the question?
Thx-RNMarsh
Last edited: