AKSA said:
Hugh,
Thanks for your very kind comments. Your easygoing approach is truly a breath of fresh air in this forum. Feel free to use any of the results in any way you wish. I'm actually an advocate of complementary designs, so it wasn't my purpose to prove or disprove anything really. I just wanted to gain more insight into the nature of the problem.
I'm not really sure this demonstrates the superiority of the SE VAS/input stage combo (see Steve Eddy's comments for example), since the total distortion of the two configurations is about the same. But it's great fodder for holy wars and other forms of confrontation!
andy_c said:I'm not really sure this demonstrates the superiority of the SE VAS/input stage combo (see Steve Eddy's comments for example), since the total distortion of the two configurations is about the same. But it's great fodder for holy wars and other forms of confrontation!
Wassat? You talkin' 'bout me, punk?
se
Hi Andy,
you are right, you talked about closed loop bandwidth implying that the GBPs were equal. On closer inspection, however, I think that even this was not the case.
The Miller cap only defines the frequency at which the open loop gain begins to roll off at 6db/oct. Now, if you change the gain of another stage, and this stage has a much larger bandwidth than the VAS, you change both the GBP and the loop gain at 20 kHz.
Assume you place the dominant pole at 200 Hz. If you have a DC gain of 100 dB, the loop gain at 20 kHz is 60 dB. If you only have a DC gain of 92 dB, loop gain at 20 kHz will be 52 dB.
Regards,
Eric
you are right, you talked about closed loop bandwidth implying that the GBPs were equal. On closer inspection, however, I think that even this was not the case.
The Miller cap only defines the frequency at which the open loop gain begins to roll off at 6db/oct. Now, if you change the gain of another stage, and this stage has a much larger bandwidth than the VAS, you change both the GBP and the loop gain at 20 kHz.
Assume you place the dominant pole at 200 Hz. If you have a DC gain of 100 dB, the loop gain at 20 kHz is 60 dB. If you only have a DC gain of 92 dB, loop gain at 20 kHz will be 52 dB.
Regards,
Eric
capslock
andy_c said that the closed loop bandwidth is the same in both cases and the closed loop gain is the same.
When the closed loop gain starts to fall off, the open loop gain equals the closed loop gain. This means that if you have a -1 slope during this time the feedback factors are equal in both topologies until the "DC gain" is reached (if we go backwards in the frequency domain).
Therefore I must say that I agree with andy_c and his ways to test the topologies.
Am I wrong?
andy_c said that the closed loop bandwidth is the same in both cases and the closed loop gain is the same.
When the closed loop gain starts to fall off, the open loop gain equals the closed loop gain. This means that if you have a -1 slope during this time the feedback factors are equal in both topologies until the "DC gain" is reached (if we go backwards in the frequency domain).
Therefore I must say that I agree with andy_c and his ways to test the topologies.
Am I wrong?
peranders
That is exactly what has been discussed during the last ten replies or so
You are right about the positive effect on distorsion because it increases the gain but andy_c was thorough enough to make sure that the feedback factor at high frequencies, at which his simulation was performed, was the same for both stages. At high frequencies the collector resistance (instead of the current mirror) does not load the diffpair in the same order as the dominant pole does
That is exactly what has been discussed during the last ten replies or so
You are right about the positive effect on distorsion because it increases the gain but andy_c was thorough enough to make sure that the feedback factor at high frequencies, at which his simulation was performed, was the same for both stages. At high frequencies the collector resistance (instead of the current mirror) does not load the diffpair in the same order as the dominant pole does
peranders said:
Hmmm... Maybe I should email Dr. Leach and tell him his amp isn't a real-world circuit too?
His diff amps have 300 Ohm emitter degeneration resistors (that's where I got the values) and 1.2 k collector load. And his VAS emitter degeneration resistors are 390 Ohms, nearly twice the value of my circuit. So his DC open-loop gain is less than mine. It's the gain-bandwidth product gm/(2*pi*Cdom) that matters. I think having been built over a period of well over 20 years by hundreds, maybe thousands of people qualifies his amp as "real world".
Oki doki. Maybe I was sleeping.Pabo said:peranders
That is exactly what has been discussed during the last ten replies or so
You are right about the positive effect on distorsion because it increases the gain but andy_c was thorough enough to make sure that the feedback factor at high frequencies, at which his simulation was performed, was the same for both stages. At high frequencies the collector resistance (instead of the current mirror) does not load the diffpair in the same order as the dominant pole does
But if we really want to compare, the circuits must be exactly the same, just doubled in the last case.
What does the symbol stand for in the schematics which looks like a ideal buffer (feeding the feedback network)?
About the last remark about realworld amp. If gain is only 3 in the first stage, the noise performance isn't optimized, is it? Ok, many have built the Leach amp so I'll guess I should rephrase maybe.
Wow, this is really getting interesting....
Eric, the 6000 schematic referred to can be found on the first post of the thread "What do you think of this schematic" on this forum. The conversion from SE to PP is done with an extra stage, as opposed to the neat way that Hugh sketched in his last post.
Thanks for clearing up the issue of compensation caps guys, I hadn't thought about it enough. I do think Peranders has a point that the two circuits simulated by andy_c were not compared on a totally "even playing field", in that the SE circuit should have a simple resistor collector load rather than a current mirror. The LTP of the SE cct would be better balanced as a result. But the results obtained were valid in that the composition of the distortion products are different between the two configurations, and this will in theory not change if the current mirror is removed from the SE cct (I think....)
andy_c, I agree with your comment to Hugh that the simulation doesn't prove the superiority of SE designs. It only shows that using asymmetrical will give you a different "flavour" of sound over symmetrical. Don't blame you Hugh for deciding on that flavour but I don't think we can say that one is superior to the other. Also the point about spectral composition of distortion to me still proves that as amplifier designers we should be aiming for the lowest possible THD (a point that andy_c alluded to I think?), with the actual harmonic distribution as a secondary concern (perhaps).
Again, I maintain that the sound of an amplifier is not purely determined by its distortion products. The Leach amp is not low distortion and yet sounds extremely uncoloured and pleasant (and does not produce simple 2nd and 3rd harmonic...). I suppose looking at the distortion products is the easiest way to objectively compare asymmetrical with symmetrical ccts....but not the whole story.
BTW, the 6000 amp produced 0.005% THD measured at 10Hz, 1kHz and 10kHz, at all power levels from 1W to 200W/8R. Interestingly constant.....the original article shows full power sine wave at 100kHz and 10kHz square, cro snapshots clean as....
Eric, the 6000 schematic referred to can be found on the first post of the thread "What do you think of this schematic" on this forum. The conversion from SE to PP is done with an extra stage, as opposed to the neat way that Hugh sketched in his last post.
Thanks for clearing up the issue of compensation caps guys, I hadn't thought about it enough. I do think Peranders has a point that the two circuits simulated by andy_c were not compared on a totally "even playing field", in that the SE circuit should have a simple resistor collector load rather than a current mirror. The LTP of the SE cct would be better balanced as a result. But the results obtained were valid in that the composition of the distortion products are different between the two configurations, and this will in theory not change if the current mirror is removed from the SE cct (I think....)
andy_c, I agree with your comment to Hugh that the simulation doesn't prove the superiority of SE designs. It only shows that using asymmetrical will give you a different "flavour" of sound over symmetrical. Don't blame you Hugh for deciding on that flavour but I don't think we can say that one is superior to the other. Also the point about spectral composition of distortion to me still proves that as amplifier designers we should be aiming for the lowest possible THD (a point that andy_c alluded to I think?), with the actual harmonic distribution as a secondary concern (perhaps).
Again, I maintain that the sound of an amplifier is not purely determined by its distortion products. The Leach amp is not low distortion and yet sounds extremely uncoloured and pleasant (and does not produce simple 2nd and 3rd harmonic...). I suppose looking at the distortion products is the easiest way to objectively compare asymmetrical with symmetrical ccts....but not the whole story.
BTW, the 6000 amp produced 0.005% THD measured at 10Hz, 1kHz and 10kHz, at all power levels from 1W to 200W/8R. Interestingly constant.....the original article shows full power sine wave at 100kHz and 10kHz square, cro snapshots clean as....
After reading some of the replies here, I decided it would be a good idea to check the loop gains of the two circuits at 20 kHz to see if they matched. To some extent, the closed-loop bandwidth can depend on the non-dominant poles. If these aren't the same between the two configurations (and in general they won't be), matching the -3 dB bandwidths won't exactly match the loop gains at 20 kHz. For the SE design, the loop gain was 26.93 dB. For the complementary design, the loop gain was 26.53 dB. So I tweaked the compensation cap of the complementary design and re-measured the distortion. The results were within about 0.1 dB of the previously reported results, so I decided not to show them.
I've attached the LTSpice project files, along with the required transistor models so that others can experiment. One idea was to use a resistive load on the diff amp of the single-ended design. My reason for using a current mirror was to have nominally equal slew rates for the two designs, as well as equal currents in their diff amps. Having a resistive load on the SE design will give it a nominal slew rate of half that of the complementary design. In other words, there's always something to complain about with the comparison. You'd also need to cut the compensation cap of the SE design in half to keep the same gain-bandwidth product if you went with a resistive load. The loop gain can be measured via the voltage source and named node "LG" I've placed in the feedback loop for that purpose. The loop gain is V(out)/V(LG).
At any rate, that's all I'm doing with this particular experiment. I'm sticking with a fully complementary design for my own project, but someday I may try a SE design as well. The slew rate for the SE design is asymetrical, and the lower of the two values is quite a bit lower than the theoretical. Self has a capacitor fix for this in his book, but it strikes me as a kluge because its value is dependent on device parameters.
More info on the FFT technique can be found by going back through the archives of the Yahoo LTSpice user's group (actually a mailing list). That's a fantastic group, and you can read it on the web like a newsgroup by subscribing to it with the "no email" option.
I've attached the LTSpice project files, along with the required transistor models so that others can experiment. One idea was to use a resistive load on the diff amp of the single-ended design. My reason for using a current mirror was to have nominally equal slew rates for the two designs, as well as equal currents in their diff amps. Having a resistive load on the SE design will give it a nominal slew rate of half that of the complementary design. In other words, there's always something to complain about with the comparison. You'd also need to cut the compensation cap of the SE design in half to keep the same gain-bandwidth product if you went with a resistive load. The loop gain can be measured via the voltage source and named node "LG" I've placed in the feedback loop for that purpose. The loop gain is V(out)/V(LG).
At any rate, that's all I'm doing with this particular experiment. I'm sticking with a fully complementary design for my own project, but someday I may try a SE design as well. The slew rate for the SE design is asymetrical, and the lower of the two values is quite a bit lower than the theoretical. Self has a capacitor fix for this in his book, but it strikes me as a kluge because its value is dependent on device parameters.
More info on the FFT technique can be found by going back through the archives of the Yahoo LTSpice user's group (actually a mailing list). That's a fantastic group, and you can read it on the web like a newsgroup by subscribing to it with the "no email" option.
capslock said:
In the first case, your gain-bandwidth product (DC open-loop gain times open-loop bandwidth) is 20 MHz. In the second case, your gain bandwidth product is 7.96 MHz. You've chosen equal open-loop bandwidths but unequal DC open-loop gains. That gives unequal gain bandwidth products. For equal gain-bandwidth products, the amp with the lower DC open-loop gain will have a higher open-loop bandwidth such that the product is the same.
In the case of my simulation, I did an AC small-signal analysis, tweaking the compensation caps of the complementary circuit until the closed-loop bandwidth was equal to that of the single-ended design. Since the closed-loop gains are equal between the two, this gives equal gain-bandwidth products which in turn gives equal loop gains at 20 kHz. I later refined this to measure the loop gain at 20 kHz directly and found the results to be within 0.4 dB.
Ok, if you did that, it's perfectly valid.
Seeing the compensation cap values being nearly equal, I figured you had just looked at the dominant pole frequency. And I suppose the reason those caps come out to be so similar is that you are right about the input impedance of the VAS dominating the LTP gain, rather than the collector impedance.
Regards,
Eric
Seeing the compensation cap values being nearly equal, I figured you had just looked at the dominant pole frequency. And I suppose the reason those caps come out to be so similar is that you are right about the input impedance of the VAS dominating the LTP gain, rather than the collector impedance.
Regards,
Eric
I still wonder what the result would be if the complementary version had current mirrors and 50pF compensation capacitors, like the single-ended version. Theoretically, one would expect this to give the same slew rate and gain-bandwidth product for both versions, and you avoid any distorted signal current which may flow through the bias resistors.
I mean, the base-emitter voltages of the second stage transistors (Q5 and Q6) will depend non-linearly on their collector currents, causing a distorted current to flow through the bias resistors R5 and R6, both below and above the open-loop pole frequency. I have no idea whether this effect is dominant, negligible or something in between.
I mean, the base-emitter voltages of the second stage transistors (Q5 and Q6) will depend non-linearly on their collector currents, causing a distorted current to flow through the bias resistors R5 and R6, both below and above the open-loop pole frequency. I have no idea whether this effect is dominant, negligible or something in between.
RFI is just one reason for bad sound.
Another is uncorrected non-linearity, another is short term instability, another is severe intermodulation. The causes are legion.
When I refer to single ended I mean the stage I am talking about. In the foregoing, this has meant a single transistor operating in Class A amplifying the voltage. I would not so much describe two voltage amplifiers working into a common output stage as push pull, but rather simply complementary.
Viewed in its basics, a single amplifying device will create asymmetrical distortion, viz sharpening at one half cycle, broadening at the other. This is primarily H2. A complementary VAS will create more symmetrical distortion, viz shape changes at the extremes of both half cycles. This is primarily H3.
Thus there should be a predominance of H3 in the complementary topology, almost from first principles. What is not clear is the effect on the sound as perceived by humans. I say, fully subjectively, that the single ended VAS sounds better. It's all a matter of religion.
What we now need (to calm the savage breast) is a definitive experiment (doubtless double blind) where the relative skewing of the harmonic content by these two different topologies is AB tested, or even ABX tested. Only then do we have a conclusion which is other than religious.
PSpice can't help here. Someone needs to build them, and everyone needs to listen. Operating conditions need to be debated and agreed to. Hmmm, tall order.
Cheers,
Hugh
Another is uncorrected non-linearity, another is short term instability, another is severe intermodulation. The causes are legion.
When I refer to single ended I mean the stage I am talking about. In the foregoing, this has meant a single transistor operating in Class A amplifying the voltage. I would not so much describe two voltage amplifiers working into a common output stage as push pull, but rather simply complementary.
Viewed in its basics, a single amplifying device will create asymmetrical distortion, viz sharpening at one half cycle, broadening at the other. This is primarily H2. A complementary VAS will create more symmetrical distortion, viz shape changes at the extremes of both half cycles. This is primarily H3.
Thus there should be a predominance of H3 in the complementary topology, almost from first principles. What is not clear is the effect on the sound as perceived by humans. I say, fully subjectively, that the single ended VAS sounds better. It's all a matter of religion.
What we now need (to calm the savage breast) is a definitive experiment (doubtless double blind) where the relative skewing of the harmonic content by these two different topologies is AB tested, or even ABX tested. Only then do we have a conclusion which is other than religious.
PSpice can't help here. Someone needs to build them, and everyone needs to listen. Operating conditions need to be debated and agreed to. Hmmm, tall order.
Cheers,
Hugh
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.