Nice one, andy_c!
This is exactly the kind of response that can make this a great place to learn.
These results pose the question: will this have a significant effect on the sound of the overall amplifier? One would expect that the trend towards odd-order harmonics would make the symmetrical design "sharper" and less warm, however this is the opposite to the subjective experience I have had. Hmmmm......I'm not convinced the harmonic content of the distortion is the main cause of the perceived difference in sound quality....
Hugh, what do you you think about the fact that the symmetrical design required much smaller miller caps to achieve the same bandwidth and how does this tie in to your theories?
This is exactly the kind of response that can make this a great place to learn.
These results pose the question: will this have a significant effect on the sound of the overall amplifier? One would expect that the trend towards odd-order harmonics would make the symmetrical design "sharper" and less warm, however this is the opposite to the subjective experience I have had. Hmmmm......I'm not convinced the harmonic content of the distortion is the main cause of the perceived difference in sound quality....
Hugh, what do you you think about the fact that the symmetrical design required much smaller miller caps to achieve the same bandwidth and how does this tie in to your theories?
Andy, nice work. Could you email me the file please or could you upload it?
pa@mbox331.tele2_remove_this.se
Interesting results but how about dist for large signal AND complex load? I think the simulation only can give you a hint about real performance. As I read in the thick brick book from AD by Walter Jung (among others), "simulation must not replace real prototyping" or something like that.
pa@mbox331.tele2_remove_this.se
Interesting results but how about dist for large signal AND complex load? I think the simulation only can give you a hint about real performance. As I read in the thick brick book from AD by Walter Jung (among others), "simulation must not replace real prototyping" or something like that.
Since we talk distortion: 0.1% or 0.03% is this important figures in this discussion (I have Pass' amp in my here)?
How about adding a buffer at last? Improves performance? Adding a cascode at the input? Adding a cascode at the VAS stage?
I have no opportunity to test this myself at the moment, Macintosh you know.
How about adding a buffer at last? Improves performance? Adding a cascode at the input? Adding a cascode at the VAS stage?
I have no opportunity to test this myself at the moment, Macintosh you know.
Hey, hey, hey, Hugh! Don't jump into conclusions so fast.AKSA said:
This is only a simulation of a no real world circuit!
You could in fact simulate your whole amp because it doesn't contain so many parts and I think most parts has SPICE parameters. ...you could ask Andy...if you don't want to try yourself.
I have tested SwitcherCAD and it's not the coolest program but it works and it's for free.
BTW: Thanks Andy for inspiring me to take the step into simulation and making the CFB amp. It gave me some visualization (how do you spell?) of some parts and what they do to the performance.
There are many instruments that generate odd order harmonics. 3rd order is not that very much less musical than 2nd: 2nd is an octave and 3rd is an octave and a fith which is still very musical. So 2nd and 3rd are both not that uneasy on the ear.
Regarding IMD however, there is a significant difference between amps that mostly generate 2nd and 3rd order harmonics.
Regards
Charles
Regarding IMD however, there is a significant difference between amps that mostly generate 2nd and 3rd order harmonics.
Regards
Charles
Lithops optica, living stone from South AfricaSteve Eddy said:
I have got a couple of emails asking the same. Haven't you checked my homepage and seen what I have as a hobby?
http://home5.swipnet.se/~w-50719/kaktus/lithops.html
http://home5.swipnet.se/~w-50719/kaktus/
http://www.diyaudio.com/forums/showthread.php?postid=203061#post203061
http://www.diyaudio.com/forums/showthread.php?s=&threadid=17535&highlight=lithops
http://www.mesembs.com/lithops/index.htm
http://www.lithop.supanet.com/
http://lithops.info/
http://www.geocities.com/RainForest/Canopy/6378/
http://users.skynet.be/fhoes/rsasucculents
Re: Amplifier topology objective effects
I am confused now. The discussion I had with Pabo was about output stages (at least he alluded to PNP-NPN combined stages at takeover etc.). Andy's simulation is a great insight, but of course is aimed to the difference between single ended and symmetrical INPUT stages.
When we talk about SE amplifiers, do we not normally mean SE OUTPUT stages? When Hugh talks about SE and others about tube amps, don't they not mean OUTPUT stages?
So this doesn't say anything at all about the difference in 3rd harmonic content of SE or symmetrical output stages. I'm still waiting on Pabo's reply to my question.
Jan Didden
andy_c said:
I am confused now. The discussion I had with Pabo was about output stages (at least he alluded to PNP-NPN combined stages at takeover etc.). Andy's simulation is a great insight, but of course is aimed to the difference between single ended and symmetrical INPUT stages.
When we talk about SE amplifiers, do we not normally mean SE OUTPUT stages? When Hugh talks about SE and others about tube amps, don't they not mean OUTPUT stages?
So this doesn't say anything at all about the difference in 3rd harmonic content of SE or symmetrical output stages. I'm still waiting on Pabo's reply to my question.
Jan Didden
simulations
The question on the compensation cap is easy to answer: both designs have roughly the same open loop gain, the asymmetrical achieves the gain doubling through the current mirror, the symmetrical by having two input stages. The two 20 p caps of the symmetrical design are effectively in parallel, so the have the same effect as the single 47 p cap of the asymmetrical design.
I am quite puzzled by the results. Using the VCVS is a neat trick, and it is perfectly valid. Distortion is high, but this can probably be explained by the large amount of degeneration used. But I am still failing to understand why the symmetrical design generates large amounts of 3rd. Given the large amount of VAS degeneration, balancing of the input stages should be ok. I seem to remember those 340/350 do not have a very linear beta, and they are not all too complementary.
Maybe using a faster, more linear and more complentary pair will improve things? Or you could cascode the VAS and use decent small signal transistors as VAS?
On a side note: assuming a VAS beta of 80, the VAS will draw 6.25 µA from the input stage. This creates an impabalance of 0.8%, which is probably tolerable. But it would be interesting to see what happens when you load the second input transistor collector with a resistor to improve balance.
Regards,
Eric
The question on the compensation cap is easy to answer: both designs have roughly the same open loop gain, the asymmetrical achieves the gain doubling through the current mirror, the symmetrical by having two input stages. The two 20 p caps of the symmetrical design are effectively in parallel, so the have the same effect as the single 47 p cap of the asymmetrical design.
I am quite puzzled by the results. Using the VCVS is a neat trick, and it is perfectly valid. Distortion is high, but this can probably be explained by the large amount of degeneration used. But I am still failing to understand why the symmetrical design generates large amounts of 3rd. Given the large amount of VAS degeneration, balancing of the input stages should be ok. I seem to remember those 340/350 do not have a very linear beta, and they are not all too complementary.
Maybe using a faster, more linear and more complentary pair will improve things? Or you could cascode the VAS and use decent small signal transistors as VAS?
On a side note: assuming a VAS beta of 80, the VAS will draw 6.25 µA from the input stage. This creates an impabalance of 0.8%, which is probably tolerable. But it would be interesting to see what happens when you load the second input transistor collector with a resistor to improve balance.
Regards,
Eric
AKSA VAS?
Hi Hugh,
I am trying to understand your argument about the O/L gain dropping. I assume you are using a bootstrapped VAS? If it is not as per D. Self, could you please post a conceptual schematic?
Thank,
Eric
PS: One argument in favor of high slew rates is that, as you approach the slew rate limit, the input pair is driven farther away from balance and becomes quite non-linear.
Hi Hugh,
I am trying to understand your argument about the O/L gain dropping. I assume you are using a bootstrapped VAS? If it is not as per D. Self, could you please post a conceptual schematic?
Thank,
Eric
PS: One argument in favor of high slew rates is that, as you approach the slew rate limit, the input pair is driven farther away from balance and becomes quite non-linear.
Re: simulations
Eric,
I didn't bother to react on it, but please note that the symmetrical case has much LESS open loop gain: the collector load of the input stage is a low-value R (1.1k IIRC) while the assymetrical case is loaded by current sources. That alone should be enough to explain the differences in 3rd harminic. IOW, this doesn't prove anything at all.
Jan Didden
capslock said:
Eric,
I didn't bother to react on it, but please note that the symmetrical case has much LESS open loop gain: the collector load of the input stage is a low-value R (1.1k IIRC) while the assymetrical case is loaded by current sources. That alone should be enough to explain the differences in 3rd harminic. IOW, this doesn't prove anything at all.
Jan Didden
owdeo said:
Hi Owdeo,
could you provide a link to the schematic or try to explain how a symmetrical VAS can be driven from a single LTP?
The solutions I have tried to think of are not very elegant. The AC signal is clearly there on the collector of the second input transistor, but the base of the second VAS sits at a much lower DC level. A current source and a resistor would do the job, or a resistor network and a coupling cap, but it is always pretty messy...
Regards,
Eric
I guess this will be a long post........
1. SE (now that's a nice pair of initials! Better than PP, huh?)
I certainly understand your point about musical instruments. Many have huge odd order harmonics, such as oboe, and some trumpets. However, is it not the distribution - the spectrum - of these harmonics which confers the timbre by which we recognize different instruments? Therefore, something which alters the harmonic spectrum will change the timbre, and we will perceive it as no longer 'natural'? There lies the rub; neither of us is wrong, we are merely concentrating more on the harmonics and rather less on their distribution.
2. Owdeo, you saw Eric's (capslock) posting on the lag compensation. Since effectively the two bases of the VASs are in phase, we have two caps in parallel, giving roughly the same equivalent capacitance since there are two sources of voltage amplification. In this instance, however, there is the very real advantage of symmetrical effect on slew rate; a single ended voltage amplifier suffers from asymmetrical charge/discharge of the lag comp capacitor, and this effect primarily leads to different slew rates for positive and negative half cycles. Beyond a certain minimum, however, around 5V/uS is my benchmark, I don't think this is such a bad thing......
3. Charles (phase_accurate) has a good point concerning IMD. Intermodulation effects are caused by uncorrected non-linearities in the voltage amplification process, and seem also to derive from the feedback loop due to group delay.
4. Cascoding the VAS leads to a couple of problems. First, with the cascoding transistor directly connected to the collector of the driver, you often find a coruscation in the trough of the negative half cycle under full output conditions. Why this happens seems to be a peculiarly local phenomenon. It can be corrected with a small value resistor between the emitter/collector which drops typically 200mV, no more. At 10mA this is 20R, not large. Of course this is local degeneration, and the OLG drops. Second, the output of the cascoded VAS is higher, which makes it more susceptible to loading effects at the speaker. No free lunch...
5. Unfortunately a resistor in the collector of the inactive branch of a diff pair does not restore current balance. The only way to guarantee this is to ensure bias paths to each base are precisely equal, the two diff transistors are perfectly matched for beta AND Vbe, and output offset is zero. A good way to obviate matching is to use at least 300mV of emitter degeneration on both transistors, but this of course seriously reduces OLG. Actually, in my amps I have never found diff pair balance to substantially affect sonics, although I do go to some pains to ensure balance because it seems to sit well with conventional wisdom.
6. Jan is absolutely right when he points out that the fully complementary input/voltage amp topology has substantially less OLG. With tiny collector loads on the input transistor, and 200R degeneration on the VASs, OLG must be four or five times lower than the other topology using current sources. This will considerably reduce feedback correction, particularly as there is no large, slow output stage on this circuit. Actually, if the OLGs were the same, we would indeed be comparing apples and apples, and I would be surprised if the third harmonic of the complementary topology were not actually smaller than the asymmetrical topology.
7. Here's a diagram I have just scanned. I hope it comes through; posting images always confuses me in this forum.....
Top schemat is a complementary VA pair driven from a single input pair, using a cascoded, DC level shifter. This design is found in the legendary pro-audio BXR 300 from Fender, a robust, good design from the eighties producing 350W into 4R. The level shifter makes this possible, though it has its problems.
Lower schematic is a typical Self bootstrap, whereby the VAS is fed from two series resistors off the positive rail. The boostrap cap, somewhere from 100uF to 470uF, is connected to the middle point of these resistors, with its negative lead to the output. The two resistors are not quite equal, typically 3K3 to rail, and 4K7 to bias generator. You can clearly see that as the reactance of the bootstrap electro rises with frequency, it becomes less and less effective at providing a constant current source via the lower resistor. The result is that the VAS collector sees a constantly stifferning load, until at very high frequencies it has little OLG, exactly as required to meet the Bode-Nyquist criteria. This situation permits use of a smaller lag compensation cap, which is good for sound quality.
Here's hoping the jpg gets through!
Cheers,
Hugh
1. SE (now that's a nice pair of initials! Better than PP, huh?)
I certainly understand your point about musical instruments. Many have huge odd order harmonics, such as oboe, and some trumpets. However, is it not the distribution - the spectrum - of these harmonics which confers the timbre by which we recognize different instruments? Therefore, something which alters the harmonic spectrum will change the timbre, and we will perceive it as no longer 'natural'? There lies the rub; neither of us is wrong, we are merely concentrating more on the harmonics and rather less on their distribution.
2. Owdeo, you saw Eric's (capslock) posting on the lag compensation. Since effectively the two bases of the VASs are in phase, we have two caps in parallel, giving roughly the same equivalent capacitance since there are two sources of voltage amplification. In this instance, however, there is the very real advantage of symmetrical effect on slew rate; a single ended voltage amplifier suffers from asymmetrical charge/discharge of the lag comp capacitor, and this effect primarily leads to different slew rates for positive and negative half cycles. Beyond a certain minimum, however, around 5V/uS is my benchmark, I don't think this is such a bad thing......
3. Charles (phase_accurate) has a good point concerning IMD. Intermodulation effects are caused by uncorrected non-linearities in the voltage amplification process, and seem also to derive from the feedback loop due to group delay.
4. Cascoding the VAS leads to a couple of problems. First, with the cascoding transistor directly connected to the collector of the driver, you often find a coruscation in the trough of the negative half cycle under full output conditions. Why this happens seems to be a peculiarly local phenomenon. It can be corrected with a small value resistor between the emitter/collector which drops typically 200mV, no more. At 10mA this is 20R, not large. Of course this is local degeneration, and the OLG drops. Second, the output of the cascoded VAS is higher, which makes it more susceptible to loading effects at the speaker. No free lunch...
5. Unfortunately a resistor in the collector of the inactive branch of a diff pair does not restore current balance. The only way to guarantee this is to ensure bias paths to each base are precisely equal, the two diff transistors are perfectly matched for beta AND Vbe, and output offset is zero. A good way to obviate matching is to use at least 300mV of emitter degeneration on both transistors, but this of course seriously reduces OLG. Actually, in my amps I have never found diff pair balance to substantially affect sonics, although I do go to some pains to ensure balance because it seems to sit well with conventional wisdom.
6. Jan is absolutely right when he points out that the fully complementary input/voltage amp topology has substantially less OLG. With tiny collector loads on the input transistor, and 200R degeneration on the VASs, OLG must be four or five times lower than the other topology using current sources. This will considerably reduce feedback correction, particularly as there is no large, slow output stage on this circuit. Actually, if the OLGs were the same, we would indeed be comparing apples and apples, and I would be surprised if the third harmonic of the complementary topology were not actually smaller than the asymmetrical topology.
7. Here's a diagram I have just scanned. I hope it comes through; posting images always confuses me in this forum.....
An externally hosted image should be here but it was not working when we last tested it.
Top schemat is a complementary VA pair driven from a single input pair, using a cascoded, DC level shifter. This design is found in the legendary pro-audio BXR 300 from Fender, a robust, good design from the eighties producing 350W into 4R. The level shifter makes this possible, though it has its problems.
Lower schematic is a typical Self bootstrap, whereby the VAS is fed from two series resistors off the positive rail. The boostrap cap, somewhere from 100uF to 470uF, is connected to the middle point of these resistors, with its negative lead to the output. The two resistors are not quite equal, typically 3K3 to rail, and 4K7 to bias generator. You can clearly see that as the reactance of the bootstrap electro rises with frequency, it becomes less and less effective at providing a constant current source via the lower resistor. The result is that the VAS collector sees a constantly stifferning load, until at very high frequencies it has little OLG, exactly as required to meet the Bode-Nyquist criteria. This situation permits use of a smaller lag compensation cap, which is good for sound quality.
Here's hoping the jpg gets through!
Cheers,
Hugh
Re: Re: simulations
I guess I accepted Andy's statement about the open loop gains being equal without too much thinking. Right you are! And because of the large emitter degeneration on the VAS, this is essentially voltage drive, so the loss in open loop gain would be substantial, on the order of 5-10. We still see a reduction in even order harmonics which is a testimony to how well the cancellation works...
janneman said:
I guess I accepted Andy's statement about the open loop gains being equal without too much thinking. Right you are! And because of the large emitter degeneration on the VAS, this is essentially voltage drive, so the loss in open loop gain would be substantial, on the order of 5-10. We still see a reduction in even order harmonics which is a testimony to how well the cancellation works...
AKSA said:
Thanks for taking the time!
3) I think you may be missing Charles' point. IMD is just another manifestation of the same evil, i.e. a non-linear ampitude transfer curve, no matter which component causes it.
When there is 2nd or any other harmonic, there must be (at least in amps) a non-linear transfer curve. This means as soon as you play two tones at the same time, you will have IMD, i.e. a mixing of frequencies. These additional frequencies are, in general, not in an octave or quint relationship to either of the original tones, so they are probably unmusical.
4) Learned a new word today, coruscation. Fortunately, it was in my Random House unabridged
I have never experienced this kind of trouble. I used BC550C/560C drivers and reasonably fast Sanyo video transistors (SD600 & complemantary, if memory serves right). The main downside, in my eyes, of using a cascode is that you waste voltage swing.
5) You are describing the perfectionist's approach
If you get the resistor wrong by 20%, you still reduce the unbalance from 0.8% to 0.17%. How much unbalance is tolerable depends on the amount of LTP degeneration. With this much LTP degeneration as in this example, 0.8% might be perfectly ok.7a) Nice trick, does the phase inversion, too. However, this is essentially current controlled voltage drive, so it needs to be used with heavy VAS emitter degeneration. Does the 6000 use the same priciple?
7b) OK, you are relying on the quality of the electrolytic. Is there a problem if the builder uses too good a cap?
Regards,
Eric
janneman
The simulation is not quite what I would do in order to compare the two alternatives SE and push pull.
Although andy_c has actually changed the VAS stage from SE to push pull. Just because you connect the collectors together doesn't mean that it isn't a push pull. The transfer characteristics is the same but without voltage gain.
It would be better though if andy_c would isolate the simulation to a single stage for example a SE emitter follower with a push pull emitter follower.
Load the stages pretty hard so that the simulation program doesn't skip harmonics beause of their low levels.
The simulation is not quite what I would do in order to compare the two alternatives SE and push pull.
Although andy_c has actually changed the VAS stage from SE to push pull. Just because you connect the collectors together doesn't mean that it isn't a push pull. The transfer characteristics is the same but without voltage gain.
It would be better though if andy_c would isolate the simulation to a single stage for example a SE emitter follower with a push pull emitter follower.
Load the stages pretty hard so that the simulation program doesn't skip harmonics beause of their low levels.
Re: Re: simulations
Note that I computed the distortion at 20 kHz, which is well above the open-loop pole frequency. Given that fact, the load impedance seen by the collectors of the input diff amp is dominated by the input capacitance of the Miller integrator, not the load resistor.
janneman said:
Note that I computed the distortion at 20 kHz, which is well above the open-loop pole frequency. Given that fact, the load impedance seen by the collectors of the input diff amp is dominated by the input capacitance of the Miller integrator, not the load resistor.
Re: Re: Re: simulations
If you read my post carefully again, you'll see that I made no such statement at all. I did, however, say that the closed-loop bandwidths of the two configurations were made equal. This implies the gain-bandwidth products of the two configurations are the same. But it is quite true that the DC open-loop gain of the complementary configuration is lower than that of the single-ended configuration. It's also irrelevant to the distortion with a 20 kHz input, since this is well above the open-loop pole frequency, and the Miller integrator input impedance dominates the load seen by the collectors of the input diff amp.
Tonight when I get home from work I'll look at the loop gains of the two configurations at 20 kHz. I'll do some fine tweaking of the compensation caps of the complementary configuration to get them equal within, say 0.1 dB. Then I'll repost the results.
capslock said:
If you read my post carefully again, you'll see that I made no such statement at all. I did, however, say that the closed-loop bandwidths of the two configurations were made equal. This implies the gain-bandwidth products of the two configurations are the same. But it is quite true that the DC open-loop gain of the complementary configuration is lower than that of the single-ended configuration. It's also irrelevant to the distortion with a 20 kHz input, since this is well above the open-loop pole frequency, and the Miller integrator input impedance dominates the load seen by the collectors of the input diff amp.
Tonight when I get home from work I'll look at the loop gains of the two configurations at 20 kHz. I'll do some fine tweaking of the compensation caps of the complementary configuration to get them equal within, say 0.1 dB. Then I'll repost the results.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.