phase_accurate
If you put a dominant pole in an active gainstage the feedback factor will increase with higher frequencies which means lower distorsion. A passive low pass filter has the opposite effect as it loads the previous gain stage at higher frequencies.
Of course you should allways put an input filter before the amplifier in order to eliminate rectification of HF-disturbances.
If you put a dominant pole in an active gainstage the feedback factor will increase with higher frequencies which means lower distorsion. A passive low pass filter has the opposite effect as it loads the previous gain stage at higher frequencies.
Of course you should allways put an input filter before the amplifier in order to eliminate rectification of HF-disturbances.
Maybe I was a little unclear with my older statement.
I did of course never mean to introduce any LPF INTO the amplifier. I was always talking of an input filter. The amp itself should always be "faster" than the signals that can pass the input LPF.
BTW: If you reduce the frequency of your dominant pole you don't incrase NFB, you reduce it.
Regards
Charles
I did of course never mean to introduce any LPF INTO the amplifier. I was always talking of an input filter. The amp itself should always be "faster" than the signals that can pass the input LPF.
BTW: If you reduce the frequency of your dominant pole you don't incrase NFB, you reduce it.
Regards
Charles
Pabo said:
Pabo, I thought janeeman's point is that how do you know that the magnitude of the 3rd order function is HIGHER in a symmetric design than in an asymmetric design?
Janneman conceded that 3rd order harmonics are more "visible" in a symmetric case. ie. they are higher vis-a-vis other distortions (2nd order or total). But that by itself doesn't mean that 3rd order harmonics is higher in a symmetric case than in an asymmetric case.
The output as RF input?
I have mentioned this before in other threads, not getting much
reaction, but since the issue of RFI is explicitly discussed here
I'll bring it up once more.
While you all seem to agree that RFI can cause problems and that
it is sensible to have an LP filter at the input, what about the
output? Rod Elliott has a theory about this, suggesting that
RFI may get in through the output. Since the output is usually
a high-impedance input for RF even a small induced current in
the cable might cause a considerable voltage that has a direct
path, through the feedback network, into the input stage,
Elliott suggests this might perhaps be a problem for many
amplifiers. (He also suggest this could be a reason why cables
sound different, ie. it would have to do with the RF properties
of the cables, the RFI spectrum at a certain location and the
amplifier outputs input impedance at RF, although, we should
probably avoid getting into a cable discussion in this thread).
Elliott suggests that it may be wise to have a Zobel filter also
at line level outputs to reduce this problem. He does
say clearly, however, that it is just a theory and that he does
not have any RFI problems where he lives and no RF test
equipment and thus has not been able to test the theory.
Any opinions on this?
I have mentioned this before in other threads, not getting much
reaction, but since the issue of RFI is explicitly discussed here
I'll bring it up once more.
While you all seem to agree that RFI can cause problems and that
it is sensible to have an LP filter at the input, what about the
output? Rod Elliott has a theory about this, suggesting that
RFI may get in through the output. Since the output is usually
a high-impedance input for RF even a small induced current in
the cable might cause a considerable voltage that has a direct
path, through the feedback network, into the input stage,
Elliott suggests this might perhaps be a problem for many
amplifiers. (He also suggest this could be a reason why cables
sound different, ie. it would have to do with the RF properties
of the cables, the RFI spectrum at a certain location and the
amplifier outputs input impedance at RF, although, we should
probably avoid getting into a cable discussion in this thread).
Elliott suggests that it may be wise to have a Zobel filter also
at line level outputs to reduce this problem. He does
say clearly, however, that it is just a theory and that he does
not have any RFI problems where he lives and no RF test
equipment and thus has not been able to test the theory.
Any opinions on this?
RF through output
I don't see any evidence for this at Rod's site, but it's plausible, especially in simple line stages where there's a compensation cap across the feedback resistor, and the feedback is taken directly from the output. If suspected in a problem case, it's an awfully easy thing to fix; depending on your religion, ferrite beads, stopper/isolation resistors, or an output inductor may prove useful individually or in combination.
I don't see any evidence for this at Rod's site, but it's plausible, especially in simple line stages where there's a compensation cap across the feedback resistor, and the feedback is taken directly from the output. If suspected in a problem case, it's an awfully easy thing to fix; depending on your religion, ferrite beads, stopper/isolation resistors, or an output inductor may prove useful individually or in combination.
RF effects
People have been putting zobel terminations on speaker cables and crossovers for some time now. Terminating line level circuits with the interconnect cables characteristic impedance is also gaining popularity and can be done for RF frequencies without having very low impedances at audio frequencies. Ben Duncan designed an external inductor for speaker cable several years ago. It is not just RF getting into amplifier circuits but the stability (and response) of the amps and preamps at these RF frequencies. I could go further and provide some references, but someone will just start another damn argument. I have talked about this before on the forum.
Output circuit from some preamp schematic I saw someware......
People have been putting zobel terminations on speaker cables and crossovers for some time now. Terminating line level circuits with the interconnect cables characteristic impedance is also gaining popularity and can be done for RF frequencies without having very low impedances at audio frequencies. Ben Duncan designed an external inductor for speaker cable several years ago. It is not just RF getting into amplifier circuits but the stability (and response) of the amps and preamps at these RF frequencies. I could go further and provide some references, but someone will just start another damn argument. I have talked about this before on the forum.
Output circuit from some preamp schematic I saw someware......
millwood
If you just look at an isolated gain stage with for example a JFET it will not generate any odd harmonics at all because of its square transfer characteristics. It will only generate second harmonic distorsion. If you combine a N- and a P-channel JFET in a push pull stage you will generate odd order harmonics because you create a third order transfer function (not a pure one though). The same will happen in a differential stage.
I'm not certain if a bipolar device or MOSFETs with their exponential transfer characteristics will generate odd harmonics in a single ended stage but I am certain that combining N- and P-channels in for example a push pull stage will increase the third order harmonic.
I will try and analyze this further when I find the time. Perhaps someone could simulate it in PSPICE?
If you just look at an isolated gain stage with for example a JFET it will not generate any odd harmonics at all because of its square transfer characteristics. It will only generate second harmonic distorsion. If you combine a N- and a P-channel JFET in a push pull stage you will generate odd order harmonics because you create a third order transfer function (not a pure one though). The same will happen in a differential stage.
I'm not certain if a bipolar device or MOSFETs with their exponential transfer characteristics will generate odd harmonics in a single ended stage but I am certain that combining N- and P-channels in for example a push pull stage will increase the third order harmonic.
I will try and analyze this further when I find the time. Perhaps someone could simulate it in PSPICE?
About balancing and odd-order distortion, I think there is no easy general answer.
Suppose you have a circuit with an instantaneous non-linear input to output transfer which can be expressed with the following Taylor series:
f(x)=a0+a1*x+a2*x^2+a3*x^3+a4*x^4+a5*x^5+...
If you make a balanced version by taking two of these circuits, giving them equal but opposite input signals, and subtracting the output signals, the result is:
f(x)-f(-x)=2*a1*x+2*a3*x^3+2*a5*x^5+...
In this case, all even-order terms are gone, but the odd-order terms have doubled. As this also applies to the linear term (2*a1*x versus a1*x), the odd-order distortion stays the same. I guess this is what happens in the fully complementary designs.
However, if you divide the input signal between the two circuits, the result is:
f(x/2)-f(-x/2)=a1*x+(1/4)*a3*x^3+(1/16)*a5*x^5+...
In this case, the odd-order distortions are also reduced, basically due to the reduced signal levels in each half of the balanced circuit.
But there are also cases where the signal is divided non-linearly between the devices, which may actually give rise to odd-order distortion that wasn't there before. For example, if you combine two perfectly quadratic devices into a differential pair, you end up with a transfer having odd-order terms. Replace the tail current source with a voltage source and the transfer becomes perfectly linear for fully balanced input signals. The reason for the difference lies in the rectified input signal occuring at the "source" node when a tail current source is used.
By the way, (MOS)FET's only have an exponential transfer when they are biased in weak inversion (below threshold). In strong inversion they are more or less quadratic, in between they are almost impossible to model properly.
Suppose you have a circuit with an instantaneous non-linear input to output transfer which can be expressed with the following Taylor series:
f(x)=a0+a1*x+a2*x^2+a3*x^3+a4*x^4+a5*x^5+...
If you make a balanced version by taking two of these circuits, giving them equal but opposite input signals, and subtracting the output signals, the result is:
f(x)-f(-x)=2*a1*x+2*a3*x^3+2*a5*x^5+...
In this case, all even-order terms are gone, but the odd-order terms have doubled. As this also applies to the linear term (2*a1*x versus a1*x), the odd-order distortion stays the same. I guess this is what happens in the fully complementary designs.
However, if you divide the input signal between the two circuits, the result is:
f(x/2)-f(-x/2)=a1*x+(1/4)*a3*x^3+(1/16)*a5*x^5+...
In this case, the odd-order distortions are also reduced, basically due to the reduced signal levels in each half of the balanced circuit.
But there are also cases where the signal is divided non-linearly between the devices, which may actually give rise to odd-order distortion that wasn't there before. For example, if you combine two perfectly quadratic devices into a differential pair, you end up with a transfer having odd-order terms. Replace the tail current source with a voltage source and the transfer becomes perfectly linear for fully balanced input signals. The reason for the difference lies in the rectified input signal occuring at the "source" node when a tail current source is used.
By the way, (MOS)FET's only have an exponential transfer when they are biased in weak inversion (below threshold). In strong inversion they are more or less quadratic, in between they are almost impossible to model properly.
Been away for a while - now to add some comments.
First, DJK, thank you for your cryptic comment:
Wrong. What happens here is that the 'slow' bootstrap cap loses effectiveness with rising impedance at very high frequencies (greater than 100KHz). This in turn increases collector loading on the voltage amplifier, which therefore returns less OLG. In turn this enables the amplifier to meet Bode-Nyquist stability criteria with less lag compensation, which has profound influence on the sonics.
Earlier comments about the fully symmetrical voltage amplifier (viz the Fender BXR and many hifi amps) relate more to the skewing of distortion spectrum than anything else. It would seem that a PNP and an NPN voltage amplifier working into a common output stage will to some extent fight each other; this mandates either use of matched devices or a moderate amount of emitter degeneration on both. This in turn limits OLG, and thus the only reason the topology might be used is to fully exploit the available drive nodes from a fully complementary input stage.
To respond to P-A, if belatedly! Symmetrical design is most pleasing to the eyes. The question remains, however - is it pleasing to the ears? Since symmetry nulls production of even order, leaving only odd order harmonics, then the harmonic spectrum of the musical signal is skewed away from the musical scale, which, for the human being, favours even order harmonics. While it is impossible to disagree that symmetry looks great in this context, and is much copied by mother nature herself, I posit that from an audiophile point of view, it is not desirable. Of course, YMMV, and obviously does!
You commented that my AKSA amplifier is slow. No, not quite true. The 3dB full power point into 8R is 45KHz on the 55W, and 42KHz on the 100W. The voltage amplifier in both cases is a 100MHz transistor, and I have found the speed of this crucial stage is very important.
Beyond a certain point, I'd say around 5V/uS, higher slew rates do not seem to improve the sound. However, you cannot go too fast on the VAS, and this bears further investigation. I can clearly hear the difference between a 30MHz and a 100MHz VAS, and perhaps it's not so hard to work out.
The bootstrap has the advantage of rapidly pulling down OLG beyond 100KHz. This helps with stability. OTOH, there are some phase shifts with a large, cumbersome capacitor between voltage amp and output, and this is a focus of present R&D.
Great debate; very fruitful.
Cheers,
Hugh
First, DJK, thank you for your cryptic comment:
Wrong. What happens here is that the 'slow' bootstrap cap loses effectiveness with rising impedance at very high frequencies (greater than 100KHz). This in turn increases collector loading on the voltage amplifier, which therefore returns less OLG. In turn this enables the amplifier to meet Bode-Nyquist stability criteria with less lag compensation, which has profound influence on the sonics.
Earlier comments about the fully symmetrical voltage amplifier (viz the Fender BXR and many hifi amps) relate more to the skewing of distortion spectrum than anything else. It would seem that a PNP and an NPN voltage amplifier working into a common output stage will to some extent fight each other; this mandates either use of matched devices or a moderate amount of emitter degeneration on both. This in turn limits OLG, and thus the only reason the topology might be used is to fully exploit the available drive nodes from a fully complementary input stage.
To respond to P-A, if belatedly! Symmetrical design is most pleasing to the eyes. The question remains, however - is it pleasing to the ears? Since symmetry nulls production of even order, leaving only odd order harmonics, then the harmonic spectrum of the musical signal is skewed away from the musical scale, which, for the human being, favours even order harmonics. While it is impossible to disagree that symmetry looks great in this context, and is much copied by mother nature herself, I posit that from an audiophile point of view, it is not desirable. Of course, YMMV, and obviously does!
You commented that my AKSA amplifier is slow. No, not quite true. The 3dB full power point into 8R is 45KHz on the 55W, and 42KHz on the 100W. The voltage amplifier in both cases is a 100MHz transistor, and I have found the speed of this crucial stage is very important.
Beyond a certain point, I'd say around 5V/uS, higher slew rates do not seem to improve the sound. However, you cannot go too fast on the VAS, and this bears further investigation. I can clearly hear the difference between a 30MHz and a 100MHz VAS, and perhaps it's not so hard to work out.
The bootstrap has the advantage of rapidly pulling down OLG beyond 100KHz. This helps with stability. OTOH, there are some phase shifts with a large, cumbersome capacitor between voltage amp and output, and this is a focus of present R&D.
Great debate; very fruitful.
Cheers,
Hugh
Good to see this thread is getting back on track to the original subject!
Hugh, it is great to have the opinion here of someone who has conducted such thourough investigations into the audible effects of different aspects of amplifier design. I must admit my subjective assesment of such things has been too random to really make definitive judgements about what I think is the best sounding way to setup an amp. My thoughts about the sound of different topologies have come from living with certain designs for extended periods, as this for me is the ultimate test of an amplifier's listenability.
On the slew rate debate: it is interesting to note that with an asymmetrical voltage amp stage, the overall amp slew rate will be asymmetrical for positive and negative signals. But assuming both are high enough this shouldn't matter - but what is high enough? (rhetorical question....please let's not go back there!)
The point made by David Tilbrook in the 6000 article about the use of symmetry was that it is inapropriate in the lower level stages where N and P devices will never have similar enough characteristics to achieve the distortion cancelling effect mentioned and indeed MarcelvdG's point above agrees with this argument. But Tilbrook believed that symmetrical VA stages performed better both objectively and subjectively, and this is the reason for his unusual 6000 topology. He believed that the dissimilarity between N and P devices in the VAS was outweighed by other advantages of symmetry in this stage. Certainly the 6000 has no lack of open-loop gain despite reasonably high degeneration in the symmetrical VAS.
I think it is difficult to argue whether even or odd harmonic is more desirable from an audiophile point of view - surely no distortion is the audiophile ideal. But from a musician's point of view, the harmonic content of an instrument is what gives it it's tone. To try to design amplifiers for a particular harmonic spread implies that the designer is aiming for a particular "tone" (ie a musical instrument) rather than the proverbial "straight piece of wire with gain". Hence valve amps, which are still so popular as the relatively high levels of distortion produced are low order and therefore musically pleasant, though not accurate.
When listening to asymmetrical amps compared to symmetrical designs I can't help but think that somehow either the frequency or phase response of the symmentrical designs is more linear, as all areas of the frequency spectrum sound equally balanced. in contrast asymmetrical designs seem to me to "project" mids forward so that vocalists tend to be more in-your-face (which can be a nice effect...). I realise that this is a highly questionable statement technically, but this is the subjective effect I have repeatedly noticed.
Indeed, a very interesting debate. Thanks guys.
Hugh, it is great to have the opinion here of someone who has conducted such thourough investigations into the audible effects of different aspects of amplifier design. I must admit my subjective assesment of such things has been too random to really make definitive judgements about what I think is the best sounding way to setup an amp. My thoughts about the sound of different topologies have come from living with certain designs for extended periods, as this for me is the ultimate test of an amplifier's listenability.
On the slew rate debate: it is interesting to note that with an asymmetrical voltage amp stage, the overall amp slew rate will be asymmetrical for positive and negative signals. But assuming both are high enough this shouldn't matter - but what is high enough? (rhetorical question....please let's not go back there!)
The point made by David Tilbrook in the 6000 article about the use of symmetry was that it is inapropriate in the lower level stages where N and P devices will never have similar enough characteristics to achieve the distortion cancelling effect mentioned and indeed MarcelvdG's point above agrees with this argument. But Tilbrook believed that symmetrical VA stages performed better both objectively and subjectively, and this is the reason for his unusual 6000 topology. He believed that the dissimilarity between N and P devices in the VAS was outweighed by other advantages of symmetry in this stage. Certainly the 6000 has no lack of open-loop gain despite reasonably high degeneration in the symmetrical VAS.
I think it is difficult to argue whether even or odd harmonic is more desirable from an audiophile point of view - surely no distortion is the audiophile ideal. But from a musician's point of view, the harmonic content of an instrument is what gives it it's tone. To try to design amplifiers for a particular harmonic spread implies that the designer is aiming for a particular "tone" (ie a musical instrument) rather than the proverbial "straight piece of wire with gain". Hence valve amps, which are still so popular as the relatively high levels of distortion produced are low order and therefore musically pleasant, though not accurate.
When listening to asymmetrical amps compared to symmetrical designs I can't help but think that somehow either the frequency or phase response of the symmentrical designs is more linear, as all areas of the frequency spectrum sound equally balanced. in contrast asymmetrical designs seem to me to "project" mids forward so that vocalists tend to be more in-your-face (which can be a nice effect...). I realise that this is a highly questionable statement technically, but this is the subjective effect I have repeatedly noticed.
Indeed, a very interesting debate. Thanks guys.
Amplifier topology objective effects
I was curious, so I thought I'd get some numbers on the 2nd/3rd harmonic issue of complementary vs. single-ended topologies. First, I simulated a Douglas Self-style single ended amplifier as shown in the first attachment below. Power supplies were +/- 56V and output voltage set to simulate an output power of 120 Watts into 8 Ohms at 20 kHz assuming an output stage were hooked up. I use a VCVS so the feedback network doesn't load down the VAS. Then I simulated a fully complementary design using the same transistors. Notice that the bias voltages and currents are the same for the complementary and single-ended designs. I tried to make component values equal so I'm doing an "apples to apples" comparison. I adjusted the compensation caps on the complementary design so the small-signal bandwidth was the same as the single-ended design. Nominal slew rates should be the same for the two.
Then, using a transient analysis and 65536-point FFT, I computed the harmonic content of the output. The details of this simulation technique can be found in the following post:
http://www.diyaudio.com/forums/showthread.php?postid=196461#post196461
And yes, I'm aware of the potential pitfalls of using SPICE to simulate harmonic distortion. I'll show the complementary design and the results in the next post.
I was curious, so I thought I'd get some numbers on the 2nd/3rd harmonic issue of complementary vs. single-ended topologies. First, I simulated a Douglas Self-style single ended amplifier as shown in the first attachment below. Power supplies were +/- 56V and output voltage set to simulate an output power of 120 Watts into 8 Ohms at 20 kHz assuming an output stage were hooked up. I use a VCVS so the feedback network doesn't load down the VAS. Then I simulated a fully complementary design using the same transistors. Notice that the bias voltages and currents are the same for the complementary and single-ended designs. I tried to make component values equal so I'm doing an "apples to apples" comparison. I adjusted the compensation caps on the complementary design so the small-signal bandwidth was the same as the single-ended design. Nominal slew rates should be the same for the two.
Then, using a transient analysis and 65536-point FFT, I computed the harmonic content of the output. The details of this simulation technique can be found in the following post:
http://www.diyaudio.com/forums/showthread.php?postid=196461#post196461
And yes, I'm aware of the potential pitfalls of using SPICE to simulate harmonic distortion. I'll show the complementary design and the results in the next post.
Attachments
Amplifier topology objective effects
The data for the single-ended amp was as follows:
Single-ended
-----------------
2nd harmonic: 61.3 dBc
3rd harmonic: 70.3 dBc
4th harmonic: 80.2 dBc
5th harmonic: 92.9 dBc
Complementary:
---------------------
2nd harmoinc: 69.6 dBc
3rd harmonic: 61.8 dBc
4th harmonic: 96.9 dBc
5th harmonic: 83.4 dBc
Total distortion is similar, but the assertions regarding improved 3rd harmonic distortion performance of the single-ended design seem to be correct.
The complementary design is shown below. I can provide the LTSpice projects if anybody is interested.
The data for the single-ended amp was as follows:
Single-ended
-----------------
2nd harmonic: 61.3 dBc
3rd harmonic: 70.3 dBc
4th harmonic: 80.2 dBc
5th harmonic: 92.9 dBc
Complementary:
---------------------
2nd harmoinc: 69.6 dBc
3rd harmonic: 61.8 dBc
4th harmonic: 96.9 dBc
5th harmonic: 83.4 dBc
Total distortion is similar, but the assertions regarding improved 3rd harmonic distortion performance of the single-ended design seem to be correct.
The complementary design is shown below. I can provide the LTSpice projects if anybody is interested.
Attachments
Andy,
Thank you for taking so much time and trouble to simulate this concept. I certainly appreciate it, as it gives me something to hang my hat on. All my comments in the past have been based on my own ideas, anecdotal at best, but essentially based on empirical experience without benefit of an expensive distortion analyzer. It certainly bears out my ideas, and I'm pleased about that.
Your post is very interesting. With appropriate attribution, could I please use something like this on my website? It is a rich argument, very rational, and crystal clear why SE is superior to the visually appealing fully complementary.....
Cheers,
Hugh
Thank you for taking so much time and trouble to simulate this concept. I certainly appreciate it, as it gives me something to hang my hat on. All my comments in the past have been based on my own ideas, anecdotal at best, but essentially based on empirical experience without benefit of an expensive distortion analyzer. It certainly bears out my ideas, and I'm pleased about that.
Your post is very interesting. With appropriate attribution, could I please use something like this on my website? It is a rich argument, very rational, and crystal clear why SE is superior to the visually appealing fully complementary.....
Cheers,
Hugh
owdeo said:
This "even order good, odd order bad" notion has never made much sense to me given that musical instruments produce huge amounts of odd order harmonics. In many cases, predominantly odd order. So if "even order good, odd order bad" actually holds true, most instruments would have to sound like absolute crap.
Actual research has shown that listiner preference tends toward low order distortion over high order distortion. And that's in keeping with single-ended triode tube amps not sounding as bad as measurements might suggest. Their typically using little or no global feedback results in low order feedback which is predominantly second and tends to drop like a rock from there.
se
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.