question and suggestion
Andy,
how do you deterimine the dBcs? Do you use the cursor to measure the level of the fundamantal and the harmonics, or is there a more elegant way?
I am not sure about numerical problems in LTSpice, but how about doing away with the feedback loop, decreasing the input signal to a few nV and looking at the output? This would allow you to assess open loop linearity directly. You could also look at individual stages.
Regards,
Eric
Andy,
how do you deterimine the dBcs? Do you use the cursor to measure the level of the fundamantal and the harmonics, or is there a more elegant way?
I am not sure about numerical problems in LTSpice, but how about doing away with the feedback loop, decreasing the input signal to a few nV and looking at the output? This would allow you to assess open loop linearity directly. You could also look at individual stages.
Regards,
Eric
A new way to look at it
So, back to my original question: how about the subjective effects of asymmetrical vs symmetrical (generalised opinions are perfectly acceptable)?
I have not posted in this area before so...Hi
Iv'e been following this thread with interest - thanks to all - I feel like I have learnt a lot
I do not have a formal electronics training and certainly not a mathtamatical understanding of complex circuits but over the years having designed and built a few amps I have tried to build up an understanding of what is going on in cct's and what features may cause what effects.
many years ago I read an article by John Linsley hood where he stated that for reasons he could not understand he had found that s/e cct's sound more 'natural' that fully symmetrical ones, which he said somehow seemed to emphasise detail in a way that made them less appealing. Since then I have been trying to get my mind around why this could be.
I often enjoy trying to find principals in nature that apply across different disciplines
Now in advance I hearby appologise to the strictly objective science crowd that don't like subjective esoteria. We are now entering an area you may not appreciate
I can't help noticing the paralells between this subject and the ancient eastern principles of Yin & Yang.
among many other qualities, Yang is the active principle, Yin is the passive principle
generally for balance and harmony to exist is any living system the two principals have to be in balance.
In symetrical designs, particularly in the driver cct in is quite clear ( at least to me ) that this balance is missing - both elements are active. However in s/e designs, wether it be an inductor, a resistor, a current source or even a bootstrap this part of the cct is passive in as much as it simply responds to the what the signal is doing elsewhere
perhaps this balance created between the active & passive emements of s/e cct's somehow corresponds to this principal of balance in nature and consequently we find the sound to be SUBJECTIVELY pleasing.
wether this corresponds to any electrical values that we can measure I don't know but personally I like the analogy. It seems elegant.
I hope there my be perhaps one or two of you with whom this strikes a chord. I certainly do not intend to get into a huge discussion about wether this is right/wrong, relavent/irrelavent etc.
cheers
mike
So, back to my original question: how about the subjective effects of asymmetrical vs symmetrical (generalised opinions are perfectly acceptable)?
I have not posted in this area before so...Hi
Iv'e been following this thread with interest - thanks to all - I feel like I have learnt a lot
I do not have a formal electronics training and certainly not a mathtamatical understanding of complex circuits but over the years having designed and built a few amps I have tried to build up an understanding of what is going on in cct's and what features may cause what effects.
many years ago I read an article by John Linsley hood where he stated that for reasons he could not understand he had found that s/e cct's sound more 'natural' that fully symmetrical ones, which he said somehow seemed to emphasise detail in a way that made them less appealing. Since then I have been trying to get my mind around why this could be.
I often enjoy trying to find principals in nature that apply across different disciplines
Now in advance I hearby appologise to the strictly objective science crowd that don't like subjective esoteria. We are now entering an area you may not appreciate
I can't help noticing the paralells between this subject and the ancient eastern principles of Yin & Yang.
among many other qualities, Yang is the active principle, Yin is the passive principle
generally for balance and harmony to exist is any living system the two principals have to be in balance.
In symetrical designs, particularly in the driver cct in is quite clear ( at least to me ) that this balance is missing - both elements are active. However in s/e designs, wether it be an inductor, a resistor, a current source or even a bootstrap this part of the cct is passive in as much as it simply responds to the what the signal is doing elsewhere
perhaps this balance created between the active & passive emements of s/e cct's somehow corresponds to this principal of balance in nature and consequently we find the sound to be SUBJECTIVELY pleasing.
wether this corresponds to any electrical values that we can measure I don't know but personally I like the analogy. It seems elegant.
I hope there my be perhaps one or two of you with whom this strikes a chord. I certainly do not intend to get into a huge discussion about wether this is right/wrong, relavent/irrelavent etc.
cheers
mike
Hi Mike
It is not fully known yet what makes an amplifier sound pleasing or not.
Some points many (i.e. still not everyone) accept as valid are:
It is not only the amount (i.e. quantity) of distortion that counts, it is also the kind of distortion (i.e. quality).
The distortion spectrum should be falling with frequency as this is the case for most natural sounds.
Even order distortion is less disturbing than uneven order distortion since it is more musically related.
The distortion should be lowest at low volumes and not vice versa, since the overtone spectra of most musical instuments behave like this as well (the instrument where this is most obvious is a Saxophone, where the sond gets more aggressive the louder it is played).
I by myself like symmetrical designs. But one basic rule has to be taken into account: Circuits that generate symmetrical distortion mainly generate odd order harmonics while circuits that distort a signal asymmetrically mainly generate even order harmonics. Thats why some asymmetrical circuits possibly might have advantages over symmetrical ones.
Furthermore any nonlinearities generate IMD, i.e. "signals" that have not been there before and which are a dependant upon input signal AND the kind of nonlinearity involved. While the usually low distortion itself might barely change a single instrument's timbre, the IMD products can disturb clarity (i.e. "air" ?) as soon as multiple instruments are played. The higher order the nonlinearity the more complex the "signal-carpet" created by IMD.
We might not forget that it is impossible to build THE perfect amplifier, so properties have to be balanced to be most pleasing, whatever this means.
Regards
Charles
It is not fully known yet what makes an amplifier sound pleasing or not.
Some points many (i.e. still not everyone) accept as valid are:
It is not only the amount (i.e. quantity) of distortion that counts, it is also the kind of distortion (i.e. quality).
The distortion spectrum should be falling with frequency as this is the case for most natural sounds.
Even order distortion is less disturbing than uneven order distortion since it is more musically related.
The distortion should be lowest at low volumes and not vice versa, since the overtone spectra of most musical instuments behave like this as well (the instrument where this is most obvious is a Saxophone, where the sond gets more aggressive the louder it is played).
I by myself like symmetrical designs. But one basic rule has to be taken into account: Circuits that generate symmetrical distortion mainly generate odd order harmonics while circuits that distort a signal asymmetrically mainly generate even order harmonics. Thats why some asymmetrical circuits possibly might have advantages over symmetrical ones.
Furthermore any nonlinearities generate IMD, i.e. "signals" that have not been there before and which are a dependant upon input signal AND the kind of nonlinearity involved. While the usually low distortion itself might barely change a single instrument's timbre, the IMD products can disturb clarity (i.e. "air" ?) as soon as multiple instruments are played. The higher order the nonlinearity the more complex the "signal-carpet" created by IMD.
We might not forget that it is impossible to build THE perfect amplifier, so properties have to be balanced to be most pleasing, whatever this means.
Regards
Charles
Hi Mike,
I also like to draw parallels between fields. I'm not sure about the yin and yang thing, but the comparison I always see is between audio and photography. Same problem, different realms.
Some observations on the significance of different aspects of "distortion".
Example: scanned photos. Like in Audio, at this point the recording process is out of reach, we can only hope the "reproduction" of stored info to be rich in information, and pleasing.
Now for instance, picture defects such as out of focus on the edges, or slight hue problems, usually don't detract from a picture. But a single piece of dust in an important bright area is a major annoyance. This represents a tiny fraction of the information of the picture "lost" - but it is highly visible. I have no problems imagining a similar effect in Audio: that 0.5% of distortion A matters far less than 0.001% of distortion B.
Incidentally other parallels come to mind: say, that a highly compressed JPG scan of a photo often actually accentuates the defects of the lens. I have pics taken with high quality Zeiss lenses, and others with lesser lenses. Some lens effects (vignetting, very low gradients of shade and intensity), and differences between lenses, can best be seen on highly compressed pictures. Translated to Audio this explains how the distortionor loss of the most "insignificant" (the least significant bit) can be the most objectionable, and the easiest to hear. Other obvious distortions may be easy to measure but less important to perception.
I also like to draw parallels between fields. I'm not sure about the yin and yang thing, but the comparison I always see is between audio and photography. Same problem, different realms.
Some observations on the significance of different aspects of "distortion".
Example: scanned photos. Like in Audio, at this point the recording process is out of reach, we can only hope the "reproduction" of stored info to be rich in information, and pleasing.
Now for instance, picture defects such as out of focus on the edges, or slight hue problems, usually don't detract from a picture. But a single piece of dust in an important bright area is a major annoyance. This represents a tiny fraction of the information of the picture "lost" - but it is highly visible. I have no problems imagining a similar effect in Audio: that 0.5% of distortion A matters far less than 0.001% of distortion B.
Incidentally other parallels come to mind: say, that a highly compressed JPG scan of a photo often actually accentuates the defects of the lens. I have pics taken with high quality Zeiss lenses, and others with lesser lenses. Some lens effects (vignetting, very low gradients of shade and intensity), and differences between lenses, can best be seen on highly compressed pictures. Translated to Audio this explains how the distortionor loss of the most "insignificant" (the least significant bit) can be the most objectionable, and the easiest to hear. Other obvious distortions may be easy to measure but less important to perception.
andy_c said:
Of course you don’t understand, you haven’t read Cherry and Hawksford!
Cherry’s “ESTIMATES OF NONLINEAR DISTORTION IN FEEDBACK AMPLIFIERS” JAES V48#4 2000 p299-313 provides a method of calculating individual component distortion mechanisms contribution to overall distortion of a amplifier, more importantly it gives a intellectual framework for reasoning about distortion and feedback in a feedback amplifier. The example calculations for a simple audio amplifier circuit in the article also would give you some perspective that will make much of this wandering about in simulation land more directed and informative.
On the cascode VAS results, Hawksford’s REDUCTION OF TRANSISTOR SLOPE DISTORTION IN LARGE SIGNAL AMPLIFIERS, M.O.J. Hawksford, JAES, vol.36, no.4, pp.213-222, April 1988 explains that the major distortion in the VAS isn’t a “simple” exponential series from gm/Vbe but the more difficult to model nonlinear CB and CE conductances. Hawksford goes on to give one of the few explanations of the “correct” version of the cascode connection I have seen (of course Pass gets this right too) and Hawksford shows the Baxendall “Super Pair” (without attribution, but no one else I’ve read seems aware of its origin either)
Both are well worth buying from the AES ($5 ea for pdf) if you can’t find them in the library (think how much of your time you’ve spent simulating; I guarantee the learning rate reading these papers is much higher)
When you linearize your VAS with the Super Pair as I've shown in the Class A headphone amplifier thread, ( < -140 dB 2nd harmonic @ 2KHz) then you might be able to see input Vbe distortions that follow the even harmonic cancellation you've spent so much time looking for
Re: Question to ANDY_C
Jan,
The capacitance seen by the collector of the input stage should be, by Miller's theorem, Ccomp * (1 - Av), where Ccomp is the compensation cap and Av is the voltage gain of the VAS. We knowl that Av is large and negative, but we don't know exactly what its value is without simulating it - because of the very high collector load impedance of the VAS. So the capacitance seen by the input stage should be quite high, much larger than Ccomp.
I'm at work now so I can only post short replies at long intervals.
janneman said:
Jan,
The capacitance seen by the collector of the input stage should be, by Miller's theorem, Ccomp * (1 - Av), where Ccomp is the compensation cap and Av is the voltage gain of the VAS. We knowl that Av is large and negative, but we don't know exactly what its value is without simulating it - because of the very high collector load impedance of the VAS. So the capacitance seen by the input stage should be quite high, much larger than Ccomp.
I'm at work now so I can only post short replies at long intervals.
jcx said:
Thanks for pointing out those articles. What is the "correct" version of the cascode circuit? I thought at least that was pretty well defined- it means using a driver and a cascode transitor of the same polarity, connecting collector and emitter together and connecting the base of the cascode transistor to a voltage source that referenced to the rail and very stiff.
Is correct cascode the same thing as super pair?
Regards,
Eric
millwood said:
It's almost by definition. If you take an assymetrical waveform, and take it apart in harmonic components, you'll find there are even order harmonics. (Sawtooth for instance) If you take a symmetrical waveform you'll find odd orfer harmonics (square wave for instance). So, any distortion that makes the sine wave assymetric generates even order thd (like in an assymmetrical or se amp), while anything that makes the wave distort symmetrically (push-pull f.i) generates odd order harmonics.
Jan Didden
janneman said:
two questions, Jan.
1) is there a difference between symmetric waveforms and symmetrci circuitry?
2) wouldn't you find both odd and even harmonics in symmetrical and asymmetrical waveforms? For example, square waves will give you both odd and even harmonics in FFT, if my memory still works.
No Logic, just subjective
Some years ago I spent a lot of time modifying and listening to the Borbely Servo100 design as published in Audio Amateur.
Here are my completely subjective results (the only thing I measured was transient response into a capacitive load to ensure stablilty):
In general I found that removing unneccessary transistors and simplifying the design improved the sound. These included removal of the driver Fets and the emitter followers between LTPs and VASs. Removal of the DC servo helped improve low level detail a lot. After doing all this I ended up with a generic dual complimentary circuit which was probably what Borbely had before he started embellishing it to reduce distortion.
I then added cascodes to the LTPs and VASs. The cascodes on the LTPs made the sound worse and I heard very little difference with/without them on the VASs. I then tried adding current sources to stabilise any current that looked like it might benefit - if you like ss harshness this is the way to go!
OK so I know all this changes the loop response of the amplifier etc etc but irrespective of the maths - simpler generally sounds better (as long as it does the job properly).
My final experiment was to remove the NPN LTP and turn the top end of the VAS into a current source (I.e. convert from dual complementary to Lin topology). I could not detect any difference in sound quality. The only reason I can see to use dual complementary front ends is that dc offset is more stable as the amp warms up.
I also tried substituting FETs in place of biplolars and they generally seemed to sound cleaner and more detailed, of course this does change the loop gain significantly so it's very difficult to attribute the change in sound to the device alone.
I now run an AKSA 100N which I built because I agreed with Hugh's philosophy and I think he has got most things right with this design. OTOH Nelson Pass's Aleph 2 looks to have all the right ingredients but I have not built one (yet! - I'm trying to resist the temptation!) 'cos I can't hack wasting 600W.
Dave
Some years ago I spent a lot of time modifying and listening to the Borbely Servo100 design as published in Audio Amateur.
Here are my completely subjective results (the only thing I measured was transient response into a capacitive load to ensure stablilty):
In general I found that removing unneccessary transistors and simplifying the design improved the sound. These included removal of the driver Fets and the emitter followers between LTPs and VASs. Removal of the DC servo helped improve low level detail a lot. After doing all this I ended up with a generic dual complimentary circuit which was probably what Borbely had before he started embellishing it to reduce distortion.
I then added cascodes to the LTPs and VASs. The cascodes on the LTPs made the sound worse and I heard very little difference with/without them on the VASs. I then tried adding current sources to stabilise any current that looked like it might benefit - if you like ss harshness this is the way to go!
OK so I know all this changes the loop response of the amplifier etc etc but irrespective of the maths - simpler generally sounds better (as long as it does the job properly).
My final experiment was to remove the NPN LTP and turn the top end of the VAS into a current source (I.e. convert from dual complementary to Lin topology). I could not detect any difference in sound quality. The only reason I can see to use dual complementary front ends is that dc offset is more stable as the amp warms up.
I also tried substituting FETs in place of biplolars and they generally seemed to sound cleaner and more detailed, of course this does change the loop gain significantly so it's very difficult to attribute the change in sound to the device alone.
I now run an AKSA 100N which I built because I agreed with Hugh's philosophy and I think he has got most things right with this design. OTOH Nelson Pass's Aleph 2 looks to have all the right ingredients but I have not built one (yet! - I'm trying to resist the temptation!) 'cos I can't hack wasting 600W.
Dave
Re: Asymetry Plus Asymetry = ?
but humans are symmetrical, our ears are symmetrical (for the most part anyway,
), electrons as far as we know now are symmetrical, moleculars are symmetrical, the sun is symmetrical (for light years away), as it the earth.
so everything else should be symmetrical,
epic psudo science,
mrfeedback said:
but humans are symmetrical, our ears are symmetrical (for the most part anyway,
), electrons as far as we know now are symmetrical, moleculars are symmetrical, the sun is symmetrical (for light years away), as it the earth. so everything else should be symmetrical,
epic psudo science,
capslock said:
For the right way to wire a cascode see:
http://www.passlabs.com/pdf/cascode.pdf fig 4 & 7b
Its easy to see that the upper cascode transistor should be referenced to the base (or, more practically the emitter) of the lower transistor to hold the Vcb of the lower device as constant as possible
What is less obvious is that the voltage reference really, really wants to be AC shorted to the lower transistor emitter to intercept the current flowing in the upper cascode transistor’s cb conductance (mostly a nonlinear capacitance) and “add” it to the regulated emitter current of the lower device, with high enough emitter load (ie degeneration resistors or ccs bias) this connection cancels the upper device cb current, eliminating the major high frequency distortion mechanism - just move those (Spice) voltage sources to the "lower" device emitters
The Baxendall Super Pair accomplishes the same Ccb cancellation and provides Darlington like input impedance with the same # of Qs http://www.diyaudio.com/forums/showthread.php?s=&threadid=17446 (my circuit bottom of 1st page; “Super” pairs Q5/7, Q6/8)
jcx said:
You've taken two unrelated quotes of mine out of context. The complete quote was:
andy_c [/i][B] So not only is the third harmonic about 15 dB better for the single-ended design than the complementary design said:
shows that you probably missed Pabo's comments that started this part of the discussion. I did these simulations after Pabo made the claim that, although the second-order distortion of a complementary power amp is generally lower than that of a single-ended amp, the third order distortion is higher and this could be verified by simulation. That was the reason - in order to verify that claim and get some data on the distribution of harmonics of the two configurations. This wasn't some sort of attempt to find the ultimate low distortion amplifier. The original simulation did bear out his claim. The second simulation was an attempt to see if the previous conclusion still held in the presence of cascoding. This led to the surprising (to me) result that the single-ended design ended up beating out the complementary design for second-order distortion as well.
Thanks for the article references. I will definitely read them if I can get them. Can non-AES members purchase these as well?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.