OK, lets get 'real' about the Quad and measurement with harmonic distortion vs IM distortion.
I have used and owned harmonic distortion measuring equipment since 1968, when I moved to the Ampex Audio Division. I had a big HP wave analyzer, and a GR oscillator on my bench. It was NECESSARY, because IM measurement is virtually useless with magnetic tape. (Please don't quibble with me about that one, as well, Bob). This combination was good enough for magnetic tape, loudspeakers, and even electronics, IF 0.03% residual is OK. I used this combination all my time at Ampex, but at home, I had a Heathkit IM tester, Heathkit oscillator, and later, I acquired my own wave analyzer. So up to 1974, I used both, BUT it was FAR easier to get a good measurement of an amp with the IM tester. The residual was almost 10 times lower. I had this combination until 1974.
Now, what was SOTA in those days? Besides HP, there was B&K, and we bought their best oscillator and harmonic multiplier for our lab in Switzerland, in 1974. This equipment, while being excellent for tape recorders for both distortion and noise spectra, was hard to use for a simple test for non-linearity, and I had a Crown IMA on my test bench for a few years. By comparison, the Crown IMA was easy, low residual (.001%), and accurate (for what it measured) and it was 'enough'. Mark Levinson had a THD meter, an HP334, and it worked OK, but it was more difficult to use than the 339, or the Crown IMA.
Now, where did the ST1700 come in? My first experience with one was with Bascom King (who I still work with, today), when he got one when it first came out, and we got him a job as technical reviewer for 'Audio' magazine. I would think that was in 1974-1975. It was a great piece of test equipment, BUT it cost the equivalent of 6 weeks pay for a younger engineer. Serious money, but not as expensive as the HP339. I could not even easily afford one, at the time. Mark Levinson did get one, so I had access to one, once in a while.
Now does this say 'popular' to you? (everyone).
I did not get my own ST1700B until I did some consulting for ST, and I still had to pay an extra $500 for the IM option, in 1976.
Now, what about the Quad power amp?
Back in 1975, Matti Otala (remember him?) wrote a paper comparing different distortion tests with different amplifiers. Quad was included in the survey. I recall Quad especially failing the 'noise insertion test' bigtime. I think that the IM was something like 0.015%, a very good figure for such a marginal circuit. This is what threw everyone off. And so it goes.
I have used and owned harmonic distortion measuring equipment since 1968, when I moved to the Ampex Audio Division. I had a big HP wave analyzer, and a GR oscillator on my bench. It was NECESSARY, because IM measurement is virtually useless with magnetic tape. (Please don't quibble with me about that one, as well, Bob). This combination was good enough for magnetic tape, loudspeakers, and even electronics, IF 0.03% residual is OK. I used this combination all my time at Ampex, but at home, I had a Heathkit IM tester, Heathkit oscillator, and later, I acquired my own wave analyzer. So up to 1974, I used both, BUT it was FAR easier to get a good measurement of an amp with the IM tester. The residual was almost 10 times lower. I had this combination until 1974.
Now, what was SOTA in those days? Besides HP, there was B&K, and we bought their best oscillator and harmonic multiplier for our lab in Switzerland, in 1974. This equipment, while being excellent for tape recorders for both distortion and noise spectra, was hard to use for a simple test for non-linearity, and I had a Crown IMA on my test bench for a few years. By comparison, the Crown IMA was easy, low residual (.001%), and accurate (for what it measured) and it was 'enough'. Mark Levinson had a THD meter, an HP334, and it worked OK, but it was more difficult to use than the 339, or the Crown IMA.
Now, where did the ST1700 come in? My first experience with one was with Bascom King (who I still work with, today), when he got one when it first came out, and we got him a job as technical reviewer for 'Audio' magazine. I would think that was in 1974-1975. It was a great piece of test equipment, BUT it cost the equivalent of 6 weeks pay for a younger engineer. Serious money, but not as expensive as the HP339. I could not even easily afford one, at the time. Mark Levinson did get one, so I had access to one, once in a while.
Now does this say 'popular' to you? (everyone).
I did not get my own ST1700B until I did some consulting for ST, and I still had to pay an extra $500 for the IM option, in 1976.
Now, what about the Quad power amp?
Back in 1975, Matti Otala (remember him?) wrote a paper comparing different distortion tests with different amplifiers. Quad was included in the survey. I recall Quad especially failing the 'noise insertion test' bigtime. I think that the IM was something like 0.015%, a very good figure for such a marginal circuit. This is what threw everyone off. And so it goes.
You need to ask Ed Simon the expert on such matters. IMNSHO these effects don't exist.
Not to neglect your question, Joao, but it is a very broad one.
What we try to use is what is BEST for the highest audio quality, and yet be practical. Jfets are higher in input impedance, low in noise over a wide range of source impedances, more linear, with less higher order distortion generation than bipolar transistors. In practical terms, they are closer in performance to tubes, that have shown the most success in making quality audio circuits.
What we try to use is what is BEST for the highest audio quality, and yet be practical. Jfets are higher in input impedance, low in noise over a wide range of source impedances, more linear, with less higher order distortion generation than bipolar transistors. In practical terms, they are closer in performance to tubes, that have shown the most success in making quality audio circuits.
I don't think they exist either but I'm curious: did you get a circuit to test ?
John, yes that's true, it is a broad question and I wanted to get many answers because this is an interesting field.
Anyway can you describe your sonical impressions?
Another point at Power (MOS)FETs is the high input capacitance.
Actually I think I could try something quick before I head out this weekend. How about 10 pieces of wirewrap wire in series just twisted together and different amounts of current run through it. I have about 130dB of gain available but only single tone generator for now.
There ain't no micro diodes. If you measure any component to an extreme enough level you will always find some sort of distortion.
The issue is at what level and what kind of distortions are important.
Now I have no idea what you will get with wire twisted for connections, but be sure to place it in a Faraday shield of a non magnetic material. Also the ground return path for the test current is an important issue.
Be sure to start with the lowest current you can test for. Also be sure no spike is introduced into the DUT during turn on. These will distort the measurements.
It sounds like you already did measurements of this type. Can you communicate what you found ?
Joao, you are asking a difficult question and the answer should become obvious from my design history over the last 45 years or so.
In the beginning of solid state, there were only single gender solid state devices for the most part. PNP Ge transistors, and NPN silicon transistors. Attempts to mix them did work, but not well.
By 1967, Motorola came out with a range of devices from small signal to complementary power transistors that allowed true complementary operation. This is when I started designing in earnest on a 'super' solid state power amp. Jfets were available for special applications, BUT they were very noisy and had low gain. Selected parts 'could' work fairly well, even then, and in 1968, while at Ampex, I made my first jfet (selected) input analog tape reproduce stage. It was a little noisy, but not much different, (except in noise sonic character) from what we normally used. I knew that it was only a matter of time when jfets would get much quieter, on average, and that it was only processing problems that kept it relatively high at the time. Still, there were only Nchannel jfets that were reasonably quiet, until the early 1970's, when Siliconix came out with a whole series of jfets, some complementary, and some that were VERY low noise, about 1nV/rt Hz at mid frequencies. This became my source of new designs, but before this, I made dozens of prototypes of amps from 15W-2000W, line drivers, etc, etc. for pro audio, all from complementary bipolar transistors, with some success. Still, the jfets held a lot of attraction to me, because by mathematical definition, they were more linear.
At first, I only changed my input stages to jfets, instead of bipolar. The circuits were faster, and tended to have lower order distortion. This was good. Then, for the Levinson JC-2 phono stage, I used complementary devices for the output followers, as well. This worked well, also. Please remember that I count on 'listener feedback' of my designs by other interested parties, to make a real determination of progress. I was then getting famous with what these circuits did in the consumer audio world with Mark Levinson's products, and in the pro world with my work for the Grateful Dead. If the 'public' did NOT like my designs, they would have made it obvious.
Still, during the 1970's, it was almost impossible to replace complementary output transistors, until the advent of Vfets, Vmos, and power mos, in the mid to late 70's. Then I could make my entire circuit with fets, and I made many examples like this. Still, I found that bipolar devices worked very well for many applications, including output devices, and I use them today, that way. That is about it.
In the beginning of solid state, there were only single gender solid state devices for the most part. PNP Ge transistors, and NPN silicon transistors. Attempts to mix them did work, but not well.
By 1967, Motorola came out with a range of devices from small signal to complementary power transistors that allowed true complementary operation. This is when I started designing in earnest on a 'super' solid state power amp. Jfets were available for special applications, BUT they were very noisy and had low gain. Selected parts 'could' work fairly well, even then, and in 1968, while at Ampex, I made my first jfet (selected) input analog tape reproduce stage. It was a little noisy, but not much different, (except in noise sonic character) from what we normally used. I knew that it was only a matter of time when jfets would get much quieter, on average, and that it was only processing problems that kept it relatively high at the time. Still, there were only Nchannel jfets that were reasonably quiet, until the early 1970's, when Siliconix came out with a whole series of jfets, some complementary, and some that were VERY low noise, about 1nV/rt Hz at mid frequencies. This became my source of new designs, but before this, I made dozens of prototypes of amps from 15W-2000W, line drivers, etc, etc. for pro audio, all from complementary bipolar transistors, with some success. Still, the jfets held a lot of attraction to me, because by mathematical definition, they were more linear.
At first, I only changed my input stages to jfets, instead of bipolar. The circuits were faster, and tended to have lower order distortion. This was good. Then, for the Levinson JC-2 phono stage, I used complementary devices for the output followers, as well. This worked well, also. Please remember that I count on 'listener feedback' of my designs by other interested parties, to make a real determination of progress. I was then getting famous with what these circuits did in the consumer audio world with Mark Levinson's products, and in the pro world with my work for the Grateful Dead. If the 'public' did NOT like my designs, they would have made it obvious.
Still, during the 1970's, it was almost impossible to replace complementary output transistors, until the advent of Vfets, Vmos, and power mos, in the mid to late 70's. Then I could make my entire circuit with fets, and I made many examples like this. Still, I found that bipolar devices worked very well for many applications, including output devices, and I use them today, that way. That is about it.
I would like to share some conclusions from one publication in russian, 2010 year, author is Chumakov.
He performed a rigorous mathematical analysis of transmission of information through a channel (an audio equipment). Analysis was focused on the transfer function. It was assumed, that TF can have some stochastic component (due to any reason, noise, parts distortions, PS effects, effects of GNFB, electro-magnetic disturbances, etc.).
The main conclusions are:
1) If the Transfer Function is strictly determined (has no stochastic non-deterministic component) there is no loss of information passing through the channel. Non-linearity of the TF does not cause loss of information. Human hearing can easily adapt to deterministic non-linearity, since hearing itself is non-linear.
2) Stochastic instability of the TF, due to, among other reasons, the "jitter" of voltage amplification factor, caused by limited velocity in GNFB loop, leads to losses of information, first of all, to losses in sibilants and transients - the most informative components of audio system.
He performed a rigorous mathematical analysis of transmission of information through a channel (an audio equipment). Analysis was focused on the transfer function. It was assumed, that TF can have some stochastic component (due to any reason, noise, parts distortions, PS effects, effects of GNFB, electro-magnetic disturbances, etc.).
The main conclusions are:
1) If the Transfer Function is strictly determined (has no stochastic non-deterministic component) there is no loss of information passing through the channel. Non-linearity of the TF does not cause loss of information. Human hearing can easily adapt to deterministic non-linearity, since hearing itself is non-linear.
2) Stochastic instability of the TF, due to, among other reasons, the "jitter" of voltage amplification factor, caused by limited velocity in GNFB loop, leads to losses of information, first of all, to losses in sibilants and transients - the most informative components of audio system.
Attachments
Last edited:
Interesting. Do you have a reference to this paper, preferably a web link ?
Also, is it available in English ?
I am not sure I understand your point: do you mean they are more linear because the quadratic transfer function is more linear than the exponential function ?
Of course, the bipolar transistors are normally "current driven" not "voltage driven", so the exponential transfer function
Ic=f(Vbe)
doesn't really matter in most of the audio circuits.
Or maybe you have a different meaning in mind ?
Here is a link to this publication.
http://www.filigrane.ru/Chumakov_Stochastic_Nonlinearity_EAT.pdf
Unfortunately, I can not find translation of this paper into English.
I could only convert it into doc format and try to use google translation.
I don't have it. Can you post here a condensed version of your findings ?
Thanks for the link, seems to be an interesting article. I am often saying that we examine deterministic periodic test signals, which is contrary to the nature of musical signals.
Hopefully I speak Russian
Here are some short translated parts of this paper:
Condition for the absence of information loss when converting the input X to output Y = G (X), including non-linear transfer function, is the presence of a single-valued inverse transform of H (V), reducing input X = H (G (X)).
Thus, in one transformation input to the output, a loss of information in a channel without noise does not occur, even if the transfer function is nonlinear.
Importance of this statement is not that any deterministic nonlinearity would not be noticeable by ear, but the fact that there is a fundamental opportunity without regard to the specific signal processing methods to obtain all the information at the output without a loss.
Consider what kind of distortions in electroacoustic channel
lead to loss of information and to the fundamental impossibility of receiving the initial information without losses.
If the signal in the channel is added by noise, the signal point is shifted from its true (expected) values. Since the direction and magnitude can not be predicted, the noise creates a small region of uncertainty around the "true" signal point. If, due to the effects of noise relation between transmitter and receiver is not single-valued - this is a mistake and part of the transmitted information is lost. Equation (3) just determines the number of potentially distinct areas of uncertainty for each dimension of the "web" receiver.
However, the unpredictable shift of signal point from its true position can be caused not only by noise, but also by other processes in the transmission channel. These random effects can be further reduced to an equivalent noise.
The fluctuations of the transfer function lead to the appearance in the output signal of
stochastic channel distortion, which can not be regarded as a change in metric space signals by random changes g (x). Stochastic distortion g (x) increases the uncertainty range around the "true" signal point and can be interpreted as an additional random noise and reduces the possibility of signals recognition.
It should be noted that the products of deterministic distortion do not enter into the specified formula, as expression (4) determines the potential maximum signal recognition capacity irrespective of the methods of signals processing. This implies an important difference in the influence of deterministic and stochastic distortions on information possibilities of electroacoustic path: Products of stochastic distortions can not be compensated by receiving information and lead to a reduction of information bandwidth. Thus, the presence in the transmission channel of stochastic components, regardless of their nature (additive noise, random modulation transfer function) decrease the bandwidth.
Information density of different genres of music signals and its individual components (instruments, parties, votes) are very different between themselves. In theory, the transfer of information is determined by the product of the signal band and its duration B = WT. Spectral analysis of sounds of different sources shows that a large information density have noise-like signals, such as sibilants, percussion (cymbals, brushes), and
and transients (attack transients). The smallest information density have tonal sounds with a small number of overtones.
- Status
- Not open for further replies.