john curl said:
Didn't they say sometime last year that they expected to have them early this year?
And if you don't mind my asking, what's the big advantage to using complimentary push-pull pairs as opposed to unipolar push-pull pairs? Wouldn't the latter allow for better matching, less distortion and not having to worry about complimentary devices which would also expand the choice of devices you could use?
se
No no. Complementary differential symmetry is really the best. The only way how to get same SR and perfectly symmetrical limitation. CCS do not make it. And even distortion cancellation, depending on part similarity, of course.
PMA said:
I am not so sure....
🙄
The 2SK389 and 2SJ109 are also not exactly each others miorror image. F. a. the 2SJ109 has much higher capacitances.
You people worry too much about superficial things. First of all, the advantage of complementary differential is that it creates two separate signals that are opposed in DC voltage from each other. This gives 2 times the gain, without any phase delay increase. It is the second stage that is truly improved in linearity. Also, the complementary differential fet input, with a floating resistor for self biasing, can drive more peak current under high rate of change than a simple differential stage. This can give higher slew rate for a given set of initial conditions.
I have never done a SPICE simulation of it, but I would predict that there is an additional advantage of having the complementary differential pair work in series with each other, as well as in parallel. There actually may be an optimum bias resistor value for a given set of devices. I have never modeled this, but I will, sooner or later.
I have never done a SPICE simulation of it, but I would predict that there is an additional advantage of having the complementary differential pair work in series with each other, as well as in parallel. There actually may be an optimum bias resistor value for a given set of devices. I have never modeled this, but I will, sooner or later.
yes, the n jfets and p jfets don't match perfectly.
are you worried about the n and p channel diff amps having different characteristics?
don't forget you are usually:
1. adding source resistors
2. cascoding the jfets
these two practices help mitigate the problems mentioned.
i'd suspect there is still lots of lower hanging fruit to be addressed elsewhere in the design before worrying about this one in excruciating detail ...
... yet
mlloyd1
are you worried about the n and p channel diff amps having different characteristics?
don't forget you are usually:
1. adding source resistors
2. cascoding the jfets
these two practices help mitigate the problems mentioned.
i'd suspect there is still lots of lower hanging fruit to be addressed elsewhere in the design before worrying about this one in excruciating detail ...
... yet
mlloyd1
john curl said:
Wasn't worrying about anything. Just curious as to the advantages of a complimentary pair.
Yeah, assuming a single-ended source. But why limit yourself to a single-ended source?
Why would there be any phase delay increase?
What if you were to use a transformer for your differential input with it driving a symmetrical pair of common-source fets?
se
One small factor that many of you might overlook is that the NONLINEAR input capacitance subtracts from its complement when it comes to nonlinearity. It is true that p channel fets still have more capacitance for a given Gm, but that doesn't matter much with drive impedances of 50Kohm or less. However, IF you are going to drive with impedances of 50Kohm or more, you should use even lower capacitance devices. They exist, but they also have lower Gm
Steve Eddy said:
Hi Steve,
You are asking some good questions, even if they may be a bit off-topic for this thread. The full-complementary differential pairs are a nice, convenient, symmetrical way to get a proper drive signal to a full-complementary VAS. I especially like they way John is able to float the tails of the complementary pairs.
However, the full complementary differential pair architecture is not the only way to skin that cat. Take a look at the front-end of my MOSFET Power Amplifier with Error Correction in the paper on my website at www.cordellaudio.com under the published papers tab. It only requires a single same-sex matched differential input JFET pair. The arrangement does not look as elegant and symmetrical on a schematic, and tends to require more transistors, however.
Cheers,
Bob
Bob, you make a good design, but it's topology goes back to the '60's. I first saw it in the Marantz 14 power amp, designed by Sid Smith.
In my experience, this basic topology has twice the distortion as a complementary symmetry design without cascodes in either design. I think that is has to do with thermal offset in the current mirror that made my designs at least, slightly compromised. Anyone can cascode and I have do so as well in many designs of this type.
If you could look at the line driver that I designed for Sound Technology in the early '80's, you would find cascodes everywhere with your selected input devices in front. The greatest difference between your design and mine was the output stage, although mine was complementary power mosfets as well, and the fact that I maintained a 100KHz open loop bandwidth.
In my experience, this basic topology has twice the distortion as a complementary symmetry design without cascodes in either design. I think that is has to do with thermal offset in the current mirror that made my designs at least, slightly compromised. Anyone can cascode and I have do so as well in many designs of this type.
If you could look at the line driver that I designed for Sound Technology in the early '80's, you would find cascodes everywhere with your selected input devices in front. The greatest difference between your design and mine was the output stage, although mine was complementary power mosfets as well, and the fact that I maintained a 100KHz open loop bandwidth.
john curl said:
Hi John,
It is not just the cascodes. You need to look more carefully at the differential current mirror load. That part of the design is very important, and, to my knowledge, does not go back to the Sixties. I'm sure your experience is correct, but I also expect that your experience does not include this circuit.
This circuit does a near-perfect job of forcing the input differential pair into current balance, something your circuit fails to do. That significantly reduces distortion. This circuit also does a near perfect job of driving the top and bottom complementary VAS transistors with identical signals, something that any gm mismatch in your circuit between the p and n channel devices will fail to do. That will also result in distortion. This circuit also provides more gain in the input stage ahead of the VAS, greatly reducing the influence of any VAS noise.
Cheers,
Bob
You are referring to the Widler 101 based current source with an added cascode making it a 4 transistor combination? Yes, I agree that that is a good way to go. I used a 4 transistor cascoded current mirror instead that also kept the temp and voltage differences to a minimum, when I designed with the general topology that you prefer. Your circuit might have less 2'nd harmonic than mine. I'll give you that.
So, it's appearing that there's no hugely compelling reason to choose one over the other, yes?
se
se
I like series simple. Parallel multiple is OK. I will also put up with 4 quadrant symmetry. It is the number of series junctions that the audio has to pass through from input to output that is significant, in my opinion.
Complementary differential is one of the most stable ways to achieve this.
Complementary differential is one of the most stable ways to achieve this.
john curl said:
Hi John,
I assume that this is based on your listening tests as opposed to any measurements, right?
Cheers,
Bob
john curl said:
Hi John,
I take it that is a yes.
So cascodes are bad because they force the signal to pass though another junction.
So emitter followers are bad because they force the signal to pass through another junction.
What do you think is the nature of the degradation that a signal suffers when it passes through a junction?
Do you have a pseudo-science reason for your assertion, or are you just saying it sounds better and you don't know why?
The latter is what I call the X Factor, and accounts for things we just don't yet understand in things that cause subjective vs objective differences, and I don't intend to imply that the X Factor is not legitimate. For example, people hear differences in cables when laboratory measurements of those cables do not reveal differences that can reasonably be correlated to the sound difference.
Cheers,
Bob
It has been my design philosophy to use a minimum of stages and devices for the last 35 years. Cascodes and followers count 1/2 in total stage count.
The biggest thing that I have in my favor, is that my circuits have been successful over the decades.
I think it is because I minimize open loop higher order distortion and attempt to maximize open loop bandwidth.
Absolute THD is not very important to me, but it is usually below what most references would consider audible.
If I did not consider this effort useful, I would just use IC's.
The biggest thing that I have in my favor, is that my circuits have been successful over the decades.
I think it is because I minimize open loop higher order distortion and attempt to maximize open loop bandwidth.
Absolute THD is not very important to me, but it is usually below what most references would consider audible.
If I did not consider this effort useful, I would just use IC's.
Subjective versus objective differences doesn't have to be called X. They already have a name: imagination, illusion, etc...
Eva said:
Simple explanations make life easier. Imagination and illusion of the listeners who were not informed about technical differences of compared systems? And they independently, in different time and session describe the same perceived distinctions?
Hi QSerraTico_Tico
---I am not so sure....---
Me too. Not at all.
At least conceptually, complementary (differential) symmetry is puzzling.
The output voltage (more precisely, the voltage at the node of the feedback network defining the voltage gain) is under the control of two loops, with two input stages of opposite polarity. I am not aware of a detailed analysis of such circuits. Having two servo-mechanisms controlling one function makes me feel uneasy. I also remember the objection from Douglas Self who seems to have never been tempted by explorations in this territory.
The main objective of complementary (differential) symmetry in low power stages seems to rely in getting a push-pull VAS for high slew-rate. However there are some examples of this being achieved with a single differential input stage :
- Hitachi low power stage for its first Mosfet amps : a current mirror loads the VAS (I see Bob Cordell's amp as a sophistication of this basic circuit).
Being a cascade of two differential stages and adding suitable current sources for the coupled emitters, these circuits should have higher PSRR than that of the complementary symmetry. Note that differential stages are push-pull circuits of same polarity devices and suffer less of characteristics dispersion than complementary pairs, and so, should be more linear.
- Sansui "diamond" circuit : the VAS is fed by a symmetrical differential intermediate stage which bears some resemblance with John Curl's Fet input stage and its floating tail. But it is bipolar and is biased by diodes networks included in the loads of the single differential input stage.
- Circuit used in some Harman-Kardon amps like the PM640 : the VAS is fed by two complementary emitter followers connected to the opposite sides of a high value resistor. One of the emitter follower is being directly connected to one of the branches of the differential input stage, and the other one is preceded by a phase reversing common emitter connected to the other branch of the differential input stage.
- Hafler 1500 circuit : same basic idea as the Harman-Kardon. One of the branches of the differential NPN input stage directly feeds the PNP transistor of the VAS stage, the other one being phase reversed by a current mirror to feed the NPN transistor of the VAS stage.
- Hafler 4000 "diamond" circuit (electronicians love diamonds…) : the differential input stage has a basic current mirror load, its output current being connected to the resistors biasing the VAS through two intermediate common base transistors.
- I would like to mention the Bissmire's circuit that I've not seen used in audio. It is not a really push-pull circuit but consists of a cascode configuration with two common base transistors in parallel, but passing different currents, one of them modulating the constant (with no signal) current source load of the other one.
At last, I wonder if a class push-pull follower loading a conventional CCS loaded VAS and feeding the output stage is not able to do exactly the same job as full symmetry without the need to look for closely matched complementary pairs.
---I am not so sure....---
Me too. Not at all.
At least conceptually, complementary (differential) symmetry is puzzling.
The output voltage (more precisely, the voltage at the node of the feedback network defining the voltage gain) is under the control of two loops, with two input stages of opposite polarity. I am not aware of a detailed analysis of such circuits. Having two servo-mechanisms controlling one function makes me feel uneasy. I also remember the objection from Douglas Self who seems to have never been tempted by explorations in this territory.
The main objective of complementary (differential) symmetry in low power stages seems to rely in getting a push-pull VAS for high slew-rate. However there are some examples of this being achieved with a single differential input stage :
- Hitachi low power stage for its first Mosfet amps : a current mirror loads the VAS (I see Bob Cordell's amp as a sophistication of this basic circuit).
Being a cascade of two differential stages and adding suitable current sources for the coupled emitters, these circuits should have higher PSRR than that of the complementary symmetry. Note that differential stages are push-pull circuits of same polarity devices and suffer less of characteristics dispersion than complementary pairs, and so, should be more linear.
- Sansui "diamond" circuit : the VAS is fed by a symmetrical differential intermediate stage which bears some resemblance with John Curl's Fet input stage and its floating tail. But it is bipolar and is biased by diodes networks included in the loads of the single differential input stage.
- Circuit used in some Harman-Kardon amps like the PM640 : the VAS is fed by two complementary emitter followers connected to the opposite sides of a high value resistor. One of the emitter follower is being directly connected to one of the branches of the differential input stage, and the other one is preceded by a phase reversing common emitter connected to the other branch of the differential input stage.
- Hafler 1500 circuit : same basic idea as the Harman-Kardon. One of the branches of the differential NPN input stage directly feeds the PNP transistor of the VAS stage, the other one being phase reversed by a current mirror to feed the NPN transistor of the VAS stage.
- Hafler 4000 "diamond" circuit (electronicians love diamonds…) : the differential input stage has a basic current mirror load, its output current being connected to the resistors biasing the VAS through two intermediate common base transistors.
- I would like to mention the Bissmire's circuit that I've not seen used in audio. It is not a really push-pull circuit but consists of a cascode configuration with two common base transistors in parallel, but passing different currents, one of them modulating the constant (with no signal) current source load of the other one.
At last, I wonder if a class push-pull follower loading a conventional CCS loaded VAS and feeding the output stage is not able to do exactly the same job as full symmetry without the need to look for closely matched complementary pairs.
- Status
- Not open for further replies.