Hi Jan,
you are very close to the answer.
They are amplifiers with 2 different output stages. The signal is connected to the outputs of the amplifiers. It is 10kHz sine, 2A peak current. The plot is output voltage vs. excitation current. So it shows output impedance trajectory of 10kHz sine. The signal starts suddenly (
) from V=0, I=0 and lasts for 200us.
1st image is an amp with NFB and classB output stage (BJT), no error correction. Iq is too low (2mA), the result looks better for say 45mA, but still not good.
2nd image is an amp with NFB, MOSFETs and error correction.
you are very close to the answer.
They are amplifiers with 2 different output stages. The signal is connected to the outputs of the amplifiers. It is 10kHz sine, 2A peak current. The plot is output voltage vs. excitation current. So it shows output impedance trajectory of 10kHz sine. The signal starts suddenly (
) from V=0, I=0 and lasts for 200us.1st image is an amp with NFB and classB output stage (BJT), no error correction. Iq is too low (2mA), the result looks better for say 45mA, but still not good.
2nd image is an amp with NFB, MOSFETs and error correction.
First plot looks like error signal for a normal NFB topology showing both proportional magnitude, phase shift (must be at some high frequency) and crossover gain drop.
Second plot look the same but for an EC topology (there is no proportionality hear because of "infinite" gain), though I am puzzled there is not crossover overdrive, unless some form of compensation has also being thrown in.
Rodolfo
Second plot look the same but for an EC topology (there is no proportionality hear because of "infinite" gain), though I am puzzled there is not crossover overdrive, unless some form of compensation has also being thrown in.
Rodolfo
Pavel, you could have waited a little longer!!
Self's XD output stage
I have reviewed the Cambridge White Paper on Self's XD output stage.
The basic idea is to pull a current from the output node of an optimally-biased Class-B stage of about 1A to one rail, such as the negative rail. This forces the crossover point to be displaced away from zero current into the load by about 1 amp. This keeps one essentially in Class A about the current-to-the-load zero crossing. His elaboration on the scheme is to make the current source voltage controlled such that its current reduces when the output voltage approaches the positive rail and increases when the output voltage approaches the negative rail. The main reason for this is to save some power dissipation at positive swing. This effective negative impedance current source placed on the output also has some effect of lightening the load, so there is some additional benefit.
The arrangement requires a third output power transistor as the controlled current source and some small-signal circuitry to drive it. So he ends up with 1-1/2 the usual output stage.
As compared to a Class AAB over-biased stage (which also keeps the crossover away from zero), he realizes maybe a net improvement in distortion of a factor of two at levels where the Class AAB has gone to Class B.
This scheme is interesting, and probably novel, but seems far inferior to just adding error correction to a bipolar Class AAB stage to kill the gm doubling effect. The cost of the EC is just four small-signal transistors, and even modest EC will deliver a distortion reduction of more than a factor of two.
I'm not impressed.
Cheers,
Bob
I have reviewed the Cambridge White Paper on Self's XD output stage.
The basic idea is to pull a current from the output node of an optimally-biased Class-B stage of about 1A to one rail, such as the negative rail. This forces the crossover point to be displaced away from zero current into the load by about 1 amp. This keeps one essentially in Class A about the current-to-the-load zero crossing. His elaboration on the scheme is to make the current source voltage controlled such that its current reduces when the output voltage approaches the positive rail and increases when the output voltage approaches the negative rail. The main reason for this is to save some power dissipation at positive swing. This effective negative impedance current source placed on the output also has some effect of lightening the load, so there is some additional benefit.
The arrangement requires a third output power transistor as the controlled current source and some small-signal circuitry to drive it. So he ends up with 1-1/2 the usual output stage.
As compared to a Class AAB over-biased stage (which also keeps the crossover away from zero), he realizes maybe a net improvement in distortion of a factor of two at levels where the Class AAB has gone to Class B.
This scheme is interesting, and probably novel, but seems far inferior to just adding error correction to a bipolar Class AAB stage to kill the gm doubling effect. The cost of the EC is just four small-signal transistors, and even modest EC will deliver a distortion reduction of more than a factor of two.
I'm not impressed.
Cheers,
Bob
Bob,
Somebody asked earlier "why would D self patent XD if not ...". I was tempted at that time to answer "commercial interests"? but I didn't.
Not that there is anything wrong with that. After a century or so of audio electronic development it gets almost impossible to come up with something truly novel.
In XD I see also elements of Nelson Pass designs where the SE output stage has a current source load that is also modulated with the output signal.
BTW I think there is something wrong with Fig 14 of the whitepaper. XD-PP is not better at all points as the txt suggests, unless I read the graphs wrong (which is possible with b/w superimposed curves).
Jan Didden
Somebody asked earlier "why would D self patent XD if not ...". I was tempted at that time to answer "commercial interests"? but I didn't.
Not that there is anything wrong with that. After a century or so of audio electronic development it gets almost impossible to come up with something truly novel.
In XD I see also elements of Nelson Pass designs where the SE output stage has a current source load that is also modulated with the output signal.
BTW I think there is something wrong with Fig 14 of the whitepaper. XD-PP is not better at all points as the txt suggests, unless I read the graphs wrong (which is possible with b/w superimposed curves).
Jan Didden
Class-XD
I’ve read the article in EW November 2006 by D. Self.
The THD figures at 10 kHz show an improvement of two to three times with respect to class-B. For such ‘giant leap forward’ Doug Self spent an additional 17 component, not to mention the extra power consumption. IMHO this so called class-XD arrangement is really a kludge. Besides there exist must simpler, better and cheaper ways to lower the THD, such as two pole compensation (not a favorite of mine), including the O/P stage in the Miller feedback loop and of course Cherry's NDFL.
I’ve read the article in EW November 2006 by D. Self.
The THD figures at 10 kHz show an improvement of two to three times with respect to class-B. For such ‘giant leap forward’ Doug Self spent an additional 17 component, not to mention the extra power consumption. IMHO this so called class-XD arrangement is really a kludge. Besides there exist must simpler, better and cheaper ways to lower the THD, such as two pole compensation (not a favorite of mine), including the O/P stage in the Miller feedback loop and of course Cherry's NDFL.
janneman said:
The original Aleph 0 had a high value constant current source
in parallel with a push pull output stage and ran Class A to
75 watts, which was the rating of the amp into 8 ohms.
A later version modulated the current source based on output
current. This was found in version 1.4 of the Aleph 0 and also
the Aleph 0s and the first version of the Aleph 1.
All later Alephs used a much different circuit.
The schematic for the Aleph 0 1.4 is a pdf file found here
www.passlabs.com/np
Hi PMA,
I find it disconcerting that you argue against me and then post your own useful reverse driven amplifier investigations with a suddenly starting 10kHz sine.
The first graph (class-AB) shows the applied current producing an initial error due to propagation delayed (C.dom or choke?) NFB control failing to maintain control, and then cycling with phase shift and showing crossover distortion.
The second appears to have a good low output impedance damping of an initial sudden start (or is this bandwidth limited?)and a good damping phase response, though maybe not linear. The error amplitude appears to be approx 7mV (~ minus 66dB) for an equivalent of 16V at 8 ohms, though of course the amplifier output stage has it easy when it is merely attempting to maintain zero output volts.
Cheers ......... Graham.
I find it disconcerting that you argue against me and then post your own useful reverse driven amplifier investigations with a suddenly starting 10kHz sine.
The first graph (class-AB) shows the applied current producing an initial error due to propagation delayed (C.dom or choke?) NFB control failing to maintain control, and then cycling with phase shift and showing crossover distortion.
The second appears to have a good low output impedance damping of an initial sudden start (or is this bandwidth limited?)and a good damping phase response, though maybe not linear. The error amplitude appears to be approx 7mV (~ minus 66dB) for an equivalent of 16V at 8 ohms, though of course the amplifier output stage has it easy when it is merely attempting to maintain zero output volts.
Cheers ......... Graham.
Re: Re: Class-XD
Glen,
Well, it depends on what else has been done. If the amplifier has been made insensitive to gm doubling, XD becomes pointless. Moreover, if the THD is already below 1ppm who cares about any further reduction and certainly not at the expense of a additional 40W power dissipation or so. Besides, the improvement is minuscule. According to some my simulations, THD drops from 0.47ppm to 0.46ppm at 10kHz and 30W.
Cheers.
G.Kleinschmidt said:
Glen,
Well, it depends on what else has been done. If the amplifier has been made insensitive to gm doubling, XD becomes pointless. Moreover, if the THD is already below 1ppm who cares about any further reduction and certainly not at the expense of a additional 40W power dissipation or so. Besides, the improvement is minuscule. According to some my simulations, THD drops from 0.47ppm to 0.46ppm at 10kHz and 30W.
Cheers.