Bob Cordell Interview: Negative Feedback

Class-XD

Hi Mike,

Let’s go back to alternatives for class-XD.
I wonder why D. Self hasn’t opted for a far more simple solution, consisting of a slight modification of the Miller loop. Only one additional C and R are needed, giving a thd reduction of 80% or more. (No, it’s NOT a two pole compensation). It’s a very old trick, as far as I know, first pioneered by Baxandall. Curiously, D. Self does know about it, but for some reasons, he has rejected it. Why?! That is my point.

Regards,
 
Graham Maynard said:
Hi Nelson,

Quite so, but it will produce an individual sound that customers can -and do- like, and with the Patent, no one else can copy that 'sound'. Hi-Fi World did a review of an XD amplifier in their May 2006 issue. Of course their early sample could have been wrongly set up or their test bench different to what Cambridge and D Self uses, but the crossover distortion and its offset were clearly visible.

They did stress however, that there is much more to reproduced sound than crossover distortion and amplifier types, with which I must agree.

Will be back shortly to examine PMA's interesting reverse testing responses.

Cheers ........ Graham.


Hi Graham,

Thanks for posting this picture of 10 KHz 1W crossover distortion of the XD amplifier. This picture is horrid on a number of levels. First, 0.04% of XO distortion at only 10 kHz is pretty bad, especially when one realises the crest factor of XO distortion. Secondly, this is at only 1 watt! If the XD circuit is working, one should not even be entering the crossover region at only one watt. Something is terribly wrong here (although I usually don't buy the argument that "this was justa bad sample from the factory").

Bob
 
Re: Class-XD

estuart said:
Hi Mike,

Let’s go back to alternatives for class-XD.
I wonder why D. Self hasn’t opted for a far more simple solution, consisting of a slight modification of the Miller loop. Only one additional C and R are needed, giving a thd reduction of 80% or more. (No, it’s NOT a two pole compensation). It’s a very old trick, as far as I know, first pioneered by Baxandall. Curiously, D. Self does know about it, but for some reasons, he has rejected it. Why?! That is my point.

Regards,

Hi Edmond,

I assume you're referring to a C-R connection between the output and the inverting input of the amp.?

If so, then it is lead compensation, and it was not, as far as I know, first developed by Baxandall.

Alternatively, if it is assumed you're referring to the connection of a small resistor in series with the Miller capacitor, then you'll accomplish exactly nothing in respect of linearity, since this resistor's only raison d'etre is to provide nominal cancellation of the RHP zero occasioned by the HF feedforward path through the Miller capacitor. This, incidentally, is of no consequence in an amplifier with a BJT transimpedance stage.
 
Re: Re: Class-XD

mikeks said:
I assume you're referring to a C-R connection between the output and the inverting input of the amp.?
Alternatively, if it is assumed you're referring to the connection of a small resistor in series with the Miller capacitor...

Hi Mike,

Nope. Neither of the two. I’ll give you one more change to figure it out.
As an expert on this subject, I’m convinced you will find the answer.

Cheers,

PS: Now, I'm going to my local pub, see you later.
 
Hi PMA,

Looking at the reverse driven scans.

#474. The traces are horizontal so ostensibly the phase response of the output stage is flat to external stimulus, however there is that lack of return to zero when the applied current (back-EMF wrt input) returns to zero.
Also the error increases with frequency like a half cycle slew energised offset, so I cannot begin to imagine what this might sound like because the error is greatest when the leading current returns/passes through zero.
If this trace was along a linear axis the zero voltage crossovers would be fractionally shifted in time yet the peak amplitude points would be linear, so this could be some sort of quadrature slew induced amplitude modulation.

#475. The symasym shows phase shifted damping, thus the error component would be greatest at peak reverse energisation currents wrt input. There is an attempt to return to zero, but here it shows a transconductance problem due to what I know to be caused by the reflection point of differenced currents between first and second stages caused by the nature of the circuit failing to prevent a reference current shift at higher frequencies.
 
Hi Graham,

the narrower is the bandwidth, the wider is the Zout(f) ellipse. Think it over. Also, try to realize the real interpretation of plots, concerning they reflect frequency response and phase shift as well. This is not a DC plot. Take into account dV(f) at the output (less than 10mV) in ratio to output voltage. For tube or non-NFB design, you would obtain volts or hundreds of milivolts respectively.

Regards,
Pavel
 
The one and only
Joined 2001
Paid Member
Bob Cordell said:
First, 0.04% of XO distortion at only 10 kHz is pretty bad, especially when one realises the crest factor of XO distortion. Secondly, this is at only 1 watt! If the XD circuit is working, one should not even be entering the crossover region at only one watt. Something is terribly wrong here

It does seem awfully odd. While I never assume a priori
that .04% is going to sound bad, I don't expect the circuit as
described to have that "feature", certainly no amplifier with a
1 Amp bias.

:cool:
 
Hi PMA,

Without the circuit you cannot expect me to guess what might be going on ! I wonder if I am expected to think that plot is satisfactory ?

There is an error in #474 at higher audio frequencies, and this is of greater magnitude than the Symasym for most of the time.

Of course we get greater error voltages with tube and non-feedback designs, and this error voltage variation in time (which is generally far greater than resistively measured THD figures) contributes towards making amplifiers sound different.

Cheers .......... Graham.
 
Tim__x said:
I think Estuart is referring to a Cherry-esque second miller loop containing the output stage, with an R in series so that the traditional miller loop becomes dominant at high frequencies to maintain the original stability.


Exactly, that’s the crux. Chapeaux!
See below for this simple but valuable trick.
Note: It doesn’t work without a preceding beta enhancement transistor in the VAS stage. Probably, that’s why D. Self misunderstood the potential of this subtle modification.

Cheers,
 

Attachments

  • splitcomp.gif
    splitcomp.gif
    8.7 KB · Views: 1,333
Using a resistor from output to Darlington VAS base can also help to reduce output stage crossover distortion and improve NFB response by flattening the open loop bandwidth.
Using an equal value resistor from other half of mirror to ground prevents start-up thump and preserves differential input balance.

This is a Sim demo cct only.

Cheers .......... Graham.
 

Attachments

  • 0504ab.gif
    0504ab.gif
    8 KB · Views: 1,303
estuart said:
Exactly, that’s the crux. Chapeaux!
See below for this simple but valuable trick.
Note: It doesn’t work without a preceding beta enhancement transistor in the VAS stage. Probably, that’s why D. Self misunderstood the potential of this subtle modification.
Cheers,

This accomplishes nothing as far as linearity is concerned: loop transmission at 20KHz is typically 25dB~30dB, while DPC can be persuaded to deliver ~60dB at the same frequency.