Bob Cordell Interview: Negative Feedback

I go for worst case slew rate, not some compromise. After all, I have to design for everyone, not just for my chosen phono cartridge or CD player. Dr. Lipshitz once looked for worst case slew rate with an old Shure phono cartridge. He didn't find much either.
I personally have found that Ortofon cartridges that could 'spit' garbage to 500kHz. Most good MC cartridges went up to 200KHz. I published this almost 30 years ago.
 
mikeks said:


If it is assumed that by ''complementary VAS'' you are referring to the second stage in your design here, then your supposition that it would necessarily require two miller capacitors for compensation is incorrect: only one is required.



''worry.....needlessly'' is right.




Not necessarily true either. One needs to define with precision what one means by ''sub-optimal slew rate'' and in what context the phrase is applied.



Not true. Your feedback compensation is actually phase lead compensation. Completely unrelated to Miller-type compensation in theory and in fact.

Moreover, this sort of compensation must, of necessity, be augmented by shunt and/or miller compensation if adequate stability margins are to obtained with your design.

Finally, see:

http://www.diyaudio.com/forums/showthread.php?s=&postid=1047674&highlight=#post1047674


Mike,

In an earlier post I described the feedback compensation I used in my MOSFET power amplifier as a variant of Miller compensation. I stated:

“…its main compensation is provided by what is effectively Miller compensation by feedback from the VAS output node back to the inverting input node of the input differential pair. The loop so formed does itself also need compensation, albeit at a much higher unity gain frequency, and that is provided by a simple series R-C shunt across the collectors of the cascoded input differential pair. This kind of compensation results in extraordinarily high slew rate capability and HF linearity without the usual compromises of classic simple Miller compensation.”

You disagreed, claiming that it is phase lead compensation. In doing so, you stated:

“The only difference between your approach and the usual phase lead approach (RC network directly shunting feedback resistor) resides in the fact that with the former the RC net (R13/C4) is connected to the second stage's output; output stage singularities are thus bypassed, and a trivial increase in gain margin accrues… Your feedback compensation is actually phase lead compensation. Completely unrelated to Miller-type compensation in theory and in fact.”

While at first blush this may seem a trivial difference since the output stage has nominally unity gain (so the two tap-off points have largely the same signal voltage) , in practice it is a HUGE difference because the output stage is the primary source of de-stabilizing excess phase in the open loop.

Here is why I refer to my compensation as a form of Miller compensation. In conventional Miller compensation, a capacitor is connected from the output of the inverting CE VAS to its input. This forms what is known as a Miller integrator, and gives the open loop response the characteristic of an integrator, with the usual 6 dB/octave rolloff. Those familiar with Miller integrators know that they are also often formed by connecting a Miller capacitor from the output of an op amp to its inverting input. In that case, the Miller capacitor spans more than one amplifying stage, since the op amp consists of multiple stages. In my compensation, I also form a Miller integrator in the forward path by connecting the Miller compensating capacitor to surround multiple stages. In my case, the capacitor surrounds both the input stage and the VAS stage. The overall effect is still the same, namely to give the open loop response a Miller integrator characteristic.

The connection I use has the added benefit that the negative feedback inherent to the Miller integrator implementation also encloses the input stage, reducing its distortion and increasing its dynamic range. This latter increase in its dynamic range is the reason why this form of “Miller” compensation breaks the constraining relationship with slew rate that conventional Miller compensation imposes.

As I mentioned in my earlier post, the local loop formed by my form of “Miller” compensation can have a very high gain crossover frequency, on the order of 10-20 MHz) because it only involves fast, small-signal transistors. Yes, this inner loop does need to be compensated, but that compensation is very light due to the high gain crossover frequency of this loop.

It is also true that the feedback nature of this modified “Miller” compensation preserves the desirable pole-splitting nature of conventional Miller compensation

What we call this form of compensation matters much less than how it works and how it performs. Call it lead compensation if you wish, and take me to task for referring to it as a form of “Miller” compensation if you wish. But that does not change its superior behavior, and does not cause it to require extra compensation elements (beyond what I have already described) to be rock-stable. Your statement here:

“Moreover, this sort of compensation must, of necessity, be augmented by shunt and/or miller compensation if adequate stability margins are to obtained with your design.”

is simply not so. The only additional compensation is that for the inner loop which I already described. Your statement here is simply evidence that you do not understand the workings of this compensation.

Cheers,
Bob
 
Bob Cordell said:
.....the output stage is the primary source of de-stabilizing excess phase in the open loop.

Hi Bob,

This is actually untrue, particularly if your preferred option of a MOSFET output stage is exercised.

Bob Cordell said:

Here is why I refer to my compensation as a form of Miller compensation............. In my compensation, I also form a Miller integrator in the forward path by connecting the Miller compensating capacitor to surround multiple stages.

In my case, the capacitor surrounds both the input stage and the VAS stage. The overall effect is still the same, namely to give the open loop response a Miller integrator characteristic.



Bob, for a start Miller compensation is not so-called because ''compensating capacitor surrounds multiple stages'', but because of attributes conferred by the shunt-derived shunt-applied minor feedback loop about an inverting forward path, something you've also acknowledged.

By definition, therefore, your compensation cannot confer the same response since, contrary to your assertion that it ''surrounds both the input stage and the VAS stage'', your compensation loop essentially constitutes a feedforward path about the output stage.

Moreover, the input stage is not included in said feedforward loop, as the only active block so-enclosed is the non-inverting output stage.

For these reasons your assertion that
It is also true that the feedback nature of this modified “Miller” compensation preserves the desirable pole-splitting nature of conventional Miller compensation.
is untrue indeed.
 
mikeks said:


Hi Bob,

This is actually untrue, particularly if your preferred option of a MOSFET output stage is exercised.





Bob, for a start Miller compensation is not so-called because ''compensating capacitor surrounds multiple stages'', but because of attributes conferred by the shunt-derived shunt-applied minor feedback loop about an inverting forward path, something you've also acknowledged.

By definition, therefore, your compensation cannot confer the same response since, contrary to your assertion that it ''surrounds both the input stage and the VAS stage'', your compensation loop essentially constitutes a feedforward path about the output stage.

Moreover, the input stage is not included in said feedforward loop, as the only active block so-enclosed is the non-inverting output stage.

For these reasons your assertion that is untrue indeed.


Have it your way Mike.

I'm sure most of the others who read my explanation understood it, and are less interested in a semantical debate. It works, and the proof is in the pudding; more than can be said for much of the product of your handwaiving and debating.

If anyone else on this board has a question about it, I will try once again to better explain it.

Bob
 
Hi, Mr. Bob,

I'm really interested in this stabilizing method. It seems not much option can be done for this.

I have 2 question.

1. Refering to your amp. From your explenation above, stabilization can be done ONLY with C4. What is the reason you also put C3?

2. Are there any other stabilization method besides putting miller cap on VAS, using C4 and C3 (in your amp)? Can i say if we put heavy local feedback anywhere (RE), for example 1-2k2 on differential, also heavy RE on VAS, the amp tends to be stable without any small cap anywhere on the CCT?
 
lumanauw said:
Hi, Mr. Bob,

I'm really interested in this stabilizing method. It seems not much option can be done for this.

I have 2 question.

1. Refering to your amp. From your explenation above, stabilization can be done ONLY with C4. What is the reason you also put C3?

2. Are there any other stabilization method besides putting miller cap on VAS, using C4 and C3 (in your amp)? Can i say if we put heavy local feedback anywhere (RE), for example 1-2k2 on differential, also heavy RE on VAS, the amp tends to be stable without any small cap anywhere on the CCT?


The capacitor C4 (20 pF), from the VAS output to the inverting input of the input differential pair, is the main compensation capacitor that sets the closed loop bandwidth of the amplifier and makes the open loop gain function one of an integrator. Resistor R13 (680) in series with C4 merely creates a zero in the forward path (at about 12 MHz), in much the same way that people often do with conventional Miller compensation. Above 12 MHz, the closed loop gain of this "inner loop" is determined by R13 against R11 (251), for a gain of about 4.

This "inner loop" needs also to be compensated. The compensation for the inner loop is provided by simple shunt compensation by C3 and R14. The need for stabilization of the inner loop is why C3 is needed.

I'm sure there are other ways of compensation that could be applied to this circuit, such as simple shunt compensation at the collector of the VAS. But in general, SOME kind of compensation is generally needed, even with additional degeneration. At some point, however, the need for the C3 portion of the compensation might go away, especially, perhaps, if a modest load resistor were placed across the collectors of the cascodes from the input differential pair.

Note also that this kind of compensation could also be applied to the popular dual-complementary differential input pair architecture that I discussed in an earlier post. It might eliminate or mitigate the problem I mentioned of the upper and lower VAS transistors tending to fight each other when two imperfectly-matched Miller capacitors are used.

Cheers,
Bob
 
Bob Cordell said:

...more than can be said for much of the product of your handwaiving and debating.
Bob

Hello Bob,

No hand waving involved.

One need only run a simple SPICE set-up to establish that pole-splitting does not and cannot occur with your modification of phase lead compensation.

This is why shunt compensation and/or proper Miller compensation are required to complement your arrangement if stability is to be reliably attained.

These are not merely articles of faith, but ice-cold, hard, unyielding fact.

Ciao!
 
mikeks said:


Hello Bob,

No hand waving involved.

One need only run a simple SPICE set-up to establish that pole-splitting does not and cannot occur with your modification of phase lead compensation.

This is why shunt compensation and/or proper Miller compensation are required to complement your arrangement if stability is to be reliably attained.

These are not merely articles of faith, but ice-cold, hard, unyielding fact.

Ciao!


So if I understand you correctly, you are saying that my MOSFET amplifier, as shown in the published schematic, has inadequate stability. This, recognizing that it does not matter whether we call that compensation lead compensation or a variant of Miller compensation.

Am I correct?

I encourage you to prove your point by properly SPICEing the circuit.

Cheers,
Bob
 
lumanauw said:
Hi, Mr. Cordell,

What kind of test do you use to determine stability?

I used to do this. The amp is playing music, and put the output to dummy load 4ohm // with 2uF capacitor. With this test, I can "judge" if my amp is already stable or not.

Is this good enough?


One of the things I use to determine stability is to feed the amplifier with a fairly wideband square wave and look at the various nodes in the circuit to see evidence of ringing and/or parasitic oscillation. The mere absence of parasitic oscillation is not enough - you need to establish that there is no HF ringing as well. I also do these investigations with differing amounts of closed loop gain. For example, if your amplifier is normally set up for a gain of 20, and you change the feedback resistor so that it has a gain of 10, you would like the amplifier to be still stable; this is an approximate way to verify that you have 6 dB of gain margin.

Similarly, I also experiment with the amplifier using compensation that is lighter than what I will ultimately use, to see if it has margin against instability in that sense as well. Finally, I do look at the sinusoidal frequency response of the amplifier out to very high frequencies. In that test, you want to see little or no peaking just prior to the closed loop frequency response rolloff. One also wants to do this test with a variety of loads connected to the output of the amplifier. It is also important to do some of these tests over a range of powers, since sometimes poles in the open loop may shift as device capacitances and/or transconductances may change a bit with signal excursion.

A rigorous lab test of amplifier stability is always a must.

Cheers,
Bob
 
Bob Cordell said:
So if I understand you correctly, you are saying that my MOSFET amplifier, as shown in the published schematic, has inadequate stability.

No, you clearly do not ''understand you correctly''.

Your design is stable because you supplement lead compansation with shunt (lag) compensation.

This is transparently noted here and here.

Bob Cordell said:
....prove your point by properly SPICEing the circuit.

I have proved this to my satisfaction. I'll leave re-examination to those who deem it necessary.
 
mikeks said:

One need only run a simple SPICE set-up to establish that pole-splitting does not and cannot occur with your modification of phase lead compensation.

Methinks, that aruments of thine art incorrect 😉
Or simply : Nope
This arrangement all in all makes output Z of VAS decrease at high frequency, so driving output buffer with lower impedance causes its pole to move to higher frequency. Pole splitiing, isn't it?