TPC vs TMC vs 'pure Cherry'

[ihan]

Hi Dave,

My spice simulations of a blameless amp (with MOSFET output stage) suggest that adding a small cap in parallel to the VAS emitter resistor makes an enormous positive difference to the phase margin. I agree with Richard that this technique can make 'pure Cherry' usable, although I have not tried it in practice (yet).

Cheers,
Ian

 

[ihan]

Hi Gents,

I'll attach my schematic and loop gain plots so you can see the difference the vas emitter caps make in pspice. Sadly, I don't use LTspice, so there is little you can do with them.

The PM is increased from 25 to 69 degrees. Notice that the gain margin is reduced, indicating no free lunch. It is important that the VAS is highly degenerated to begin with (110R resistors), in order to get plenty of GM that can be 'traded' for more PM.

I have sent a board to manufacture to (hopefully) validate this theory. I hope to have some results early next year - dependent upon 'proper' work commitments. I am grateful to Richard for bringing the effect of the VAS emitter cap to my attention. I have been a student of amplifier design for over 20yrs, and wasn't aware of it.

Thanks,
Ian

 

[Dave Zan]

I'll attach my schematic and loop gain plots so you can see the difference the vas emitter caps make in pspice. Sadly, I don't use LTspice, so there is little you can do with them.

I don't know details about compatibility of Pspice and LTspice but the netlist should be similar.
Can you post the netlist of the amp, no Gain probes etc. It should be possible to import it into LTspice.

The PM is increased from 25 to 69 degrees. Notice that the gain margin is reduced...

This is classic Bode feedback maximisation, well done.
The plots indicate your capacitor resistor combination is not quite optimal.

It is important that the VAS is highly degenerated to begin with (110R resistors), in order to get plenty of GM that can be 'traded' for more PM...

This is the area that Ric and I discussed a while back.
Educational, but we both felt it needed more study so thanks for the input.

Best wishes
David

 

[Dave Zan]

... a small cap in parallel to the VAS emitter resistor makes an enormous positive difference to the phase margin....

When Richard and I discussed there were cases where the capacitor made much less difference than he expected.
Here it seems that it does make a substantial difference, it is not clear to me exactly when it is effective.
Cherry wrote a JAES paper on this that I have read repeatedly but still haven't really conceptualized, or whatever the word is when you have no intuition for the problem.

Best wishes
David

 

[ihan]

Hi Dave,

I think I understand what you are saying about it being not quite optimised. Are you suggesting that I change the cap value so that I achieve the maximum possible PM? I have 'optimised' to achieve a PM of >60 degrees, and a GM of >12dB.

It would be useful to know what you consider as optimum (I understand you are writing an article, so you may not want to disclose this yet).

* PSpice Schematics Netlist without loopgain probe*

C_C98 0 b 100p
R_R556 $N_0002 $N_0001 3950
R_R563 $N_0002 $N_0003 56
R_R388 $N_0005 $N_0004 8k
R_R567 0 $N_0005 8k
R_R582 0 vin 100k
R_R560 $N_0002 vas- 10
C_C189 $N_0006 b 100u
R_R583 vin $N_0006 47
C_C184 vee $N_0005 100u
R_R456 $N_0004 $N_0007 56
C_C178 0 $N_0008 100n
R_R539 $N_0009 vout 5m
R_R540 $N_0008 feed 10
L_L14 feed $N_0009 1.3u
C_C180 $N_0002 $N_0010 100u
M_M80 $N_0010 $N_0001 $N_0003 $N_0003 IRF540
Q_Q302 $N_0011 $N_0004 $N_0012 KSC3503C
Q_Q303 $N_0004 $N_0012 vee KSC3503C
M_M90 vcc $N_0014 $N_0013 $N_0013 FDA20N50F
M_M91 vcc $N_0015 sense+ sense+ FDA20N50F
M_M92 vee $N_0016 $N_0017 $N_0017 IXTQ26P20P
M_M93 vee $N_0018 sense- sense- IXTQ26P20P
R_R544 ngate $N_0015 120
R_R543 ngate $N_0014 120
Q_Q282 vcc vas+ ngate MJE15032C
R_R552 pgate $N_0016 120
R_R553 pgate $N_0018 120
Q_Q283 vee vas- pgate MJE15033C
R_R610 pgate ngate 820
M_M94 vcc $N_0020 $N_0019 $N_0019 FDA20N50F
R_R611 ngate $N_0020 120
M_M95 vee $N_0021 $N_0022 $N_0022 IXTQ26P20P
R_R612 pgate $N_0021 120
R_R613 $N_0022 feed 0.22
R_R614 feed $N_0019 0.33
R_R546 feed $N_0013 0.33
R_R548 $N_0017 feed 0.33
R_R549 sense- feed 0.33
R_R547 feed sense+ 0.33
R_R615 vout feed 10
R_R352 $N_0023 b 2k
R_R354 0 $N_0023 68
C_C95 $N_0023 e 2200u
R_R383 vee $N_0012 100
R_R601 $N_0010 $N_0024 1m
R_R542 0 vout 2
R_R357 e feed 2k
R_R455 vee $N_0025 110
R_R619 vee $N_0026 110
Q_Q292 $N_0002 $N_0007 $N_0025 KSC3503C
Q_Q304 $N_0002 $N_0007 $N_0026 KSC3503C
R_R558 $N_0001 $N_0024 7.13k
V_V46 vin 0 DC 0 AC 0
+SIN 0 1.33 20k 0 0 0
R_R370 $N_0027 vcc 22
C_C101 0 $N_0027 2200u
R_R377 $N_0028 $N_0027 47
R_R376 $N_0029 $N_0027 47
Q_Q301 vasin b $N_0030 KSC3503C
R_R358 $N_0030 $N_0011 220
R_R359 $N_0011 $N_0031 220
R_R559 $N_0010 vas+ 10
Q_Q295 0 vasin $N_0032 KSA1381C
R_R569 $N_0032 $N_0033 470
Q_Q298 $N_0034 $N_0034 $N_0029 KSA1381C
Q_Q300 $N_0034 e $N_0031 KSC3503C
Q_Q299 vasin $N_0034 $N_0028 KSA1381C
V_V40 vcc 0 71V
V_V41 0 vee 71V
C_C173 vasin feed 47p
C_C214 $N_0033 $N_0027 270p
C_C216 $N_0035 $N_0027 270p
R_R464 $N_0033 $N_0027 110
R_R620 $N_0035 $N_0027 110
Q_Q305 $N_0024 $N_0032 $N_0035 KSA1381C
Q_Q296 $N_0024 $N_0032 $N_0033 KSA1381C

*** End of netlist ****

MOSFET models:

.MODEL FDA20N50F-X NMOS
LEVEL=3
L=2.0000E-6
W=88
KP=1.0302E-6
RS=10.000E-3
RD=.24532
VTO=5.1535
RDS=4.0000E6
TOX=2.0000E-6
CGSO=38.100E-12
CGDO=455.00E-15
CBD=2.3613E-9
RG=38.830
IS=1.0000E-18
RB=1.0000E-9
GAMMA=0
KAPPA=0


.MODEL IXTQ26P20P-X PMOS
LEVEL=3
L=2.0000E-6
W=62
KP=520.48E-9
RS=10.000E-3
RD=.15551
VTO=-3.9774
RDS=1.3333E6
TOX=2.0000E-6
CGSO=42.600E-12
CGDO=1.6100E-12
CBD=2.0243E-9
RG=10.000E-3
IS=1.0000E-18
RB=1.0000E-9
GAMMA=0
KAPPA=0
UO=300

All transistor models are Cordell's.

Cheers,
Ian

 

[Dave Zan]

...It would be useful to know what you consider as optimum ... you may not want to disclose this yet

I think we're alone now.
I just noticed Richard has been sin binned and most people seem to prefer debates without objective data.
So what I meant was that your phase has a peak at 10 MHz but your unity Return Ratio frequency is a few MHz lower.
I would try to move the peak frequency down until it's directly on top.
This will certainly improve your PM and I suspect it will actually improve the GM a little too, because it should be a more efficient use of the available phase.

I will try to import your netlist and see what I discover.

Best wishes
David

 

[ihan]

This is the area that Ric and I discussed a while back.
Educational, but we both felt it needed more study so thanks for the input.

Hi Dave,

My understanding (backed up by extensive spice work over the last few months, is as follows):

As the VAS emitter resistance is increased, the transconductance of the VAS is decreased, thereby reducing the loop gain. As a result, the ULGF decreases, and the GM increases.

My design 'process': Once sufficient GM has been established, by VAS emitter degeneration, the parallel cap (zero) can then be added in order increase the PM. Steadily increasing the cap (up-to a certain value) will increase the PM, whilst reducing the GM. I aim to get PM above 60 degrees, whilst maintaining a GM of at least 12dB. If no capacitor value can achieve that criteria, I increase the VAS emitter degeneration resistor and repeat the capacitor selection exercise.

My apologies if you knew all this already, I don't want to sound as if I'm preaching to the choir!

Best wishes,
Ian

 

[Dave Zan]

As the VAS emitter resistance is increased, the transconductance of the VAS is decreased, thereby reducing the loop gain. As a result, the ULGF decreases, and the GM increases.

It is not entirely clear to me how the transconductance of the VAS interacts when it works into a current source and there is Cherry compensation.
It is well known that in a Miller compensated amp the VAS transconductance matters little because the Miller capacitor sets the Return Ratio.
Whether the VAS transconductance makes a difference depends on the load and the feedback.

...if you knew all this already, I don't want to preach... to the choir!

Cherry did the maths but said he couldn't explain it simply.
So it seems no one really knows all this already.
I appreciate the chance to bounce ideas around.
Funny how much we learn when we try to teach.

Best wishes
David

Not much luck with the netlist.

 

[ihan]

Hi Dave,

If I attempt to optimise by increasing the caps to 360pF, the attached plot is the result. I get a larger PM, but my GM drops below 12dB, hence my preference for 270pF.

Cheers,
Ian

 

[ihan]

It is not entirely clear to me how the transconductance of the VAS interacts when it works into a current source and there is Cherry compensation.
It is well known that in a Miller compensated amp the VAS transconductance matters little because the Miller capacitor sets the Return Ratio.
Whether the VAS transconductance makes a difference depends on the load and the feedback.


Cherry did the maths but said he couldn't explain it simply.
So it seems no one really knows all this already.
I appreciate the chance to bounce ideas around.
Funny how much we learn when we try to teach.

Best wishes
David

Not much luck with the netlist.

Unfortunately, I must admit, it is not entirely clear to me either.

I also struggle with a simple and complete explanation. I visualise the VAS as a gain block within the Cherry compensation loop. Decreasing the gain of the VAS (by degeneration), decreases loop gain and thereby increases the GM. I appreciate that is a very simplistic view, and that many more questions need answering. It would seem that maths is the best domain for this problem (rather than words). Sadly, it is now 20yrs since university, and as a result, by mathematical abilities have faded. I look forward to trying to understand your future articles though.

Incidentally, I also have a loopgain sim of a 3 pole compensated amplifier (if you are interested).

Best wishes,
Ian

 

[Nikitas]

Hi Dave,

Are you going to discuss any real circuit in your coming article, or strictly simulations?

Cheers

 

[Dave Zan]

...Itwould seem that maths is the best domain for this problem (rather than words). Sadly, it is now 20yrs since university, and as a result, by mathematical abilities have faded. I look forward to trying to understand your future articles though.

I use mostly visual techniques like Bode plots, Nichols charts etc. to keep it understandable.
I have tried hard not to do maths in the articles but to explain what the maths does.
The problem is that the better it is explained then the more obvious the result seems and the less obvious how difficult it was to write

Incidentally, I also have a loopgain sim of a 3 pole compensated amplifier (if you are interested).

Most interested, thank you.
I will reply to your other points when I have an adequate answer.

Best wishes
David.

 

[Dave Zan]

...to discuss any real circuit... or strictly simulations?

The aim is to produce a family of real amplifiers.
600 W into 4 ohms.
300 W / 8.
150 W / 8.
75 W / 8.
They are very low distortion and ultra low noise.

I have started on layout of OPS but not totally finalized the circuit yet.
That article is still a few issues away so I have time.
The next article is fairly well finalized.

Best wishes
David

 

[kgrlee]

The PM is increased from 25 to 69 degrees. Notice that the gain margin is reduced, indicating no free lunch. It is important that the VAS is highly degenerated to begin with (110R resistors), in order to get plenty of GM that can be 'traded' for more PM.

PM & GM are just 2 points on the Nyquist Curve.

It's the distance of Nyquist from (-1,0) that determines how much peaking takes place. This comes from the denominator in the

G(s)/[1 + G(s)] feedback formula.

The closer, Nyquist gets to (-1,0), the more the peaking ... which becomes infinite when it touches or encircles.

So try to get the closest approach as far from (-1,0) as possible rather than maximise one or the other.

LTspice has a nice Nyquist plot.

A Nichols plot is similar but has contours with the amount of peaking. You can read the peaking directly from the plot. I think Dave had some success making LTspice do this.
_____________

Guru Zan, I'm out on good behaviour pending my appeal on the grounds of innocence.

Please don't provoke me with certain names. I'm on parole.

 

[ihan]

Thanks Richard, I certainly appreciate PM and GM are just two points on a Nyquist plot.

I'm pretty certain my antiquated simulator won't do Nyquist plots, but I'll look into it. At some stage (when I have time), I will move over to LTSPICE, but I must admit when I tried it years ago, the user interface put me off.

Dave, attached are the 3 pole compensated schematic and loop gain plots.
I've termed it 3 pole, since the loop gain initially falls at almost 18dB/octave (with a corresponding loop phase of <0).

Notice that I used a cap across the feedback resistor (zero) so that the loopgain slope is sensible at the ULGF, giving a decent PM (84 degrees).

Best wishes,
Ian

 

[dadod]

PM & GM are just 2 points on the Nyquist Curve.

It's the distance of Nyquist from (-1,0) that determines how much peaking takes place. This comes from the denominator in the

G(s)/[1 + G(s)] feedback formula.

The closer, Nyquist gets to (-1,0), the more the peaking ... which becomes infinite when it touches or encircles.

So try to get the closest approach as far from (-1,0) as possible rather than maximise one or the other.

LTspice has a nice Nyquist plot.

Like this one? How would you interpret it?
Damir

 

[Dave Zan]

Thanks Richard, I certainly appreciate PM and GM are just two points on a Nyquist plot.

Your appreciation is excellent. I spoke too soon when I expected to improve the compensation easily.

...(when I have time), I will move over to LTSPICE, but I must admit when I tried it years ago, the user interface put me off.

I don't like the UI much either but I'm used to it now.
In other aspects the product is impressive, a pity there is not more documentation, especially on how to interface extensions.

Dave, attached are the 3 pole compensated schematic and loop gain plots.

Thank you.

Notice that I used a cap across the feedback resistor (zero) so that the loopgain slope is sensible at the ULGF, giving a decent PM (84 degrees).

Yes, I like this technique too.
Bob Cordell mentions increased risk of RFI but I think that should be dealt with elsewhere.
I notice your feedback resistance is very low, any particular reason?

Best wishes
David

 

[Dave Zan]

PM & GM are just 2 points on the Nyquist Curve...

Horowitz and also Lurie discuss the shape of the curve near the critical point.
On a Nichols plot a specified GM and PM leave a rectilinear shape.
A smoother shape is possibly better.
I think it is pretty much academic because in practice there is not much control at that frequency and the curve is inevitably smooth anyway.
2 points looks sufficient to define the stability in practice I think, you don't?

Best wishes
David

 

[catalin]

Dave, attached are the 3 pole compensated schematic and loop gain plots.
I've termed it 3 pole, since the loop gain initially falls at almost 18dB/octave (with a corresponding loop phase of <0).

Notice that I used a cap across the feedback resistor (zero) so that the loopgain slope is sensible at the ULGF, giving a decent PM (84 degrees).

Best wishes,
Ian

Hello Ian ,My question is what is happening at 11khz and 90khz .I am seeing that the phase is 0 degrees at this frequencies meaning an returning signal at the negative input with gain .

 

[ihan]

Hello Catalin,

You are quite correct. Do not worry, it is a stable design.

I suspect (like me) that you were taught at an academic institution, that for a system to be stable, the loop gain cannot exceed unity (0dB) at frequencies where the loop phase is 0. That isn't strictly true.

Best wishes,
Ian