Bob Cordell's Power amplifier book
 User Name Stay logged in? Password
 Home Forums Rules Articles Store Gallery Blogs Register Donations FAQ Calendar Search Today's Posts Mark Forums Read Search

 Solid State Talk all about solid state amplification.

 Please consider donating to help us continue to serve you. Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving
diyAudio Member

Join Date: Dec 2006
Quote:
 Originally Posted by jcx This sim moves the simple Middlebrook gain probe to the “inner” feedback loop, measuring all of the gain around the output stage I think you can see it is a much more useful way to view the stability margin of TMC To look at the phase margin numbers with the added delay we have to convert the delay time into phase – since delay has linear phase with frequency the phase depends on the frequency For the CMC loop 100 nS delay at the 977KHz gain intercept is: 100ns * 977kHz * 360 degrees = 35 degrees added phase shift the cursor reading gives 77.8 degrees phase margin for the CMC at 0 delay, 42.6 degrees at 100 ns (append @5 to both V in the waveform formula, ie V(d_cmc)@5/V(b_cmc)@5) 77.8 - 35.2 = 42.6 which agrees with the idea that the added delay is a fair measure of phase margin the numbers clearly don’t work so neatly in the outer loop gain/phase plot of TMC because the gain intercept is a moving target the Inner loop Bode gain/phase measurement of TMC below Is well behaved and we can do the same calculation: 100ns * 1.69MHz * 360 = 60.8 degrees added phase shift the cursor reading gives 58.7 degrees phase margin for the CMC at 0 delay, -2 degrees at tau = 100 ns, step 5 58.7 – 60.8 = -2.1 which again agrees and predicts that the TMC loop will be unstable with 100 ns added delay – as the tran sim pic above shows as a further check note that 100nS is just over the 0 degree line and the previous step at 75ns delay has poor but positive phase margin – and the tran sim shows it does settle out instead of oscillate (Thanks to Mike Engelhardt of Linear Technology – you may have noticed the stepped waveform colors didn’t agree in the previous post Bode plot but they do match in this sims pic – the 1st time I’ve had a bug fixed over lunch! ) I assume anyone wanting to verify can mod the .asc in my previous post
A first question: I do not understand why in the simple Miller loop, with the values given, the loop transmission crossover frequency is not 10Mhz/16 =625 Khz instead of 1 Mhz.

Thanks

JPV

 4th January 2011, 11:49 PM #1342 diyAudio Member   Join Date: Feb 2003 Location: .. because you missed the doubling of the diff pair gain by the mirror? 2*(1/(2*(470 Ohm + 26 mv/0.5 ma))) ~= 1/500 mho with ~15 pF Cdom ~= 20 MHz GBW then with the beta of 1/20 pretty much gives the observed 1 MHz corner freq for more fun with my delay sim you can edit both circuits to TMC and measure the outer loop gain in one and inner loop gain in the other the final 100 ns delay step @5 gives very close to the same `1.7 MHz intercept and -2 to -4 degrees phase margin - so you could say measuring the outer loop gets you the same stability result if you add delay externally - it just doesn't help you predict the stability limit from the 0 delay plot a good question is why the resistance to measuring the inner loop? Probably because we expect just measuring the outer loop to be right from experience with 1st order loops as in the CMC case there are complex rules for measuring multiloop circuit stability by "loop cutting" methods - the simplest is that measuring the global outer loop gain/phase margin is sufficient if all internal loops are negative feedback and are stable - I think the "fail" is that TMC is a positive feedback loop - admittedly expected to be stable if the output stage gain is less than unity I think the probe position in my sims is adequate because it breaks both the global and local output stage feedback at the same time and the remaining “inner” miller feedback around the VAS is negative feedback and stable to anticipate a possible objection - the delay element could be put in-between the driver and output Q with little effect on the result if you just like the visual appearance of not modifying feedback component connections the delay following the output Q and RL is probably a little better since the pre/driver see the output Q's base load - I don't know how to make a "reciprocal" delay Last edited by jcx; 4th January 2011 at 11:58 PM.
diyAudio Member

Join Date: Sep 2006
Quote:
 Originally Posted by wahab Hi, Bob In fact, a cap as low as 0.2pF is enough to get rid of the TPC OL gain peaking, so in real implementation, it s unprobable that such peakings occur as the parasistic layout and components leads caps are enough to create this cap, while not decreasing available gain. Though, as soon as the loop is closed, the usual step response peaking will remain, since the TPC loop becomes then dominant. As i already pointed, a TMC network pondered with a TPC one is generaly the optimmum in respect of the global behaviour, although i m not still sure if there isnt undesirable by products. cheers, w
Hi wahab,

The amount of bridging capacitance needed to suppress the loop gain peaking will of course be highly dependent on the particular design. 0.2pF was not enough to do it in my design. Other designs, which use higher overall values of compensation capacitance will need even more bridging capacitance. Of course, a design that does not use an emitter follower in front of the VAS transistor will typically not need the bridging capacitor because that function will be performed by the collector-base capacitance. But I don't think we are talking about such low-performance designs here.

I would not leave the bridging capacitance to chance. In fact, I think a decent layout would not have anywhere near 0.5 pF (or even 0.2 pF) from the VAS collector to the input of the emitter follower feeding the VAS transistor. On the other hand, one could do the calculations and deliberately create some PWB inter-trace capacitance to fulfill the role of C3.

"As i already pointed, a TMC network pondered with a TPC one
is generaly the optimmum in respect of the global behaviour,
although i m not still sure if there isnt undesirable by products."

Cheers,
Bob

diyAudio Member

Join Date: Sep 2006
Quote:
 Originally Posted by jcx because you missed the doubling of the diff pair gain by the mirror? 2*(1/(2*(470 Ohm + 26 mv/0.5 ma))) ~= 1/500 mho with ~15 pF Cdom ~= 20 MHz GBW then with the beta of 1/20 pretty much gives the observed 1 MHz corner freq for more fun with my delay sim you can edit both circuits to TMC and measure the outer loop gain in one and inner loop gain in the other the final 100 ns delay step @5 gives very close to the same `1.7 MHz intercept and -2 to -4 degrees phase margin - so you could say measuring the outer loop gets you the same stability result if you add delay externally - it just doesn't help you predict the stability limit from the 0 delay plot a good question is why the resistance to measuring the inner loop? Probably because we expect just measuring the outer loop to be right from experience with 1st order loops as in the CMC case there are complex rules for measuring multiloop circuit stability by "loop cutting" methods - the simplest is that measuring the global outer loop gain/phase margin is sufficient if all internal loops are negative feedback and are stable - I think the "fail" is that TMC is a positive feedback loop - admittedly expected to be stable if the output stage gain is less than unity I think the probe position in my sims is adequate because it breaks both the global and local output stage feedback at the same time and the remaining “inner” miller feedback around the VAS is negative feedback and stable to anticipate a possible objection - the delay element could be put in-between the driver and output Q with little effect on the result if you just like the visual appearance of not modifying feedback component connections the delay following the output Q and RL is probably a little better since the pre/driver see the output Q's base load - I don't know how to make a "reciprocal" delay
Hi jcx,

My understanding and sims do not indicate the the inner loop around the output stage is a positive feedback loop. This is the loop that starts at the output of the amplifier, passes through R1, and gets to the input of the VAS through C1. Although C2 is connected to the input of the output stage and to the output of the VAS, it provides less transmission to the output stage than the path through C1. If you open the global loop, and also open the local loop at R1, then apply a signal to R1, you will see the loop gain of the local loop.

You are certainly correct that just looking at the global loop is insufficient. The total amount of feedback enclosing the output stage derives from two paths, one the global feedback path and the other the local loop path. The sum of the loop signals of both of these paths is what causes the output stage to be enclosed by the feedback that has the peak and typically has a higher gain crossover frequency.

Cheers,
Bob

diyAudio Member

Join Date: Oct 2008
Quote:
 Originally Posted by Bob Cordell Hi wahab, Of course, a design that does not use an emitter follower in front of the VAS transistor will typically not need the bridging capacitor because that function will be performed by the collector-base capacitance. But I don't think we are talking about such low-performance designs here. Cheers, Bob

diyAudio Member

Join Date: Sep 2006
Quote:
 Originally Posted by stinius Please remember, there is more than one road to Rome.
Hi Stinius,

This is certainly true, but it is only my opinion that before one applies TPC or TMC to get really high performance out of a design, they should have at least given the design a decent VAS (with an emitter follower in front of it, what we often call a Darlington VAS). Just my opinion on priorities, and others may of course differ.

Cheers,
Bob

diyAudio Member

Join Date: Nov 2003
Location: Amsterdam
stability test

Quote:
 Originally Posted by jcx I don't suppose a direct sim of robustness to added output delay will cause anyone to rethink? As I stated earlier in this thread a sticking point seems to be people wanting to believe that the outer loop gain/phase margin Bode plot of TMC looking like CMC means you managed greater loop gain/distortion reduction - "for free" I find it hard to believe the resistance to the TANSTAAFL principle among a engineering oriented community

Hi jcx,

I'm not sure if you consider me part that 'engineering oriented community', but years ago I already said that any compensation trick (and even HEC) that lowers the distortion, has an impact on the stability. So I fully endorse the TANSTAAFL principle.

Quote:
 Specifically I have pointed to Bode's Integral relations and BJ Lurie's work that shows how this "conservation law" for feedback is applied My earlier output conductance sim clearly established that the added feedback is externally visible - simply by "tugging" on the output node you can tell TMC from CMC the Bode Integral relations lead me to expect more feedback has a cost in loop gain slope and phase margin - the extra distortion reduction performance of TMC - is not "free"
Referring to post 1106, where I stated that TMC is essentially a 1st order system, that is, at least with idealized components, your output conductance sim becomes pointless, as the output impedance is zero.
(just teasing a bit)

Quote:
 Now I have another sim - this time I packaged up Bob's amp, bringing out the VAS connections so compensation can be seen as external components I've added a pure delay to the output of the amp - between the loaded output and the normally output connected feedback components to sim variable "excess poles" - which Bob mentions in his book can be approximated as delay. [snip] In sum: You Have been bitten by those inner loops - TMC is not as stable as CMC, the gain/phase measured in the outer loop is not informative in the TMC case
Till now, I ran similar stability tests by means of a capacitor (5...100nF) directly tied to the output and gnd. Together with the lead or trace inductances, additional poles were created. It appeared that, compared to CMC, TMC amps becomes already unstable with smaller caps.

However, your stability test, using a simple delay, provides less ambiguous results, as it doesn't rely on inductances, which may vary from amp to amp. IOW, it's more 'robust'.

Cheers,
E.
__________________
Een volk dat voor tirannen zwicht, zal meer dan lijf en
goed verliezen dan dooft het licht…(H.M. van Randwijk)

diyAudio Member

Join Date: Dec 2006
Quote:
 Originally Posted by jcx because you missed the doubling of the diff pair gain by the mirror? 2*(1/(2*(470 Ohm + 26 mv/0.5 ma))) ~= 1/500 mho with ~15 pF Cdom ~= 20 MHz GBW then with the beta of 1/20 pretty much gives the observed 1 MHz corner freq
I didn't miss that. Simply I took the 30 pF of Bob's book instead of your 15which gives as GBW: gm/(2pi C) = 1/ 500 2 pi 30 ) = 10Mhz.
I read on your shematic a resistor of 3000 instead of 2000 which gives a beta of 16 instead of 20, therefore my mistake

 5th January 2011, 03:21 PM #1349 Banned   Join Date: Oct 2010 Edmond, you mean you give distortion and get linearity and stability in your 1st order system? Sounds like a good deal.
 5th January 2011, 03:34 PM #1350 diyAudio Member   Join Date: May 2003 Location: Northern Va. Oh boy! The beginnings of a new obtuse discussion....or is it the same one?

 Posting Rules You may not post new threads You may not post replies You may not post attachments You may not edit your posts BB code is On Smilies are On [IMG] code is On HTML code is OffTrackbacks are Off Pingbacks are Off Refbacks are Off Forum Rules

 Similar Threads Thread Thread Starter Forum Replies Last Post Bouvett Everything Else 30 13th August 2012 02:43 AM giulianodes Equipment & Tools 12 3rd August 2010 01:57 PM pixie Everything Else 27 11th June 2010 08:36 PM Karl71 Solid State 46 6th October 2008 03:47 AM mikee12345 Solid State 8 16th November 2003 01:16 AM

 New To Site? Need Help?

All times are GMT. The time now is 08:43 PM.