Ounce again here is the sim:
http://www.diyaudio.com/forums/soli...terview-negative-feedback-55.html#post1160809
My concern was more with changes in the output stage transfer function and sticking.
I'll try to be more clear/specific here. There are two, perhaps more reasons to consider different compensation schemes. Stability, distortion reduction, and best recovery and stability during overload. Jcx's and Megajock's models are completely linear, however I do understand that they will probably say that with X amount of loop gain available at a given frequency we know what the distortion reduction will be. Still, this does not explain why Bob's simulations of complete amplifiers provide lower distortion with TMC. Anyone care to explain and show how to make Bob's full amp sims provide equal distortion with both types of compensation? I've not looked at them in much detail - will go take a look now. Also, I'd expect there to be greater front end overload and inferior global loop stabilty with TPC when the output stage becomes slower in overload/saturation.
Simple linear models provide some insight into the behavior of these systems, however they do not tell the whole story and it is short sighted to not look further into the details.
Simple linear models provide some insight into the behavior of these systems, however they do not tell the whole story and it is short sighted to not look further into the details.
Last edited:
But the loop gain around the output stage, as can be simulated by linear models, is just what you want to know to be able to predict stability if there are changes in the output stage+load transfer function.
It is when the front end itself deviates significantly from the linear model you need something more sophisticated.
As far as distortion is concerned, there is no contradiction in that the distortion changes in a more sophisticated model/reality because the distortion contribution from the front end may well be different even though the loop gain and distortion reduction from the output stage is about the same for both compensation networks.
Moreover, weren't Bob's simulations done with different compensation component values for TPC and TMC, giving different loop crossover frequencies and/or phase margins in the loop gain around the output stage?
It is when the front end itself deviates significantly from the linear model you need something more sophisticated.
As far as distortion is concerned, there is no contradiction in that the distortion changes in a more sophisticated model/reality because the distortion contribution from the front end may well be different even though the loop gain and distortion reduction from the output stage is about the same for both compensation networks.
Moreover, weren't Bob's simulations done with different compensation component values for TPC and TMC, giving different loop crossover frequencies and/or phase margins in the loop gain around the output stage?
Last edited:
Good points, Pete!
We also should take into account the effect during switch-on and switch-off. At lower supply voltages, the trans-conduction of one or more stages is lower. With an already poor phase margin of TPC, this might provoke instabilities.
edit: anything that affects the phase margin should be considered bad practice.
I've compared many amplifier configuration with TPC and TMC (equal TPC/TMC caps and resistors) and in all cases TMC gives lower distortion. Sometimes only marginal, sometimes very distinct.
Here are the results from a blameless amp, C1=120pF, C2 = 560pF and R = 470R
TPC: THD20 = 35.7ppm, BW=200kHz; THD20 = 72.7ppm, BW=500kHz
TMC: THD20 = 31.5ppm, BW=200kHz; THD20 = 38.7ppm, BW=500kHz
And why is TMC clearly better? Because the VAS has less work to do. At 100kHz (for example) the AC collector current of the VAS is 20dB lower!
Cheers,
E.
Last edited:
This is untrue.
With TPC C1 should be larger than C2, but the difference in the audio band as far as loading the second stage is concerned is negligible.
Not at all..
A blameless style amp is favorable to TPC compensation,
as the current mirrored ltp has about 60db gain, while
the vas account for another 60db.
Therefore, the gain loss due to vas loading is small,
but not negligible, 6db typicaly at 20khz, increasing to 10db
at 60khz...
With a high gain vas, in the 80/90db range, vas loading
become important and lead to as much as 15db lower OLG
largely including the audio band...
Hi Mike,
When saying that C1 should be larger than C2 for TPC, are you referring to C1 as being the capacitor that is connected to the input of the VAS?
In your opinion, what is the optimum ratio of C1 to C2 for TPC?
Thanks,
Bob
Okay, no problem. So let's make C1 > C2: C1=560pF, C2 = 120pF and R = 470R
TPC: THD20 = 49ppm, BW=200kHz; THD20 = 82ppm, BW=500kHz
Are you happy now?
Last edited:
Compensation
Hello Edmond, Bob, Wahab, a,d all the others too ...
Since I am reading Bob's book (congrats Bob!!) I understood to simulate also other (not only output) nodes. So the first interesting node for me was the feedback input of the IPS.
While the output node shows a nice rise curve w/o overshoot (simple miller compensation) at 60V/us, the feedback input for the IPS acts a bit weird to me.
It rises very steeply. Much more steep than the output node. Almost like the input node which is set to have a rise time of 1ns. (remark : there is no input filter cap for HF roll-off). Around halfway this rising it suddenly drops again before going back up but this time following the output node curve.
The delay between the moment the signal falls again and the moment where it starts to follow the output rising is approx. 60ns... so it is very short.
What is this?
I will try however to see what this represents when I insert a HF rolloff Cap at the input.
Here is a link to the graphs :
http://www.diyaudio.com/forums/soli...ogy-construction-troubles-32.html#post2411951
Greetz,
Olivier
Hello Edmond, Bob, Wahab, a,d all the others too ...
Since I am reading Bob's book (congrats Bob!!) I understood to simulate also other (not only output) nodes. So the first interesting node for me was the feedback input of the IPS.
While the output node shows a nice rise curve w/o overshoot (simple miller compensation) at 60V/us, the feedback input for the IPS acts a bit weird to me.
It rises very steeply. Much more steep than the output node. Almost like the input node which is set to have a rise time of 1ns. (remark : there is no input filter cap for HF roll-off). Around halfway this rising it suddenly drops again before going back up but this time following the output node curve.
The delay between the moment the signal falls again and the moment where it starts to follow the output rising is approx. 60ns... so it is very short.
What is this?
I will try however to see what this represents when I insert a HF rolloff Cap at the input.
Here is a link to the graphs :
http://www.diyaudio.com/forums/soli...ogy-construction-troubles-32.html#post2411951
Greetz,
Olivier
Hi Olivier,
I'm glad you're enjoying my book!
I think that what you are seeing is the transfer of the sharp-rising input signal to the other side of the input differential pair (where feedback is applied) as a result of the base-emitter capacitances of the turned-on differential pair transistors. If you choos a slower input simulation risetime than 1ns, this effect should become much smaller.
Cheers,
Bob
Hi Bob,
Same minds, same thoughts. see: http://www.diyaudio.com/forums/soli...ogy-construction-troubles-32.html#post2412815.
What do you think about my latest sims, post 1169 and 1174?
Cheers,
E.
Same minds, same thoughts.
see: http://www.diyaudio.com/forums/soli...ogy-construction-troubles-32.html#post2412815.What do you think about my latest sims, post 1169 and 1174?
Cheers,
E.
Yes.
There is no such thing as an optimum ratio. The only requirement is that the series combination of C1 and C2 should equal the value of the capacitor you would use for single pole compensation.
Howdy,
a whimsical, unpredictable dynamic transfer function makes any static analysis meaningless, including dependable analyses. One way to in real terms reduce distortion is to reduce the amplification factor, thus the susceptibility to overdrive, proceeding with keeping the number of stages at a minimum. This gives a badly needed increase in bandwidth, less phase errors, less need for compensation and generally a better sound. No kind of compensation network, diode clamp and other trickery will accomplish that; will only make things worse.
a whimsical, unpredictable dynamic transfer function makes any static analysis meaningless, including dependable analyses. One way to in real terms reduce distortion is to reduce the amplification factor, thus the susceptibility to overdrive, proceeding with keeping the number of stages at a minimum. This gives a badly needed increase in bandwidth, less phase errors, less need for compensation and generally a better sound. No kind of compensation network, diode clamp and other trickery will accomplish that; will only make things worse.