Hi TheeAldeen,
Thank you for your kind words.
How it will perform sonically, I can't tell you yet, as I have to build this amp first, of course. And that will take a lot of time (month!), as it's rather complicated and contains a huge amount of components.
Not to disappoint you, but I hope this amp will sound just neutral. That might even mean it doesn't sound spectacular. The point is that a good amp, IMHO, has no 'sound' at all.
Cheers,
E.
Thank you for your kind words.
How it will perform sonically, I can't tell you yet, as I have to build this amp first, of course. And that will take a lot of time (month!), as it's rather complicated and contains a huge amount of components.
Not to disappoint you, but I hope this amp will sound just neutral. That might even mean it doesn't sound spectacular. The point is that a good amp, IMHO, has no 'sound' at all.
Cheers,
E.
Edmond, these resistors can be used to increase open loop bandwidth alot. Open loop gaan wil be decreased by these also what makes the freq compensation easier.
If you know other options to reduce the VAS gain, I am very interested.
Ben
you seem to have wandered into the wrong thread
we really don’t “know” why one should want to do either at the expense of reducing open loop gain at all audio frequencies - which is what the loading R does - without even the possible advantage of forming a additional local feedback loop that could "use" the gain to linearize some other aspect of the amp
most here actually seem to know enough EE to evaluate Otala's prescription against the subsequent analysis, measurements by Cordell, et al
the idea really is as "dead" as it gets - initiated by a flawed assumption, demonstrated in theory and Hardware to be incorrect
initially "converts", Walt Jung, Marshall Leach both published "retractions" - subsequent analysis showing high loop gain and low PIM do not require Otala's "flat loop gain"
we really don’t “know” why one should want to do either at the expense of reducing open loop gain at all audio frequencies - which is what the loading R does - without even the possible advantage of forming a additional local feedback loop that could "use" the gain to linearize some other aspect of the amp
most here actually seem to know enough EE to evaluate Otala's prescription against the subsequent analysis, measurements by Cordell, et al
the idea really is as "dead" as it gets - initiated by a flawed assumption, demonstrated in theory and Hardware to be incorrect
initially "converts", Walt Jung, Marshall Leach both published "retractions" - subsequent analysis showing high loop gain and low PIM do not require Otala's "flat loop gain"
Edmond, the image is a thumbnail, if you click on it it will zoom out. I wish the forums had a better image resizing algorithm.
Guys, chill out. I was using Symasym as an example. I know about the 47k resistors, PSRR and all that jazz. Let's not get myopically critical here.
And no, the 100p cap is not just doing nothing. It is a critical part of the stability compensation. It is equivalent to shunt compensation.
Please Edmond, check out the 100p cap, it is not just there for looks, it responds to the voltage at the current mirror input (the frontend rail filters fix PSRR), which itself responds to the current loading of the output stage. The 330pF caps augment this response with voltage loading to make it stable at HF. The result of this compensation is a 2nd-order capacitive input stage loading, like two-pole compensation.
MikeB was very clever, but he didn't design his amps for the lowest numbers, and it seemed people just did not want to understand him. I think it's strange how no one recognizes the Symasym compensation scheme.
To disambiguate, I am just trying to discuss compensation schemes and their performance, I am not interestested in the "sound" right now, and if I was I would discuss it in another thread.
The 47k resistors are inconsequential, they do not affect the concept of the compensation scheme. Let's get on track.
Here is a schematic of the Symasym 4, the first version, which not be so confusing. If it looks corrupted, click on it to see if it is a thumbnail. The later implementation with the 100p cap was possibly made because the voltage at the current mirror input (one heavily degenerated diode) was more linear than the voltage across the second differential bases (two non-degenerated diode junctions). Although this does not redeem the performance in other aspects.
Guys, chill out. I was using Symasym as an example. I know about the 47k resistors, PSRR and all that jazz. Let's not get myopically critical here.
And no, the 100p cap is not just doing nothing. It is a critical part of the stability compensation. It is equivalent to shunt compensation.
Please Edmond, check out the 100p cap, it is not just there for looks, it responds to the voltage at the current mirror input (the frontend rail filters fix PSRR), which itself responds to the current loading of the output stage. The 330pF caps augment this response with voltage loading to make it stable at HF. The result of this compensation is a 2nd-order capacitive input stage loading, like two-pole compensation.
MikeB was very clever, but he didn't design his amps for the lowest numbers, and it seemed people just did not want to understand him. I think it's strange how no one recognizes the Symasym compensation scheme.
To disambiguate, I am just trying to discuss compensation schemes and their performance, I am not interestested in the "sound" right now, and if I was I would discuss it in another thread.
The 47k resistors are inconsequential, they do not affect the concept of the compensation scheme. Let's get on track.
Here is a schematic of the Symasym 4, the first version, which not be so confusing. If it looks corrupted, click on it to see if it is a thumbnail. The later implementation with the 100p cap was possibly made because the voltage at the current mirror input (one heavily degenerated diode) was more linear than the voltage across the second differential bases (two non-degenerated diode junctions). Although this does not redeem the performance in other aspects.
Last edited:
TSC
Hi Ben,
You are perfectly right, one can 'increase' the open loop bandwidth (defined by the -3dB corner frequency) by those stupid resistors, but nothing is gained by this utter nonsense (sorry for my harsh words).
>If you know other options to reduce the VAS gain, I am very interested.
Sure, a short circuit to ground, very effective!
@JCX: Thanks for your explanation.
@Keane: Don't worry, I do know it was just an example and I do like to discuss with you the possibilities of TSC.
PS: Symasym 4, i.e. 330pF + 100R, makes more sense to me.
Cheers,
E.
Hi Ben,
You are perfectly right, one can 'increase' the open loop bandwidth (defined by the -3dB corner frequency) by those stupid resistors, but nothing is gained by this utter nonsense (sorry for my harsh words).
>If you know other options to reduce the VAS gain, I am very interested.
Sure, a short circuit to ground, very effective!
@JCX: Thanks for your explanation.
@Keane: Don't worry, I do know it was just an example and I do like to discuss with you the possibilities of TSC.
PS: Symasym 4, i.e. 330pF + 100R, makes more sense to me.
Cheers,
E.
Last edited:
The 330p+100R in the Symasym 4 and the 100p cap in Symasym 5 are equivalent as far as compensation goes. That much makes sense from a design standpoint.
I had a really good response typed out, but almost at the end I accidentally pressed the CTRL key instead of shift, and deleted the tab. This always happens, without fail, every time I write something long worth posting. Too bad Firefox does not remember content from deleted tabs - what's the point of recovering them at all then?
Bensen, try a resistor from the C-B of the VAS. This reduces gain without loading the VAS so much (loads the frontend instead).
Actually, local feedback at the expense of local stage gain, decreases the distortion of the stage itself, but only when the input signal magnitude remains the same. If an outer feedback loop is used, it must produce a larger input signal to make up for the lost gain. So there doesn't seem to be an point to local feedback at individual stages, unless local gain is conserved. This seems be be the logic behind condemning Dado's C-B resistor a the VAS, but it seems this does not somehow apply to miller compensation as well?
Furthermore, the VAS sees basically a zero source impedance because of the feedback loop, so any current flowing through the miller cap can't cause local feedback at audio frequencies can it?
On to TSC. First, let's talk about shunt and Miller compensation. Where stability is concerned, in an amp which must be stable into any load AND cannot have any output network, shunt compensation will usually be more stable with higher PSRR, but will distort more (and is too large to put on a chip - which explains it's relative obscurity). If we instead decide from the beginning to use a passive output network to eliminate load-dependent stability issues, this allows us to lower internal compensations greatly and we end up with a simpler design with less distortion. This is my chosen method.
I think the significant difference between shunt and miller compensation is an impedance matching issue. Shunt and Miller compensation serve dual purposes. Shunt compensation responds to current loading, whereas Miller compensation responds to voltage loading. Because there can be various I/V conversions in any amplifier, local shunt compensation can perform the function of global miller compensation, and vice-versa, which is evidently a point of confusion over the Symasym 5 amplifier.
My contention is this:
TSC in combination with an RC across the VAS input (for a common-emitter VAS) or PLIL is a possible alternative to TMC and TPC. It combines the low VAS loading of TMC with the low frontend loading of TPC, yet it does not have the PSRR reduction of either (in the case of a common-emitter VAS), and it does not excessively conduct voltage/switching distortion from the output stage. Furthermore it changes the configuration of parasitics around the VAS and the input of the OPS, I am not sure but I think for the better. TSC on my bench has been a breeze to troubleshoot.
One critical difference between TMC and this compensation type is that it responds to the VAS current load, rather than it's voltage load, and this affects it's application. For instance in a double EF where the VAS current load is high, TSC+shunt will distort more, but in a triple EF it will distort less.
You can imagine what I'm thinking by taking the Symasym schematic, with a triple EF output stage, and replacing the 330p caps with TSC. The Super TIS is a bit different, but in my simulation it works with PLIL and TSC (in this case PLIL performs the same function as the RC across the VAS input - are they the same thing? I can't find the papers on PLIL).
Shunt compensation does not necessarily need a capacitor to ground at the VAS output. If the shunt compensation can be made extreme enough, it will work with the parasitic elements of the output stage. While this results in the lowest VAS loading, it results in disproportionately large compensation values, which magnify the output stage distortion. The TSC at the output stage input therefore assists the compensation by providing a large enough load at stability frequencies to allow the shunt compensation values to be minimized. There is a point at which distortion conducted through the TSC becomes larger than that conducted through the first shunt compensation, so in terms of distortion there is a balance between these two compensations. At the same time, the optimal values for distortion are not necessarily the best values for bandwidth. The TSC can instead be optimized for highest BW, in which case it effectively becomes a compensation for the OPS input impedance, and the preceding shunt compensation is resized to attain the desired stability margins. In Symasym a bare capacitor is used instead of TSC; using TSC here gives us the option of increasing the capacitor and decreasing the preceding shunt compensation as it's distortion contribution decreases significantly.
Does this compensation deserve a name? Could it be called TSASC (Transitional Shunt Assisted Shunt Compensation) or TSAPLIL (Transitional Shunt Assisted PLIL)? Is there not some better way to organize and refer to compensation techniques? There must be some mathematical way to calculate the max number of unique compensation types for a given amplifier configuration, and give them all unique, sensible names.
Whew. I hope that made some sense.
I had a really good response typed out, but almost at the end I accidentally pressed the CTRL key instead of shift, and deleted the tab. This always happens, without fail, every time I write something long worth posting. Too bad Firefox does not remember content from deleted tabs - what's the point of recovering them at all then?
Bensen, try a resistor from the C-B of the VAS. This reduces gain without loading the VAS so much (loads the frontend instead).
Actually, local feedback at the expense of local stage gain, decreases the distortion of the stage itself, but only when the input signal magnitude remains the same. If an outer feedback loop is used, it must produce a larger input signal to make up for the lost gain. So there doesn't seem to be an point to local feedback at individual stages, unless local gain is conserved. This seems be be the logic behind condemning Dado's C-B resistor a the VAS, but it seems this does not somehow apply to miller compensation as well?
Furthermore, the VAS sees basically a zero source impedance because of the feedback loop, so any current flowing through the miller cap can't cause local feedback at audio frequencies can it?
On to TSC. First, let's talk about shunt and Miller compensation. Where stability is concerned, in an amp which must be stable into any load AND cannot have any output network, shunt compensation will usually be more stable with higher PSRR, but will distort more (and is too large to put on a chip - which explains it's relative obscurity). If we instead decide from the beginning to use a passive output network to eliminate load-dependent stability issues, this allows us to lower internal compensations greatly and we end up with a simpler design with less distortion. This is my chosen method.
I think the significant difference between shunt and miller compensation is an impedance matching issue. Shunt and Miller compensation serve dual purposes. Shunt compensation responds to current loading, whereas Miller compensation responds to voltage loading. Because there can be various I/V conversions in any amplifier, local shunt compensation can perform the function of global miller compensation, and vice-versa, which is evidently a point of confusion over the Symasym 5 amplifier.
My contention is this:
TSC in combination with an RC across the VAS input (for a common-emitter VAS) or PLIL is a possible alternative to TMC and TPC. It combines the low VAS loading of TMC with the low frontend loading of TPC, yet it does not have the PSRR reduction of either (in the case of a common-emitter VAS), and it does not excessively conduct voltage/switching distortion from the output stage. Furthermore it changes the configuration of parasitics around the VAS and the input of the OPS, I am not sure but I think for the better. TSC on my bench has been a breeze to troubleshoot.
One critical difference between TMC and this compensation type is that it responds to the VAS current load, rather than it's voltage load, and this affects it's application. For instance in a double EF where the VAS current load is high, TSC+shunt will distort more, but in a triple EF it will distort less.
You can imagine what I'm thinking by taking the Symasym schematic, with a triple EF output stage, and replacing the 330p caps with TSC. The Super TIS is a bit different, but in my simulation it works with PLIL and TSC (in this case PLIL performs the same function as the RC across the VAS input - are they the same thing? I can't find the papers on PLIL).
Shunt compensation does not necessarily need a capacitor to ground at the VAS output. If the shunt compensation can be made extreme enough, it will work with the parasitic elements of the output stage. While this results in the lowest VAS loading, it results in disproportionately large compensation values, which magnify the output stage distortion. The TSC at the output stage input therefore assists the compensation by providing a large enough load at stability frequencies to allow the shunt compensation values to be minimized. There is a point at which distortion conducted through the TSC becomes larger than that conducted through the first shunt compensation, so in terms of distortion there is a balance between these two compensations. At the same time, the optimal values for distortion are not necessarily the best values for bandwidth. The TSC can instead be optimized for highest BW, in which case it effectively becomes a compensation for the OPS input impedance, and the preceding shunt compensation is resized to attain the desired stability margins. In Symasym a bare capacitor is used instead of TSC; using TSC here gives us the option of increasing the capacitor and decreasing the preceding shunt compensation as it's distortion contribution decreases significantly.
Does this compensation deserve a name? Could it be called TSASC (Transitional Shunt Assisted Shunt Compensation) or TSAPLIL (Transitional Shunt Assisted PLIL)? Is there not some better way to organize and refer to compensation techniques? There must be some mathematical way to calculate the max number of unique compensation types for a given amplifier configuration, and give them all unique, sensible names.
Whew. I hope that made some sense.
a
To follow up on the subject, I have acquired a copy of Lurie's second book. Much more readable than his first, thank you for the recommendation.
Helpful treatment of nested loops and Bode optimization that looks applicable to audio power amplifiers.
Surely I can't be the only person who has seen this.
Is there any published work on this or an amp that uses it?
Best wishes
David
Last edited:
100pF
Hi Keane ,
I have to admit that I completely overlooked the subtle but essential interaction between the degen. resistors of the CM and compensation capacitors. Indeed, very clever. Probably, I was too distracted by the 47k resistors to recognize the importance of this subtlety.
In the mean time I've simmed the circuit, which confirmed the importance of the degen. resistors. If zero, the 100pF cap has virtual no effect and the amp becomes far less stable.
I also simmed the TSC compensation, but the reduction in distortion wasn't spectacular, most likely because the compensation was already of the 2nd order. This leaves little room for further improvements.
Cheers,
E.
Hi Keane ,
I have to admit that I completely overlooked the subtle but essential interaction between the degen. resistors of the CM and compensation capacitors. Indeed, very clever. Probably, I was too distracted by the 47k resistors to recognize the importance of this subtlety.
In the mean time I've simmed the circuit, which confirmed the importance of the degen. resistors. If zero, the 100pF cap has virtual no effect and the amp becomes far less stable.
I also simmed the TSC compensation, but the reduction in distortion wasn't spectacular, most likely because the compensation was already of the 2nd order. This leaves little room for further improvements.
Cheers,
E.
Hello Edmond
Did you ever try a sim of "Not TMC" That is, still a resistor and and capacitor in parallel back to the Miller capacitor but both tied to the same point? It would lack the transitional benefits but it would not be Miller compensation because of the extra time constant and it would be educational to compare it.
Best wishes
David
Did you ever try a sim of "Not TMC" That is, still a resistor and and capacitor in parallel back to the Miller capacitor but both tied to the same point? It would lack the transitional benefits but it would not be Miller compensation because of the extra time constant and it would be educational to compare it.
Best wishes
David
It is possible to define each different compensation scheme as an array of transfer functions from node to node. So for n nodes there are n**2 functions, each of which can be arbitrarily complex. So this identifies them all but is hardly friendly or convenient. If we assume that a typical audio amp has it's sub circuits in a cascade then I think it's a bit simpler but still not useful.
Ed Cherry tackles NDFL this way in some JAES articles but it's hard work to read. Well, for me anyway.
Even some of the Control Theory gurus complain that the matrix formulation is practically opaque.
If anyone has a better idea I'd love to read it!
Best Wishes
David Zan
In Fig. 1c of your website put the 1k resistor in parallel with C2.
So for TPC it's tied to earth.
For TMC it's tied to OUT (of the OPS).
For comparison it's tied to the "VAS" output.
So, not transitional but still has the two time constants of TMC.
Is that clearer?
Best wishes
David
So you think that localyl feedbacked VAS is condemned.
Here are three simulations:
1.TMC pure
2.TMC with feedback resistor and bridged with small capacitor
3.TMC bridged with small capacitor
In the case of pure TMC phase drops to 177 deg. and bridged to 165 deg. I think that this gives better stability.
dado
Attachments
I was meaning in terms of distortion, not stability. And, it was more of a question than a statement. And I think you miss my point entirely.
If gain ever drops enough for stability at those frequencies to matter, then I think stability will be the least of your worries. Ideally some sort of protection scheme will engage before that point.
If gain ever drops enough for stability at those frequencies to matter, then I think stability will be the least of your worries. Ideally some sort of protection scheme will engage before that point.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.