Re: Re: Re: Sr
It also confirms that Low Capacitance VAS transistors are mandatory for an LTP with Re degeneration.
Is there any more to hear?
if that is all there is to it, then it seems obvious to conclude that zero to tiny Cdom and high values of LTP Re are all that is required to get best use of this effect.Edmond Stuart said:
It also confirms that Low Capacitance VAS transistors are mandatory for an LTP with Re degeneration.
Is there any more to hear?
Re: Re: Sr
Cc is a constant/linear capacitor. It does not create distortions by itself.
In a Cdom configuration the product between loop gain and frequency is constant and equal to the unity gain frequency (here, 1MHz). Following, as long as the unity gain frequency is constant, the amount of loop gain useful to reduce distortions is constant as well. Therefore, any variation in the THD is due to the base amp distortions and that is because of driving the input stage in non-linearities while loading Cdom (a larger Cdom requires more current). Call it "loading effect", "slew effect" or "LTP linearity", it's the same story.
Adding degeneration (re1) can be viewed as a local NFB loop, linearizing the LTP. In the time domain, this LTP linearizing maps to an increase of SR.
AndrewT said:
Cc is a constant/linear capacitor. It does not create distortions by itself.
In a Cdom configuration the product between loop gain and frequency is constant and equal to the unity gain frequency (here, 1MHz). Following, as long as the unity gain frequency is constant, the amount of loop gain useful to reduce distortions is constant as well. Therefore, any variation in the THD is due to the base amp distortions and that is because of driving the input stage in non-linearities while loading Cdom (a larger Cdom requires more current). Call it "loading effect", "slew effect" or "LTP linearity", it's the same story.
Adding degeneration (re1) can be viewed as a local NFB loop, linearizing the LTP. In the time domain, this LTP linearizing maps to an increase of SR.
Re: Re: Re: Re: Sr
The problem is not really the value of Cob (it's usually much smaller than Cdom anyway) but the fact that Cob is strongly nonlinear. The VAS, being a gain stage, has a rather high ac voltage excursion across Cob, therefore the need to keep Cob as small as possible to minimize nonlinearities and hence THD.
AndrewT said:
The problem is not really the value of Cob (it's usually much smaller than Cdom anyway) but the fact that Cob is strongly nonlinear. The VAS, being a gain stage, has a rather high ac voltage excursion across Cob, therefore the need to keep Cob as small as possible to minimize nonlinearities and hence THD.
thanks Syn,
your two posts are confirming the technicalities behind the often ignored advice to minimise the capacitance loading of the LTP.
Edmond,
just to satisfy my curiosity, could you return the Cc to the inverting input, instead of VAS base/LTP collector, and post the results? all three cases as before might give useful background for further discussion.
Sorry, we are miles off topic. Are we ready for a thread split?
If so someone (me?) should compile a list of post numbers for the Mods to do the required separation.
your two posts are confirming the technicalities behind the often ignored advice to minimise the capacitance loading of the LTP.
Edmond,
just to satisfy my curiosity, could you return the Cc to the inverting input, instead of VAS base/LTP collector, and post the results? all three cases as before might give useful background for further discussion.
Sorry, we are miles off topic. Are we ready for a thread split?
If so someone (me?) should compile a list of post numbers for the Mods to do the required separation.
Cdom
Hi Andrew,
If I do that, the distortion becomes, as expected, vanishing small.
Bob Cordell, for example, did that. But it is not without risk, because the combined phase shift of LTP, I-mirror and VAS becomes easily too large and makes the compensation loop unstable.
In my latest design I have also followed this path, but circumvented the additional phase shift by using a CFB input stage and applying 'DTMC' see: C4, C5 and R17
in: http://www.diyaudio.com/forums/showthread.php?postid=1367089#post1367089
With such arrangement, the contribution of the front-end to closed loop distortion at 20kHz can be made as low as 50ppb.
Cheers,
Edmond.
AndrewT said:
Hi Andrew,
If I do that, the distortion becomes, as expected, vanishing small.
Bob Cordell, for example, did that. But it is not without risk, because the combined phase shift of LTP, I-mirror and VAS becomes easily too large and makes the compensation loop unstable.
In my latest design I have also followed this path, but circumvented the additional phase shift by using a CFB input stage and applying 'DTMC' see: C4, C5 and R17
in: http://www.diyaudio.com/forums/showthread.php?postid=1367089#post1367089
With such arrangement, the contribution of the front-end to closed loop distortion at 20kHz can be made as low as 50ppb.
Cheers,
Edmond.
Re: Cdom
Edmond is right, that kind of Miller input feedback, as I call it, is very effective. As I pointed out in my original MOSFET power amplifeir paper, that local loop formed that spans the input stage and the VAS does need to be compensated, as does any but the most trivial feedback loop. It does not impose any more risk than compensation of an ordinary global feedback loop. In my circuit, a single series R-C did the job. In some circuits a very light series RC from the VAS output node to ground is also helpful.
One very big benefit of this circuit is that it encloses the input stage in a feedback loop. Indeed, because small signal fast transistors are involved, that loop can have a gain crossover frequency in the 10-20 MHz neighborhood. The technique allows very high slew rates to be obtained. In my experimental 50W MOSFET amplifier, I obtained over 300V/us.
Cheers,
Bob
Edmond Stuart said:
Edmond is right, that kind of Miller input feedback, as I call it, is very effective. As I pointed out in my original MOSFET power amplifeir paper, that local loop formed that spans the input stage and the VAS does need to be compensated, as does any but the most trivial feedback loop. It does not impose any more risk than compensation of an ordinary global feedback loop. In my circuit, a single series R-C did the job. In some circuits a very light series RC from the VAS output node to ground is also helpful.
One very big benefit of this circuit is that it encloses the input stage in a feedback loop. Indeed, because small signal fast transistors are involved, that loop can have a gain crossover frequency in the 10-20 MHz neighborhood. The technique allows very high slew rates to be obtained. In my experimental 50W MOSFET amplifier, I obtained over 300V/us.
Cheers,
Bob
Splitting hairs.
Hi Bob,
Allow to split some hairs. The input stage (in your EC) is only partly enclosed by the Miller feedback loop (i.e. the inverting half). Your remarks might suggest that such inclusion itself reduces the distortion of the input stage. As a matter of fact, the subtraction unit (LTP) is the only part of any (non-inverting) amplifier that lies inevitably outside a global, Miller or whatever NFB loop. In other words, the inherent distortion of an LTP cannot be reduced by feedback.
In practice however, it is true that the distortion is lower, some times much lower (and the SR higher), but it is solely due to the fact that the input stage is much less stressed, i.e. it has to deliver less current to the next stage.
BTW, this mechanism sharply contrasts with another 'inclusion trick', namely E.M. Cherry's (heavily criticized) inclusion of the output stage into the Miller loop.
Cheers,
Edmond.
Bob Cordell said:
Hi Bob,
Allow to split some hairs. The input stage (in your EC) is only partly enclosed by the Miller feedback loop (i.e. the inverting half). Your remarks might suggest that such inclusion itself reduces the distortion of the input stage. As a matter of fact, the subtraction unit (LTP) is the only part of any (non-inverting) amplifier that lies inevitably outside a global, Miller or whatever NFB loop. In other words, the inherent distortion of an LTP cannot be reduced by feedback.
In practice however, it is true that the distortion is lower, some times much lower (and the SR higher), but it is solely due to the fact that the input stage is much less stressed, i.e. it has to deliver less current to the next stage.
BTW, this mechanism sharply contrasts with another 'inclusion trick', namely E.M. Cherry's (heavily criticized) inclusion of the output stage into the Miller loop.
Cheers,
Edmond.
Just to get back on-topic, here are a few thoughts on the effects of output coils.
All power amplifiers have output inductance whether they employ an output coil or not and whether they employ negative feedback or not. It is just a matter of degree.
The effective output inductance of an amplifier can be inferred from looking at how its high-frequency response changes as a function of load. An amplifier with no output inductance will generally have its high-frequency 3-dB-down frequency be the same for light resistive loads and heavy resistive loads. Indeed, if the damping factor decreases as frequency increases, this tends to correspond to an inductive component in the output impedance.
If we look at the 3-dB high frequency response point with an 8-ohm load, and compare that with the 3-dB down frequency driving a 2-ohm load, we can infer effective output inductance. Put another way, if we look at the gain of the amplifier at 50 kHz with an 8-ohm load and with a 2-ohm load we can also deduce output inductance.
Consider an ideal amplifier with two poles, one at 500 kHz and the other at 250 kHz. The one at 500 kHz could be the closed-loop bandwidth pole of the negative feedback amplifier, or it could be the forward active-circuit bandwidth of an amplifier that does not employ negative feedback. The 250 kHz pole could be the passive input filter pole. These poles do not include the effect of an output coil, if used.
The frequency response of this amplifier will be as follows:
20 kHz: -0.033 dB
50 kHz: -0.21 dB
100 kHz: -0.81 dB
250 kHz: -4.0 dB
500 kHz: -10.0 dB
1 MHz: -19.3 dB
Now assume that the amplifier has a low-frequency damping factor of 100, corresponding to a resistive output impedance of 0.08 ohms. Further assume that the amplifier has a 1 uH output inductor. Amplifier response will be as follows with 8-ohm and 2-ohm loads, respectively:
1 kHz: -0.088 dB -0.345 dB
10 kHz: -0.097 dB -0.358 dB
20 kHz: -0.123 dB -0.396 dB
50 kHz: -0.307 dB -0.659 dB
As expected, the finite damping factor causes a net drop of 0.257 dB at low frequencies as the load is changed from 8 ohms to 2 ohms.
At 50 kHz, the combined effect of the resistive and inductive output impedance causes a net drop of 0.352 dB at 50 kHz. This is 0.095 dB more than at low frequencies and is suggestive of the output inductance effect. The excess gain loss at 50 kHz due to the presence of 1 uH of output inductance is thus about 0.1 dB.
Note that a 1 uH inductance forms a pole with a 2-ohm load resistance at 318 kHz, and that such a pole creates 0.106 dB of loss at 50 kHz. This is just a sanity check.
If a 0.01 uF capacitor is connected across the load, these frequency response results up to 50 kHz are changed by less than 0.01 dB.
If the inclusion of a 1 uH coil in the output of an amplifier makes an audible difference, it is very hard to ascribe it to an alteration in frequency response. If there is an audible difference, it would seem to be due to some other factor, like magnetic egress or ingress effects that might be linear or nonlinear.
Cheers,
Bob
All power amplifiers have output inductance whether they employ an output coil or not and whether they employ negative feedback or not. It is just a matter of degree.
The effective output inductance of an amplifier can be inferred from looking at how its high-frequency response changes as a function of load. An amplifier with no output inductance will generally have its high-frequency 3-dB-down frequency be the same for light resistive loads and heavy resistive loads. Indeed, if the damping factor decreases as frequency increases, this tends to correspond to an inductive component in the output impedance.
If we look at the 3-dB high frequency response point with an 8-ohm load, and compare that with the 3-dB down frequency driving a 2-ohm load, we can infer effective output inductance. Put another way, if we look at the gain of the amplifier at 50 kHz with an 8-ohm load and with a 2-ohm load we can also deduce output inductance.
Consider an ideal amplifier with two poles, one at 500 kHz and the other at 250 kHz. The one at 500 kHz could be the closed-loop bandwidth pole of the negative feedback amplifier, or it could be the forward active-circuit bandwidth of an amplifier that does not employ negative feedback. The 250 kHz pole could be the passive input filter pole. These poles do not include the effect of an output coil, if used.
The frequency response of this amplifier will be as follows:
20 kHz: -0.033 dB
50 kHz: -0.21 dB
100 kHz: -0.81 dB
250 kHz: -4.0 dB
500 kHz: -10.0 dB
1 MHz: -19.3 dB
Now assume that the amplifier has a low-frequency damping factor of 100, corresponding to a resistive output impedance of 0.08 ohms. Further assume that the amplifier has a 1 uH output inductor. Amplifier response will be as follows with 8-ohm and 2-ohm loads, respectively:
1 kHz: -0.088 dB -0.345 dB
10 kHz: -0.097 dB -0.358 dB
20 kHz: -0.123 dB -0.396 dB
50 kHz: -0.307 dB -0.659 dB
As expected, the finite damping factor causes a net drop of 0.257 dB at low frequencies as the load is changed from 8 ohms to 2 ohms.
At 50 kHz, the combined effect of the resistive and inductive output impedance causes a net drop of 0.352 dB at 50 kHz. This is 0.095 dB more than at low frequencies and is suggestive of the output inductance effect. The excess gain loss at 50 kHz due to the presence of 1 uH of output inductance is thus about 0.1 dB.
Note that a 1 uH inductance forms a pole with a 2-ohm load resistance at 318 kHz, and that such a pole creates 0.106 dB of loss at 50 kHz. This is just a sanity check.
If a 0.01 uF capacitor is connected across the load, these frequency response results up to 50 kHz are changed by less than 0.01 dB.
If the inclusion of a 1 uH coil in the output of an amplifier makes an audible difference, it is very hard to ascribe it to an alteration in frequency response. If there is an audible difference, it would seem to be due to some other factor, like magnetic egress or ingress effects that might be linear or nonlinear.
Cheers,
Bob
Re: Splitting hairs.
Edmond,
You are not only splitting hairs, you are trying to split apart an LTP
.
Indeed, the distortion of an LTP due to common-mode effects cannot be reduced by this enclosure of the input stage. And, of course, in a non-inverting amplifier there will be a common-mode signal equal to the input signal.
It is true that the LTP distortion is reduced by the fact that in this design it must drive much less output current because it is not having to drive into a miller compensation capacitor. This is just another way of looking at distortion reduction by negative feedback, and it is called input-referred distortion analysis. This kind of distortion reduction mechanism applies to many stages with feedback around them when analyzed in this way.
It works. It reduces the distortion very substantially. You are indeed splitting hairs. But at the same time you are not technically wrong.
Cheers,
Bob
Edmond Stuart said:
Edmond,
You are not only splitting hairs, you are trying to split apart an LTP
.Indeed, the distortion of an LTP due to common-mode effects cannot be reduced by this enclosure of the input stage. And, of course, in a non-inverting amplifier there will be a common-mode signal equal to the input signal.
It is true that the LTP distortion is reduced by the fact that in this design it must drive much less output current because it is not having to drive into a miller compensation capacitor. This is just another way of looking at distortion reduction by negative feedback, and it is called input-referred distortion analysis. This kind of distortion reduction mechanism applies to many stages with feedback around them when analyzed in this way.
It works. It reduces the distortion very substantially. You are indeed splitting hairs. But at the same time you are not technically wrong
.Cheers,
Bob
Edmond Stuart said:
The first proposal of the L//R + C across the speaker terminal that I'm aware of was by that "idiote" Nevile Thiele.
There are advantages and drawbacks to every network. With decent grounding the shunt cap does indeed provide superior RFI suppression, but at some possible cost of resonances due to load inductances. However such resonances are well dampened in virtually all cases as very few speaker loads come anywhere near an ideal inductor.
With no load connected at all, the input impedance of the network is still reasonably low at the series resonate frequency of the inductor and the shunt capacitor as the Q of the inductor is neutered by the parallel R.
This really is a no-brainer, but any amplifier using such an output network should have a reasonable degree (from a stability POV) of immunity to load impedance variation in the first place.
I have been using the R//L+C network (with 100nF shunt capacitors) for years on simple / low power designs as such with some of the most horrible loads imaginable and I have never had a problem.
Here are some unloaded and loaded squarewave stability tests for one such amplifier using one such output network.
http://www.diyaudio.com/forums/showthread.php?postid=1488531#post1488531
For little historical perspective, the L//R+C (with C up to 150nF) was the reliable defacto standard (since Neville Thiele) in Electronics Australia magazine for all discrete power amps (HiFi / guitar / active speaker / etc) published over a 30+ year period, representing literally 10’s of thousands of hobbyist built units. It is still the standard for every single design I’ve seen published in Silicon Chip Magazine over the last 20 years or so, representing 1000's more amplifiers built by hobbyists.
Such a track record for reliability says something, I think.........
Cheers,
Glen
AndrewT said:
The story of those "idiotes" started here: http://www.diyaudio.com/forums/showthread.php?postid=1558590#post1558590
Cheers,
Edmond.
PS: Using these so called 'scare quotes' might create confusing. In English they are used to draw attention, while in Dutch (for example) they mean 'so called ...', often meant ironically or sarcastically.
Sorry for bringing this thread back to top.
Regardless of whether an output RL network is a good idea or not I have some questions:
In most amplifiers the inductor is wound around the resistor. So the inductor is no longer an air core inductor. Is becomes a resistor core inductor. This must change the behavior of the inductor, doesn't it?
Do you think it is a good idea to have the inductor on the amplifier pcb? I'm concerned about the stray magnetic field coupling into the circuit of the amplifier. Maybe it is better to place the inductor at the speaker terminals?
Thanks.
Lee
Regardless of whether an output RL network is a good idea or not I have some questions:
In most amplifiers the inductor is wound around the resistor. So the inductor is no longer an air core inductor. Is becomes a resistor core inductor.
This must change the behavior of the inductor, doesn't it? Do you think it is a good idea to have the inductor on the amplifier pcb? I'm concerned about the stray magnetic field coupling into the circuit of the amplifier. Maybe it is better to place the inductor at the speaker terminals?
Thanks.
Lee
I started out thinking the whole Thiele Network should be at the speaker terminals.
I have changed my mind.
The first R+C Zobel should be very closely coupled with the output stage, not just on the PCB.
The middle L//R can be in the cable route from PCB to speaker terminals. I agree that the possible effects of the coil on the amp circuit may cause a problem.
The last R+C Zobel should be across the speaker terminals.
I am, as always, open to be educated.
I have changed my mind.
The first R+C Zobel should be very closely coupled with the output stage, not just on the PCB.
The middle L//R can be in the cable route from PCB to speaker terminals. I agree that the possible effects of the coil on the amp circuit may cause a problem.
The last R+C Zobel should be across the speaker terminals.
I am, as always, open to be educated.
Dear,
Interesting Topic. Just read all 35 pages
In practice I can tell, that a RC zobel is a must. The series inductor and resistor can save output-stages. But not all manufactures seems to use them. Rotel leave them out in many designs.
I saw an interesting Zobel in the dutch Sphinx Project 12 amplifier.
RC>RL>RC
With kind regards,
Bas
Interesting Topic. Just read all 35 pages
In practice I can tell, that a RC zobel is a must. The series inductor and resistor can save output-stages. But not all manufactures seems to use them. Rotel leave them out in many designs.
I saw an interesting Zobel in the dutch Sphinx Project 12 amplifier.
RC>RL>RC
With kind regards,
Bas
