Exactly so; this is the penalty, which has to weighed against the advantage.
E.g. in my own transistor case the input requirement was 0,6Vrms into 5K1, which I supplied from a unity-gain pair of Zout of 130E at well below audible distortion. But it is naturally included in the whole equation - almost part of the power amp.
E.g. in my own transistor case the input requirement was 0,6Vrms into 5K1, which I supplied from a unity-gain pair of Zout of 130E at well below audible distortion. But it is naturally included in the whole equation - almost part of the power amp.
Johan Potgieter said:
Right;
each particular case is different. No optimal system may be built optimizing each component of it. The sound track represents equation with multiple variables, and the art is to find the optimal combination of component that results in the optimal solution of the whole system. To simplify combinations and re-combinations there were standards developed for studio equipment that specify impedances and signal levels, but in case of home audio where we have more flexibility they are not optimal.
input impedance distortion
Some late night cogitating:
Several of the commonly used input subtraction techniques for feedback affect the input impedance seen by the signal source. These usually cause the input Z to increase due to the tracking voltage difference across the differencing device(s) as mentioned above. This is fine as long as it is consistant.
But the actual distortion components do not track the input signal and so will modulate the input Z and thereby require error currents from the input signal. These will corrupt the reference signal if not derived from a much lower impedance.
The output difference of a high loop gain amplifier's input stage is almost exclusively the error difference signal. At least high loop gain reduces the difference error down to a very small amount. But low loop gain, or low global feedback, designs would be susceptible to this distortion mechanism.
This has probably been written about in journals under the heading "interface distortion", but I haven't heard much about it mentioned on the input side. Mainly just regarding the speaker interface. If anyone has some references on this, please let us know.
Using a Fet or Vacuum tube differential input design in non-inverting mode to get high input Z doesn't fix this. The comparison is still being made between the first source/cathode follower output Z and the feedback Fet source or cathode follower Z, so impedances are comparable at the tail differencing point. So still allows error differences to mistrack the Zs and corrupt the reference. (Of course thats how the diff. stage actually works to get current differencing at its output. So not much can be done I guess. If the impedance changes were linear for both elements, then there wouldn't be the odd harmonic distortion generally seen. )
Even the inverting mode differencing approach with resistors is susceptible, since the - input is still moving around with the error signal. Only the linear parts cancel to a (near) null at the - input.
Of course, more loop gain reduces the error magnitude to give a better voltage null.
Maybe what is needed is a hi Z current drive from the input signal side, then error voltages won't matter.
Don
Some late night cogitating:
Several of the commonly used input subtraction techniques for feedback affect the input impedance seen by the signal source. These usually cause the input Z to increase due to the tracking voltage difference across the differencing device(s) as mentioned above. This is fine as long as it is consistant.
But the actual distortion components do not track the input signal and so will modulate the input Z and thereby require error currents from the input signal. These will corrupt the reference signal if not derived from a much lower impedance.
The output difference of a high loop gain amplifier's input stage is almost exclusively the error difference signal. At least high loop gain reduces the difference error down to a very small amount. But low loop gain, or low global feedback, designs would be susceptible to this distortion mechanism.
This has probably been written about in journals under the heading "interface distortion", but I haven't heard much about it mentioned on the input side. Mainly just regarding the speaker interface. If anyone has some references on this, please let us know.
Using a Fet or Vacuum tube differential input design in non-inverting mode to get high input Z doesn't fix this. The comparison is still being made between the first source/cathode follower output Z and the feedback Fet source or cathode follower Z, so impedances are comparable at the tail differencing point. So still allows error differences to mistrack the Zs and corrupt the reference. (Of course thats how the diff. stage actually works to get current differencing at its output. So not much can be done I guess. If the impedance changes were linear for both elements, then there wouldn't be the odd harmonic distortion generally seen. )
Even the inverting mode differencing approach with resistors is susceptible, since the - input is still moving around with the error signal. Only the linear parts cancel to a (near) null at the - input.
Of course, more loop gain reduces the error magnitude to give a better voltage null.
Maybe what is needed is a hi Z current drive from the input signal side, then error voltages won't matter.
Don
I don't remember who in the middle of the previous sentury first raised the question of distorted feedback, there was a belief that small amount of feedback does not help, instead the entire feedback signal is getting distorted adding more distortions of distortions, so instead of just the 2'nd harmonic higher order harmonics are produced. I believe it was wrong interpretation of other facts such as phase shifts that prevented cancellation of nonlinearities (lot of reactive elements), and additional nonlinearities in the feedback loop that differs from nonlinearities in the main signal path (cathode VS grid, for example).
On the two last posts:
Yes, the input source impedance has an impact on the distortion. There have been some articles (can't remember exactly from whom, I thought Walt Jung among others) explaining it. Some modern new opamps have such a low distortion that this effect becomes dominant, and their spec sheets say that you should keep the source impedance below 600 ohms to get the spec'ed performance.
Peter Baxandall and other british authors have documented the relation between feedback factor and distortion. What they found was that generally speaking the distortion initially increases if you go from zero to low feedback, and that with increasing feedback the distortion RMS level decreases but you get more harmonics. In fact, to decrease higher harmonics needs more and more feedback the higher the harmonic.
One other thing to remember is that the total amp may have a gain largely determined by the feedback, the amplifying element itself will ALWAYS work open loop! That is the reason for the small effective input voltage. And now it also is clear that if the open loop gain is linear, the effective input voltage is more linear and the effects of input Z are less.
Jan Didden
Yes, the input source impedance has an impact on the distortion. There have been some articles (can't remember exactly from whom, I thought Walt Jung among others) explaining it. Some modern new opamps have such a low distortion that this effect becomes dominant, and their spec sheets say that you should keep the source impedance below 600 ohms to get the spec'ed performance.
Peter Baxandall and other british authors have documented the relation between feedback factor and distortion. What they found was that generally speaking the distortion initially increases if you go from zero to low feedback, and that with increasing feedback the distortion RMS level decreases but you get more harmonics. In fact, to decrease higher harmonics needs more and more feedback the higher the harmonic.
One other thing to remember is that the total amp may have a gain largely determined by the feedback, the amplifying element itself will ALWAYS work open loop! That is the reason for the small effective input voltage. And now it also is clear that if the open loop gain is linear, the effective input voltage is more linear and the effects of input Z are less.
Jan Didden
janneman said:
It would be interesting to find, in which case. No doubt, feedback was distorted. Phase shifts and/or different transfer characteristics for input and feedback points.
janneman said:
Ok, now let's review couple of voltage amplifyer stages.
Suppose, we want to add a local feedback to the second stage.
To minimize effective voltage swing on it's grid let's add a parallel feedback by voltage like it was suggested in one post (it was called "partial feedback"). It means as well, that the first stage will see lower resistance. If it is a triode that may be roughly viewed as a voltage controlled resistor, such trick forces it to produce higher second harmonic. Also, its gain will be less so more input swing needs on its grid to get the same output swing, that means more distortions. Trying to linearize the 2'nd cascade we increased overall distortions.
Now, let's add a series feedback by current to the same 2'nd cascade instead of a parallel one by voltage, increasing resistance in it's cathode tail. As the result, the 1'st triode will see higher resistance load, so it's distortions will be less. Also, it's gain will be higher, so less input swing is needed for the same voltage output of overall amplifier. However, it will have higher output resistance as the result of the feedback by current, but it is the different story.
Re: input impedance distortion
Very important, Don.
I do not have references but believe that the priciples of interface distortion can be applied here in as much as this input impedance forms the load for a preceeding stage just as a loudspeaker forms it for a power output stage.
A first approximation can be that everything that feeds the input can be characterised as a distortionless generator followed by a resistance (the source impedance). It is this impedance that gets "modulated", thus it is important how this (hopefully low) impedance is arrived at. The problem arises if this is not simply the internal resistance of the feeding stage, but an impedance arrived at from global feedback. Then interface distortion is right back in the picture.
(Many will realise that this is one of the major reasons why a low impedance reached by NFB is to be carefully handled. The often quoted high damping factor of power amplifiers, when realised mainly by NFB, can be a time bomb source of distortion. I am sure the defining article years ago by Otala and Lamasniemi on interface distotion is known.)
regards.
smoking-amp said:
Very important, Don.
I do not have references but believe that the priciples of interface distortion can be applied here in as much as this input impedance forms the load for a preceeding stage just as a loudspeaker forms it for a power output stage.
A first approximation can be that everything that feeds the input can be characterised as a distortionless generator followed by a resistance (the source impedance). It is this impedance that gets "modulated", thus it is important how this (hopefully low) impedance is arrived at. The problem arises if this is not simply the internal resistance of the feeding stage, but an impedance arrived at from global feedback. Then interface distortion is right back in the picture.
(Many will realise that this is one of the major reasons why a low impedance reached by NFB is to be carefully handled. The often quoted high damping factor of power amplifiers, when realised mainly by NFB, can be a time bomb source of distortion. I am sure the defining article years ago by Otala and Lamasniemi on interface distotion is known.)
regards.
I am going to venture posting some spectrum diagrams here (following post), although they do not show distortion/NFB magnitude as promised earlier. Rather they show the often neglected influence of certain basic circuit elements on distortion.
Fig. 1 shows a basic power amplifier circuit simply to illustrate the three resistances (impedances) under discussion here. R1 and R2 are the usual sub-ohm compensation resistors in the power emitters, while Rdrive is the output impedance of T3. (T3 should be seen as a stage with adjustable output impedance, not necessarily a single transistor.)
The amplifier used is also not this circuit. It was a 70W amplifier using 27 dB of NFB operated at about 50W. The rise between 20KHz and 25KHz should be ignored - it was an artifact of the analyser programme. The input signal was 1 KHz, and that as well as the 2nd and 3rd harmonic products are off scale to facilitate presenting the higher order products; they were not exorbitant.
Figures are numerated rather backwards! Fig. 5 shows the spectrum of the typical circuit often encountered. The row of odd order harmonics would be noted - they extend beyond 40KHz.
Figs. 2 and 4 show the separate influence of zero R1, R2, and lowering R drive respectively (note the different vertical scale). Fig. 3 shows the reult of these measures combined. This is noticably better than fig. 5, especially concerning higher order harmonics.
Making R1, R2 zero could be regarded as theoretical. Alternative temperature compensation can however be effected by thermal feedback between power transistors and the T3 current source.
If the programme obliges the graphics should appear in the next post.
Regards
Fig. 1 shows a basic power amplifier circuit simply to illustrate the three resistances (impedances) under discussion here. R1 and R2 are the usual sub-ohm compensation resistors in the power emitters, while Rdrive is the output impedance of T3. (T3 should be seen as a stage with adjustable output impedance, not necessarily a single transistor.)
The amplifier used is also not this circuit. It was a 70W amplifier using 27 dB of NFB operated at about 50W. The rise between 20KHz and 25KHz should be ignored - it was an artifact of the analyser programme. The input signal was 1 KHz, and that as well as the 2nd and 3rd harmonic products are off scale to facilitate presenting the higher order products; they were not exorbitant.
Figures are numerated rather backwards! Fig. 5 shows the spectrum of the typical circuit often encountered. The row of odd order harmonics would be noted - they extend beyond 40KHz.
Figs. 2 and 4 show the separate influence of zero R1, R2, and lowering R drive respectively (note the different vertical scale). Fig. 3 shows the reult of these measures combined. This is noticably better than fig. 5, especially concerning higher order harmonics.
Making R1, R2 zero could be regarded as theoretical. Alternative temperature compensation can however be effected by thermal feedback between power transistors and the T3 current source.
If the programme obliges the graphics should appear in the next post.
Regards
Sorry; I should perhaps mention:
The results were not influenced by a significant change in loop gain as a result of changes in the T3-stage to give the different Rdrives. That was adjusted. The global NFB of about 27dB was maintained.
The results were not influenced by a significant change in loop gain as a result of changes in the T3-stage to give the different Rdrives. That was adjusted. The global NFB of about 27dB was maintained.
Hello Johan,
Interesting graphs. Are these from a real speaker load or a fixed resistor load?
On the idea of displaying increasing higher harmonic distortion (although lower THD) with increasing NFB, it has been commented somewhere that as the device characteristic curvature is straightened out by increasing NFB, it is described by a power law math. model with power coefficient dropping toward unity but never quite getting there.
A coefficient of say 1.1 being the same as the fraction 11/10, this being equivalent to the 10th root of the 11th power, so no real surprise that higher harmonics are generated.
I recall an article in EW+WW a few years back where it was proposed to build an amplifier using square law Mosfets which would only generate 2nd harmonics, then they cancelled the 2nd harmonics with a push-pull design. Of course, real Mosfets aren't quite perfect square law unfortunately.
regards,
Don
Interesting graphs. Are these from a real speaker load or a fixed resistor load?
On the idea of displaying increasing higher harmonic distortion (although lower THD) with increasing NFB, it has been commented somewhere that as the device characteristic curvature is straightened out by increasing NFB, it is described by a power law math. model with power coefficient dropping toward unity but never quite getting there.
A coefficient of say 1.1 being the same as the fraction 11/10, this being equivalent to the 10th root of the 11th power, so no real surprise that higher harmonics are generated.
I recall an article in EW+WW a few years back where it was proposed to build an amplifier using square law Mosfets which would only generate 2nd harmonics, then they cancelled the 2nd harmonics with a push-pull design. Of course, real Mosfets aren't quite perfect square law unfortunately.
regards,
Don
Don,
Resistor load - these are actually Spice graphs, because I could load the graphics here more easily. But I did check with a simulated loudspeaker load - no mentionable difference. The load was not severe - 8 ohm system going to 16 ohm main resonance and down to 6 ohm; phase shift staying between +/- 15 degrees (equalised). Actual amplifier tests with spectrum analyser into resistive load showed confirmation, but I could not photograph those.
Your further explanation makes sense. Mathematics (a series) confirm that NFB increases high order harmonic products, e.g 3rd generated from 3rd, etc. I did not do such analysis, but it would seem that such products remain very low in amplitude, mostly in noise floor. It would appear to be harmonic multiplication but at very low amplitude. Products can go on over 100KHz, but at negligible amplitude, and so will any intermodulation products also be. This is of course only the case if the open-loop design itself is relatively clean to begin with - old principle.
.... and yes, unfortunately real life products are never that obliging!
Resistor load - these are actually Spice graphs, because I could load the graphics here more easily. But I did check with a simulated loudspeaker load - no mentionable difference. The load was not severe - 8 ohm system going to 16 ohm main resonance and down to 6 ohm; phase shift staying between +/- 15 degrees (equalised). Actual amplifier tests with spectrum analyser into resistive load showed confirmation, but I could not photograph those.
Your further explanation makes sense. Mathematics (a series) confirm that NFB increases high order harmonic products, e.g 3rd generated from 3rd, etc. I did not do such analysis, but it would seem that such products remain very low in amplitude, mostly in noise floor. It would appear to be harmonic multiplication but at very low amplitude. Products can go on over 100KHz, but at negligible amplitude, and so will any intermodulation products also be. This is of course only the case if the open-loop design itself is relatively clean to begin with - old principle.
.... and yes, unfortunately real life products are never that obliging!
- Status
- Not open for further replies.