Has it been tryed to compensate the non linearity of a class B output stage adding an opposite non linearity made of passive components.
This targeted at gm doubling and beta drooping in BJT output stages.
I guess this is difficult, asking for a very stable bias and very stable components. So I wonder whether there exist something worthwhile about this technique.
This targeted at gm doubling and beta drooping in BJT output stages.
I guess this is difficult, asking for a very stable bias and very stable components. So I wonder whether there exist something worthwhile about this technique.
Indeed, NFB is the usual way to linearize.
This is not what I am speaking about.
I am talking about linearization with passive components which excludes NFB an active technique.
I look for linearization done on the open loop system, before looping to apply NFB.
So forget about NFB in this topic.
This is not what I am speaking about.
I am talking about linearization with passive components which excludes NFB an active technique.
I look for linearization done on the open loop system, before looping to apply NFB.
So forget about NFB in this topic.
This would only work if you can accurately predict the errors to be corrected. This is generally not possible, not the least because they vary with operating conditions, temperature, signal, etc. You cannot predect it to a sufficient accuracy that it gives results better than can be obtained otherwise.
Jan
Two questions:
1. Has the OP heard of something called feedforward? It is used when NFB cannot be used, such as some RF systems.
2. Is a diode a passive component? If not, how does he propose to get nonlinear response from passive components (which are almost all fairly linear)?
1. Has the OP heard of something called feedforward? It is used when NFB cannot be used, such as some RF systems.
2. Is a diode a passive component? If not, how does he propose to get nonlinear response from passive components (which are almost all fairly linear)?
A diode is a passive component. Indeed, I had diodes in mind.
May be the name of what I am looking for is "feed forward".
As a start I am ploting the DC transfert of a typical ouput stage.
It has a slope near 1 that varies with the load at the output, and a different slope at zero crossing.
Not much surprise so far, but no idea at the moment how to correct this to hopefully get a constant slope.
May be the name of what I am looking for is "feed forward".
As a start I am ploting the DC transfert of a typical ouput stage.
It has a slope near 1 that varies with the load at the output, and a different slope at zero crossing.
Not much surprise so far, but no idea at the moment how to correct this to hopefully get a constant slope.
Last edited:
Feed test signal through your OPS, and then provide all needed correction in digital domain before your DAC.
For example - MiniDSP Dirac:
https://www.minidsp.com/products/dirac-series
For example - MiniDSP Dirac:
https://www.minidsp.com/products/dirac-series
Try to use current source to drive a class b push pull output stage. One of the example is output stage inclusive miller compensation.
What do you mean in "output stage inclusive Miller compensation" ?
The usual place for Miller compensation is not in the output stage, it is ahead.
The usual place for Miller compensation is not in the output stage, it is ahead.
You may be interested in Ian Hegglun's 'squareA' or 'cubeA' amplifier topology. He uses complementary characteristics of output devices to obtain a linear transfer from the combination. He has a few threads here.
Jan
As I said, feedforward is generally used when a designer would really like to use NFB but this is not possible. It is inferior to NFB, as it cannot automatically track any changes in the output stage (e.g. due to bias shift arising from signal or temperature). It suffers the same problem as NFB: reducing low order distortion at the cost of generating high order distortion.
Look up Dr Hawksford's "Amplified Diode" circuit. It is geared towards the use of Darlington BJT output stages but Bob Cordell also used the concept with vertical mosfets, with their very non-linear Gm and capacitances when used as push pull class AB. One major advantage that I like is that this scheme can be placed within the global loop as 'nested' feedback thus feeding back a 'linearized' signal to the global loop, since we all know that global feedback alone is very poor at 'linerizing' a non-linear transfer.🙄 Also the global loop is not offended by the non-linear transients created by driving a speaker(s), which is no longer reflected onto the VAS.🙂
I had tried, nested feedback, I know how difficult it is to achieve stability.
I will look again in this way.
The main issue is the phase lag of the inner loop that is required for the outer loop stability.
I will look again in this way.
The main issue is the phase lag of the inner loop that is required for the outer loop stability.
As has been mentioned, you cannot compensate for non-linearity with linear components, but it can be done with active components
It is not particularly easy, but it can be done, and I am talking about true error correction, not some kind of cleverly obfuscated NFB scheme: all of the so-called "error-correction" schemes are in fact using some kind of NFB in disguise.
This scheme doesn't use NFB: rather, it is a "feed-backwards" system, but certainly not negative, it is in fact positive:
http://www.diyaudio.com/forums/solid-state/185501-unigabuf-follower-cut-out-leader.html
The circuit looks rather complicated, but its core, its "engine" is the Q7/Q5 combo.
Q5 is the actual follower stage, and Q7 is the error correcting transistor: the current through Q7 is sampled, scaled and mirrored by Q4/Q6 and sent to Q7.
Because of the exponential nature of I-V relationship in a junction, the variation caused to the Vbe of Q5 by the variation in output current is exactly mirrored in Q7 with an opposite sign, even though the absolute level of current may be widly different.
This means that the circuit makes Vin=Vout, whatever:
the dynamic resistance of the output transistor is cancelled together with the non-linearities it brings.
Keantoken has reused this technique in his error-correcting buffer.
Of course, a power amplifier cannot be single-ended (at least preferably not), it should be class B or AB.
You could in principle associate two complementary such circuits, but this leads to thorny issues: both the positive and the negative sides will try to impose their version of a perfect Vin to the output, which could work in an ideal world, but not in the real world: there will always be small offsets, differences, meaning the two halves, each having a zero output impedance will spend their energy battle one another instead of providing a useful output.
This means that some kind of arbitration is needed, which is doable, but complicates things significantly.
Here, rather than a complementary circuit, the single-ended circuit is associated with two power slaves providing the large output currents; but the master always stays in control, and the whole circuit doesn't require or use NFB to achieve its good performances.
There are real-world adjustments and tradeoffs, because the low current PNP cannot exactly compensate for the high current NPN, etc., but basically it works as advertised.
Note that the input impedance of this circuit is negative... it should therefore be driven from a low impedance source, or compensated by a parallel positive resistance
With LTspice on a class aB CFP output stage, I see a DC gain a bit lower than 1 in the B region and a gain a bit higher than 1 in the narrow A region.
The width of the A region depends of the bias.
I know how to make a very stable bias.
These two gains vary with the output load ( and I presume with Bjt current gains) .
This I have no control, can only design for a specific nominal load like 8 ohm which is resistive load far from an actual 8 ohm speaker.
I do not understand why gm doubling induces a gain a bit higher than 1.
The width of the A region depends of the bias.
I know how to make a very stable bias.
These two gains vary with the output load ( and I presume with Bjt current gains) .
This I have no control, can only design for a specific nominal load like 8 ohm which is resistive load far from an actual 8 ohm speaker.
I do not understand why gm doubling induces a gain a bit higher than 1.
Oups,
Forget about gain seen over 1.
This was from an error of my own about a DC offset I had overseen.
With a correct calculation, taking care of the offset, I have a gain in the A region lower than 1 and higher than the gain in the B regions.
So, gains are as expected in a Class AB output stage.
Now, how could I compensate these two gains for a constant gain.
I am thinking of resistor switching.
Forget about gain seen over 1.
This was from an error of my own about a DC offset I had overseen.
With a correct calculation, taking care of the offset, I have a gain in the A region lower than 1 and higher than the gain in the B regions.
So, gains are as expected in a Class AB output stage.
Now, how could I compensate these two gains for a constant gain.
I am thinking of resistor switching.
Generally, bias for class AB is a compromise and is very difficult to achieve a perfect unity gain. However, the feedback/feedforward amplified diode circuit nullifies the majority of non-linear Gm created by the low bias of the output transistors. In the A bias region the Gm of both output devices sum, but is only one or the other at B bias levels. Gm is generally lower at low current (sum of the two in A bias) and this is indicated by graphing the Gm vs Ic (Id).
- Status
- Not open for further replies.