"Class-B" Audio Power Amp with "Current Pumping"

[AlexCaledin]

TL;DR Summary

- this seems to be a new way to cancel the Crossover Distortion in "Class-B" complementary transistor amplifier -

Hello everyone, I was thinking how to cancel the class B crossover distortion, may I suggest to discuss an interesting variant.
Let's start from this:

- there is crossover distortion (and no current protection).
The signal source is "full voltage, zero current".
Let's make it "full current, zero voltage" :

Now the crossover distortion is cancelled, but the 5A source is too hard to make.
Let's try an input current transformer.
It must be without stray inductance and capacitance, to be equivalent to single coil with current:

In reality, adding a big input coil to the coil shown above causes stray HF resonance and great headache when developing the current source for high dV/dt and applying the general feedback to the whole amp.

But - how about moving the speaker to the right?

Now , the source current is 51 times less than output - so the source is quite possible to build even with the full voltage across it.
The voltage magnetizing the coil core is the follower error, practically only the crossover distortion, so the core non-linearity hardly matters.

My practical MicroCap simulation model seems working good, the current source is made of 4 Mosfets (2 in the differential input, 2 amplifying the current); the open-loop gain and overall feedback stability seem as good as in any classic amplifier.

The problem is very old MicroCap version I am using ; so if someone joins with a modern simulator, with modern components, the improvement may be very valuable.

 

[eq1]

I got a bit hung up by the schematics you posted so I admit I haven't gone through the whole post.

Specifically starting with the voltage follower circuit in the first image: The single-cell batteries connected across the gates of the power FETs are to pre-bias them a bit and put some space in voltage across the gates. Given that space and the fact that the output is wired to the negative input of the LT1028, I don't think this circuit has any meaningful amount of crossover distortion. It is not clear how the more complicated circuits would reduce it further as the LT1028 itself is the limiting factor.

Assuming you're using this definition of cross-over distortion: https://en.wikipedia.org/wiki/Crossover_distortion

 

[AlexCaledin]

- those batteries are MicroCap symbols for voltage sources, they are 2V and 4.2V for IRF530 and IRF9540 Mosfets. Yes, the opamp does reduce the crossover distortion - at the lowest frequencies, up to million times - but the remaining distortion is proportional to the frequency, as the opamp gain falls. So the distortion may be microvolts at 30Hz and millivolt at 10kHz. But with that current-pump coil, it can be kept well under microvolt, whatever the frequency is, within audible range.

It works simple: 1/51 of the follower distortion error is applied across the short right part of the "pump" inductor creating the distortion current that passes through the speaker. So that current can be kept negligible by sufficient coil inductance.
The coil wire resistance is important, too, allowing the DC control.

 

[gneill]

The problem is very old MicroCap version I am using ; so if someone joins with a modern simulator, with modern components, the improvement may be very valuable.

LTSpice is a modern product and is free to download. It's provided by Linear Technology, now part of Analog Devices. They use it in-house and make it available for free, presumably because the they can provide their product lines in the built-in components, so it's both extremely useful and a marketing tool of sorts. It's regularly maintained with automatic updates being available.

So it's a free design tool for their stuff, but of course you can always install other components if you have the appropriate spice descriptor files.

 

[Tom.G]

Conceptually, isn't that tapped inductor just an additional gain stage for the amplifier? With the effective gain increased by a factor of 50, of course there will be lower closed-loop distortion.

Clever approach though!

Cheers,
Tom

 

[AlexCaledin]

Conceptually, isn't that tapped inductor just an additional gain stage for the amplifier? With the effective gain increased by a factor of 50, of course there will be lower closed-loop distortion.

- well, yes, the open-loop gain is the product of the current source transconductance, that factor of 50 and the speaker resistance, like a tube amp, but that transconductance is waay greater than any tube - so what is to be emphasized is that two big-gain loops are hitched together, the "overall" one and that inside the voltage follower.
Speaking of tubes, a good big one may well be used here, as that current source)

 

[eq1]

- well, yes, the open-loop gain is the product of the current source transconductance, that factor of 50 and the speaker resistance, like a tube amp

I am still pretty confused. cross-over distortion is a property of the DC transfer function.

I think what you're looking to do is improve the THD of the speaker amplifier. For example: if the input was a pure 10KHz sine the output should also be a pure 10KHz sine, i.e. no additional harmonics, a.k.a THD as low as possible.

I don't get how the coil increases the open-loop gain. I didn't analyze it but here is a quick thought experiment. Pick any sine with a low enough frequency to make the coil a short. Because the coil is a short the opamp sees zero error (P=N) so it's basically uncontrolled, or there is no current gain. I am not sure which without doing the math.

If you're worried about THD at audible frequencies I think the answer is to get an opamp with more GBW. In 2019 those are readily available. :)

 

[AlexCaledin]

Pick any sine with a low enough frequency to make the coil a short. Because the coil is a short the opamp sees zero error (P=N) so it's basically uncontrolled, or there is no current gain.
. . .
I think the answer is to get an opamp with more GBW.

- although it's the physics forums here, the coils are not necessarily superconductive!
In my model, the resistance of the coil wire is ( 5 + 0.1 ) Ohm which is enough because the currents are sufficient. (The same wire must be used for the left and right parts.) So, at DC, the currents in the coil are kept at the same ratio of 50, if the opamp DC offset is not too big.

Well, the opamp choice was restricted by the high capacitance of those powerful MOSFETs, and high DC gain required, and good supply voltage, and low input current, and high dV/dt, not to mention the old age of my simulator library.

 

[eq1]

In my model, the resistance of the coil wire is ( 5 + 0.1 ) Ohm which is enough because the currents are sufficient. (The same wire must be used for the left and right parts.) So, at DC, the currents in the coil are kept at the same ratio of 50, if the opamp DC offset is not too big.

Now I get it. If there is a real part to the coil that's near in magnitude to the speaker impedance then it's a current mirror. The current in the right coil is mirrored by the left and the ratio is controlled by where the speaker taps the coil.

Ignore phase delay and amp stability, which is probably ok as the amp is probably much faster than the coil then:

Vi=voltage @ current source red dot, Vo=power fet node, Vs=voltage @ speaker input
ZR=impedance of right coil, ZL=impedance of left coil

KCL@speaker => (Vi-Vs)/ZL+(Vo-Vs)/Zr-Vs/Rs=0 => ((Vi-Vs)/ZL)((ZL+ZR)/ZR)=Vs/Rs => Io/Isrc = (ZR+ZL)/ZR

Io/Isrc gives the current gain of the amp and it's a function of the left and right impedances which shows how it moves with the tap point. The KCL equation shows pretty clearly how the mirror is working.

You picked ZR=1 and ZL=50 hence current gain = 51 which matches the simulation. Assuming the coil is well designed (i.e. they lose equivalent L with B@5A) it should hold well over the audio band.

But does it have less THD? I can't make up my mind. I think it does because the coil makes an LR filter with the speaker which should suppress harmonics but the thing that's really throwing me is the speaker is common mode to the amp. But I guess that's always the case in all the schematics given. Also, I think higher frequencies will have to slew harder in voltage at the amp output due to the coil which could lead to distortion when the amp tries to follow. For sure phase margin is worse but it very likely had margin to spare so meh. Would probably have to do a proper analysis over frequency to know for sure.

Definitely an interesting design. Thanks for sharing.

 

[AlexCaledin]

... the coil makes an LR filter ... higher frequencies will have to slew harder in voltage at the amp output due to the coil ...

- the thing is, the coil is kept at "zero" voltage - at the audible frequencies; - and at megahertz frequencies it makes a divider with the traditional several-microHenri inductor placed before the speaker in series. The AC analysis shows the phase margin of 80 degrees, at 2 MHz, where the open loop gain gets under 0 dB slightly slower than -20dB/ decade, the phase shift is -280 degrees at 2 MHz.

As for distortion, to me, it consists of crossover pulses rather than frequencies. So , that coil is placed in the way of quick crossover pulses - to reject the quick-transient error of the follower - but still transparent for the correct current. It can perhaps be called follower-assisted current transformer.

 

[trurle]

- although it's the physics forums here, the coils are not necessarily superconductive!
In my model, the resistance of the coil wire is ( 5 + 0.1 ) Ohm which is enough because the currents are sufficient. (The same wire must be used for the left and right parts.) So, at DC, the currents in the coil are kept at the same ratio of 50, if the opamp DC offset is not too big.

Well, the opamp choice was restricted by the high capacitance of those powerful MOSFETs, and high DC gain required, and good supply voltage, and low input current, and high dV/dt, not to mention the old age of my simulator library.

I think you will have problems with input 0.1A driver. What point of having low-noise (1nV/sqrt(Hz))LT1028 as main amplifier if you need to drive it with something as noisy as 150mA-capable LT1010 (20 nV/sqrt(Hz))?

 

[AlexCaledin]

I think you will have problems with input 0.1A driver. What point of having low-noise (1nV/sqrt(Hz))LT1028 as main amplifier if you need to drive it with something as noisy as 150mA-capable LT1010 (20 nV/sqrt(Hz))?

Here is it, that driver, it makes up to 0.2 A in my simulation.

No point of several amperes those IRF MOSFETs are capable of - but their capacitance seems perfect forming the dominant pole. (The 2 Ohm resistor is connected to the voltage clipping diodes. The diodes on the picture are current limiting. )

 

[AlexCaledin]

- the thing is, the coil is kept at "zero" voltage

- I am sorry, it ought to be , The voltage across the coil is kept at "zero" by the precision follower. The whole coil is at the full output potential of course.

 

[Svein]

Applying a large amount of feedback around an amplifier is prone to introduce TIM (Transient Intermodulation Distortion, see Matti Otala's original paper here or Google it.

 

[AlexCaledin]

Applying a large amount of feedback around an amplifier . . .

- yes, that is not necessary since the crossover distortion is suppressed "locally" in the output part.
So I am working with this lower voltage variant where the overall feedback is not stronger than what is necessary to keep the output resistance not too great - similar to a tube amp feedback:

- both opamps are employed to make simple voltage followers - which are arranged to magnify the current , 100 and 20 times (in the first follower, the BJTs are working in "Class-A" mode) - so the resulting transconductance is 20 A/V which makes the output resistance about 1/4 Ohm if K=5.

 

[Svein]

Hm. I would personally have made changes - such as introducing emitter resistors in Q2 and Q3 (about 33ohms). The same goes for M3 and M4 (about 0.1ohm). I am also at a loss trying to understand the connections to K1.

BTW - TIM is not the same as crossover distortion. The usual way to minimize TIM is to use local feedback in the stages - like for example using the emittter resistors / source resistors I mentioned.

 

[AlexCaledin]

Hm...emitter resistors in Q2 and Q3 ... M3 and M4
...
the connections to K1
...
to minimize TIM is to use local feedback in the stages...

Those resistors decrease the maximum voltage achievable with 12V supply, 10 or more Amperes maximum current for "1 Ohm" speaker system. Q2 and Q3 are BD135 and BD136, their emitter properties seem very special with the resistance already more than usual, and they are operating without crossover here, Class A mode, 0.6A maximum current. Still, yes, the problem of thermal stabilization may require some emitter resistors even though each extra Ohm subtracts 0.6V from the output maximum.

The transformer is connected in order to implement that same tapped coil that is imposing the right current on the imperfect output follower. Only , in this low voltage case, the turns ratio is 20 instead of 50, because the power dissipated by the current driver still seems tolerable.

The opamps may well be LT1122 which seem to be practically free from intermodulation, even at unity-gain, see the datasheet. They are never overloaded according to simulation. So the feedbacks are local except the "general" feedback which can be "loosened" by increasing the source resistors at the input.

 

[Svein]

Take a look at this 100W amplifier (http://electronics-diy.com/electronic_schematic.php?id=440):

 

[AlexCaledin]

Ohh that looks great of course, with those special MOSFETs. I have been already warned , to never try to outperform such big amps which are Class AB mode of the output. But it's interesting to try making a simple low voltage amplifier despite the fact that the crossover problem is especially challenging when the speaker resistance is low. It could be powered from two old car batteries, with a simplest low power converter making 18V for the opamps. Then the components for repairing it would be always available in our local shops.

 

[Svein]

What I meant by "take a look at" is all the standard building blocks of a high performance amplifier:

• Differential pair on the inpout
• "Cascode" connection in the input stage
• "Constant current" generators
• Current mirrors
• "Current feedback" in all stages

 

[Rive]

The voltage magnetizing the coil core is the follower error, practically only the crossover distortion, so the core non-linearity hardly matters.

Well, I've just started to look into the depths of this design, but what I would worry about is not the nonlinearity, but the hysteresis there. The core seems to change polarity exactly around the point where you want to kill the crossover distortion.
Just first thought: I may be wrong.

 

[AlexCaledin]

... worry about ... the hysteresis

The idea is to keep the distortion current low, when the follower error - applied across the coil - increases proportional to the frequency. The ideal inductor current does not increase. Now the hysteresis adds its special current of same phase as voltage, consuming energy according to the hysteresis loop. If the frequency is 2 times higher , then there are 2 times more hysteresis cycles - of the same loop size, because the error voltage , though 2 times greater, has half time to magnetize . So the hysteresis current does not increase, right? The power loss remains proportional to the voltage.
Then, the hysteresis current remains always less than the proper inductor current - which is itself caused by the error voltage ; so , what the hysteresis ( as well as non-linearity ) is doing is simply distorting the negligible error a little more.

 

[Rive]

...so , what the hysteresis ( as well as non-linearity ) is doing is simply distorting the negligible error a little more.

I'm not really familiar with the low intensity magnetics, but it's in the feedback loop, so what it does (I think) is exactly about adding a false error signal into the feedback - so it get amplified.
I think LtSpice has some ways to simulate coils with core hysteresis: I would try to check it anyway.

 

[AlexCaledin]

adding a false error signal into the feedback - so it get amplified. .

Look at a simple tube amplifier, such as in an old radio - there's a very similar feedback with a good iron transformer within it - and the whole output audio signal is magnetizing the iron; and still many people say it sounds better than usual modern amplifiers.

 

[Tom.G]

Look at a simple tube amplifier

Speculation!

I think LtSpice has some ways to simulate coils with core hysteresis: I would try to check it anyway.

Simulate it, prove it!

 

[AlexCaledin]

Let me make this point once more: my feedback loop is not to be suspicious about, because it's the same thing as used in the good old tube amplifiers. In either case, the overall feedback is applied to the transconductance of 10...20 A/V in order to stabilize - no more than necessary! - the actual voltage gain (and of course to improve linearity) .

 

[Svein]

still many people say it sounds better than usual modern amplifiers.

Those "many people" are usually guitar players who do not care about distortion (they usually add distortion to the guitar output anyway),.