Odd amp

sounds kind of like class-D, but not quite right.

what's the content of the PWM?

--
     ?

I don't know. Without further info I see 2 possibilities:

1. The PWM is indeed a class D output waveform - but if so, why mix the analog in with it? Doesn't make sense.
2. The PWM is not, in which case I don't see how the result would amplify.

I'll see if I can find more info later.

NT

Add the analog signal to a fast triangle, feed into a fixed- threshold comparator to get PWM and then low-pass, maybe?

Jeroen Belleman

Sounds like you want pulse-density modulation, e.g. delta-sigma. PDM generally doesn't encode specific amplitude values in the width of each pulse like PWM (special case of PDM)

That is to say in a strict PWM waveform the cycle period is fixed and the pulse width is proportional to the instantaneous signal amplitude, while with PDM the pulse width is proportional to the instantaneous input signal frequency and it's the average of the PDM signal over a full cycle that corresponds to the amplitude (however long a cycle may be.)

It sounds like splitting hairs maybe for many applications but there is a difference here, for example a DC signal of 1 volt forever is a perfectly cromulent PDM signal too while it wouldn't be a PWM signal with any meaning.

Given a linear adder and a linear lowpass filter, that won't work. The two signals won't interact.

If there's a comparator or equivalent after the adder, that can.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

This works well. The analog signal is added to a PWM before driving the hydraulic servo valve. The PWM serves to add dither to overcome the valve stiction, and adjusting the PWM mark/space offsets the signal as required. The low pass function (dither removal) is provided by the valve/actuator inertia.

The analog input is amplified - a low power into the servo valve can produce a much higher output power.

Cheers

--
Clive

Is it possibly square waves to the rails and analog gain to the transistors ?

Sounds like something Crown might do.

It sounds like the equivalent of "rocking" a car to bust it free of a snowdrift but if the actuator inertia is large enough to filter the PWM to DC I'm not seeing how the PWM signal simultaneously overcomes the valve stiction.

Couldn't you use an initial burst of low frequency large-amplitude pulses to bust it out of its deadband and then revert to high-frequency PWM to reduce power loss in the drive amp once it's up and moving?

quite. I vaguely wondered whther it might then go through diodes but that would only work for smallish signals.

Maybe that's what's done. I clearly don't have the full story. I failed to find it again.

NT

Just found it, and it turns out to be not what I suspected:

NT

Oh, a classic PWM multiplier. Before uPs, people made very precise analog multipliers, dividers, and square-rooters with that technique.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

l with a PWM signal, then low pass. Amplified analogue is supposed to come out, but I don't get how it would. Anyone?

e

at would only work for smallish signals.

to find it again.

Built one of those at Kent Instruments in 1974. 0.1% accurate, but slow.

Used to multiply essentially DC process control signals, with the chopping going on at a kHz or so.

Don't know if it ever went into production. We used saturating bipolar tran sistors as our choppers - two in series with the second one inverted - and CMOS did a much better job when it hit the market a year or so later, when I used it to make a 0.1% D/A converter, again relying on PWM. That did go i nto production - there was at least one at DRAX - but it was never going to be a high volume product.

--
Bill Sloman, Sydney

These days PWM chopper switch signals can be up to 1MHz. Anybody wants to experiment with this idea, ask for one of my RIS-778 unpopulated PCBs, see schematic, pics at

Idea was to evaluate a super-low-noise programmable-gain circuit. Hah, made PCB a year ago, but didn't find time to build one. Email address to snipped-for-privacy@yahoo.com

--
 Thanks, 
    - Win

Hi

The adding and the low-pass seem to be in the wrong order.

There are (or used to be) RF amplifier designs out there, where a main output stage with a high power but also high distortion was combined (in a sense, like paralleled) at the output with a lower power stage that had more bandwidth and higher fidelity. The lower power stage was driven with a correction signal, which, when amplified and added to the output of the main power stage, would result in a signal with a considerably lower total distortion.

The smaller and faster additional stage was thus providing a difference signal that was needed to "subtract out" the distortion from of the output of the main power stage in real time.

I'm not sure if anyone ever used such a thing in production, but it would at least "not be unthinkable" that the same principle be applied to a (presumably big and powerful but noisy and slow) PWM output stage of a Class-D amplifier together with a (presumably lower power but quieter and high speed) Class-AB to provide a fast correction signal.

The resulting combined amplifier could have the high power and most of the high efficiency of the Class-D main stage at lower frequencies, (where power is typically needed the most) and the high bandwidth of the Class-AB auxiliary stage at higher frequencies (where output power requirements are typically lower but the main stage can't provide BW).

Additionally, if the auxiliary stage is used in a closed loop, and if it has the needed dV/dt capability and drive strength, it could even correct the switching noise out of the PWM stage's output, making the whole thing both efficient and quiet. This would however require the auxiliary stage to have enough bandwidth to compensate the PWM as well as its strongest harmonics, which is orders of magnitude higher than the actual output signal. For example, an "audio" type Class-D stage running at 150 kHz PWM could be combined with a Class-AB stage that has 1.5 MHz analog bandwidth, which is enough to correct out any remaining PWM as well as its harmonics up to the 9th. The combined device could have an impressive combination of power efficiency, signal bandwidth and ripple.

For this whole scheme to work, however, the addition of the signals needs to be done after the low-pass filtering rather than before it.

Also in certain topologies the Class-AB auxiliary stage may need to be configured as a current source (high impedance) rather than a voltage source (low impedance), but this depends strongly on how exactly the combiner circuitry is designed. If the correction stage can drive the "GND end" of the filter capacitors of the PWM stage directly, it would need to be of a very low impedance instead.

Regards Dimitrij

e.g. "Iterative error take-off":

I think there are a lot of assumptions there about the performance of the amplifiers individually required to make such that you get all the benefits. It sounds too good to be true, so I'm a bit skeptical that a scheme like the one in the link I posted would have all the claimed benefits.

The reason that it's not often seen in practice is that the improvement you actually see in reality doesn't justify the design time and expense of a second (or third) independent error-correcting amp. Passive filters remove PWM ripple pretty well. Large open-loop gains on a single amp can solve problems via a single negative feedback loop pretty well, too.

Some of these ideas seem to come from the assumption that two amplifiers with lower amounts of global NFB working together must somehow intrinsically produce a better performing amplifier than a single amp with a higher amount. Or that there are certain magical types of distortion that NFB is unable to correct for. I don't see much evidence this is actually the case

How exactly would you combine the 2 outputs? Putting one ouput on the ground out line means it takes the full current of the bigger amp. Injecting current instead of v onto the + out line has 2 issues:

1. Speakers aren't steady impedance, so you'll only get reduction & cancellation some of the time, sometimes it'll make it worse
2. By changing what goes down the nfb path of the big amp you make its distortion worse.

NT

Related:

Note that it can't be done at frequency, because the summing node of each stage has a lagging phase shift. In the context of RF amps, it's feedforward, and no, it's not the easiest transfer function to solve for!

For the class D amp, the improvement might be useful, if maybe not practical for audio purposes. For lab purposes, let's say? Where also for whatever reason, the high efficiency is required?

You wouldn't get a strong reduction in harmonics, but those would be handled by the filter. You'd design the output filter to have some damping at f_c (so its performance is reasonably independent of speaker impedance), an inductive input impedance above f_c (which restricts the curve to Butterworth or looser), and a high enough order to get adequate attenuation at f_sw and/or harmonics.

The correction amplifier could simply be wired in series; the current capacity isn't a big deal, and the low voltage can be supplied from an isolated winding.

Alternately, an output transformer could be used (how quaint!), since the class D amp will be quite good by itself at DC, and only AC correction is really required. (Speaking of, a 2nd order control loop could be used, so that the class D amp has good accuracy at DC and first order AC (DC error of "zero", ramp error proportional to rate). I'm not sure how feasible that is with the output filter included.)

Anyway, the correction amp would be fed the input signal, after passing a filter equivalent to the output filter (it could be a delay network that merely approximates it, or an op-amp implementation of the same response). This does two things: one, it limits high frequency content of the input signal, reducing the chance that the correction amp will be called upon to do far more work than it can; and two, it matches the response of the class D amp and filter, except for errors due to its finite loop gain. A little feedback on this, and you address the gain droop, phase shift and distortion of the class D stage, and do a modest part to the output ripple. (Ripple could be better compensated by feeding some of it forward from the inverter, out of phase, so it can be nulled by design or adjustment.)

I think this assumption is used a lot in the audiophile community... but they're quite fond of making assumptions!

Tim

--
Seven Transistor Labs, LLC 
Electrical Engineering Consultation and Contract Design 
Website: https://www.seventransistorlabs.com/