Simple PWM tracking supply for class A output stage?

Hi all,

I'm a novice with an idea I'm hoping to get a word of validation on before I sink any resources in to it. As the title says...

Start with an opamp driving a 555 based PWM circuit like this one (555 Delta-Sigma Modulator). Feed the square wave output to a gate driver IC driving a single mosfet. Follow this with a typical LC filter. The output will swing somewhere between ground and the supply rail.

Use this as a tracking power supply to feed a linear output stage, operating in single ended class A. If the filtered PWM output is biased a few volts above the linear output, the linear device will generate little heat, greatly reducing the size/cost of the required heat sink, as well as related energy losses.

I'd like to keep everything as simple and cheap as possible - minimize the parts count and use through hole construction. The power supply will be a laptop brick around 20V. I'd like it all to work with typical 4Ohm - 8Ohm speakers with passive crossovers.

I haven't thought too much about the design of the class A stage yet. I'm leaning towards a follower, but I'll probably end up trying a bunch of options.

Here's my first stab at the active components:

NJM4580D NJR Corporation/NJRC | Integrated Circuits (ICs) | DigiKey

LMC555CN/NOPB Texas Instruments | Integrated Circuits (ICs) | DigiKey

Gate driver
IRS2127PBF Infineon Technologies | Integrated Circuits (ICs) | DigiKey

IRFP250MPBF Infineon Technologies | Discrete Semiconductor Products | DigiKey

Does this look viable? Are there other components I should look at? Is there any hopes of it actually sounding better than just ok? All feed back welcome.
Joined 2007
Paid Member
It a good idea although there may be practical issues. As you alter supply voltage dynamically you will probably cause a shift in the operating conditions of the analogue Class A amp that will show as a 'bounce' in the output as the rail alters.

The other issue is of getting a supply to follow the audio through transients and so on. The supply would just not react quickly enough to follow the transient changes.

I once had an idea of altering the bias of an amp in relation to the expected signal level, in other words the bias was set to a point that related to the volume control setting and so the maximum output that could ever be seen in that condition. The amp would thus remain in Class A at all signal levels.

New idea for low dissipation Class A amplifier.
Not wishing to place too much of a damper on your design dreams but starting out with a 555 to implement what will be a buck convertor suggests your level of experience is going to cause you a lot of grief. If I were to have a play with such a thing I would start off with a ready made Class D amplifier. The forum is stuffed full of peoples experiences with E-Bay chip amplifiers. Consider using one of those supplied from your brick to generate the Class A supply. You would have to interface to its input with something like a full wave rectified, op-amp circuits exist for this, version of the input signal with some DC offset.
Thanks for the responses.


I had actually imagined that the tracking rail would improve the operating conditions of the output device, much as a cascode does. There will of course be some jitter introduces in the form of the audio distortion present in the PWM output. But that output is a relatively constant x-volts above the linear output, the voltage across the linear device would be similarly constant.

Regarding transient response, how is this different from normal class D? Note that I specified a cmos 555, which can run up to 3MHz.

I've pondered using system volume to modulate power, but not quite like you propose. I'm going to noodle on your concept for a while - thanks.


I had considered starting with a turn-key class D for the supply. However, it hadn't dawned on me that I could bias the input to keep the output on one side of the rail - neat trick. One aesthetic element to my approach it would be nice to hold on to is the fact that no one IC does too much - there's a conceptual element of discrete components. The most magical part I've proposed is the humble 555.

Can you elaborate on where you think I'm going to run in to problems? To me, an engineer, hearing why something won't work isn't much different from hearing why it could :)
Joined 2007
Paid Member
The problem I think you would run up against would be (for example) having the output stage running at (say) 10 volts and then having an audible peak come along that requires your full 20 volts supply to be available.

That peak may only last a few milliseconds and I would think in practice that a supply could not follow and rise the required amount in so little time. Your 20 volt laptop supply puts a real limit on things anyway, its not like having an amp running on say 30 volts with a maximum of 70 or volts supply available.

Don't let me put you off though :)

A first step might be to look at a sample of music on a scope as you listen in real time and then to see if you feel a supply could track the peak levels you see.

If you were only testing with different amplitude sines and with a substantial time gap between non abrupt level changes then I could see it could possibly work... but tracking audio millisecond to millisecond.
Sounds like you're describing clipping, not a transient. There are plenty of 100W class D boards that run on 24V. (I'm open to >20V, so long as it's available in a brick.)

Assuming I've set the gain to fit the constraints of the supply voltage, is there a difference you're seeing between this and a normal class D run from the same supply that would cause problems with transients?
I didn't say it won't. I suggested you might not know enough to make it will. My silly too. The output of your Class A plus a bit becomes the supply to your Class A so that would be the reference to your switching regulator.

Your biggest problem with a single ended buck would be slew rate for a decreasing load. A Class D amplifier is in effect a synchronous buck. There is active reset of the output inductor.

If you have your heart set on discrete then have a look at the basic UCD circuits. Like as not you can bend it to get your desired result.
Joined 2007
Paid Member
Maybe I'm not understanding you :) as I read it (post #1) to be a variable PWM type power supply whose output is determined by tracking the audio, and this is then used as a variable voltage rail to power a conventional linear Class output stage.
Mooly - correct! To my understanding, this is synonymous with class D: PWM tracking audio input. My sited 555 reference design is, after all, presented as a class D implementation.

MorbidFractal - falling slew rate is something I had not considered and makes total sense. However, worst case result looks like a slight decrease in efficiency, right? Admittedly, that's exactly what I'm hoping to improve, but I'm seeing this effect as relatively small compared to the overall improvement I should get.

I'm aware of the UCD design. It has something like 14 active components; compare to my proposed 4. I suppose I'm looking for a happy medium between building a ucd from scratch vs. a turnkey class d board; break the problem down in to bite size chunks. Everybody knows what opamps do, a 555 is simple and worth learning about, and the gate driver is just a buffer with some safety features. I could trade the gate driver for a class D driver, but that ushers in the dead time problem and we end up with more magic happening in one IC, plus a higher parts count with the low side switch.

I've been mulling over this concept of a crappy class D-ish stage followed by a linear stage to clean things up for quite a while. Hitting on the notion of a simple 555 implementation with a single switch is the one that grabbed me.
Use this as a tracking power supply to feed a linear output stage, operating in single ended class A. If the filtered PWM output is biased a few volts above the linear output, the linear device will generate little heat, greatly reducing the size/cost of the required heat sink, as well as related energy losses.

I think you have to track both rails, a class A output stage has two devices, both dissipate lots of heat as the quiescent point is mid-rail.

You could use a class D stage to drive a floating low voltage supply used to power the class A output stage. You'd need good filtering of the class D output, and this would need to be very close to zero-phase across the audio band so it tracks the analog signal well enough. The class D would need good feedback around the filter to keep it stiff against load variation.
Mark, single ended class A uses a single rail and output device. Push-pull class A uses two.

A floating LDO did cross my mind. I was thinking feed back in the class A stage might accomplish the same task. Phase lag is of course a concern, as is propagation delay through the PWM stage. I could potentially put a phase lead network in front of the comparator to even out the output filter lag, but that's more parts and complexity, which I'd like to avoid. In either case, I think the end result is inefficiency, as the PWM output would have to be high enough above the output signal to make room for the error: more volts across the output device equals more wasted power.

Would a constant current source in the linear stage help with load variation? Or is that effectively accomplished in the PWM stage?