Class D audio amp

Try the LM311. It has a 200μs response time. Note that it requires a pull up resistor on its output.

 

For your interest, here's the LTspice simulation of a PWM circuit using a transistor sawtooth generator with about a 250kHz frequency, along with a fast LT1720 comparator:
The Demod output is from a low-pass filter to show the demodulated PWM signal.

 

You would normally use a triangular wave rather than a sawtooth for best linearity.

 

How does that affect linearity?
I would think it would be the linearity of the wave slope that determines the linearity.

 

How does whether its a triangle or a sawtooth affect the duty-cycle accuracy of the square-wave?
Either a triangle or a sawtooth will generate a 50% duty-cycle when the modulation input is 1/2 the PP voltage of either wave-shape.

 

You're right of course, but for a given frequency, the slope (dv/dt) of the ramp of a sawtooth of a given amplitude is only half as fast as that of the sides of a triangle.

But which augers for better accuracy/fidelity is another question entirely.

 

Using a saw tooth will skew the PWM pulses leading to distortion.
Re the LM311, my post should have read 200ns not 200μs.

 

The "skew" is just a shift in time of each PWM pulse from the center of the sawtooth.
Since that is a constant shift for each pulse, I don't see how that can add distortion to the modulated signal.

 

It's been some time since I worked in this stuff, but from memory, the comparator without sample-and-hold implements what was called "natural sampling" which has advantages. Since the effective sampling rate is not significantly higher than the signal (low-pass) bandwidth, this may have some advantage. Sort of depends on the integrity that you need in the end.
For many applications, this, as suggested, might be just a theoretical concern.
Quite prepared to be wrong. Just from memory.

 

What "sample and hold" ?
You seem to referring to a different thread.

 

Sorry, I assume this was a bit of a theoretical aside, of little practical impact.
.
It's a one-bit converter with "natural sampling". That is, the edges occur when the waveform crosses a threshold, there is no pre-defined sample time nor sample-and-hold as might be more normal in conversion.
Natural sampling, allowing the waveform to choose the sampling point had some advantages; the details or justification of which, I frankly can't remember. Regarding the sawtooth versus the triangle; could speculate that having two such times, may have advantages over having one.

 

Sorry but I have no idea what that means in regard to generating a PWM signal.
There is no "sampling" as such.

 

Actually, now that I have taken time to actually read some of the offered paper, it does introduce these concepts, natural versus uniform sampling, on page 9 and explains it somewhat.
In my terms, each upward or downward slope of your sawtooth takes a single sample between the times of the extremes. In this comparator approach the timing of that sample is determined by the waveform and the slope waveform. In more traditional conversion, the signal level would be samples at some point in the period and converted. The paper and reference [4] suggest this natural sampling has benefits.

As pointed out, using a symmetrical up & down waveform effectively doubles the sample rate as you get one transition & natural sampling point on the up & one on the down.

 

That is exactly the point. It is the shift in time that creates the distortion.

 

But the shift is constant from pulse to pulse so show me how that creates distortion.