I'm thinking of your statement, and comparing it to an amplified "shark-tooth" RC Charge/Discharge plot. The upper and lower points are a bit thicker than a regular triangle wave, but not by much, and the slope across the midpoint is greatly different than that of a triangle wave.This may generate some controversy:
I haven't done the math, but my gut tells me that the PWM from an exponential triangle generator will be as accurate as that from a linear triangle. This assumes that the rising and falling time constants are identical, and that the comparator's propagation delay is independent of input slew rate. I am referring to the comparator that generates the PWM, not the one involved in generating the triangle wave.
OTOH, maybe I'm the only one who thinks that this might be controversial.
I am not implying that, with real-world parts, the exponential triangle is as good as the linear triangle for generating PWM.
It would be interesting to hear the change in sound if one swapped out an RC waveform into a Class D amp, then switched between that and the triangle wave. When getting to nearly square waveforms, there will be a big problem, but the RC waveforms would only "mess up" the zero point, so to speak. I'm now curious to the human ear picking up the change, assuming the RC plot was very close to a prefect triangle wave.