Hi g_u_e_s_t,
a warm welcome on the journey of pushing the audio quality from 1-bit (and 2-bit) outputs to new frontiers! Raising audio quality from such limited systems is a challenging task but I am amazed how far we get. Some years ago I also shared the opinion that without an additional DAC it would be hard to get audio out of an AVR that sound far better than telephone.
I read your writeups about PWM distortion and - at least - tried to understand them. I think you did a very good theoretical work, even I didn't understand everything yet. But I think you did a really great work toward raising audio quality out of such constrained systems and breaking actual known limits. Some time ago I started my journey by coding a high-order delta-sigma-modulator that converts audio-files to pulse-dense-modulated single bit-streams that I played back through the SPI port of the AVR at rates of up to 2 MBit/s. That way I moved most of the quantization noise out of the audible band, but for whatever reason, the quality on real-world hardware was worse than expected and because it is not possible to do mixing easily with pulse-width-modulated data I stopped my efforts. I wanted then to make something that makes good audio quality playback as easy as possible to use and with minimum hardware effort and started SimpleSDAudio.
With your research results I think we should try to get a step further in quality, but it's time to leave theoretic and try it at the real thing. Let's try if and when phase-correct PWM really sound better than fast PWM. Does it make sense for 2-bit systems to go from 16-bit to 14-bit (or even less) for better results? How about pre-filtering the input audio files to something like PWM-Freq/4? We will loose some high-frequency audio-information but maybe reduce perceived noise also.
You measured -110dB noise floor, I can't believe this really. When SimpleSDAudio stopped at a fixed PWM value, there is really no audible noise, but how about when it is accessing SD card and the AVR has to work harder? I thought the noise is much higher than but maybe I am wrong and that noise is really only from those artifacts you described - I have to try again...
Also interesting that Ti-paper about logic outputs. I wondered why it is so hard to drive speakers at reasonable volume, but with a 74AC14 it should be again worth a try to drive a speaker as this thing cost only a quarter Euro.
Keep on your good work! I look forward to see what happens if you apply your thoughts to real hardware...
@WilliamK: Interesting thing that AM modulation. BTW: hook up a laser-pointer or a focused LED to the digital audio output pin and receive it over quite a distance using a small solar-panel from garden-light connected to an amplifier - it works great.