I've checked the output with a scope: It's indeed a PWM signal, but not the canonical PWM with fixed frequency and varying duty-cycle but something more sophist

I've checked the output with a scope: It's indeed a PWM signal, but not the canonical PWM with fixed frequency and varying duty-cycle but something more sophisticated. PWM nevertheless, probably there's some shaping being done by the Videocore-Code (through which the audio data is routed). I've tested the output with a sound-card (M-Audio Delta 44, via a Yamaha MV12/4 Mixing console as preamp) that has (at the test-settings) a noise-floor of ~-130dB/Hz. If I let the raspberry-pi play a sine-wave, I get harmonics at ~60-66dB down from full-scale. I suspect that this is unshaped quantization noise, so this would compare the suspicion that the PWM resolution is 11 or 12 bit. Here's a screenshot of a frequency analyzer plugin running while recording the audio output of the RPI, for comparison I put in the same sine being played by the 16bit DAC of my mobile phone, here the harmonics are down -90dB or more. https://picasaweb.google.com/lh/photo/fWlV5aan-94e9GOpXJaLm9MTjNZETYmyPJy0liipFm0 https://picasaweb.google.com/lh/photo/fWlV5aan-94e9GOpXJaLm9... Sinewave has been generated with the command $ sox -n -t wav - synth 60 sine 1013.0 | aplay on the raspberry pi.