**stv014**

Well, with dithering (the simple +/-1 LSB TPDF type), I can subtract the quantized signal from its original version, and the difference (the quantization error) sounds and looks (in a spectrum analyzer) like white noise at a constant RMS level no matter what the input signal is. This is easy to verify in practice

If the signal has sufficiently large variance, then quantization noise looses correlation to the input signal, including the non-dithered quantization. Are you referring to non-shaped TPDF generating white noise?