Sigma-Delta versus Binary Weighted AD/DA conversion, what is the most promising?

Though conversion of analog signals to and from digital data is supposed to be a mature, well defined technology, the application in digital signal processing like digital audio unveiled a number of neglected artifacts. Also the correlation between audibility and type of imperfection in the converted signal is only partly covered. Test signals as used on CD records explore only a small part of the converter characteristic with sufficient detail. Bias-dependent glitches, slew-rate distortion, noise-switching, clock-jitter and parasitic coupling, which can seriously degrade the signal, are mostly ignored. The reproduction of low-level signals requires differential linearity to be much better than usually specified. Low-level distortion is produced at every place in the chain between analog input signal and analog output signal where quantization or conversion errors are introduced. In general this distortion is reduced by adding noise as a dither signal to de-correlate errors. The amplitude of the dither varies from one LSB step to several LSB steps, depending on the error to be de-correlated, at the expense of a considerable loss in dynamic range. Error feed-back by noise-shaping can be used to reduce quantization distortion if over-sampling is sufficiently high. Dithering or noise-shaping however can emphasize certain non-linearities generated by the D/A converter, e.g. like level-dependent glitches. Converters that guarantee good low-level reproduction are oversampled noise-shaping coders. These coders obtain linearity by time-averaging in stead of by analog precision.