rs55611 (12/20/82)
Someone asked if the signal to noise ratio (SNR) of digital recordings was as good at low signal levels as at high signal levels, due to the effects of quantization distortion. In order to get a SNR that is essentially constant down to very low signal levels, the quantizing algorithm used is some form of logarithmic encoding, rather than linear encoding. In other words, the "step size" between successive code values increases as the sampled signal level increases. This gives a constant SNR, down to signal levels just slightly larger than the smallest step size. In the telephony world, the two commonly used logarithmic encoding schemes (8-bit Pulse Code Modulation) are u-255 (US and some others), and A-Law (most of the rest of the world.) While I don't think there has been standardization of any 14-16 bit audio encoding algorithms yet, I'm sure that most of the current schemes use logarithmic encoding. If linear encoding is used, the main impact is that more bits/sample are needed to get equivalent SNR values. (8 bit logarithmic encoding is as good as 10-11 bit linear encoding, for example). Bob Schleicher ihuxk!rs55611