The most common codec's are the ones used in telephony, which are "companding" codecs. These premphasize parts of the audio spectrum such that the information-rich portion of the frequency range is emphasized in order to preserve the part of that signal that's important to intellegibility. The result is that, while they send only eight bits per sample, they have essentially 13-14 bits of inherent dynamic range.

A linear codec doesn't fuss around with the signal. It just converts it and sends the converted bits as PCM. Not all of them send sample data, as they send differences or deltas rather than the whole signal, in favor of using less bandwidth.

RE