With standard linear PCM (pulse code modulation), as used
on audio-CDs, all frequencies are stored with equal precision (as 16 bits) but although the human auditory system has a
frequency range of ca. 20Hz to 20kHz, at frequencies above 3kHz and below 200Hz it becomes less sensitive. It is therefore not necessary to store audio in those frequencies with the same detail as they can't be heard anyway. The PCM signal also contains redundant information that can be removed by the coder and replaced by the decoder. The apt-X100 algorithm used for DTS incorporates these principles.
AC-3 (Dolby Digital), ATRAC (SDDS) and MUSICAM (L.C.Concept) algorithms use another feature of human hearing system: auditory masking. The principle is that a loud sound will mask out quieter sounds that occur at similar or near frequencies. So, if the human ear can't perceive them, there's no point of storing them. Because not all information is stored, it is called lossy encoding.
Both techniques divide the frequency range of the human hearing system in subbands. The subbands containing the frequencies the human ear is most sensitive to will be stored with greater accuracy.