The spectrum (0..22.5kHz) for 44.1kHz sample rate is divided into 16 bands of equal spectral width of 1406,25Hz. The division into bands of same width is a direct consequence of the DCT synthesis scheme. Some of these 16 bands are further devided into bandlets, as at low frequencies the same spectral width occupies a higher relative spectrum range (i.e. a higher pitch interval). The bandlet data has to be combined into band data before the IDCT synthesis can be performed, so all processing steps in time domain work on band basis or even more coarse, while frequency-domain processing often is performed on the bandlet level.

The selection of spectral dither noise amplitude uses an even more coarse scheme of dither groups that contain several bandlets. Borders of dither groups coincide with borders of bands except for the first dither group that leaves out the first two bandlets of the first band.

The layout is like this

• Per substream data:
  • highest non-zero bandlet (used to calculate highest non-zero band and highest non-zero dither group)
  • Mute flag
• Per channel data
• Per dither-group data
  • Frequency-domain dither level
• Per band data
  • Joint-stereo info
  • Envelopes
  • Tone data
  • Time-domain dither info
• Per bandlet data
  • Spectral quantizers
  • Spectral levels
  • Coefficient encoding tree selection
  • Coefficients