or Tone Ordering
Tone ordering [1]
A DMT time-domain signal has a high peak-to-average ratio (its amplitude distribution is almost Gaussian), and large values may be clipped by the D/A-converter. The error signal caused by clipping can be considered as an additive negative impulse for the time sample that was clipped. The clipping error power is almost equally distributed across all tones in the symbol in which clipping occurs. Clipping is therefore most likely to cause errors on those tones that have been assigned the largest number of bits (and therefore have the densest constellation). These occasional errors can be reliably corrected by the FEC coding if the tones with the largest number of bits have been assigned to the interleave buffer.
The number of bits and the relative gains to be used for every tone are calculated in the ATU-R receiver, and send back to the ATU-C. The pairs of numbers are typically stored, in ascending order of frequency or tone number , in a bit and gain table.
The ``tone-ordered'' encoding assigns the first bytes (8 bits) from the symbol buffer to the tones with the smallest number of bits assigned to them, and the remaining bytes (8 bits) to the remaining tones.
[1]
http://www.cs.tut.fi/tlt/stuff/adsl/node13.html