underflow beyond the range of values allowed by the precision of the input. However, the division by two which is part of

some of the prediction calculations shall be approximated by an arithmetic-shift-right by one bit.

coding in the hierarchical mode of operation are calculated modulo 65 536, thereby restricting the precision of these

differences to a maximum of 16 bits. The modulo values are calculated by performing the logical AND operation of the

two's complement difference with X'FFFF'. For purposes of coding, the result is still interpreted as a 16 bit two's

complement difference. Modulo 65 536 arithmetic is also used in the decoder in calculating the output from the sum of

the prediction and this two's complement difference.