For non-reversible systems this value corresponds to ceil(log2(D')), where D' is the quantization step size normalized to data of a dynamic range of 1. The true quantization step size is (2^R)*D', where R is ceil(log2(dr)), where 'dr' is the dynamic range of the subband samples, in the corresponding subband.

For reversible systems the exponent value in 'exp' is used to determine the number of magnitude bits in the quantized coefficients. It is, in fact, the dynamic range of the subband data.

In general the index of the first subband in a resolution level is not 0. The exponents appear, within each resolution level, at their subband index, and not in the subband order starting from 0. For instance, resolution level 3, the first subband has the index 16, then the exponent of the subband is exp[3][16], not exp[3][0].

The true step size D is obtained as follows: D=(2^R)*D', where 'R=ceil(log2(dr))' and 'dr' is the dynamic range of the subband samples, in the corresponding subband.

This value is 'null' for reversible systems (i.e. there is no true quantization, 'D' is always 1).

In general the index of the first subband in a resolution level is not 0. The steps appear, within each resolution level, at their subband index, and not in the subband order starting from 0. For instance, if resolution level 3, the first subband has the index 16, then the step of the subband is nStep[3][16], not nStep[3][0].