Bit allocation is of fundamental importance, since it determines the amplitude of the quantization noise. In particular, the SMR value depends on bit allocation and can be:
| - | larger than SMR, so that quantization noise can not be perceived (MNR > 0) |
| - | smaller than SMR, thus leaving a part of the quantization noise in the audible domain (MNR < 0) |
From SNR and SMR (values which are available only in the encoder), some of the following situations can be verified for every frame:
Fig. 3.1 - Possible situations that can arise for Signal, Noise and Mask. For every one them, the effect at listening is considered. Such effect is only theoretical and its correspondence to reality depends solely on the accuracy of the perceptive model.
Each configuration is examined and the theoretical consequences at the perceptive level are examined. Note therefore that MNR, SNR and SMR vary for each subband of the FRAME, and that therefore the total quantity for the audio sequence is given by the whole set of situations which arise in all subbands and all FRAMEs.
The maximum signal level is the one that determines the scaling factor, i.e. the highest among the 36 frequency samples of the subband.
Fig. 3.2 - Optimum case for which quantization noise is masked. In such subband, the decoded signal appears perceptively identical to the original signal. The following holds:
The larger the MNR, the larger the margin for signal processing.
Fig. 3.3 - The signal has the highest amplitude and the quantization noise is higher than the masking, i.e. in the subband there is some quantization noise that can be heard, depending on the nature of the signal and the influence of the adjacent bands. In this case, the following holds:
Fig. 3.4 - Such case can never happen, because it is impossible that the quantization noise is higher than the signal amplitude; SNR is always a positive quantity and indicates the distance in dB between Signal and Noise.
Fig. 3.5 - In this case the signal is imperceptible and the quantization noise, being lower than the signal, is masked. The following holds:
The band will be allocated only if the bit rate provides additional bits, i.e. if after allocating all S_M_N bands there are still some bits in the Frame available. The band, even though it plays no role in regard to sound perception, can carry important information for future processing.
Fig. 3.6 - This is a configuration similar to N_M_S, which can never happen because the signal would be higher than the threshold while quantization noise is higher than the signal, which is impossible.
Fig. 3.7 - Situation similar to M_S_N again: it is impossible that the quantization noise is higher than the signal!
Therefore, bands with masked signals are allocated only in case of availability of bits.
[ Index | Main Paragraph | Previous paragraph | Next paragraph ]