Compression of Audio Signals to MPEG and Dolby Digital - Digital Video and Audio Broadcasting Technology - page 153

8.5 Subband Coding in Accordance with MPEG Layer I, II

First the method of subband coding will be discussed. In accordance with
MPEG layer I and II (Fig. 8.10.), the audio signal is passed through a filter
bank of 32 filters that split the signal into frequency subbands of 750 Hz.
For each subband there is a separate quantizer controlled by an FFT block
and a psychoacoustic model. The quantizer either completely suppresses
the subband in question or reduces the number of quantization steps. In the
case of layer II coding, FFT is carried out every 24 milliseconds on 1024
samples. This means that the information fed to the psychoacoustic model
changes every 24 milliseconds. During the 24 ms intervals, the subbands
are subjected to irrelevance reduction in accordance with the information
received from the psychoaoustic model block. In other words, the signal is
treated as if its composition had not altered for 24 ms.

Audio in

BP Q

BP Q

compressed
audio out

Frequency
subbands

BP Q

Bandpass
filter

Quantizer

Example:
MPEG Layer I, II

Psycho
acoustic
model

FFT

512 point FFT
@MPEG Layer I,
1024 point FFT
@ Layer II;
every 24ms

Fig. 8.10. Subband coding using 32 bandpass filters in MPEG-1 and MPEG-2
layer I, II

Because of the different audibility thresholds, bit allocation and thus
quantization is different for the different subbands. Quantization must be
finest at low frequencies; it may be reduced towards higher frequencies.
Fig. 8.11. illustrates the principle of irrelevance reduction in audio
transmission by means of two examples. In one subband, there is a signal
at about 5 kHz with a level above the masking threshold. In the case of this