Waveform Quantization and Compression - Digital Signal Processing: Fundamentals and Applications - page 528

(a)

Bit strea m

Linear
quantizer

Layer 1, or layer 2
frame formatting

Filter banks:
32 subbands

PCM
samples

Side-information
coding

Psycho-acoustic
model

1024-point
FFT

Layer 1 and Layer 2

(b) Non-uniform

quantizer

Filter banks:
32 subbands

PCM
samples

M-DCT

Side-information
coding

1024-point
FFT

Psychoaccoustic
model

Huffman coding

Bit str eam

Layer 3
frame formatting

Layer 3

FIGURE 11.18

Encoder block diagrams for layers 1 and 2 and for layer 3.

over layers 1 and 2 include use of the scale-factor band, where the W-MDCT coefficients are
regrouped from the original 32 uniformly divided subbands into 25 actual critical bands based on
the human auditory system. Then the corresponding scale factors are assigned, and a nonlinear
quantizer is used.
Finally, Huffman coding is applied to the quantizer outputs to obtain more compression. Partic-
ularly in CD-quality audio, MP3 (MPEG-1 layer 3) can achieve CRs varying from 12:1 to 8:1, cor-
responding to bit rates from 128 kbps to 192 kbps. Besides the use of DCT in MP3, MPEG-2 audio
coding methods such as AC-2, AC-3, ATRAC, and PAC/MPAC also use W-MDCT coding. Readers
can further explore these subjects in Brandenburg (1997) and Li and Drew (2004).

11.5 LABORATORY EXAMPLES OF SIGNAL QUANTIZATION USING THE
TMS320C6713 DSK
Linear quantization can be implemented as shown in C Program 11.1. The program only demonstrates
left channel coding; right channel coding can be easily extended from the program. The program
consists of both an encoder and decoder. First, it converts the 16-bit 2's complement data to the sign-
magnitude format with truncated magnitude bits as required. Then the decoder converts the
compressed PCM code back to the 16-bit data. The encoding and decoding are explained in
Example 11.12.