Databases Reference
In-Depth Information
Spectral
envelope
coding
Bit
allocation
MDCT
Mantissa
coding
Framing
F I GU R E 17 . 14
The Dolby AC-3 algorithm.
17.5.1 Bit Allocation
The Dolby AC-3 scheme has a very interesting method for bit allocation. Like the MPEG
schemes, it uses a psychoacoustic model that incorporates the hearing thresholds and the
presence of noise and tone maskers. However, the input to the model is different. In the
MPEG schemes the audio sequence being encoded is provided to the bit allocation procedure
and the bit allocation is sent to the decoder as side information. In the Dolby AC-3 scheme
the signal itself is not provided to the bit allocation procedure. Instead a crude representation
of the spectral envelope is provided to both the decoder and the bit allocation procedure. As
the decoder then possesses the information used by the encoder to generate the bit allocation,
the allocation itself is not included in the transmitted bitstream.
The representation of the spectral envelope is obtained by representing the MDCT coeffi-
cients in binary exponential notation. The binary exponential notation of a number 110.101 is
0
2
3
, where 110101 is called the mantissa and 3 is the exponent. Given a sequence
of numbers, the exponents of the binary exponential representation provide an estimate of the
relative magnitude of the numbers. The Dolby AC-3 algorithm uses the exponents of the bi-
nary exponential representation of the MDCT coefficients as the representation of the spectral
envelope. This encoding is sent to the bit allocation algorithm, which uses this information in
conjunction with a psychoacoustic model to generate the number of bits to be used to quantize
the mantissa of the binary exponential representation of the MDCT coefficients. To reduce
the amount of information that needs to be sent to the decoder, the spectral envelope coding is
not performed for every audio block. Depending on how stationary the audio is, the algorithm
uses one of three strategies [
200
].
.
110101
×
The D15 Method
When the audio is relatively stationary, the spectral envelope is coded once for every six
audio blocks. Because a frame in Dolby AC-3 consists of six blocks, during each block we
