IEEE Consumer Electronics Magazine - May/June 2021 - 100

Innovations in Consumer Electronics
Figure 4. Solution to preecho: Long block is applied
to non-attack blocks as in (a) and (b) whereas short
block is used for a block with attack as in (c).
Figure 3. Mechanism of preecho generation:
Evenly distributed quantization noise is audible
before attack and inaudible afterward.
preceding the attack have a near-zero power and
are not sufficiently large to mask the quantization
noise, resulting in audible preecho. In
Figure 3(c), preecho is visible in Section P preceding
the attack, whereas it is invisible and
inaudible in Section Q due to masking.
Because the quantization noise is evenly
distributed within the block over the samples,
the adjacent blocks are free from preecho. For
ashortblocksizeasdepicted inFigure3(d)
where the block size is one fourth of (c), the
preecho exists only in the third block for a
shorter duration. Moreover, a short preecho
preceding an attack is likely to be inaudible
due to premasking in the temporal domain10.
From Figure 3(c) and (d) , it looks that short
blocks are better than long blocks, however,
it is not that simple. Side information is transmitted
with the quantized samples block by
block. It means that the short block in Figure 3
(d) transmits side information four times as
frequent as (c), leading to four times the total
size of the side information per second.
100
Accordingly, the main information, i.e., the
quantized audio samples, is reduced and the
audio quality after decoding is degraded. A
long block-size is more efficient in transmission
of side information and worse in audio
quality with preecho at attacks than a short
block-size, and vice versa. In summary, there
is a tradeoff in block-size selection between
efficient transmission of side information and
quality of the decoded audio signal.
This tradeoff is alleviated by adaptively
changing the block-size depending on the presence
of an attack as shown in Figure 4. As far as
there is no attack in a specified long block, the
most efficient long block-size is applied because
there is no preecho problem. On the contrary,
presence of an attack selects a predetermined
short block-size for sufficient audio quality and
compromised efficiency. The size of the long
block is designed to be a multiple of the short
block-size. The adaptive block-size MDCT
described here is the technique to make preecho
inaudible in MPEG-1 and subsequent MPEG
Audio standards.
When this example is looked at from the
Toyota's Five Why's point of view, the problem
is preecho and the identified cause, at
whichever the step, is the tradeoff in blocksize
selection between audio quality and datacompression
efficiently with side information.
It is worthwhile to apply one more " why " to
arrive at a further cause that is presence of
an attack. Then, a possible and most straightforward
solution to this cause is to erase the
attack by flattening. It is also possible to view
this problem as insufficient source signal
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - May/June 2021

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - May/June 2021
