Digital Signal Processing Reference
In-Depth Information
signals and parameters. The spectrally whitened input signal
e
nb
(
n
) is used for the esti-
mation of the broadband excitation signal
ˆ
bb
(
n
) while the extracted spectral envelope,
described by its predictor coefficient vector
a
nb
(
n
)
¼
[
a
nb,1
(
n
),
a
nb,2
(
n
),
...
,
a
nb,
N
pre,nb
(
n
)]
T
(7
:
33)
is utilized for estimating the broadband spectral envelope. We will describe the latter
term also by a vector containing the coefficients of a prediction filter
a
bb,
N
pre,bb
(
n
)]
T
a
bb
(
n
)
¼
[
a
bb,1
(
n
),
a
bb,2
(
n
),
...
,
:
(7
:
34)
Usually, this vector consists of more coefficients than its narrowband counterpart
N
pre,bb
N
pre,bb
:
(7
:
35)
Finally, both model parts are combined using an inverse predictor error filter
with coefficients
ˆ
bb,
i
(
n
) that is excited with the estimated broadband excitation
signal
ˆ
bb
(
n
)
N
pre,bb
i¼
1
s
(
n i
)
a
bb,
i
(
n
)
:
s
(
n
)
¼ e
bb
(
n
)
X
(7
:
36)
In some implementations a power adjustment of this signal is necessary. Since we want
to reconstruct the signal only in those frequency ranges that are not transmitted over
the telephone line a bandstop filter is applied. Only those frequencies that are not trans-
mitted should pass the filter (e.g. frequencies below 200 Hz and above 3.8 kHz).
Finally the reconstructed and upsampled telephone signal is added (see Fig. 7.9).
7.5.1 Generation of the Excitation Signal
For the generation of the broadband excitation signal mainly three classes of
approaches exist. These classes include modulation techniques, nonlinear processing,
and the application of function generators. We will describe all of them briefly in the
next sections.
Modulation Techniques
Modulation technique is a term that implies the pro-
cessing of the excitation signal in the time domain by performing a multiplication with
a modulation function
e
bb
(
n
)
¼ e
nb
(
n
)
2 cos(
V
0
n
)
:
(7
:
37)
Search WWH ::
Custom Search