Digital Signal Processing Reference
In-Depth Information
time-domain input samples
Signal vectors are compared
in the time-domain. Exploits the
linearity of the Fourier transform.
static input symbol vector (
n
samples)
in
1
in
2
in
n
ε
1
ε
2
ε
n
Convert error vector to
the frequency domain.
DFT
frequency-domain error vector
Remove affected
frequency bands.
DFT-based subband filter
Return to time-domain
representation.
IDFT
DFT-based channel filter
G
G
G
Integrator in feedback loop
forces error vector to
zero, for valid subbands.
s
s
s
out
1
out
2
out
n
Force the output signal to
adopt a QPSK constellation
shape.
ISSR output vector
Figure 3.4.
Internal functional structure of the issr decoder.
Remember that, as it is our purpose to reconstruct the original signal, isi-
corrupted data is available at the input. The intersymbol interference is due
to the fact a part of the input data is missing: some of the subbands were ze-
roed out due to problems caused by destructive fading or narrowband interfer-
ence. As a consequence, the error vector
[
]
from Figure 3.3 does not contain
any valid information in those particular frequency bands. To prevent that this
invalid information starts poisoning the loop, the affected bands should be re-
moved from the system before they are able to re-enter the system. In practice,
this can be achieved by blocking specific frequency bands using a reconfig-
urable dft filter as illustrated in Figure 3.4. It is important to note that the
digital filter from Figure 3.4 operates on the error vector and uses it as a time-
domain signal. In other words, the consecutive samples of [
]
n
of the error vector represent the time-domain input of the dft filter. It is crucial
not to confuse the time-domain input signal of the issr decoder with the time-
domain dimension of the impulse response of the internal feedback system,
which has a vertical signal flow in Figure 3.4.
The attentive reader may correctly remark that the current system is unable to
reconstruct the isi-free qpsk signal vector: a linear system can never produce
signal components in frequency bands for which the input contains no spectral
=
1
,
2
,
···
