Digital Signal Processing Reference
In-Depth Information
based on mostly erroneous information. This remarkable observation leads to
the assumption that the issr decoder does not have a deteriorating effect on the
signal quality for any practical E
b
/N
tot
value. In this sense, the decoder does not
inject additional noise that could compromise the performance of subsequent
decoding stages.
At the start of the issr algorithm, it is important to initialize the output vec-
tor of the issr algorithm to the received input vector. When doing this, sim-
ulations show a fairly stable crossover value around
3 dB, even for a low
number of iterations in the issr routine. By avoiding a randomly chosen ini-
tial population, the optimization loop converges much faster to the final steady
state value, which is of special interest for applications with a limited process-
ing power. Also note that, unlike most pso implementations, the issr routine
does not employ probabilistic mutations during the optimization process. This
ensures a smooth migration from the initial input signal to the reconstructed
signal vector. In most other optimization contexts, mutations are introduced to
prevent that the algorithm gets trapped into local optima. However, initializing
the output vector of the issr algorithm to the input signal appears to result in a
fairly smooth optimization surface. Also, the effect of relentless grinding away
useless frequency bands is smeared out over a wide range of symbols in the
time domain. This can be attributed to the repeated conversions between the
time and the frequency domain representation of the signal. This results in rel-
atively smooth transitions between the consecutive qpsk vectors and a fairly
progressive increase of signal quality at the output of the issr subsystem.
A final observation from Figure 3.9 is that, when the contribution of awgn
noise to the total noise figure increases, the ber performance curve of the issr
decoder reverts to the scores of hard-decision qpsk demapping. This finding
is consistent with the internal workings of the signal reconstruction algorithm:
issr offers no protection whatsoever against the unpredictable nature of white
noise in the time domain: the information contained in a single qpsk symbol
sample cannot be retrieved from surrounding symbols as issr is focussed on
the exploitation of frequency diversity, instead of diversity in time. This is
also the reason why the performance of issr does not show the characteristic
steep slope as typically seen in sequence-based coding schemes. In order to
fully benefit from its advantages, issr should thus always be used as the inner
decoder of a compound error correction system. The result is a robust coding
system that exploits diversity in both the time and the frequency dimension.
If correctly dealt with, frequency diversity and intersymbol interference can
actually become one of the major advantages of a wideband, frequency-
selective channel: a narrowband communication system will always suffer
from flat fading in a multipath channel. Increasing the transmission power in
a narrowband channel does not significantly increase its performance because
−
