Nom-1-2015-sait pdf

Синева Ирина Сергеевна, 
Московский технический университет связи и информатики
Россия, Москва, профессор,
[e-mail] 
68

---

T-Comm #1-2015
69
МАТЕМАТИКА

---

T-Comm #1-2015
70
МАТЕМАТИКА

---

T-Comm #1-2015
71
МАТЕМАТИКА

---

T-Comm #1-2015
72
МАТЕМАТИКА

---

T-Comm #1-2015
73
МАТЕМАТИКА

---

MATHEMATICS
T-Comm #1-2015
74
COMPARATIVE ANALYSIS OF ERROR-CORRECTING PROPERTIES OF GENETIC
NOISE IMMUNITY CODING ALGORITHMS FOR CLUSTERED SOURCE SPACES
Alexey Batalov
postgraduate student of MTUCI, Moscow, Russia, 
[e-mail]
Irina Sineva
professor of MTUCI, Moscow, Russia,
[e-mail]
Abstract
Noise immunity is one of the main aspects of the messaging system. As a rule, high noise immunity is achieved by introducing
of the additional redundancy. However, in some cases, it is impossible or impractical. To improve noise immunity without
introducing additional redundancy has been developed a distinct class of algorithms, which thus improves the decoding result
if an error occurs in the communication channel. This class of algorithms based on the principles of genetics, when there is a
parent population and the population of descendants. Population is a set of codewords. Wherein for each step of the parental
population of codewords generated codewords offspring population. This, in turn, will be the parent population in the next
step. To improve noise immunity source messages are encoded using genetic algorithms. The basic operating principle is that
the algorithms to encode the original message source space is that for neighbor messages in code space correspond also close
message in the initial space. It turns out that, when an error occurs in the communication channel, this error distorts the com-
bination slightly. The resulting combination is close to the original one, as in the codeword space and in the original metric
source space. Quite often the source space can be clustered for some reasonable number of clusters. One of the key features
of genetic algorithms is its flexibility. Separately for each class or clustering type spaces can develop a separate type of algo-
rithm that will select the starting point on the basis of existing clusters. First we need to examine in greater detail the behav-
ior of genetic algorithms without any modifications in comparison with a random source coding messages. The article
describes the clustering and non-clustering metrized spaces that are well described by the following distribution: normal, uni-
form and Cauchy distribution. The choice of models is determined by the fact that the Gaussian random fields have a strong
localization, uniform do not have it at all, and the distribution of the Cauchy have weak localization, in which even the center
of the scattering due to the divergence of integrals is not defined and it is necessary to use the geometric characteristics to
select the starting point of the original algorithm .