Cellular Nonlinear Networks: State of the Art and Applications - Systematic Design for Emergence in Cellular Nonlinear Networks - page 10

cellular computers may have hundreds of thousands of cells they can process large
databases such as images or other multi-dimensional signals. For instance cellular
computing systems can extract edges, corners and other features of interest from
an image regardless the size of that image. Besides pattern recognition, cellular
systems are capable of various linear and nonlinear filtering tasks.
Ciphering is another popular application of cellular computing. Indeed several
patents have been filled for such applications where cells are designed such that a
complex, chaotic dynamics, emerge in the array of cells. Unlike other methods for
random number generation (e.g. the linear feedback shift registers) the CA-based
method is scalable, i.e. one can add more cells without changing the essence of the
dynamics behavior. By employing a larger number of cells, the probability of de-
ciphering decreases making such systems extremely reliable in terms of security.
In a recent paper [33] the use of cellular automata as ciphering systems is carefully
investigated and several benchmarks are computed showing that highly reliable
random number generators can be obtained using relatively simple cells (e.g. four
input Boolean cells) arranged in one-dimensional arrays of several hundreds of
cells.
The pseudo-random and complex dynamics of cellular systems is also exploited
for built in self test (BIST) systems. Such systems are required to perform func-
tional analysis of complex circuits and detect functional failures. In doing so, a
convenient solution is to embed a cellular system acting as a pattern generator.
The CA is designed such that a large sequence of patterns is generated as a result
of the CA dynamics. The length of the sequence is optimized as a tradeoff be-
tween a reasonable testing time (demanding thus not a very long sequence) and
enough information in the sequence to detect certain failures. In the parlance of
emergence ideal BIST sequence generators are operated in the “edge of chaos”
regime, i.e. they are neither random signal generators with very long cycles nei-
ther orderly systems with very small length limit cycles.
Signal compression is another interesting application of cellular automata. Several
solutions were reported so far. For instance [34] proposes a solution called a “CA
transform” where a signal (image or sound) is decomposed as a binary weighted
sum of basis signals. The basis signals are generated by properly tuned cellular
automata with certain genes (cell parameters). In order to reconstruct the signal
one needs only the set of binary weights and the (relatively) short description of
the CA cells generating the basis signals.
In order to demonstrate the idea of image compression using cellular automata
we proposed recently a novel approach where generalized cellular automata
(GCA) can be used. Using the following simple Matlab programs, a wide palette
of images can be obtained (one pixel corresponds to one cell in the image), part of
which are depicted in Fig. 2.2. Each image in Fig. 2.2 displays above the set of 17
generating parameters.
Only a few parameters (17 parameters, underlined above) control the diversity of
the obtained images. Such images or part of them can be combined to approximate