Hans Peter Bernhard, Gernot Kubin,
"A Fast Mutual Information Calculation Algorithm"
: European Speech and Signal Processing Conference EUSIPCO94 Signal Processing VII: Theories and Applications, Elsevier, Seite(n) 50-53, 9-1994
Original Titel:
A Fast Mutual Information Calculation Algorithm
Sprache des Titels:
Englisch
Original Buchtitel:
European Speech and Signal Processing Conference EUSIPCO94 Signal Processing VII: Theories and Applications
Original Kurzfassung:
In modern signal analysis, chaos-theoretic interpretation is becoming more
and more popular. One of the most important parameters of chaotic systems
is their information gain or entropy rate. Its measurement provides
performance limits for nonlinear signal models in source coding and
prediction applications. We have developed a fast algorithm for the
estimation of the entropy rate for attractors reconstructed from
finite-length signals with the method of delays. With conventional
algorithms, the computational load increases almost exponentially with the
reconstruction dimension. We exploit dynamical memory organization
principles to arrive at a linear increase, i.e.~whenever we
increment the reconstruction dimension by one, the calculation time of the
new algorithm increases only by $1/10$ of the calculation time needed in
two dimensions. Furthermore, the calculation time
is proportional to $N \log_2 N$ where $N$ is the signal length.