Cellular Automaton Based Random Noise Generator with Post-Processing for DT-CNN Annealing

Inhalt:
A Cellular Automaton (CA) based Random Number Generator (RNG) for Discrete Time Cellular Neural Network (DT-CNN) annealing is proposed. The DT-CNN solves quadratic assignment problem, and is expected to apply various problems. The DT-CNN annealing achieves global optimization by means of injecting a noise term in its convergent process. We use a chaotic behavior of CA as a RNG for DT-CNN annealing, because of their structural affinity. The performance of the DT-CNN annealing is degraded by using row CA based random number due to the bias of its distribution. To improve the performance we propose the use of a post-processing for the random numbers. We investigate the statistical characteristics of the CA based RNG using two statistical test suits for random numbers: FIPS and NIST. Random numbers generated with ‘XOR’ post-processing passes all test items in these test suits. The performance of DT-CNN annealing also improves with these post-processed random numbers.