COMBINATORIAL OPTIMIZATION OF SYSTEMS OF NEURAL NETWORK CRYPTOGRAPHIC DATA PROTECTION

Abstract

The problem of improving the reliability of cryptographic data protection in neural network systems with flexible configuration is considered. To ensure the possibility of encrypting/decrypting messages it is proposed to use combinatorial optimization methods for the tasks of forming encoded sequences with improved quality indicators for correcting ability, noise immunity, and autocorrelation properties. The basis of combinatorial optimization is the principle of optimal structural relationships, the essence of which is to achieve the maximum diversity of the system under the established restrictions on the number of structural elements and their mutual placement in space-time. It is proposed to use signal-code sequences for neural network data protection, which are characterized by high noise immunity and low level of the autocorrelation function, using various types of optimized code sequences depending on the set of requirements for work under specific conditions, taking into account restrictions on the duration of sending encrypted messages and the presence of noise in communication channels. The system for neural network cryptographic data protection has been developed using encoded signal sequences, where the number of binary characters of different names differs by no more than one character, which minimizes the value of the autocorrelation function of the encoded signal at a fixed bit depth. To ensure high technical and economic indicators of the cryptosystem, it is advisable to equip it with specialized modules of neuro-similar elements of the network with the possibility of training and flexible configuration for cryptographic data encryption. The relationship between the parameters of optimized encoded signal sequences, in which the value of the autocorrelation function is minimized, and the maximum achievable number of detected and corrected errors has been established. It is proposed to use unique properties of combinatorial configurations with a non-uniform distribution of structural elements, which are distinguished by the fact that the set of all ring sums of their numerical values occurs a fixed number of times. A comparative analysis of cryptographic methods for data protection and transfer using non-standard codes built on the so-called IRB code sequences together with other signal-code constructions was carried out.