The development of basic components of the neural network protection system, data transmission coding based on an integrated approach, which includes an improved method of neural network encryption (decryption) and the method of adaptive barker-like coding (decoding) of data, which focuses on modern element base. The principles of specialization and adaptation of hardware and software to the structure of algorithms for neuro-like encryption (decryption) of data, neural network architecture, and barker-like code are used to develop the system. The architecture of the system is proposed, which takes into account the variable composition of the equipment and modularity. The method of neural network encryption (decryption) of data has been improved. The time of neural network encryption and decryption of data depends on the size of the tables of macroparticle products. The size of the tables of pre-calculated macroparticle products is based on the provision of encryption and decryption of data in real-time. A method of adaptive barker-like encoding (decoding) has been developed, which, due to the signal-to-noise ratio, provides high noise immunity and reduces data transmission time. The hardware of the system, which was created using the developed basic components of neural network protection and barker-like data encoding, is described. When creating hardware, ready-made components and modules of industrial production are used as much as possible, and the availability of appropriate means of software code development is taken into account. Means of neural network cryptographic encryption (decryption) of data of the mobile part of the system are implemented using a microcomputer-based on SoC. Not the most powerful microcomputer of the NanoPi Duo type from FriendlyElec has been especially used to test the means of neural network cryptographic encryption (decryption) of data. Using the created system, it is determined that the performance of neural network cryptographic encryption (decryption) of data blocks based on a microcomputer is carried out in close to real-time. The time of formation and training of the neural network is about 200 ms, and the implementation of encryption and decryption procedures is about 35 ms and 30 ms, respectively, and does not depend significantly on the chosen configuration of the neural network.
Read full abstract