S. E. Mishchenko¹, V. V. Shatskiy², P. N. Bashly³, M. S. Mishchenko4

1, 2 Rostov-on-Don Research Institute of Radio Communication (Rostov-on-Don, Russia)

3, 4 Russian Customs Academy, Rostov-on-Don Branch (Rostov-on-Don, Russia)

Receiving digital antenna arrays (DAA) element base development makes it possible to implement more complex logical signal processing devices in comparison to logical noise compensators. However, the modern antenna synthesis theory does not consider the DAA logical synthesis issues, despite the fact that the appropriate methods creation and development would allow to overcome physical limitations when forming antenna directional characteristics and to solve a number of problems concerning signal processing in radar and communication systems.

The purpose of the research is to increase the DAA radio-electronic systems noise immunity and resolution based on the logical signal processing and neural networks application.

Logical signal processing devices applied in receiving DAA can be represented as the simplest convolutional neural networks (CNN), constituting a limited number of layers and elements. The link ratios between the CNN elements can be selected in the analytical way. However, with the input signals number increase, as well as with the neural network required response forming rules number increase, it is non-feasible to determine the CNN simplest neural network minimum required layers number, the neurons number in each layer, as well as the link ratios between neurons in the analytical way. To solve this problem, the machine learning algorithms are necessary to be applied.

To solve the problem of synthesizing DAA with logical signal processing, the model of the DAA with a diagram-forming circuit in the form of the simplest CNN has been considered. The method of CNN learning for the formation of a logical radiation pattern has been proposed. This method is based on the backpropagation algorithm and characterized by the initial coefficients of the convolutional layer correspond to the complex weighting coefficients for the forming the receiving beams fan.

On the example of a linear equidistant DAA, consisting of 32 elements, mathematical modeling has been performed, the results of which confirmed the performance of the proposed synthesis method. The main regularities of the forming of logical radiation patterns using convolutional neural networks have been performed. It has been shown that the logical directional pattern can have narrower beam than the physically realizable radiation pattern, and beam scanning is possible without retraining the neutron network.

Mishchenko S.E., Shatskiy V.V., Bashly P.N., Mishchenko M.S. Method of synthesis of a digital antenna array with logic signal processing. Antennas. 2021. № 3. P. 66–75. DOI: https://doi.org/10.18127/j03209601-202103-09 (in Russian)

1. Minkovich B.M., Yakovlev V.P. Teoriya sinteza antenn. M.: Sov. radio. 1969. (in Russian)
2. Maksimov M.V., Bobnev M.P., Krivitskij B.Kh., Gorgonov G.I., Stepanov B.M., Shustov L.N., Il'in V.A. Zashchita ot radiopomekh. Pod red. M.V. Maksimova. M.: Sov. radio. 1976. (in Russian)
3. Tatuzov A.L. Nejrosetevaya obrabotka radiolokatsionnoj informatsii. Sb. materialov 4-j Mezhdunar. Konf. DSPA–2002. Moskva. 2002. T. 2 [Elektronnyj resurs]. URL: http://autex.spb.su/download/dsp/dspa/dspa2002/tom2_76.pdf. (in Russian)
4. Grigor'ev L.N. Tsifrovoe formirovanie lucha v fazirovannykh antennykh reshetkakh. M.: Radiotekhnika. 2010. (in Russian)
5. LeCun Y., Broser B., Denker J.S., et.al. Backpropagation applied to handwritten zip code. Neural Computation. 1989. V. 1. № 4. P. 541–551.
6. LeCun Y., Bottou L., Bengio Y., et.al. Gradient-based learning applied to document recognition. Proceedings of the IEEE. 1998. V. 86. № 11. P. 2278–2324.
7. Nikolenko S., Kadurin A., Arkhangel'skaya E. Glubokoe obuchenie. SPb.: Piter. 2018. (in Russian)
8. Gudfellou Ya., Bendzhio I., Kurvill' A. Glubokoe obuchenie. M.: DMK Press. 2018. (in Russian)