I. A. Kalmykov, R. A. Voronkin, D. N. Rezenkov, Ya. V. Yamarlukova, A. A. Falko

For increase of speed of digital signals processing (DSP) it is offered to proceed (pass) from one-dimensional processing signals to multivariate on the basis of application of polynomial systems of deductions classes (PSDC) in which not brought minimal multinominals are used as modules of non-positional system. Thus use neural network logic basis will allow to provide a real time scale at realization of parallel calculations. Characteristic feature of all arithmetic devices DSP functioning in PSDC, is necessity of performance of summation on the module two. The basic realizations of multiport adders on the module two use neural network basis are considered. The expediency of use at training of multiport of neural network adder of majority genetic algorithm with the allocated dominant is shown. Application of the developed algorithm at training of neural network adder on the module two has allowed to change its architecture in such a manner that weight factors and displacement of neural began to belong to three-element set {-1;0;1}. Thus the synthesized neural network is characterized by smaller circuit expenses in comparison with earlier known realizations.