Neural Network Implementation of the Probabilistic Detector in the Digital Stabilizer of Distorted Pulse Sequences

The paper contains the description on how to replace the Bayesian detector (“probabilistic relay”) in the developed digital stabilizer of the distorted pulse sequences by its neural network implementation. The main idea has been inspired by the proven abilities of neural networks to tune their weight coefficients in more flexible way providing improved accuracy for estimated parameters. As in the previous case, there are considered the pulse sequences being under affect by multiplicative and low frequency interferences along with the additive noise. The simulation modeling results have shown that under the same noise variances and interferences' behavior, the usage of the neural network implementation allows decreasing the mean square errors of pulse amplitude stabilization and improving the pulses' form restoration with 5–7 % comparing to the “classical” implementation of the detector as a computing routine procedure.