AUTOMATIC SIGNAL TYPE RECOGNITION USING THE CYCLIC CONTROL METHOD OF NEURAL NETWORK LEARNING RATE

Abstract

In the modern radio frequency systems, there are problems when it is necessary to receive and demodulate signals about which there is no initial information. Solving such problems involves recognizing the type of signal using certain methods and algorithms, before demodulating the signals.

         Among the various methods developed, a relatively modern and effective method is the automatic recognition of the type of signal modulation using artificial neural networks. An artificial neural network is a mathematical model of a biological neural network, endowed with the necessary capabilities to solve classification problems. Both in the biological analog and mathematical model, the network must be trained in advance, only then it will be able to make accurate decisions. Training efficiency depends on many hyperparameters. Hyperparameters  do not obey any rules, they can only be obtained from experimental results. One of the most important hyperparameters of the artificial neural network is the learning rate. There are many methods to control the learning rate that can be selected, depending on the type of problem at hand and the results of the experiments. To solve any problem, it is necessary to choose the most effective and fastest method, since the accuracy of the result depends on this choice. The learning rate cyclical control method is a relatively modern method.

         For the initial stage of the development of the system for automatic recognition of the signal modulation type, it was decided first to check the developed algorithms on the task of automatic signal type recognition. In this work, the effectiveness of the cyclic learning rate control method for the problem of automatic signal type recognition was also evaluated.