<p>The authors consider the application of machine learning methods to digital signal processing in telecommunication systems, in particular, automatic modulation classification and subsequent signal demodulation. Four machine learning methods for automatic modulation classification are investigated: multinomial regression, nearest neighbors method, Gaussian mixture method, and convolutional neural network. Experimental results on artificial data showed the recognition accuracy for five modulation classes ranging from 96% to 99%. The convolutional neural network provided the highest accuracy (99%). However, the other three methods, which have a simpler structure, demonstrate a satisfactory compromise between accuracy and implementation complexity. Testing on 89 signals from real modems showed that the nearest neighbors method attains the highest classification accuracy (100%), while the other methods provide 99% accuracy. This indicates that high classification accuracy can be attained using much simpler methods compared to convolutional neural networks. A method of block demodulation of signals based on multinomial linear regression and a feedforward neural network is proposed, which has a simple practical implementation compared to other known methods. It is shown that in the case of high noise levels, the proposed method provides higher signal recovery accuracy compared to the traditional demodulation method based on the Gardner and Costas loops, and also uses fewer parameters than other known methods.</p>