Non-linear FM Demodulation with Neural Networks in Through-the-earth Mine Communications

Classical methods of FM demodulation used in through-the-earth mine communications rely heavily on an assumption of Gaussian random noise. Such assumption ignores many possible non-linear effects which may affect signal-to-noise ratio at the receiver and lead to increase in bit error. We propose a simple technique based on convolutional neural network which is robust to possible non-linear effects. Our field study demonstrates that such effects do occur in through-the-earth communications and that the distorted signal at the receiver can still be demodulated with reasonably low bit error rate.