A ResNet Based End-to-End Wireless Communication System under Rayleigh Fading and Bursty Noise Channels

Deep learning has been applied recently in the wireless communication area such as modulation classification, channel estimation and signal detection. Many of the wireless communication problems can be modeled as classification problems. Residual learning has proven to have a crucial role in image recognition for providing fascinating classification accuracy. This paper proposes a residual learning-based autoencoder model that can jointly optimize the transmitter and the receiver while communicating over Rayleigh flat fading and bursty noise channels. Depending on the number of bits per symbol at the transmitter, the proposed system can automatically learn the constellation mapping and reconstruct the transmitted bits with very low loss metrics. Simulation studies show that the block error rate performance of the proposed model is superior to the convolutional layer based autoencoder system as well as the conventional modulation system under Rayleigh flat fading and bursty noise channels.