Transform Domain End-to-end Learning Communication System

Unknown time-varying interference is a common and practical problem for communication system. An end-to-end learning communication system with transform domain interference suppression is proposed in this paper to suppress the real-time interference for the autoencoder(AE) neural networks. The basic idea is to dynamical shape the waveform in transform domain at both the AE transmitter and the receiver to avoid interfered spectral regions. The simulation shows that the proposed method can save 2dB Eb/N0 at bit error ratio 10−4 than AE-based communication system with 20% wideband time-varying interference.