Deep Learning for Interference Cancellation in Non-orthogonal Signal Based Optical Communication Systems

Non-orthogonal waveforms are groups of signals, which improve spectral efficiency but at the cost of interference. A recognized waveform, termed spectrally efficient frequency division multiplexing (SEFDM), which was a technique initially proposed for wireless systems, has been extensively studied in 60 GHz millimeter wave communications, optical access network design and long haul optical fiber transmission. Experimental demonstrations have shown the advantages of SEFDM in its bandwidth saving, data rate improvement, power efficiency improvement and transmission distance extension compared to conventional orthogonal communication techniques. However, the achieved success of SEFDM is at the cost of complex signal processing for the mitigation of the self-created inter carrier interference (ICI). Thus, a low complexity interference cancellation approach is in urgent need. Recently, deep learning has been applied in optical communication systems to compensate for linear and non-linear distortions in orthogonal frequency division multiplexing (OFDM) signals. The multiple processing layers of deep neural networks (DNN) can simplify signal processing models and can efficiently solve un-deterministic problems. However, there are no reports on the use of deep learning to deal with interference in non-orthogonal signals. DNN can learn complex interference features using backpropagation mechanism. This work will present our investigations on the performance improvement of interference cancellation for the non-orthogonal signal using various deep neural networks. Simulation results show that the interference within SEFDM signals can be mitigated efficiently via using properly designed neural networks. It also indicates a high correlation between neural networks and signal waveforms. It verifies that in order to achieve the optimal performance, all the neurons at each layer have to be connected. Partially connected neural networks cannot learn complete interference and therefore cannot recover signals efficiently. This work paves the way for the research of simplifying neural networks design via signal waveform optimization