Neural Network-Based Successive Interference Cancellation for Non-Linear Bandlimited Channels

TL;DR

This work tackles achieving near-JDD information rates on bandlimited nonlinear channels with manageable receiver complexity. It introduces model-based NN equalizers that implement SIC using periodically time-varying Bidirectional RNNs, mirroring forward-backward state recursions to approximate APPs stage-by-stage. The results on short-reach fiber links with square-law detection show NN-SIC can nearly reach JDD performance, delivering substantial spectral efficiency and energy gains (up to about 3 dB) while reducing complexity relative to mismatched FBA and Gibbs sampling. The approach holds promise for practical direct-detection optical links and highlights avenues for coded modulation integration, robustness testing, and potential hardware implementations.

Abstract

Reliable communication over bandlimited and non-linear channels usually requires equalization to simplify receiver processing. Equalizers that perform joint detection and decoding (JDD) achieve the highest information rates but are often too complex to implement. To address this challenge, model-based neural network (NN) equalizers that perform successive interference cancellation (SIC) are shown to approach JDD information rates for bandlimited channels with a memoryless nonlinearity and additive white Gaussian noise. The NNs are chosen to have a periodically time-varying and recurrent structure that imitates the forward-backward algorithm (FBA) in every SIC stage. Simulations for short-haul fiber-optic links with square-law detection show that NN-SIC nearly doubles current spectral efficiencies, and bipolar or complex-valued modulations achieve energy gains of up to 3dB compared to state-of-the-art intensity modulation. Moreover, NN-SIC is considerably less complex than equalizers that perform JDD, mismatched FBA processing, and Gibbs sampling.

Figures (16)

• Figure 1: Bandlimited channel with a memoryless nonlinear device and additive noise.
• Figure 2: Nonlinear functions for a (a) PA: $\mathop{\mathrm{\xi}}\nolimits\lparen*\rparen{|x|} = |x|/\sqrt[\leftroot{1}\uproot{2}4]{(1+|x|^{4})}$ and (b) SLD: $\mathop{\mathrm{\xi}}\nolimits\lparen*\rparen{x} = \lvert x\rvert^2$.
• Figure 3: Magnitude-squared response of an SSMF channel with parameters in Tab. \ref{['tab:simparams']}, length 30km and $g(t) \propto \mathrm{sinc}(Bt) * g_\text{SSMF}(t)$; circles show samples with $N_\text{sim} = 2$.
• Figure 4: SIC with $S=3$ stages and $n=15$ input symbols.
• Figure 5: SIC receiver with SDD for each stage.