Neural Network Equalizers and Successive Interference Cancellation for Bandlimited Channels with a Nonlinearity

TL;DR

Neural networks inspired by the forward-backward algorithm (FBA) are used as equalizers for bandlimited channels with a memoryless nonlinearity to approach the information rates of joint detection and decoding with considerably less complexity than JDD and other existing equalizers.

Abstract

Neural networks (NNs) inspired by the forward-backward algorithm (FBA) are used as equalizers for bandlimited channels with a memoryless nonlinearity. The NN-equalizers are combined with successive interference cancellation (SIC) to approach the information rates of joint detection and decoding (JDD) with considerably less complexity than JDD and other existing equalizers. Simulations for short-haul optical fiber links with square-law detection illustrate the gains.

Figures (9)

• Figure 1: Bandlimited channel with a memoryless non-linearity and additive noise.
• Figure 2: Encoding with $S=3$ stages, $N=5$ and $n=15$ input symbols.
• Figure 3: SIC receiver with SDD for each stage.
• Figure 4: Bidirectional time-varying RNN for SIC stage $s$.
• Figure 5: 4-ASK, $L_\text{fib}=30km$. The rates increase with stages $S=1,\ldots,4$; the lower curve of a certain style marks SDD; the upper curve marks $S=4$.