High Dynamic Range 100G PON Enabled by SOA Preamplifier and Recurrent Neural Networks

In recent years the PON research community has focused on future systems targeting 100 Gb/s/ $\lambda$ and beyond, with digital signal processing seen as a key enabling technology. Spectrally efficient 4-level pulse amplitude modulation (PAM4) is cost-effective solution that exploits ready availability of cheaper, low-bandwidth devices, Semiconductor Optical Amplifiers (SOA) are being investigated receiver preamplifiers to compensate PAM4's high signal-to-noise ratio requirements meet demanding 29 dB loss budget. However, SOA gain saturation-induced patterning distortion concern in context burst-mode signalling, 19.5 loud-soft packet dynamic range expected by most ITU-T 50G standards. this paper we propose recurrent neural network equalisation technique based gated units (GRU-RNN) not only mitigate affecting loud bursts, but also exploit their remarkable effectiveness at compensating non-linear impairments unlock saturated regime. Using such an equaliser demonstrate notation="LaTeX">$ > 28$ system Gb/s PAM4 using compression conjunction GRU-RNN equalisation. We find our proposed similar capabilities Volterra, fully connected network, long short-term memory equalisers, observe feedback-based RNN equalisers more suited varying levels impairment inherent signalling due low input tap requirements. Recognising issues surrounding hardware implementation RNNs, investigate multi-symbol scheme lower feedback latency GRU-RNN. Finally, compare complexities performances according trainable parameters real valued multiplication operations, finding than those networks or elsewhere.