Reservoir Computing for Short-reach Optical Communication

We reviewed two promising NN based equalizers for IM/DD transmission links:
1. Time delay Feedforward Neural Networks
   • Trainable with backpropagation but with size growing linearly with the required memory
2. Reservoir computing
   • Single ridge regression training but potentially more hardware demanding (high redundancy)

Both schemes showed good improvement especially in conjunction with a novel spectrally sliced receiver proposed

Simplified NN based equalizers are promising tools for extending the reach of IM/DD transmission systems