Fiber-length agnostic all-optical equalisation using silicon photonics neural networks
Abstract
Signal integrity in high-speed intensity-modulation direct-detection (IM/DD) fibre links is limited by chromatic dispersion, Kerr-induced nonlinearities, and the frequency-dependent response of the transmitter–receiver chain. We demonstrate an integrated photonic reservoir with a programmable optical readout for pre-detection equalisation of a 28 Gbps on–off keying (OOK) signal. Using a fixed reservoir architecture and delay configuration, experimental bit-error rates as low as 4×10-7 are achieved, remaining below the generic forward-error-correction (GFEC) threshold of 5.8×10-5 for fibre lengths up to 50 km at a launch power of 10 dBm and launch powers up to 17 dBm at a fiber length of 25 km. Numerical simulations are used to assess the impact of data-rate mismatch relative to the reservoir design point. Robust performance is obtained without reconfiguration of the photonic hardware, requiring only optimisation of the readout weights for different operating conditions.
Fields of Study & Subjects
Neural Networks and Reservoir Computing
Optical Network Technologies
Advanced Photonic Communication Systems
Publication details
- Journal / source
- AI and Optical Data Sciences
- Pages
- 43–43
- Publication date
- 4 March 2026
- Language
- EN