Skip to Main Content
This paper discusses new systems for nonlinear optical recording channels. The systems combine neural network structure into simple detectors such as the multilevel decision-feedback equalizer (MDFE) and the discrete matched filter (DMF). The latter (denoted as DFNE/DMF) provides powerful nonlinear tolerance, while the former (denoted as NMDFE) shows poor tolerance because of conditional training property in the MDFE. When compared with a partial response neural equalizer with maximum-likelihood (PRNE/ML), the proposed DFNE/DMF proves to be very hardware-efficient and able to support high data rates. Simulation results show that the DFNE/DMF increases bloom tolerance up to roughly 30% at a density of S = 6.