Aldaya, Ivan [UNESP]Giacoumidis, EliasTsokanos, AthanasiosJarajreh, MutsamWen, YannuoWei, JinlongCampuzano, GabrielAbbade, Marcelo L. F. [UNESP]Barry, Liam P.2021-06-252021-06-252020-10-15Optics Letters. Washington: Optical Soc Amer, v. 45, n. 20, p. 5820-5823, 2020.0146-9592http://hdl.handle.net/11449/208891A novel nonlinear equalizer based on a multiple-input multiple-output (MIMO) deep neural network (DNN) is proposed and experimentally demonstrated for compensation of inter-subcarrier nonlinearities in a 40 Gb/s coherent optical orthogonal frequency division multiplexing system. Experimental results reveal that MIMO-DNN can extend the power margin by 4 dB at 2000 km of standard single-mode fiber transmission when compared to linear compensation or conventional single-input single-output DNN. It is also found that MIMO-DNN outperforms digital back propagation by increasing up to 1 dB the effective Q-factor and reducing by a factor of three the computational cost. (C) 2020 Optical Society of America5820-5823engArtigo10.1364/OL.403778WOS:000581198200047