A DSP-Assisted IF-Over-Fiber System With Extreme Learning Machine-Based Linearization

This work proposes a densely allocated DSP-supported intermediate-frequency-over-fiber (IFoF) system with real-value single-hidden layer feedforward network (SLFN) post-equalization. The SLFN is trained using downsampled multichannel training signals (DMTSs) and the extreme learning machine (ELM) algorithm. The transmission of multichannel orthogonal frequency-division multiplexing (OFDM)-based IF signals over a 12.4-km direct-detection optical link is demonstrated, achieving adjacent channel leakage ratio (ACLR) improvements of up to 10 dB and significant reductions in error vector magnitude (EVM) due to the ELM-SLFN linearization scheme.