Wideband Delta-Sigma Radio-Over-Fiber Embedding a Sparse Nonlinear ISI Model Derived From Neural Networks
This paper presents a novel sparse nonlinear intersymbol interference model (SnISIM) for delta-sigma radio-over-fiber (DS-RoF) systems, designed to address severe pulse distortion caused by narrowband bandpass filters in wideband millimeter-wave applications. SnISIM is derived from a filtered time-delayed neural network (FTDNN) and optimized through structured pruning and compressive sensing, enabling high modeling accuracy with minimal computational complexity. Unlike conventional pulse distortion models (PDMs), the proposed approach efficiently captures higher-order nonlinearities without activation functions, reducing complexity while improving distortion compensation. Experimental validation using a 400-MHz 256-QAM OFDM signal over a 10-km fiber demonstrates that SnISIM-embedded DS-RoF achieves substantial gains in signal quality—up to 1.9 dB error vector magnitude (EVM) improvement for real-valued models and 0.7 dB for complex-valued models—while meeting stringent 5G requirements for EVM and adjacent channel leakage ratio (ACLR). To the best of our knowledge, this is the first DS-RoF architecture to simultaneously satisfy both in-band and out-of-band specifications for such wideband signals, highlighting its potential for cost-effective deployment in beyond-5G distributed MIMO systems.