AI-Driven Equalization for Compressive Optical Fronthaul in Next-generation Radio Access Network

Abstract

The increasing demands for data throughput and efficiency in next-generation Radio Access Networks (RAN) necessitate innovative solutions for the optical fronthaul. This work presents an AI-driven equalization framework integrated with digital automatic gain control (DAGC) for compressive optical fronthaul systems. The proposed system uniquely performs joint phase and timing recovery while mitigating channel impairments, noise, and compression-induced signal loss. Leveraging a Convolutional Neural Network (CNN) architecture optimised through iterative pruning and quantisation, the model achieves high sparsity and reduced computational complexity. Experimental results demonstrate significant enhancements in Error Vector Magnitude (EVM) over the conventional technique, enabling efficient support for high-order modulation schemes. By addressing key challenges in lossy compression and signal distortion, this approach ensures robust performance, scalability, and low latency in practical 5G and beyond fronthaul scenarios, making it particularly suited for Open RAN architectures. Index Terms—radio access networks, optical fiber communication, digital signal processing, neural networks, data compression