Photonic Streaming Convolutional Acceleration for Sub-microsecond Interference Recognition in 6G Radio Access Networks

Abstract

The escalating complexity of wireless environment in next-generation radio access networks (RANs) has made interference a primary constraint on network efficiency and reliability. Low latency and accurate interference recognition is thus a critical prerequisite at base stations. Conventional electronic approaches face inherent architectural bottlenecks that hinder real-time processing capabilities. Here, for the first time, we adapt the photonic streaming convolutional computing architecture previously demonstrated in image and general signal processing to the task of interference identification in 6th Generation mobile communication (6G) Radio Access Network (RAN). Leveraging the high parallelism of light, our microcomb-based optical convolution system is able to directly process massive interference optically and adaptively. The experimental results demonstrate 88% accuracy for 10 types of interference classified from PRB (Physical Resource Block) power statistics of China Mobile’s live networks, which are acquired via the OMC (Operation and Maintenance Center) northbound interface. The total hardware end-to-end latency of the photonic streaming convolutional accelerator is approximately 220 ns, with a core optical computing latency as low as 20 ns, indicating three orders of magnitude faster than electronic counterparts. System-level analyses of latency, power consumption, scalability, and fidelity collectively confirm the deployment potential of this architecture. This work presents the first application of photonic streaming convolution acceleration for wireless interference management, providing key experimental evidence for the deployment of photonic computing in next-generation edge networks.