AI-Aided Opportunistic Quantization for Channel Aging Mitigation and Fronthaul Overhead Reduction in O-RAN Systems

Modern base station (BS) employ distributed architecture, wherein the overall processing is divided between a radio unit (RU) and a distributed unit (DU), that are connected by a fronthaul link. In open-RAN (O-RAN) 7.2 split architecture, RU performs only a few simple operations, while most of the BS processing happens at the DU, providing the benefits of centralized computing and control. A key challenge in 7.2 split is to limit the overhead on the fronthaul link within acceptable limits. To address this issue, a common solution is to perform combining operation on the received signals at RU antennas, and send low-dimensional signals to DU over fronthaul. The combining matrix used at the RU is typically based on older/aged CSI received from DU, resulting in performance degradation, which is referred as channel aging problem. An alternate approach to limit the fronthaul overhead is to quantize the frequency-domain (i.e., post-FFT) samples at RU to fewer bits (say, 3 or 4 bits) and send the low-resolution signals over fronthaul. Since this approach does not use aged channels, its performance is not affected by channel aging, but is mainly limited by quantization noise. It is recently shown that, under certain conditions, quantization approach achieves superior error performance compared to combining, while also maintaining lower overhead. Motivated by this observation, the present work proposes an AI-based solution to opportunistically switch from combining to quantization under favorable conditions, thereby mitigating channel aging while also reducing fronthaul overhead.