Centralized Digital Predistortion in 6G: Distributed Task Offloading and Scheduling for Complexity-Reduced PA Linearization

To improve power efficiency while using ultra-wide bandwidth, digital predistortion (DPD) is essential in 6G base station (BS) design to linearize the nonlinearity distortion of power amplifiers (PA). However, traditional DPD methods rely exclusively on localized PA distortion estimation in distributed BSs, which could become extremely complex and costly due to the increased number of BSs and growing scale of antenna arrays in 6G. Therefore, simplified and centralized DPD design strategies tailored to 6G are urgently required to reduce the DPD complexity in distributed BSs and the exponentially in-creased costs throughout the network. In this paper, we propose a centralized DPD scheme by offloading and scheduling the distributed PA distortion estimation tasks from local BSs to a centralized training device so that the overall complexity of the PA linearization is significantly reduced. To avoid unnecessary PA data transmission, lightweight onsite analysis is designed at each BS by comparing the PA operating conditions, which can minimize the communication resource wastage for task offloading. Long-term distortion analysis is further conducted at the training device to support intelligent scheduling of distortion re-estimation and model sharing among similar PAs. Simulation results demonstrate that the proposed approach significantly achieves efficient PA distortion estimation and linearization without sacrificing accuracy.