Automated Source Selection for Enhanced Inter-gNB Transfer Learning Through Modeling Propagation Environments

With the rapid advancement and widespread adoption of 5G technologies, the deployment of new gNBs has become increasingly common for purposes such as coverage extension, capacity enhancement, and improved service quality. In such scenarios, numerous artificial intelligence (AI) predictive models must be either trained from scratch or fine-tuned to accommodate the new operational environments of newly deployed gNBs. Both approaches require substantial amounts of data, which may not yet be available at a new gNB. Transfer learning (TL) emerges as an effective solution in this context. However, a critical challenge in TL is deciding which gNB should be selected for the transfer. Addressing this question is essential to ensure the performance of TL and to avoid negative TL (the scenario where TL degrades system performance). In this paper, we present the first attempt to formulate the problem of node selection in TL between gNBs. Powered by advanced Raytracing techniques, we introduce a novel framework for source node selection based on modeling and characterizing the propagation environment of base stations. The experimental results demonstrate the efficacy of the proposed technique across various environments. A 30% reduction in the packet error rate has been achieved by adopting the proposed approach for source node selection, which in turn, has a clear advantage for energy saving at base stations.