Channel Knowledge Map Updating with Machine Unlearning

Dynamic environmental changes such as urban development and construction activities render existing channel knowledge map obsolete, necessitating efficient update mechanisms. Traditional model updating approaches face significant challenges. For example, complete retraining requires extensive data collection. The channel knowledge map updating strategy proposed in this paper classifies transmitter-receiver pairs into Line-of-Sight and Non-Line-of-Sight sets based on environmental changes. Since Line-of-Sight propagation remains relatively stable during environmental transitions, only Non-Line-of-Sight data requires updating, substantially reducing data collection requirements. The machine unlearning mechanism systematically removes outdated environmental information while preserving knowledge of unchanged features, followed by targeted fine-tuning with new environmental data. Numerical results demonstrate that the proposed strategy achieves good prediction accuracy and reduces data collection requirement.