Online Gaussian Process for Dynamic Radio Map Updating

Radio maps estimate spatial radio-frequency characteristics from spectrum measurements and are essential for applications such as spectrum management and network planning. As spectrum measurements continuously arrive at the monitoring center, conventional batch processing methods become computationally prohibitive as they require reprocessing all accumulated measurements for each radio map update. In this paper, we propose an online Gaussian process (GP) method to effectively update radio maps with dynamically arriving spectrum measurements. Our method employs a sparse approximation strategy based on variational inference to achieve incremental online updates of the GP model by minimizing a hybrid objective function. This function enables our model to update the posterior distribution using new measurements while simultaneously leveraging a memory subset of previous measurements to enhance inference accuracy and reduce error accumulation. Experimental results show our method outperforms existing batch and online baselines in reconstruction accuracy and computational efficiency.