Mitigating the Impact of Location Uncertainty on Radio Map-Based Predictive Rate Selection via Noisy-Input Gaussian Process

This paper proposes a predictive rate-selection framework based on Gaussian process (GP)-based radio map construction that is robust to location uncertainty. Radio maps are a promising tool for improving communication efficiency in 6G networks. Although they enable the design of location-based maximum transmission rates by exploiting statistical channel information, existing discussions often assume perfect (i.e., noiseless) location information during channel sensing. Since such information must be obtained from positioning systems such as global navigation satellite systems, it inevitably involves positioning errors; this location uncertainty can degrade the reliability of radio map-based wireless systems. To mitigate this issue, we introduce the noisy-input GP (NIGP), which treats location noise as additional output noise by applying a Taylor approximation of the function of interest. Numerical results demonstrate that the proposed NIGP-based design achieves more reliable transmission-rate selection than pure GP and yields higher throughput than path loss-based rate selection.