Predictive Rate Selection for Ultra-Reliable Communication using Statistical Radio Maps
This paper proposes exploiting the spatial correlation of wireless channel statistics beyond the conventional received signal strength maps by constructing statistical radio maps to predict any relevant channel statistics to assist communications. Specifically, from stored channel samples acquired by previous users in the network, we use Gaussian processes (GPs) to estimate quantiles of the channel distribution at a new position using a non-parametric model. This prior information is then used to select the transmission rate for some target level of reliability. The approach is tested with synthetic data, simulated from urban micro-cell environments, highlighting how the proposed solution helps to reduce the training estimation phase, which is especially attractive for the tight latency constraints inherent to ultra-reliable low-latency (URLLC) deployments.