Where to Perform Channel Measurements for CKM Construction: A Random Field Theory Analysis

What the math gives to ML

The paper provides a constructive framework for choosing spatial measurements by minimizing integrated reconstruction error under a Gaussian random-field model. Its transferable asset is not the wireless application, but the ordinary-kriging error formula and the resulting uncertainty-aware greedy placement rule, which can be used to decide where a coordinate neural field should acquire labels or expensive simulator evaluations. The most direct ML use is active training of neural fields or neural operators: fit a spatial covariance or variogram from current residuals, estimate kriging uncertainty over candidate coordinates, and query locations with the largest expected reconstruction error rather than using uniform samples.