Predictive Line-of-Sight Geometry Compatibility in High-Rate GNSS: Formulation, Validity, and Causal Limits

Abstract

High-rate Global Navigation Satellite System (GNSS) measurements can resolve dynamic motion, but detecting a spectral component does not identify its physical source. This study formulates Predictive Geometry-Normalized Structural Compatibility (PGNSC) as a post-detection, single-receiver test in which a model fitted without one satellite predicts the omitted complex time-differenced carrier-phase (TDCP) response through measured line-of-sight (LOS) geometry. PGNSC is characterised by the open TRANSFORM2 TM3 campaign, comprising 50 Hz GNSS observations and 47 controlled sinusoidal events at 0.5 and 2 Hz with independent accelerometers and additional GNSS receivers. Conventional variometric processing recovered the imposed frequencies; PGNSC then tested whether the detected response generalised through LOS geometry. Because the six-GPS formulation has limited redundancy and high leverage, a GPS+Galileo sensitivity analysis examined numerical robustness. It reduced PRESS fragility and yielded campaign-internal event-versus-quiet area under the receiver operating characteristic curve (AUC) values of 0.959 at 0.5 Hz and 0.9998 at 2 Hz; separate dependence-aware checks on the historical expanded branch also retained strong separation. Algebraically, disturbances within the tested LOS-model space can preserve both fitted-model residuals and leave-one-satellite-out prediction errors. An exchangeable in-model-space verification accordingly produced chance-level discrimination (CPAUC = 0.491; 95% CI, 0.457–0.524). CP quantifies predictive LOS compatibility within the tested campaign; causal attribution requires independent physical evidence beyond LOS geometry.