Receiver-Side Reconstruction of Vessel Roll and Pitch Under Simulated Packet Loss and Delay: Accuracy–Freshness Trade-Offs

Vessel roll and pitch measurements transmitted over communication links may be delayed or lost. We compared six receiver-side reconstruction methods: hold, linear interpolation, linear extrapolation, adaptive selection, constant-velocity Kalman filtering, and Rauch–Tung–Striebel smoothing. Recorded motion data from MITO 31 and Seafar were processed under three synthetic communication scenarios: independent 10% packet loss with Gamma delays (S1), bursty 10% loss with time-varying delays (S2), and bursty 20% loss (S3). Reconstruction used only packets received before reporting deadlines, with permitted lags of 0–1 s. Parameters were fixed on training data; 50 paired channel realizations were evaluated per segment. Increasing lag from 0.2 to 0.5 s reduced interpolation error by 40–77%. Under S1 at 0.5 s, smoothing reduced MITO 31 roll and pitch errors by 38% and 52% relative to interpolation. Under S3, smoothing increased roll error by 30%, while adaptive selection improved on interpolation by 4.3% for roll and 3.5% for pitch. Doubling bursty loss increased interpolation error by 33–47%. Kalman tuning transferred poorly between vessels. No method consistently dominated: smoothing benefited from timely post-target observations but deteriorated across prolonged bursts. Interpolation withhold fallback remained robust under bursty loss. Method selection should reflect vessel motion and measured link characteristics.