A Causal Framework for Quantifying Task-Driven Selection Bias in Historical Deployment of Integrated Underwater Communication and Positioning Networks

Abstract

To address task-driven selection bias in historical deployment records, this study proposes a structural-causal-model-based framework for quantifying bias in the utility assessment and deployment-effect estimation of integrated underwater communication and positioning networks. Four datasets were constructed under unbiased, communication-dominant, positioning-dominant, and joint-biased sampling mechanisms, and double machine learning (DML) was adopted to analyze overall utility and its communication/localization components under decision factors. The simulation results demonstrate that communication-dominated data overestimates overall utility by 23.5%, while location-dominated data underestimates by 21.4%. CATE analysis further identifies noise spectral level as the strongest effect modifier (feature importance 0.76). The sea trial results show close agreement between simulated CRLB and measured RMSE, which supports the physical plausibility of the positioning utility model and the WOA23-based simulation pipeline underlying the causal analysis.

Bibliographic Information