A Bi-Layer Optimization Scheme for Enhanced Detection of Indoor Pseudolite Interference Signals

Machine learning is frequently used to detect multipath (MP) and nonline-of-sight (NLOS) signals in indoor pseudolite systems. Signal information redundancy and how to find algorithm’s optimal hyperparameters pose significant challenges to this task. To this end, a bi-layer optimization scheme (BOS) is proposed in this article. In the first layer, a result-data-driven principal component analysis (PCA) adjustment strategy is proposed. This strategy eliminates the correlation among the feature parameters of the original pseudolite signals and constructs a feature space with optimal dimensionality. It is contributing to reducing information redundancy in signals. In the second layer, an enhanced dung beetle optimizer (DBO) is proposed. The algorithm incorporates the good point set, opposition-based learning, and CauchyGauss Mutation strategies, and has been demonstrated to achieve faster convergence and better global optimization capability. It is employed for the adaptive selection of hyperparameters. With BOS optimization, the classification accuracy of the support vector machine (SVM) algorithm improved by 6.0% and 6.1% on the two datasets, respectively, while the classification precision of line-of-sight (LOS) signals improved by an average of 12.3%. This confirms the applicability and practical value of the BOS in indoor pseudolite systems.