An Initialization Method for Cellular Signals of Opportunity/MEMS-INS Integrated System: Eliminating Dependence on GNSS

In global navigation satellite system (GNSS)-denied environments, opportunistic cellular signals have emerged as a promising alternative positioning source, and their integration with microelectromechanical-system-based inertial navigation systems (MEMS-INS) enables continuous low-cost navigation. The integrated system requires accurate initial position and attitude information, and its performance depends strongly on the initialization accuracy. However, existing cellular/MEMS-INS initialization approaches typically rely on GNSS/INS assistance, while directly using magnetometer- and accelerometer-derived attitude may introduce non-negligible errors and degrade subsequent Extended Kalman Filter (EKF) performance. To overcome these limitations, this paper proposes a two-stage GNSS-free initialization framework based on pseudorange-difference measurements. In the first stage, a pseudorange-difference two-step weighted least squares (PD-TWLS) method provides coarse initial position, while attitude estimates are calculated from the outputs of the magnetometer and accelerometer. In the second stage, an INS-aided backward recursion (IABR) algorithm refines these estimates by jointly exploiting inertial error propagation and pseudorange-difference observations. Simulation results demonstrate the robustness of the proposed method under different base-station geometries, observation noise levels, and initial heading errors. Field experiments further confirm that EKF positioning initialized by the proposed IABR achieves root-mean-square horizontal errors of 4.8 m in an open area and 10.5 m in a signal-occluded environment, comparable to GNSS-assisted initialization. The results show that the proposed method provides accurate, robust, and dynamically adaptive initialization for low-cost cellular/MEMS-INS integrated systems.