An electromagnetic particle optimization algorithm with FPGA acceleration for robust 6-DOF localization

Abstract Existing optimization methods for electromagnetic localization—such as improved particle swarm optimization, hippopotamus optimization, and grey wolf optimization—suffer from high failure rates in singular configurations due to premature convergence, overreliance on a single best solution, and static parameter settings. To overcome these limitations, this paper presents a new algorithm specifically designed for six-degree-of-freedom electromagnetic positioning. The proposed method decouples the search process into two distinct phases: a global exploration phase that maintains population diversity through dual guidance from a perturbed elite candidate and the mean of a random subgroup, and a local refinement phase that performs adaptive neighborhood searches with iteration-dependent step sizes. Furthermore, the algorithm dynamically adjusts search boundaries based on both the current best estimate and the iteration count, enabling a smooth transition from coarse exploration to fine exploitation. Experimental results demonstrate that the new approach reduces the divergence rate at singular poses from over 50% to just 6%, achieving a global convergence success rate of 98.01%. To meet real-time demands, a three-stage pipelined hardware architecture is also implemented on a field-programmable gate array, accelerating 50 iterations from 46.49 seconds to 4.45 seconds.