Engineering-Constrained Mixed-Encoding Genetic Algorithm Optimization of Bias Coils for Cold Atom Gravimeters

The sensitivity of cold atom gravimeters is strongly governed by the spatial uniformity of their internal bias magnetic field. This paper presents an engineering-constrained optimization framework based on a mixed-encoding genetic algorithm for designing a coaxial three-coil bias magnetic field system with configurable uniform field regions. Unlike conventional analytical methods that produce uniform fields inherently centered at the coil geometry, the proposed approach allows the target uniform region to be specified according to the actual vacuum chamber layout. The genetic algorithm employs a mixed-encoding scheme—real-valued for positions, integer for turns, and binary for current directions—to handle the heterogeneous variable types arising from engineering constraints. The axial magnetic field uniformity was used as the fitness function, producing a design with non-uniformity below 1% in the specified region. To verify the fabricated coils, the magnetic field distribution inside the vacuum chamber was measured in situ using scanning Raman spectroscopy. Experimental results showed good agreement with the GA-predicted distribution, with a measured non-uniformity of approximately 0.96%. This study demonstrates that the mixed-encoding GA framework provides an effective and configurable solution for bias magnetic field design in cold atom quantum sensors, and the method can be extended to other quantum devices requiring large-scale uniform magnetic fields.