The application belongs to the technical field of array
signal processing, and relates to a three-dimensional polarized array near-field source parameter
estimation method based on a
quaternion neural network, which comprises three-dimensional polarized array near-field source receiving
signal modeling, establishment of a near-field receiving
signal model of a three-dimensional co-point orthogonal
magnetic loop dipole array, and
coupling of receiving direction vectors of the three-dimensional COLD array with three-dimensional space parameters and two-dimensional polarization domain parameters; network input data is constructed;
training set data is constructed; a
quaternion neural network architecture is constructed and trained. The application uses the
quaternion neural network to realize space-polarization domain parameter
estimation of a near-field source in the three-dimensional COLD array, extracts multi-dimensional signal characteristics of the three-dimensional COLD array by using the quaternion neural network, and decouples space and polarization domain parameters by constructing polarization coherent and non-polarization coherent two-way quaternion
covariance matrix data, so that the training cost growth problem caused by sample combination in multi-dimensional parameter
estimation is greatly reduced.