Near-Field Nulling Control Beamfocusing Optimization for Multi-User Interference Suppression

Abstract

This paper presents comprehensive full-wave simulation-based studies of near-field beam radiation patterns for large-scale arrays, accounting for realistic electromagnetic wave characteristics, heterogeneous element radiation patterns, and array element interactions. These simulations thoroughly investigate and illustrate the radiation behaviors of antenna arrays at different observation distances. To leverage the advantages offered by distance-dependent radiation patterns in the near-field, we consider two nulling control beamfocusing algorithms to effectively mitigate multi-user interference (MUI) in massive multiple-input multiple-output (mMIMO) systems by achieving considerable focusing gain differences between the desired and interference locations. Firstly, a linear constraint minimum variance (LCMV) scheme to effectively control radiation nulls in the Fresnel region is developed. By adjusting the array feeding magnitudes and phase shifters, an average gain difference of 29.2 dB between desired and undesired users can be achieved, with minimal gain degradation of 0.4 dB at the desired user compared to the maximum directivity beamfocusing scheme. Moreover, a constant-modulus beamfocusing scheme based on a perturbation-based nulling control beamfocusing algorithm employing particle swarm optimization is proposed. Using only phase shifters, an average gain difference of 26.1 dB between desired and undesired users can be achieved. Iterative full-wave simulations are conducted to investigate how the achievable beamfocusing gain difference varies with different desired and interference user locations. Finally, a deep neural network (DNN) is trained for MUI suppression based on the LCMV-generated beamfocusing vectors. The model achieves a phase error of less than 0.021 radians and a magnitude error of 0.17 dB in the predicted feeding weights. The resulting near-field beam patterns using the LCMV-based vector and the DNN-predicted vector show good agreement.