New methods for moving antennas in wireless systems

Researchers developed ways to position antennas and manage signals to improve data rates when channel information is imperfect.

New methods help manage multiuser movable antenna (MA) systems. These methods work even when channel state information (CSI) is imperfect.

The work uses a discrete prolate spheroidal sequence (DPSS)-based a posteriori channel model. This model addresses the mismatch between spatial statistics and the discrete Fourier transform (DFT) basis. It exploits the spatial covariance structure of MA channels to model uncertainty.

An iterative algorithm maximizes the expected sum rate. It uses a low-complexity position replacement scheme to handle antenna positioning and precoding. A reinforcement learning framework uses tensor equivariance (TE) to improve efficiency and scalability.

The framework uses pairwise user-interaction and CSI-conditioned equivariant attention mechanisms. These encode channel statistics and fuse them with available CSI. Distance-biased spatial attention incorporates the geometry of candidate positions into attention weighting.

Simulation results show improved sum-rate performance and reduced online runtime. The learning framework generalizes across different numbers of users and MAs without retraining. It achieves near-optimal performance under perfect CSI.

Key facts

  • The proposed methods improve sum-rate performance and reduce online runtime under imperfect CSI.
  • The learning framework works across different numbers of users and MAs without retraining.
  • The methods achieve near-optimal performance under perfect CSI by matching the exhaustive-search method.

All sections

Edition