Semantic embeddings improve beam selection from fewer pilot signals

Semantic CSI Feedback for Beam Selection: When Task-Aware Embeddings from Sparse Pilots Outperform Full-Bandwidth Reconstruction

Artificial IntelligenceMachine Learning

Summary

Sending detailed channel information from user devices to base stations is usually done by reconstructing the entire signal, which can be large and inefficient. The authors show that sending a small, carefully learned summary focused just on picking the best beam can work better than sending full details. Their method uses just 8 numbers derived from a smaller set of pilot signals to accurately predict beams, even better than traditional methods with much more data. This proves that focusing communication on the actual task, rather than on perfectly reconstructing signals, can save resources and improve performance.

What this means in practice

  • For cellular network engineers: Use semantic embeddings from sparse pilot data to improve beam selection accuracy and reduce feedback overhead in massive MIMO systems.
  • For wireless device developers: Implement compact task-specific feedback in user equipment to efficiently convey channel information relevant for beamforming decisions.

Authors

Cristian J. Vaca-Rubio, Konstantinos Vandikas, Aneta Vulgarakis Feljan

Abstract

Classical CSI feedback in FDD massive MIMO transmits a compressed reconstruction of the channel, optimizing fidelity to the original signal regardless of the downstream task. We propose a semantic communication perspective: instead of reconstructing the channel, the UE transmits a learned \emph{semantic embedding} optimized end-to-end for beam selection at the gNB. Comparing reconstruction-oriented feedback (CsiNet) against task-aware semantic feedback across two input domains and three observation scenarios, we show that a semantic embedding of just $d=8$ real values from only 43 NR CSI-RS pilots in the angular-delay domain achieves the highest beam prediction accuracy, outperforming every method with access to the full 512-subcarrier channel. The key insight is that beam-relevant information is intrinsically low-dimensional: the semantic encoder learns to discard reconstruction-irrelevant structure and retain only a compact representation that is relevant to beam selection, realizing the core principle of semantic communication: transmit the intent, not the signal.
View PDFOpen arXiv