Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CSI Compression using Channel Charting

Baptiste Chatelier, Vincent Corlay, Matthieu Crussière, Luc Le Magoarou

TL;DR

This work tackles CSI reporting overhead in FDD MU-MIMO with large antenna arrays by using channel charting (CC) to compress CSIs into low-dimensional chart coordinates. An end-to-end architecture pairs a CC-based encoder with a random Fourier features (RFF) decoder to map chart locations to precoders, optimized with a task-oriented loss to maximize sum-rate. Encoder subsampling reduces learnable parameters by up to ~90x while maintaining performance, enabling very high compression ratios (e.g., $\\gamma \\approx 1024$ at $d=2$). Experiments on realistic DeepMIMO and Sionna data show superior beamforming performance at high compression compared to baselines, highlighting CC’s potential as a practical encoding strategy for task-based CSI compression in MU-MIMO.

Abstract

Reaping the benefits of multi-antenna communication systems in frequency division duplex (FDD) requires channel state information (CSI) reporting from mobile users to the base station (BS). Over the last decades, the amount of CSI to be collected has become very challenging owing to the dramatic increase of the number of antennas at BSs. To mitigate the overhead associated with CSI reporting, compressed CSI techniques have been proposed with the idea of recovering the original CSI at the BS from its compressed version sent by the mobile users. Channel charting is an unsupervised dimensionality reduction method that consists in building a radio-environment map from CSIs. Such a method can be considered in the context of the CSI compression problem, since a chart location is, by definition, a low-dimensional representation of the CSI. In this paper, the performance of channel charting for a task-based CSI compression application is studied. A comparison of the proposed method against baselines on realistic synthetic data is proposed, showing promising results.

CSI Compression using Channel Charting

TL;DR

This work tackles CSI reporting overhead in FDD MU-MIMO with large antenna arrays by using channel charting (CC) to compress CSIs into low-dimensional chart coordinates. An end-to-end architecture pairs a CC-based encoder with a random Fourier features (RFF) decoder to map chart locations to precoders, optimized with a task-oriented loss to maximize sum-rate. Encoder subsampling reduces learnable parameters by up to ~90x while maintaining performance, enabling very high compression ratios (e.g., at ). Experiments on realistic DeepMIMO and Sionna data show superior beamforming performance at high compression compared to baselines, highlighting CC’s potential as a practical encoding strategy for task-based CSI compression in MU-MIMO.

Abstract

Reaping the benefits of multi-antenna communication systems in frequency division duplex (FDD) requires channel state information (CSI) reporting from mobile users to the base station (BS). Over the last decades, the amount of CSI to be collected has become very challenging owing to the dramatic increase of the number of antennas at BSs. To mitigate the overhead associated with CSI reporting, compressed CSI techniques have been proposed with the idea of recovering the original CSI at the BS from its compressed version sent by the mobile users. Channel charting is an unsupervised dimensionality reduction method that consists in building a radio-environment map from CSIs. Such a method can be considered in the context of the CSI compression problem, since a chart location is, by definition, a low-dimensional representation of the CSI. In this paper, the performance of channel charting for a task-based CSI compression application is studied. A comparison of the proposed method against baselines on realistic synthetic data is proposed, showing promising results.
Paper Structure (5 sections, 9 equations, 6 figures, 1 table, 1 algorithm)

This paper contains 5 sections, 9 equations, 6 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Problem formulation
  3. Proposed approach
  4. Experiments
  5. Conclusion and perspectives

Figures (6)

  • Figure 1: Overview of the proposed approach: encoder (left), decoder (right)
  • Figure 2: Encoder LLM22Yassine22 and decoder architectures
  • Figure 3: CDF of the squared cosine similarities: Sionna dataset
  • Figure 4: Squared cosine similarity maps: Sionna dataset
  • Figure 5: Median $\rho$ evolution with the compression ratio
  • ...and 1 more figures