Study on Downlink CSI compression: Are Neural Networks the Only Solution?

TL;DR

The paper tackles the uplink overhead of downlink CSI feedback in FDD massive MIMO by evaluating a PCA-based CSI compression approach against AI/ML methods that use auto-encoders. It systematically analyzes two channel representations, Angular-Delay $($AD$)$ and Eigenvector $($EV$)$, and provides explicit PCA pipelines, overhead formulas, and reconstruction strategies, comparing them to CSINet and EVCSINet architectures. Results show that PCA can achieve reconstruction performance comparable to neural networks while avoiding generalization across channel scenarios and inter-vendor interoperability concerns, as the compression and reconstruction matrices are per-instance. This suggests PCA-based CSI feedback as a practical alternative for real deployments, with further work planned to test 3GPP channel models and to explore non-linear dimensionality reduction techniques such as manifold learning.

Abstract

Massive Multi Input Multi Output (MIMO) systems enable higher data rates in the downlink (DL) with spatial multiplexing achieved by forming narrow beams. The higher DL data rates are achieved by effective implementation of spatial multiplexing and beamforming which is subject to availability of DL channel state information (CSI) at the base station. For Frequency Division Duplexing (FDD) systems, the DL CSI has to be transmitted by User Equipment (UE) to the gNB and it constitutes a significant overhead which scales with the number of transmitter antennas and the granularity of the CSI. To address the overhead issue, AI/ML methods using auto-encoders have been investigated, where an encoder neural network model at the UE compresses the CSI and a decoder neural network model at the gNB reconstructs it. However, the use of AI/ML methods has a number of challenges related to (1) model complexity, (2) model generalization across channel scenarios and (3) inter-vendor compatibility of the two sides of the model. In this work, we investigate a more traditional dimensionality reduction method that uses Principal Component Analysis (PCA) and therefore does not suffer from the above challenges. Simulation results show that PCA based CSI compression actually achieves comparable reconstruction performance to commonly used deep neural networks based models.

Figures (2)

• Figure 1: AI/ML induced CSI compression framework over-view.
• Figure 2: Percentage of variance captured by each principal component for CDLA300 channel represented as (a) Angle Delay Domain data and (b) Eignevector data