Aging-Resistant Wideband Precoding in 5G and Beyond Using 3D Convolutional Neural Networks

TL;DR

This work tackles wideband MIMO precoding under CSI acquisition impairments by proposing a two-part neural architecture: a neural compensator that treats channel aging and CE/CC errors as denoising noise to recover a clean CSI, and a 3D-CNN neural precoder that maps the compensated CSI to a wideband precoding tensor across all subcarriers. The approach is validated within a unified TDD/FDD 5G Advanced CSI framework and trained in a differentiable simulation environment, incorporating on-the-fly data generation and BER-based optimization. The proposed framework yields significant improvements over a regularized zero-forcing baseline, particularly under severe aging and estimation errors, while maintaining low additional complexity and scalability to larger bandwidths. This architecture thereby enables more robust, high-rate communications in future 5G/6G systems with practical CSI impairments.

Abstract

To meet the ever-increasing demand for higher data rates, 5G and 6G technologies are shifting transceivers to higher carrier frequencies, to support wider bandwidths and more antenna elements. Nevertheless, this solution poses several key challenges: i) increasing the carrier frequency and bandwidth leads to greater channel frequency selectivity in time and frequency domains, and ii) the greater the number of antennas the greater the the pilot overhead for channel estimation and the more prohibitively complex it becomes to determine the optimal precoding matrix. This paper presents two deep-learning frameworks to solve these issues. Firstly, we propose a 3D convolutional neural network (CNN) that is based on image super-resolution and captures the correlations between the transmitting and receiving antennas and the frequency domains to combat frequency selectivity. Secondly, we devise a deep learning-based framework to combat the time selectivity of the channel that treats channel aging as a distortion that can be mitigated through deep learning-based image restoration techniques. Simulation results show that combining both frameworks leads to a significant improvement in performance compared to existing techniques with little increase in complexity.

Figures (9)

• Figure 1: Block diagrams and time charts related to TDD and FDD CSI acquisition frameworks: (a) block diagram of 5G data transmission and CSI acquisition at the PHY layer in TDD mode; (b) time chart of CSI acquisition in TDD mode for $T_\mathrm{CSI} = 2$ and $T_\mathrm{offset} = 0$; (c) time chart of CSI acquisition in FDD mode for $T_\mathrm{CSI} = 4$ and $T_\mathrm{offset} = 0$; and (d) block diagram proposed for data transmission and enhanced CSI acquisition at the PHY layer in FDD mode as per the related study item of rel'18 (CSI feedback enhancement) 3gpp_ran114.
• Figure 2: Block diagram of the implemented model (abstracted system model).
• Figure 3: Kronecker pilot pattern for $N_\mathrm{FFT} = 16, \mathbf{k}_\mathrm{guard}=[0,0], N_s=2$, and $\mathbf{L}=(2, 11)$. A zero power, i.e., $0$, symbol is transmitted on the masked REs.
• Figure 4: Subsystems in the proposed solution.
• Figure 5: NN architectures: (a) residual block; (b) proposed neural compensator to mitigate the CSI acquisition impairments, i.e., aging and CE/CC errors, based on Proposition \ref{['prop:prediction_as_denoising']}; and (c) proposed neural precoder based on 3D convolutional layers. The tensors' dimensions are written in black, whereas the CNN parameters (i.e., kernel size and number of filters) are written in blue.

Theorems & Definitions (2)

• Proposition 1
• proof