Transformer-Driven Neural Beamforming with Imperfect CSI in Urban Macro Wireless Channels

TL;DR

The paper tackles beamforming for MU-SIMO in urban macro channels when CSI is imperfect. It introduces NNBF, an unsupervised deep-learning framework that fuses depthwise separable convolutions with transformer-based stacked multi-channel attention to map imperfect CSI $\hat{\mathbf{H}}$ to beamforming weights $\mathbf{W}_{nn}$ by minimizing the negative sum-rate loss $\mathcal{L} = -\sum_{i=1}^N \alpha_i \log(1+\gamma_i)$ under the constraint $\mathrm{tr}(\mathbf{W}^H \mathbf{W}) \le N$, where $\gamma_i$ is the per-user SINR. The architecture comprises a Convolutional Residual Network for local feature extraction and a Stacked Multi-Channel Attention module to capture intra- and inter-channel dependencies, followed by a final convolutional head that outputs $\mathbf{W}_{nn}$. Experimental results show that NNBF consistently surpasses zero-forcing and MMSE beamforming across a wide SNR range, mobility scenarios, and modulation orders, achieving higher spectral efficiency and lower BLER while remaining robust to CSI imperfections. This indicates practical potential for scalable, data-driven beamforming in dense urban deployments where perfect CSI is unattainable.

Abstract

The literature is abundant with methodologies focusing on using transformer architectures due to their prominence in wireless signal processing and their capability to capture long-range dependencies via attention mechanisms. In particular, depthwise separable convolutions enhance parameter efficiency for the process of high-dimensional data characteristics of MIMO systems. In this work, we introduce a novel unsupervised deep learning framework that integrates depthwise separable convolutions and transformers to generate beamforming weights under imperfect channel state information (CSI) for a multi-user single-input multiple-output (MU-SIMO) system in dense urban environments. The primary goal is to enhance throughput by maximizing sum-rate while ensuring reliable communication. Spectral efficiency and block error rate (BLER) are considered as performance metrics. Experiments are carried out under various conditions to compare the performance of the proposed NNBF framework against baseline methods zero-forcing beamforming (ZFBF) and minimum mean square error (MMSE) beamforming. Experimental results demonstrate the superiority of the proposed framework over the baseline techniques.

Figures (8)

• Figure 1: Uplink multi-user SIMO system in a dense urban environment, where single-antenna UEs transmit data streams on the same time/frequency resources and M-antenna BS apply digital beamforming on the received signal $\mathbf{y}$.
• Figure 2: The block diagram of system model.
• Figure 3: Deep neural network architecture.
• Figure 4: Stacked multi-channel attention module.
• Figure 5: Performance comparison of NNBF with baseline methods ZFBF and MMSE when the coding rate can be $\frac{1}{2}$ and $\frac{3}{4}$. The results are shown for a system with $M=8$, $N=2$, 4QAM modulation, and for stationary UEs.