Deep Learning Based Joint Multi-User MISO Power Allocation and Beamforming Design

TL;DR

The paper tackles joint downlink power allocation and beamforming for sum-rate maximization in MU-MISO networks, addressing non-convexity and real-time computation. It introduces NNBF-P, an unsupervised deep neural network that ingests frequency-domain channel data $\mathbf{H}$ and outputs beamforming weights $\tilde{\mathbf{W}}$ and powers $\mathbf{p}$ under the total power constraint $P_{\max}$ to maximize $\sum_{i=1}^N \alpha_i \log(1+\gamma_i)$, as captured by the loss $\mathcal{L}(\boldsymbol{\theta};\mathbf{H}) = -\sum_{i=1}^N \alpha_i \log(1+\gamma_i)$. The architecture builds on stacked BasicBlock modules with 1D convolutions, processing the frequency-domain channel and producing $\tilde{\mathbf{W}}$ and $\mathbf{p}$ through fully connected layers with a Softmax-normalized output for power. Training is unsupervised and end-to-end, using the target objective rather than labeled beamforming solutions, and the approach is shown to outperform baselines such as ZFBF and MMSE, as well as the non-joint NNBF configuration, across diverse channel models and modulation schemes. These results indicate a practical, real-time capable framework for joint resource management in 5G/6G networks, with potential extensions to more complex antenna configurations and dynamic scenario settings.

Abstract

The evolution of fifth generation (5G) wireless communication networks has led to an increased need for wireless resource management solutions that provide higher data rates, wide coverage, low latency, and power efficiency. Yet, many of existing traditional approaches remain non-practical due to computational limitations, and unrealistic presumptions of static network conditions and algorithm initialization dependencies. This creates an important gap between theoretical analysis and real-time processing of algorithms. To bridge this gap, deep learning based techniques offer promising solutions with their representational capabilities for universal function approximation. We propose a novel unsupervised deep learning based joint power allocation and beamforming design for multi-user multiple-input single-output (MU-MISO) system. The objective is to enhance the spectral efficiency by maximizing the sum-rate with the proposed joint design framework, NNBF-P while also offering computationally efficient solution in contrast to conventional approaches. We conduct experiments for diverse settings to compare the performance of NNBF-P with zero-forcing beamforming (ZFBF), minimum mean square error (MMSE) beamforming, and NNBF, which is also our deep learning based beamforming design without joint power allocation scheme. Experiment results demonstrate the superiority of NNBF-P compared to ZFBF, and MMSE while NNBF can have lower performances than MMSE and ZFBF in some experiment settings. It can also demonstrate the effectiveness of joint design framework with respect to NNBF.

Figures (7)

• Figure 1: Downlink massive MIMO where BS transmit data streams to single-antenna UEs with allocated powers on the same time/frequency resources.
• Figure 2: Basic block structure (dashed part) with 2 input channels and 16 output channels, $\mathrm{BB (2,16)}$.
• Figure 3: Deep neural network architecture with joint training procedure.
• Figure 4: Performance comparison of NNBF and NNBF-P with baseline methods ZFBF and MMSE when the channel delay profile is TDL-C with delay spread of $300$ ns and modulation type is 16QAM for $M=4$, $N=4$.
• Figure 5: Performance comparison of NNBF and NNBF-P with baseline methods ZFBF and MMSE when the channel delay profile is TDL-C with delay spread of $300$ ns and modulation type is 16QAM for different antenna configurations: (a) $M=8$, $N=4$ (b) $M=16$, $N=4$.