On Purely Data-Driven Massive MIMO Detectors

2024-01-15

arXiv.org

AI Summary

Existing learning-based MIMO detection methods for massive MIMO suffer from poor flexibility and high computational complexity—often relying on channel/noise priors or iterative architectures. To address this, we propose ChannelNet: the first fully data-driven, end-to-end MIMO detector requiring no prior physical model. Its core innovation lies in embedding the channel matrix as a learnable linear layer—rather than treating it as raw input—thereby overcoming high-dimensional modeling bottlenecks and achieving low O(NₜNᵣ) detection complexity. We theoretically prove that ChannelNet is a permutation-equivariant universal approximator for probability distributions. Experiments demonstrate that ChannelNet matches or surpasses state-of-the-art detectors across diverse channel conditions, while offering superior efficiency, robustness, and generalization. This work provides the first empirical validation of the feasibility of purely data-driven paradigms for wireless physical-layer detection tasks.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

To enhance the performance of massive multi-input multi-output (MIMO) detection using deep learning, prior research primarily adopts a model-driven methodology, integrating deep neural networks (DNNs) with traditional iterative detectors. Despite these efforts, achieving a purely data-driven detector has remained elusive, primarily due to the inherent complexities arising from the problem's high dimensionality. This paper introduces ChannelNet, a simple yet effective purely data-driven massive MIMO detector. ChannelNet embeds the channel matrix into the network as linear layers rather than viewing it as input, enabling scalability to massive MIMO scenarios. ChannelNet is computationally efficient and has a computational complexity of $mathcal{O}(N_t N_r)$, where $N_t$ and $N_r$ represent the numbers of transmit and receive antennas, respectively. Despite the low computation complexity, ChannelNet demonstrates robust empirical performance, matching or surpassing state-of-the-art detectors in various scenarios. In addition, theoretical insights establish ChannelNet as a universal approximator in probability for any continuous permutation-equivariant functions. ChannelNet demonstrates that designing deep learning based massive MIMO detectors can be purely data-driven and free from the constraints posed by the conventional iterative frameworks as well as the channel and noise distribution models.

Problem

• Overcoming high dimensionality in learning-based MIMO detectors
• Reducing computational complexity in massive MIMO detection
• Enhancing scalability and flexibility in data-driven MIMO systems

Innovation

• Purely data-driven MIMO detector ChannelNet
• Channel-embedded layers and shared processors
• Low complexity O(NtNr) scalable solution