Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GCEPNet: Graph Convolution-Enhanced Expectation Propagation for Massive MIMO Detection

Qincheng Lu, Sitao Luan, Xiao-Wen Chang

TL;DR

GCEPNet incorporates data-dependent attention scores into Chebyshev polynomial for powerful graph convolution with better generalization capacity and empirically achieves the state-of-the-art (SOTA) MIMO detection performance with much faster inference speed.

Abstract

Massive MIMO (multiple-input multiple-output) detection is an important topic in wireless communication and various machine learning based methods have been developed recently for this task. Expectation Propagation (EP) and its variants are widely used for MIMO detection and have achieved the best performance. However, EP-based solvers fail to capture the correlation between unknown variables, leading to a loss of information, and in addition, they are computationally expensive. In this paper, we show that the real-valued system can be modeled as spectral signal convolution on graph, through which the correlation between unknown variables can be captured. Based on such analysis, we propose graph convolution-enhanced expectation propagation (GCEPNet). GCEPNet incorporates data-dependent attention scores into Chebyshev polynomial for powerful graph convolution with better generalization capacity. It enables a better estimation of the cavity distribution for EP and empirically achieves the state-of-the-art (SOTA) MIMO detection performance with much faster inference speed. To our knowledge, we are the first to shed light on the connection between the system model and graph convolution, and the first to design the data-dependent coefficients for graph convolution.

GCEPNet: Graph Convolution-Enhanced Expectation Propagation for Massive MIMO Detection

TL;DR

GCEPNet incorporates data-dependent attention scores into Chebyshev polynomial for powerful graph convolution with better generalization capacity and empirically achieves the state-of-the-art (SOTA) MIMO detection performance with much faster inference speed.

Abstract

Massive MIMO (multiple-input multiple-output) detection is an important topic in wireless communication and various machine learning based methods have been developed recently for this task. Expectation Propagation (EP) and its variants are widely used for MIMO detection and have achieved the best performance. However, EP-based solvers fail to capture the correlation between unknown variables, leading to a loss of information, and in addition, they are computationally expensive. In this paper, we show that the real-valued system can be modeled as spectral signal convolution on graph, through which the correlation between unknown variables can be captured. Based on such analysis, we propose graph convolution-enhanced expectation propagation (GCEPNet). GCEPNet incorporates data-dependent attention scores into Chebyshev polynomial for powerful graph convolution with better generalization capacity. It enables a better estimation of the cavity distribution for EP and empirically achieves the state-of-the-art (SOTA) MIMO detection performance with much faster inference speed. To our knowledge, we are the first to shed light on the connection between the system model and graph convolution, and the first to design the data-dependent coefficients for graph convolution.
Paper Structure (16 sections, 30 equations, 3 figures)

This paper contains 16 sections, 30 equations, 3 figures.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Notation
  4. Graph and Convolution
  5. Spectral graph convolution
  6. System Model
  7. Machine learning for MIMO Detection
  8. GNN-aided Expectation Propagation
  9. Efficient Graph Convolution-Based MIMO Detector
  10. The Graph Convolution Form of the System Model
  11. Architecture of GCEPNet
  12. Data-dependent Graph Convolution Coefficients
  13. Readout Module for Graph-based Estimation
  14. Computational Complexity Analysis
  15. Numerical Results
  16. ...and 1 more sections

Figures (3)

  • Figure 1: The structure of GCEPNet. The left panel shows the main workflow, where an iteration $t$ contains three modules that perform EP calculation, graph convolution and GRU gating respectively. Arrows indicate the data flow. The right panel illustrates the graph convolution process, that computes \ref{['eq: hidden-feature']} and \ref{['eq: gcepnet']}, using $N_t = 3$ as an example.
  • Figure 2: The total inference running time (in seconds) for $5 \times 10^4$ samples with $N_t = N_r$ and the same number of EP iterations ($T=9$). Both GCEPNet and GEPNet use 2 GNN layers in each EP iteration. The implementation of GEPNet is from kosasih2022graph. All methods are under CUDA acceleration.
  • Figure 3: The SER performance comparison for 64-QAM