MMNet is a massive MIMO signal detection scheme based on light online learning with neural networks that extends to correlated channel scenarios.

0. Introduction
1. Citation
2. Repository structure

This repository contains MMNet signal detection model, the channels dataset, and benchmarking detection schemes discusssed in the paper "Adaptive Neural Signal Detection for Massive MIMO" (https://arxiv.org/abs/1906.04610). On i.i.d. Gaussian channels, MMNet requires two orders of magnitude fewer operations than existing deep learning schemes but achieves near-optimal performance. On spatially-correlated channels, it achieves the same error rate as the next-best learning scheme (OAMPNet) at 2.5dB lower SNR and with at least 10x less computational complexity. MMNet is also 4--8dB better overall than a classic linear scheme like the minimum mean square error (MMSE) detector.

You may cite this project as:

@article{khani2019adaptive,
  title={Adaptive Neural Signal Detection for Massive MIMO},
  author={Khani, Mehrdad and Alizadeh, Mohammad and Hoydis, Jakob and Fleming, Phil},
  journal={arXiv preprint arXiv:1906.04610},
  year={2019}
}

Find MMNet and other learning based schemes in ./learning_based directory. Minimum mean square error (MMSE), Approximated message passaing (AMP), Semidefinite relaxation (SDR), Multistage interference cancelation (BLAST), and Maximum-likelihood optimal (ML) are located under ./classic. In order to reproduce the simulated correlated channels using 3D-3GPP model please refer to ./channels directory.