# MaMIMO CSI-based positioning using CNNs: Peeking inside the black box

arXiv: https://arxiv.org/abs/2003.04581 · HTML: https://mlsanity.com/paper/2003.04581/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 10 Mar 2020.

## Built on by (most cited first, 5 of 5)

- [CSI-fingerprinting Indoor Localization via Attention-Augmented Residual Convolutional Neural Network](https://mlsanity.com/paper/2205.05775/) (2022)
- [DyLoc: Dynamic Localization for Massive MIMO Using Predictive Recurrent Neural Networks](https://mlsanity.com/paper/2101.07848/) (2021)
- [Towards Sustainable Deep Learning for Wireless Fingerprinting Localization](https://mlsanity.com/paper/2201.09071/) (2022)
- [Resource-aware Deep Learning for Wireless Fingerprinting Localization](https://mlsanity.com/paper/2211.01759/) (2022)
- [High-Precision Machine-Learning Based Indoor Localization with Massive MIMO System](https://mlsanity.com/paper/2303.03743/) (2023)

## Abstract (arXiv)

Massive MIMO (MaMIMO) Channel State Information (CSI) based user positioning systems using Convolutional Neural Networks (CNNs) show great potential, reaching a very high accuracy without introducing any overhead in the MaMIMO communication system. In this study, we show that both these systems can position indoor users in both Line-of-Sight and in non-Line-of-Sight conditions with an accuracy of around 20 mm. However, to further develop these positioning systems, more insight in how the CNN infers the position is needed. The used CNNs are a black box and we can only guess how they position the users. Therefore, the second focus of this paper is on opening the black box using several experiments. We explore the current limitations and promises using the open dataset gathered on a real-life 64-antenna MaMIMO testbed. In this way, extra insight in the system is gathered, guiding research on MaMIMO CSI-based positioning systems using CNNs in the right direction.