Pilot Allocation in Multi-Cell MIMO Systems Based on Missing Data Imputation

We consider a pilot allocation algorithm to alleviate pilot contamination in a multi-cell massive MIMO network with missing channel information. We utilize a machine learning framework for the imputation of large-scale channel state information (CSI) features in neighboring cells. In the offline pha...

Full description

Saved in:
Bibliographic Details
Main Authors: Bushra Shaikh, Pere Garau Burguera, Hanan Al-Tous, Markku Juntti, Bilal Muhammad Khan, Olav Tirkkonen
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11072106/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:We consider a pilot allocation algorithm to alleviate pilot contamination in a multi-cell massive MIMO network with missing channel information. We utilize a machine learning framework for the imputation of large-scale channel state information (CSI) features in neighboring cells. In the offline phase, a dataset with full information is created; the covariance CSI feature is estimated for each user at all base stations. The CSI features are exploited to create a multi-point channel chart (CC), where relative positions of UEs are preserved. We assume an incomplete information setup in the online phase, in which user channel covariance is known only at the serving BS. The CC locations in the online phase are predicted using a supervised machine learning framework. We consider a constrained weighted graph coloring approach to allocate the pilots. We make use of CC distances to construct similarity weights between users; nearby users have large similarities, whereas faraway users have small similarities. Numerical results show that the CC based approach outperforms the solution utilizing angle-of-arrival with full information, and nears the performance of the one based on covariance with full information. Furthermore, we consider a machine learning framework to predict the channel covariance matrices at other BSs. The performance of this scheme is slightly better than that of the CC based approach. However, its communication overhead and computational complexity is much larger, compared to the CC based scheme.
ISSN:2169-3536