Indoor Massive MIMO: Uplink Pilot Mitigation Using Channel State Information Map

TL;DR

This paper introduces a machine learning-based method using a CSI map to predict channel state information, reducing uplink pilot contamination and feedback in indoor Massive MIMO systems, thereby improving sum-rate performance.

Contribution

It proposes a novel CSI prediction method using a graph-based machine learning approach to mitigate pilot contamination in indoor Massive MIMO scenarios.

Findings

Increased downlink sum-rate observed in simulations.

Significant reduction in feedback overhead.

Effective CSI prediction with the proposed graph-based method.

Abstract

Massive MIMO brings both motivations and challenges to develop the 5th generation Mobile wireless technology. The promising number of users and the high bitrate offered per unit area are challenged by uplink pilot contamination due to pilot reuse and a limited number of orthogonal pilot sequences. This paper proposes a solution to mitigate uplink pilot contamination in an indoor scenario where multi-cell share the same pool of pilot sequences, that are supposed to be less than the number of users. This can be done by reducing uplink pilots using Channel State Information (CSI) prediction. The proposed method is based on machine learning approach, where a quantized version of Channel State Information (QCSI) is learned during estimation session and stored at the Base Station (BS) to be exploited for future CSI prediction. The learned QCSI are represented by a weighted directed graph, which is responsible to monitor and predict the CSI of User Terminals (UTs) in the local cell. We introduce an online learning algorithm to create and update this graph which we call CSI map. Simulation results show an increase in the downlink sum-rate and a significant feedback reduction.