Machine Learning Enabled Preamble Collision Resolution in Distributed Massive MIMO

TL;DR

This paper introduces a machine learning framework that leverages distributed massive MIMO and deep neural networks to effectively resolve preamble collisions in grant-free random access, significantly improving uplink data rates.

Contribution

The paper proposes a novel deep learning-based approach combined with AP clustering to identify and mitigate preamble collisions in GFRA, enhancing performance over traditional methods.

Findings

Deep neural network achieves high accuracy in preamble multiplicity estimation.

AP clustering effectively organizes APs for collision resolution.

Proposed schemes significantly improve uplink achievable rates.

Abstract

Preamble collision is a bottleneck that impairs the performance of random access (RA) user equipment (UE) in grant-free RA (GFRA). In this paper, by leveraging distributed massive multiple input multiple output (mMIMO) together with machine learning, a novel machine learning based framework solution is proposed to address the preamble collision problem in GFRA. The key idea is to identify and employ the neighboring access points (APs) of a collided RA UE for its data decoding rather than all the APs, so that the mutual interference among collided RA UEs can be effectively mitigated. To this end, we first design a tailored deep neural network (DNN) to enable the preamble multiplicity estimation in GFRA, where an energy detection (ED) method is also proposed for performance comparison. With the estimated preamble multiplicity, we then propose a K-means AP clustering algorithm to cluster the neighboring APs of collided RA UEs and organize each AP cluster to decode the received data individually. Simulation results show that a decent performance of preamble multiplicity estimation in terms of accuracy and reliability can be achieved by the proposed DNN, and confirm that the proposed schemes are effective in preamble collision resolution in GFRA, which are able to achieve a near-optimal performance in terms of uplink achievable rate per collided RA UE, and offer significant performance improvement over traditional schemes.