Stochastic Geometry-based Uplink Analysis of Massive MIMO Systems with Fractional Pilot Reuse

TL;DR

This paper models and analyzes the uplink performance of massive MIMO networks using stochastic geometry, fractional pilot reuse, and cell region partitioning to improve spectral efficiency and reduce pilot contamination.

Contribution

It introduces a novel system model with Poisson point process BS locations, Johnson-Mehl cell regions, and fractional pilot reuse, deriving analytical expressions for key performance metrics.

Findings

Fractional pilot reuse improves CE user spectral efficiency.

Analytical expressions accurately predict coverage probability and spectral efficiency.

Partitioning rule for pilot sequences enhances overall network performance.

Abstract

In this work, we analyze the performance of the uplink (UL) of a massive MIMO network considering an asymptotically large number of antennas at base stations (BSs). We model the locations of BSs as a homogeneous Poisson point process (PPP) and assume that their service regions are limited to their respective Poisson-Voronoi cells (PVCs). Further, for each PVC, based on a threshold radius, we model the cell center (CC) region as the Johnson-Mehl (JM) cell of its BS while rest of the PVC is deemed as the cell edge (CE) region. The CC and CE users are located uniformly at random independently of each other in the JM cell and CE region, respectively. In addition, we consider a fractional pilot reuse (FPR) scheme where two different sets of pilot sequences are used for CC and CE users with the objective of reducing the interference due to pilot contamination for CE users. Based on the above system model, we derive analytical expressions for the UL signal-to-interference-and-noise ratio (SINR) coverage probability and average spectral efficiency (SE) for randomly selected CC and CE users. In addition, we present an approximate expression for the average cell SE. One of the key intermediate results in our analysis is the approximate but accurate characterization of the distributions of the CC and CE areas of a typical cell. Another key intermediate step is the accurate characterization of the pair correlation functions of the point processes formed by the interfering CC and CE users that subsequently enables the coverage probability analysis. From our system analysis, we present a partitioning rule for the number of pilot sequences to be used for CC and CE users as a function of threshold radius that improves the average CE user SE while achieving similar CC user SE with respect to unity pilot reuse.