Historically, stochastic geometry analyzed wireless networks using Homogeneous Poisson Point Processes (HPPPs). In an HPPP model, base stations and users are dropped completely at random across an infinite plane, entirely independent of one another. While mathematically elegant, this uniformity assumption breaks down in the real world. Mobile users naturally aggregate in hotspots—such as shopping malls, stadiums, or downtown business districts—and network operators intentionally co-locate infrastructure to meet this localized demand.
Assuming uniformity leads to severe miscalculations. If users are modeled as uniformly distributed when they are actually clustered, a model will underestimate the local interference they experience from neighboring devices and miscalculate the coverage provided by infrastructure. To bridge this gap, a robust body of research leverages clustered spatial point processes to capture the spatial correlation inherent in modern wireless deployments.
The Anatomy of a Cluster Process
To mathematically construct a network that is clustered yet still analytically tractable, researchers turn to two-tier point processes, most notably the Poisson Cluster Process (PCP).
A PCP is generated in two distinct steps:
- Parent Points: A set of root locations is generated using a standard point process. These represent the centers of hotspots, macro base stations, or central access points.
- Child Points: Around each parent point, a cluster of offspring points (representing users, small cells, or IoT devices) is scattered according to a specific probability density function.
Depending on how the children are distributed around the parent, different mathematical flavors emerge to match real-world topologies:
- Matern Cluster Process: The children are distributed uniformly within a disk of a fixed radius $R$ around the parent. This is highly effective for modeling sharp, bounded hotspots like a building or a localized sensor network. For instance, Zhijin Qin et al. utilized this approach to model LoRa networks where receivers serve as parents and active transmitters cluster within a specific radius. Similarly, Yujie Qin et al. used it to evaluate terrestrial base stations serving dense user hotspots.
- Thomas Cluster Process: The children are scattered according to a symmetric Gaussian (normal) distribution around the parent. This models a dense core that gradually thins out, capturing realistic urban sprawl or device-to-device (D2D) groupings. This variant was leveraged by Wenqiang Yi et al. to model high-frequency millimeter-wave (mmWave) D2D networks, and by Minwei Shi et al. to capture the spatial coupling between users and small cells in traffic hotspots.
Homogeneous PPP (Uniform) Poisson Cluster Process (Clustered)
+-----------------------+ +-----------------------+
| . . . . | | ... .. |
| . . . | | ..... .... |
| . . . . | --> | ... .. |
| . . . | | |
| . . . . | | .... |
+-----------------------+ +-----------------------+
Why It Works: Rethinking Interference and Geometry
The power of clustered modeling lies in how it fundamentally alters the calculation of the Signal-to-Interference-plus-Noise Ratio (SINR). In a uniform network, the distance to the nearest interfering node follows a simple exponential-like distribution. In a clustered network, if a user is part of a cluster, their closest neighbors are statistically much closer than they would be in a uniform distribution.
This spatial coupling means that intra-cluster interference dominates the network performance. For example, in Non-Orthogonal Multiple Access (NOMA) networks, multiple users share the same frequency and time resources within a tight geographic area. By using a PCP, Wenqiang Yi et al. demonstrated how power allocation and antenna configurations must be tuned differently when users are normally distributed around a base station rather than spread uniformly. Their work proved that NOMA consistently outperforms traditional Orthogonal Multiple Access (OMA) under realistic clustering constraints.
This spatial reality also extends to emerging hardware deployments. Chao Zhang 0048 et al. applied PCPs to model Reconfigurable Intelligent Surfaces (RISs) grouping around users, accurately capturing the localized path loss characteristics and reflecting channel enhancements that uniform models obscure.
Unlocking New Network Architectures
Clustered spatial modeling has become indispensable for evaluating next-generation, non-terrestrial, and heterogeneous architectures where uniform assumptions fail entirely.
UAV and Aerial Networks
Drones are uniquely capable of flying directly over user hotspots to provide on-demand capacity. Modeling this requires a coupled approach where the aerial base station acts as the parent point directly above a cluster of ground users. Research by Wenqiang Yi et al. and Yujie Qin et al. shows that while this positioning maximizes line-of-sight signal strength, it creates a "double-edged sword." The line-of-sight links that boost the desired signal also amplify the interference received from neighboring aerial nodes, a nuance that requires precise 3D geometric modeling to optimize drone altitude and antenna array sizes.
Post-Disaster and Heterogeneous Deployments
When infrastructure is partially destroyed or unevenly distributed, uniform models cannot capture the geographic variance. In these scenarios, researchers utilize Inhomogeneous Poisson Point Processes (IPPP), where node density varies continuously across space based on a localized intensity function.
Maurilio Matracia et al. applied IPPPs to disaster zones where infrastructure survival probabilities drop near an epicenter, allowing them to compare the efficacy of drone-mounted base stations against aerial smart mirrors. Furthermore, Serkan Ak et al. used non-homogeneous processes to prove that as cellular networks become ultra-dense and heterogeneous, the aggregate wireless interference mathematically converges to a simple Gaussian distribution via Berry-Esseen bounds, offering a powerful shortcut for network designers.
From Average Performance to Fine-Grained Reliability
Beyond calculating simple average coverage probabilities, clustered spatial modeling allows researchers to evaluate the SINR meta distribution—the distribution of the conditional coverage probability. In simple terms, while standard geometry tells you the percentage of users who are connected on average, the meta distribution tells you the reliability of individual links across the network (e.g., what fraction of users achieve a specific reliability threshold).
Using this framework, Yujie Qin et al. evaluated uplink performance in UAV-assisted networks. They discovered that when users are tightly clustered, increasing a drone's altitude is far more effective at boosting overall system reliability than merely increasing transmit power.
This counter-intuitive insight emerges precisely because changing the altitude alters the 3D geometric relationship between the cluster's internal users and external interferers. By shifting the geometry, the network designer mitigates the dominant intra-cluster interference—a design breakthrough that is completely invisible when using traditional uniform models.
Key Foundational Work
Understanding the Benefits of Open Access in Femtocell Networks: Stochastic Geometric Analysis in the Uplink (2013) — contributed 1 concept elements to this direction.
Gaussian approximation for the downlink interference in heterogeneous cellular networks (2016) — contributed 1 concept elements to this direction.