Abstract
In wireless communication systems, the accurate and reliable evaluation of channel capacity is believed to be a fundamental and critical issue for terminals. However, with the rapid development of wireless technology, large-scale communication networks with significant random interference have emerged, resulting in extremely high computational costs for capacity calculation. In ultra-dense wireless networks with extremely large numbers of base stations (BSs) and users, we provide fast estimation methods for determining the capacity. We consider two scenarios according to the ratio of the number of users to the number of BSs, $\beta_m$. First, when $\beta_m\leq1$, the FIsher-Spiked Estimation (FISE) algorithm is proposed to determine the capacity by modeling the channel matrix with random interference as a Fisher matrix. Second, when $\beta_m>1$, based on a closed-form expression for capacity estimation requiring solely simple computations, we prove that this estimation stabilizes and remains invariant with increasing $\beta_m$. Our methods can guarantee high accuracy on capacity estimation with low complexity, which is faster than the existing methods. Furthermore, our approaches exhibit excellent generality, free of network area shapes, BS and user distributions, and sub-network locations. Extensive simulation experiments across various scenarios demonstrate the high accuracy and robustness of our methods.
Introduction
As wireless networks evolve toward 6G, they are becoming "ultra-dense," featuring massive numbers of base stations (BSs) and users. A critical metric for these networks is channel capacity—the maximum rate at which information can be reliably transmitted. Traditionally, calculating capacity involves computing the determinant of large-dimensional matrices, which requires $O(J^3)$ operations for a network with $J$ antennas. In ultra-dense scenarios, this becomes a massive computational bottleneck.
Furthermore, existing "fast" methods often rely on oversimplifying assumptions, such as treating interference as a fixed diagonal matrix. This leads to significant inaccuracies in real-world, dynamic environments where interference is random and structured. This paper introduces a high-speed, high-accuracy framework leveraging Random Matrix Theory (RMT) to reduce computational complexity to linear $O(J)$ while maintaining less than 2% error compared to exact baseline methods.
Methodology
The proposed framework addresses capacity estimation through two distinct strategies based on $\beta_m$, the ratio of users to base stations in a network cluster:
1. FISE Algorithm ($\beta_m \leq 1$)
For scenarios where users are fewer than or equal to BSs, the authors introduce the Fisher-Spiked Estimation (FISE) algorithm. Instead of calculating every eigenvalue of the channel matrix, FISE leverages the theory of "spiked Fisher matrices":
- Spiked Eigenvalues: It estimates the few dominant ("spiked") eigenvalues that carry the majority of the signal power.
- Bulk Approximation: It uses the Limiting Spectral Distribution (LSD) of Fisher matrices to approximate the "bulk" of the remaining non-spiked eigenvalues. This approach treats both the signal and the interference as random matrices, providing a more realistic model than assuming deterministic interference.
2. Stability Theorem ($\beta_m > 1$)
When users outnumber BSs, the authors prove a "Stability Theorem." They demonstrate that as the user-to-BS ratio increases, the capacity estimation stabilizes and remains invariant. This implies that in extremely crowded areas, the mathematical complexity does not scale infinitely; researchers can calculate capacity once for a specific threshold, saving significant time and power.
Experimental Results
The algorithms were validated using extensive numerical simulations across two primary synthetic scenarios:
- Scenario S1 (Uniform): A circular network where BSs and users follow a homogeneous Poisson Point Process (PPP).
- Scenario S2 (Non-Uniform): A square network where nodes follow a truncated normal distribution, mimicking urban "hotspots."
Key Findings
- Speed: The FISE algorithm operates at $O(J_m)$ complexity, making it significantly faster than traditional $O(J_m^3)$ determinant-based calculations and the $O(J_m^2)$ MPM method.
- Accuracy: Across both uniform and non-uniform distributions, the proposed methods maintained a relative error of less than 2% compared to the exact Cholesky decomposition baseline.
- Robustness: Unlike previous methods like TOSE, FISE does not require manual parameter tuning, making it highly adaptable to different network shapes and cluster locations.
Discussion and Future Work
A core insight of this work is that interference in ultra-dense networks has a specific spectral signature. Modeling the system as a Fisher Matrix (the ratio of two random covariance matrices) allows for high accuracy without the computational overhead of traditional methods.
While the current study focuses on single-antenna nodes, the framework is theoretically adaptable to multi-antenna systems. Future research will explore the limiting theory of Hadamard products to achieve even more refined estimations and test these algorithms on real-world software-defined radio (SDR) testbeds.
Summary updated