Spatial-Temporal-DBSCAN-Based User Clustering and Power Allocation for Sum Rate Maximization in Millimeter-Wave NOMA Systems

Huu-Trung Hoang, Quoc-Viet Pham, Won-Joo HwangSymmetry 2020 · DOI

TL;DR

This paper tackles the challenge of maximizing data rates in millimeter-wave (mmWave) Non-Orthogonal Multiple Access (NOMA) systems by proposing a novel spatial-temporal clustering algorithm (ST-DBSCAN) to group users. This clustering, combined with optimized power allocation and beamforming, significantly boosts system performance compared to existing methods.

What Problem Does This Paper Address?

Fifth-generation (5G) mobile networks are experiencing a massive surge in data traffic, necessitating higher bandwidth and efficiency. Millimeter-wave (mmWave) frequencies offer this bandwidth but suffer from directional transmission and susceptibility to interference. Non-Orthogonal Multiple Access (NOMA) allows multiple users to share the same resource block, enhancing spectrum efficiency. Combining mmWave and NOMA (mmWave-NOMA) promises high capacity but introduces complexities: users within the same beam can interfere with each other (intra-beam interference), and users in different beams can also interfere (inter-beam interference). Efficiently grouping users (clustering) and allocating power (power allocation) are crucial for mitigating these interferences and maximizing the overall system sum-rate. Traditional clustering methods often struggle with the spatial and temporal dynamics of user distribution and channel conditions in mmWave-NOMA systems.

Methods

The authors propose a three-pronged approach:

  • Spatial-Temporal DBSCAN (ST-DBSCAN) for User Clustering: This is the core innovation. ST-DBSCAN is an unsupervised learning algorithm that extends the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm. It leverages both the spatial proximity of users (their geographical coordinates) and their channel gains to form clusters. This is a key departure from methods that solely rely on channel gains, as it accounts for the directional nature of mmWave. The algorithm automatically determines the optimal number of clusters, a significant advantage over methods requiring pre-defined cluster counts.
  • Inter-Cluster Power Allocation (PA): After clustering, the problem of allocating power to each cluster is addressed. The authors decompose this into intra-cluster and inter-cluster PA. They propose an inter-cluster PA algorithm that aims to maximize the sum-rate while considering the interference between clusters.
  • Hybrid Beamforming (HBF) with BC-PSO: To further suppress inter-cluster interference, the paper incorporates an HBF solution. This involves an Approximate Zero-Forcing (AZF) method for digital beamforming and a Boundary-Compressed Particle Swarm Optimization (BC-PSO) algorithm for analog beamforming. BC-PSO is used to handle the non-convexity of the analog beamforming problem.

Datasets

Not discussed in the paper. The paper focuses on a simulated system model and evaluates performance through simulations.

Key Insights and Ideas

  • Spatial-Temporal Clustering is Key: The paper highlights that considering both user location and channel gain simultaneously is crucial for effective clustering in mmWave-NOMA, leading to better performance than methods relying on only one factor.
  • Automatic Cluster Number Detection: ST-DBSCAN's ability to automatically determine the optimal number of clusters simplifies system design and avoids suboptimal performance that can arise from manually setting this parameter.
  • NOMA's Superiority over OMA: The results clearly demonstrate that NOMA, when properly managed with intelligent clustering and power allocation, significantly outperforms Orthogonal Multiple Access (OMA) in terms of sum-rate for mmWave systems.
  • Interference Mitigation is Achievable: The proposed combination of ST-DBSCAN, inter-cluster PA, and HBF effectively mitigates both intra- and inter-cluster interference, leading to performance close to an "ideal" scenario with no inter-group interference.

Key Results

  • The proposed ST-DBSCAN-based user clustering algorithm achieves a higher sum-rate compared to benchmark algorithms like K-means and Cluster-Head, under both Line-of-Sight (LoS) and Non-Line-of-Sight (NLoS) channel models.
  • The mmWave-NOMA system with the proposed ST-DBSCAN clustering and PA significantly outperforms the mmWave-OMA system.
  • The proposed inter-cluster PA algorithm shows a substantial improvement in sum-rate compared to a scheme with no inter-cluster PA.
  • The paper's approach achieves sum-rate performance that is asymptotic to an ideal scenario without inter-group interference, indicating effective interference management.

Limitations and Future Work

  • Computational Complexity: While ST-DBSCAN is more efficient than some other advanced clustering methods, its complexity is noted as being higher than random clustering.
  • Future Work: The authors plan to investigate the effect of ST-DBSCAN on NOMA systems with more cells and users. They also aim to explore its application in UAV clustering for improved communication performance, moving beyond traditional methods.

Summary updated

BibTeX
@article{hoang2020spatial,
  author = {Huu-Trung Hoang and Quoc-Viet Pham and Won-Joo Hwang},
  title = {{Spatial-Temporal-DBSCAN-Based User Clustering and Power Allocation for Sum Rate Maximization in Millimeter-Wave NOMA Systems}},
  journal = {Symmetry},
  year = {2020},
  doi = {10.3390/sym12111854},
}
Figure