This paper presents a robust framework for optimizing uplink communications in 5G-and-beyond networks by combining two powerhouse technologies: Millimeter Wave (mmWave) and Non-Orthogonal Multiple Access (NOMA). The authors introduce a novel hierarchical clustering approach for user grouping and an iterative power allocation strategy that significantly boosts both spectral and energy efficiency while avoiding the "beam overlapping" interference common in hybrid beamforming systems.
TL;DR
The authors propose a joint optimization framework for mmWave-NOMA systems that uses hierarchical clustering (AGNES) to group users and a successive interference-avoidance scheme to select beams. By recasting non-convex power allocation into a solvable convex form, they achieve superior spectral and energy efficiency with linear computational complexity compared to traditional OMA and state-of-the-art NOMA methods.
What Problem Does This Paper Address?
As we push toward 5G and 6G, we face a dual challenge: the need for massive connectivity (supporting many users at once) and massive bandwidth. NOMA addresses the former by allowing multiple users to share the same frequency/time resources via the power domain. mmWave addresses the latter by tapping into high-frequency bands, though it requires directional beamforming to overcome high path loss.
However, merging these in a Hybrid Beamforming (HBF) structure—where a few expensive RF chains serve many antennas—creates a massive optimization headache. Specifically, the authors tackle the beam overlapping problem. As illustrated in Figure 1, if two different user groups share a scatterer or select highly correlated beams, the digital beamformer cannot effectively separate their signals, leading to crippling inter-group interference.
Methods
The authors decompose the joint optimization of user grouping, beam selection, and power allocation into two manageable parts:
- The Integer Problem (Grouping & Beam Selection):
- AGNES Clustering: Instead of using K-means (which is highly sensitive to initial conditions), the authors use Agglomerative Nesting (AGNES). This hierarchical approach builds groups "spontaneously" based on channel correlation. They identify "Complete Linkage" (see Table 1 and Figure 2) as the best method for ensuring users within a group are highly correlated enough to be served by the same beam.
- SUC-AGNES: To solve the beam overlapping issue, they propose a Successive AGNES scheme. After selecting a beam for one group, it uses a Gram-Schmidt-like process to "remove" that beam's component from the remaining users' channels. This actively steers subsequent groups toward less-interfered spatial paths.
- The Continuous Problem (Power & Digital Beamforming):
- Quadratic Transform (QT): The original power allocation problem is non-convex and difficult to solve. The authors use the Quadratic Transform to recast the sum-of-ratio objective functions (for Spectral Efficiency and Energy Efficiency) into a convex form.
- Iterative Optimization: They then use a block coordinate ascent approach to iteratively optimize the power levels and digital beamforming weights until convergence.
Datasets
The paper relies on numerical simulations rather than static datasets. They use a widely adopted double directional mmWave channel model to simulate realistic sparse propagation environments.
- Parameters: 64-antenna Base Station (BS), single-antenna users, 6 propagation paths per user.
- Evaluation: Results are averaged over 3000 Monte Carlo simulations to ensure statistical significance.
Key Insights and Ideas
- Correlation is King in mmWave: Traditional NOMA grouping often focuses on channel gain differences. In mmWave, the authors show that channel correlation (spatial directionality) is a far more critical metric for grouping to ensure effective beamforming gain.
- Active vs. Passive Interference Mitigation: While standard HBF-NOMA tries to fix interference at the digital stage, the SUC-AGNES scheme prevents it at the analog stage by updating the "residual" channel during the selection process.
- Spontaneous Grouping: By using AGNES, the system avoids the "bad initialization" trap of K-means, leading to more stable and higher-performing user clusters. Figure 3 demonstrates that "Complete Linkage" provides the best spectral efficiency across various SNR levels.
Key Results
- Spectral Efficiency (SE): The proposed SUC-AGNES scheme consistently outperforms conventional OMA and existing NOMA grouping algorithms (like those based solely on gain difference).
- Energy Efficiency (EE): The iterative power allocation successfully finds the "sweet spot" where EE is maximized. The authors note that simply increasing transmit power eventually leads to a decrease in EE due to the fixed power costs of the hardware.
- Computational Complexity: The proposed algorithms reduce the complexity from exponential (required for a brute-force global optimum) to linear ($O(K^2)$), making them feasible for real-time implementation in 5G base stations.
Limitations and Future Work
- Perfect CSI Assumption: The study assumes perfect Channel State Information (CSI) is available at the BS. In real-world mmWave systems, channel estimation is notoriously difficult due to the high number of antennas and low SNR before beamforming.
- Single-Antenna Users: While the authors mention the framework can be extended to multi-antenna users, the primary evaluation focuses on single-antenna devices.
- Flat Fading: The current model assumes a flat-fading channel. Extending this to frequency-selective channels (common in wideband 5G) would be a valuable next step.
Summary updated