TL;DR
This paper tackles the challenge of optimizing spectral and energy efficiency in mmWave MIMO-NOMA systems by proposing novel user grouping and power allocation strategies. They introduce an AGNES-based clustering approach and two efficient schemes (DIR-AGNES and SUC-AGNES) to manage beam overlapping and interference, outperforming existing methods.
What Problem Does This Paper Address?
The paper addresses the complex optimization problem in 5G and beyond wireless systems, specifically focusing on the integration of Non-Orthogonal Multiple Access (NOMA) with millimeter wave (mmWave) Massive MIMO (Multiple-Input Multiple-Output) systems employing hybrid beamforming (HBF). While NOMA offers increased capacity and mmWave provides high bandwidth, their combination presents significant challenges. These include:
- Spectral and Energy Efficiency (SE/EE) Optimization: Finding the optimal balance between maximizing data throughput and minimizing power consumption is a difficult non-linear programming problem.
- User Grouping and Beam Selection: Efficiently grouping users and selecting appropriate beams is crucial, especially in mmWave systems where channels are highly directional. Poor grouping can lead to "beam overlapping," where different user groups unintentionally interfere with each other due to similar directional paths.
- Inter-Group Interference: In NOMA, users within a group share resources, but interference between different groups (especially when they share scatters) can degrade performance.
- Computational Complexity: Exhaustive search for optimal solutions is computationally infeasible. Existing methods often use sequential optimization, which can lead to suboptimal results.
Methods
The authors propose a multi-faceted approach to tackle these challenges:
- Decomposition Strategy: They decompose the overall optimization problem into two main parts:
- An integer problem involving user grouping and beam selection.
- A continuous problem involving power allocation and digital beamforming design.
- Initial User Grouping (AGNES): They introduce a novel initial user grouping algorithm based on Agglomerative Nesting (AGNES) clustering. This method leverages channel correlation (specifically, angle-of-arrival similarity) to group users that are likely to benefit from the same beam, thereby minimizing inter-group interference. They compare various linkage methods within AGNES, finding "complete linkage" to be most effective.
- User Grouping and Beam Selection Schemes: To address the beam overlapping problem and reduce complexity, they propose two low-complexity schemes:
- DIR-AGNES (Direct AGNES): This scheme first groups users using AGNES and then iteratively assigns beams. Once a beam is assigned to a group, it's removed from the available codebook for subsequent groups, preventing direct beam reuse and mitigating overlapping.
- SUC-AGNES (Successive AGNES): This more advanced scheme builds upon DIR-AGNES. It not only removes assigned beams but also actively updates the channels of remaining users using a Gram-Schmidt-based procedure. This allows the system to better adapt to the remaining available beams and further mitigate inter-group interference in the analog domain.
- Power Allocation (Quadratic Transform - QT): To handle the non-convex power allocation problem, they employ the Quadratic Transform (QT) method. This technique transforms the non-convex problem into a sequence of convex problems that can be solved iteratively. They apply this for both SE maximization and EE maximization, with a nested QT (NQT) approach for EE.
- Digital Beamforming: They utilize a standard approach for digital beamforming, often involving Zero-Forcing (ZF) techniques, to combat residual inter-beam interference after analog beamforming.
Datasets
Not discussed in the paper. The evaluation is based on extensive simulations using synthetic channel models.
Key Insights and Ideas
- Channel Correlation is Key for Grouping: The directionality of mmWave channels makes channel correlation (specifically, angle-of-arrival similarity) a critical factor for effective user grouping. Users with similar channel characteristics are good candidates for sharing a beam.
- Beam Overlapping is a Major Hurdle: Simply grouping users isn't enough. The paper highlights the "beam overlapping problem" where similar beam patterns for different groups lead to significant interference, a critical issue in mmWave NOMA.
- Iterative Refinement for Interference Mitigation: The SUC-AGNES scheme's channel updating mechanism is a clever way to proactively reduce inter-group interference by adapting to the available beams and the remaining users.
- Decomposition Simplifies Complexity: Breaking down the complex joint optimization into an integer (grouping/beam selection) and a continuous (power allocation/digital beamforming) problem is a practical strategy to manage computational complexity.
- AGNES Offers Spontaneous Grouping: Unlike K-means, AGNES provides a deterministic and spontaneous grouping without the need for random initializations, potentially leading to more consistent performance.
Key Results
- Outperformance of Proposed Schemes: Both DIR-AGNES and SUC-AGNES schemes consistently outperform conventional OMA and other state-of-the-art NOMA schemes in terms of Spectral Efficiency (SE).
- SUC-AGNES Superiority: The SUC-AGNES scheme generally achieves better SE than DIR-AGNES, attributed to its more sophisticated inter-group interference mitigation.
- Robustness to User Number: The proposed schemes show stable superiority across varying numbers of users, unlike some existing methods (e.g., algorithm in [9]) which degrade significantly when the user-to-RF chain ratio increases.
- Efficiency Gains: The paper demonstrates improvements in both SE and Energy Efficiency (EE) compared to baseline methods.
- Reduced Complexity: The proposed algorithms significantly reduce computational complexity compared to exhaustive search, moving from exponential to a more manageable level.
- Convergence Speed: The QT algorithm for power allocation shows comparable results to the Dinkelbach method but with a slightly better convergence speed in the initial iterations.
Limitations and Future Work
- Perfect CSI Assumption: The paper assumes perfect Channel State Information (CSI) is known at both the BS and users. In real-world scenarios, CSI acquisition is imperfect and can significantly impact performance.
- Single Antenna Users: The current model assumes users have single antennas. The authors mention that the user grouping can be extended to multi-antenna users, which would be a natural area for future work.
- Specific Channel Model: The simulations are based on a double directional channel model. Evaluating performance with more realistic and complex channel models would be beneficial.
- Hardware Imperfections: The study does not explicitly account for hardware impairments like phase noise or non-linearities, which are common in mmWave systems.
- Future Work: The authors suggest extending the user grouping to multi-antenna users and exploring more advanced interference cancellation techniques. Further investigation into the impact of imperfect CSI and hardware imperfections would also be valuable.
Summary updated