TL;DR
This paper introduces a novel precoding scheme for large-scale MIMO systems that significantly reduces computational complexity while maintaining the power efficiency of the optimal linear precoder (OLP). By applying a truncated polynomial expansion (TPE) on a per-user basis and leveraging an asymptotic analysis, they achieve near-optimal performance with drastically lower computational demands.
What Problem Does This Paper Address?
Modern cellular networks are experiencing an exponential growth in data traffic, driving the adoption of large-scale Multiple-Input Multiple-Output (MIMO) systems. These systems, characterized by base stations (BSs) with a massive number of antennas (M) serving multiple users (K), offer significant improvements in data rates and spectral efficiency. A key challenge in these systems is designing precoding matrices at the BS that minimize total power consumption while ensuring users receive data at their target rates. The "optimal linear precoder" (OLP) achieves this but requires computationally intensive matrix inversions for every new channel realization. This complexity makes OLP impractical for real-time implementation, especially in systems with highly dynamic channels or a large number of antennas and users.
Methods
The authors propose a "User Specific Truncated Polynomial Expansion" (US-TPE) precoding scheme. Instead of applying TPE to the entire precoding matrix, they apply it individually to each user's precoding vector. This "per-user" approach allows for a convex reformulation of the optimization problem, making it solvable.
To further reduce complexity and derive closed-form solutions, they employ an "asymptotic regime" analysis. This assumes that both the number of BS antennas (M) and users (K) grow large simultaneously, with a bounded ratio. Under this assumption, the signal-to-noise-plus-interference ratio (SINR) and transmit power can be approximated by deterministic quantities that depend only on channel statistics, rather than instantaneous channel realizations. This simplification leads to a novel US-TPE precoder that can be computed efficiently.
Datasets
Not discussed in the paper. The evaluation is based on simulated channel realizations.
Key Insights and Ideas
- Per-User TPE is Key: Applying TPE individually to each user's precoding vector, rather than the entire matrix, is crucial for achieving a convex optimization problem and enabling closed-form solutions.
- Asymptotic Analysis for Tractability: The assumption of large M and K with a bounded ratio is a powerful tool for simplifying complex problems in large-scale MIMO. It allows for deterministic approximations of performance metrics, moving away from computationally expensive Monte Carlo simulations for analysis.
- Complexity-Performance Trade-off: The truncation order (J) of the polynomial expansion directly impacts the trade-off between performance and complexity. Higher J leads to better performance but increased complexity.
- Practical Implications for Green Networks: The ability to achieve near-optimal power efficiency with significantly reduced complexity is vital for the development of energy-efficient ("green") cellular networks.
Key Results
- The proposed US-TPE precoding scheme achieves performance (in terms of total transmit power) that is very close to the optimal linear precoder (OLP).
- For a truncation order J=3, the US-TPE requires almost the same amount of power as OLP, with only a marginal increase compared to PA-RZF and traditional TPE schemes.
- The computational complexity of US-TPE is O(KM), which is significantly lower than OLP and PA-RZF (which involve O(K²M) operations per coherence period).
- The paper demonstrates that increasing the truncation order J improves performance, but with diminishing returns. J=3 is shown to be a good option for achieving performance close to OLP.
- The performance of US-TPE is also robust to variations in user target rates, maintaining near-optimal power efficiency.
Limitations and Future Work
- Truncation Order Selection: While J=3 is suggested as a good option, the paper notes that higher values of J might be needed for very high target rates (e.g., rmax ≥ 4 bits/sec/Hz) to further enhance performance. The optimal selection of J for specific scenarios could be a direction for future work.
- Channel State Information: The paper assumes perfect channel state information (CSI) at the BS. In practical systems, CSI is often estimated and imperfect. Investigating the performance of US-TPE under imperfect CSI would be a valuable extension.
- Beyond Single-Cell: The current work focuses on a single-cell scenario. Extending the US-TPE concept to multi-cell environments, where inter-cell interference is a significant factor, would be a natural next step.
Summary updated