TL;DR
This paper tackles the challenge of designing efficient communication constellations for noncoherent MIMO multiple-access channels, where channel information isn't known at the transmitter. The authors introduce novel design criteria based on minimizing detection error probability, leading to a simple partitioning-based construction that outperforms existing methods.
What Problem Does This Paper Address?
In wireless communication, especially with multiple-input multiple-output (MIMO) systems, knowing the exact channel conditions (channel state information or CSI) at the transmitter allows for optimized signal transmission. This is known as coherent communication. However, in many scenarios, obtaining this CSI is difficult or impossible, leading to noncoherent communication. This paper focuses on noncoherent MIMO multiple-access channels (MACs), where multiple users share a common MIMO channel without the transmitter knowing the CSI. The core problem is designing the signal constellations (the set of possible signals each user can send) for all users jointly to minimize the probability of errors when the receiver tries to decode the signals. This is particularly challenging because the users' signals are combined at the receiver, and their individual signal designs impact the overall system's performance.
Methods
The authors propose novel design criteria derived from analyzing the maximum likelihood (ML) detection error probability in noncoherent MIMO MACs. They first analyze the pairwise error probability (PEP), which is the probability that the detector mistakenly identifies one transmitted symbol as another.
Key methodological contributions include:
- Deriving New Design Metrics: They introduce two primary metrics:
- Max-emin: Maximizing the minimum expected pairwise log-likelihood ratio (PLLR) over all possible pairs of transmitted symbols. This metric is related to the Kullback-Leibler (KL) divergence and is particularly relevant in massive MIMO scenarios.
- Max-dmin: Maximizing the minimum value of a specific term related to the signal matrix product, which is shown to be a key factor in the error probability at high signal-to-noise ratios (SNR). This metric is inspired by single-user noncoherent MIMO design.
- Partitioning-Based Construction: Based on these metrics, they propose a practical and efficient construction method. This involves taking a well-designed single-user constellation and partitioning it into subsets, with each subset assigned to a different user. This simplifies the joint design problem by leveraging existing single-user constellation design principles.
- Numerical Optimization: For general cases, they employ numerical optimization techniques, including an alternating optimization approach where each user's constellation is optimized iteratively, and a manifold-constrained optimization using the Manopt toolbox.
Datasets
Not discussed in the paper. The paper focuses on theoretical analysis and numerical simulations rather than training on specific datasets.
Key Insights and Ideas
- Joint Design is Crucial: The performance of noncoherent MIMO MACs is highly dependent on how users' constellations are designed together, not just individually.
- Error Probability as a Guide: Analyzing the ML detection error probability provides a direct path to effective design criteria, moving beyond simpler distance-based metrics.
- Simplicity Can Be Effective: The proposed partitioning-based construction, which leverages single-user constellation designs, offers a practical and performant solution. This suggests that structured approaches can be very competitive.
- New Metrics Outperform Old: The proposed metrics (Max-emin and Max-dmin) lead to constellations that achieve better symbol error rates compared to constellations optimized with older criteria (Min-m1, Min-m2) and even a common pilot-based scheme.
Key Results
- The proposed design metrics (Max-emin and Max-dmin) are shown to be meaningful and effective for joint constellation design.
- Constellations optimized using these new metrics achieve significantly lower joint symbol error rates (SER) compared to existing methods, including a pilot-based scheme and constellations optimized with older criteria (Min-m1, Min-m2).
- The partitioning-based construction method provides a simple yet effective way to generate high-performing joint constellations.
- Numerical results demonstrate that the proposed constellations perform best, especially at higher SNRs, for the same transmission rate and power constraints.
Limitations and Future Work
- High Complexity of ML Detector: The paper focuses on optimizing for the joint maximum-likelihood (ML) detector, which is computationally complex. The authors acknowledge that it would be interesting to develop structured joint constellations that allow for low-complexity detection, similar to what has been achieved in the single-user case.
- Suboptimal Partitioning Assumption: While the partitioning approach is practical, the assumption that individual constellations follow from USTM is suboptimal. Future work could explore more optimal partitioning strategies.
- Convergence of Alternating Optimization: The alternating optimization approach is proposed for simplification, but its convergence is not guaranteed, and even when it converges, it may not reach the global optimum.
Summary updated