Transmit Correlation Diversity: Generalization, New Techniques, and Improved Bounds

Fan Zhang, Khac–Hoang Ngo, Sheng Yang, Aria Nosratinia

Abstract

In Multiple-Input Multiple-Output (MIMO) broadcast channels, users often experience distinct spatial transmit correlation matrices. When these matrices possess mutually exclusive eigenspaces, a phenomenon known as "transmit correlation diversity" (TCD) allows for non-interfering training and transmission. This paper generalizes the concept of TCD to scenarios where users' eigenspaces partially or fully overlap. By introducing a hybrid framework of product superposition and pre-beamforming (rate splitting), the authors derive achievable degrees of freedom (DoF) and rate regions for both two-user and $K$-user systems. The results demonstrate significant performance gains in Frequency-Division Duplex (FDD) Massive MIMO systems by optimizing pilot alignment and reclaiming resources typically lost to training overhead.

Introduction

In modern wireless communications, spatial correlation—the tendency of signals to travel along specific spatial paths—is traditionally viewed as a limitation that reduces the available Degrees of Freedom (DoF). However, recent research into Transmit Correlation Diversity (TCD) has shown that if users have distinct spatial signatures, the transmitter can exploit these differences to serve multiple users simultaneously.

While early TCD models required users to have perfectly orthogonal (mutually exclusive) correlation structures, real-world environments rarely offer such clean separation. This paper addresses the critical gap of overlapping eigenspaces, providing a theoretical and practical framework to harvest diversity gains even when users share spatial paths. This is particularly relevant for Massive MIMO systems operating in FDD mode, where the overhead of channel state information (CSI) acquisition often limits system throughput.

Methodology and Technical Framework

The authors propose a framework that "carves" the transmission signal space into common and private components based on the geometric overlap of users' correlation matrices. To achieve this, they employ several key techniques:

1. Rate Splitting and Pre-beamforming

The transmitter divides messages into a common part, which is decodable by multiple users within the overlapping eigenspaces, and private parts directed at specific users in non-overlapping regions. This allows the system to manage interference more granularly than traditional beamforming.

2. Product Superposition

This signaling technique involves "multiplying" data for one user onto the pilot signals intended for another. By embedding data within the training phase, the system effectively reduces the time dedicated solely to pilots, increasing the overall spectral efficiency.

3. Interference Graph Analysis

To generalize the 2-user model to a $K$-user system, the authors model the overlapping eigenspaces as an undirected graph. By applying graph coloring techniques (specifically identifying the chromatic number), they determine the minimum number of pilot dimensions required to prevent interference. This systematic approach allows for optimized pilot scheduling in complex network topologies.

4. The Zassenhaus Algorithm

The paper utilizes the Zassenhaus algorithm to calculate the specific precoding matrices required to align signals with the intersection and sum of the users' eigenspaces, ensuring that the mathematical theory translates into implementable signal design.

Key Insights and Theoretical Results

Overlap as a Resource

Rather than treating overlapping eigenspaces as unavoidable interference, the paper treats them as a structured resource. By identifying the rank of the overlap ($r_0$), the transmitter can optimize the balance between common and private data streams. The study proves that a hybrid approach—combining rate splitting and product superposition—is necessary to achieve the optimal DoF region across various antenna configurations.

Tightness of Bounds

For the two-user case with Channel State Information at the Receiver (CSIR), the authors demonstrate that their proposed achievable DoF region is "tight" (optimal) for several common configurations of receive antennas and channel correlation ranks. This provides a definitive benchmark for the maximum possible performance in correlated MIMO channels.

Massive MIMO Efficiency

In simulations of an FDD Massive MIMO system (e.g., $M=32$ antennas), the proposed TCD schemes significantly outperformed standard Time Division Multiple Access (TDMA). By reclaiming the "lost" time spent on pilots through product superposition, the system achieves a marked increase in the sum rate, proving the practical utility of the generalized TCD framework.

Limitations and Future Directions

While the paper provides a robust solution for symmetric and nested correlation structures, it acknowledges that highly irregular or "messy" correlation structures in large-scale networks remain mathematically complex to optimize. Furthermore, the framework assumes the transmitter has access to long-term statistics (correlation matrices). While these statistics change slowly compared to instantaneous CSI, accurate feedback mechanisms are still required.

Future research may extend these techniques to general interference channels involving multiple transmitters and refine power allocation strategies to enhance performance at moderate Signal-to-Noise Ratio (SNR) levels, where the benefits of TCD are most impactful for end-users.

Conclusion

This work broadens the scope of transmit correlation diversity from a niche theoretical case to a generalized principle applicable to modern wireless standards. By leveraging the structured nature of spatial correlation through graph theory and sophisticated signaling, the authors provide a roadmap for more efficient FDD Massive MIMO systems, turning spatial limitations into a source of multiplexing gain.

Summary updated

Abstract

When the users in a MIMO broadcast channel experience different spatial transmit correlation matrices, a class of gains is produced that is denoted transmit correlation diversity. This idea was conceived for channels in which transmit correlation matrices have mutually exclusive eigenspaces, allowing non-interfering training and transmission. This paper broadens the scope of transmit correlation diversity to the case of partially and fully overlapping eigenspaces and introduces techniques to harvest these generalized gains. For the two-user MIMO broadcast channel, we derive achievable degrees of freedom (DoF) and achievable rate regions with/without channel state information at the receiver (CSIR). When CSIR is available, the proposed achievable DoF region is tight in some configurations of the number of receive antennas and the channel correlation ranks. We then extend the DoF results to the $K$-user case by analyzing the interference graph that characterizes the overlapping structure of the eigenspaces. Our achievability results employ a combination of product superposition in the common part of the eigenspaces, and pre-beamforming (rate splitting) to create multiple data streams in non-overlapping parts of the eigenspaces. Massive MIMO is a natural example in which spatially correlated link gains are likely to occur. We study the achievable downlink sum rate for a frequency-division duplex massive MIMO system under transmit correlation diversity.
BibTeX
@article{zhang2021transmit,
  author = {Fan Zhang and Khac–Hoang Ngo and Sheng Yang and Aria Nosratinia},
  title = {{Transmit Correlation Diversity: Generalization, New Techniques, and Improved Bounds}},
  journal = {IEEE Transactions on Information Theory},
  year = {2021},
}
Figure