Transmit Designs for the MIMO Broadcast Channel With Statistical CSI

Pablo Piantanida, Yongpeng Wu, Shi Jin, Xiqi Gao, Matthew R. McKay, Chengshan XiaoIEEE TSP 2014 · DOI

TL;DR

This paper tackles the challenge of designing efficient transmission strategies for MIMO broadcast channels when only statistical channel state information (CSI) is available at the transmitter. They introduce a novel low-complexity approach that leverages second-order CSI statistics, offering performance close to theoretical limits and significantly outperforming existing methods.

What Problem Does This Paper Address?

In wireless communication systems, particularly those with multiple antennas (MIMO), efficiently transmitting data to multiple users simultaneously is crucial. The MIMO broadcast channel (BC) is a key model for this. While optimal strategies like dirty-paper coding (DPC) exist, they often rely on perfect, instantaneous knowledge of the channel state information (CSI) at the transmitter. This assumption is often unrealistic, especially in mobile scenarios where channels fluctuate rapidly, making accurate tracking at the transmitter difficult.

This paper focuses on the more practical scenario where only statistical CSI is available at the transmitter. This means the transmitter knows the statistical properties (like average power or correlation) of the channel, but not its exact instantaneous realization. The challenge is to design transmit strategies that maximize the overall data rate (sum-rate) under these statistical CSI constraints, while also being computationally feasible. Existing work has explored capacity limits but often lacked explicit, practical transmit designs.

Methods

The authors propose a two-pronged approach:

  1. Iterative Optimization (Algorithm 1): They first derive necessary conditions for optimal transmit designs (linear assignment matrices and precoding matrices) and formulate a multidimensional optimization problem. This is solved using an iterative alternating optimization method with gradient descent. However, this method relies on Monte Carlo simulations for expectation calculations, leading to high computational complexity.
  2. Low-Complexity Design Based on Second-Order Statistics (Algorithm 2): To overcome the complexity of Algorithm 1, they derive an upper bound on the achievable rate using Jensen's inequality. This upper bound is shown to be surprisingly tight. Crucially, they then derive a closed-form expression for the linear assignment matrix that maximizes this upper bound, relying solely on second-order CSI statistics (like channel correlation matrices). This approach avoids computationally intensive averaging and offers a significant speedup.

They also investigate special cases, such as single-antenna transmitters and i.i.d. fading channels, to derive optimal or near-optimal strategies.

Datasets

Not discussed in the paper. The evaluation relies on simulated channel models with specified correlation properties.

Key Insights and Ideas

  • Statistical CSI is Sufficient for Near-Optimal Performance: The paper demonstrates that even with only statistical CSI, performance very close to the theoretical upper bound can be achieved. This is a significant finding for practical system design.
  • The Power of Jensen's Inequality for Bounds: The clever application of Jensen's inequality to a difference of terms in the achievable rate formula yields a strict upper bound that is both theoretically sound and practically useful. This technique is highlighted as a valuable tool for future research in similar communication problems.
  • Second-Order Statistics Mimic DPC Structure: The derived closed-form linear assignment matrix, based on second-order CSI, exhibits a structure remarkably similar to that of DPC designs that use instantaneous CSI. This suggests that statistical CSI can effectively capture some of the interference-handling capabilities of DPC.
  • Computational Complexity Reduction is Key: The stark contrast in computational complexity between the iterative Monte Carlo approach and the closed-form statistical CSI approach underscores the practical importance of the latter for real-world deployment.

Key Results

  • Algorithm 2 Achieves Near-Optimal Performance: The proposed low-complexity design (Algorithm 2) achieves sum-rate performance very close to the exact iterative optimization method (Algorithm 1) and significantly outperforms the baseline "TDMA" scheme.
  • Significant SNR Gains: For example, the proposed designs offer SNR gains of around 4.5 dB and 7 dB over the TDMA scheme at a sum-rate of 10 b/s/Hz in different scenarios.
  • Massive Computational Speedup: The running time for Algorithm 2 is orders of magnitude faster than Algorithm 1, making it highly practical for real-time implementation. For instance, evaluating the gradient in Algorithm 2 takes milliseconds, while Algorithm 1 takes seconds for similar tasks.
  • Convergence: Algorithm 1, despite its complexity, converges within a few iterations, indicating its stability.

Limitations and Future Work

  • Monte Carlo Dependency in Algorithm 1: The primary limitation of the initial iterative algorithm is its reliance on Monte Carlo simulations, which can be computationally intensive.
  • Approximation vs. Exact Optimality: While Algorithm 2 provides a near-optimal solution based on an upper bound, it's not guaranteed to be the absolute optimum. The paper notes that the performance is "close" to the exact optimum.
  • Future Work: The authors suggest that the derived closed-form expressions and the understanding of the upper bound's tightness can pave the way for applying existing MIMO BC algorithms (designed for instantaneous CSI) to the statistical CSI scenario. They also mention the potential for further exploration of Rician fading channels to understand the impact of line-of-sight paths.

Summary updated

BibTeX
@article{piantanida2014transmit,
  author = {Pablo Piantanida and Yongpeng Wu and Shi Jin and Xiqi Gao and Matthew R. McKay and Chengshan Xiao},
  title = {{Transmit Designs for the MIMO Broadcast Channel With Statistical CSI}},
  journal = {IEEE TSP},
  year = {2014},
  doi = {10.1109/tsp.2014.2336637},
}
Figure